Expansion of a CD8+ Temra Population and Activating T-Cell Interactions Characterize the Graft Versus Leukemia Response in Relapsed AML

Graft-versus-leukemia (GvL) activity underlies the curative potential of allogeneic hematopoietic stem cell transplantation. Its effectiveness is well-exemplified by donor lymphocyte infusion (DLI), an established adoptive cellular therapy for treatment of relapsed disease. Whereas ~80% of chronic myelogenous leukemia (CML) patients achieve durable remissions to DLI, responses for acute myelogenous leukemia (AML) are only 15-20%. We hypothesized that systematic characterization of immune cell composition and cellular interactions within the marrow tumor microenvironment in relation to DLI would illuminate the molecular basis of the disease-specific susceptibility to GvL. Given the heterogeneous nature of AML, we suspected AML response to comprise more complex leukemic-immune cell interactions than in CML. To these ends, we generated 5’ single-cell RNA- and TCR-sequencing (scTCRseq) and cell surface marker profiling (scCITE-seq) on 24 bone marrow biopsy samples collected before and after DLI from 9 patients with relapsed AML (5 responders [R], 4 nonresponders [NR]). We merged this dataset with our previously reported CML dataset (14 patients, 8 R, 6 NR [PMID: 34758319]) and batch corrected these data with scVI to obtain a total of 414,000 high-quality single-cell transcriptomes for analysis. Evaluation of the immune cell subset proportions over time revealed 10 populations with distinct dynamics in AML responders to DLI but not CML, while 3 other populations showed similar dynamics across diseases. Both CML and AML DLI response were associated with pre-treatment enrichment of an NK cell population highly expressing XCL1, XCL2, CCL3, CCL4, CCL4L2, and CCL5, consistent with a tumor-infiltrating expression profile. Unique to CML (i.e. not observed in AML), in addition to post-DLI enrichment of the TPEX population previously identified, we found a γδ-like T cell population and a classical dendritic cell population exhibiting post-DLI enrichment associated with response to DLI. In AML, response to DLI was most consistently associated with expansion of a CD8+ T cell population (6.03% of total pre-DLI cells vs 18.31% post-DLI for R [p=0.016] compared to 8.68% pre-DLI vs 9.55% post-DLI for NR [p=0.454]) and a naïve B cell population (4.02% pre-DLI vs 11.68% post-DLI for R [p=0.002] compared to 5.42% pre-DLI vs 5.17% post-DLI for NR [p=0.471]). The CD8+ T cell population was characterized by high expression of cytotoxicity genes (e.g. GZMA, GZMB, GZMH, GZMM, PRF1, KLRG1, TOX, EOMES, TBX21) and was most consistent with a T effector memory re-expressing RA (TEMRA) profile, supported by high CD45RA expression by scCITE-seq. This population also highly expressed ZNF683, which encodes the transcription factor Hobit, known to regulate cytotoxicity in CD8+ T cells. Differential gene expression between AML R and NR TEMRA cells revealed enriched expression of ZNF683, GZMH, NKG7, CCL5, and CXCR4 in R and the exhaustion marker TIGIT and the senescence marker KLRG1 in NR. scTCR-seq of the TEMRA subset demonstrated clonal expansion and differentiation into a more proliferative expression profile post-DLI. Among AML cases, 2-20% of the post-DLI repertoire of responders originated from the DLI product while no TCR from the DLI product was detected in the single NR patient available for analysis. We applied predictive interaction modeling among all 50 cell type pairs using CellPhoneDB (PMID: 32103204), which utilizes a publicly available curated database of receptor-ligand interactions. Predicted interactions unique to R between CD8+ TEMRA and AML leukemia cells included CD226-PVR, CD6-ALCAM, and CD46-JAG1, which have been associated with T cell activation. Interactions unique to NR between CD8+ TEMRA and AML leukemia included the inhibitory interactions TIGIT-NECTIN and PTPRC-MRC1. These interactions were not observed in CML patients in either R or NR between CD8+ TEMRA and CML cells, suggesting disease-specific response and resistance mechanisms. Here we show that response to DLI in AML is associated with largely distinct expansion of immune cell populations compared to CML, including not only T cells but an orchestration of B, NK, and myeloid cell subsets. AML responders were uniquely characterized by expansion of a TEMRA subset marked by high expression of the pro-cytotoxic gene ZNF683. Further investigation is needed to evaluate the role of this population in mediating GvL response.