Single-cell atlas of PBMC and CAR T cells reveals post-infusion expansion of CD8+ T cells
We investigated eight patients from a clinical trial21 (Supplementary Table S1) with relapsed or refractory B cell malignancy following HSCT who were subsequently treated with second generation CAR T cells (Methods). Seven of these patients (P1-P5, P7, P8) received IP generated from cells obtained from the same sibling HSCT donor, and one patient (A1) had an autologous CAR T transfer expressing an identical CAR to the allogeneic (donor-derived) IP. All patients had engrafted before infusion of CAR T cells except for P3, who experienced partial donor marrow chimerism at the time of administration of CAR T and thus the endogenous immune repertoire was not allogenic to the CAR T cells. Patients underwent cyclophosphamide/fludarabine lymphodepletion conditioning for three days prior to infusion, resulting in cytopenia (Fig. 1A). Robust expansion of CAR T cells was observed with peak count occurring within the first 2 weeks (Fig. 1B). As previously reported, two patients (P2 and P8) developed CAR T cell derived lymphoma21,22.
We sought to understand the evolution of the reconstituting immune populations, and of the allogenic CAR T cells during the initial stage of therapy (Fig.1A). Single cell multi-omics data were thus generated from peripheral blood mononuclear cells (PBMC) one day prior to CAR T cell infusion and up to a month post-infusion. CAR T cells were obtained from the IP and from flow cytometric sorting of CD3+CAR+ cells from post-infusion PBMC samples (Fig. 1A, Extended Data Fig. 1A). A total of 106,976 cells were analysed from four patients (P1, P2, P7 and P8), comprising of single cell measures of gene and extracellular protein expression, as well as full length T cell receptor (TCR) sequences across 20 samples including 11 PBMC and 8 CAR T cell sorted populations and PBMCs from one healthy donor as a reference population (Supplementary Tables S2 and S3). Mass cytometry (CyTOF) was also utilised to longitudinally characterise protein expression profiles of 41 extracellular and intracellular protein markers for PBMCs and CAR T cells obtained from 45 samples of 8 patients (Supplementary Table S4, Fig. 1A). We also identified CAR+ T cells from the scRNA-seq data by detecting CAR T transcripts (Methods), which showed a positive correlation with the number of circulating CAR+ T cells identified using an anti-CAR antibody in the PBMC samples (Extended Data Fig. 1B). Circulating CAR T cells were at low frequencies within the PBMC populations, hence we will refer to CAR+ cells as those isolated via flow cytometry using the anti-CAR antibody, and CAR- cells as cells identified from the PBMC samples.
Dimensionality reduction and clustering of gene and protein expression profiles revealed eight clusters identifying canonical cell types (Fig. 1, C and D), with gene and protein expression profiles consistent between patients (Extended Data Fig. 1C, Supplementary Table S5). These comprised of CD8+ T cells, CD4+ T/regulatory T CD25+FOXP3+ cells, NK/NKT, 𝛾δ T cells, B cells, dendritic cells (DCs), megakaryocytes, and monocytes (Fig. 1C). The population of monocytes decreased in post-infusion samples across all patients (Fig. 1E). As expected, only a small number of B cells were identified in the blood.
The largest cluster was formed by CD8+ T cells including both CAR- (Fig. 1E) and CAR+ (Fig. 1F) cells, which was consistent with the CyTOF data for all eight patients (Extended Data Fig. 1, D and E). Both CAR+ and CAR- CD8+ T cells expressed intermediate levels of cytotoxic (GZMA, PRF1, IL32) and NK-like markers (CD56, CD16, CD161, NKG7, FCGRA) (Fig. 1D, Extended Data Fig. 1F), suggesting that both the CAR- and CAR+ CD8+ T population comprised of cytotoxic NK-like (innate-like) CD8+ T cells23-25. All patients demonstrated an increase in the proportion of CAR-CD8+ T cells (Fig. 1E, Extended Data fig. 1, G and H), and similar increases were found for γδ T cells (Fig. 1E). The majority of CAR+ T cells were found in the CD8+ subset except for the IP in patients P2, P4 and A1 which were dominated by CD4+ (Fig. 1F, Extended Data fig. 1E). Following infusion, CAR+ T cells increased in the CD8+ and CD4-CD8- subsets (Fig. 1F, Extended Data fig 1E). In summary, single-cell multi-omics and mass cytometry revealed significant expansion of CD8+ and γδ T cells following infusion, with CAR+ and CAR- T cells sharing similar molecular profiles.
CAR- and CAR+ T cells form functionally distinct subsets including an NK-like phenotype
To further explore the heterogeneity of the CD8+ T cell population we performed unsupervised analysis within the subset of cells identified as CD3+CD4-CD8+, which comprised of both CAR- and CAR+ T cells. After removing a cluster with high levels of CD56 residing in the NK/NKT subset (Fig. 1C), we identified 6 clusters (C1-C6) (Fig. 2A) each including product, pre- and post-infusion CAR- and CAR+ T cells (Fig. 2, B and C, Extended Data fig. 2A).
Differential gene and protein expression analysis (Supplementary Table S5, Fig. 2D) was then performed to identify cluster phenotypes. Cells in C6 comprised of resting cells and included both naïve and central memory T cells expressing CD27 and CD127 (Fig. 2D), and transcription factors associated to resting state, such as TCF7, IL7R, SELL (Fig. 2D). Cluster C5 revealed an effector memory profile, and proteins associated with tissue homing (CCR4, CCR6), and unconventional mucosal associated invariant T (MAIT) cells expressing high levels of KLRB1 (CD161), and genes IL18R1, CXCR4, MAF, SLC424,26-28. Cells within C4 expressed activation protein markers CD38 and HLA-DR, and genes associated to proliferation, e.g., MKI67, HMGB2, BIRC5, as well as homeostatic cytokine receptors IL15RA. Cells in C4 expressed exhaustion proteins Tim-3, PD-1, and genes TOX, TIGIT, CX3CR16,28,29, as well as Protein Tyrosine Phosphatase Non-Receptors PTPN11 (encoding SHP-2) and PTPN6 which have been associated with exhaustion30,31 (Fig. 2D). This profile, together with a distinct cell-cycle signature (Extended Data Fig. 2B), is consistent with the highly proliferative dysfunctional subset identified within tumour-infiltrating CD8+ T cells in melanoma29. Clusters C2 and C3 comprised of cells with both effector memory phenotype, expressing activation markers CD38 and HLA-DR, and CD27, with C3 largely formed by CAR+ T cells (Fig. 2C, Extended Data fig. 2A). These two clusters revealed progenitor of exhausted cells (TPEX), having intermediate levels of proteins PD-1, Tim-3 and LAG-3, (Fig. 2E) and genes TIGIT, GZMK, TOX, PTPN1 and PTPN11 (SHP2)32 (Fig. 2D). These cells retained levels of AP-1 transcription factors FOS and JUN, which are known to be associated with resistance to exhaustion in CAR T cells33.
We discovered a shared NK-like signature among CAR+ and CAR- cells within cluster C1. C1 expressed predominantly a TEMRA phenotype (CD45RA+CD27-CD28-) and expressed NK-like markers CD56, CD16, CD11c, and KLRB1 (Fig. 2D) at comparable levels to those expressed by NK and NKT cells (Extended Data Fig. 1F). This NK-like profile in CD8+ T cells has been previously observed in viral infection34 and cancer23, as well as in CAR T cells under persisting stimulation in vitro15. These cells also expressed CD154 (and gene CD40L) which is known to induce T cell expansion via interactions with dendritic cells producing IL-12p7035, as well as IL10RA and TGFB, which are involved in the inhibition of cytotoxicity in NK cells36. Besides expression of classical CD8 cytotoxicity marker genes (e.g., PRF1, IFNG, GZMB) (Fig. 2D), these cells expressed NKG7, KLRG1, KLRD1, and IL18R1 (Fig. 2D), and transcription factors KLF2, ZEB2, and TBX21 which are involved in regulation of T cell differentiation and trafficking37-39. The NK-like profile was also detected in the TPEX subsets (C2 and C3), although the expression of genes and proteins characterising this signature was reduced compared to C1 (Fig. 2D). These three clusters (C1-C3) shared signatures of activation (LCK) and homing (CCL5), as well as high levels of HLA-E, and HLA-C, which are ligands for NKG2A and KIR, respectively, which indicate interactions with NK cells .
Gene Set Enrichment Analysis (GSEA) confirmed the distinct gene signatures of each cluster, notably the enrichment of both CD8 T cell cytotoxicity as well as NK-like signature in clusters C1 (Extended Data Fig. 2C, Supplementary Table S6). Cells in clusters C1, C2 and C3 had enriched exhaustion profile in line with in vitro findings15.
We next sought to identify phenotypic changes of CAR+ cells as they molecularly and phenotypically evolve from product to circulating post-infusion, and CAR- cells from pre- to post-infusion. The key finding is that both CAR- and CAR+ CD8+ T cells had an up-regulation of NK-like markers at post-infusion, suggesting evolution towards an NK-like profile. Specifically, focusing first on CAR+ cells, we found circulating CAR+CD8+ T cells had increased cytotoxic, exhausted and NK-like profiles, and decreased memory-like phenotype compared to the infusion product (Extended Data Fig. 3, A and B). Activation (CD27, CD28), inhibitory (PD-1), and NK-like (KLRB1 (CD161)) markers were up-regulated in circulating cells compared to product, across all clusters (Extended Data Fig. 3A). Gene expression revealed extensive differences (26, 971 differential expressed genes, adjusted p-value < 0.1, Extended Data Fig. 3B, Supplementary Table S5) which confirmed circulating cells had increased cytotoxic (GZMH), NK-like signatures (IFNG, KLRC1, KLRG1, FCRL6, CD160), TCR signalling (ZAP70), and TPEX markers (TIGIT, LAG-3, TOX, and CX3CR1)32,40. Circulating cells also had increased levels of transcription factors associated with regulation of T cell differentiation such as BATF41,42, and IRF9 and TGFB which are known to regulate T cell effector functions43,44 (Extended Data Fig. 3B). In contrast, cells in the IP expressed higher levels of memory markers IL7R, IL2RA as well as transcription factors controlling T cell activation and favouring memory formation, EGR1, BATF341, and BACH242. CAR- CD8+ T cells had significantly less differentiated genes from pre- to post-infusion across all clusters (424 genes, adjusted p-value <0.1, Extended Data Fig. 3C, Supplementary Table S5), but showed similar increases in cytotoxic (GZMB, GZMH) and NK-like markers (FCGR3A (CD16), NKG7) from pre- to post-infusion (Extended Data Fig. 3C).
Finally, we validated the heterogenous CD8+ T clusters including the NK-like subset across the 8 patients, utilising mass cytometry data (Extended Data Fig. 4, A-E). These data confirmed the presence of phenotypically distinct subsets of CD8+ T cells and four major clusters (M1-M4) classified into cytotoxic M1 (KLRG1, Perforin, Granzyme B, T-bet), memory TPEX M2 (CD69, CXCR3, CD39), naïve and central memory M4 (SELL, IL7R, CD28, CD45RA), and finally a proliferation/activation subset M3 (CD38, HLA-DR, KI67). Notably, the NK-like signature we previously observed in cytotoxic C1 and to a less extent within the memory TPEX clusters (C2 and C3) was confirmed via mass cytometry in all patients, with a high expression of CD56, CD16, and CD49D23,24 in the cytotoxic (M1) and to a less extent in the memory (M2) clusters (Extended Data Fig. 4, C and E).
In summary, our findings from both multi-omics and mass cytometry identified the presence of cytotoxic, exhausted, NK-like CD8+ T cells which are present in both CAR- and CAR+ subsets.
Circulating CAR+ CD8+ T cells differ from CAR- CD8+ T cells by increased activation and proliferation with distinct NK-like profile
Focusing on post-infusion samples, we reasoned that despite the similarities, CAR+ and CAR- CD8+ T cells also differ in molecular and phenotypic markers within each cluster. We observed that CAR+ T cells had a higher level of surface markers consistent with increased activation (CD27, CD38 and HLA-DR) and exhaustion (Tim3, PD-1) across all clusters (Fig. 2E). Transcriptionally, we observed even greater distinction between CAR+ and CAR- T cells within clusters. Notably, CAR+ T cells within cytotoxic (C1) and memory TPEX (C3) clusters showed up-regulation of cytotoxic (GZMK, FGFBP2), NK-like (KLRC1 encoding NKG2A) and exhaustion markers (TIGIT, CXCR3, NELL2, PRDX3)27,28 (Fig. 2G, Supplementary Table S5). In contrast, CAR- T cells up-regulated cytotoxic gene GMZH, chemokine CXCR3, NK-like marker KLRD1 (CD94), and transcription factors MYBL1 and KLF3, and ZEB2 which is known to drive T cell differentiation39. CAR- T cells in memory TPEX C3 and effector C5 also had up-regulation of transcription factors TBX21 and ZNF683 (Hobit), and Fc-gamma receptors (FCGR3A, FCER1G, and FCRL6). GSEA also identified post-infusion CAR+ T cells were enriched in pathways for metabolic activities (oxidative phosphorylation, electron transport), DNA replication but a reduced enrichment in regulatory/inhibitory functions mediated by TNF and NK receptors (Supplementary Table S6), compared to CAR- cells. Taken together, this suggests that although post-infusion CAR- and CAR+ CD8+ T cells share an up-regulated cytotoxic-NK like phenotype compared to the product and pre-infusion samples, there are specific cytotoxic and NK-like markers which delineate CAR- and CAR+ cells at post-infusion. Other distinguishing cluster specific differences were also observed, related to differentiation, and regulation of activation and proliferation.
CAR- and CAR+ CD8+ T cells shared a differentiation trajectory through TPEX towards cytotoxic and NK-like state
We investigated the extent of clonal expansion of both CAR- and CAR+ CD8+ T cells in the previously identified CD8+ T phenotypes, reasoning that the observed shared distribution of cell states reflects the presence of common mechanisms underlying the differentiation process of both subsets. In each subset, clonal expansion was quantified by identifying full-length αβ-TCR sequences from the single cell data utilising long-read sequencing45. A total of 2054 and 1016 clones were identified in CAR-CD8+ and CAR+CD8+ T cells, respectively (Extended Data Fig. 5, A and E). Focusing first on CAR-CD8+ T cells, we observed an increase in the number of expanded clones from pre- to post-infusion (Fig. 3A), and the Shannon evenness index, which is a measure of the degree of clonal diversity, revealed a reduction in diversity in post-infusion samples (Extended Data Fig. 5B). Clonal expansion in post-infusion samples occurred mainly in cytotoxic (C1) and memory TPEX clusters (C2, C3) (Fig. 3B), which contrasted with the distribution of clones identified in pre-infusion samples, which were expanded in the cytotoxic (C1) and the proliferative (C4) subsets (Extended Data Fig. 5C). The majority of clones that were detected in pre-infusion samples persisted following the infusion of CAR T cells across all clusters, with the majority of them mapped within the cytotoxic cluster (Fig. 3C). Persisting clones showed transition between clusters from pre- to post-infusion samples, thus confirming intra-clonal phenotypic plasticity (Extended Data Fig. 5D).
CAR+CD8+ T cells were characterised by a more diverse clonal repertoire with reduced clonal expansion when compared to the endogenous subset (Extended Data Fig. 5F). CAR+CD8+ clones were found across all clusters, predominantly within the proliferating (C4) and TPEX (C3) clusters (Fig. 3D, Extended Data fig. 5G), largely reflecting the higher proportion of CAR+ T cells within these two clusters. TCR clonal analysis of two IPs which were administered in P2 and P8 revealed a very diverse repertoire with no evidence of large clones, thus suggesting a highly diverse TCR repertoire in the IP.
We interrogated the VDJdb database of TCR sequences with known epitope specificity46 to identify epitope-specificity of these receptors. For both CAR- and CAR+ T cells, we identified TCRs with specificity for influenza, CMV and EBV epitopes (Supplementary Table S7). An exception was P7, where a single clone comprised 61% of the circulating CAR+ cells (Fig. 3E). This clone was identified in each of the 6 clusters and carried a TCR αβ that matched a known sequence specific for human cytomegalovirus (CMV) epitope (Supplementary Table S7). Collectively, this data suggests that both CAR- and CAR+ T cells undergo clonal expansion and displayed intra-clonal phenotypic plasticity, with evidence of clonal expansion of pathogen specific T cells.
We next set to investigate the dynamics and differentiation trajectory between cell clusters, and identify the relationship between the observed cell states. For this purpose, we applied established bioinformatic tools on the gene expression data, by first constructing a graph depicting inter-cluster connectivity using PAGA47 (Fig. 3, F and G). There was a strong predicted connectivity between C6, C2 and C1 for CAR- T cells (Fig. 3F), and between clusters C6, C2, C3, and C1 for CAR+ T cells (Fig. 3G). To identify the directional transitions between these clusters and hence construct the differentiation trajectory, we applied RNA velocity48, which confirmed the inferred connectivity between clusters and identified a differentiation process from naïve and central memory C6 into an intermediate state identified as the memory TPEX cluster C2 (CAR-) or C2 and C3 (CAR+), and finally towards the terminally differentiated cytotoxic NK-like state (C1) (Extended Data Fig. 6, A and B). These two trajectories differed only in the addition of the intermediate cluster C3 in the CAR+ trajectory, consistent with the prevalence of this cluster in CAR+ T cells (Extended Data Fig. 2A). Further evidence of these trajectories was provided by inference of directional transition probabilities between clusters utilising CellRank25, which revealed a shared cluster transition pattern for both CAR- (Extended Data Fig. 6C) and CAR+ T subsets (Extended Data Fig. 6D). Both RNA velocity and CellRank also revealed that cells within the C4 and C5 clusters were characterised by a self-cycling state (Extended Data Fig. 6, A-D).
We calculated pseudotime values associated with the CAR- (Fig. 3H) and CAR+ (Fig. 3I) trajectories to quantify the differentiation and fate from the root cluster C6 to the terminal state C1. Along these two trajectories, we quantified the changes in gene signature scores (Fig. 3J, Extended Data fig. 6E) which revealed increasing TCR signalling, cytotoxicity and NK-like activation signatures, consistent with an increased clonal expansion into the terminal state (C1) (Fig. 3J). In contrast, declining trends were detected for TPEX cells and AP-1 transcription factors. CAR- T cells had an increased NK-like suppression signature compared with CAR+ cells but decreased metabolic signatures, e.g., oxidative phosphorylation and glycolysis, and MTOR signalling (Fig. 3J, Extended Data fig. 6E). These profiles were confirmed by protein expression profiles along the trajectory (Extended Data Fig. 6F), showing increased expression of CD45RA and CD56, confirming the increase of TEMRA and NK-like signatures.
In summary, phenotypic and clonal data are consistent with a shared trajectory between CAR- and CAR+ T cells which terminated at a cytotoxic NK-like phenotype and identified a TCR-dependent contribution to the evolution of CD8+ T cells.
Clonally expanded γδ T cells revealed phenotypic plasticity and repertoire diversity
We next analyzed transcriptional and protein profiles of γδ T cells under the hypothesis that the heterogeneous molecular and phenotypic subsets form a differentiation trajectory similar to that followed by CD8+ T cells. Gene expression profile of γδ T cells, identified as CD3+CD4-CD8-TRDC+ subset (Extended Data Fig. 7A), revealed 7 clusters (G0-G6) (Fig. 4A), which confirmed a similar molecular profile with CD8+ T cells (Fig. 4B, fig. 2D).
In contrast to the single cytotoxic cluster (C1) identified in CD8+ T cells, we identified two distinct clusters of cytotoxic γδ T cells, G0 and G1, with cluster G0 presenting higher expression of GNLY, IFNG and GZMA when compared to G1, despite similar protein expression profile consistent with a TEMRA phenotype (CD45RA+, CD27-) and high expression of CCR4 (Fig. 4B). Clusters G2 and G3 comprised of effector memory (CD127+, CD45RO+), expressing CD69, TCF7, FOS and JUN, as well as exhaustion markers (TIGIT, LAG3, Tim3), suggesting evidence of TPEX. Cells in G0-G3 expressed CD56, CD154 (CD40L), KLRB1, and TNFAIP3 (encoding A20), consistent with an NK-like signature49,50. Similar to the C4 cluster in CD8+ T cells, γδT in G4 had a proliferating (MKI67, STMN1, HMGB2), activation (CD38, HLA-DR) and inhibitory (PD-1, TIM3) phenotype. Cluster G5 comprised of a small subset of cells resembling the effector MAIT phenotype observed in CD8+ T cells. Cluster G6 comprised a small subset of cells with a memory-like signature (IL7R). Most clusters expressed the butyrophilin gene BTN3A3, and HLA genes (Supplementary Table S5), which indicate interactions between γδ T cells via TCR mediated recognition of minor HLA molecules17,19. In summary, γδ T cells comprised of similar phenotypes with the CD8+ T cells, including a cytotoxic NK-like phenotype shared between clusters G0-G3.
To investigate the intra-clonal phenotypic heterogeneity and cell fate of γδ T cells, full-length TCR sequences were used to identify clonally expanded subsets, which were then mapped onto the identified clusters. Full-length sequences were recovered for up to 80% of cells (Extended Data Fig. 7B). Clonotypes were observed in all samples and across all clusters (Fig.4C, Extended Data Fig. 7C), and with the clone size increasing from pre- to post-infusion (Fig. 4D). The majority of expanded clones identified at pre-infusion samples carried a TRDV2-TRGV9 genotype (Extended Data Fig. 7D), which are commonly found in circulation in humans19,50, while clones at post-infusion were dominated by TRDV1 or TRDV3 genes (Fig. 4E), with TRDV1 paired with TRGV9, TRGV8, or TRGV4 (Extended Data Fig. 7D). TRDV1, TRDV2 and TRDV3 genotypes occupied all clusters (Extended Data Fig. 7E). Similar to CD8+ cells, we also found intra-clonal phenotypic heterogeneity within clones, revealing a transcriptional transition between clusters across time points (Extended Data Fig. 7F).
To study the dynamics between γδ T cell states, we performed similar trajectory analysis to the CD8+ T cells with RNA velocity (Fig. 4A), as well as by inferring cluster connectivity and differentiation trajectories using PAGA and CellRank (Extended Data Fig. 7, G and H). γδ T cells differentiated along a trajectory largely overlapping with the CD8+ T cells, originating from the naïve central memory subset (G6) and evolving through the memory-TPEX clusters (G2/G3), and finally terminating towards the cytotoxic states (G1, G0) (Fig. 4F). The directionality of this trajectory correlated with the increased clonal expansion measured in cytotoxic clusters G0 and G1 (Fig. 4C). Transition probabilities between clusters confirmed the self-cycling nature of the highly proliferating cells in G4, with evidence of transition towards memory G3 and cytotoxic clusters G0 (Extended Data Fig. 7H).
Gene signature scores along the inferred trajectory revealed an increase in exhaustion, cytotoxic and NK-like profiles towards the terminal cytotoxic cluster (G0), while NK-like suppression peaked in the intermediate cell states (G2, G3) (Fig. 4G). Along this trajectory, transcription factors involved in memory differentiation such as TCF7, GATA3, JUNB, NR4A2 declined, while TBX21 increased (Fig. 4H).
Overall, these results revealed significant clonal expansion of unconventional γδ T cells predominantly due to an expanding TRDV1 clonotype, recapitulated known gene heterogeneity among γδ T cells50, and demonstrated a shared differentiation trajectory and phenotypic plasticity with CAR+ and CAR- CD8+ T cells, which differentiated into a cytotoxic, NK-like and exhausted state.
CAR+ and CAR- T-cells are associated with altered serum cytokines levels following infusion
We next investigated whether the heterogenous clusters we discovered were associated with changes in serum cytokine levels, which were measured for the eight patients before CAR T cell infusion and at time-points within the first 30 days post-infusion21,22. Comparison with pre-infusion baseline levels revealed an increase at post-infusion in pro-inflammatory cytokines IL-18, IL-12p40, and a decrease in MCP1 (encoded in CCL2) and homeostatic cytokine IL-15 (p-value<0.05, Fig. 5A). For the four patients with single cell multi-omics data (P1, P2, P7, P8), we assigned a gene-set score quantifying the degree to which each immune cell subset was expressed in pre- and post-infusion, including the CD8+ clusters, and calculated the Spearman correlation between these gene-set scores and measured cytokine levels to form an association score. This association score thus quantifies the degree to which cytokines simultaneously increases or decreases with each cluster gene signature. We also calculated this association between measured CAR+ T counts and clusters. Significant associations were found between pro-inflammatory cytokines IL-18, IL12p40, TNF-α, and IFN-γ, with γδT and NK cells (Fig. 5B), and with CAR+ T counts (Fig. 5B). These cytokines were also associated with CAR+ CD8+ T cells in clusters C1 (cytotoxic), C2 and C3 (memory TPEX) (Fig. 5C), but not with the corresponding CAR- CD8+ subsets (Fig. 5D). These associations are consistent with the increased proportion of NK and NK-like CD8+ subsets in circulation, which are known to respond to innate-like cytokine stimulation such as IL-12 and IL-18 to make IFN-𝛾23-25. In contrast, decline in homeostatic cytokine IL-15 was associated with a corresponding reduction of CAR+ and CAR- CD8+ T cells in clusters C4 and C6 (Fig. 5, C and D), as well as monocytes and CD4+ cells (Fig. 5B).
We used mass cytometry to validate the cytokine-CD8+ T cluster associations on the remaining patients for CAR+ (Fig. 5E) and CAR- CD8+ T cells (Fig. 5F). Pro-inflammatory cytokines (TNF-α, IL-18, IL12p40, and IFN-γ) were positively associated with CAR+ T cells with cytotoxic and effector markers (T-bet, perforin, and granzyme B), but also CD49D, CD56, and CCR6 which confirmed the associations with the NK-like signature from multi-omics data23,24. In both the CAR- and CAR+ T compartment, homeostatic cytokine IL-15 was associated with proliferation (HLA-DR, CD38) and exhaustion markers (CD39 (encoded in ENTPD1), PD-1, and Tim-3), (Fig.5, E and F). Finally, CAR+CD8+ T cells expressing chemokine receptors (CXCR3, CXCR5) were positively associated with IL-15, which were found expressed on the memory -TPEX and proliferative T cells subsets (Fig. 5E).
We next investigated whether the association between the cell populations and the cytokine levels were consistent with cell-to-cell interactions between endogenous CAR- immune cells and CAR+ T cells, for instance through cytokine-cognate receptor interactions. To test this hypothesis, we predicted the cell-to-cell interactions between the circulating immune cells and CAR+ T cells by computing statistically significant interactions mediated by chemokines, cytokines and HLA molecules following infusion using CellphoneDB51. We identified a total of 9143 interactions (Fig. 5G) between CAR+ T cells and other immune cells. These cells interacted predominantly with CAR+ T cells in C4, C5, and C6 via HLA-E expression (Fig. 5H). CAR+ CD8+ T cells in the clusters of memory TPEX (C2) and proliferative subset (C4) were predicted to interact with endogenous CD8+ and 𝛾ẟ T cells via chemokine receptors CCR5 (Fig. 5H). A regulatory set of interactions was also discovered between CAR+ and CAR- T cells, via expression of co-inhibitory molecules such as KLRB1-CLEC2D, TIGIT-NECTIN3, HAVCR2 (Tim-3)-LGALS9, TGFB, and TNF receptors such as TNFRSF25-TNFRSF12 (Tweak), and TNFRSF14, which interact with multiple ligands influencing both activation and regulation52 (Fig. 5H).
Together, these data revealed associations between NK/NKT and NK-like CD8+ T cells with pro-inflammatory cytokines, including the innate-like IL-18 and IL-12, while proliferating and activated CD8+ T cells were associated with homeostatic cytokine IL-15. Lastly, cell-cell interactions with CAR T cells contribute to homing and regulatory functions of recovering immune cells following infusion.