In the present investigation, male C57BL/6J mice, aged 6-8 weeks, were obtained from Ensiwer Corporation and utilized in the ICB-resistant MC38 tumor model. The colorectal cancer MC38 cell line, adjusted to a concentration of 1 × 10^5 cells, was subcutaneously inoculated into these mice. On the seventh day post-inoculation, the animals were administered either IgG or anti-CTLA4 antibodies (Bio X cell, 9H10) at a dosage of 5 mg/kg for a continuous period of five days. At the culmination of the designated experimental timelines, the tumor tissues were meticulously excised for further analysis. In a subsequent experimental paradigm, mice bearing MC38 tumors were subjected to treatment with either a control vehicle or PHA-793887 (MedChemExpress, HY-11001), administered at a dosage of 10 mg/kg. Post-treatment, the excised tumor tissues underwent flow cytometric analysis to elucidate the impact of PHA-793887 on the TME. Concurrently, tumor dimensions were meticulously monitored biweekly throughout the experimental period.

Live cells were assessed using the Fixable Viability Dye eFluor 450. To evaluate cytokine production, the cells were stimulated with the Cell Stimulation Cocktail and subsequently labeled with anti-IFN-γ (BioLegend, XMG1.2) and anti-TNF-α (BioLegend, MP6-XT22) antibodies. Other antibodies included an-CD45 (Biolegend, 2D1), anti-CD11b (BioLegend, M1/70), anti-CD8a (BioLegend, 53-6.7), anti-CD4 (BioLegend, GK1.5), anti-FOXP3 (BioLegend, MF-14), anti-F4/80 (BioLegend, BM8).

Analysis of stained cells was conducted using a BD FACSCanto II Flow Cytometer in conjunction with BD FACSDiva software (BD Biosciences), and the resulting data were processed using FlowJo software (version 10.5.3).

Total RNA was isolated from MC38 tumor tissues using the Trizol reagent (Invitrogen, catalog number 15596026). RNA samples were forwarded to ANNOROAD for construction of sequencing libraries and subsequent sequencing on the NovaSeq platform (Illumina). The resultant raw fastq files were processed to quantify gene expression as transcripts per million (TPM) using the htseq-count tool, facilitating the downstream analysis. Differential expression analysis of genes (DEGs) was performed using the “DESeq2” R package, applying stringent filtering criteria: a fold change threshold greater than 2, an adjusted P-value of less than 0.05, and a mean log2-TPM in the high-expression cohort exceeding 0.

In the present investigation, we procured five distinct public datasets from the NCBI Gene Expression Omnibus (GEO) repository. Our approach entailed the application of scRNA-seq datasets, namely GSE231559 and GSE166555, to dissect the heterogeneity of fibroblast populations within both normal and neoplastic colorectal tissues. Additionally, we leveraged the COAD cohort from the TCGA database and an aggregate of three bulk RNA sequencing datasets, GSE17536, GSE29621, and GSE38832, for the development and substantiation of our prognostic model. It is important to note that for analyses based on publicly available datasets, neither patient consent nor ethical board approval is required.

The processing of scRNA-seq data was executed utilizing the ‘Seurat’ R package (version 5.0.2), following established protocols (12). Initially, a stringent filtration criterion was applied, excluding cells with gene expression levels below the 200-gene threshold or exceeding the 6000-gene ceiling, as well as those with mitochondrial gene expression surpassing the 5% mark. This exclusionary process was pivotal for the retention of a substantial cell population representative of the datasets in question. Subsequently, the SCTransform function was utilized to standardize and normalize the raw count data, which was then subjected to principal component analysis (PCA) to identify the underlying patterns. To address and neutralize batch effects inherent in the dissociated scRNA-seq datasets, the “Harmony” R package, was strategically implemented. Clustering was conducted by assessing the edge weights connecting pairs of cells, culminating in the construction of a shared nearest-neighbor graph. This graph was adeptly derived using the Louvain algorithm, facilitated by the FindNeighbors and FindClusters functions. The outcome of this process was a visual representation of the cells, rendered through the UMAP algorithm, providing a comprehensive overview of cellular distribution. The “FindMarkers” function was employed to pinpoint genes that were preferentially expressed within specific clusters, in addition to identifying DEGs. Each resultant cell cluster was annotated with reference to established cell-type marker genes, enhancing the interpretability of the data. To elucidate the distinct expression profiles of the identified genes at the single-cell level, the “scRNAtoolVis” package was utilized, providing a graphical interface that enabled precise and clear visualization of gene expression patterns.

To explore genes associated with FGFRS-positive fibroblasts, we conducted a hdWGCNA using the “hdWGCNA” package. We created metacells for each sample and cell cluster, with 50 cells per metacell, and applied a standard pipeline of functions to analyze gene expression patterns and visualize module relationships in a reduced-dimensional space.

As previous published study (13, 14), we identified FGFRS-fibroblast-related genes using hdWGCNA and validated their predictive potential in tumor development with three transcriptome datasets, GSE17536 GSE29621, and GSE38832. Using the “mlr3” package, we developed a predictive model incorporating seven machine learning algorithms. Following cross-validation, we selected the most accurate model and assessed its predictive ability on an independent test set.

Employing an unsupervised approach, pseudotemporal analysis was performed using the “Monocle” package, applying the DDR-Tree algorithm with its default parameters to delineate the developmental trajectory of fibroblasts. Following the pseudotemporal trajectory mapping, the “plot_pseudotime_heatmap” function was engaged to craft a heatmap. This visual tool effectively depicted the fluctuating expression patterns of a cohort of genes, illustrating their dynamic behavior along the pseudotime trajectories of fibroblasts. Furthermore, to uncover potential cellular interactions, both intracellular and extracellular, the “CellChat” package was deployed using its default settings and recommended pipeline configurations. This application facilitated the identification of communication networks among fibroblasts and other cellular components within the TME.

The Seurat package’s “FindMarkers” function was deployed to discern DEGs within each delineated cell subcluster. The selection criteria for these genes were stringent, requiring a fold change surpassing a threshold of 2 and an adjusted p-value below the significance level of 0.05. Subsequently, leveraging the identified DEGs, a comprehensive GSEA and KEGG enrichment analyses were conducted to explore the functional profiles of the cell subgroups. These analyses were executed utilizing the “clusterProfiler” package, which provided a robust framework for assessing the overrepresentation of specific gene sets and biological processes. To visually represent the functional enrichment results, the “GseaVis” package was employed. This tool facilitated the creation of intuitive and informative visualizations that encapsulated the enriched biological themes and pathways associated with the cell subclusters under investigation.

To explore the diversity of FGFRS subtypes, we applied the NMF algorithm from the “NMF” package (15). The objective was to identify distinct subtypes characterized by unique gene expression patterns. To assess the prognostic value of these genes, we conducted survival analysis using the “survival” package. Additionally, we used the “ggrisk” package to analyze the survival and risk profiles of cancer patients, categorizing them into high- and low-risk groups. The analyses were considered statistically significant for P values below 0.05.

All computational analyses and graphical representations were executed utilizing the R software (v4.3.2). The strength and direction of the linear relationships between pairs of continuous variables were evaluated through Pearson’s correlation coefficients. In the case of quantitative datasets, statistical comparisons among subgroups were made using either a two-tailed, unpaired Student’s t-test for two-group comparisons or a one-way analysis of variance (ANOVA) complemented with Tukey’s post hoc test for multiple group analyses. The threshold for statistical significance was set at a P-value of less than 0.05.

To elucidate the role of the fibroblast growth factor receptor (FGFR) signature in tumors, we analyzed single-cell RNA sequencing (scRNA-seq) data from colorectal cancer contained in the Gene Expression Omnibus (GEO) database (GSE231559, GSE166555) to identify potential fibroblast subsets that are positive for FGFRS (FGFRS⁺). In the GSE231559 dataset, we annotated cell types using established markers, including T cells, malignant cells, neutrophils, myeloid cells, fibroblasts, B cells, epithelial cells, and plasma cells, each characterized by their unique gene expression profiles ( Figure 1A ). Notably, the analysis highlights genes like GZMK, GZMM, CD3G, which are expressed in T cells, and genes such as CD79A and CD19, which are characteristic of B cells. The myeloid and plasma cells are associated with genes like C1QC and IGHA1, respectively ( Figure 1B ). In the GSE166555 dataset, we identified 22 clusters and further annotated these clusters to 10 main clusters based classic markers, including B cells, DC cells, endothelial cells, fibroblasts, malignant cells, mast cells, monocyte/macrophage, plasma and T cells ( Supplementary Figure S1A ). These established markers including of MZB1, MS4A1, EPCAM, CD3D, CPA3, CD163, COL1A2, VWF, and IDO1 ( Supplementary Figure S1C ). Significantly, each cluster exhibited different gene expression profiles ( Supplementary Figure S1D ). Furthermore, to analyze the effect of FGFR signaling, we constructed an FGFR signature (FGFRS) using 86 genes involved in the FGFR signaling pathway from Molecular Signatures Database (MSigDB). This signature was then applied to score the identified cell groups using the “AddModuleScore” function in the Seurat package. Our findings reveal that fibroblast cell groups exhibit the highest FGFRS score, indicating a significant involvement of FGFR signaling in these cells ( Figures 1C , Supplementary Figure S1B ). To further dissect the role of FGFRS in fibroblast cell cluster, we isolated these cells and identified nine main fibroblast subgroups. Notably, in tumor tissues, a distinct fibroblast subgroup was found to have an increased proportion relative to normal tissues and showed highest FGFRS score (here after as “FGFRS⁺ fibroblast”) ( Figures 1D , Supplementary Figure S1E ). This observation suggested a potential role of FGFRS in the transformation and proliferation of fibroblasts in the TME. Furthermore, Gene Set Enrichment Analysis (GSEA) was performed on the transcriptome of this specific fibroblast subsets. The enrichment analysis identified several key cells signaling pathways that are potentially modulated by FGFRS. In the GSE231559 dataset, these pathways include receptor-mediated endocytosis, positive regulation of phagocytosis, and the regulation of miRNA metabolic processes, among others. In the FGFRS⁺ fibroblast, some signaling pathways were significantly enriched, including epithelial tube morphogenesis and collagen fibril organization ( Figure 1E ). In the GSE16655 dataset, FGFRS⁺ fibroblasts exhibited upregulated signaling pathways, including cytokine-cytokine receptor interactions, while downregulated pathways included the Wnt and Hippo signaling pathways ( Supplementary Figure S1F ). In summary, the integration of scRNA-seq data analysis highlighted that FGFRS is enriched in the fibroblast cluster within the TME, and these relevant signaling pathways may be critical for the role of FGFRS⁺ fibroblasts in tumor progression.

To identify potential core genes associated with FGFRS⁺ fibroblasts, we conducted high-dimensional Weighted Gene Co-expression Network Analysis (hdWGCNA), a comprehensive methodology for analyzing co-expression networks in the scRNA-seq data. This analysis aimed to detect co-expressed gene modules and unravel their functional roles within FGFRS⁺ fibroblasts. Subsequently, we constructed a scale-free co-expression network by applying an optimal soft thresholding power of 12 ( Supplementary Figure S2A ). From this analysis, we distinguished a total of 19 distinct gene co-expression modules, identified the top 10 hub genes from these modules, and constructed protein-protein interaction (PPI) networks for the identified hub genes in each module ( Supplementary Figure S2B, C , Figure 2B ). Additionally, we investigated the correlation between each module ( Supplementary Figure S2D ), where modules 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, 15, 18, and 19 displayed significant activation primarily in FGFRS⁺ fibroblasts ( Figure 2C , Supplementary Figure S2E ). Subsequently, 325 genes from these modules underwent univariate Cox regression analysis in the TCGA COAD cohort, leading to the identification of 13 genes significantly associated with overall survival in CRC patients ( Figure 2D ). Additionally, by integrating data from the TCGA COAD and GEO database cohorts, we constructed robust models using 101 algorithmic combinations and calculated the area under the curve (AUC) for each model across all cohorts to assess their predictive capacity. ( Figure 2A ). Among the 101 models, the Step Cox (direction = both) algorithm in conjunction with a Random Survival Forest (RSF) demonstrated the highest AUC, serving as the basis for the final model creation. We further utilized RSF analysis to assess the prognostic relevance of various genes in predicting patient survival outcomes systematically. Notably, genes such as JDP2, HEYL, NRG1, RPS17, and MANF exhibited substantial predictive value, as indicated by their low Minimal Depth and high Variable Importance scores, thus influencing the accuracy of the survival model significantly ( Figures 2E, F , Supplementary Figure S2F ). The gene signature, comprising JDP2, HEYL, NRG1, RPS17, and MANF, known as the FRS, has demonstrated substantial predictive capabilities for patient survival at the 1-year, 3-year, and 5-year intervals. This integrated genetic profile serves as a valuable prognostic tool, providing insights into anticipated clinical outcomes for patients over the specified time frames ( Figure 2G ). Based on the FRS, cancer patients were categorized into high- and low-risk groups, with high-FRS patients exhibiting a poor prognosis ( Figure 2H ). These results were further validated using cohorts from the TCGA and GEO databases. In the TCGA cohort, Kaplan-Meier survival curve analysis revealed adverse outcomes for high-FRS patients, with an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.7334 serving as a critical metric for model evaluation. These comprehensive results highlight the predictive capacity of the model and align with the analysis of COAD cohorts from the GEO database ( Figures 2I-L ).

To further analyze the correlation between the FRS and tumor immunity, we employed Non-negative Matrix Factorization (NMF), a robust technique for clustering analysis, to investigate the FRS of cancer patients based on the FGFRS. Our analysis specifically aimed to identify the optimal number of clusters (K) that would best capture the heterogeneity within the patient cohort. Following the NMF-based clustering, we observed that when K was set to 2, the consensus matrix revealed a clear distinction between two distinct patient groups. This finding suggests that dividing the patients into two groups at K=2 provides the most meaningful separation in terms of their FRS ( Figure 3A , Supplementary Figure S3A ). Notably, patients in cluster 1 exhibited a poorer prognosis compared to those in cluster 2 ( Figure 3B ). Upon TME score analysis, cluster 1 displayed lower immune scores, including stromal score, immune score, and ESTIMATE score, in contrast to cluster 2 ( Supplementary Figure S3B ). Additionally, CIBERSORT analysis was conducted for both clusters. Notably, tumors from patients in cluster 1 demonstrated reduced CD8⁺ T cell infiltration but increased M0-like macrophages compared to those in cluster 2 ( Figure 3C ). Moreover, tumors from patients in cluster 1 exhibited decreased expression levels of immunostimulatory molecules ( Supplementary Figure S3C ). These findings imply that tumors from patients in cluster 2 might possess heightened antitumoral immunity within an immune-stimulating TME. The cell-cell interactions within the TME play a crucial role in tumor progression. Therefore, to further understand the interactions between FGFRS⁺ fibroblasts and T cells, we conducted CellChat analysis. While the total interaction strength was notably increased in tumor tissues compared to normal tissues, the number and weight of interactions showed no significant differences ( Supplementary Figures S3D, E ). Subsequently, we investigated whether specific signaling interactions between FGFRS⁺ fibroblasts and T cells were altered. Utilizing signal flow analysis, a quantitative method for assessing information transmission in biological systems, revealed substantial changes in signaling pathways between tumor and normal tissues. These results highlight a significant divergence in the flow of information within distinct tissue states ( Figure 3D ). In our subsequent analysis, we delved into the ligand-receptor interactions within the TME, focusing on the FGFRS⁺ fibroblasts and their potential binding partners. Utilizing a comprehensive approach, we identified a significant interaction between collagen (COL)-related ligands and the CD44 receptor on T cells. Notably, this interaction was found to be markedly enhanced in the TME compared to normal tissues ( Figure 3E ). Our findings underscore the complex interplay between the extracellular matrix components and immune cells. Specifically, the pairs COL4A2-CD44, COL6A1-CD44, and COL1A2-CD44 demonstrated a pronounced increase in their relative contribution to the signal transduction within the tumor context ( Figures 3F , Supplementary Figure S3F ). Moreover, there was a significant increase in COLLAGEN signaling between FGFRS-positive fibroblasts and T cells in the tumor tissue compared to normal tissues ( Figures 3G, H ). This heightened interaction between collagen ligands and the CD44 receptor on T cells implies a potential role in influencing the immune response within the TME.

In our investigation of the influence of the FRS on T cell dynamics, we conducted a comprehensive analysis of scRNA-seq data from the GEO database. Using unsupervised clustering techniques, we analyzed T cell populations and identified seven primary clusters. By annotating these clusters with characteristic markers, we delineated six major T cell subsets: naïve T cells, regulatory T cells (Tregs), effector T cells (Teffs), central memory T cells (Tcms), natural killer T cells (NKTs), and exhausted T cells (Tex). Each subset exhibited distinct expression patterns of specific markers ( Figures 4A, D , Supplementary Figure S4A, B ). Our results revealed a notable shift in the proportion of T cell subsets between tumor tissues with high FRS scores and those with low FRS scores. Specifically, there was a significant increase in the proportion of Teff and NKT in tumors with low FRS scores, whereas the proportion of Tex was markedly reduced in comparison to tissues with high FRS scores ( Figures 4B, C ). Given the pivotal role of T cell function in determining the success of tumor immunity, we further analyzed the impact of FRS on the functionality of Teff cells. We observed a significant upregulation of effector molecules in Teff cells within tumors with low FRS scores, including PRF1, GZMH, GZMB, and IFNG ( Figure 4E ). Additionally, we utilized pseudotime analysis to elucidate the temporal evolution of T cell subset differentiation. Our findings indicated that in the early stages of differentiation, the T cell population was predominantly composed of Treg cells. As pseudotime progressed, there was a transition towards a predominance of naive T cells. In the late stages of differentiation, the population was characterized by an increase in Tex cells and NKT cells ( Figures 4F , Supplementary Figure S4C ). Importantly, the expression levels of CD8, GZMB, and IFNG and other markers associated with T cell function were significantly increased in the late stages of differentiation ( Supplementary Figures S4D, E ).

Utilizing the transcriptomic sequencing data from the TCGA COAD cohort, we embarked on a comparative analysis of gene expression differences between tumor tissues with high and low levels of FRS score. This analysis unveiled a spectrum of differentially expressed genes (DEGs), which we subsequently subjected to a co-expression network analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. Notably, genes such as NRG1, SAMD13, MFAP3L, SALL1, ZNF704, SEMA5A, and HLA-DQB1 were identified as having altered expression patterns, potentially reflecting the diverse roles of FGFRS in tumor biology. The presence of immune-related genes such as CX3CL1, HLA-DQBY, and KIR2DL4, along with cytotoxic markers like GZMH and GZMB, suggests an intricate interplay between FRS signaling and immune response modulation within the TME ( Figure 4G ). Furthermore, the KEGG analysis elucidated the biological pathways and cellular processes significantly enriched among the differentially expressed genes (DEGs). Pathways such as the NF-kappa B signaling pathway, the STAT signaling pathway, and the TNF signaling pathway were prominently highlighted, indicating the roles of immune response and cell signaling in tumors altered by the FRS. Additionally, the identification of processes, including neutrophil extracellular trap formation and natural killer cell-mediated cytotoxicity, underscores the influence of the FRS on immune cell functions ( Figure 4H ). Collectively, FRS exhibited a negative correlation with T cell function, thereby impairing antitumor immunity and facilitating tumor progression.

The expression of FRS is inversely associated with T cell function, but it is unclear whether this correlation contributes to immunotherapy resistance. In order to elucidate the role of FRS in colon tumors following immune checkpoint blockade (ICB) therapy, we developed an ICB-resistant MC38 tumor model and conducted RNA sequencing analysis. Within our experimental framework, mice engrafted with MC38 tumors were treated with either IgG or anti-CTLA4 antibodies, and we meticulously monitored changes in tumor volume. Upon completion of the treatment, we classified tumors that exhibited a significant reduction in volume following anti-CTLA4 treatment as “responder” phenotypes, in stark contrast to those that showed minimal change with IgG treatment, which we labeled as “nonresponder” phenotypes ( Figure 5A ). Of particular interest, the expression levels of HEYL, RPS17, and JDP2 were significantly elevated in tumors from nonresponders compared to those from responders ( Figure 5B ). To investigate the potential correlation between FRS expression and immune checkpoint blockade (ICB) resistance, we stratified nonresponders into two groups based on the gene set variation analysis (GSVA) score of the FRS and identified differentially expressed genes (DEGs) between high and low FRS-expressing tumors ( Figures 5C, D ). Utilizing these DEGs, we performed GSEA, revealing that several signaling pathways were significantly enriched in low FRS -expressing tumors, including those involved in extracellular matrix organization, collagen formation, and oxidative stress response. Notably, key genes associated with collagen and oxidative stress were found to be significantly downregulated in low FRS -expressing tumors ( Supplementary Figures S5A-F ). Further investigation into the impact of FRS on the TME of nonresponsive MC38 tumors using MCPcounter analysis indicated elevated levels of endothelial and fibroblast infiltration, alongside a diminished presence of T cells in the TME of tumors with high FRS expression ( Figure 5E ). Additionally, the expression of GZMA, a marker for cytotoxic T lymphocytes, was significantly higher in tumors with low FRS expression ( Figure 5F ). Our data suggest that reduced FRS may enhance T cell function within the TME. To comprehensively assess the influence of FRS expression on ICB-resistant tumor development, we conducted pseudotime analysis on RNA sequencing data from nonresponsive tumors. The findings revealed that the expression patterns of three key genes closely mirrored one another and were significantly elevated during the advanced stages of tumor progression ( Figures 5G, H ). Collectively, these results suggest that the FRS is positively associated with ICB resistance and may facilitate tumor progression.

As our analysis, the correlation between FRS expression and tumor immunity has been established, yet the extent to which FRS inhibitors can curtail tumor progression through the enhancement of antitumor immunity is not fully understood. To address this, we utilized the CTRP and PRISM databases to prognosticate potential inhibitors that target the FRS and subsequently validated the antitumoral efficacy of these inhibitors in murine tumor models. Through the amalgamation of data from both databases, we identified PHA-793887, recognized as a quintessential cyclin-dependent kinase (CDK) inhibitor, as a promising candidate with potential inhibitory effects on the FRS ( Figures 6A, B ). Then, we developed a mouse xenograft model of colorectal cancer by subcutaneously inoculating MC38 tumor cells and evaluated the impact of PHA-793887 on tumor progression and the TME via oral administration ( Figure 6C ). Notably, the treatment did not induce any discernible harm to the vital organs of the mice, encompassing the heart, liver, intestine, stomach, and lungs ( Figure 6D ). Importantly, the PHA-793887 significantly suppressed tumor growth and reduced the tumor burden relative to the control cohort ( Figure 6E ). Further examination of the TME utilizing flow cytometry disclosed that PHA-793887 markedly augmented the secretion of cytotoxic T cell-associated cytokines, including interferon-gamma (IFN-γ), tumor necrosis factor-alpha (TNF-α), and granzyme B (GZMB), within the tumor milieu ( Figure 6F ). This suggests an enhancement of T cell functionality. Moreover, in tumors treated with PHA-793887, there was a significant increase in the infiltration of CD8⁺ T cells and effector CD4⁺ T cells, along with a decrease in tumor-associated macrophages (TAMs) and regulatory T cells (Tregs) ( Figures 6G–J ). These findings underscore that PHA-793887, acting as an FRS inhibitor, can profoundly rewire the immunosuppressive TME, bolster anti-tumor immune responses, and reduce tumor growth.

To further elucidate the association between FRS and the therapeutic efficacy of ICB in cancer patients, we conducted an integrative analysis using TCGA and the Tumor Immune Dysfunction and Exclusion (TIDE) database. Our analysis revealed that patients with tumors exhibiting high FRS score had a significantly lower survival rate compared to those with low FRS score within their tumors ( Figure 7A ). Additionally, by examining patients within the TIDE database who underwent ICB treatment, we observed a notable increase FRS among patients with stable disease (SD) or progressive disease (PD) compared to those with partial response (PR) ( Figure 7B ). Utilizing the TIDE scoring system, we identified a higher prevalence of responders among patients with low FRS expression ( Figure 7C ). Furthermore, the SubMap analysis revealed that the low FRS group exhibited a high likelihood of response to anti-PD-1 in the immunotherapy cohorts (IMvigor210) ( Figure 7D ). In the context of patients undergoing ICB therapy, we conducted a detailed analysis of the prognostic impact of key genes within the FRS. Our findings indicate that tumors with elevated expression levels of HEYL, RPS17, and JDP2 are associated with a poorer prognosis for cancer patients ( Figures 7E-G ). These results suggest that the FRS may serve as a prognostic indicator for ICB treatment outcomes, with a lower FRS score potentially correlating with a more favorable response to ICB therapy.