Effector Vδ1 γδ T cells mediate anti-tumor immunity and are reprogrammed by imatinib in human gastrointestinal stromal tumor

Abstract

Gamma-delta (γδ) T cells are emerging effectors of anti-tumor immunity, but their relevance to gastrointestinal stromal tumor, the most common human sarcoma, is not defined. Here, we integrate single-cell transcriptomics, immunophenotypic and T cell receptor profiling, and functional assays of circulating and intratumoral γδ T cells across 68 human GIST specimens. Vδ1 cells predominated within tumors, and their abundance correlated with improved outcomes across 3 separate patient cohorts. γδ T cells from untreated tumors were cytotoxic ex vivo, but tumors that had become resistant to the tyrosine kinase inhibitor imatinib had lower Vδ1 composition, effector function, and altered clonality, and were enriched for apoptosis, IL-17 signaling, and checkpoint pathways. In a genetically engineered murine GIST model, PD-L1 blockade enhanced imatinib efficacy and restored tumor γδ T cell function. Thus, γδ T cells contribute to tumor immunity in GIST and are reprogrammed during imatinib resistance, making them an attractive immunotherapy target.

Introduction

Gastrointestinal stromal tumor (GIST) is the most common human sarcoma and usually harbors either a KIT or PDGFRA mutation¹. The tyrosine kinase inhibitor (TKI) imatinib mesylate improves survival but is rarely curative due to acquired drug resistance, which typically occurs within 2 years¹. Although we found that imatinib initially increased intratumoral CD8⁺ T cells in a genetically engineered Kit^V558del/+ mouse model of GIST^2,3, in human GIST CD4 and CD8A expression were not prognostic⁴ and conventional T cell checkpoint blockade was ineffective⁵; these findings prompted us to examine other immune effectors.

γδ T cells recognize a variety of antigens on tumor cells and unlike αβ T cells, exert cytotoxicity independently of HLA⁶. There are different subsets of human γδ T cells based on their γ and δ chains. Each TCRδ chain associates with one or more dominant TCRγ chains. The Vγ9Vδ2 subset predominates in human peripheral blood and can be readily expanded in vivo and in vitro. In contrast, human solid tissues are enriched for Vδ2-negative γδ T cells, primarily Vδ1 cells, which perform immune surveillance and possess cytotoxic potential⁶. A smaller Vδ3 subset, most often paired with Vγ2, has been described, although its functional role is unclear. Effects of γδ T cells in human solid tumors depend on tumor type and γδ T cell subset composition⁷. Anti-tumor effects are driven through interferon-γ (IFNγ) production and innate-like cytotoxicity; increased γδ T cell tumor infiltration has been associated with better clinical outcome in some malignancies^8,9. However, in other contexts, tumor-infiltrating γδ T cells may acquire inhibitory phenotypes associated with impaired effector function and reduced IFNγ secretion¹⁰. IL17A-producing γδ T cells have variable effects on tumors, as they are anti-tumoral in colorectal cancer¹¹ but mediate tumor resistance in breast cancer¹². In murine models, single-cell RNA sequencing and γδ TCR profiling have demonstrated that γδ cell function is strongly influenced by TCR usage¹³. Individual γδ TCRs have an intrinsic ability to participate in both innate and adaptive immune features driven by CDR3δ length and composition¹⁴.

Here, we establish a γδ T cell atlas in human GIST by integrating single-cell transcriptomics, immunophenotyping, TCR profiling, and functional assays of circulating and tumor-infiltrating γδ T cells across 68 tumor samples. Our findings establish γδ T cells as a core component of GIST immunosurveillance and provide a framework for γδ T cell-directed immunotherapy in GIST.

Results

Single-cell transcriptomics reveal γδ T cell heterogeneity.

To dissect the role of γδ T cells in human GIST, we sorted CD45⁺CD3⁺TCRγδ⁺ cells from untreated human GISTs and matched peripheral blood mononuclear cells (PBMCs) to profile them using single-cell RNA sequencing (scRNA-seq) (Fig. S1, S2A-B). After quality control, we obtained 55,228 cells from 4 blood and 2 tumor samples. We initially defined Vδ1, Vδ2, and Vδ3 T cell clusters based on canonical marker (TRDV1, TRDV2, and TRDV3) expression (Fig. 1A). Vδ2 cells were more prevalent in PBMCs, while Vδ1 and Vδ3 subsets were more frequent in tumors, consistent with reports that Vδ2-negative cells predominate within solid tumors^15,16. Flow cytometry in 13 patients revealed that the percentages of Vδ1 T cells correlated between PBMCs and tumors (Fig. S3).

Fig. 1: scRNA-seq analysis of γδ T cells in PBMCs and tumors from patients with GIST.

γδ T cells were sorted from PBMCs and tumor specimens and processed for 10x Genomics scRNA-seq. A) left, UMAP of γδ T cells annotated by broad class (Vδ1, Vδ2, Vδ3) defined by canonical marker expression. B) UMAP of γδ T cells annotated by cluster (55,228 cells across 4 PBMC and 2 tumor samples). C) top, Bar plots showing the distribution of each cell cluster across PBMC and tumor samples and bottom, across the 6 samples. D) Dot plot of selected differentially expressed genes in each cluster. Genes relevant to Vδ1, Vδ2, and Vδ3 subsets are indicated by color. Dot size indicates the percentage of cells in the cluster expressing the gene, and color indicates Z-score-normalized average expression between clusters. E) Differentially expressed genes between PBMCs and tumor samples shown as left a volcano plot and right gene set enrichment (GSEA) analysis of differentially expressed pathways. Representative significant pathways (FDR<0.1) are shown. No pathways were significantly upregulated in PBMC samples. F) top, Volcano plot of differentially expressed genes between Vδ1 and Vδ2 cells within tumor samples. bottom, GSEA analysis of differentially expressed pathways between tumor Vδ1 and Vδ2 cells. Representative significant pathways (FDR<0.1) are shown. E-F) Hallmark (H) and KEGG (K) pathways are indicated in parentheses. FDR, false discovery rate. NES, normalized enrichment score. G) CytoTRACE stemness score across γδ T cells shown as left UMAP and right violin plots separated by Vδ1, Vδ2, and Vδ3 subsets in PBMCs and tumor samples.

Unbiased clustering of our integrated tumor/PBMC γδ T cell atlas demonstrated 7 clusters for Vδ1, 4 for Vδ2, and 1 for Vδ3 based on differential gene expression (Fig. 1B–C, S2C-D, Table S1). The Vδ1-Cytotoxic T lymphocyte (CTL) cluster highly expressed classic markers for effector function (Fig. 1D), such as cytotoxicity (GZMH, NKG7, KLRD1, KLRK1, GNLY, KLRF1) and other effector and exhaustion markers (FCGR3A, FCRL6, NCR1 (NKp46), CD244, CD8A, KLRC2, IFNG, and TIGIT). There was also high expression of TBX21, ZEB2, and EOMES, the master transcriptional regulators of conventional T cell effector function. In contrast, the Vδ1-naive cluster possessed high expression of innate markers (TCF7, LEF1, LTB, XCL1, CCR7, MAL) as well as the transcription factor ZNF683 encoding PRDM1 (BLIMP1) that regulates T cell differentiation and lineage. The Vδ1-Inflammatory cluster had the highest expression of KLRK1 (NKG2D), IL2RB, and IKZF2 (Helios) and increased expression of genes involved in T cell-mediated inflammation and activation (AOAH, PRKCH, ZEB2, PIK3AP1, BCL2, RUNX1). The Vδ3 subset most closely resembled the Vδ1-CTL cluster and had the highest expression of killer-cell immunoglobulin-like receptors (KIRs) and GNLY, implying cytotoxicity based on HLA class I expression.

Of the 4 Vδ2 T cell clusters, Vδ2-CTL predominated and was enriched for hallmark Vδ2 markers (CEBPD, KLRB1, GZMK, KLRC1, IL7R, and CD23R)¹⁷. Compared to Vδ1-CTL, Vδ2-CTL had lower expression of cytotoxic molecules, exhaustion markers, and IFNG. Another Vδ2 cluster highly expressed CD44, CD69, ZBTB16 (encoding PLZF), RORA, RUNX1, CAMK4, a calmodulin-dependent protein kinase (CaMK), and BTN3A1, an essential membrane protein for TCR–dependent activation of human Vγ9Vδ2 T cells¹⁸. This Vδ2 cluster was also highly enriched for other unique markers (e.g., ANXA1, IL17RA, IL23R, IFNGR1, CXCR6, AREG, THEMIS) and was annotated as Vδ2-type 3 immunity, based on a prior report¹⁹. Smaller populations of Vδ2 T cells (Vδ2-Naïve and Vδ2-Differentiation) shared some common gene markers with Vδ2-CTL, but showed higher expression of SELL and CCL4, respectively.

Tumor γδ T cells had unique gene expression programs compared to PBMC γδ T cells (Fig. 1E, left), including heat shock proteins (HSPA1B and HSPA1A), stress response genes (DNAJB1), the integrin ITGA1, and other differentiation-related genes (e.g., ZNF683, XCL1, DUSP4, and CXCR3). Gene set enrichment analysis (GSEA) revealed multiple signaling pathways in tumor-infiltrating versus circulating γδ T cells, such as cytokine-receptor interactions, inflammatory response, IL2/STAT5 signaling, and IFNγ (Fig. 1E, right). Compared to tumor Vδ2 cells, tumor Vδ1 cells upregulated most Vδ1 signature genes (e.g. TRDV1, TRDV3, TRGV8, KLRD1, FCGR3A, KLRC2, KLRF1, IKZF2, FCRL6, XCL1, NCR1, ITGAD) and other markers of cytotoxicity and exhaustion. Similarly, tumor Vδ2 cells showed increased expression of Vδ2 signature genes (TRDV2, TRGV9, IL7R, THEMIS, ANXA1, RORA, and RUNX2; Fig. 1F, top). GSEA demonstrated that tumor Vδ1 cells were enriched for cell cycle pathways, oxidative phosphorylation, antigen processing, and NK cell-mediated cytotoxicity (Fig. 1F, bottom), the latter two of which were also enriched in circulating Vδ1 cells (Fig. S2E). Lastly, using CytoTRACE to infer relative differentiation states among γδ T cell populations²⁰, tumor-infiltrating γδ T cells were more differentiated than circulating γδ T cells (Fig. 1G). Taken together, transcriptomes of purified PBMC and tumor γδ T cells of GIST patients indicated considerable heterogeneity and broad functional plasticity.

Regulatory network analysis identifies Vδ1 γδ T cell transcriptional programs.

To investigate the transcription factors driving γδ T subsets, we used single-cell regulatory network interference and clustering (SCENIC) to identify potential regulatory mechanisms²¹. Regulons mediating lymphocyte trafficking (e.g., RUNX3, KLF2, PRDM1, CEBPD) were highly activated in circulating γδ T cells across all clusters (Fig. 2A–B). Predicted CEBPD target genes included both Vδ1 and Vδ2 signature genes (e.g., ZEB2, ZBTB16, ITGA1, THEMIS, RORA, IL7R), which mediate T cell differentiation and activation (Fig. 2C). In addition, regulons of EOMES, TBX21 (Tbet), and RORC, which regulate conventional T cell (T-conv) activation, were also higher in PBMC γδ T cells (Fig. 2A). On further analysis, RORC was mainly detectable in Vδ2 (Fig. 2B–C), but not Vδ1 cells. Conversely, EOMES and TBX21 were primarily in the Vδ1 subset, especially in the Vδ1-CTL cluster (Fig. 2C), suggesting they functionally resembled CD8⁺ αβ T cells. Regulons linked to inflammation or T cell exhaustion induced by stress antigens (e.g., STAT1, NFAT5, IRF8) were enriched across all tumor-infiltrating γδ T cells, suggesting presence of IFNγ signaling, inflammation, and exhausted T cell differenation^22–24.

Fig. 2: Regulon activity and potential gene-regulatory networks based on SCENIC of PBMC and tumor γδ T cells.

A) Differential transcriptional regulons as inferred by SCENIC between PBMCs and tumor γδ T cells. B) UMAP visualization of γδ T cells colored by regulon activity scores calculated using AUCell across the dataset. C) Heatmap of selected regulon activity across γδ T cells subsets. Regulon activity is Z-score normalized by row. D) Violin plots of selected regulon activities across PBMC and tumor γδ T cells stratified by broad class (Vδ1, Vδ2, Vδ3). Significance between tumor Vδ1 and tumor Vδ2 cells was assessed using the Wilcoxon rank-sum test. E) Gene module scores of broad γδ T cell classes within PBMCs and tumors as in D. F) Enrichment analysis using Gene Ontology pathways was conducted on the predicted target genes of each regulon. Selected significant pathways are shown.

Tumor Vδ1 and Vδ2 T cells had distinct regulon activity (Fig. 2D, S4). EOMES, TBX21, and POLR3G were enriched in Vδ1 cells, particularly in the Vδ1-CTL cluster (Fig. 2C). NFAT5 and IRF6 were specifically increased in the Vδ1-Inflammatory cluster. In contrast, RORC was more activated in the tumor Vδ2 subset. RORA and RUNX2 primarily regulated the Vδ2-Type 3 immunity phenotype (Fig. 2C). To identify potential functional differences mediated by these regulons, we computed immune module scores based on selected gene markers for memory, exhaustion, proliferation, activation, and cytotoxicity (Table S2)^25–27. Tumor Vδ1 cells had higher exhaustion, activation, and cytotoxicity scores, whereas tumor Vδ2 cells had higher memory scores (Fig. 2E). Lastly, enrichment analysis using Gene Ontology was conducted based on predicted target genes of each regulon (Fig. 2F). Consistent with their transcriptional profiling, targets of EOMES and POLR3G were enriched for T cell activation, cytotoxicity, and lymphocyte chemotaxis. Meanwhile, RORC and RUNX2 were associated with pathways in T cell differentiation, activation, and leukocyte adhesion. In summary, we defined diverse functional phenotypes in γδ T cells and inferred their regulatory mechanisms.

γδ TCR analysis uncovers clonotype diversity and TKI alteration.

To examine the γδ TCR repertoire, we performed next-generation sequencing of genomic DNA for TCR delta chains in sorted γδ T cells from 5 pair-matched PBMC and tumor samples (3 untreated and 2 with acquired resistance to imatinib) and 1 additional resistant tumor. Most (>75%) of the PBMC and tumor γδ TCRs were minimally expanded with fewer than 5 clones (Fig. 3A). Patients with resistant tumors had a higher frequency of highly expanded clones (clone size >10), particularly in the blood (Fig. 3B).

Fig. 3: Compositional and clonal analysis of γδ T cells in PBMCs and tumors.

γδ T cells were sorted from human PBMC and GISTs. gDNA was processed for TCRαδ CDR3 next-generation sequencing. A) Clonotype frequency in all specimens (n=11). B) Frequency of highly expanded (>10) clonotypes in 5 matched PBMC and tumors. C) Frequency of TCRDV gene segment usage in matched untreated (UT) and resistant (Res) specimens. D) Tree plots illustrating the composition of TCR repertoires in matched specimens. Each square represents a clone, with its area proportional to the clone size and its color indicating the V gene usage. E) Scatter plot of cell count distribution of a clone in PBMC versus tumor. Each point is a clone and the x,y axes denote the log transformed cell count in PBMC and tumor. Clones which have significantly high expansion are highlighted with FDR controlled q value. F) The alluvium plot for the VDJ combination of clones which have significantly high expansion in untreated and resistant tumors. G) The frequency change between the untreated and resistant patient groups for each VDJ recombination in clones which are highly expanded in tumor. VDJ recombinations with significant change are highlighted with “***”. H) Frequency of Vδ1 and Vδ2 γδ T cells determined by flow cytometry in PBMCs and tumors. I) left, Frequency of Vδ1 and Vδ2 γδ T cells in tumors or right, PBMCs based on treatment status in untreated (UT; n=15), and resistant (Res; n=10) specimens. Medians indicated. For all panels, * p<0.05, ** p<0.01, *** p<0.001; **** p<0.0001.

In 4 of 5 patients, TCRDV1 usage was higher in tumors than paired PBMCs (Fig. 3C), consistent with our scRNA-seq data. Notably, TCRDV1 usage was lower in the 2 tumors that developed resistance to imatinib compared to the 3 untreated tumors, while TCRDV2 gene usage was higher in the 2 resistant tumors. The tree plot summarizes TCRDV usage. TCRDV1 usage was higher in tumors than PBMCs in the untreated samples, both overall and in highly expanded clones. Furthermore, it was higher in untreated versus resistant patients in the tumor compartment by both metrics (Fig. 3D). We next performed differential expansion analysis for each clone between paired PBMC and tumor samples (Fig. 3E). In resistant tumors, there was a higher frequency of highly expanded clones compared to untreated tumors (2.78 vs. 0.89%, p=3.91×10⁻¹¹). The recombinations V1D1J1 and V2D1J1 predominated in both groups, suggesting that the primary variation lies in V gene usage (Fig. 3F). In the untreated group, V1D1J1 and V2D1J1 appeared at a similar frequency, whereas in the resistant group, V2D1J1 was enriched while V1D1J1 was reduced (Fig. 3G). To confirm our findings, we performed flow cytometry for Vδ1 and Vδ2 in human PBMCs and tumors in 29 patients. Consistent with our scRNA-seq and TCR sequencing in sorted γδ T cells, the Vδ1 subset dominated in the tumor, whereas the Vδ2 subset was more abundant in paired PBMCs (Fig. 3H). Furthermore, TKI-resistant GISTs had lower tumor Vδ1 and higher Vδ2 composition than untreated tumors (Fig. 3I, left), even though the relative proportion of the 2 subsets did not vary with treatment status in PBMCs (Fig. 3I, right). Overall, resistant tumors had an altered γδ TCR repertoire.

Tumor Vδ1 T cells are cytotoxic and prognostic of clinical outcome.

To validate our scRNA-seq atlas, which showed effector and exhaustion markers in the Vδ1 subset, we performed flow cytometry on 10 untreated GISTs. Tumor-infiltrating Vδ1 T cells expressed CD8A, FCGR3A (encoding CD16), PDCD1 (PD1), TIGIT, and EOMES (Fig. S5A), as identified in Fig. 1D. In addition, expression of CCR7, HLA-DR, and the cytotoxic degranulation marker CD107a was higher in tumor-infiltrating Vδ1 cells compared to Vδ2 cells (Fig. 4A, S5B) and PBMC Vδ1 cells (Fig. 4B), indicating enhanced tumor trafficking, activation, and cytotoxic potential. To assess the functional relevance of tumor-infiltrating Vδ1 cells ex vivo, we sorted γδ T cells from 9 additional human GISTs (7 untreated and 2 imatinib-resistant). Six tumors had a high proportion of Vδ1 cells among γδ T cells (mean 59.2%), while 3 had Vδ1 lower percentages (mean 32.6%) (Fig. 4C). Tumors with a high Vδ1 proportion had increased expression of TRGC2, TRDV3, IFNG and KLRK1 (encoding NKG2D) (Fig. 4D), thereby validating our scRNA-seq findings shown in Figure 1D.

Fig. 4: Tumor-infiltrating Vδ1 T cells exhibit enhanced cytotoxicity and are prognostic of improved clinical outcomes.

A) Comparison of selected marker expression by flow cytometry between tumor-infiltrating Vδ1 and Vδ2 T cells. Data represent mean±SEM. B) Summary of selected markers in PBMC-derived (n=10) and tumor-derived (n=8–16) Vδ1 T cells. Data represent mean±SEM. C) Tumor γδ T cells were sorted and grouped by relatively high Vδ1 (mean 59.2%, n=6) or low Vδ1 frequency (mean 32.6%, n=3) using a cut-off of 50% Vδ1 frequency. D) qPCR showing mRNA expression of TRGC2, TRDV3, IFNG and KLRK1 in sorted tumor γδ T cells with high or low Vδ1 proportions. Data represent mean±SEM. E) Composition of γδ T cell subsets (n = 6,970 cells) in scRNA-seq data from untreated (n=3) and resistant (n=2) GISTs based on sorted γδ T cell reference (Fig. 1A). F) left, Volcano plot of differentially expressed genes between γδ T cells in untreated (n=3) and resistant (n=2) GISTs based on scRNA-seq data. Key Vδ1 and Vδ2 marker genes (Fig. 1F) are highlighted. right, Corresponding Dot plot of Vδ1 and Vδ2 marker genes. G) Over representation analysis using Gene Ontology (GO), KEGG (K), and Reactome (R) gene sets in the top differentially expressed genes between untreated (top) and resistant γδ T cells (bottom). Representative significant pathways are shown (FDR <0.05). H) qPCR analysis showing relative mRNA expression of TRGC2, TRDV3 and IFNG in untreated (n=15) and resistant (n=14) GISTs. Medians indicated. I) Flow cytometry analysis of CD107a expression on tumor Vδ1 γδ T cells from untreated (UT; n=8) and resistant GISTs (Res; n=5). Data represent mean ±SEM. J) Summary data of effector status of tumor Vδ1 T cells from untreated (UT; n=9) and imatinib-resistant tumors based on flow cytometry (Res; n=3). Medians indicated. T_EM - effector memory T cells, T_EMRA, - terminally differentiated effector memory T cells, T_CM - central memory T cells, T_N – naïve T cells. K) Kaplan-Meier overall survival analysis of a published Japanese cohort of 32 patients with untreated primary GISTs, stratified by median expression score of γδ T cell signature genes (TRGC2, TRDC, TRDV3, and TRGV2). L) Scatterplot of predicted 5-year recurrence-free survival (RFS) based on median expression score of γδ T cell signature genes in our published cohort of bulk RNA-seq in 30 untreated primary GISTs. M) Scatterplot of predicted 5-year RFS in 17 patients with untreated primary GIST based on Vδ1 percentage of γδ T cells determined by flow cytometry. For all panels, *p<0.05; **p<0.01; ***p<0.001, **** p<0.0001.

The TCR data of sorted γδ T cells suggested that resistant GISTs had altered clonotype diversity and subset composition. Therefore, we performed additional scRNA-seq analysis on 3 untreated and 2 resistant GISTs, without sorting γδ T cells. CD3⁺ T cells were identified and clustered across all samples (Fig. S6A). Two clusters were annotated as γδ T cells based on canonical gene expression (Fig. S6B). Using reference mapping from our γδ T cell scRNA-seq atlas (Fig. 1A), we identified a substantial shift in γδ T cell subset composition, from Vδ1 dominance in untreated tumors to a more balanced Vδ2 proportion in resistant tumors (Fig. 4E). Differential gene expression analysis revealed enrichment of tumor-specific Vδ1 signature genes in untreated tumor γδ T cells and Vδ2 genes in resistant γδ T cells (Fig. 1F, 4F, left). Notably, higher expression of cytotoxic effector genes, including GNLY, PRF1, KLRF1, KLRD1, and FCGR3A, was observed in untreated tumors, indicating enhanced cytotoxic activity (Fig. 4F, right). Over-representation enrichment analysis (ORA) confirmed that untreated tumor γδ T cells were significantly enriched for immune effector processes, leukocyte-mediated immunity, cell killing, and NK cell-mediated cytotoxicity. In contrast, γδ T cells in resistant tumors had increased apoptosis, T cell receptor signaling, and lipid metabolism, as well as PD-1/PD-L1 checkpoint, NF-kappa B, and IL-17 signaling pathways (Fig. 4G).

To further validate these findings, we analyzed RNA from 29 human GISTs (15 untreated and 14 resistant tumors) by quantitative real-time PCR (qPCR). Expression of TRGC2 and TRDV3, reflecting the abundance of Vδ1 and Vδ3 cells, respectively, trended lower in imatinib-resistant tumors than in untreated GISTs (Fig. 4H), corroborating our flow cytometry and scRNA-seq data showing reduced Vδ1 cells in resistant GISTs, with concordant IFNG reduction. CD107a expression was decreased in tumor-infiltrating Vδ1 cells from resistant GISTs (Fig. 4I), indicating reduced effector function. Vδ1 T cells in untreated tumors predominantly displayed an effector phenotype, with high frequencies of T cell effector memory (T_EM) populations, in contrast to resistant tumors (Fig. 4J).

To determine the clinical relevance of tumor Vδ1 T cells in GIST, we analyzed 3 independent cohorts of untreated patients. In a bulk tumor RNA microarray of 32 untreated primary human GISTs from Japan²⁸, we applied a published γδ T cell signature (TRGC2, TRDC, TRDV3, and TRGV2)²⁹ and stratified high versus low signature expression based on the median. High signature expression was associated with improved overall survival (Fig. 4K). Next, we evaluated this γδ T cell signature in our published cohort of 30 untreated primary GISTs³⁰ and found that it correlated with predicted recurrence-free survival (RFS)³¹ (Fig. 4L). Finally, we quantified Vδ1 T cell abundance by flow cytometry in 17 untreated primary GIST tumors and again found a correlation to predicted RFS (Fig. 4M). Thus, γδ T cells, and Vδ1 T cells in particular, are favorably prognostic in untreated human GIST.

γδ T cells are cytotoxic in patient-derived tumor models.

To assess whether γδ T cells kill GIST tumors, we first expanded PBMC-derived γδ T cells from healthy donors (n=4) with an anti-γδ TCR antibody and IL-2 (Fig. S7A)³². Healthy donors with a high frequency of Vδ1 cells after 8 days of expansion (Fig. S7B) were used for in vitro assays (n=2). Expanded γδ T cells demonstrated dose-dependent cytotoxicity against luciferase-expressing human GIST T1 cells (Fig. 5A). Fluorescently labeled γδ T cells migrated into GIST spheroids and accomplished tumor cell killing, albeit at a higher E:T ratio (Fig. 5B–C). To evaluate γδ T cell function in a more physiologic setting, we utilized an ex vivo human GIST tumor slice model. Primary tumor slices from 3 untreated patients were co-cultured with expanded γδ T cells from healthy donors. After 24h, tissue slices were dissociated into single-cell suspensions and viable KIT⁺ tumor cells were quantified by flow cytometry. Expanded γδ T cells infiltrated tumor slices (Fig. S7C) and reduced KIT⁺ tumor cells (Fig. 5D).

Fig. 5: Expanded circulating and tumor-infiltrating γδ T cells exhibit cytotoxicity against GIST.

A). Expanded γδ T cells from healthy donors were co-cultured with luciferase-expressing GIST-T1 cells at the indicated effector-target (E:T) ratios. After 24h, cytotoxicity was assessed by measuring luciferase activity. B) Representative images of a tumor spheroid assay. GIST-T1 cell spheroids were co-cultured with expanded γδ T cells, E:T ratio of 2.5:1, for 24h. Tumor cells expressed luciferase (green), and γδ T cells were labeled with CellTrace (red). C) Cytotoxicity in spheroids after co-culture with γδ T cells for 72h at the indicated E:T ratios as measured by luciferase assay. D) left, Histograms of KIT expression on tumor cells and right, quantification of KIT⁺ tumor cell killing in human GIST tissue slices co-cultured with expanded γδ T cells at an E:T ratio of 5:1 for 24h, assessed by flow cytometry after tissue slice dissociation. E) Representative pictures showing expansion of tumor-derived γδ T cells on day 6. Tumor-infiltrating lymphocytes (TILs) from untreated GISTs were isolated using CD45 microbeads and expanded with anti-γδ TCR antibody supplemented with IL-2 (200IU/ml). F) Representative flow plots showing robust expansion of Vδ1 cells at 24h following co-culture of expanded tumor-derived γδ T cells with tumor single-cell suspensions (tumor SC). G) Cytotoxicity against autologous KIT⁺ tumor cells was evaluated by co-culturing expanded tumor-derived γδ T cells as shown in F. left, Flow cytometry histograms of KIT expression on tumor cells and right, quantification of KIT+ tumor cell killing. H) left, Flow cytometry histograms of MICA expression on tumor cells (KIT⁺) and T cells (CD45⁺CD3⁺) and right, qPCR showing relative MICA mRNA expression in KIT⁺ tumor cells and KIT⁻ cells from GIST patients. I) Tumor-infiltrating lymphocytes (TILs) were treated with recombinant MICA (rhMICA, 10 μg/ml) and IL-2 for 72h and analyzed by flow cytometry, demonstrating an increased percentage of Vδ1 γδ T cells, and greater frequency of HLA-DR among the same. J) Sorted tumor γδ T cells from untreated GISTs were co-cultured with rhMICA or anti-NKG2D antibody for 24h, and IFNγ production was measured by ELISplot. Data represent mean±SEM, For all panels, *p<0.05; **p<0.01; ***p<0.001, **** p<0.0001.

Although tumor-infiltrating γδ T cells are thought to possess limited anti-tumor activity in certain cancers due to their exhausted phenotypes⁶, our scRNA-seq data revealed enrichment of a NK cell-mediated cytotoxicity pathway in tumor Vδ1 cells. Therefore, we expanded tumor-infiltrating lymphocytes (TILs) from 2 untreated GIST patients for 6 days using an anti-γδ TCR antibody and IL-2, and cultured them with matched tumor single-cell suspensions for 24h (Fig. 5E–F). Co-culture reduced the number of viable KIT⁺ tumor cells (Fig. 5G). Given that tumor-associated ligands can activate γδ T cells, we examined the role of MICA, an NKG2D ligand expressed by KIT⁺ tumor cells in human GIST (Fig. 5H). Plate-bound recombinant MICA plus IL-2 expanded Vδ1 cells within TILs and induced higher expression of HLA-DR (Fig. 5I). MICA stimulation also induced IFNγ production, which was attenuated by NKG2D blockade (Fig. 5J). Thus, both circulating and tumor-infiltrating γδ T cells, particularly the Vδ1 subset, retain cytotoxic effector function following expansion and activation.

Initial imatinib response alters tumor γδ T cell composition and function in mice.

Our human studies utilized untreated tumors and tumors that acquired resistance to imatinib. To investigate the early effect of imatinib on tumor γδ T cells, we administered imatinib for 1 week to Kit^V558del/+ mice³, which develop an intestinal GIST. Bulk RNA-seq of sorted tumor γδ T cells showed that imatinib upregulated 414 genes and downregulated 577 (Fig. 6A). GSEA revealed increased oxidative phosphorylation, fatty acid metabolism, and glycolysis/gluconeogenesis pathways. Murine IL17A-secreting γδ T cells rely on oxidative lipid metabolism³³, and previously we found that imatinib expanded mouse IL17A-producing γδ T cells⁴. Here, we further identified that imatinib reduced γδ T cell IFNγ response, antigen presentation, and NK cell-mediated cytotoxicity (Fig. 6B–C). IFNγ–producing murine γδ T cells, particularly the Vγ7/Vγ1 subset, display antitumor activity in various models, whereas IL-17A–producing γδ T cells, mainly the long-lived Vγ6 subsets, have variable pro-tumoral function¹³. In our model, by bulk gene expression and our published murine scRNA-seq data⁴, Vγ6 cells (cluster 1) predominated and were expanded by imatinib, while the Vγ7/Vγ1 subset (cluster 4) was reduced (Fig. 6D–F, S8). Accordingly, imatinib upregulated γδ-Th17 signature genes and reduced γδ-Th1 related and cytotoxicity genes (e.g., Gzmb, Klrd1, Klrc2), respectively. Bulk RNA-seq also showed that imatinib increased expression of Pdcd1 (PD1) and Tnfrsf9 (4–1BB ligand receptor), which were validated by qPCR (Fig. 6G).

Fig. 6: Imatinib drives functional and compositional reprogramming of tumor γδ T cells in vivo.

γδ T cells were sorted from Kit^V558del/+ mice treated with vehicle or imatinib for 1 week and processed for bulk RNA-seq. A) Volcano plot of differentially expressed genes in tumor γδ T cells based on treatment (n=3/group). Significant genes defined with p-adjusted < 0.05 and log₂FC > 1. B) Gene set enrichment analysis (GSEA) of Hallmark (H) and KEGG (K) pathways in tumor γδ T cells based on treatment. C) Selected enrichment plots from B. NES, normalized enrichment score; FDR, false discovery rate. D) Heatmap of selected tumor γδ T cell signature genes across samples. Color represents Z-score normalized expression for each gene. E) UMAP of previously published scRNA-seq data of murine γδ T cells (1,641) cells, separated by treatment and colored by Tcrgv6 gene expression. F) Deconvolution of bulk RNA-seq data with the single-cell reference in E using CIBERSORTx. Significance was assessed using a Welch’s 2-sided t-test. G) qPCR of expression of selected genes from sorted γδ T cells. n=4–6/group. H) Tumor weight after Kit^V558del/+ mice were treated for 1 week with vehicle, imatinib, or imatinib plus anti-PDL1. n=9/group. I) The frequency of Vγ7 and Vγ6 high subsets among tumor γδ T cells from Kit^V558del/+ mice treated in H as assessed by flow cytometry. J) The frequency of PD1⁺ cells in Vγ7 and Vγ6 high subsets among tumor γδ T cells in each group. K-L) The frequency of IFNγ⁺ cells (K) and IL17A⁺ cells (L) among tumor γδ T cells from Kit^V558del/+ mice treated in H. Data represent mean±SEM. n=4/group. For all panels, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

Since checkpoint inhibition of PD-1 and PD-L1 can augment γδ T cell cytotoxicity in other solid tumors^16,34, we hypothesized that targeting the PD1/PDL1 pathway would affect γδ T cell immunity in GIST. The combination of imatinib and an anti-PD-L1 blocking antibody further reduced tumor weight as expected³⁵, but also restored the proportion of Vγ7⁺ cells, downregulated their PD1 expression, and increased γδ T cell IFNγ while maintaining IL-17 production (Fig. 6H–L). Taken together, these data suggest that checkpoint inhibition increased the anti-tumor efficacy of γδ T cells during imatinib response.

Discussion

By integrating single-cell transcriptomic, TCR repertoire profiling, and functional analyses, we established a comprehensive framework for understanding tumor and circulating γδ T cell heterogeneity, differentiation, clonotype diversity, and function in GIST, revealing previously unappreciated mechanisms linking oncogenic signaling, innate-like cytotoxic immunity, and therapeutic response. Vδ1 cells were more prevalent in human tumor samples than in matched blood, regulated by EOMES and TBX21, and enriched for a T_EM phenotype. Vδ1 cells expanded in vitro from healthy donor PBMCs killed tumor cells in 3 different patient-derived models, including freshly procured GIST tumor slices. TILs from GIST specimens expanded with an anti-TCR antibody were also cytotoxic against autologous tumor cells. The baseline γδ T frequency in PBMCs influence ex vivo expansion capacity³⁶. In contrast, Vδ2 cells were driven by RORC, predominated in tumors that had acquired resistance to imatinib, and expressed fewer effector molecules. Analysis of 3 independent patient cohorts demonstrated the prognostic value of total, as well as Vδ1-specific, γδ T cell abundance in untreated GISTs, as observed in other tumors.²⁹ Notably, including Vδ1- and Vδ3-specific genes (TRGC2 and TRDV3, respectively) improved upon the prognostication of TRDC alone, which we used previously⁴.

The TCR repertoire of sorted γδ T cells provides insight into the mechanisms underlying γδ T cell activation in GIST. Untreated GIST tumors displayed broad Vδ1 TCR diversity with relatively few highly expanded clones, implying limited engagement of a few dominant immunogenic tumor antigens. Known γδ TCR ligands, including Ephrin type-A receptor 2 (EphA2) and HLA/MHC class I chain-related proteins A and B (MICA/B), can trigger γδ T cell differentiation and expansion within tumors^37,38. Indeed, MICA expression by human GIST tumor cells expanded the Vδ1 subset more efficiently than the Vδ2 subset and induced IFNγ production and HLA-DR upregulation. In contrast, the Vδ2 subset predominantly recognizes phosphoantigens (e.g., zoledronate) via the BTN2A1-BTN3A1 complex for expansion³⁹. Admittedly, our TCR analysis focused on the δ chain, potentially underestimating clonotypic complexity. Nevertheless, these observations support the model that GIST contains multiple γδ immune circuits, with Vδ1 cells preferentially linked to tumor stress recognition and direct cytotoxic effector function.

Imatinib therapy remodeled the γδ T cell compartment in GISTs. While the Vδ1 subset was prevalent in untreated human GISTs, tumors that had acquired resistance to imatinib exhibited reduced Vδ1 representation, diminished T_EM differentiation, increased apoptosis, reduced cytotoxicity, altered cellular metabolism, and increased IL-17 signaling. These changes reinforce that while imatinib directly targets KIT signaling in tumor cells, the GIST tumor microenvironment is also remodeled, as we have shown previously with alterations in intratumoral T cells, tumor-associated macrophages, and dendritic cells^2,40,41.

Our murine Kit^V558del/+ model allowed us to investigate γδ T cells during the early (i.e., 1 week) tumor response to imatinib. Previously, we showed that γδ T cells in GIST contribute to tumor control through interleukin-17 (IL-17) production, as antibody-mediated depletion of either IL-17 or γδ T cells accelerated tumor growth in otherwise untreated mice⁴. The role of IL-17 in human GIST remains uncertain. Here, we found that the early response to imatinib in Kit^V558del/+ mice upregulated tumor γδ T cell fatty acid metabolism and oxidative phosphorylation, as detected by bulk RNA-seq analysis of sorted γδ T cells. Furthermore, imatinib decreased tumor γδ T cell IFNγ in addition to increasing IL-17 production. Combining PD-L1 blockade with imatinib increased tumor cytotoxicity, tumor Vγ7 cells, and tumor γδ T cell IFNγ production. These findings are consistent with Vδ1 cells responding favorably to checkpoint inhibition in melanoma¹⁶ and Merkel cell carcinoma³⁴. Human GIST tumor samples during early response to imatinib are not clinically available. Moreover, technical challenges precluded the isolation and analysis of sufficient γδ T cells after chronic therapy with imatinib, which is typically administered for at least 6 months prior to surgery to reduce tumor size.

Our study establishes Vδ1 γδ T cells as critical mediators of anti-tumor immunity in GIST and highlights γδ T cells as an actionable immunotherapeutic axis in GIST and other solid tumors that are refractory to conventional αβ T cell-directed immunotherapy. Emerging clinical efforts to harness γδ T cells, including CAR-engineered Vδ1 therapies currently under investigation^42,43, further highlight the translational relevance of these immune cells.

Methods

Human GIST samples.

Tumor specimens and PBMCs were obtained from patients with GIST undergoing surgical resection under an Institutional Review Board-approved protocol (additional supporting data files). All patients provided written informed consent prior to surgery, and studies were conducted in accordance with the Declaration of Helsinki. Patient information was de-identified and encrypted in compliance with the Health Insurance Portability and Accountability Act regulations (HIPAA). Tumor tissues were sectioned and processed into single-cell suspensions. PBMCs were isolated from whole blood using CPT tubes and density gradient centrifugation.

Cell isolation, sorting, and flow cytometry.

Single-cell suspensions were prepared from PBMCs, human GISTs, and murine GISTs derived from Kit^V558del/+ mice. In brief, tumor tissues were minced and digested with a mixture of 2.5 mg/ml collagenase IV and 50 μg/ml DNase I at 37°C for 30 minutes. They were processed into single-cell suspensions as previously described 35. Flow cytometry was performed using an LSRFortessa (BD Biosciences). Prior to antibody staining, cells were incubated with an anti-CD16/32 antibody to block Fc receptors. The following fluorochrome-conjugated antibodies were used for murine cells: CD45 (30-F11), CD3 (17A2), CD4 (RM4–5), CD8 (53–6.7), NK1.1 (PK136), TCR γδ (GL3), IL17A (TC11–18H10.1), PD-1 (29F.1A12), IFN-γ (XMG1.2), TCR Vγ1 (2.11), Vγ2 (UC3–10A6), Vγ4 (49.2), Vγ6 (1C10–1F7), Vγ7 (F2.67). For human samples, the following antibodies were used: CD45 (2D1), CD3 (SK7), CD8 (RPA-TB), TCR γδ (B1), TCR Vδ1 (TS8.2), TCR Vδ2 (B6), TCR Vγ9 (B3), TIGIT (TgMab-2), CD45RA (HI100), PD-1 (EH12.2H7), CD69 (FN50), IFN-γ (4S.B3), CCR7 (G043H7), HLA-DR (G46–6), CD27 (L128), CD107a (H4A3), NKG2D (1D11), MICA/B (6D4). Viable cells were identified using LIVE/DEAD Fixable Aqua (Themo Fisher Scientific). Transcription factor staining was performed with the Foxp3/Transcription Factor Staining Buffer Set (eBioscience), and intracellular cytokine staining was conducted using the Cytofix/Cytoperm Kit (BD Biosciences). For cell sorting, fluorescent-labeled antibodies were used to stain single-cell suspensions, and sorting was performed on a FACSAria III (BD Biosciences). Human γδ T cells were defined as CD45⁺CD3⁺TCRγδ⁺, and murine γδ T cells as CD45⁺CD3⁺NK1.1⁻TCRγδ⁺. Flow cytometry data were analyzed using FlowJo software (version 10.8.2; BD Biosciences).

Human γδ T cell expansion and in vitro assays.

PBMCs were isolated from the peripheral blood of healthy donors or GIST patients. Tumor infiltrating lymphocytes (TILs) were isolated from tumor single-cell suspensions using CD45 microbeads (Miltenyi Biotec). For γδ T cell expansion, PBMCs or TILs (2x10⁶ cells per well) were seeded in 24-well plates pre-coated with anti-γδ TCR antibody (0.5 μg/ml; Immunotech) and cultured in RPMI1640 medium supplemented with IL-2 (200U/ml, PeproTech) for 6 to 10 days.

For in vitro cytotoxicity assays, luciferase-expressing GIST T1 cells (1x10⁴ per well) were seeded as target cells in 96-well flat-bottom plates. Expanded γδ T cells were added as effector cells at the indicated effector-to-target (E:T) ratio and co-cultured for 24h. Cells were then lysed, and cytotoxicity was quantified using the Luciferase Assay System (E1501, Promega) on a GloMax Discover Microplate Reader, according to manufacturer’s instructions. For three-dimensional (3D) spheroid assays, 96-well plates were pre-coated with 50 μl of 1.0% agarose (Invitrogen). Luciferase-expressing GIST-T1 cells (5x10³ per well) were seeded and allowed to form spheroids for 24h. CellTrace-labeled γδ T cells (CellMask deep red, Invitrogen) were then added at the indicated E:T ratios. γδ T cell infiltration was analyzed by fluorescence microscopy at 24h and cytotoxicity was measured at 72h using the luciferase assay.

For GIST tissue slice co-culture assays, 6 mm tumor cores were collected in the operating room and maintained on ice. Specimens were embedded in low melting point agarose and sectioned into 300-μm slices using a vibrating microtome as described previously⁴⁴. Tissue slices were placed into 0.4 μm Millicell inserts (Milipore) in 24-well plates and cultured in RPMI1640 medium supplemented with IL-2. One slice was dissociated to determine the number of target cells. CellTrace-labeled γδ T cells were added to the top of tissue slices at an E:T ratio 5:1. After 24h of co-culture, slices were washed of free cells with PBS and dissociated into single-cell suspensions with collagenase and DNase I for flow cytometry analysis. To assess cytotoxicity of tumor-derived γδ T cells against autologous KIT⁺ tumor cells, TILs were isolated and expanded for 6 days with an anti-γδ TCR antibody in the presence of IL2. Expanded TILs were then co-cultured with autologous tumor single-cell suspensions at an E:T ratio of 5:1 for 24h in vitro. Live KIT⁺ tumor cells were quantified by flow cytometry and normalized to tumor single-cell suspensions cultured alone as the control.

For human MICA stimulation, TILs were cultured with recombinant MICA (rhMICA; 10 μg/ml; R&D Systems) or mouse IgG1 isotype control (50 ng/ml; BioLegend) for 72h and analyzed by flow cytometry. To examine the role of NKG2D signaling, sorted γδ T cells (predominantly Vδ1) from untreated GISTs were cultured in 96-well plates with rhMICA in the presence or absence of an anti-NKG2D blocking antibody (10 μg/ml; clone 1D11; BioLegend) added prior to plating. IL-2 was included in the culture medium. After 24h, IFNγ production was measured by ELISpot (ImmunoSpot) following the manufacturer’s instructions.

RNA isolation and real-time PCR.

Total RNA was extracted from human GIST specimens and murine Kit^V558del/+ tumors using the RNeasy Mini Kit (Qiagen). Reverse transcription was performed with TaqMan reagents (Applied Biosystems). TaqMan gene expression assays (Applied Biosystems) included the following probes: mouse Pdcd1 (Mm01285676_m1), Havcr1 (Mm01294183_m1) and Tnfrsf9 (Mm00441899_m1); and human Trgc2 (Hs00827007_m1), Trdv3 (Hs01669899_m1), Ifng (Hs009899291_m1), Mica (Hs00792195_m1), and Klrk1 (Hs00183683_m1). qPCR was performed using a QuantStudio real-time PCR system (Applied Biosystems). Relative gene expression was calculated using the 2^−ΔΔCT method, as described in the manufacturer's instructions, and results were expressed as fold increase for murine samples or as median values for human samples relative to the indicated controls.

Processing and quality control filtering of human γδ T cell scRNA-seq data.

Human γδT cells sorted from PBMCs and tumors were immediately processed for scRNA-seq using the 10x Genomics platform at the Center for Applied Genomics of the Children’s Hospital of Philadelphia. Read count matrices from scRNA-seq data were generated from raw FASTQ files using CellRanger v.7.1.0, with alignment to the GRCh38–2020-A human transcriptome reference. The resulting count matrices were processed and analyzed using Seurat v5⁴⁵. Quality control filtering was applied to each cell, using filters of 100 < nFeatureRNA < 10000 mitochondrial read percentage < 10%. Doublets were then called using DoubletFinder (v2.0.4) using SCTransform normalization, an expected doublet rate of the total cell number / 125,000, and other parameters as default. Then, predicted doublets were removed and additional filters of nFeature_RNA > 300 and nCount_RNA > 1000 were applied.

Sample integration, clustering, and cell type annotation of human γδ T cell scRNA-seq data.

Analysis of scRNA-seq data was conducted using Seurat v5⁴⁵ and R v4.4.0. Samples were normalized individually using the SCTransform function with regression of the mitochondrial read percentage, and a PCA was computed. Then, samples were integrated using the IntegrateLayers function with RPCA integration and default parameters. A UMAP was constructed using the first 30 dimensions of the integrated reduction. Cells were clustered using the FindNeighbors function with the first 30 dimensions of the integrated reduction followed by the FindClusters function. The clustering resolution was chosen empirically. Cell cycle phase was predicted using the CellCycleScoring function after conversion of the default human gene lists to mouse orthologues. Reference mapping was performed with the FindTransferAnchors and Mapquery functions. Differential gene expression was conducted using the FindMarkers function with the default Wilcoxon rank-sum test on log-normalized counts, and clusters were manually annotated based on their differentially expressed genes. For construction of volcano plots, cells were downsampled to a maximum of 1,500 per condition to avoid p-values of zero. For pathway analysis, the log2FC across all genes was used as input to a pre-ranked gene set enrichment analysis (GSEA) using the fgsea package. Pathways were sourced from the Molecular Signatures Database (MsigDB) Hallmark gene sets²⁷ and the KEGG database⁴⁶. Overrepresentation enrichment analysis was performed on the top genes with predefined filtering parameters from select gene set collections using the gost function from the gprofiler2 package (minimum expressing cells = 20%, |log-fc| >1, FDR < 0.05). Differentiation trajectory was inferred using CytoTRACE (v0.3.3) with default parameters using the log-normalized counts.

SCENIC transcriptional regulatory network (TRN) analysis.

The Single-Cell Regulatory Network Inference and Clustering (SCENIC) algorithm⁴⁷ was used to identify transcription factors regulating γδ T cell states. It was run using the pySCENIC implementation as previously described⁴⁸ on the scRNA-seq data, after conversion of the Seurat object to loom format. The GRNBoost2 algorithm was used for GRN inference. To predict transcription factor regulons, we used the human v9 cisTarget motif collection and the hg38_refseq-r80 databases with the 500bpUp100Dw and TSS+/−10kb search spaces. All relevant databases were obtained from: https://resources.aertslab.org/cistarget/. The pySCENIC implementation of AUCell was used to score the activity of regulons for each cell. The SCopeLoomR package (https://github.com/aertslab/SCopeLoomR) was then used to extract the regulons and AUCell matrix from the resulting loom file, and the final AUC matrix was added to the initial Seurat object for visualization and analysis. Differential regulons were identified using the FindAllMarkers function using the Wilcoxon rank-sum test. Gene sets enriched within regulon target genes were analyzed using the enrichGO function of clusterProfiler with the Biological Process Gene Ontology pathways.

TCR Sequencing and analysis of TCRαδ repertoires.

Human γδ T cells were sorted from PBMCs (n=5) and tumor samples (n=6), and genomic DNA (gDNA) was extracted. Next-generation DNA sequencing of the CDR3 regions of human TCRα and TCRδ chains was performed using the ImmunoSEQ^® Assay (Adaptive Biotechnologies). gDNA was amplified by a multiplex PCR method, followed by high-throughput sequencing. Sequence output was filtered on in-frame CDR3s and normalized across all samples. All analyses were carried out using CDR3 amino sequences in the TCRδ chain to track clonal expansion within an individual. Briefly, to identify the TCRδ repertoires, data were pre-processed based on the productive rearranged TCRδ for each sample and TCRs with TCRα expression were first filtered out. Then, the count of TCR was quantified on the amino acid level for each sample. To measure clonotype frequency of TCRδ repertoires, we classified TCR into 5 expansion groups, using 1, 5, 10, 100 as boundaries according to the total count of TCRδ sequence in each sample. The frequency of each expansion group was calculated by the count of unique TCRs in the group divide by the total number of unique TCRs within the same sample. The frequency of the highly expanded group (clone size > 10) was calculated in the same way and compared between PBMC and tumor sample for each patient by the Wilcoxon rank-sum test. We applied Fisher’s exact test to assess clonal expansion between paired PBMC and tumor samples for each individual clone. To compare the number of clones significantly expanded in the tumor relative to PBMC between untreated and resistant patients, we also used Fisher’s exact test. Additionally, differences in the number of clones within each VDJ recombination group between the 2 patient groups were evaluated using the same test.

Mice and treatments.

Kit^V558del/+ mice (8–10 weeks old) on a B6 background (Jackson Laboratory) were maintained in a pathogen-free animal facility. Mice were age- and sex-matched and randomized for each experiment. Animal experiments were approved by the Institutional Animal Care and Use Committee at the University of Pennsylvania (IACUC). Imatinib mesylate (Novartis) was dissolved in the drinking water at 600 mg/L; isotype (rat IgG2b; BioXcell) and anti-PDL1 blocking antibody (10F.9G2; BioXcell) were administered intraperitoneally at a dose of 200 μg per mouse for 1 week as previously described³⁵.

Bulk RNA-seq analysis of murine γδ T cells.

To isolate γδ T cells from Kit^V558del/+ mice, CD45⁺CD3⁺NK1.1⁻TCRγδ⁺ T cells were sorted from vehicle and imatinib treated mice using a FACSAria III (BD Biosciences). Total RNA was extracted, and next-generation RNA sequencing was performed on the Illumina HiSeq 2500 platform at the High-Throughput Sequencing Core of the Children’s Hospital of Philadelphia. Sequencing reads were aligned to the mouse genome (mm10) using STAR aligner with default parameters⁴⁹. Expression was quantified using featureCounts⁵⁰. Differentially expressed genes (DEGs) were identified using the DESeq2 package v1.46.0 and fold changes were shrunk with apeglm. The relevant differential expression coefficient statistic was used as input for Gene Set Enrichment Analysis (GSEA) using the fgsea⁵¹ package. Pathways were sourced from the Molecular Signatures Database (MsigDB)⁵². For a comparison to our published murine scRNA-seq data, we utilized CIBERSORTx⁵³ using the scRNA-seq data as a reference with S-mode batch correction.

Statistical analysis.

Statistical analyses for the scRNA-seq and TCRδ repertoires datasets are described above. Predicted 5-year recurrency free survival was calculated based on patient characteristics utilizing a published nomogram³¹. Overall survival analysis was performed using Kaplan-Meier method with the R package “survival.” Significance was assessed with Logrank test, and hazard ratio with a Cox proportional hazard model. Signature expression scores were calculated using a robust Z-scoring framework on normalized expression data from previously published datasets^28,30. Each gene was median-centered and scaled by a robust spread estimate combining the standard deviation and median absolute deviation (MAD) across samples. Standardized values were then averaged across signature genes to yield a per-sample score. Correlations between continuous variables were evaluated using Spearman’s rank correlation coefficient. Differences between 2 experimental groups were analyzed using an unpaired, 2-tailed Welch’s t-test or Mann-Whitney U test. All in vitro and ex vivo experiments were performed using at least three biological samples. Statistics analyses were carried out using Prism 9.0 (Graph Pad Software) and R software (R Foundation for Statistical Computing). Data are shown as mean ± SEM or median, as specified in the Fig. legends. A p value < 0.05 was considered statistically significant.

Supplementary Material

Supplementary Files

This is a list of supplementary files associated with this preprint. Click to download.

• MSSupplFigsFinal.pdf

Data availability.

scRNA-seq data from sorted human γδ T cells and GIST tumors have been deposited to NCBI GEO under accession number GSE301351. Bulk RNA-seq data from sorted murine γδ T cells are available under accession number GSE300283. TCRαδsequencing data from sorted human γδ T cells are accessible via the ImmunoSEQ platform under the project name UPenn_00131343_R-01. Murine scRNA-seq used in this study were previously deposited in the Sequencing Read Archive (SRA) under accession number PRJNA859907. All remaining data supporting the findings of this study are available within the article and its supplementary materials or from the corresponding author upon reasonable request.

Code availability.

Source code will be made available upon publication, and any code can be made available to the reviewers upon request.