Single-cell characterization of T cells in newly diagnosed AML patients with RUNX1::RUNX1T1

To explore the heterogeneity of T cell compartment and the characteristics of tumor-reactive T cells, we collected BM T cells from 5 newly diagnosed AML patients with RUNX1::RUNX1T1 and performed scRNA-seq and scV(D)J-seq (Fig. 1A). AML patients’ characteristics are described in Supplementary Table S1. To identify potentially tumor-reactive T cells (pTRTs) and help cluster annotation of the BM T cell landscape, we integrated a T cell reference from 20 healthy donors [22] (HDs) with AML patients’ data. In total, 57,186 T cells (25,868 cells from AML patients and 31,318 cells from HDs) were integrated after quality control, doublet removal, and batch correction (Supplementary Figure S1A). In addition, to evaluate T cell activities in AML patients, we performed scV(D)J-seq and identified paired productive ɑ and β chain in 22,214 T cells.

Fig. 1

The single cell profile of T cells from newly diagnosed AML patients’ BM with RUNX1::RUNX1T1. A Schematic overview of the study. BM samples were collected from 5 AML patients with RUNX1::RUNX1T1. The CD3+ T cells were sorted for scRNA-seq and scV(D)J-seq. This figure is edited by BioRender.com. B UMAP visualization of identified 9 CD8+ T cell clusters. The subgroups corresponding to each color in the UMAP plot are indicated in the right-side legend. C Dot plot showing the relative expression of marker genes across different CD8+ T cells. Bubble size is proportional to the percentage of cells expressing a gene and color intensity is proportional to average scaled gene expression. D The diffusion map of CD8+ T cells revealing a continuum of cellular states. The color shows the pseudotime performed by Palantir [19]. E Comparison of the pseudotime (calculated with Palantir) between CD8+ T cells in RUNX1::RUNX1T1 (red) and HD (blue) group showed by boxplot (top) and density plot (bottom). The significance was calculated by the Wilcoxon rank-sum test. F Boxplot showing the proportion of CD8+ T cell in the RUNX1::RUNX1T1 group (red) and the HD (blue) group. Student’s t-test was used to measure the differences between the two groups. p-value: no significance (ns) > 0.05, * ≤ 0.05; ** ≤ 0.01; ***≤ 0.001, **** ≤ 0.0001

Based on classical gene markers, we defined 15,176 CD8+ T cells and annotated unsupervised clusters (Fig. 1B-C, Supplementary Figure S1B). Overall, the CD8+ T cells could be classified into various cell types, including CD8+ naïve T cells (Tn, CCR7), CD8+ central memory T cells (Tcm, naïve markers and CD44), CD8+ effector memory T cells (Tem, GZMK) CD8+ effector T cells (Teff, GZMB), CD8+ mucosal-associated invariant T cells (MAIT, SLC4A10, TRAV1-2, Supplementary Figure S1C), and CD8+ terminally differentiated effector T cells (Temra, TBX21). Through trajectory inference, we discovered that T cells differentiated into two pathways: from naive T cells to MAIT or Temra (Fig. 1D, Supplementary Figure S1D). Compared to healthy donors, T cells in AML BM were significantly more advanced in pseudotime (Fig. 1E, Supplementary Figure S1E). From the distribution of cell clusters, we found that the total percentage of Tn in the RUNX1::RUNX1T1 group was much lower (15.56% vs. 42.45%, p = 0.04), while the percentage of Temra increased (CD8_Temra_FGFBP2: 15.95% vs. 5.72%, p = 0.007; CD8_Temra_EOMES: 7.52% vs. 0.42%, p = 0.04).

In addition, we also defined the 5,555 CD4+ T cells based on classical gene markers (Supplementary Figure S1F-I). Similarly, we could also observe a decrease in Tn (16.45% vs. 73.44%, p < 0.0001) and an increase in more differentiated clusters of RUNX1::RUNX1T1 group (CD4_Tcm: 42.65% vs. 8.30%, p = 0.0007; CD4_Tem_LMNA: 24.43% vs. 5.97%, p = 0.004).

These data suggested that, in the BM of AML patients with RUNX1::RUNX1T1, the immune reserve reduced and differentiated T cells increased, indicating immune mobilization.

The pTRTs exhibited heterogeneity in AML patients with RUNX1::RUNX1T1

Since AML patients exhibited abnormal distribution of T cell proportions in BM, it suggested that T cells might be activated in the presence of tumor antigen stimulation, indicating the presence of pTRTs. As pTRTs, they should be selectively enriched in the AML microenvironment, with TCR capable of recognizing AML tumor antigens and activating downstream signaling pathways, leading to T cell proliferation and clonal expansion (Fig. 2A). Therefore, to distinguish pTRTs from bystander T cells, we used four criteria mainly applied from STARTRAC [23], namely tumor enrichment index (Supplementary Figure S2A), clonal expansion index (Supplementary Figure S2B), proliferation index (Supplementary Figure S2C), and TCR signaling pathway activation [5] (Fig. 2B-C). Based on the four characteristics, we identified three Temra clusters (CD8_Terma_FGFBP2, CD8_Temra_EOMES, and CD8_Temra_KLRC2) as pTRTs and validated them by published gene sets [24] (Supplementary Figure S2D).

Fig. 2

The identification and characteristics of tumor-reactive T cells in AML. A The schema plot for identification of pTRTs. This figure is edited by BioRender.com. B The scatter plot of the tumor enrichment index and clonal expansion index in each CD8+ T cell cluster. The proliferation index is shown in gradient color and the color bar is annotated on the right margin. C Heatmap showing the TCR downstream signal activation level of different clusters. TCR-signaling-related genesets were obtained from https://www.gsea-msigdb.org/gsea/msigdb/. TCR_sgnl: REACTOME_TCR_SIGNALING, Downstream_TCR_sgnl: REACTOME_DOWNSTREAM_ TCR_SIGNALING, pCD3_and_TCRZ: REACTOME_PHOSPHORYLATION_OF_CD3_AND_TCR_ZETA_CHAINS, Second_messenger: REACTOME_GENERATION_OF_SECOND_MESSENGER_MOLECULES, TCRa_pathway: BIOCARTA_TCRA_PATHWAY, ZAP70_to_immuosynapse: REACTOME_TRANSLOCATION_OF_ZAP_70_TO_IMMUNOLOGICAL_SYNAPSE. D UMAP visualization of clusters annotated by TCR repertoire. Red represents pTRT, blue represents pTRT-relavant cells and the rest of the cells are annotated as pTRT-irrelevant cells in grey. E Bar plot of pTRT-relevant cell frequency in each cluster. F The UMAP plot of tumor-reactive T cells. C0_stem: stem-like cluster, C1_transitory: transitory cluster, C2_TD: terminally differentiated cluster. G Dot plot showing the relative expression of marker genes across different tumor-reactive T cell clusters. Bubble size is proportional to the percentage of cells expressing a gene and color intensity is proportional to average scaled gene expression. H The diffusion map of tumor-reactive T cells. The color annotations are on the right margin. The differential trajectory is indicated by the grey arrow

To explore the origin of pTRTs, we conducted TCR-based tracing. We defined a group of cells that shared TCR with pTRTs as pTRT-relevant cells (Fig. 2D). The main components of pTRT-relevant cells belong to CD8_Teff_GZMB (17.80%) and CD8_Tem_GZMK (12.22%), which are in line with pan-cancer analysis results [5] (Fig. 2E).

Next, we performed unsupervised clustering on pTRTs and pTRT-relevant cells and defined three clusters based on different gene signatures [25] (Fig. 2F). The stem-like C0 cluster exhibited a relatively high level of naïve markers like TCF7 and IL7R (Fig. 2G) which was mainly composed by CD8_Tem_GZMK (Supplementary Figure S2E-F). The transitory C1 cluster was characterized by acute-TCR-engagement-related genes (NR4A1, FOS, and JUN) [26] and effector makers. Compared to C1, C2 exhibited reduced expression of transitory genes and high expression of effector markers and was thus defined as the terminally differentiated (TD) cluster (Fig. 2G). Additionally, GZMK was highly expressed in C0, while GZMB was highly expressed in C1/C2, suggesting that differential expression of granule enzymes could reflect the developmental state of cells and serve as markers.

Therefore, our results indicated the existence of tumor-reactive T cells (pTRTs and pTRT-relevant cells) in the BM of AML patients carrying the RUNX1::RUNX1T1 which showed a continuous spectrum of differentiation from the stem-like cluster to the terminally differentiated cluster (Fig. 2H, Supplementary Figure S2G).

The tumor-reactive T cells showed distinct non-exhausted cytotoxic transcriptional features in AML

To address the molecular characteristics of tumor-reactive T cells in AML with RUNX1::RUNX1T1, we compared the transcriptional regulon (Fig. 3A) and gene expression profile (Fig. 3B, Supplementary Table S2) between tumor-reactive T cells and bystander T cells. The transcriptional programs regulated by TBX21, ZNF595, FOSL2, ZNF83, ZNF484, IKZF2, and ELK3 were found to be significantly activated in tumor-reactive T cells. The TBX21 regulon, whose target genes included ADGRG1, CST7, CX3CR1, and so on, was identified as a Temra-specific regulon in CD8+ T cells [27]. The FOSL2 was an AP1 family member, functioning downstream of TCR signaling and driving CD8+ effector differentiation [28]. Consistent with the regulon profile, tumor-reactive T cells exhibited higher expression of pro-inflammatory effector molecules such as IFNG, TNF, and PRF1 (Fig. 3C) and lower expression of genes like LBT, CD27, and CD28.

Fig. 3

Comparing the characteristics between tumor-reactive T cells and bystander T cells. A Volcano plot showing the specifically activated regulons in the tumor-reactive and bystander T cells. The upregulated target genes are shown in the black dashed box. The vertical dashed line represents a log2 (fold change) value of 0.5, and the horizontal dashed line represents an adjusted p-value of 0.05. B Scatter plot showing the differentially expressed genes in tumor-reactive and bystander T cells. The x-axis represents the log2 (fold change) value with a threshold value of 3, and the y-axis represents the difference in expression ratios of genes between tumor-reactive T cells and bystander T cells (threshold value: 0.3), with the color of the points indicating the -log10 (p-value). The upregulated genes are partly shown in the red dashed box. C The stacked violin plot showing the pro-inflammatory effector molecules and exhaustion-related molecules expression in tumor-reactive T cells and bystander T cells. D The UMAP plot illustrating the naïve/cytotoxicity/exhaustion score in CD8+ T cells. The color bar is shown on the right margin. E Proportions of TOX/PDCD1/KLRG/KLRK-positive cells in each T cell cluster. The pTRT-related Temra clusters are drawn in red, while the rest of the clusters are in blue. F Representative pathways enriched in the identified genes as determined by gene set enrichment analysis (GSEA) [57]. The normalized enrichment score (NES) and the false discovery rate (FDR) are annotated in the top right corner. G The bar plot showing the senescence-like/NK-related gene expression between bystander T cells (blue) and tumor-reactive T cells (red). The Wilcoxon rank sum test was performed for calculation. The p-values for each gene are all less than 0.0001

Interestingly, tumor-reactive T cells demonstrated high cytotoxicity scores and low exhaustion scores (Fig. 3D). The expression levels (Fig. 3C) and proportions (Fig. 3E) of exhaustion-related markers like PDCD1 and TOX were also low. This suggested that tumor-reactive T cells in AML displayed a unique non-exhausted but cytotoxic profile compared to solid tumors.

Next, we performed functional enrichment analysis on tumor-reactive T cells and found enrichment of the TCR downstream pathway validating the feature of tumor reactivity. Notably, the NK-mediated cytotoxicity pathway was enriched (Fig. 3F) and NK-related makers like KLRG1 and KLRK1 were significantly upregulated (Fig. 3B, Fig. 3E, Fig. 3G). These tumor-reactive cells did not express invariant Vα24-Jα18 chain and Vβ11 chain (Supplementary Figure S3), which excluded the possibility of invariant NKT (iNKT) [29].

Previous studies have indicated that Temra cells lacking CD27/CD28 expression and expressing NK-related markers such as KLRG1 exhibit a senescent-like but functionally active profile after antigen stimulation [30,31,32,33,34] and our results also confirmed such characteristics (Fig. 3F-G). Therefore, the tumor-reactive T cells in AML exhibited a specific non-exhausted cytotoxic Temra profile with upregulated NK-related molecules.

The ADGRG1 was highly and almost specifically expressed in tumor-reactive T cells

In solid tumors, the tumor-reactive T cells are characterized by high expression levels of CXCL13 or exhaustion signatures [35]. However, tumor-reactive T cells in AML did not exhibit significant expression of CXCL13 or classical exhaustion molecules (Fig. 4A). To explore the distinctive maker of tumor-reactive T cells in AML, we examined the average gene expression pattern of tumor-reactive T cells and bystander T cells (Fig. 4B). Genes such as ADGRG1, TBX21, S1PR5, FCRL6, and FCGR3A showed relatively selective expression in tumor-reactive T cells (the average expression in bystander T cells < 0.5, Supplementary Table S3). For the top 100 genes based on the delta value (Supplementary Methods), we calculated the proportion of tumor-reactive T cells among the cells expressing those genes (Fig. 4C, Supplementary Table S4). Particularly, we found that ADGRG1 had the highest proportion (74.90%) and was almost specifically expressed in tumor-reactive T cells (Fig. 4D-E). Additionally, ADGRG1+ cells exhibit reduced TCR clone diversity (Fig. 4F) and a significant increase in TCR clone size (Fig. 4G-H), indicating a state of clonal expansion.

Fig. 4

The tumor-reactive T cells expressed a high level of ADGRG1. A The feature plots showing the expression levels of tumor-reactive T cells’ marker genes in solid tumors. The color bars for each graph are placed on their respective right margin. B Scatter plot showing the average gene expression of tumor-reactive T cells versus bystander T cells. The horizontal axis represents the average expression level of genes in bystander T cells, while the vertical axis represents the average expression level in tumor-reactive T cells. The 45-degree red dashed line indicates where the average expression levels are equal in both cell types. Some gene names that are highly expressed in tumor-reactive T cells are labeled near their corresponding points. C The frequency of pTRTs in certain gene expressed cells. The ADGRG1 gene is shown in red dot. D The UMAP plot visualizing the ADGRG1 expression in CD8+ T cells. E The distribution pattern of ADGRG1+/−CD8+ T cells. (Left) matrix plot showing distribution prevalence estimated by Ro/e. (Right) violin plot showing the ADGRG1 expression among pTRT, pTRT-relevant, and pTRT-irrelevant groups. F Bar plot showing the normalized Shannon entropy of ADGRG1− and ADGRG1+ cells. G UMAP plot of the log-transformed clone size value in the RUNX1::RUNX1T1 group. The color bar is shown on the right side. H Bar plot showing the clone size of ADGRG1− and ADGRG1+ cells. I The functional module involved by ADGRG1 based on STRING [58]. The edge stroke color is scaled by combined score which was computed by combining the probabilities from the different evidence channels and corrected for the probability of randomly observing an interaction [59]. The gene group annotation is shown on the right margin. In Fig. 4E and H, the Wilcoxon rank-sum test was applied to calculate the p-value (**** ≤ 0.0001)

ADGRG1, also known as GPR56, is a member of the adhesion G protein-coupled receptor (GPCR) family. ADGRG1 is expressed in NK cells, gdT cells, and terminal effector T cells in healthy human BM (Supplementary Figure S4A-B) [22, 36]. In mouse, Adgrg1 is expressed in hematopoietic stem/progenitor cells (Supplementary Figure S4C-D) [37, 38]. Notably, ADGRG1 has also been found to serve as a marker for leukemia stem cells [39]. Recent studies have revealed that ADGRG1+CD8+ T cells, following allogeneic HSCT, function as allo-reactive cytotoxic T cells that are capable of recognizing the patients’ original AML blasts [40]. By performing network analysis, we demonstrated that the upregulated genes in the ADGRG1+ group like IFNG, S1PR5, and CD81 were involved in a functional module with ADGRG1 (Fig. 4I), indicating the relation between ADGRG1 and T cell effector function.

Thus, we observed an elevated expression of ADGRG1 specifically in T cells of AML, particularly in tumor-reactive T cells, suggesting its potential utility as a marker for tumor-reactive T cells.

The Adgrg1 was specifically expressed in tumor-reactive T cells in the Runx1::Runx1t1 mouse model

Due to the complexity of patient genetic backgrounds, we established the Runx1Runx1t1/+; Mx1-Cre mouse model by conditionally knocking in the Runx1::Runx1t1 gene (Supplementary methods) with a cleaner and more controlled background to further validate the previous results (Fig. 5A). We sorted mCherry− T cells indicating the absence of Runx1::Runx1t1 fusion gene expression from Runx1Runx1t1/+; Mx1-Cre mouse (the Runx1::Runxt1 group) and Runx1Runx1t1/+; w/o Mx1-Cre mouse (the control group), respectively. The freshly sorted cells were profiled by scRNA-seq and scV(D)J-seq from 10× Genomics. We obtained a total of 43,450 high-quality cells after quality control (21,410 from the control group and 22,040 from the Runx1::Runxt1 group) (Fig. 5B). Besides, to examine the T cell patterns, we performed scV(D)J-seq and identified paired productive α and β chains in 35,379 T cells.

Fig. 5

The T cell profile in the BM of Runx1Runx1t1/+; Mx1-Cre mouse. A Overview of Runx1Runx1t1/+; Mx1-Cre mouse model construction. The conditional inducible knock-in C57BL/6 murine model was established with an inducible Cre recombinase (Mx1-Cre) and heterozygous conditional knock-in of Runx1::Run1t1 in the hematopoietic system (the Runx1::Runx1t1 group), in which the mCherry+ cell indicated the Runx1::Runx1t1 fusion gene expression. The mouse with heterozygous conditional knock-in of Runx1::Run1t1 and without Mx1-Cre was categorized as the control group which is mCherry−. This figure is edited by BioRender.com. B UMAP visualization of mouse BM Cd8+ T cells. Red represents the Runx1::Runx1t1 group and blue represents the control group. C UMAP plot showing the identified 10 Cd8+ T cell clusters. The subgroups corresponding to each color in the UMAP plot are indicated in the right-side legend. D The scatter plot of the tumor enrichment index and clonal expansion index in each mouse Cd8+ T cell cluster. The proliferation index is shown in gradient color and the color bar is annotated on the right margin. E Heatmap showing the TCR downstream signal activation level of different mouse Cd8+ clusters. The abbreviation of TCR signaling pathways is the same as that in Fig. 2C. F UMAP visualization of clusters annotated by TCR repertoire in mouse Cd8+ T cells. Red represents pTRT, blue represents pTRT-relevant cells and the rest of the cells are annotated as pTRT-irrelevant cells in grey. G The violin plot showing the naïve score (left) and the cytotoxicity score (right) among pTRT, pTRT-relevant, and pTRT-irrelevant groups. H Scatter plot showing the average gene expression of tumor-reactive T cells versus bystander T cells in mouse Cd8+ T cells. I The frequency of pTRTs in certain gene expressed cells. The Adgrg1 gene is shown in red dot. J, The UMAP plot showing the Adgrg1 expression in mouse Cd8+ T cells. K The distribution pattern of Adgrg1+/−Cd8+ T cells. (Left) matrix plot showing distribution prevalence estimated by Ro/e. (Right) violin plot showing the Adgrg1 expression among pTRT, pTRT-relevant, and pTRT-irrelevant groups. In Fig. 5G and K, the Wilcoxon rank-sum test was applied to calculate the p-value between pTRT/pTRT-relevant group with pTRT-irrelevant group: (**** ≤ 0.0001)

Overall, unsupervised clustering identified 10 clusters of Cd8+ T cells (Fig. 5C) and 15 Cd4+ T cells (Supplementary Figure S5D), each with unique signature genes (Supplementary Figure S5A, Supplementary Figure S5E-F). Cd4+ T cells and Cd8+ T cells both exhibited a reduction in immune reserve cells like Tn and an increase in terminally differentiated T cells like cytotoxic T cells (Tcytox, Supplementary Figure S5B, Supplementary Figure S5G).

Following the steps described in patients’ data analysis, we identified pTRTs in the BM of Runx1Runx1t1/+; Mx1-Cre mouse, including the Cd8_Tcytox_Maf and Cd8_Tcytox_Tnfrsf9 cluster, which were specifically TCR activated and clonal expanded in the Runx1::Runx1t1 group (Fig. 5D-F). Tumor-reactive T cells in the Runx1Runx1t1/+; Mx1-Cre mouse model also exhibited reduced naïve score, high cytotoxic score, and high senescence score (Fig. 5G, Supplementary Figure S5C).

To explore the gene specifically expressed in tumor-reactive T cells, we compared the gene expression level between tumor-reactive T cells and bystander T cells (Fig. 5H, Supplementary Table S5). The Adgrg1, Pdcd1, Osgin1, Lag3, and Chn2 were predominantly expressed in tumor-reactive T cells. In particular, 89% of Adgrg1+CD8+ T cells were tumor-reactive T cells (Fig. 5I-K, Supplementary Table S6). Moreover, as all mice were under specific-pathogen-free (SPF) conditions without continuous virus antigen stimulation, the T cells with virus-specific TCR identified by trained prediction model (Supplementary methods) did not exhibit Adgrg1 expression (Supplementary Figure S6, Supplementary methods).

In summary, the results obtained from Runx1Runx1t1/+; Mx1-Cre mouse validated the specific role of ADGRG1 in tumor-reactive T cells of AML.

ADGRG1+CD8+T cells were characterized as tumor-reactive T cells in AML with RUNX1::RUNX1T1

Combining the results from both mouse and human samples mentioned above, we hypothesized that ADGRG1+CD8+ T cells in the BM of AML patients with RUNX1::RUNX1T1 represent tumor-antigen-experienced cytotoxic T cells. To confirm this hypothesis, we first applied anti-CD33 CAR-T and Molm13 as an artificial effector-target cell model (Fig. 6A). When only under CD3/CD28 Dynabeads stimulation, neither CAR-T nor vector T cells showed significant ADGRG1 expression. After coculturing with K562 (CD33−), the anti-CD33 CAR-T also lacked ADGRG1 expression. However, under CD33 antigen stimulation (Molm13), the ADGRG1 was significantly upregulated on anti-CD33 CAR-T cells (p = 0.016, Fig. 6B, Supplementary Figure S7A), suggesting only the TCR-antigen interaction induces the evident ADGRG1 upregulation. For ADGRG1+CD8+ anti-CD33 CAR-T after coculture with Molm13, bulk RNA-seq data (Supplementary Figure S8, Supplementary Table S7) showed that its TCR signaling pathway was activated and the cytotoxic pathway was enriched, indicating higher cell-killing ability compared with ADGRG1−CD8+ anti-CD33 CAR-T cocultured with Molm13. Like the results in scRNA-seq, ADGRG1+CD8+ anti-CD33 CAR-T also showed NK-related features and were enriched in the senescence pathway.

Fig. 6

The function analysis of ADGRG1+CD8+T cells. A The coculture study workflow. T cells isolated from healthy donors’ peripheral blood were activated with CD3/CD28 Dynabeads and then transduced with anti-CD33 CAR-T lentivirus. On day 7 of production, anti-CD33 CAR-T cells were cocultured with AML cell line MOLM13 (CD33+) or K562 (CD33−) cells. B Bar plot showing the ADGRG1 expression detected by flow cytometry in CD8+ T cells. Blue represents vector T cells and red represents anti-CD33 CAR-T cells. C Schematic visualizing the sorting process for bulk RNA-seq from 4 AML BM. The ADGRG1+CD8+ T cells and ADGRG1−CD8+ T cells were sorted for following analysis. D The bar plot showing the TCR diversity of ADGRG1− cells and ADGRG1+ cells reconstructed by TRUST4 [41]. E The volcano plot of the differentially expressed genes (DEGs) between ADGRG1+ cells (n = 3 after quality control) and ADGRG1− cells (n = 4). The blue dots represent downregulated genes (689 genes), the red dots represent upregulated genes (708 genes), and the gray dots represent genes with no significant change. The vertical dashed line represents a log2 (fold change) value of 2, and the horizontal dashed line represents an adjusted p-value of 0.05. F Representative pathways enriched in the DEGs as determined by GSEA. The NES value and FDR value are annotated in the top right corner. G Bar plot showing the enriched pathways of DEGs. Blue represents pathways enriched in DEGs upregulated in ADGRG1− cells and red represents pathways enriched in DEGs upregulated in ADGRG1+ cells. H The IFN-γ -releasing level of ADGRG1+CD8+ T cells and ADGRG1−CD8+ T cells from AML patients with RUNX1::RUNX1T1. The cells from the same patient are connected by lines. The IFN-γ was measured when T cells were cultured without blast cells on the left plot, and IFN-γ was measured 48 h after T cells cocultured with the corresponding patient’s BM CD34+ leukemia blast cells on the right plot. The p-value was calculated by paired t-test. I The ADGRG1+CD8+ T cell ratio calculated by deconvolution in remission patients and refractory/relapsed (R/R) patients so far. The p-value was performed by Student’s t-test. J Survival analysis of 37 AML patients with RUNX1::RUNX1T1. The p-value was determined by the log-rank test. K The cell killing assay workflow. Mononuclear cells were isolated from AML BM. ADGRG1+CD8+ T cells, ADGRG1−CD8+ T cells, and blast cells were sorted from each patient. Then ADGRG1+/−CD8+ T cells were cocultured with matched blasts with E: T = 1:1 for 24 h. L A representative flow cytometry plot of the cell killing assay result. The residual blasts are gated and the cell percentage is labeled in green. The X-axis represents CD3 (APC/Cy7) and the Y-axis represents CD123 (PE). M The residual blast cell count ratio between ADGRG1− and ADGRG1+ groups. The cells from the same patient are connected by lines. The p-value was calculated by paired t-test. p-value: ns > 0.05, * ≤ 0.05, ** < 0.01, *** < 0.001, **** < 0.0001

To validate the findings in AML patients, we sorted ADGRG1+/−CD8+ T cells from 4 AML with RUNX1::RUNX1T1 BM samples for bulk RNA-seq (Fig. 6C, Supplementary Figure S7B, Supplementary Table S8). After TCR reconstruction based on TRUST4 [41], we found that the TCR diversity of ADGRG1+CD8+ T cells was significantly lower than that of the ADGRG1− group, indicating a state of TCR clonal expansion (Fig. 6D). Compared with the ADGRG1− group, the ADGRG1+ group upregulated effector molecules like GZMB, NK-related genes like FCGR3A, Temra-related genes like CX3CR1, and ADGRG1 upstream regulatory gene ZNF683 [42] (Fig. 6E, Supplementary Table S9). Additionally, ADGRG1+CD8+ T cells exhibited antigen-response characteristics (Fig. 6F) and enriched pathways such as cytotoxicity and cell killing (Fig. 6G).

To prove that the upregulation of ADGRG1 in AML CD8+ T cells signifies tumor-reactive T cells, we isolated ADGRG1+CD8+ T cells and ADGRG1−CD8+ T cells from the BM samples of newly diagnosed AML patients with RUNX1::RUNX1T1 (Supplementary Table S8) and co-cultured them with the matched patient’s BM CD34+ leukemia blast cells (Fig. 6H, Supplementary Figure S7C). We observed that ADGRG1+CD8+ T cells secreted higher levels of IFN-γ in the absence of leukemia cells. Upon co-culture 48 h with leukemia blasts, ADGRG1+CD8+ T cells exhibited increased IFN-γ secretion, significantly surpassing the levels observed in ADGRG1−CD8+ T cells, suggesting the tumor reactivity of ADGRG1+CD8+ T cells.