IGSF9 promotes immunotherapy resistance in colon cancer by orchestrating an immunosuppressive tumor microenvironment and enables combinatorial targeting strategies.

1
Introduction
Colon cancer (CC), the third most common malignancy worldwide, poses a significant global health burden, with an estimated 2.2 million new cases and 1.1 million deaths annually [1,2]. Despite advancements in treatment, patients with advanced or metastatic CC often face poor prognosis [3], particularly those with immunosuppressive tumor microenvironments (TME) that limit the efficacy of immunotherapy [4,5].
Nonetheless, the efficacy of immunotherapy exhibits marked heterogeneity across cancer types [6]. Despite substantial breakthroughs in the field, a considerable subset of patients remains unresponsive to immunotherapeutic interventions [7,8]. This disparity underscores the critical need to dissect the biological determinants of treatment response [9]. TME has emerged as a central regulator of immunotherapy outcomes, with its complex architecture dictating the balance between anti-tumor immunity and immune evasion [10]. Dynamic interactions among tumor cells, immune cell subsets (e.g., T cells, macrophages), and stromal elements (e.g., fibroblasts, endothelial cells) within the TME orchestrate either permissive or suppressive microenvironments for immune effector function [11,12].
Immunoglobulin superfamily member 9 (IGSF9) [13], a cell surface protein involved in cell adhesion and signaling, has emerged as a potential regulator of tumor biology. Previous studies have demonstrated that IGSF9 promotes lung cancer invasion and metastasis through GSK-3β [14]. Additionally, a research team validated in mice that IGSF9 blocks the efficacy of immunotherapy by inhibiting T-cell proliferation [15]. These findings collectively indicate that IGSF9 exhibits tumor-promoting functions and influences treatment efficacy by regulating the TME.
TME in CC is highly heterogeneous, with immunosuppressive features such as reduced immune cell infiltration and elevated levels of regulatory T cells (Tregs) and cancer-associated fibroblasts (CAFs) contributing to treatment resistance [16,17]. IGSF9 has been shown to modulate TME components, including CD8+ T cells proliferation which are linked to diminished anti-tumor immune responses [15]. However, in CC, which is primarily characterized by immunosuppressive TME, how IGSF9 orchestrates TME reprogramming remains elucidated.
This study aims to characterize the role of IGSF9 in the immunosuppressive TME of CC, investigate its correlation with immunotherapy resistance, and explore potential therapeutic interventions to overcome this resistance. Our analytical approach begins with an initial pan-cancer discovery phase to establish the broader context of IGSF9 dysregulation, followed by colon cancer-focused validation, which includes single-cell transcriptomics, spatial profiling, and preclinical therapeutic experiments. By integrating multi-omics analyses, single-cell sequencing, and preclinical models, we seek to establish IGSF9 as a prognostic and predictive biomarker and identify novel combination therapies for CC patients with high IGSF9 expression.

2
Methods
2.1
Data sources
Expression profiles and clinical datasets for TCGA, GTEx, and CCLE were obtained from the UCSC XENA platform (https://xenabrowser.net/datapages/). Tumor tissue data were sourced from TCGA, whereas normal tissue annotations were supplemented by GTEx. This integrated approach effectively addressed the limitation of insufficient or missing normal tissue samples in select TCGA cohorts. For tumor cell line expression analysis, data were retrieved from the CCLE repository.

2.2
Data preprocessing and normalization
For transcriptomic data obtained from TCGA, we utilized the log2(TPM+1) normalized expression values provided by the UCSC XENA platform, which had been uniformly processed using the TOIL pipeline to ensure cross-sample comparability. For GTEx normal tissue data, the same TOIL-processed log2(TPM+1) values were used to enable direct comparison with TCGA tumor samples. When integrating TCGA and GTEx data for tumor-normal comparisons, we note that batch effects may exist despite uniform processing; therefore, we applied ComBat (from the R package, version 3.46.0) batch correction when performing differential expression analyses across datasets to minimize technical variation while preserving biological signals.
For single-cell RNA sequencing analyses, we utilized pre-processed and annotated datasets from the IMMUcan and TISCH databases, which apply standard quality control filtering (removing cells with <200 genes, >20 % mitochondrial content), SCTransform normalization, and Harmony integration for batch correction across samples.
For spatial transcriptomics data from the CROST database, spot-level expression matrices were normalized using SCTransform and spatially-aware deconvolution algorithms as implemented in the original database processing pipelines.
For the immunotherapy cohorts (GSE61676 and GSE135222), we used the processed expression matrices provided by the Gene Expression Omnibus (GEO), which were normalized using the robust multi-array average (RMA) method for microarray data (GSE61676) and DESeq2-normalized counts for RNA-seq data (GSE135222), respectively.
Drug sensitivity data from GDSC2 were analyzed using pre-computed IC50 values and cell line expression data (log2(TPM+1)) provided by the database, ensuring consistency in cross-study comparisons.

2.3
Enrichment analysis
Pearson correlation analysis was applied to explore the relationship between the target gene and all other genes across 33 different cancer types. After obtaining the correlation data, the 50 genes with the strongest positive and negative correlations were selected to create heatmaps illustrating these associations. To further characterize the biological significance, functional enrichment analyses based on Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were performed using the clusterProfiler package (version 4.14.4). In this analysis, gene sets were assembled from the top 300 genes showing the highest positive correlations (including the target gene) and used for pathway enrichment, with results displayed as bubble plots and bar graphs.
For gene set-level evaluation, Gene Set Enrichment Analysis (GSEA) was performed across all 33 cancer types using clusterProfiler, integrating annotations from GO, KEGG, and Reactome databases. In parallel, Gene Set Variation Analysis (GSVA, version 2.0.5) [18] was conducted with the updated HALLMARK gene sets (50 canonical pathways) from the Molecular Signatures Database (MsigDB) to assess pathway activity patterns. GSVA scores were generated for each tumor type, and Pearson correlation coefficients between the target gene's expression and pathway scores were calculated. These correlations were visualized in an integrated heatmap to reveal key functional relationships.

2.4
Mutation, copy number, and methylation analysis
The cBioPortal platform (http://www.cBioPortal.org) was employed to systematically explore the mutation and amplification landscapes of the target gene across various cancer types. Structural variant data were examined to uncover characteristic patterns of genetic alterations. For each cancer type, Pearson correlation analysis was conducted to evaluate the relationship between gene expression levels (log2-transformed TPM values) and copy number variation (CNV) z-scores. In parallel, promoter methylation (derived from HM450 array data available in cBioPortal) was analyzed for its correlation with gene expression using Pearson's method, given that promoter hypermethylation is typically associated with reduced transcriptional activity. The findings were presented through graphical plots, emphasizing statistically significant correlations (p < 0.05) while excluding tumor types without relevant CNV or methylation data from the analyses.

2.5
ESTIMATE analysis
The ESTIMATE package in R was applied to compute stromal scores, immune scores, overall ESTIMATE scores (calculated as the combined stromal and immune scores), and tumor purity levels within cancer tissues [19]. The associations were illustrated using various visual formats, including correlation line graphs, radar charts, lollipop plots, and heatmaps summarizing pairwise correlations, with detailed statistics provided in supplementary files for further examination.

2.6
TIMER2.0 analysis
To characterize immune cell infiltration, the Tumor Immune Estimation Resource 2.0 (TIMER2.0, http://timer.cistrome.org/) was utilized. This web-based platform integrates several immune deconvolution methods [20]. Using the “Immune Association” module, the gene symbol (IGSF9) was entered, and correlations with specified immune cell populations were visualized through the “Gene” function. A total of 21 immune cell subsets—including various T cell types, macrophages, and dendritic cells—were systematically evaluated for their infiltration levels and correlation with IGSF9 expression.

2.7
CIBERSORT and ImmuCellAI
CIBERSORT [21] and ImmuCellAI (http://bioinfo.life.hust.edu.cn/ImmuCellAI/#!/) algorithms were applied to estimate the abundance of 26 immune cell populations based on TCGA transcriptomic data. Relationships between cell infiltration levels and gene expression were displayed through heatmaps and circular correlation plots.

2.8
Gene mutation analysis
Somatic mutation characteristics between tumors with high versus low gene expression levels were analyzed using the maftools R package (version 2.18.0). Differentially mutated genes (DMGs) were identified via Wilcoxon rank-sum tests, and statistically significant genes (adjusted p < 0.05) were illustrated using forest plots. These plots detailed the mutation types (e.g., missense, nonsense) and effect sizes, supplemented with bar charts that summarized mutation burden (mutations per megabase) for each group.

2.9
Drug resistance analysis
Data from the Genomics of Drug Sensitivity in Cancer (GDSC2, https://www.cancerrxgene.org/) database were leveraged to explore links between gene expression and sensitivity (IC50 values) to 198 anticancer agents.

2.10
Cell culture and animal experiments
Animal experiments were approved by the Animal Care and Ethics Committee of Jinhua Central Hospital. Female Balb/c mice (4–6 weeks old) were obtained from Changzhou Cavens Model Animal Co., Ltd., and acclimated for a week. CT26 cells were cultured to 70–80 % confluence overnight in serum-supplemented medium. Cells were collected by centrifugation at 1150 rpm for 5 min, rinsed twice in cold PBS, counted, and resuspended at ∼3 × 106 cells/mL. Mice were prepared and injected subcutaneously with 1 × 106 CT26 cells in 200 μL PBS. When tumors reached volumes of 150–250 mm3 (average ∼200 mm3) on day 12 post-inoculation, mice were randomly allocated into four groups: (1) saline control, (2) Doramapimod (10 mg/kg, n = 4), (3) anti-PD-1 (30 mg/kg, n = 4), and (4) combination therapy (n = 4). Treatments were given intraperitoneally every 3 days, for a total of five doses.

2.11
Multiplex immunofluorescence staining
Multiplex staining was conducted with validated antibodies against murine CD8A (rabbit monoclonal, Abcam, ab237723) and FOXP3 (rabbit polyclonal, Abcam, ab20034). The Opal™ 7-color Automated IHC Kit (PerkinElmer) was used following the manufacturer's instructions. Slides were scanned on the Vectra® 3 imaging platform, and marker expression was quantified using inForm® software (version 2.4).

2.12
Statistical analysis
All analyses were performed in R (version 4.2.3). Categorical variables were analyzed by Pearson's chi-squared or Fisher's exact tests (where applicable), while continuous variables were compared using Student's t-test or Wilcoxon rank-sum test depending on normality (assessed via Shapiro-Wilk test). Kaplan-Meier survival curves with log-rank tests were used to assess overall survival (OS), progression-free interval (PFI), disease-free interval (DFI), and disease-specific survival (DSS), using the survival and survminer R packages. Univariate Cox regression identified variables associated with survival (p < 0.05), which were then entered into multivariate Cox models. Forest plots summarized hazard ratios (HR) and 95 % confidence intervals (CI) from univariate Cox analyses of the target gene across OS, DSS, DFI, and PFI. All p-values were two-sided. For multiple comparison corrections, we employed the following approaches: (1) In differential expression analyses across multiple cancer types, the Benjamini-Hochberg method was used to control the false discovery rate (FDR), with adjusted p < 0.05 considered statistically significant; (2) In pathway enrichment analyses (GO, KEGG, GSEA), adjusted p-values corrected by the Benjamini-Hochberg method were reported, with a significance threshold set at adjusted p < 0.05; (3) For gene mutation analysis using maftools, adjusted p-values (FDR-corrected) were used as stated in the original text; (4) For correlation analyses (Pearson and Spearman), survival analyses (Kaplan-Meier and Cox regression), and immune infiltration correlations, nominal p-values <0.05 were considered statistically significant, as these analyses were typically performed within specific cancer types rather than across multiple comparisons.

3
Results
3.1
Differential expression of IGSF9 in tumor versus normal tissue samples
By integrating RNA expression data from TCGA and GTEx databases, we analyzed IGSF9 expression across 33 tumor types. IGSF9 exhibited significant differential expressions in 27 cancers. Elevated IGSF9 expression was observed in tumor tissues compared to normal counterparts in cancers such as bladder urothelial carcinoma (BLCA), breast cancer (BRCA), cervical squamous cell carcinoma (CESC), colon adenocarcinoma (COAD), diffuse large B-cell lymphoma (DLBC), esophageal carcinoma (ESCA), glioblastoma (GBM), head and neck squamous cell carcinoma (HNSC), acute myeloid leukemia (LAML), lower grade glioma (LGG), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), ovarian serous cystadenocarcinoma (OV), pancreatic cancer (PAAD), rectal cancer (READ), gastric cancer (STAD), thyroid cancer (THCA), thymoma (THYM), uterine corpus endometrial carcinoma (UCEC), and uterine carcinosarcoma (UCS). Conversely, higher IGSF9 levels were seen in normal tissues relative to tumors in adrenocortical carcinoma (ACC), clear cell renal carcinoma (KIRC), papillary renal carcinoma (KIRP), hepatocellular carcinoma (LIHC), prostate cancer (PRAD), skin melanoma (SKCM), and testicular germ cell tumors (TGCT). For cancers such as cholangiocarcinoma (CHOL), kidney chromophobe (KICH), mesothelioma (MESO), pheochromocytoma/paraganglioma (PCPG), sarcoma (SARC), and uveal melanoma (UVM), expression differences were not statistically significant, likely due to limited normal controls (Fig. 1A).
Stage-specific analysis (Fig. 1B) showed that IGSF9 expression was significantly higher in advanced (stage III/IV) than early-stage (I/II) tumors in ACC, COAD, HNSC, KIRC, and THCA, suggesting its potential role as a diagnostic and staging biomarker.

3.2
IGSF9 alterations and genomic associations in pan-cancer
Data from cBioPortal revealed that IGSF9 alterations—including mutations, amplifications, and deletions—were widespread across cancers. Amplifications were most frequent in cholangiocarcinoma (CHOL), while deep deletions occurred mainly in prostate cancer (PRAD) and non-clear cell renal carcinomas. Certain tumors, such as thyroid cancer, ocular melanoma, seminomas, and non-seminomatous germ cell tumors, showed no detectable alterations (Fig. 2A).
Analysis of copy number variation (CNV; Fig. 2B) and promoter methylation (Fig. 2C) demonstrated significant CNV-expression correlations in ∼66 % of tumor types, including THYM, BRCA, and BLCA. Promoter methylation levels were inversely associated with IGSF9 expression in about the most tumor types, except LAML, PCPG, KIRC, READ, GBM, KICH, UVM, DLBC, and OV.

3.3
IGSF9 as a prognostic indicator in multiple cancers
Through univariate Cox proportional hazards modeling, we assessed the prognostic value of IGSF9 expression in relation to OS, DSS, PFI, and DFI. Elevated IGSF9 levels were consistently linked to poorer OS in LGG, ACC, PAAD, COAD, THYM, SKCM, and MESO (Fig. 3A). Similarly, high IGSF9 expressions were a risk factor for DSS in cancers including LGG, ACC, PAAD, BLCA, COAD, and HNSC. For DFI, elevated IGSF9 expression predicted adverse outcomes in PAAD and ACC (Fig. 3C), while in PFI analyses, IGSF9 overexpression was associated with unfavorable prognosis in LGG, PAAD, ACC, KIRC, PRAD, GBM, and COAD (Fig. 3D). Collectively, these results underscore the significant impact of IGSF9 on survival outcomes across multiple cancer types, highlighting its potential as a robust prognostic biomarker.

3.4
Pathway and functional associations of IGSF9
We conducted correlation analyses between IGSF9 expression and the transcriptome across tumors, identifying both positively and negatively co-expressed gene sets. Enrichment of these gene sets through HALLMARK analyses (Fig. S1) revealed significant correlations between IGSF9 expression and key oncogenic pathways, particularly the KRAS and Notch signaling cascades, suggesting potential associations with cell cycle dynamics, DNA replication fidelity, and apoptosis. It should be noted that these correlational findings do not establish direct mechanistic involvement of IGSF9 in these pathways. In COAD specifically (Fig. 4A), strong correlations were found between IGSF9 expression and KRAS pathway activity, suggesting a potential association that may warrant further mechanistic investigation to inform treatment strategies for KRAS-mutant CC.
In addition, we showed 50 genes with a positive correlation to IGSF9 expression (Fig. 4B) and genes with a negative correlation to IGSF9 expression (Fig. 4C). GO and KEGG pathway analyses revealed that IGSF9 was associated with biological processes such as selective autophagy and ERBB signaling (Fig. 4D), as well as pathways like VEGF-mediated angiogenesis (Fig. 4E), reinforcing its role in tumor progression and microenvironment modulation.

3.5
Immune microenvironment and IGSF9 expression
To comprehensively understand the regulatory function of IGSF9 within the TME across various cancer types, we initiated our analysis by leveraging an immunotherapy - related gene signature (Fig. S2A). The findings revealed that EMT1 is linked to nearly all cancer types and is up-regulated in most tumors. In COAD, we observed that EMT1 and EMT3 are both up-regulated in patients with high IGSF9 expression, suggesting a potential association between IGSF9 and tumor metastasis in COAD [22]. Additionally, the expression of IGSF9 is significantly correlated with mismatch repair [23] and antigen processing machinery [24], indicating that IGSF9 could potentially serve as a biomarker for immunotherapy in COAD patients.
ESTIMATE analyses (Fig. S2B) indicated that higher IGSF9 expression correlated with lower immune and stromal scores in most tumor types, while showing a positive association in THCA. Focusing on COAD (Fig. 5), the expected negative correlation between tumor purity and immune/stromal content was disrupted in high-IGSF9 tumors, highlighting a complex relationship between IGSF9 expression and immune cell exclusion within the tumor microenvironment.

3.6
The relationship between IGSF9 within pan-cancer immune cell infiltration
To gain deeper insight into the specific TME components associated with IGSF9 expression, we systematically examined immune cell infiltration patterns using both ImmuCellAI and TIMER2 platforms. Our analyses revealed that IGSF9 expression was markedly correlated with infiltration levels of various immune cell subsets across multiple cancer types, including CD8+ naive T cells, Th17 cells, neutrophils, monocytes, as well as certain CD4+ naive T cell populations (Fig. S3A). To further validate and refine these associations, TIMER2 analysis was performed, which consistently demonstrated a robust positive correlation between IGSF9 expression and neutrophil infiltration across nearly all tumor types, corroborating the ImmuCellAI findings. In the context of COAD, IGSF9 expression showed significant positive associations not only with neutrophil infiltration but also with elevated levels of regulatory T cells (Tregs) and cancer-associated fibroblasts (CAFs) (Fig. S3B). These results suggest that IGSF9 expression is associated with features of an immunosuppressive TME and epithelial-mesenchymal transition (EMT)-related signatures in COAD. However, these correlational findings do not establish that IGSF9 directly regulates these immune cell populations or EMT processes. The observed associations may reflect common upstream regulators or parallel biological programs rather than direct causal relationships. It is important to note, however, that due to the inherent heterogeneity of immune landscapes across different malignancies, the strength and nature of these correlations varied to some extent among tumor types.

3.7
High expression of IGSF9 is associated with poor response to immunotherapy
Given the close correlation between IGSF9 and the heterogeneity of the immune microenvironment, we investigated its relationship with immunotherapy efficacy. Tumor mutational burden (TMB) [25] and microsatellite instability (MSI) status [26] are two clinical indicators closely linked to immunotherapy response. Analysis of two independent datasets revealed that COAD patients with high IGSF9 expression had lower TMB levels (Fig. 6A), and MSI scores (Fig. 6B) were near the lowest across all tumors analyzed. Spearman correlation analyses further showed negative correlations between IGSF9 and both TMB and MSI (Fig. 6C and D), suggesting that CC patients with high IGSF9 expression are likely non-responsive to immunotherapy.
We performed exploratory correlation analyses using two publicly available immunotherapy datasets to assess whether IGSF9 expression patterns observed in colon cancer may have broader implications for immunotherapy response across cancer types. It is important to note the characteristics and limitations of these cohorts: GSE61676 comprises 64 patients with advanced NSCLC treated with bevacizumab plus erlotinib combination therapy, representing an anti-angiogenic and EGFR-targeted regimen rather than immune checkpoint blockade. GSE135222 includes 121 melanoma patients treated with anti-PD-1/PD-L1 monotherapy or combination immunotherapy, with heterogeneous treatment protocols. In the GSE61676 dataset [27] (Fig. 7A), patients with high IGSF9 expression showed significantly more non-responders to immunotherapy compared to low IGSF9 expressers. Survival analysis also indicated shorter survival duration after immunotherapy in high IGSF9 patients. Similarly, in the GSE135222 dataset [28] (Fig. 7B), all high IGSF9 expression patients experienced relapses after immunotherapy and had significantly shorter survival times. These cross-cancer exploratory analyses provide preliminary evidence suggesting that IGSF9 may serve as a biomarker associated with treatment resistance; however, the lack of colon cancer-specific immunotherapy cohorts substantially limits the direct clinical applicability of these findings to colon cancer patients. Prospective validation in dedicated colon cancer immunotherapy trials is essential before any clinical implementation.

3.8
Single-cell and spatial transcriptomics reveal IGSF9 is predominantly expressed in colorectal cancer epithelium
To localize IGSF9 expression, we analyzed its distribution across single-cell datasets of multiple CRC samples using the IMMUcan database [29] (Fig. 8A). Boxplot analysis showed IGSF9 was primarily expressed in endothelial and epithelial cells (Fig. 8B). We then selected the GSE146771 dataset with high IGSF9 expression for UMAP dimensionality reduction and cell type annotation (Fig. 8C), mapping IGSF9 expression onto the UMAP plot. Results demonstrated that IGSF9 was predominantly expressed in malignant tumor epithelium.
Subsequent validation using spatial transcriptomics data from GSE146771 further supported these findings. Following spatial transcriptomics annotation (Fig. 9A), colocalization of “Malignant” cell clusters with IGSF9 expression was observed, reconfirming the single-cell data (Fig. 9B). To quantify cellular expression patterns, Spearman correlation analysis was performed. Consistent with spatial localization, IGSF9 expression positively correlated with malignant tumor epithelial cells (Fig. 9C) but negatively correlated with endothelial cells (Fig. 9D), fibroblasts (Fig. 9E), and proliferative T cells (Tprolif, Fig. 9F). Data from the ProteinAtlas database (Fig. S4) also confirmed high IGSF9 expression in malignant epithelium of CC.

3.9
Doramapimod sensitizes immunotherapy responses in IGSF9-high CC patients
In an effort to explore potential precision therapeutic strategies for CC patients exhibiting high IGSF9 expression, we conducted a comprehensive drug sensitivity screening based on data from the GDSC2 database (Fig. S5). This analysis highlighted Doramapimod—a potent MAPK pathway inhibitor—as the most promising agent for IGSF9-high CC, consistent with our earlier observations linking IGSF9 dysregulation to aberrant activation of the KRAS signaling axis. To experimentally validate these in silico findings and assess the potential synergy between Doramapimod and immunotherapy, we established a subcutaneous xenograft model using CT26 tumor-bearing Balb/c mice (Fig. 10A). Treatment with Doramapimod alone significantly impeded tumor growth, while the combination of Doramapimod and anti-PD-1 therapy yielded even greater tumor regression compared to either monotherapy (Fig. 10B and C).
Further examination of the tumor immune microenvironment through histological analysis (Fig. 10D) revealed that combined treatment substantially reduced the infiltration of FOXP3+ Tregs relative to either single-agent group (Fig. 10E). Concurrently, this regimen significantly enhanced the infiltration of cytotoxic CD8+ T cells (Fig. 10F). Comparative analyses across the different treatment arms underscored Doramapimod's capacity to diminish immunosuppressive Treg populations, thereby amplifying the efficacy of anti-PD-1 therapy through increased CD8+ T cell recruitment and activation. Collectively, these findings support Doramapimod as a promising adjunctive agent capable of sensitizing IGSF9-high CC tumors to immune checkpoint blockade, likely through modulation of the tumor's immunosuppressive microenvironment.

4
Discussion
Our study, through an integrated pan-cancer discovery and CC-focused validation approach, demonstrates that IGSF9 is a promising biomarker particularly relevant to CC biology. While our initial pan-cancer analyses revealed IGSF9 dysregulation across multiple malignancies, we specifically prioritized colon cancer for in-depth characterization based on: (1) the significant upregulation and prognostic value of IGSF9 in COAD, (2) the strong correlation with immunosuppressive TME features, and (3) the clinical need for novel biomarkers in colon cancer immunotherapy. The subsequent single-cell, spatial transcriptomics, and preclinical therapeutic experiments were therefore focused exclusively on CC to provide mechanistic and translational insights specific to this malignancy. These finding aligns with previous reports showing IGSF9 promotes tumor progression by enhancing EMT and suppressing T-cell proliferation in lung cancer [14]. In COAD, high IGSF9 expression is associated with reduced immune cell infiltration and an immunosuppressive TME, characterized by increased Tregs and CAFs. These results extend prior observations in mouse models where IGSF9 blocked immunotherapy efficacy by inhibiting T-cell function, highlighting its conserved role in creating the immune desert-like microenvironment across species.
Notably, our single-cell and spatial transcriptomics analyses reveal IGSF9 is predominantly expressed in malignant epithelial cells of CC, rather than immune or stromal cells. This epithelial-specific expression pattern contrasts with some immune checkpoint molecules [30] (e.g., PD-L1, which is often expressed on both tumor and immune cells), suggesting IGSF9 may exert its immunosuppressive effects indirectly, such as secreting factors that recruit immunosuppressive myeloid cells or modulating the epithelial-stromal crosstalk.
Functional enrichment analyses reveal correlations between IGSF9 expression and KRAS and Notch signaling pathways, which are known to be critical for cell cycle regulation and epithelial-mesenchymal transition (EMT) [31]. While these associations are intriguing, we acknowledge that enrichment-based analyses cannot establish direct mechanistic involvement. The correlation between IGSF9 and KRAS pathway activity may reflect co-regulation by shared upstream factors, parallel activation in aggressive tumor phenotypes, or indirect relationships through intermediate molecular mediators. Direct experimental validation—such as IGSF9 knockdown/overexpression studies with assessment of KRAS/ERK phosphorylation, rescue experiments, and co-culture systems—would be required to establish mechanistic causality. This finding is consistent with a recent study showing IGSF9 activates GSK-3β/β-catenin signaling to promote lung cancer metastasis. Interestingly, Hui et al. found that IFN-γ is a key cytokine that induces IGSF9 expression, which may further explain why IGSF9 is upregulated in the tumor microenvironment [32]. In COAD, the correlation between IGSF9 and KRAS pathways may explain treatment resistance in KRAS-mutated tumors, as KRAS mutations are known to drive immunosuppressive TME remodeling [33]. Our drug sensitivity analysis further identifies Doramapimod [34], a MAPK inhibitor targeting KRAS downstream pathways, as a potential therapeutic agent for IGSF9-high CC. Combined with anti-PD-1 therapy, Doramapimod reduces Treg infiltration and enhances tumor regression in mouse models, providing preclinical validation for targeting IGSF9-associated pathways to overcome immunotherapy resistance.
A key finding of this study is the strong association between high IGSF9 expression and poor response to immunotherapy in CC. Patients with high IGSF9 exhibit lower TMB and MSI scores, which are established predictors of immunotherapy efficacy [35,36]. This is further validated in two independent immunotherapy datasets, where high IGSF9 expression correlates with increased non-responders and shorter survival after treatment. These results are consistent with mechanistic studies showing IGSF9 suppresses T-cell infiltration and promotes immune evasion, highlighting its role as a predictive biomarker for immunotherapy resistance. Notably, while previous studies have focused on immune checkpoint molecules (e.g., PD-1/PD-L1) in TME [37,38], our work identifies IGSF9 as a novel regulator of immunosuppression of TME that acts through distinct mechanisms. For example, unlike PD-L1, which directly interacts with T-cell surface receptors, IGSF9 may indirectly inhibit immunity by recruiting immunosuppressive cell populations (e.g., neutrophils, Tregs) and activating EMT programs. This suggests that IGSF9 could serve as a complementary biomarker to PD-L1, particularly in PD-L1-negative tumors where alternative resistance mechanisms are prevalent.
Doramapimod (BIRB 796) is a potent and selective p38 MAPK inhibitor that has been evaluated in multiple clinical trials for inflammatory conditions, including rheumatoid arthritis and Crohn's disease [39,40]. Pharmacokinetic studies have demonstrated favorable oral bioavailability and a manageable safety profile in humans, with dose-limiting toxicities primarily related to hepatic enzyme elevations at higher doses. However, it is important to note that Doramapimod has not yet been evaluated in oncology clinical trials, and its safety profile in combination with immune checkpoint inhibitors remains to be established. The potential for additive hepatotoxicity when combining Doramapimod with anti-PD-1 agents—which can themselves cause immune-related hepatic adverse events—warrants careful consideration in future clinical trial design. Additionally, the optimal dosing schedule, pharmacokinetic interactions, and long-term safety of this combination require systematic investigation in phase I/II studies before clinical implementation. Our preclinical findings provide a rationale for such clinical development, but we acknowledge that substantial translational work remains necessary.
However, several limitations must be addressed. First, our findings are primarily based on publicly available databases and preclinical models, without validation in prospective clinical trials specific to colon cancer immunotherapy. Notably, the immunotherapy datasets used in this study-GSE61676 and GSE135222-are not colon cancer-specific cohorts. GSE61676 comprises non-small cell lung cancer patients treated with bevacizumab/erlotinib, while GSE135222 includes melanoma patients receiving anti-PD-1/PD-L1 therapy. While these datasets provide preliminary evidence linking IGSF9 expression to immunotherapy outcomes across various cancer types, the absence of large-scale, prospective validation in colon cancer-specific immunotherapy cohorts limits the generalizability of our predictive findings. Future studies using dedicated colon cancer immunotherapy trials with IGSF9 expression profiling are essential to establish its clinical utility as a predictive biomarker. Second, the role of IGSF9 in the TME varies across cancer types, as evidenced by its differential correlation with immune cell infiltration in pan-cancer analyses. Third, the preclinical CT26 murine model used in this study, while providing valuable proof-of-concept evidence, has inherent limitations in replicating the complexity of the human colon cancer microenvironment. Differences in immune system dynamics between mice and humans—including variations in immune cell subsets, checkpoint molecule expression patterns, and cytokine networks—may impact the translatability of our therapeutic findings. The CT26 model represents MSS murine CRC, which may not fully capture the heterogeneity observed in human colon cancer subtypes. Future validation studies utilizing patient-derived xenografts in humanized mouse models or organoid co-culture systems with patient-derived immune cells would provide more clinically relevant evidence to support the therapeutic potential of combining Doramapimod with immune checkpoint inhibitors.

5
Conclusion
In summary, this study establishes IGSF9 as a key regulator of immunosuppressive TME in CC, providing a rationale for its use as a prognostic and predictive biomarker. The combination of MAPK inhibition and immune checkpoint blockade offers a promising strategy to overcome resistance in IGSF9-high tumors, warranting further clinical investigation.

Consent for publication
Not applicable.

Ethics approval and consent to participate
Written informed consent was acquired from every patient, and study was approved by the Clinical Research Ethics Committee of Affiliated Jinhua Hospital, Zhejiang University School of Medicine ((2024) Ethics approval No. (91)).

Declaration of generative AI and AI-assisted technologies in scientific writing
During the preparation of this work, the authors utilized ChatGPT for language editing and polishing. The authors have thoroughly reviewed and edited the output and take full responsibility for the content of this publication.

CRediT authorship contribution statement
Kangfu Dai: Investigation, Supervision, Writing – original draft. Xingxing Yu: Project administration, Writing – original draft. Shicheng Zhou: Data curation, Formal analysis. Zhekang Jin: Methodology, Resources. Wuzhen Dong: Project administration, Supervision. Pengcheng Yu: Resources. Chenyang Ge: Investigation, Supervision. Zhifeng Zhong: Methodology, Software. Jianping Wang: Resources, Validation, Writing – review & editing.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.