The remission status of AML patients after allo-HCT is associated with a distinct single-cell bone marrow T-cell signature

Abstract

Acute myeloid leukemia (AML) is a hematologic malignancy for which allogeneic hematopoietic cell transplantation (allo-HCT) often remains the only curative therapeutic approach. However, incapability of T cells to recognize and eliminate residual leukemia stem cells might lead to an insufficient graft-versus-leukemia (GVL) effect and relapse. Here, we performed single-cell RNA-sequencing (scRNA-seq) on bone marrow (BM) T lymphocytes and CD34+ cells of 6 patients with AML 100 days after allo-HCT to identify T-cell signatures associated with either imminent relapse (REL) or durable complete remission (CR). We observed a higher frequency of cytotoxic CD8+ effector and gamma delta (γδ) T cells in CR vs REL samples. Pseudotime and gene regulatory network analyses revealed that CR CD8+ T cells were more advanced in maturation and had a stronger cytotoxicity signature, whereas REL samples were characterized by inflammatory tumor necrosis factor/NF-κB signaling and an immunosuppressive milieu. We identified ADGRG1/GPR56 as a surface marker enriched in CR CD8+ T cells and confirmed in a CD33-directed chimeric antigen receptor T cell/AML coculture model that GPR56 becomes upregulated on T cells upon antigen encounter and elimination of AML cells. We show that GPR56 continuously increases at the protein level on CD8+ T cells after allo-HCT and confirm faster interferon gamma (IFN-γ) secretion upon re-exposure to matched, but not unmatched, recipient AML cells in the GPR56+ vs GPR56- CD8+ T-cell fraction. Together, our data provide a single-cell reference map of BM-derived T cells after allo-HCT and propose GPR56 expression dynamics as a surrogate for antigen encounter after allo-HCT.

Graphical abstract

Figure 1.

Single-cell characterization of BM HSPCs and T cells from patients with REL and those with CR. (A) Overview of the experimental setup. (B) Uniform manifold approximation and projection (UMAP) of 8492 post-QC cells representing BM HSPCs and T cells from 6 patients with AML on day 100 after allo-HCT. Cells are colored according to cell type. HSPCs included myeloid/lymphoid progenitors (MLP), B-cell precursors (preB), T-cell precursors (pre/proT), megakaryocyte-erythroid progenitors (MEP), neutrophil progenitors (NP), monocyte-dendritic progenitors (MDP), late monocytic precursors/monocytes (MP/mono), and dendritic cells (pDCs, cDCs). (C-D) Scaled expression (C) of marker genes and TF activity (D) are shown for each cell type cluster in panel B as heat maps. Values are averaged across all cells in the cluster. TF activity is obtained from SCENIC (“Methods”). (E-G) UMAP as shown in panel B indicating the normalized gene expression of selected genes, with the range of normalized gene expression indicated in the parenthesis of each UMAP. QC, quality control. The BM biopsy illustration in the lower part of panel A was adapted from the original work of Cancer Research UK under a CC BY-SA 4.0 license.

Figure 2.

Altered BM T-cell composition is associated with the remission status. (A) UMAP highlighting CR and REL cells (green, REL; blue, CR). (B) Absolute numbers of cells across CD3⁺ T-cell types. The different colors indicate REL (green) and CR (blue) samples. (C) Differential abundance per cell type, within the CD3⁺ population. The bars represent the log2 odds ratios (Fisher exact test, P value after Bonferroni correction; n.s., not significant; ∗∗∗P < .0001). (D) (Left) Scaled expression (z-score) of publicly available cellular indexing of transcriptome and epitopes (CITE)-seq data. Features (x-axis) with antibody (AB) suffix indicate that the measurement was performed on protein level. (Right) Percentage of cells per cluster (x-axis) that map to public reference clusters (y-axis). Bold cluster names indicate CD8⁺ T_EMRA subsets (CD45RA⁺ CCR7^–). (E) Comparison of the CD8⁺ pseudotime (calculated with Monocle3) between CR and REL using naive CD8⁺ as the starting population. Box plot (top) and density plot (bottom) depicting the pseudotime of CD8⁺ cells in CR (blue) and REL (green). Difference between CR and REL was assessed using t test. (F) Heat map depicting the scaled expression across pseudotime of selected effector, memory, and exhaustion genes in CD8⁺ cells. (G) Diffusion maps for the CD8⁺ cells colored according to the inferred pseudotime using Monocle3 split per condition (top) and based on the clusters (bottom).

Figure 3.

CD8⁺ T cells in REL samples have lower cytotoxic potential. (A) Differential TF activity analysis using SCENIC. Heat map indicating log2 odds ratios; only significantly enriched TFs are colored (Fisher exact test false discovery rate < 0.05) (left). Characterization of target genes based on known gene sets (right). Assignment of target genes to known functions was performed using publicly available gene sets (supplemental Methods). The colored bars represent the fraction of target genes per TF, which belong to the different gene sets (IFN, IFN response; Activation, Immune cell activation; TNF, TNF signaling). (B) CD8⁺ T gene regulatory network (GRN) of exemplary differentially active TFs (TBX21, REL, and FOS) and their target DEGs. (C) UMAP highlighting the T-cell clusters that belong to the CD8⁺ effector memory (EM) cells and were used for the differential expression analysis. (D) Heat map depicting scaled expression (z-score) across all 6 samples of 235 DEGs (CR, 144 genes; REL, 91 genes) in CD8⁺ EM clusters. Analysis was performed using the MAST algorithm (log2FC > 0.5; adjusted P value [p.adj] < 0.05 after Bonferroni correction). (E) Gene ontology and hallmark enrichment analysis on the DEGs from D. Terms were selected from the top-enriched terms. Full list provided in supplemental Table 8. (F) UMAP indicating the normalized gene expression of ADGRG1/GPR56. (G) GPR56 and CD27 expression across pseudotime of CD8⁺ cells, split per condition. The bar plots below indicate the percentages of GPR56⁺ and CD27⁺ cells in REL and CR samples when considering all CD8⁺ clusters. (H) Percentage of GPR56 positive cells in the indicated fractions determined by flow cytometry in the same 6 samples used for scRNA-seq. (Top left cartoon) Gating strategy used to identify naive, central memory (T_CM), EM (T_EM), and CD45RA⁺ EM (T_EMRA) cells using CCR7 and CD45RA. P values were calculated using Student t test. (I) Volcano plot illustrating the DEGs between GPR56⁺ (purple) and GPR56^– (orange) CD8⁺ T_EM cells of patients in CR. The y-axis represents P value after Bonferroni correction (p.adj), and points were colored according to absolute log2FC > 0.5 and P.adj < .05 (purple/orange). (J) Box plots illustrating the percentage of GZMB⁺ (top) and PRF1⁺ (bottom) T cells in the GPR56⁺ and GPR56⁻ fractions assessed by intracellular flow cytometry. Connected points indicate fractions originating from the same sample. P value was calculated using paired Wilcoxon test, n = 10.

Figure 3.

CD8⁺ T cells in REL samples have lower cytotoxic potential. (A) Differential TF activity analysis using SCENIC. Heat map indicating log2 odds ratios; only significantly enriched TFs are colored (Fisher exact test false discovery rate < 0.05) (left). Characterization of target genes based on known gene sets (right). Assignment of target genes to known functions was performed using publicly available gene sets (supplemental Methods). The colored bars represent the fraction of target genes per TF, which belong to the different gene sets (IFN, IFN response; Activation, Immune cell activation; TNF, TNF signaling). (B) CD8⁺ T gene regulatory network (GRN) of exemplary differentially active TFs (TBX21, REL, and FOS) and their target DEGs. (C) UMAP highlighting the T-cell clusters that belong to the CD8⁺ effector memory (EM) cells and were used for the differential expression analysis. (D) Heat map depicting scaled expression (z-score) across all 6 samples of 235 DEGs (CR, 144 genes; REL, 91 genes) in CD8⁺ EM clusters. Analysis was performed using the MAST algorithm (log2FC > 0.5; adjusted P value [p.adj] < 0.05 after Bonferroni correction). (E) Gene ontology and hallmark enrichment analysis on the DEGs from D. Terms were selected from the top-enriched terms. Full list provided in supplemental Table 8. (F) UMAP indicating the normalized gene expression of ADGRG1/GPR56. (G) GPR56 and CD27 expression across pseudotime of CD8⁺ cells, split per condition. The bar plots below indicate the percentages of GPR56⁺ and CD27⁺ cells in REL and CR samples when considering all CD8⁺ clusters. (H) Percentage of GPR56 positive cells in the indicated fractions determined by flow cytometry in the same 6 samples used for scRNA-seq. (Top left cartoon) Gating strategy used to identify naive, central memory (T_CM), EM (T_EM), and CD45RA⁺ EM (T_EMRA) cells using CCR7 and CD45RA. P values were calculated using Student t test. (I) Volcano plot illustrating the DEGs between GPR56⁺ (purple) and GPR56^– (orange) CD8⁺ T_EM cells of patients in CR. The y-axis represents P value after Bonferroni correction (p.adj), and points were colored according to absolute log2FC > 0.5 and P.adj < .05 (purple/orange). (J) Box plots illustrating the percentage of GZMB⁺ (top) and PRF1⁺ (bottom) T cells in the GPR56⁺ and GPR56⁻ fractions assessed by intracellular flow cytometry. Connected points indicate fractions originating from the same sample. P value was calculated using paired Wilcoxon test, n = 10.

Figure 4.

GPR56 is dynamically upregulated by CAR-T cells upon target recognition. (A) Schematic visualizing the experimental setup: Peripheral blood mononucleated cells (PBMCs) from 4 healthy donors were first activated, activated T cells (ATCs) were then transduced with a retroviral vector comprising a CD33.CAR construct. On day 15 of production, CAR-T cells were coincubated with the AML cell line HL60, expressing CD33 on the surface, or with HL60 cells, in which CD33 was knocked out using CRISPR/Cas9 (HL60 CD33 KO). (B) Representative FACS plots showing CD27 and GPR56 expression on CAR-T cells after activation and transduction, but without contact to leukemia cells (upper left), after 5-day coculture with CD33⁺ HL60 (upper right), after coculture with HL60 CD33 KO cells (lower left), and on nontransduced cells after coculture with HL60 CD33⁺ cells (lower right). Note that GPR56 upregulation occurs exclusively when CAR-T cells carrying the CD33.CAR were incubated with HL60 CD33⁺ cells. This suggests that GPR56 upregulation occurs only upon antigen recognition by the T cell receptor. (C) Statistical analysis of the experiment shown in 4B. ∗∗∗ P < .0005. (D) Percentage of GPR56⁺CD27⁺ fractions (blue) and CD15⁺ HL60 cells (orange) in the 4 individual donors during the 5 serial 5-day challenges. (E) (Top) Experimental setup of sorting experiment; CAR-T cells were challenged once with HL60 for 5 days. Then, cultures were sorted for CD8⁺ and the following 4 quadrants: GPR56⁺CD27^–, GPR56⁺CD27⁺, GPR56^–CD27⁺, and GPR56^–CD27^–. Subsequently, equal numbers of sorted cells were re-exposed to HL60 cells for 5 days and subsequently analyzed for surface marker expression. All fractions were capable of eliminating HL60 WT cells except for 1 culture with double-negative cells (data not shown), confirming that all fractions contained the CAR construct. (Lower left) Representative FACS plots of the sorted CAR-T cells after re-exposure to target cells. The label above the plots indicates the originally sorted phenotype. (Lower right) Stacked bar graph showing the distribution of the 4 quadrants in the 4 different conditions. Mean and standard deviation of the 4 donors are shown. The x-axis labels indicate the originally sorted fraction. Colors of bars indicate the output phenotype according to the legend. nt, nontransduced; wt, wild type.

Figure 5.

GPR56 increases after non-self recognition by allo-reactive T cells after allo-HCT. (A) Percentage of CD8⁺ T_EM in BM (left) and percentage of GPR56⁺ on CD8⁺ T_EM (right). Numbers below plots show the median percentage. Numbers between groups of patients without (no-Allo), before (pre-Allo), and after (post-Allo) allo-HCT indicate the P values (unpaired Wilcoxon test). Note that the total fraction of CD8⁺ T_EM in BM does not significantly differ between the 3 groups. Box plots showing medians and quartiles, each dot represents an individual sample. (B) Representative FACS plots showing CD27 and GPR56 expression on CD3⁺CD8⁺ cells in healthy BM (left), in a patient with low GPR56 upregulation (patient 1 [middle]), and a patient with a dominant GPR56⁺CD27^– fraction (patient 2 [right]). (C) Time course of percentage of GPR56^–CD27⁺ (blue), GPR56⁺CD27⁺ (purple), and GPR56⁺CD27^– (pink) in the CD8⁺ compartment in patients with CR after allo-HCT. Numbers above the box plots indicate the median percentages. Box plots represent medians, quartiles, and outliers. (D) Proposed model of CD8⁺ T-cell phenotype switch after allo-HCT. (E) Percentage of GPR56⁺ on CD8⁺ T_EMRA in recipients with CMV IgG negativity (left) and positivity (right). Numbers above box plots indicate the median percentages. Box plots represent medians, quartiles, and outliers. (F) Time course of the percentage of the indicated cell types in patient GXW165. As indicated, CMV status was positive for the recipient and negative for the donor (CMV R/D pos/neg); donor sex was male. The text below the x-axis provides clinical information on the course of the disease. Note the drop in chimerism, neutrophils, and GPR56 positivity accompanied by FACS MRD positivity around day +100. IS was reduced, and the patient developed skin GVHD requiring steroids, which was accompanied by a steady increase in GPR56⁺, re-establishment of full donor chimerism, and CR. (G) Schematic illustrating the experimental strategy for measuring IFN-γ production using ELISpot. GPR56⁺ and GPR56^– CD8⁺ T cells were sorted using FACS from PBMCs of 6 patients with AML in CR after allo-HCT. Sorted populations were then cocultured for 24 hours with primary AML blasts from initial diagnosis of the same patients. ELISpot assay was performed to detect and quantify IFN-γ production by the T cells in response to AML blasts. (H) Representative wells from the ELISpot assay after coculturing primary AML blasts with GPR56^– (orange) and GPR56⁺ (purple) CD8⁺ T cells for 24 hours. (I) ELISpot results showing the numbers of spots generated in the GPR56⁺ and GPR56^– T-cell fractions of 6 patients with AML in remission within 24 hours of contact with matched AML blasts. P value was calculated using paired Wilcoxon test. Individual points indicate biological replicates (mean across technical replicates) and connected points indicate fractions originating from the same sample. Chim, donor chimerism; IgG, immunoglobulin G; IS, under immunosuppression; MRD, minimal residual disease.

Figure 5.

GPR56 increases after non-self recognition by allo-reactive T cells after allo-HCT. (A) Percentage of CD8⁺ T_EM in BM (left) and percentage of GPR56⁺ on CD8⁺ T_EM (right). Numbers below plots show the median percentage. Numbers between groups of patients without (no-Allo), before (pre-Allo), and after (post-Allo) allo-HCT indicate the P values (unpaired Wilcoxon test). Note that the total fraction of CD8⁺ T_EM in BM does not significantly differ between the 3 groups. Box plots showing medians and quartiles, each dot represents an individual sample. (B) Representative FACS plots showing CD27 and GPR56 expression on CD3⁺CD8⁺ cells in healthy BM (left), in a patient with low GPR56 upregulation (patient 1 [middle]), and a patient with a dominant GPR56⁺CD27^– fraction (patient 2 [right]). (C) Time course of percentage of GPR56^–CD27⁺ (blue), GPR56⁺CD27⁺ (purple), and GPR56⁺CD27^– (pink) in the CD8⁺ compartment in patients with CR after allo-HCT. Numbers above the box plots indicate the median percentages. Box plots represent medians, quartiles, and outliers. (D) Proposed model of CD8⁺ T-cell phenotype switch after allo-HCT. (E) Percentage of GPR56⁺ on CD8⁺ T_EMRA in recipients with CMV IgG negativity (left) and positivity (right). Numbers above box plots indicate the median percentages. Box plots represent medians, quartiles, and outliers. (F) Time course of the percentage of the indicated cell types in patient GXW165. As indicated, CMV status was positive for the recipient and negative for the donor (CMV R/D pos/neg); donor sex was male. The text below the x-axis provides clinical information on the course of the disease. Note the drop in chimerism, neutrophils, and GPR56 positivity accompanied by FACS MRD positivity around day +100. IS was reduced, and the patient developed skin GVHD requiring steroids, which was accompanied by a steady increase in GPR56⁺, re-establishment of full donor chimerism, and CR. (G) Schematic illustrating the experimental strategy for measuring IFN-γ production using ELISpot. GPR56⁺ and GPR56^– CD8⁺ T cells were sorted using FACS from PBMCs of 6 patients with AML in CR after allo-HCT. Sorted populations were then cocultured for 24 hours with primary AML blasts from initial diagnosis of the same patients. ELISpot assay was performed to detect and quantify IFN-γ production by the T cells in response to AML blasts. (H) Representative wells from the ELISpot assay after coculturing primary AML blasts with GPR56^– (orange) and GPR56⁺ (purple) CD8⁺ T cells for 24 hours. (I) ELISpot results showing the numbers of spots generated in the GPR56⁺ and GPR56^– T-cell fractions of 6 patients with AML in remission within 24 hours of contact with matched AML blasts. P value was calculated using paired Wilcoxon test. Individual points indicate biological replicates (mean across technical replicates) and connected points indicate fractions originating from the same sample. Chim, donor chimerism; IgG, immunoglobulin G; IS, under immunosuppression; MRD, minimal residual disease.