Tumor-Restricted CXCR4 and MIF Signaling Axis is Associated with T Cell-Mediated Macrophage Exclusion in Gastrointestinal Stromal Tumors

Abstract

Purpose

Gastrointestinal stromal tumors (GISTs), driven by KIT/PDGFRA mutations and treated with tyrosine kinase inhibitors (TKIs), remain challenged by drug resistance, suggesting that poorly understood mechanisms sustain an immunosuppressive tumor microenvironment (TME). This study aimed to investigate the cellular and molecular basis of the immune architecture in GISTs through integrated multi-omics analysis.

Patients and Methods

We integrated bulk RNA-sequencing data from 20 GISTs and 20 matched normal tissue samples (GSE225819) with single-cell RNA-sequencing data from 2 GISTs and 2 normal tissue samples (GSE162115) to dissect the cellular and molecular basis of TME. Weighted gene co-expression network analysis (WGCNA), immune deconvolution (CIBERSORTx), cell-cell communication inference (CellChat), and pseudotime trajectory analysis (Monocle 3) were employed. In vitro functional validation was performed using the GIST-T1 cell line, and clinical specimens were analyzed by RT-qPCR and flow cytometry.

Results

Transcriptomic profiling of GISTs revealed an inverse correlation between activated T-cell signatures and macrophage abundance, alongside significant transcriptional remodeling (1071 upregulated and 1211 downregulated genes). scRNA-seq further identified a tumor cell subpopulation with high C-X-C Motif Chemokine Receptor 4 (CXCR4) expression that, together with activated CD4⁺ memory T cells, produced macrophage migration inhibitory factor (MIF). Ligand-receptor analysis identified the MIF pathway as the dominant intercellular signaling network, primarily targeting macrophages, thereby offering a mechanistic basis for their reduced presence in the TME. In vitro, CXCR4 knockdown suppressed MIF mRNA expression and cell proliferation. In clinical specimens, tumor tissue CXCR4 and MIF transcripts were elevated and positively correlated with T-cell-derived MIF protein, which inversely correlated with macrophage infiltration.

Conclusion

Our study identifies a CXCR4⁺ GISTs tumor subpopulation associated with CD4⁺ T cell activation and MIF signaling amplification, which correlates with reduced macrophage infiltration. These findings nominate the CXCR4/MIF axis as a candidate therapeutic target, although further functional validation is warranted.

Introduction

Gastrointestinal stromal tumors (GISTs) are the most common mesenchymal tumors in the digestive tract, primarily driven by pathogenic mutations in the KIT or PDGFRA receptor tyrosine kinases.¹ The clinical application of tyrosine kinase inhibitors (TKIs), such as imatinib, has significantly improved patient outcomes.^2–4 However, both intrinsic and acquired resistance remain major clinical challenges, underscoring the limitations of targeting tumor cell-intrinsic oncogenic pathways alone.⁵ This has prompted a shift in research focus to the tumor microenvironment (TME), which plays a crucial regulatory role in treatment response and disease progression.^6,7 In GISTs, the immune landscape presents an intriguing paradox: immunohistochemical and transcriptomic studies have documented abundant T cell infiltration and tertiary lymphoid structures, particularly in untreated lesions, suggesting potential for immune recognition. However, these tumors typically exhibit a marked paucity of tumor-associated macrophages and respond poorly to immune checkpoint blockade. This unique configuration suggests the existence of incompletely understood mechanisms that sustain an inhibitory microenvironment.

Significant progress has been made in determining the cellular composition and some functional characteristics of the immune microenvironment of GISTs. Studies have confirmed the presence of infiltrating lymphocytes within the tumors, especially in untreated lesions, and their impact on prognosis has also been confirmed. At the same time, although the tumor-promoting roles of tumor-associated macrophages, especially M2-polarized macrophages, have been extensively characterized in other cancer types,^8–10 the regulatory networks governing the specific immune architecture of these tumors remain poorly defined.¹¹ The mechanisms underlying the discordance between abundant T cell infiltration and macrophage paucity in GISTs have not been elucidated. Previous studies have largely focused on individual cell types or used bulk profiling approaches that cannot resolve cell-type-specific transcriptional programs. Elucidating these multicellular interactions is essential for explaining the immune evasion capacity of GISTs and identifying therapeutic vulnerabilities beyond kinase inhibition.

We hypothesized that a specific tumor cell subpopulation may function as a key regulator of the local immune response by engaging T cells through specific chemokine-cytokine signaling axes, thereby establishing a state of macrophage exclusion. To systematically dissect the TME of GISTs, we combined bulk and single-cell transcriptomic analyses with in vitro functional validation. Our integrative analysis identified a distinct tumor cell subset characterized by elevated expression of C-X-C Motif Chemokine Receptor 4 (CXCR4). Pseudotime trajectory inference revealed that this subset is transcriptomically primed for T cell interaction. Furthermore, we delineated a macrophage migration inhibitory factor (MIF)-centered signaling network driven by these tumor cells and amplified by activated T cells, and computational modeling predicts this network may contribute to reduced macrophage presence in the tumor microenvironment.

Together, these findings suggest a model in which a CXCR4⁺ tumor subpopulation is associated with CD4⁺ T cell activation and MIF-mediated signaling, providing a potential explanation for the co-occurrence of T cell abundance and macrophage scarcity in GISTs. The present study offers an integrative transcriptomic framework and identifies the CXCR4/MIF axis as a candidate target for further investigation.

Materials and Methods

Bulk RNA-Sequencing Data Analysis

Publicly available bulk RNA-sequencing datasets for Gastrointestinal Stromal Tumors (GISTs) and matched normal tissues were downloaded from GEO (GSE225819), which contains 20 GISTs tumor samples and 20 matched normal tissue samples. Raw read counts were normalized and variance-stabilized using the DESeq2 R package. Principal Component Analysis (PCA) was performed on the top 500 most variable genes to assess overall transcriptomic differences. Differential expression analysis between GISTs and normal samples was conducted with DESeq2, applying an adjusted p-value (FDR) cutoff of < 0.05 and an absolute log2 fold-change threshold of > 1. Gene Set Enrichment Analysis (GSEA) and over-representation analysis (ORA) were performed using the clusterProfiler package to identify enriched biological pathways. Immune cell infiltration scores were estimated from bulk transcriptomes using the CIBERSORTx algorithm with the LM22 signature matrix. Weighted Gene Co-expression Network Analysis (WGCNA) was implemented to construct gene modules associated with clinical traits and immune cell infiltration, using a soft-thresholding power determined by scale-free topology criterion.

Single-Cell RNA-Sequencing Data Processing and Analysis

Single-cell RNA-seq data for GISTs samples were obtained from GEO (GSE162115), which contains 2 GISTs tumor samples and 2 normal tissue samples. Treatment status and mutation profiles (KIT/PDGFRA) were not available for these samples. Raw gene expression matrices were processed using the Seurat R package (version V4). Cells with a low number of detected genes (< 200), high mitochondrial gene percentage (> 20%), or high hemoglobin gene expression were filtered out. Data were normalized using the LogNormalize method, and the top 2000 highly variable genes were identified for downstream analysis. Dimensionality reduction was performed using Principal Component Analysis (PCA), followed by graph-based clustering and visualization in two dimensions with Uniform Manifold Approximation and Projection (UMAP). As samples were processed on the same 10x Genomics platform with a consistent protocol and UMAP visualization showed good inter-sample mixing after standard preprocessing, no additional batch effect correction was applied. Cell clusters were annotated into major cell types (tumor cells, T cells, macrophages, etc.) based on the expression of well-established canonical marker genes. To explore tumor cell heterogeneity, tumor cells were subsetted and re-clustered following the same workflow.

Trajectory Inference and Cell-Cell Communication

Pseudotime trajectory analysis on the re-clustered tumor cells was performed using the Monocle 3 R package to infer potential transcriptional relationships among tumor cell subpopulations. Pseudotime trajectories represent computationally inferred relationships, not experimentally validated developmental lineages. Branched expression analysis modeling (BEAM) was used to identify genes differentially expressed at key branch points. Cell-cell communication analysis was performed using the CellChat R package, which models the probability of ligand-receptor interactions based on scRNA-seq expression data. Significantly over-represented signaling pathways (eg, MIF, MDK) were identified, and the communication networks were visualized to depict predicted signaling flows between cell types.

Integration of Bulk and Single-Cell Analyses

WGCNA was performed to identify gene modules associated with immune phenotypes in GISTs. A signed co-expression network was constructed using the normalized expression matrix from the 20 GISTs tumor samples in GSE225819. The soft-thresholding power was set to 12 to achieve a scale-free topology. Modules were detected using the dynamic tree-cut algorithm with a minimum module size of 60 genes. Modules with eigengene correlation > 0.75 were merged. Module eigengenes were correlated with immune cell infiltration scores (from CIBERSORTx) to identify immune-related modules. To integrate bulk and single-cell analyses, genes within key modules (eg, turquoise) that showed high correlation with immune phenotypes were intersected with differentially expressed genes from pseudotime branch points—genes co-expressed with immune phenotypes at the bulk level and dynamically regulated during tumor cell state transitions at the single-cell level are more likely to be functionally relevant. Overlapping genes were extracted for downstream functional enrichment analysis using clusterProfiler.

Cell Line

Human embryonic lung fibroblast cell line WI-38 (STM-CL-5348) and human gastrointestinal stromal tumor cell line GIST-T1 (STM-CL-5520) were purchased from TST Biotechnology Co., Ltd (Shanghai, China). GIST-T1 was cultured in high glucose Dulbecco’s Modified Eagle’s Medium (DMEM) (Gibco) supplemented with 10% fetal calf serum (FCS, Hycyte), 100 U/mL penicillin and 100 ug/mL streptomycin (Gibco) at 37°C and 5% CO2. WI-38 was cultured in Minimum Essential Medium (MEM) (Gibco) supplemented with 10% fetal calf serum (FCS, Hycyte), 100 U/mL penicillin and 100 ug/mL streptomycin (Gibco) at 37°C and 5% CO2.

RT-qPCR

Total RNA from cell lines and sample tissues was isolated according to the following procedure: resuspended the cell pellet with 1 mL of Trizol and incubated on ice for 5 min; added 0.2 mL of chloroform, mixed thoroughly and incubated at room temperature for 5 min; Centrifuged at 4°C, 13000rpm for 15 min and retained the upper layer; added 0.4 mL of isopropanol, invert to mix; centrifuge at 4°C, 13000rpm for 10 min, discard the supernatant and wash twice with 75% ethanol. RNA concentration and purity were assessed using a NanoDrop spectrophotometer. All samples had an A260/A280 ratio between 1.8 and 2.0. Reverse transcription was performed using using Hifair^® III 1st Strand cDNA Synthesis SuperMix for qPCR following kit instructions. cDNA was analyzed by Hieff^® qPCR SYBR^® Green Master Mix real time using CFX Connect Real-Time PCR Detection System. Each reaction was run in triplicate and three independent biological replicates were performed. mRNA expression was normalized to Gapdh using the formula (ΔCt = Cttarget gene - CtGapdh) and represented as relative mRNA expression (2-(ΔCtsample - ΔCtcontrol)). Primer sequences for all genes are listed in Table 1.

Table 1.

Primer Sequences for RT-qPCR

Gene	Sense (5ʹ→3ʹ)	Antisense (5ʹ→3ʹ)
CXCR4	GGGCAGAGGAGTTAGCCAAG	CCACCTTTTCAGCCAACAGC
MIF	GTGGTGTCCGAGAAGTCAGG	TTGCTGTAGGAGCGGTTCTG

Cell Transfection

GIST-T1 cells in exponential growth were plated in six-well plates at a density of 2×105 cells per well and allowed to attach overnight until reaching ~50% confluence. Transfection was performed using GP-transfect-Mate as directed. Two separate 200 μL aliquots of serum-free DMEM were prepared in microcentrifuge tubes, one receiving 7.5 μL siRNA and the other 6 μL transfection reagent. Each was mixed by gentle pipetting, incubated at room temperature for 5 min, pooled into a single tube, and incubated for an additional 15 min to allow complex formation. After washing the cells with PBS, complete DMEM was added per well, followed by the dropwise addition of the entire complex mixture. Cells were harvested at 48 h post-transfection for RNA extraction and RT-qPCR analysis. The sequences of the siRNA are provided in Table 2.

Table 2.

Sequences of the Si-RNA

Category	Sense (5ʹ→3ʹ)	Antisense (5ʹ→3ʹ)
si-CXCR4	GCAAGAUCAUCACCGUGAU	AUCACGGUGAUGAUCUUGC
si-NC	UUUUCCGAACGUGUCACGUTT	ACGUGACACGUUCGGAGAATT

MTT Assay

Cell viability was determined using the MTT assay. Briefly, GIST-T1 cells transfected with si-NC or si-CXCR4 were seeded in 96-well plates at a density of 5×10³ cells per well and allowed to adhere overnight. 10 µL of MTT solution (5 mg/mL) was added to each well and the plates were incubated for 4 hours at 37°C. The resulting formazan crystals were then dissolved by adding 100 µL of dimethyl sulfoxide (DMSO). The absorbance was measured at 490 nm using a microplate reader at 0, 24 and 48 hour.

Flow Cytometry

Fresh tumor tissues from patients with GISTs were processed into single-cell suspensions (n = 9). Cells were incubated with Fc receptor blocking reagent (Invitrogen, 14-9161-73) for 15 min at 4°C to prevent non-specific antibody binding, and then stained with fluorochrome-conjugated anti-human antibodies (dilution at 1:100) for 30 min at 4°C in the dark. The following antibodies were used: 7-AAD (Invitrogen, 00-6993-50), CD45-PE (Invitrogen, 12-0459-42), CD4-APC-A750 (Invitrogen, 47-0049-42), CD8-PE-CF594 (Elabsciecne, AN00427P), MIF (Invitrogen, A700-141-T), AF700-conjugated secondary antibody (Invitrogen, A-21038), CD14-PE (Elabscience, E-AB-F1209D), CD11b-FITC (Elabscience, E-AB-F1081C), CD45-APC-A750 (Invitrogen, 47-0459-42). Fluorescence compensation was performed using single stained UltraComp Beads (Invitrogen, 01-2222-42) and compensation matrices were calculated and applied. Data were acquired on Beckman CytoFLEX LX Flow Cytometer and analyzed by Flowjo software (version 10.10.0). Single cells were gated based on forward scatter area (FSC-A) versus forward scatter height (FSC-H) to exclude doublets. Live cells were identified as 7-AAD negative. CD45⁺ leukocytes were selected within live cells. CD4⁺ and CD8⁺ T cells were identified within the CD45⁺ gate respectively. MIF expression was assessed in each T cell subset. Macrophage were identified as CD45⁺CD14⁺CD11b⁺ cells.

Statistical Analysis

Statistical analyses were performed in R (version 4.2.1) and GraphPad software (version 10.4.2). For comparisons between two groups, Student’s t-test (for normally distributed data) or Wilcoxon rank-sum test (for non-parametric data) was used. For correlations, Pearson’s or Spearman correlation coefficient was calculated as appropriate. A p-value of < 0.05 was considered statistically significant. All data visualizations were generated using ggplot2 and related R packages.

Results

Computational Deconvolution Reveals Reduced Macrophage Infiltration Scores in GISTs, Which Correlate with Activated Memory T Cell Signatures

To elucidate the transcriptomic landscape of GISTs, we analyzed bulk RNA-sequencing data from 20 GISTs and 20 matched normal tissue samples (GSE225819). Principal Component Analysis (PCA) demonstrated a pronounced divergence in the global mRNA expression profiles between GISTs specimens and normal tissue controls (Figure 1A). Systematic differential gene expression analysis identified 1071 significantly upregulated and 1211 significantly downregulated genes in GISTs (|log₂FC| > 1, adjusted p < 0.05; Figure 1B). Hierarchical clustering of the most highly upregulated transcripts, including LY6H, GDF10, SFRP1, BCHE, and NPR3, further delineated the GISTs-specific expression landscape (Figure 1C). Notably, LY6H (Lymphocyte Antigen 6 Family Member H) is a gene located within the MHC class III locus on chromosome 6 and has been implicated in promoting tumor proliferation, metastasis and therapeutic resistance in multiple cancer types.¹² This body of evidence prompted our investigation into lymphoid involvement in GISTs pathogenesis.

Figure 1.

Bulk transcriptomic profiling reveals an immune-activated transcriptional landscape with reduced macrophage infiltration scores in GISTs. (A) PCA of mRNA expression profiles shows clear separation between GISTs tumors (n = 20) and normal tissue samples (n = 20), the ellipses indicate the 95% confidence intervals for each group. (B) Volcano plot of DEGs (|log₂FC| > 1, adjusted p < 0.05). Genes without a significant difference are indicated as “NS”. (C) Heatmap of top upregulated genes. (D) Immune cell infiltration scores by CIBERSORTx showing reduced macrophage scores and elevated T cell association in GISTs. (E) WGCNA module–immune correlation heatmap. (F) GO enrichment of macrophage-associated WGCNA modules. (G) Unsupervised clustering of GISTs samples. (H) Functional enrichment of the turquoise module.

To characterize the immune microenvironment, we estimated immune cell infiltration scores using CIBERSORTx deconvolution. Macrophage scores were substantially lower in GISTs compared to normal tissues, whereas T cell signatures showed a strong positive association with the GISTs cohort (Figure 1D). These deconvolution-based estimates reflect relative rather than absolute cell fractions and have not been validated histologically. To unravel the co-regulation networks underlying this phenotype, we performed Weighted Gene Co-expression Network Analysis (WGCNA). This identified key modules (MEturquoise, MEtan, MEgreenyellow) exhibiting high correlation with macrophage abundance, which also showed significant overlap with signatures of activated CD4⁺ T cells (Figure 1E), suggesting co-regulated transcription, though this does not establish direct functional interaction. Functional enrichment analysis of these modules consistently revealed a significant overrepresentation of gene sets related to T cell activation and immune response (Figure 1F). Unsupervised clustering of GISTs samples based on transcriptional similarity, followed by WGCNA, pinpointed the turquoise module as a hallmark of a distinct GISTs subgroup (Cluster 1) (Figure 1G). This module was functionally enriched for T cell-mediated immunity, representing a concerted T cell response signature within the GISTs microenvironment (Figure 1H). These computational analyses indicate an inverse relationship between an active T cell transcriptional state and reduced macrophage infiltration scores in GISTs.

Single-Cell Analysis Identifies the MIF Pathway as a Candidate Mediator of Reduced Macrophage Presence in GISTs

To mechanistically dissect the interplay between T cells and macrophages, we turned to single-cell resolution. Initial correlation analysis confirmed a strong positive relationship between macrophage and CD4⁺ T cell (both resting and activated) frequencies (Figure 2A), suggesting a potential functional relationship. We then analyzed single-cell RNA sequencing (scRNA-seq) data from GISTs patients. Dimensionality reduction, clustering, and canonical marker-based annotation resolved the major cellular constituents of the GISTs ecosystem (Figure 2B and C). Samples from the tumor core and invasive margin, which displayed congruent profiles, were analyzed jointly. Systematic interrogation of intercellular signaling using CellChat identified the MIF and MDK pathways as the most prominent communication networks, predominantly orchestrated by tumor cells and T cells to engage other immune compartments (Figure 2D). Notably, the MIF pathway emerged as the top-ranked signaling axis in GISTs (Figure 2E). MIF is a pleiotropic cytokine with reported roles in cancer-associated inflammation and cachexia.¹³ Our computational analysis predicts that tumor- and T-cell-derived MIF primarily targets macrophages (Figure 2F). We therefore hypothesize that locally elevated MIF signaling contributes to macrophage exclusion in GISTs. Direct macrophage migration assays will be required to functionally validate this hypothesis.¹⁴ Ligand-receptor analysis demonstrated the highest MIF ligand expression in tumor cells, while cognate receptors (CD74, CXCR4, CD44) were abundantly expressed on T cells, macrophages, and other immune cells (Figure 2F). Quantitative analysis further predicted macrophages as the primary targets of MIF signaling within the TME (Figure 2G). These single-cell analyses identify tumor and T-cell-derived MIF signaling as a candidate mechanism that may contribute to reduced macrophage presence in GISTs.

Figure 2.

Single-cell analysis identifies MIF as a dominant signaling pathway associated with reduced macrophage presence in GISTs. (A) Immune cell correlation network. (B) UMAP of scRNA-seq data. (C) Cell type annotation by canonical markers. (D) CellChat analysis showing MIF and MDK as top signaling pathways from tumor cells and T cells. (E) Signaling pathway ranking. (F) MIF ligand and receptor expression across cell types. (G) Ligand–receptor interaction scores predicting macrophages as MIF targets.

Pseudotime Trajectory Analysis Identifies a CXCR4⁺ Tumor Subpopulation Transcriptionally Associated with T Cell Engagement

We next sought to identify the tumor-intrinsic programs that engage this immune axis. Re-clustering of malignant cells from the scRNA-seq data defined five transcriptionally distinct tumor subpopulations (Figure 3A). To explore potential transcriptional relationships among these subpopulations (clusters 0–5), we applied pseudotime trajectory inference, a computational method that orders cells along a continuum based on transcriptomic similarity (Figure 3A). Branch point analysis identified differentially expressed genes associated with distinct predicted transcriptional states (Figure 3B and C). Branch Node 1 defined a divergence from a transcriptional state enriched for immune-negative regulation toward two trajectories: one associated with antigen presentation functions and another associated with protein metabolism. Branch Node 2, originating from the latter branch, further diverged into an antigen presentation-associated state and a distinct state characterized by T cell activation and chemotaxis gene signatures (Figure 3D and E). This inferred topology suggested that Clusters 1 and 4 converge on two predicted terminal states: Clusters 0 and 5 shared antigen-presenting capacity, but only Cluster 5 was uniquely associated with T-cell-recruiting and activating transcriptional programs. To molecularly anchor this functional prediction, we intersected the genes from the T cell/macrophage-correlated turquoise WGCNA module with genes from the pseudotime branch nodes. The turquoise module showed the most extensive overlap with Nodes 1 and 2 (95 and 124 genes, respectively) (Figure 3F). Deconstructing the nodes into pre-branch, cell fate 1, and cell fate 2 states revealed that the overlapping genes were overwhelmingly concentrated in cell fate 2—the predicted state associated with T-cell-attracting transcriptional programs (Figure 3G). Concordantly, pseudotime mapping localized this cell fate 2 predominantly within Cluster 5. Interrogation of MIF pathway components specifically identified high expression of the receptor CXCR4 as a defining feature of Cluster 5 (Figure 3H and I). These integrative analyses identify a CXCR4⁺ tumor cell subpopulation transcriptionally associated with T cell engagement within the GISTs microenvironment.

Figure 3.

Pseudotime analysis identifies a CXCR4⁺ tumor subpopulation transcriptionally associated with T cell engagement. (A) UMAP of tumor cell subclustering (Clusters 0–5). (B) Pseudotime trajectory overlaid on UMAP. The dotted arrows indicate the predicted differentiation lineages along the pseudotime trajectory, starting from the root state to the differentiated terminal states. (C) Differentially expressed genes at Branch Nodes 1 and 2. (D and E) Functional enrichment of branch node genes showing transition toward antigen presentation and T cell–attracting programs. (F) Venn diagrams of turquoise module and branch node gene overlap. (G) Intersection with trajectory states (pre-branch, cell fate 1, cell fate 2). (H) Violin plot of CXCR4 in Cluster 5. (I) Feature plot of CXCR4 across clusters.

CD4⁺ Activated Memory T Cells are Identified as Major Contributors to MIF Signaling Among Lymphocytes

To define the role of T cells as potential signal amplifiers in this circuit, we performed detailed characterization of the T cell compartment. Subclustering of T cells from the scRNA-seq data identified major subsets: CD4⁺ naïve/resting memory, CD4⁺ activated memory, regulatory T cells (Tregs), CD8⁺ T cells, and γδ T cells (Figure 4A and B). All major T cell subsets uniformly expressed high levels of both ligands and receptors comprising the MIF pathway (Figure 4C). Cell-cell communication analysis focused on MIF signaling pinpointed CD4⁺ activated memory T cells as the most interactively active T cell subset, serving as the major contributors to MIF-mediated signaling among lymphocytes (Figure 4D). This was reinforced by evidence of elevated transcriptional activity of genes central to MIF signaling, such as CXCR4 and CD74 (Figure 4E). Based on these analyses, we propose that CD4⁺ activated memory T cells serve as important contributors to MIF signal amplification, consistent with a model in which T-cell-derived MIF participates in regulating macrophage presence. Functional co-culture experiments would be needed to confirm this inference.

Figure 4.

CD4⁺ activated memory T cells are identified as major contributors to MIF signaling among lymphocytes. (A) UMAP of T cell subsets from intra- tumor tissue (I) or peri- tumor tissue (P). (B) Canonical subset markers. (C) MIF ligand and receptor expression. (D) MIF signaling contributions by subset (CellChat). (E) Transcriptional activity of CXCR4 and CD74 across subsets.

Functional Validation of the CXCR4/MIF Axis in vitro and in Clinical Specimens

To dissect the CXCR4-MIF axis, we first examined its role in GISTs cells. The GIST-T1 cell line showed significant overexpression of CXCR4 and MIF mRNA compared to control cells (Figure 5A). Knockdown of CXCR4 robustly diminished MIF expression (Figure 5B) and compromised cellular proliferation (Figure 5C), suggesting that CXCR4 sustains MIF expression and supports GISTs cell growth. Consistent with this, clinical samples confirmed upregulation of CXCR4 and MIF in tumor (T) versus normal adjacent tissues (NAT) (Figure 5D), and the detailed clinicopathological characteristics of the GISTs patients are summarized in Table 3.

Figure 5.

Association of GISTs CXCR4 with T cell MIF and macrophage infiltration. (A) The mRNA expression of CXCR4 and MIF in WI-38 and GIST-T1 assessed using RT-qPCR. (B) The transfection efficiency of CXCR4 into GIST-T1 (left) and mRNA expression of MIF in GIST-T1 after transfection (right) assessed using RT-qPCR. (C) MTT assay of GIST-T1 after transfection. (D) The mRNA expression of CXCR4 and MIF in normal adjacent to tumor (NAT) and GSIT tumor (T) evaluated by RT-qPCR. (E) The proportion of CD4⁺ and CD8⁺ T cells in GISTs assessed by flow cytometry. (F) The correlation between mRNA expression of CXCR4 and the proportion of CD4⁺ and CD8⁺ T cells in GISTs tumors. (G) The proportion of macrophage in GISTs assessed by flow cytometry. (H) The correlation between the proportion of macrophage and the proportion of CD4⁺ and CD8⁺ T cells in GISTs tumors. *p < 0.05, **p < 0.01, ***p < 0.001, ****p<0.0001.

Table 3.

Clinicopathological Characteristics of the GISTs Patients

Patient ID	Age (Years)	Sex	Tumor Location	Histological Subtype	NIK Risk Category	Treatment Status
GIST-01	62	M	Stomach	Spindle cell	Low	Treatment-naïve
GIST-02	58	F	Small intestine	Spindle cell	High	Treatment-naïve
GIST-03	71	M	Stomach	Epithelioid	Intermediate	Treatment-naïve
GIST-04	55	F	Stomach	Spindle cell	Low	Treatment-naïve
GIST-05	67	M	Small intestine	Spindle cell	Low	Treatment-naïve
GIST-06	48	M	Stomach	Spindle cell	Intermediate	Imatinib-resistant
GIST-07	74	F	Small intestine	Spindle cell	Low	Treatment-naïve
GIST-08	63	M	Stomach	Spindle cell	Low	Treatment-naïve
GIST-09	69	F	Stomach	Mixed	High	Imatinib-resistant

We next asked whether the immune compartment also contributes to the MIF pool within the tumor microenvironment. Flow cytometric profiling further demonstrated tumor-infiltrating CD4⁺ and CD8⁺ T cells express MIF protein, identifying T cells as an additional potential source beyond the tumor cells. (Figure 5E). To explore potential crosstalk between tumoral CXCR4 and T cell-derived MIF, we correlated tumor tissue CXCR4 mRNA levels (as a surrogate for tumor cell CXCR4) with MIF expression in T cells. A significant positive correlation was observed (Figure 5F), suggesting that higher tumoral CXCR4 expression is associated with elevated MIF production by T cells, although the cellular origin of the CXCR4 signal cannot be definitively assigned from this bulk analysis.

Crucially, the level of MIF in these T cells was inversely correlated with macrophage infiltration (Figure 5G and H) These findings suggest a model in which MIF within the GISTs microenvironment, supplied by tumor cells and T cells, may participate in restricting macrophage accumulation.

Discussion

Our study identifies a CXCR4-MIF-centered intercellular communication network associated with the distinctive immune architecture of GISTs, particularly the co-occurrence of T cell infiltration and reduced macrophage presence. Through integrative multi-omics analysis, we identify a CXCR4⁺ tumor subpopulation transcriptionally associated with T cell engagement and MIF-mediated signaling. Our data suggests a model in which tumor-derived MIF, amplified by activated CD4⁺ T cells, may contribute to limiting macrophage accumulation in the TME. This model provides a mechanistic framework for understanding the paradoxical immune landscape of GISTs, though direct functional validation, including macrophage migration assays and in vivo perturbation, is required to establish causality.

Our findings underscore the necessity of a systemic perspective that integrates tumor-intrinsic programs with multicellular crosstalk to decode cancer complexity. The CXCR4/MIF axis is a pharmacologically accessible target: CXCR4 inhibitors (eg, AMD3100/plerixafor) and MIF inhibitors (eg, ISO-1) are clinically available or under development.^15–17 In our in vitro experiments, CXCR4 knockdown reduced MIF expression and cell proliferation in GIST-T1 cells, providing preliminary evidence for direct anti-tumor effects. CXCR4/MIF inhibition could offer a dual benefit: impairing tumor cell proliferation while modulating the immune microenvironment. This strategy may be particularly relevant in TKI-resistant disease, where CXCR4 upregulation has been linked to drug resistance. Beyond the current findings, pharmacological interrogation of the CXCR4/MIF axis in TKI-resistant GISTs models will be essential to translate these observations into therapeutic strategies.¹⁸ Concurrently, the evaluation of T cell-derived MIF as a correlative biomarker in longitudinal patient cohorts may further solidify its clinical relevance.

Among the key questions that remain are the plasticity of the CXCR4⁺ subpopulation and whether disrupting the CXCR4/MIF axis can restore macrophage infiltration. MIF-mediated immune modulation has been implicated in immune evasion across diverse tumor types, including breast cancer, lung squamous cell carcinoma, and endometrial cancer,^19–22 suggesting a conserved mechanism. However, macrophage biology can differ substantially depending on tumor context, and whether this mechanism operates similarly in other mesenchymal tumors remains to be investigated. Future studies incorporating in vivo perturbation models, spatially resolved approaches, and functional macrophage migration assays will be essential to establish the causal role of this axis and explore its therapeutic potential.

This study has several limitations. First, most mechanistic inferences—including macrophage exclusion and cell-cell communication—are derived from computational analyses (transcriptomic deconvolution, CellChat modeling, and pseudotime inference) and do not establish direct causality; macrophage reduction in GISTs has not been validated by immunohistochemistry. Second, functional validation was performed in a single GISTs cell line (GIST-T1); validation in additional lines with different mutational backgrounds would strengthen generalizability. Third, the scRNA-seq analysis was based on one public dataset (GSE162115, 2 tumor and 2 normal samples) without treatment or mutation information; independent cohort validation is needed. Fourth, macrophage migration assays, pharmacological inhibition, and in vivo perturbation experiments were not performed. Fifth, the clinical cohort was limited in size and lacked longitudinal follow-up. Finally, spatial validation (eg, multiplex IHC or spatial transcriptomics) was not available. Despite these limitations, the integrative multi-omics framework and candidate mechanism identified here provide a foundation for future mechanistic and translational studies.

Conclusion

In summary, our study identifies a CXCR4⁺ GISTs tumor subpopulation that is transcriptionally activates associated with CD4⁺ T cell activation and MIF signaling amplification, which correlates with reduced macrophage infiltration in the tumor microenvironment. These findings nominate the CXCR4/MIF axis as a candidate therapeutic target. Future studies incorporating in vivo models, pharmacological inhibition, and macrophage functional assays will be needed to validate the causal role of this axis. This study provides a conceptual framework for how tumor-intrinsic transcriptional programs may shape the immune landscape of GISTs.

Funding Statement

The authors declared that financial support was received for this work and its publication. The research reported in this project was generously supported by Hebei Provincial Medical Scientific Research Project (Project number: 20221609).

Abbreviations

GISTs, Gastrointestinal stromal tumors; TKIs, tyrosine kinase inhibitors; TME, tumor microenvironment; CXCR4, C-X-C Motif Chemokine Receptor 4; MIF, migration inhibitory factor; WGCNA, Weighted Gene Co-expression Network Analysis; NAT, normal adjacent tissues; scRNA-seq, single-cell RNA sequencing; PCA, Principal Component Analysis; UMAP, Uniform Manifold Approximation and Projection; DEGs, differentially expressed genes; Tregs, regulatory T cells.

Data Sharing Statement

The transcriptomic datasets analyzed for this study are publicly available in the Gene Expression Omnibus (GEO) repository under accession numbers GSE162115²³ (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE162115) and GSE225819²⁴ (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE225819). These datasets are cited in the Methods section where first used. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE162115. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE225819.

Ethical Approval and Consent to Participate

The studies involving humans were approved by the Medical Ethics Committee of The First Hospital of Qinhuangdao (ethics approval number: 2021C014) and were conducted in accordance with the local legislation, institutional requirements, and the Declaration of Helsinki. The participants provided their written informed consent to participate in this study.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.