Immune Microenvironment Dynamics and Therapeutic Targets in GIST Revealed by Multi‐Omics and Functional Validation
Xingyuan Li, Jun Xiang, Zewen Chang, Qingchao Tang
TL;DR
This study explores how the immune environment changes in gastrointestinal stromal tumors and identifies new genes that could improve immunotherapy treatments.
Contribution
The study identifies MYBL1 and AIF1L as key genes that regulate CD8+ T cell function and PD-1 response in GIST, offering new therapeutic strategies.
Findings
Mendelian randomization analysis found 18 immune cell phenotypes with causal relationships to GIST.
scRNA-seq revealed dynamic remodeling of the tumor immune microenvironment.
MYBL1 and AIF1L showed synergistic anti-tumor effects when combined with PD-1 blockade.
Abstract
The tumour immune microenvironment (TIME) significantly influences the progression and treatment response of gastrointestinal stromal tumours (GIST), but the causal mechanisms and therapeutic targets remain incompletely understood. We integrated multi‐omics analyses, including Mendelian randomization analysis of 731 immune cell phenotypes to assess their causal relationship with GIST, mediation analysis to identify plasma metabolite mediators, single‐cell RNA sequencing (scRNA‐seq) to characterise TIME heterogeneity and bulk RNA‐seq to screen for differentially expressed genes. Using the GIST‐882 cell and CD8 + T cell co‐culture model, combined with functional assays such as proliferation, migration, invasion and protein uptake tracing, we validated the roles of candidate genes MYBL1 and AIF1L. Additionally, we evaluated their combined effects with PD‐1 inhibitors. Mendelian…
Genes, proteins, chemicals, diseases, species, mutations and cell lines named across the full text — each resolved to its canonical identifier and authoritative record.
Click any figure to enlarge with its caption.
FIGURE 1
FIGURE 2
FIGURE 3
FIGURE 4
FIGURE 5
FIGURE 6- —Noncommunicable Chronic Diseases‐National Science and Technology Major Project 2024ZD0520100.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsGastrointestinal Tumor Research and Treatment · Single-cell and spatial transcriptomics · Cancer Immunotherapy and Biomarkers
Introduction
1
Gastrointestinal stromal tumours (GIST) are the most common mesenchymal neoplasms of the gastrointestinal tract, with 50%–60% occurring in the stomach, 30%–40% in the small intestine, and occasional cases arising outside the gastrointestinal tract [1]. Unlike benign or malignant tumours of epithelial origin, GISTs derive from the interstitial cells of Cajal. As a borderline tumour, GIST accounts for 0.1%–3% of all gastrointestinal malignancies, with a global annual incidence of approximately 10–15 per million [2, 3]. Current diagnostic approaches for GIST primarily involve enhanced CT, endoscopic ultrasound and other imaging modalities combined with pathological examination. Initial differentiation from other submucosal tumours—such as leiomyomas, neuroendocrine tumours, schwannomas, lipomas and ectopic pancreas—is based on features including density, size, boundary definition and blood flow signals, with confirmation through immunohistochemistry. The diagnosis of GIST has evolved from relying solely on morphological assessment to a precision pathology practice that incorporates molecular characteristics. The detection of immunohistochemical biomarkers is central to confirming GIST, with high expression of CD117 and DOG‐1 providing critical evidence for the vast majority of cases and effectively distinguishing GIST from other spindle cell tumours. Building on this, mutation profiling of driver genes such as KIT and PDGFRA further elucidates the molecular nature of the tumour, serving not only to validate the diagnosis but also to define molecular subtypes, predict targeted drug sensitivity and assess prognosis. The integration of IHC and mutation profiling together forms an indispensable cornerstone in the modern diagnostic workflow for GIST, laying a solid pathological foundation for achieving personalised treatment [2, 3, 4]. Risk stratification into very low‐, low‐, intermediate‐ and high‐risk categories is determined by tumour size, mitotic index, primary site and rupture status. Higher risk categories are associated with more malignant features and increased risks of recurrence and distant metastasis. The five‐year survival rate for malignant GIST ranges between 35% and 65% [4]. To date, surgical resection remains the most effective treatment, and it may be combined with tyrosine kinase inhibitors such as imatinib postoperatively—based on risk classification—to reduce the risk of recurrence and metastasis [5].
Although growing basic and clinical research, along with advances in diagnosis and treatment, have deepened our understanding of GIST in recent years, knowledge of this disease remains limited compared to other gastrointestinal malignancies such as gastric and colorectal cancer. Current studies suggest that mutations in KIT or PDGFRA (platelet‐derived growth factor receptor alpha) play a role in GIST pathogenesis [6]; however, the precise aetiology remains unclear, and related research is still relatively scarce. One major reason for the poor prognosis of advanced GIST is the development of resistance to tyrosine kinase inhibitors. Meanwhile, immunotherapy has been widely validated as effective in gastrointestinal solid tumours like gastric and colorectal cancer [7], yet its application in GIST is still in the early exploratory stages. Although GIST is not typically considered an ‘immunologically hot’ tumour, its tumour microenvironment exhibits significant immune cell infiltration, forming a unique immunosuppressive landscape that profoundly influences both disease development and treatment response [8, 9]. A key feature of this microenvironment is the dominant infiltration of immunosuppressive cells. These cells actively shape an immunosuppressive milieu by secreting inhibitory factors, impairing effector T cell function and promoting angiogenesis. This not only aids tumour cells in evading immune surveillance but also directly drives tumour progression, invasion, metastasis and resistance to targeted therapy, making the immune microenvironment a critical participant in GIST pathogenesis [10].
To systematically decipher the immune microenvironment of GIST and identify potential targets for immunotherapy, this study integrated multi‐omics approaches including Mendelian randomization (MR), single‐cell RNA sequencing (scRNA‐seq) and bulk RNA‐seq. First, we employed MR to investigate the causal relationships between 731 immune cell phenotypes and GIST, and further analysed the mediating effects of 1400 plasma metabolites. Subsequently, using scRNA‐seq data, we characterised the cellular composition, lineage differentiation and intercellular communication networks within the GIST tumour microenvironment, dynamically illustrating shifts in immune cell states during tumour progression. Building on this, by integrating multiple bulk RNA‐seq datasets, we identified key genes—MYBL1 and AIF1L—that show graded expression changes between tumour and normal tissues and are significantly correlated with the immune checkpoint molecule PD‐1. Finally, through in vitro functional experiments, we validated the roles of these two genes in regulating malignant phenotypes of GIST cells and their interactions with CD8^+^ T cells, and explored their synergistic anti‐tumour effects when combined with PD‐1 inhibitors. This study aims to provide new theoretical insights and experimental evidence for a deeper understanding of GIST immunobiology, discovery of novel therapeutic targets and development of combined immunotherapy strategies.
Methods
2
Mendelian Randomization
2.1
Design and Data Source
2.1.1
In this study, a two‐step Mendelian randomization (MR) approach was used to investigate the causal relationship between immune cells and gastrointestinal stromal tumours (GISTs), as well as whether this causal relationship is mediated by plasma metabolites; the first step involved using two‐sample MR to explore the causal relationships between 731 types of immune cells and GISTs, the second step employed two‐sample MR to investigate the causal relationships between 1400 types of plasma metabolites and GISTs, and finally, we examined the mediating effect of plasma metabolites in the relationship between immune cells and GISTs. The application of MR satisfies three key assumptions: (i) instrumental variables must have a stable and strong correlation with the exposure; (ii) instrumental variables are independent of any confounding factors that affect the ‘exposure‐outcome’ relationship; (iii) instrumental variables can only influence the outcome through the exposure, rather than acting on the outcome via other pathways [11].
Data on 731 immune cell phenotypes (Ebi‐a‐GCST0001391 to Ebi‐a‐GCST0002121) were obtained from the GWAS database (https://gwas.mrcieu.ac.uk/), covering immune subsets including B cells, dendritic cells, T cells, monocytes, myeloid cells, TBNK (T/B/NK cells) and regulatory T cells (Tregs); phenotypes related to 1400 plasma metabolites (GCST90199621–GCST90201020) were also retrieved from the GWAS database, which included 8299 European individuals and covered 1091 plasma metabolites along with 309 metabolite ratios; data related to GISTs were obtained from the FinnGen database, which included 351 GIST cases and 378,695 controls (https://r12.finngen.fi).
Selection of Genetic Instrumental Variables
2.1.2
For immune cell‐related phenotypes, we set the GWAS threshold at p < 5e‐6; for plasma metabolite‐related phenotypes, the GWAS threshold was set at p < 1e‐5. All SNPs showed a strong correlation with their corresponding phenotypes (F > 10), and linkage disequilibrium analysis was performed for all to ensure independence [2] (r < 0.001, kb > 10,000). We manually excluded palindromic sequences, as they can affect gene expression [12, 13].
Statistical Analysis
2.1.3
We used three methods—Inverse Variance Weighted (IVW), MR‐Egger and Weighted Median—to estimate the causal relationship between exposure and outcome. Among these, the IVW method served as the primary research method, while MR‐Egger and Weighted Median methods were used as supplements to verify the consistency of the causal relationship. In the sensitivity analysis, to ensure the robustness of the results, we employed Cochrane's Q test to assess heterogeneity and used the intercept term of MR‐Egger regression analysis to detect horizontal pleiotropy [14]; the leave‐one‐out method was applied to verify whether the results were affected by a single SNP, and the Steiger test was used to examine the direction of the causal relationship. In addition, we performed multiple test correction using the Bonferroni method. A result was considered meaningful when the following conditions were met: p < 0.05/number of exposures in the IVW method, consistent directions of Beta values across all research methods, and no presence of horizontal pleiotropy. When 0.05/number of exposures < p < 0.05, we considered a potential causal relationship between exposure and outcome [15]. All our analyses were conducted using R software (Version 4.2.3) with the TwoSampleMR package (Version 0.5.7).
Mediation Analysis
2.1.4
Mediation analysis can be used to evaluate the role of a third variable in the mechanism by which an exposure factor leads to an outcome factor. Through mediator Mendelian randomization (MR) studies, we identified pathways from immune cell traits to plasma metabolites and then to gastrointestinal stromal tumours (GISTs), which helps clarify the potential mechanisms by which immune cell traits promote GIST development. First, we used mediation analysis to assess the potential associations between immune cell traits and plasma metabolites screened by MR. Subsequently, through a two‐stage MR workflow, we quantified the ‘indirect’ effect of immune cell traits on GISTs mediated by plasma metabolites. The specific calculation formulas are as follows: Mediation proportion = β (mediating effect)/β (total effect); β (mediating effect) = β (direct effect A) × β (direct effect B) [16]. Here, the total effect refers to the causal effect of immune cell traits on GISTs; direct effect A refers to the causal effect of immune cell traits on plasma metabolites; and direct effect B refers to the causal effect of plasma metabolites on GISTs.
Single‐Cell and Bulk Transcriptomic Analysis
2.2
Collection and Processing of Omics Data
2.2.1
Publicly available scRNA‐seq and bulk RNA‐seq data from gastrointestinal stromal tumour (GIST) patients were collected from the Gene Expression Omnibus (GEO) database. The included scRNA‐seq datasets comprise GSE254762 and GSE162115 [17, 18]. The bulk RNA‐seq datasets include GSE225819, GSE155800 and GSE136755 [19]. All bulk RNA‐seq data were uniformly normalised and expressed in TPM (Transcripts Per Million) format.
Analysis of Single‐Cell Data
2.2.2
The R package ‘Seurat’ was utilised to convert scRNA‐seq data into Seurat objects. Three quality control strategies were applied: only genes detected in at least five cells were retained; cells expressing fewer than 100 genes were removed; and cells with mitochondrial gene expression exceeding 5% were excluded. Subsequently, the data were normalised using the ‘NormalizeData’ function, and the top 2000 highly variable genes were identified via the ‘FindVariableFeatures’ function. Principal component analysis (PCA) was then performed using the ‘RunPCA’ function, and the top 20 principal components were selected for cell clustering analysis. Differential expression genes (DEGs) for each cluster were calculated using the ‘FindAllMarkers’ function, with a threshold set at FDR < 0.05 and |log_2_(fold change)| > 0.25. Cell types for each cluster were manually annotated.
Trajectory Analysis
2.2.3
The Monocle algorithm was employed to infer potential differentiation trajectories among distinct cell phenotypes. The Seurat object was imported into Monocle, and dimensionality reduction was performed using the DDRTree method, with conserved and differentially expressed subtype‐specific genes in GIST applied. Branch‐associated dynamically expressed genes were identified using Monocle's BEAM test, and genes with the q < 0.001 were selected for heatmap visualisation.
Cell–Cell Interaction Analysis
2.2.4
This study used the Python package CellPhoneDB to construct cell–cell interaction networks within the GIST microenvironment, primarily following its default parameter settings. This tool predicts potential interaction strengths between different cell subtypes based on the expression of ligand–receptor pairs. Interactions were subsequently filtered according to statistical significance (p < 0.05). Specifically, if epithelial cells expressed a ligand or receptor, the corresponding interactions were defined as outgoing or incoming, respectively. Biologically relevant ligand–receptor pairs were further analysed across different cell subtypes.
Statistical Analysis
2.2.5
All statistical analyses were conducted using R software. All statistical tests were two‐sided, and a p < 0.05 was considered statistically significant.
Cell‐Based Functional Assays
2.3
Cell Lines and Culture Conditions
2.3.1
The human gastrointestinal stromal tumour (GIST) cell line GIST‐882 was obtained from the American Type Culture Collection (ATCC) and maintained in RPMI‐1640 medium (HyClone) supplemented with 10% foetal bovine serum (FBS, Gibco) and 1% penicillin–streptomycin (Sigma‐Aldrich). Cells were cultured at 37°C in a humidified incubator with 5% CO_2_. Human CD8^+^ T cells were isolated from peripheral blood mononuclear cells (PBMCs) of healthy volunteers using an immunomagnetic bead separation kit (Miltenyi Biotec). These cells were activated and expanded in X‐VIVO 15 medium (Lonza) containing 10% human AB serum and 100 U/mL IL‐2 (PeproTech) for subsequent co‐culture experiments.
Plasmid Construction and Establishment of Stable Cell Lines
2.3.2
To investigate the biological functions of MYBL1 and AIF1L, stable genetically modified GIST‐882 cell lines were established using lentiviral transduction.
MYBL1 Knockdown Model: Specific shRNA sequences targeting human MYBL1 (shMYBL1) and a negative control sequence (shNC) were cloned into the pLKO.1 lentiviral vector (Addgene). Lentiviruses were packaged in 293 T cells by co‐transfecting psPAX2 and pMD2. G packaging plasmids using Lipofectamine 3000 (Invitrogen). GIST‐882 cells were infected with the viral supernatant, and stable pools were selected using 2 μg/mL puromycin (Sigma). AIF1L Overexpression Model: The full‐length coding sequence of human AIF1L was cloned into the pCDH‐CMV‐MCS‐EF1‐Puro lentiviral expression vector (System Biosciences), with a C‐terminal green fluorescent protein (GFP) tag introduced via molecular cloning. Lentiviral packaging and infection procedures were performed as above, yielding stable overexpression cell lines (oeAIF1L). Cells infected with empty vector (Vector) served as controls.
The knockdown or overexpression efficiency in all stable cell lines was validated by quantitative real‐time PCR (qRT‐PCR) and Western blot analysis.
Establishment of CD8
- T Cell Co‐Culture System
2.3.3
An in vitro co‐culture system was established to model the tumour immune microenvironment. GIST‐882 cells were seeded in 6‐well plates at a density of 1 × 10 [5] cells/well. After adherence, pre‐activated CD8^+^ T cells were added at an effector‐to‐target ratio of 5:1. Co‐cultures were maintained in RPMI‐1640 medium with 10% FBS. GIST‐882 cells cultured alone served as the control group. Co‐culture duration ranged from 8 to 72 h, depending on the specific experimental requirements.
Colony Formation Assay
2.3.4
Treated GIST‐882 cells were plated at low density (500 cells/well) in 6‐well plates and cultured under co‐culture or mono‐culture conditions for 14 days. Colonies were fixed with 4% paraformaldehyde, stained with 0.1% crystal violet and colonies larger than 50 μm in diameter were counted under a microscope. Experiments were independently repeated three times.
Cell Proliferation Assay
2.3.5
Cell proliferation was assessed using a CCK‐8 kit (Dojindo). GIST‐882 cells were seeded in 96‐well plates at 5 × 10 [3] cells/well. After 24, 48 and 72 h of co‐culture or mono‐culture, 10 μL of CCK‐8 solution was added to each well. Following 2 h of incubation, the absorbance at 450 nm was measured using a microplate reader (BioTek).
Cell Migration and Invasion Assays
2.3.6
Transwell chambers (8 μm pore size, Corning) were used for evaluation. GIST‐882 cells (1 × 10/chamber) [5] were resuspended in serum‐free medium and placed in the upper chamber, while the lower chamber contained medium with 10% FBS as a chemoattractant. After 24 h of co‐culture, non‐migrated cells on the upper surface were removed with a cotton swab. Cells that migrated to the lower surface were fixed with 4% paraformaldehyde, stained with crystal violet and counted in five randomly selected fields under a microscope.
Live/Dead Cell Staining Assay
2.3.7
After 48 h of co‐culture, cell viability was assessed using the Calcein‐AM/PI dual‐fluorescence viability/cytotoxicity kit (Invitrogen). Cells were stained and observed under a fluorescence microscope (Nikon). Calcein‐AM labels live cells with green fluorescence, while propidium iodide (PI) stains nuclei of dead cells with red fluorescence. The cell death rate (dead cells/total cells × 100%) was calculated using ImageJ software.
Protein Uptake Tracing Using the ‘Split‐GFP’ System
2.3.8
To directly and specifically demonstrate that MYBL1 and AIF1L proteins derived from tumour cells are internalised by CD8^+^ T cells—rather than merely adhering to the cell surface—a protein complementation system based on ‘split green fluorescent protein (split‐GFP)’ was constructed. System Design Principle: The system is based on splitting GFP into two non‐fluorescent fragments: a large fragment containing β‐strands 1–10 (GFP1‐10) and a short peptide tag containing the 11th β‐strand (GFP11). Functional GFP fluorescence is reconstituted only when these two fragments meet inside a cell. Cell Engineering: Donor Cells (GIST‐882): Stable GIST‐882 cell lines were generated to overexpress MYBL1‐GFP11 or AIF1L‐GFP11 fusion proteins. The fusion proteins, consisting of the target protein covalently linked to the GFP11 tag via a flexible peptide linker, were normally synthesised by the tumour cells and secreted extracellularly. Recipient Cells (CD8^+^ T cells): CD8^+^ T cells were engineered to stably express the GFP1‐10 fragment, which is constitutively present in the cytoplasm and is non‐fluorescent on its own. Experimental Procedure and Detection: The two engineered cell types were co‐cultured at a donor‐to‐recipient ratio of 1:5. At 8, 16 and 24 h post‐co‐culture, live‐cell time‐lapse fluorescence microscopy (Nikon Ti‐E) was performed for dynamic observation. Interpretation and Mechanism: Fluorescent GFP signal reconstitution inside CD8^+^ T cells occurs only if the GFP11‐tagged MYBL1 or AIF1L fusion protein secreted by GIST‐882 cells is actively taken up and enters the cytoplasm. The GFP11 tag then binds to the pre‐existing GFP1‐10 fragment, reconstituting a functional GFP molecule within the CD8^+^ T cell. A time‐dependent increase in green fluorescence intensity within the cytoplasm of CD8^+^ T cells was observed. This result provides direct evidence for the internalisation of tumour‐derived proteins by CD8^+^ T cells. The mean fluorescence intensity inside cells was quantified using NIS‐Elements software.
Statistical Analysis
2.3.9
All experimental data are presented as mean ± standard deviation. Statistical analyses were performed using GraphPad Prism 9.0 software. Multiple group comparisons were analysed by one‐way ANOVA followed by Tukey's post hoc test, while comparisons between two groups were conducted using Student's t‐test. A p‐value < 0.05 was considered statistically significant. All experiments were independently repeated at least three times.
Results
3
The Overall Causal Impact of Immune Cells Phenotypes on GIST
3.1
Two‐sample Mendelian randomization analysis identified 18 immune cell phenotypes with potential causal relationships to GIST (Figures 1A–E and S1), including 10 risk factors: Plasma Blast—Plasma Cell %lymphocyte (OR: 1.17 [1.06, 1.30]); Plasma Blast—Plasma Cell %B cell (OR: 1.21 [1.04, 1.41]); CD25 on CD4 regulatory T cell (OR: 1.26 [1.03, 1.55]); IgD—CD38 + B cell Absolute Count (OR: 1.20 [1.02, 1.42]); CD127 on granulocyte (OR: 1.23 [1.03, 1.48]); IgD + CD24—B cell Absolute Count (OR: 1.14 [1.01, 1.28]); Monocyte Absolute Count (OR: 1.16 [1.01, 1.33]); CD66b on CD66b++ myeloid cell (OR: 1.18 [1.01, 1.37]); TCRgd T cell Absolute Count (OR: 1.30 [1.01, 1.68]); TCRgd T cell %T cell (OR: 1.17 [1.00, 1.36]); Protective factors comprising eight types: CD45RA + CD28—CD8 + T cell %T cell (OR: 0.998 [0.996, 0.9996]); CD16 + monocyte %monocyte (OR: 0.686 [0.504, 0.935]); CD20 on CD24 + CD27 + B cell (OR: 0.733 [0.562, 0.957]); CD20 on memory B cell (OR: 0.818 [0.688, 0.972]); CCR7 on naive CD4 + T cell (OR: 0.867 [0.759, 0.992]); CD62L—plasmacytoid Dendritic Cell %Dendritic Cell (OR: 0.817 [0.675, 0.989]); CD3—lymphocyte Absolute Count (OR: 0.733 [0.539, 0.997]); HLA DR on HLA DR + CD8 + T cell (OR: 0.725 [0.526, 0.998]), among these positive findings, the CD45RA + CD28—CD8 + T cell %T cell was characterised by an excessively narrow confidence interval. Despite a statistically significant P value, the conclusion derived therefrom lacked sufficient robustness. Thus, this result was reclassified as a negative finding. The symmetry of the funnel plot demonstrates the robustness of the results (Figure S2). No horizontal pleiotropy was observed during the study, and the leave‐one‐out method indicated that the results were not driven by a single SNP (Figure S3).
The causal relationship between immune cells and gastrointestinal stromal tumours. (A) Forest plot for the causal relationship between 18 immune cell types and gastrointestinal stromal tumour. (B) Circular heatmap for the causal relationship between 18 immune cell types and gastrointestinal stromal tumours. (C) Circular heatmap of the causal relationship between 731 immune cell types and gastrointestinal stromal tumours. (D) Volcano plot of the Mendelian randomization analysis between 731 immune cell types and gastrointestinal stromal tumours. (E) Manhattan plot depicting the causal relationship between exposure and outcome.
The Causal Influence of Plasma Metabolite Phenotypes on GIST
3.2
Two‐sample Mendelian randomization identified causal relationships between 41 plasma metabolites and gastrointestinal stromal tumour (GIST) (Figure 2A–C). Among these, the risk factors included levels of 1‐oleoyl‐GPE (18:1), 4‐allylphenol sulphate, 5α‐pregnan‐3β,20α‐diol monosulphate (2), pregnenediol sulphate (C21H34O5S), palmitoyl sphingomyelin (d18:1/16:0), 2‐methoxyresorcinol sulphate, furaneol sulphate, heptenedioate (C7:1‐DC), tetradecadienoate (14:2), 3‐ethylcatechol sulphate (2), 2′‐deoxyuridine, serine, X‐13553, X‐21742, X‐24544, X‐24978, as well as the sphingosine‐to‐phosphate ratio, cysteinylglycine‐to‐glutamate ratio, adenosine 5′‐diphosphate (ADP)‐to‐glucose ratio and phosphoethanolamine‐to‐choline ratio. The protective factors included levels of 1‐myristoyl‐2‐palmitoyl‐GPC (14:0/16:0), 1‐linoleoylglycerol (18:2), 1,3,7‐trimethylurate, octadecanedioate, glycolithocholate sulphate, androstenediol (3β,17β) disulphate (2), perfluorooctanesulfonate (PFOS), pentose acid, 3‐ureidopropionate, pipecolate, cholesterol, alanine, X‐11852, X‐12818, X‐21442, as well as the alpha‐ketoglutarate‐to‐aspartate ratio, adenosine 5′‐diphosphate (ADP)‐to‐glycerol 3‐phosphate ratio, isoleucine‐to‐phosphate ratio, citrulline‐to‐ornithine ratio and adenosine 5′‐diphosphate (ADP)‐to‐choline phosphate ratio. Notably, horizontal pleiotropy was detected between the outcome and the following exposures: levels of 3‐methylxanthine, carnitine C14, 4‐oxo‐retinoic acid, oleoylcholine, 3‐hydroxy‐2‐methylpyridine sulphate, 5‐hydroxy‐2‐methylpyridine sulphate, nicotinamide, X‐12740, as well as the cysteine‐to‐5‐oxoproline ratio, adenosine 5′‐diphosphate (ADP)‐to‐glucose ratio, and choline phosphate‐to‐phosphoethanolamine ratio. Consequently, the results for these exposures were not robust. Additionally, given the highly similar nomenclature and functions of various amino acids and derivatives, nucleotides and derivatives, as well as unidentified compounds identified in the positive exposure findings, we conducted multivariate Mendelian randomization analyses (Table S1‐3). The results revealed that X‐11852 levels, X‐13553 levels, X‐21742 levels, X‐24978 levels, the ADP to glycerol 3‐phosphate ratio, the ADP to choline phosphate ratio and serine levels were no longer statistically significant. As such, they were thus considered negative findings. For the remaining exposures, no horizontal pleiotropy was observed with the outcome, and the leave‐one‐out method did not identify any single SNP that significantly influenced the results.
The causal relationship between plasma metabolites and gastrointestinal stromal tumours. (A) Forest plot of the causal relationship between 41 plasma metabolites and gastrointestinal stromal tumours. (B) Circular heatmap of the causal relationship between 41 plasma metabolites and gastrointestinal stromal tumours. (C) Volcano plot of the Mendelian randomization analysis between 1400 plasma metabolites and gastrointestinal stromal tumours. (D) The proportion of the mediating effects of the four metabolites. (E) The mediating effects of the four metabolites. (F) Forest plot depicting the causal relationship between immune cells and plasma metabolites.
The Causal Effect of Immune Cells Phenotypes on Plasma Metabolite Phenotypes
3.3
Two‐sample MR identified causal relationships between 12 pairs of immune cells and plasma metabolites (Figure 2F), among which 9 pairs were positively correlated: IgD‐CD38+ B cell Absolute Count—Pentose acid levels (OR: 1.046 [1.013, 1.080], p = 0.006), IgD‐CD38+ B cell Absolute Count—4‐allylphenol sulphate levels (OR: 1.047 [1.013, 1.082], p = 0.007), Monocyte Absolute Count—3‐ureidopropionate levels (OR: 1.046 [1.014, 1.079]), CD16+ monocyte %monocyte—Glycolithocholate sulphate levels (OR: 1.100 [1.034, 1.172]), TCRgd T cell %T cell—1,3,7‐trimethylurate levels (OR: 1.069 [1.014, 1.127]), CD20 on CD24+ CD27+ B cell—Serine levels (OR: 1.070 [1.013, 1.131]), CD20 on memory B cell—Pentose acid levels (OR: 1.051 [1.011, 1.093]), CD127 on granulocyte—X‐21742 levels (OR: 1.074 [1.018, 1.133]), CD25 on CD4 regulatory T cell—Isoleucine to phosphate ratio (OR: 1.073 [1.010, 1.141]); three pairs were negatively correlated: TCRgd T cell Absolute Count—Furaneol sulphate levels (OR: 0.882 [0.798, 0.976]), TCRgd T cell %T cell—Furaneol sulphate levels (OR: 0.934 [0.878, 0.994]), CD20 on CD24+ CD27+ B cell—1,3,7‐trimethylurate levels (OR: 0.900 [0.848, 0.956]). Sensitivity analyses did not detect the presence of horizontal pleiotropy, and the leave‐one‐out method showed that the results were not driven by a single SNP.
The Mediating Role of Plasma Metabolites Between Immune Cells and GIST
3.4
We conducted mediation analysis on the screened results where the associations between exposure and mediator, exposure and outcome, as well as mediator and outcome were all positive. Eventually, four mediation pathways were identified (Figure 2D,E): IgD‐CD38+ B cell Absolute Count—4‐allylphenol sulphate levels—GIST: mediation proportion was 11.285% [1.138% ~ 21.431%], CD16+ monocyte %monocyte—Glycolithocholate sulphate levels—GIST: mediation proportion was 6.359% [0.98% ~ 11.737%], CD20 on memory B cell—Pentose acid levels—GIST: mediation proportion was 11.777% [0.304% ~ 23.25%], CD127 on granulocyte—X‐21742 levels—GIST: mediation proportion was 15.415% [0.638% ~ 30.192%]. Due to the relatively limited sample size in this study, the confidence intervals observed in the mediation analysis were consistently wide. Although the point estimates suggest certain proportions of mediation effects, the broad confidence intervals reflect considerable uncertainty in the current estimates. This may reduce the precision of the effect estimates, leaving the strength and direction of the mediation pathways ambiguous from a biological interpretation standpoint. Therefore, the clinical or mechanistic implications of these pathways should be interpreted with caution.
Single‐Cell Transcriptomic Analysis Reveals Cellular Heterogeneity and Dynamic Evolution in the Tumour Microenvironment
3.5
Analysis of tumour samples and their matched tissues using single‐cell RNA sequencing (scRNA‐seq) identified 30 distinct cell populations with unique transcriptional signatures (Figure 3A). UMAP clustering revealed these cells could be annotated into multiple types, including tumour cells, epithelial cells, immune cells (CD8^+^ T cells, B cells, NK cells, dendritic cells, macrophages, etc.), fibroblasts, and endothelial cells (Figure 3B). Sample distribution analysis (Figure 3C) demonstrated significant enrichment of tumour cells, exhausted T cells, Treg cells and neutrophils in tumour and metastatic tissues, while plasma cells and CD8^+^ T cells were predominantly located in normal tissues. This distribution pattern indicates systematic remodelling of immune composition and cellular proportions within the tumour microenvironment during tumour initiation and progression.
Single‐cell transcriptomic analysis unveils the cellular composition, developmental trajectories and communication networks within the tumour microenvironment. (A) UMAP clustering of single‐cell transcriptomes shows the distribution of 30 distinct cell populations identified in the samples. (B) Cells were annotated into major types based on marker genes, including tumour cells, epithelial cells, immune cells (CD8+ T cells, B cells, NK cells, dendritic cells, macrophages, etc.), fibroblasts and endothelial cells. (C) Proportional differences of each cell type across samples from different tissue origins (normal, tumour, metastatic). (D–F) Pseudotime analysis of CD8+ T cells. The trajectory demonstrates a gradual shift from cytotoxic CD8+ T cells toward an exhausted T cell phenotype, accompanied by decreased expression of the effector gene EEF1A1 and increased expression of the exhaustion marker gene TMSB10. (G–I) Pseudotime trajectory and gene dynamics of the B cell lineage. The analysis reveals progressive differentiation of B cells into plasma cells along the trajectory, with gradually increasing expression levels of TP1 and MALAT1 during differentiation. (J) Cell–cell communication network based on ligand–receptor interactions. (K) Heatmap of intercellular communication strength, showing the enrichment of major signalling pathways among different cell types.
To further investigate the functional dynamics of immune cells in the tumour microenvironment, we performed pseudotime analysis on major immune cell subsets. The results showed CD8^+^ T cells following a differentiation trajectory from cytotoxic CD8^+^ T cells towards an exhausted T cell phenotype (Figure 3D,E), reflecting the gradual loss of effector function under persistent antigen stimulation and immunosuppressive conditions. Gene expression analysis revealed progressive downregulation of the effector gene EEF1A1 along the pseudotime axis, while the exhaustion‐associated gene TMSB10 was significantly upregulated (Figure 3F), further supporting the dynamic process of T cell functional exhaustion. Additionally, pseudotime trajectory analysis of the B cell lineage revealed a differentiation path from naïve B cells to plasma cells (Figure 3G,H), indicating active humoral immune responses within the tumour microenvironment. Analysis along this trajectory showed progressively increasing expression levels of TP1 and MALAT1 (Figure 3I), suggesting significant transcriptional activation during B cell differentiation into plasma cells.
To further understand the interactions between different cell types within the tumour microenvironment, we constructed a cell–cell communication network based on ligand–receptor pairs. The results (Figure 3J) revealed highly intricate and densely interconnected signalling interactions among tumour cells, immune cells and fibroblasts, with particularly prominent communication along the tumour cell‐macrophage‐fibroblast axis. This suggests that cross‐talk among these cell populations may play a critical role in regulating immunosuppression and promoting tumour progression. Heatmap analysis further elucidated the specific distribution of signalling pathways among different cell types (Figure 3K). Notably, tumour cells communicated with macrophages, monocytes and NK cells primarily through signalling pathways such as APP‐CD74 and CXCL12‐CXCR4, suggesting their potential involvement in regulating key processes including cell migration, immune evasion and stromal remodelling.
Differential Expression Analysis Reveals Conserved Alterations and Correlations of Key Immune‐Regulatory Genes
3.6
To identify key differentially expressed genes demonstrating consistent expression trends between tumour and normal tissues, we performed differential expression analysis on multiple publicly available transcriptomic datasets (GSE136755, GSE155800, GSE225819). Volcano plots (Figure 4A–C) revealed numerous upregulated and downregulated differentially expressed genes (|logFC| > 1, FDR < 0.05) across all datasets. Subsequent Venn diagram analysis comparing common differentially expressed genes across the three datasets (Figure 4D,G) identified seven genes consistently upregulated and four genes consistently downregulated in all samples.
Consistent alteration patterns of differentially expressed genes across multiple transcriptomic datasets and their correlation analysis. (A–C) Volcano plots from differential expression analysis of three public transcriptomic datasets (GSE136755, GSE155800, GSE225819). Red dots represent upregulated genes, blue dots represent downregulated genes and grey dots represent genes without significant differences. A substantial number of significantly differentially expressed genes (|logFC| > 1, FDR < 0.05) were observed in each dataset, suggesting extensive transcriptional changes between tumour and normal tissues. (D) Venn diagram analysis of upregulated genes across the three datasets. (E) Among the upregulated genes, the expression of MYBL1 showed a significant positive correlation with the immune checkpoint gene PDCD1 (R = 0.82, p = 1.9 × 10−5), suggesting that MYBL1 activation may promote PDCD1 expression and thereby participate in the regulation of immune responses. (F) MYBL1 was significantly upregulated in tumour tissues across three independent datasets (GSE225819, GSE155800, GSE136755), with the highest expression levels observed in metastatic samples. (G) Venn diagram analysis of downregulated genes across the three datasets. (H) Among the downregulated genes, AIF1L expression showed a significant negative correlation with PDCD1 (R = −0.63, p = 0.0035), suggesting its potential involvement in immune escape within the tumour microenvironment through the negative regulation of immune signalling transduction. (I) AIF1L was significantly downregulated in tumour tissues across multiple datasets, exhibiting the highest expression levels in normal tissues.
Among the upregulated genes, the expression of MYBL1 demonstrated a significant positive correlation with the immune checkpoint gene PDCD1 (R = 0.82, p = 1.9 × 10^−5^) (Figure 4E). This association indicates that MYBL1 expression co‐varies with PDCD1 at the tissue level during GIST progression, potentially reflecting shared immune‐context–dependent regulation rather than a direct mechanistic relationship. Conversely, among the downregulated genes, AIF1L expression showed a negative correlation with PDCD1 (R = −0.63, p = 0.0035) (Figure 4H). This inverse association suggests that reduced AIF1L expression is linked to tissue states characterised by higher immune checkpoint expression, consistent with an immunosuppressive tumour microenvironment, rather than indicating a direct inhibitory effect on immune signalling.
Further validation of the expression patterns of these key genes (Figure 4F,I) confirmed that MYBL1 was significantly upregulated in tumour tissues across all three independent datasets, with the highest expression levels observed in metastatic samples. In contrast, AIF1L was significantly downregulated in tumour tissues, exhibiting markedly higher expression in normal tissues compared to tumour samples. Collectively, these results indicate that MYBL1 and AIF1L exhibit opposite expression trends relative to PDCD1 at the tissue level. Their associations with PDCD1 likely reflect distinct immune‐context–dependent states within the GIST microenvironment, rather than direct regulation of PD‐1 expression.
In summary, consistent gene expression patterns across multiple independent datasets reveal stable characteristics of tumour‐associated immune regulatory genes. The positive correlation between MYBL1 and PDCD1, coupled with the negative correlation between AIF1L and PDCD1, suggests that these two molecules may participate in immune activation and immunosuppression pathways, respectively. These findings provide important insights for elucidating the molecular mechanisms of the tumour immune microenvironment and identifying potential therapeutic targets.
The Inhibitory Effect of MYBL1 and AIF1L on the Malignant Phenotype of GIST‐882 Depends on CD8+ T Cells
3.7
In this study, we systematically evaluated the changes in malignant phenotypes of GIST‐882 cells under co‐culture with CD8^+^ T cells by establishing a stable MYBL1 knockdown model. The colony formation assay demonstrated that MYBL1 knockdown significantly suppressed colony‐forming ability, and this inhibitory effect was further enhanced in the presence of CD8^+^ T cell co‐culture (Figure 5A). CCK‐8 assay results revealed more pronounced suppression of tumour cell proliferation in the co‐culture system (Figure 5B). Transwell assays further confirmed that MYBL1 knockdown and CD8^+^ T cell co‐culture exhibited synergistic effects in inhibiting cell invasion (Figure 5C). In the AIF1L overexpression model, we observed similar synergistic inhibition: the combination of AIF1L overexpression with CD8^+^ T cell co‐culture resulted in significantly stronger suppression of GIST‐882 cell proliferation, migration and invasion capabilities compared to overexpression alone (Figure 5D–F). To further investigate the underlying mechanism, we generated GFP‐tagged MYBL1 and AIF1L overexpressing GIST‐882 cells and conducted dynamic co‐culture experiments with CD8^+^ T cells. Time‐lapse fluorescence microscopy observations showed time‐dependent enhancement of GFP fluorescence signals within CD8^+^ T cells at 8, 16 and 24 h of co‐culture, indicating effective uptake of tumour‐derived MYBL1 and AIF1L proteins by CD8^+^ T cells (Figure 5G,H). In summary, our findings demonstrate that the inhibitory effects of MYBL1 and AIF1L on the malignant behavioural phenotypes of GIST‐882 cells are CD8^+^ T cell‐dependent.
Co‐culture with CD8+ T cells can enhance the inhibitory effects of MYBL1 knockdown or AIF1L overexpression on the malignant phenotype of the GIST‐882 cell line. (A) Representative images and quantitative analysis of colony formation by GIST‐882 cells with MYBL1 knockdown under co‐culture or mono‐culture conditions with CD8+ T cells. (B) Effect of MYBL1 knockdown on GIST‐882 cell proliferation assessed by CCK‐8 assay under co‐culture or mono‐culture conditions with CD8+ T cells. (C) Representative images and quantitative analysis of Transwell migration and invasion assays of GIST‐882 cells with MYBL1 knockdown under co‐culture or mono‐culture conditions with CD8+ T cells. (D) Representative images and quantitative analysis of colony formation by GIST‐882 cells with AIF1L overexpression under co‐culture or mono‐culture conditions with CD8+ T cells. (E) Effect of AIF1L overexpression on GIST‐882 cell proliferation assessed by CCK‐8 assay under co‐culture or mono‐culture conditions with CD8+ T cells. (F) Representative images and quantitative analysis of Transwell migration and invasion assays of GIST‐882 cells with AIF1L overexpression under co‐culture or mono‐culture conditions with CD8+ T cells. (G–H) Uptake of tumour‐derived MYBL1 and AIF1L proteins by CD8+ T cells visualised using the split‐GFP system.
MYBL1 Knockdown or AIF1L Overexpression Combined With a PD‐1 Inhibitor Exerts a Synergistic Anti‐Tumour Effect in GIST
3.8
In the CD8^+^ T cell co‐culture model, this study established four experimental groups: shNC control, shMYBL1 knockdown, anti‐PD‐1 monotherapy and combined shMYBL1 knockdown with anti‐PD‐1 treatment. A systematic evaluation through multiple functional assays demonstrated the inhibitory effects of these treatments on malignant phenotypes of GIST‐882 cells. In the colony formation assay, the combined treatment group exhibited the most significant suppression of colony formation (Figure 6A). CCK‐8 cell proliferation assays confirmed that the combined treatment more effectively inhibited cellular proliferative activity (Figure 6B). Transwell assays further revealed that the combined treatment produced significantly stronger inhibition of cell migration and invasion capabilities compared to any single treatment group (Figure 6C). Live/dead cell staining assays also verified the highest cell mortality rate in the combined treatment group (Figure 6D). In the AIF1L overexpression (oeAIF1L) model, we observed consistent trends: the combination of AIF1L overexpression with anti‐PD‐1 treatment demonstrated superior synergistic inhibitory effects on colony formation, migration, invasion and proliferation compared to individual treatment groups (Figure 6E–H). This study confirms that in the GIST immune microenvironment model, targeting either MYBL1 or AIF1L significantly enhances the anti‐tumour efficacy of PD‐1 inhibitors, demonstrating a synergistic effect (Figures S4 and S5). These findings provide experimental evidence for developing novel GIST treatment strategies based on synergistic immune regulation.
Targeting MYBL1 or AIF1L potentiates the efficacy of PD‐1 blockade in a CD8+ T cell‐mediated co‐culture model. (A) Representative images and quantitative analysis of colony formation by GIST‐882 cells under CD8+ T cell co‐culture treated with MYBL1 knockdown, anti‐PD‐1 monotherapy or their combination. (B) CCK‐8 assay assessing proliferation of GIST‐882 cells under CD8+ T cell co‐culture treated with MYBL1 knockdown, anti‐PD‐1 monotherapy or their combination. (C) Representative images and quantitative analysis of Transwell migration and invasion assays of GIST‐882 cells under CD8+ T cell co‐culture treated with MYBL1 knockdown, anti‐PD‐1 monotherapy or their combination. (D) Live/Dead cell staining of GIST‐882 cells under CD8+ T cell co‐culture treated with MYBL1 knockdown, anti‐PD‐1 monotherapy or their combination (green: live cells; red: dead cells). (E) Representative images and quantitative analysis of colony formation by GIST‐882 cells under CD8+ T cell co‐culture treated with AIF1L overexpression, anti‐PD‐1 monotherapy or their combination. (F) CCK‐8 assay assessing proliferation of GIST‐882 cells under CD8+ T cell co‐culture treated with AIF1L overexpression, anti‐PD‐1 monotherapy or their combination. (G) Representative images and quantitative analysis of Transwell migration and invasion assays of GIST‐882 cells under CD8+ T cell co‐culture treated with AIF1L overexpression, anti‐PD‐1 monotherapy or their combination. (H) Live/Dead cell staining of GIST‐882 cells under CD8+ T cell co‐culture treated with AIF1L overexpression, anti‐PD‐1 monotherapy or their combination (green: live cells; red: dead cells).
Discussion
4
As a relatively common mesenchymal neoplasm of the gastrointestinal tract, gastrointestinal stromal tumour (GIST) has established targeted therapy combined with surgery as the cornerstone of its treatment. Over the past few decades, the emergence of tyrosine kinase inhibitors, notably imatinib, has transformed the management of recurrent and unresectable GIST [20]. Nevertheless, the efficacy of TKIs remains substantially limited by the development of primary or secondary resistance [21], highlighting the urgent need for novel therapeutic targets and agents to improve outcomes in malignant GIST. Previous studies have confirmed the rich immune landscape of GIST, demonstrating that the infiltration of immune cells—particularly T lymphocytes—is significantly associated with tumour development, progression, patient survival and prognosis [22], Moreover, CD8^+^ T cells, as central players in tumour immunotherapy, exhibit PD‐1/PD‐L1 expression levels closely linked to the effectiveness of immune checkpoint blockade [23, 24]. In this study, we integrated single‐cell transcriptomics, bulk transcriptomics and Mendelian randomization analysis to explore the immune infiltration landscape in primary and metastatic GIST. We identified causal relationships between various immune cell subtypes and GIST, along with potential mediating factors. Furthermore, we pinpointed two genes, MYBL1 and AIF1L, which show gradient expression changes across metastatic GIST, primary GIST and peritumoral normal tissues, and are correlated with PD‐1 expression. Subsequent functional validation through cellular experiments provided further insights into their roles, offering important implications for future immunotherapy research in GIST.
We identified 18 immune cell subsets with a causal association with gastrointestinal stromal tumour (GIST) development using Mendelian randomization, and most of these 18 immune cell subsets belong to classic subgroups. Among T cell subsets, CD4 and CD25 are classic markers for regulatory T cells, with well‐defined immunosuppressive functions [25]; however, single‐cell analysis revealed that the abundance of these cells gradually increases in peritumoral tissue, primary lesions and metastatic lesions, consistent with the results of Mendelian randomization analysis. HLA‐DR + CD8+ T cells and CD45RA + CD8+ T cells represent different differentiation stages of CD8+ T cells, respectively [26, 27], and have tumour‐killing effects. Single‐cell analysis showed that the abundance of these cells gradually decreases in peritumoral tissue, primary lesions and metastatic lesions, corresponding to the findings of Mendelian randomization analysis. CCR7 is a classic marker on naive CD4+ T cells, and Th0 is a transient, elusive activated state of these cells [28], Mendelian randomization analysis indicated that naive CD4+ T cells are a protective factor against GIST, while the abundance of Th0 cells gradually decreases in peritumoral tissue, primary lesions and metastatic lesions, which precisely reflects the process in which ‘immune pioneers’ are gradually disrupted, leading to immune escape, as normal Cajal interstitial cells develop into GIST and eventually metastasize. For TCRγδ T cells, our single‐cell analysis did not identify a subset that clearly corresponds to them, but Mendelian randomization analysis suggested that the infiltration of these cells is a risk factor for GIST, providing a direction for future research. In B cell subsets, plasma cells are the terminal stage of B cell differentiation [29], and both IgD‐CD38+ B cells and IgD + CD24‐ B cells are important precursors of plasma cells [30], as functionally diverse tumour‐associated cells, B cells can increase GIST risk by producing immunosuppressive antibodies or activating pro‐tumour immune cells, and Mendelian randomization analysis identified these three subsets as protective factors against GIST, consistent with the results of single‐cell analysis. CD24 + CD27+ B cells are memory B cells with immunomodulatory effects; single‐cell analysis did not identify a subset corresponding to memory B cells, but Mendelian randomization results indicated that CD20 on memory B cells is a protective factor against GIST, so it may serve as a potential target for immunotherapy, and future studies could explore immunomodulators that activate the corresponding B cell functions as drugs for GIST treatment. In monocyte subsets, Mendelian randomization analysis revealed that monocyte count is a risk factor for GIST, and single‐cell analysis showed that the abundance of these cells gradually increases in peritumoral tissue, primary lesions and metastatic lesions; notably, the CD16+ monocyte subset, which is associated with immune surveillance and antigen presentation, is a protective factor against GIST, and increasing the proportion of this subset among monocytes may be a direction for future therapeutic strategies [31, 32]; CD66b is a typical marker for neutrophils, and CD66b++ myeloid cells represent neutrophils; both Mendelian randomization and single‐cell analysis consistently indicated that these cells are a risk factor for GIST, with their abundance gradually increasing in peritumoral tissue, primary lesions and metastatic lesions. Recent studies have shown that neutrophils in the tumour microenvironment are not mere ‘bystanders’ but active ‘accomplices’ [33]. They release specific substances to provide a fertile ‘soil’ for epithelial cells carrying oncogenic mutations, thereby promoting tumour survival, proliferation and malignant progression, so neutrophils in GIST may also play an important role in immunotherapy. Although other subsets did not correspond to those identified in single‐cell analysis, they still provide insights for GIST immunotherapy. In addition, our Mendelian randomization analysis identified four plasma metabolites (4‐allylphenol sulphate, Glycolithocholate sulphate, Pentose acid, X‐21742) that exert mediating effects, which also provide a theoretical basis for more precise immunotherapy to a certain extent. In summary, we have delineated the process in which pro‐tumour immune cells gradually increase and a suppressive immune microenvironment is progressively established during the transition from peritumoral tissue to primary GIST and ultimately to metastatic GIST.
MYBL1, a member of the MYB transcription factor family, functions as a DNA‐binding protein primarily localised in the nucleus. It plays crucial roles in transcriptional regulation, cell cycle progression and proliferation, and has been established as a proto‐oncogene involved in the pathogenesis of various malignancies [34]; AIF1L, a calcium‐binding protein, is known to participate in organelle formation and sperm acrosome development. Recent investigations have revealed its involvement in breast, thyroid and colorectal cancers [35, 36, 37], with demonstrated associations with activated CD8^+^ T cells and participation in multiple immune‐related pathways. Through bulk transcriptomic analysis, we identified genes showing progressive expression changes across the peritumoral tissue‐primary GIST‐metastatic GIST spectrum. Correlation analysis with PD‐1 expression revealed that MYBL1 demonstrated progressively increasing expression along this disease continuum, showing positive correlation with PD‐1 levels. Conversely, AIF1L exhibited a gradual decrease in expression with an inverse correlation to PD‐1. Given that PD‐1 represents the most established marker of T cell exhaustion, we hypothesised that both MYBL1 and AIF1L might be involved in T cell exhaustion and anti‐tumour immunity. Functional validation using the GIST‐882 cell line demonstrated that under mono‐culture conditions, MYBL1 knockdown or AIF1L overexpression only modestly reduced proliferative, migratory and invasive capabilities. However, in co‐culture with CD8^+^ T cells, these inhibitory effects were substantially enhanced. This indicates that the suppressive effects of AIF1L overexpression and MYBL1 knockdown on GIST malignant phenotypes are highly dependent on the CD8^+^ T cell‐mediated anti‐tumour immune microenvironment. Subsequently, we conducted co‐culture experiments using GFP‐tagged MYBL1 and AIF1L overexpressing cell lines with CD8^+^ T cells. Notably, although MYBL1 is a nuclear transcription factor and AIF1L is not a secreted protein, we observed that both proteins were effectively taken up by co‐cultured T cells. This suggests that these proteins might be exported through alternative pathways such as exosomal transport [38, 39], a mechanism worthy of further investigation. Finally, we demonstrated that under CD8^+^ T cell co‐culture conditions, the combination of anti‐PD‐1 treatment with either MYBL1 knockdown or AIF1L overexpression produced superior inhibition of GIST malignant phenotypes compared to monotherapies. Currently, mono‐immunotherapy has shown limited benefits in GIST treatment, with several immune checkpoint inhibitor trials failing to achieve breakthrough clinical efficacy. Based on these findings, we propose a potential novel immunotherapeutic strategy for GIST: targeting MYBL1 or AIF1L to enhance tumour sensitivity to PD‐1 inhibitor therapy. The limited efficacy of immune checkpoint inhibitors in GIST may be attributed to its unique tumour microenvironment characteristics and relatively low immune cell infiltration. Our research suggests that MYBL1 upregulation or AIF1L downregulation may contribute to a microenvironment conducive to T cell exhaustion and immune escape, thereby attenuating the efficacy of PD‐1/PD‐L1 blockade.
This study represents the first integrated analysis combining Mendelian randomization, single‐cell transcriptomics and bulk transcriptomics to investigate the immune microenvironment and potential therapeutic targets in GIST, with experimental validation providing new insights for GIST immunotherapy. However, several limitations should be acknowledged. First, the Mendelian randomization analysis relied primarily on GWAS summary statistics from European populations. The generalisability of these genetic findings requires validation across diverse ethnic groups, which may limit the broad applicability of our conclusions; moreover, due to the low incidence of gastrointestinal stromal tumours, the available data are limited by a relatively small sample size. This resulted in no significant findings after multiple testing correction, which reflects a lack of statistical power rather than the absence of a causal relationship. Therefore, we adopted multiple complementary research methods and performed sensitivity analyses for validation. In the future, we look forward to the availability of larger‐scale databases to further support our research. Second, due to the relatively limited sample size in this study, the confidence intervals for the mediation effect estimates are wide. Although some pathways show a certain proportion of mediation, the precision of the effect sizes is relatively low, which may affect the reliability of statistical inference. In the future, larger‐scale prospective studies or experimental validation are needed to further confirm the robustness and biological significance of these mediation pathways. Third, both the single‐cell and bulk transcriptomic data were obtained from public databases. Although we integrated multiple datasets to enhance robustness, the constrained sample size and lack of treatment‐response data limited our ability to establish deeper correlations between molecular findings and long‐term patient outcomes or therapeutic responses. Finally, while we confirmed the uptake of MYBL1 and AIF1L by CD8^+^ T cells using the split‐GFP system and observed synergistic effects when combined with PD‐1 inhibitors, several fundamental questions remain unanswered. The mechanisms underlying protein delivery, the intracellular fate of these internalised proteins in T cells, and their direct signalling pathways remain to be elucidated. The precise immunoregulatory mechanisms—whether through direct modulation of T cell function or indirect pathways such as antigen presentation—also require further investigation in future mechanistic studies.
In conclusion, this study delineates the dynamic evolution of the immune microenvironment during GIST progression and identifies MYBL1 and AIF1L as pivotal regulators with opposing functions in modulating the PD‐1 pathway and CD8^+^ T cell activity. MYBL1 acts as a promoter while AIF1L serves as a suppressor of PD‐1 expression, and their targeting can potently enhance the efficacy of PD‐1 blockade. These findings not only deepen our understanding of GIST immunobiology but also provide a compelling rationale for a novel combination immunotherapy strategy, paving the way for overcoming resistance to current treatments.
Author Contributions
X.L. and Z.C. were responsible for the Mendelian randomization analysis and cell experiments, while J.X. was responsible for the single‐cell and bulk transcriptomic analyses. Q.T. made contributions to the conception of work and revised the manuscript. All authors had no objections to the final manuscript.
Funding
Noncommunicable Chronic Diseases‐National Science and Technology Major Project (2024ZD0520100).
Ethics Statement
The authors have nothing to report.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Figure S1: Scatter plot depicting the causal relationship between immune cells and gastrointestinal stromal tumours. Figure S2: Funnel plot depicting the causal relationship between immune cells and gastrointestinal stromal tumours. Figure S3: Leave‐one‐out plot depicting the causal relationship between immune cells and gastrointestinal stromal tumours. Figure S4: Combination Index Plot of shMYBL1 Combined with PD‐1 Inhibitor. Figure S5: Combination Index Plot of oeAIF1L Combined with PD‐1 Inhibitor. Table S1: Multivariable Mendelian Randomization Analysis of Unknown Compounds. Table S2: Multivariable Mendelian Randomization Analysis of Nucleotides and Their Derivatives. Table S3: Multivariable Mendelian Randomization Analysis of Amino Acids and Their Derivatives.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1J.‐Y. Blay , Y.‐K. Kang , T. Nishida , et al., “Gastrointestinal Stromal Tumours,” Nature Reviews Disease Primers 7, no. 1 (2021): 22.10.1038/s 41572-021-00254-533737510 · doi ↗ · pubmed ↗
- 2Y. Shinomura , K. Kinoshita , S. Tsutsui , and S. Hirota , “Pathophysiology, Diagnosis, and Treatment of Gastrointestinal Stromal Tumors,” Journal of Gastroenterology 40, no. 8 (2005): 775–780.16143881 10.1007/s 00535-005-1674-0 · doi ↗ · pubmed ↗
- 3A. Nowain , H. Bhakta , S. Pais , G. A. R. Y. Kanel , and S. U. M. I. T. A. Verma , “Gastrointestinal Stromal Tumors: Clinical Profile, Pathogenesis, Treatment Strategies and Prognosis,” Journal of Gastroenterology and Hepatology 20, no. 6 (2005): 818–824.15946127 10.1111/j.1440-1746.2005.03720.x · doi ↗ · pubmed ↗
- 4H. Joensuu , “Risk Stratification of Patients Diagnosed With Gastrointestinal Stromal Tumor,” Human Pathology 39, no. 10 (2008): 1411–1419.18774375 10.1016/j.humpath.2008.06.025 · doi ↗ · pubmed ↗
- 5C. M. Kelly , L. G. Sainz , and P. Chi , “The Management of Metastatic GIST: Current Standard and Investigational Therapeutics,” Journal of Hematology & Oncology 14, no. 1 (2021): 2.33402214 10.1186/s 13045-020-01026-6PMC 7786896 · doi ↗ · pubmed ↗
- 6M. C. Heinrich , C. L. Corless , G. D. Demetri , et al., “Kinase Mutations and Imatinib Response in Patients With Metastatic Gastrointestinal Stromal Tumor,” Journal of Clinical Oncology 41, no. 31 (2023): 4829–4836.37890277 10.1200/JCO.22.02771 · doi ↗ · pubmed ↗
- 7X. Chong , Y. Madeti , J. Cai , et al., “Recent Developments in Immunotherapy for Gastrointestinal Tract Cancers,” Journal of Hematology & Oncology 17, no. 1 (2024): 65.39123202 10.1186/s 13045-024-01578-x PMC 11316403 · doi ↗ · pubmed ↗
- 8Y. Tan , J. C. Trent , B. A. Wilky , et al., “Current Status of Immunotherapy for Gastrointestinal Stromal Tumor,” Cancer Gene Therapy 24, no. 3 (2017): 130–133.28186088 10.1038/cgt.2016.58 · doi ↗ · pubmed ↗