Abstract
Convincing evidence suggests that FoxO signaling (FS) dysfunction is associated with cancer progression and tumorigenesis but its effect on the tumor microenvironment (TME) and immunotherapy response remains unclear. Here, we retrieved FS-related genes from the KEGG database, first analyzing their differential expression in 31 TCGA cancer types with matched Genotype-Tissue Expression (GTEx) normal tissue data. We then calculated FS scores via ssGSEA for 33 TCGA cancer types, systematically exploring their prognostic value and correlations with TME characteristics and immunotherapy response. High FS scores were consistently associated with an immunosuppressive TME and poorer OS across multiple tumor types. Analysis of the IMvigor210 immunotherapy cohort further revealed elevated FS scores correlated with anti-PD-1 treatment resistance. Focusing on gliomas, we identified FS-based molecular subtypes and constructed a PCA score that robustly predicted glioma prognosis and immune checkpoint inhibitor response. External validation was achieved in independent CGGA and GEO cohorts. Our findings indicate that FS is a key modulator of the immunosuppressive TME and support FS and PCA scores as potential prognostic and immunotherapeutic biomarkers for gliomas with their clinical utility pending experimental validation.
Keywords: FoxO signaling, glioma, immunotherapy, prognosis, tumor immune microenvironment
Introduction
The World Health Organization (WHO) classifies gliomas, highly invasive central nervous system tumors that constitute the majority of brain malignancies, into 4 distinct grades. Low-grade gliomas (LGG) refer to WHO grade 1 to 2 gliomas, and high-grade gliomas (HGG) mainly refer to WHO grade 3 to 4 glioblastoma (GBM), the most malignant subtype.1,2 LGG has the potential to progress to HGG. The prognosis worsens with higher grades with median OS times of 11.6 years, 7 years, and 15 months for patients with LGG, grade 3, and grade 4 glioma, respectively.1 Current medical treatments for gliomas have limitations in improving patient survival and most patients may develop resistance to temozolomide.3 Therefore, exploring novel markers and potential therapeutic targets for gliomas is essential.
Due to their diverse molecular characteristics, transcription factors are critical targets for drug development aimed at inhibiting and regulating oncogenic progression in human cancers. Drugs targeting transcription factors, such as those involved in the proteolysis of transcription factor-derived proteins, represent some of the few available therapeutic strategies for cancer treatment.4 Concrete examples include proteolysis-targeting chimeras that degrade oncogenic transcription factors via the ubiquitin-proteasome pathway, MDM2 inhibitors (eg, idasanutlin, navtemadlin), which restore p53 transcriptional activity, and BET bromodomain inhibitors (eg, JQ1, OTX015), which disrupt transcription factor-chromatin interactions.
FoxO signaling (FS), a crucial pathway in cancer progression, plays a fundamental role in tumor proliferation and spread.5,6 Moreover, various cellular functions, including proliferation, differentiation, apoptosis, and DNA damage repair are regulated by FS, which also serve as mediators of oxidative stress.7,8 FS is also linked to the tumor immune microenvironment. For instance, FoxO1 regulates MHC-II expression and the antitumor effects of tumor-associated macrophages.9 The expression of lymphoid organ-homing molecules is promoted by FoxO1, thereby influencing T cell migration.10–12 Thus, the effect of FS on the tumor microenvironment (TME), particularly in the context of immunotherapy, has not been fully elucidated. Advances in cancer immunotherapy, driven by immune checkpoint inhibitors (ICIs), aim to sustain effector immune cell activity within the TME. However, many patients face challenges regarding the duration of response or resistance to ICIs, highlighting the need for further advancements.13–15 The ineffectiveness of immunotherapy may be linked to an unfavorable immunosuppressive TME. Thus, a detailed examination of the relationship between FS and the TME is crucial.
Initially, a comprehensive analysis of 33 tumor types was performed to investigate genes associated with FS genes using datasets from The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx), focusing on their expression patterns and prognostic implications. Subsequently, the FS score was assessed and its correlation with the TME was explored. Additionally, unsupervised clustering based on FS gene expression resulted in the identification of 2 glioma subtypes (Clusters A and B). Notably, the survival outcomes of these 2 patient groups differed significantly, emphasizing the need for further investigation into the mechanisms related to FS. The characteristics of the 2 clusters, encompassing clinical features, differentially expressed genes (DEGs), signaling pathways, and immune cell infiltration, were compared through further analyses. Moreover, a prognostic model for glioma patients was developed using the identified DEGs. The study also investigated the correlation between the prognostic models, TME characteristics, and immunotherapy efficacy. Among all 33 cancer types examined, glioma (TCGA-LGG and TCGA-GBM) stood out as the cancer type with the most pronounced FS score elevation relative to normal tissue, the strongest association between FS score and unfavorable prognosis, and the most heterogeneous immune microenvironment between FS-high and FS-low groups. Moreover, FS has established mechanistic links to glioma-relevant processes including TGF-β signaling, T cell exclusion, and IDH-mutation biology. These observations collectively provided the rationale for focusing subsequent in-depth analyses on gliomas. The overall analytical workflow, including data sources, cohort composition, and all analytical steps, is summarized in Figure S1.
Methods
Data sources
Data on gene expression and clinical features were procured from TCGA and GTEx using the UCSC Xena database. The TCGA pan-cancer dataset encompasses 33 cancer types and was used for FS score computation, prognostic analyses, enrichment, and immune infiltration analyses. Differential expression analyses comparing tumor and matched normal tissue required GTEx data and were restricted to 31 cancer types. (MESO (mesothelioma) and UVM (uveal melanoma) lack matched normal-tissue RNA-seq data in GTEx). Full per-cancer sample sizes for all 33 cancer types are provided in Table S1. Immune cell infiltration data for TCGA were retrieved from the ImmuCellAI and TIMER2 databases and published literature.16 The immunotherapy cohort IMvigor210 was accessed through the R package “IMvigor210CoreBiologies.”17
Expression analysis and risk score analysis of FS genes in the pan-cancer
FoxO signaling genes were sourced from the kyoto encyclopedia of genes and genomes (KEGG) database (https://www.kegg.jp/pathway/hsa04068), encompassing a total of 131 genes. Using the “limma” package in R. FoxO signaling gene expression differences were analyzed between tumor and matched normal-tissue groups across TCGA cancer types. A prognostic value analysis of FS genes in these tumor types was conducted using univariate Cox regression (uniCox) analysis with results visualized using the R packages “survival” and “pheatmap.” The risk score for FS genes was calculated as the sum of individual gene scores across all cancers, assigning +1 for risk factors and −1 for protective factors.
Value of FS scores in pan-cancer prognosis
The ssGSEA function in the R (version 4.1.1) GSVA package was used to calculate individual FS scores for all available TCGA tumor samples across 33 cancer types. FoxO signaling score distributions and their associations with OS were then assessed for each cancer type independently. Kaplan–Meier analyses were undertaken with “survminer” and “survival” in R explored associations between FS scores and patient OS, with P values < .05 considered significant.
Pan-cancer gene enrichment analyses
A computer approach termed GSVA (gene set variation analysis) is used to evaluate gene set activity in gene expression data. For this, using the MsigDB database, gene sets for each of the 50 HALLMARK pathways were obtained (http://software.broadinstitute.org/gsea/msigdb/index.jsp). Gene expression data from the TCGA pan-cancer were utilized, and the gsva function in “GSVA” in R was applied to compute scores for the 50 HALLMARK pathways.
Tumor microenvironment analysis in pan-cancer
Based on a previously published study, scores were calculated for TME-associated pathways, including immune (immune checkpoints, CD8 T effector cells, antigen processing), stromal (Pan_F_TBRs, EMT1-3), and DNA repair (base excisions, mismatches, nucleotide excisions, responses to DNA damage, and DNA replication) pathways.18 The median FS score derived from pan-cancer was employed as a cutoff to classify samples into high- and low-expression groups. Pathway scores in these 2 groups were examined and visualized using “ggpubr” in R, while “estimate” was employed to determine immune scores, stroma scores, and tumor purity for patients diagnosed with tumors.
Infiltration of immune cells in pan-cancer
Samples were categorized into high- and low-expression groups, using the median FS score as cutoff. Pathway scores in these 2 groups were analyzed with “ggpubr” in R. To determine tumor purity, stroma scores, and immunological scores for tumor patients, the R package “estimate” was used.
Immunotherapy response analysis
Using the IMvigor210 dataset, the relationship between FS scores and the effectiveness of immunotherapy treatment was evaluated. The comparison of FS scores between complete/partial response (CR/PR) and stable/progressive disease (SD/PD) groups was conducted using the “limma” package in R. Further, using the TIDE database (http://tide.dfci.harvard.edu), receiver operating characteristic (ROC) curves between FS scores and known immunotherapy predictive indicators were compared.
Unsupervised clustering of the LGG and GBM in the TCGA cohort
The merged glioma cohort comprised TCGA-LGG (n = 522) and TCGA-GBM (n = 165), yielding a combined cohort of 687 glioma samples (male n = 395, female n = 292). The R package “ConsensusClusterPlus” facilitated unsupervised clustering, while the proportion of ambiguous clustering (PAC) metric guided the selection of the optimal number of subtypes. Moreover, subtype distributions were then assessed using principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) analysis.
External validation of the PCA score in independent glioma cohorts
The prognostic performance of the PCA score was externally validated in independent GBM and LGG cohorts obtained from CGGA (Chinese Glioma Genome Atlas; CGGA_301, CGGA_325, CGGA_693) and GEO (GSE4271, GSE16011, GSE61335, GSE74187, GSE83300, GSE108474 for GBM; GSE108474 for LGG). For each dataset, PCA scores were computed de novo by applying principal component analysis (prcomp, R; scale. = TRUE) to the expression matrix of the 2423 prognostic genes identified by univariate Cox regression in the TCGA training cohort, restricted to genes available in both the training list and the respective validation dataset. The PCA score was defined as the sum of PC1 and PC2 loading scores. The optimal survival cut-off was determined with surv_cutpoint (survminer), Kaplan–Meier analysis was performed (log-rank test), and model discrimination was quantified by concordance index (C-index, 95% CI) using the survcomp package (R).
Validation of FS score in immunotherapy cohorts
To validate the association between FS scores and immunotherapy outcomes, ssGSEA-derived FS scores were computed for independent immunotherapy cohorts retrieved from GEO (GSE93157, GSE135222) and published data (Van Allen cohort). Patients were stratified by optimal FS score cut-off (surv_cutpoint, survminer) and Kaplan–Meier analysis was performed (log-rank test) for progression-free survival (PFS).
Statistical analysis
Groups were compared using t-tests, and data were visualized using “ggpubr” or “ggplot2” in R (version 4.1.1). Correlation analysis involved calculating Pearson coefficients, with statistical significance defined as two-tailed P values below .05.
Results
FS gene expression and prognostic values
A total of 131 FS genes were analyzed across 31 TCGA cancer types using matched TCGA tumor and GTEx normal-tissue data. The mRNA levels of FS genes varied among most tumor types (Figure S2). For instance, CCNB1 and CCNB2 were highly expressed in 29 of 31 tumors. The prognostic value of each gene across 33 tumor types was analyzed using uniCox. In 14 of the 33 tumor types, CCNB1 was recognized as a risk factor, while CCNB2 was found to be a risk factor in 12 of the 33 tumor types (Figure 1A and B).
Figure 1.
The prediction regarding FoxO signaling (FS) genetic mutations. (A) Heatmap of FS genes in various tumors. (B) Risk scores of individual FS genes in pan-cancer (Count the number of genes that have significance in Cox analysis in all tumors. Scores of +1 indicate protection and scores of −1 indicate risk).
Determination, distribution, and analysis of survival for FS score
The FS scores for all 33 TCGA cancer types were computed using the ssGSEA method. KIRC exhibited the highest FS score, whereas TGCT had the lowest FS score (Figure 2A). An unfavorable prognosis in patients with LGG was associated with a high FS score, as demonstrated by Kaplan–Meier survival analysis (Figure 2B).
Figure 2.
Distribution and prognostic values of FoxO signaling (FS) scores in pan-cancer. (A) FS score distribution was analyzed across 33 tumor types in the The Cancer Genome Atlas (TCGA) cohort. (B) Kaplan–Meier curves of FS scores across 33 tumors; only significant results (P < .05) are presented.
Enrichment analysis of FS score
Gene set variation analysis was performed to explore pathways related to FS scores. Figure 3 demonstrates the relationship between GSVA scores and FS score expression in pan-cancer. The FS score positively correlated with several malignancy-associated pathways, including TGF-β signaling, PI3K-AKT-mTOR signaling, hypoxia, angiogenesis, and inflammatory response. Moreover, KEGG analysis demonstrated that the FS score in GBM correlated with multiple biological pathways, including FoxO, HIF-1, thyroid hormone, sphingolipid, and prolactin signaling, in addition to growth hormone regulation, endocytosis, and the PD-L1/PD-1 checkpoint pathway (Figure S3A). In LGG, FS score enrichment was observed in pathways including TGF-β, neurotrophin, and FS, as well as actin cytoskeleton regulation, hepatitis B response, PD-L1-associated pathway in cancer, and endocrine resistance (Figure S3B).
Figure 3.
Gene set variation analysis (GSVA) of FoxO signaling (FS) scores. Heatmap of FS scores and 50 HALLMARK pathways in pan-cancer. *P < . 05; **P < . 01; ***P < . 001; ****P < . 0001.
Relationship between FS and TME
The relationship between FS scores and TME-related scores in pan-cancer was evaluated, with a focus on pathways associated with immune response, stromal activity, and DNA repair (Figure 4A). Significant associations were found between FS scores and immune pathways, including regulation of immune checkpoints, CD8 T effector function, and mechanisms of antigen presentation. Stromal pathways such as EMT2, Pan F TBRs (TGFB1-related), and EMT3 were also significantly associated. In GBM and LGG, TME-related scores were assessed between cohorts with high and low FS scores (Figure 4B and C).
Figure 4.
Association of FoxO signaling (FS) scores with the tumor microenvironment (TME). (A) Heatmap of FS scores and TME-related pathways. In glioblastoma (GBM) (B) and Low-grade gliomas (LGG) (C), the boxplots illustrate the variation in TME-related pathways between high and low FS scores. *P < . 05; **P < . 01; ***P < . 001; ****P < . 0001.
Building on previous findings indicating a strong association between FS scores and immune system pathways, the correlation between immune cells in the TME and FS scores was investigated. Analysis using the TIMER2 database indicated marked a positive association between FS scores and infiltration of immune cells, especially tumor-associated and endothelial cells, cancer-associated fibroblasts (CAFs), and macrophages (TAMs) (Figure 5A). According to data from the ImmuCellAI database, TAMs and iTregs are significantly correlated with FS scores in a variety of cancer types (Figure 5B). The FS scores are associated with CD4 + T cells, TAMs, and neutrophils, according to further association analysis with immune cell infiltration data from previous research (Figure S4A). A positive correlation was observed between FS scores and both stromal and immune scores in most tumors, as shown by the TME analysis (Figure S4B). Furthermore, FS scores were found to be positively associated with immune-related gene sets, including genes associated with immunosuppression (Figure 6A), chemokines (Figure 6B), chemokine receptors (Figure 6C), and MHC molecules (Figure 6D).
Figure 5.
Infiltration of immune cells. (A) Associations between FoxO signaling (FS) scores and infiltration of immune cells, shown by TIMER2. (B) Associations between FS scores and immune cell profiles, shown by ImmuCellAI. *P < . 05; **P < . 01; ***P < . 001; ****P < . 0001.
Figure 6.
Associations between FoxO signaling (FS) scores and immune-related genes. The association between FS scores and immune-related genes, including immunosuppressive genes, was examined across pan-cancer datasets (A), chemokines (B), chemokines receptors (C), and MHC genes (D). *P < . 05; **P < . 01; ***P < . 001; ****P < . 0001.
Association of FS with immunotherapy response
The correlation between FS scores and immunosuppressive cells and genes suggested a potential impact of FS on immunotherapy effectiveness. To validate the hypothesis, FS scores were derived from immunotherapy datasets. In the IMvigor210 cohort (urothelial/bladder carcinoma patients receiving anti-PD-L1 atezolizumab therapy, n = 348), patients with SD/PD responses showed higher FS scores than those with CR/PR outcomes (Figure 7A). High FS scores were associated with reduced OS (Figure 7B). The Kaplan–Meier curves indicated that 82% of cases with high FS scores fell into the SD/PD category, in comparison with 66% of those with low scores (Figure 7C). Moreover, area under the curve (AUC) values derived from ROC curves for FS scores and other immunotherapy-related indicators were also presented across multiple datasets to assess immunotherapy efficacy (Figure S4C).
Figure 7.
Associations between FoxO signaling (FS) scores and responses to immunotherapy. (A) The FS score displays variations among the specified response groups in the IMvigor210 cohort (urothelial/bladder carcinoma, n = 348, anti-PD-L1 atezolizumab). (B) Kaplan–Meier curves of FS scores in IMvigor210 cohort. (C) Cases with complete/partial response (CR/PR) and stable/progressive disease (SD/PD) based on high- and low-FS score groups within the IMvigor210 cohort.
To validate the association between FS scores and immunotherapy outcomes, we evaluated the FS score in independent ICI cohorts. In the GSE93157 cohort (n = 65), high FS scores were associated with significantly shorter PFS (log-rank P = .024; Figure S5). Similarly, in GSE135222 (n = 27), high FS scores predicted worse PFS (P = .018; Figure S5). However, in the Van Allen melanoma cohort (n = 39), high FS scores showed a trend toward improved OS (OS, P = .013) and PFS (P < .001; Figure S5). These findings provide independent cohort-based evidence supporting the association between FS activity and immunotherapy efficacy.
FS gene-related glioma subtypes
Based on FS gene expression, 2 glioma subtypes, Clusters A and B, were identified using the R package “ConsensusClusterPlus” (Figure 8A). Based on FS gene expression, 2 glioma subtypes were identified in the merged TCGA glioma cohort using the R package “ConsensusClusterPlus” (PAM algorithm, Euclidean distance, k = 2) (Figure 8A). Cluster A comprised 248 samples (GBM n = 147, 59.3%; LGG n = 101, 40.7%), and Cluster B comprised 439 samples (GBM n = 18, 4.1%; LGG n = 421, 95.9%). A difference in survival between the subtypes was found to be significant, with Cluster A exhibiting lower survival rates than Cluster B (Figure 8B). Figure 8C illustrates the clinical characteristics and FS expression distribution in both clusters. For example, TGFB1, TGFB2, TGFB3, TGFBR1, and TGFBR2 were highly expressed in Cluster A, suggesting that TGF-related pathways are enriched in this cluster, potentially contributing to the poor survival rates observed. Additionally, the majority of GBM patients (147/165, 89.1%) were classified into Cluster A, while Cluster B was predominantly composed of LGG patients (421/439, 95.9%). Cluster A therefore effectively represents a high-grade-enriched subtype, whereas Cluster B represents a low-grade-enriched subtype.
Figure 8.
FoxO signaling (FS) gene-related subtypes. (A) Consensus matrix heatmap of clusters and correlation areas. (B) Kaplan–Meier curves for the clusters. (C) Clinical characteristics and FS-gene levels in the 2 clusters.
Increased scores in pathways involved in immune responses, including interferon-α and -γ responses, were found in Cluster A by further GSVA analysis (Figure 9A). KEGG analysis indicated elevated levels of focal adhesion and mismatch repair pathways in Cluster A (Figure 9B). Reactome analysis showed the upregulation of immune-related pathways, including interleukin and interferon signal transduction, in Cluster A (Figure 9C). The primary distinction between the 2 subtypes appears to stem from differences in the immune microenvironment.
Figure 9.
Enrichment in the 2 clusters. (A) HALLMARK pathways enriched in the 2 clusters. (B) Comparative performance analysis of HALLMARK pathways between the 2 clusters. (C) Reactome pathway enrichment in the clusters.
Significant differences between the 2 clusters were further confirmed by PCA analysis (Figure 10A). Most immune cell levels, calculated using the R software package “CIBERSOFT,” were strongly enriched in Cluster A (Figure 10B), consistent with the findings of the GSVA analysis. Based on these results, we conclude that the TME of patients in Cluster A contains a higher density of both pro- and anti-cancer immune cells.
Figure 10.
Infiltration of immune cells. (A) Sample distribution across the 2 identified clusters. (B) Immune cell infiltration levels were compared across clusters A and B.
Identification of gene subtypes based on DEGs
Screening with the “limma” package, applied to the full merged TCGA glioma cohort (n = 687, LGG + GBM combined), identified 2427 DEGs between Cluster B and Cluster A (|logFC| > 1, P < .05; Table S2). The results of the enrichment analyses for the GO categories Biological Process, Molecular Function, and Cellular Component ([BP], [MF], [CC]), and KEGG pathways are displayed in Figure 11A and B.
Figure 11.
Enrichment analyses of differentially expressed gene (DEGs). (A) GO analyses. (B) KEGG analyses.
The prognostic value of the 2427 DEGs was evaluated using univariate analysis, and genes significantly correlated with OS were identified (P < .05; Table S3). These genes were used in an unsupervised clustering based on prognostic DEG expression, which classified cases into gene clusters A and B (Figure 12A). Cluster B exhibited better survival outcomes than Cluster A (Figure 12B).
Figure 12.
Construction of PCA score. (A) Heatmap of the consensus matrix depicting cluster definitions and inter-cluster correlation areas. (B) Analysis of 2 geneClusters in glioma patients using Kaplan–Meier method. (C) Survival analysis of glioma patients in the 2 geneClusters. (D) The visualization of the Sankey diagram depicting the relationship among different clusters, geneClusters, PCA scores, and the survival outcomes of patients with glioma.
Construction of prognostic model
The PCA method was utilized for the construction of a prognostic model using the DEGs. Patients with high PCA scores showed better survival outcomes (Figure 12C). Using an optimal survival cut-off (surv_cutpoint), patients were classified into PCA-High (n = 471, 68.6%) and PCA-Low (n = 216, 31.4%) groups. Patients in the PCA-High group showed significantly better survival outcomes (Figure 12C). The Sankey plot illustrates the correlation between clusters, gene clusters, prognostic models, and survival status (Figure 12D). Additionally, the PCA score distribution across different clusters and gene clusters is presented (Figure 13A and B). An inverse correlation was observed between PCA scores and immune cell infiltration levels, indicating a potential regulatory influence of PCA scores on the immune microenvironment (Figure 13C). The association between PCA scores and clinical characteristics in glioma was further analyzed. Significantly higher PCA scores were observed in patients with LGG compared with GBM (Figure 13D), consistent with the predominance of LGG in Cluster B and reflecting that the PCA score inversely tracks tumor grade. Regarding survival status, PCA scores were higher in samples from patients who were still alive (Figure 13E). The PCA scores of the male and female groups did not vary significantly (Figure 13F). The results suggest that higher PCA scores are typically related to more positive clinical characteristics.
Figure 13.
The relationship between clinical features and PCA score. (A) The scores of PCA for the clusters identified as A and B. (B) The PCA score observed in geneGroup A and geneGroup B. (C) The relationship between immune cell infiltration and PCA score. Red indicates a positive correlation; blue indicates a negative correlation. The percentage of glioma patients with various disease statuses (D), survival status (E), and gender (F) in high and low PCA score groups.
The role of PCA score in predicting the efficiency of ICI treatment
The expression levels of immune checkpoints have been shown to predict the therapeutic response to ICI treatment. Studies have shown that ICI effectiveness positively correlates with immune checkpoint expression levels.19,20 In the high PCA score group, we observed notable downregulation of common immune checkpoints, specifically TGFB1, PDCD1, LAG3, CTLA4, and CD274 (Figure 14A). Based on the observed findings, resistance to ICI therapy may be associated with elevated PCA scores in glioma patients. Due to the absence of publicly available immunotherapy datasets for gliomas, the TCIA database was utilized for predicting the success of ICI treatment by calculating the immune phenotype score (IPS). Individuals with higher PCA scores showed decreased responsiveness to ICI therapy with PD-1 inhibitors (Figure 14B). Results highlight the potential of PCA scores as predictors of ICI therapeutic efficacy in glioma.
Figure 14.
PCA prediction of immune checkpoint inhibitors (ICI) treatment efficacy. (A) Levels of immune checkpoints in high- and low-PCA-score groups. (B) Association between immune phenotype score (IPS) and PCA scores.
External validation of PCA score in independent glioma cohorts
To validate the prognostic performance of the PCA score beyond the TCGA training cohort, we applied it to independent GBM and LGG datasets from CGGA and GEO. In GBM, significant associations between PCA score and OS were observed in 8 of 9 evaluated cohorts (Figure S6). In LGG, all 4 independent cohorts confirmed the prognostic significance of the PCA score (Figure S7). These results confirm the robust prognostic value of the PCA score in independent glioma cohorts.
Discussion
FoxO signaling is involved in various cellular functions, including differentiation, cell death, growth, and damage repair, and acts as a conduit for oxidative stress.7,8 These functions are crucial in cancer development. Data suggest a link between FS dysfunction and cancer progression and tumor formation.21,22 However, its impact on the TME and immunotherapy remains unclear.
This study investigated the association between FS and TME by analyzing TCGA pan-cancer data. Differential expression and survival significance of FS were assessed. CCNB1 and CCNB2, known target genes of FoxO1,23 were found to be overexpressed in 29 of 31 tumor cases. Specifically, CCNB1 was a risk factor in 14 of 33 tumor types, while CCNB2 was a risk factor in 12 of 33 types. We established an FS scoring system using ssGSEA for various cancer types. Gene set variation analysis results showed a strong correlation between FS scores and immune-related pathways. A significant relationship was observed between FS scores and immune-associated markers, such as immune checkpoints, CD8 T effector cells, and factors involved in antigen presentation.
The relationship between FS scores and aspects of tumor immunobiology across different cancer types has not been thoroughly explored. A strong association was observed between FS scores and immunosuppressive cell populations, TAMs, CAFs, iTregs, and endothelial cells, based on immune cell infiltration data from the ImmuCellAI and TIMER2 databases. These results indicate that a high FS score may be indicative of an immunosuppressive TME, potentially associated with unfavorable prognoses. An immunosuppressive microenvironment could reduce the effectiveness of immunotherapy.24–26 These observations suggest that patients with high FS scores may harbor a less favorable immunotherapy microenvironment; however, these analyses are correlative and based on transcriptomic surrogates. To address this limitation, we validated the association of FS scores with immunotherapy survival outcomes in independent ICI cohorts: high FS scores were associated with significantly shorter PFS in GSE93157 (n = 65, P = .024) and GSE135222 (n = 27, P = .018). Notably, in the Van Allen melanoma cohort, high FS scores showed a trend toward better OS and PFS, which differs from most other tumors. This discrepancy likely reflects tumor-type specificity as melanoma is highly immunogenic with a distinct immune and genomic landscape compared with glioma and most other solid tumors. FoxO signaling may exert context-dependent immunomodulatory effects rather than a uniform immunosuppressive role across all cancers. The small sample size and clinical heterogeneity in the melanoma cohort may also contribute to this divergent result, consistent with the dual and context-dependent functions of FoxO family members. While these findings support the association between elevated FS activity and reduced immunotherapy efficacy in most cancers, prospective clinical validation will be required to establish causality.
We identified 2 glioma subtypes based on FS gene expression (Clusters A and B). Compared to Cluster B, Cluster A individuals exhibited more severe clinical characteristics and poorer survival outcomes. Prediction of immune cell infiltration levels showed that Cluster A had higher immune cell infiltration. We further screened DEGs related to prognosis and constructed a PCA score based on these DEGs. Elevated PCA scores were linked to improved survival. Furthermore, the PCA scores showed negative associations with the infiltration of immune cells. Exploration of the PCA score and clinical features of glioma indicated that high PCA scores often correlated with better clinical characteristics.
Immune checkpoints are key predictors of ICI efficacy.27,28 ICI treatment effectiveness correlates with immune checkpoint expression levels. Lower expression levels of typical immune checkpoint molecules, including CD274, PDCD1, LAG3, CTLA4, and TIGIT, were found in the group with high PCA scores. This suggests reduced responsiveness to ICI therapy in individuals with high PCA scores. This pattern is consistent with a possible association between high PCA scores and reduced predicted responsiveness to ICI therapy. To address the concern about validation, the PCA score was independently confirmed in external GBM and LGG cohorts from CGGA and GEO: significant prognostic associations were observed in 8 of 9 GBM cohorts and all 4 LGG cohorts tested, validating the robustness of the PCA model. Given the limited availability of glioma immunotherapy data, the TCIA immunophenoscore was utilized to estimate the potential response to ICI therapy. Individuals with high PCA scores showed lower predicted ICI responsiveness based on the TCIA immunophenoscore surrogate, a finding consistent with the observed immunosuppressive phenotype in Cluster A.
In conclusion, this comprehensive analysis of FS reveals its extensive influence on the TME, clinical features, and patient outcomes in gliomas. A PCA score was established to evaluate its predictive capacity for prognosis and ICI therapy response. The results emphasize the clinical importance of FS and suggest its use in guiding immunotherapy decisions in glioma management. However, this study has several inherent limitations that should be acknowledged. First, the present study is entirely based on in silico analysis of public transcriptomic datasets from TCGA, GTEx, CGGA and gene expression omnibus (GEO). Although our results were validated in multiple independent cohorts and immunotherapy datasets, the biological mechanisms underlying the association between FS and the tumor immune microenvironment, as well as the proposed clinical implications, remain speculative. Further experimental validation using in situ detection methods and in vivo functional assays is warranted to verify our conclusions and elucidate the underlying molecular mechanisms. Second, the complexity of multistep computational analyses may limit the immediate clinical translation of our findings. Prospective clinical studies with a large sample size and experimental validation are needed to further confirm the clinical utility of the FS-related signatures (FS genes, FS score and PCA score) in glioma prognosis and immunotherapy response prediction. Third, due to the lack of publicly available glioma immunotherapy datasets with detailed clinical follow-up, we used the TCIA immunophenoscore as a surrogate to predict the response to ICI therapy, which may not fully reflect the actual clinical efficacy of ICIs in glioma patients.
Supplementary Material
Contributor Information
Yuetao Wen, Department of Neurology, The Second Affiliated Hospital of Chongqing Medical University, Chongqing, China; Department of Neurosurgery, Chongqing University Jiangjin Hospital, Chongqing, China.
Ya He, Department of Physical Examination Center, Chongqing University Jiangjin Hospital, Chongqing, China.
Jingang Li, Department of Neurosurgery, Chongqing University Jiangjin Hospital, Chongqing, China.
Hui Zhou, Department of Neurosurgery, Chongqing University Jiangjin Hospital, Chongqing, China.
Fei Wang, Department of Neurosurgery, Chongqing University Jiangjin Hospital, Chongqing, China.
Tao Xu, Department of Neurology, The Second Affiliated Hospital of Chongqing Medical University, Chongqing, China.
Yangmei Chen, Department of Neurology, The Second Affiliated Hospital of Chongqing Medical University, Chongqing, China.
Xiaolan Yang, Department of Neurology, The Second Affiliated Hospital of Chongqing Medical University, Chongqing, China.
Supplementary material
Supplementary material is available at Journal of Neuropathology & Experimental Neurology online.
Conflicts of interest
All authors declare that they have no conflict of interests.
Funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.
Data availability
Data and materials generated that are relevant to the results are included in this article. Other data are available from the corresponding author upon reasonable request.
Ethics approval and consent to participate
TCGA and GTEx using the UCSC Xena database belong to public databases. The patients involved in the database provided ethical approval and users can download relevant data for free for research and publish relevant articles. Our study is based on open-source data. The Second Affiliated Hospital of Chongqing Medical University does not require research using publicly available data to be submitted for review to their ethics committee. Therefore, there are no ethical issues or other conflicts of interest.
References
- 1. Xu S, Tang L, Li X, et al. Immunotherapy for glioma: current management and future application. Cancer Lett. 2020;476:1-12. [DOI] [PubMed] [Google Scholar]
- 2. Yu L, Hu X, Zhu H. Predictive value of procollagen c-protease enhancer protein on the prognosis of glioma patients. Heliyon. 2024;10:e28089. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Tomar MS, Kumar A, Srivastava C, et al. Elucidating the mechanisms of Temozolomide resistance in gliomas and the strategies to overcome the resistance. Biochim Biophys Acta Rev Cancer. 2021;1876:188616. [DOI] [PubMed] [Google Scholar]
- 4. Garcia-Fructuoso I, Gomez-Bravo R, Schettini F. Integrating new oral selective oestrogen receptor degraders in the breast cancer treatment. Curr Opin Oncol. 2022;34:635-642. [DOI] [PubMed] [Google Scholar]
- 5. Farhan M, Wang H, Gaur U, et al. FOXO signaling pathways as therapeutic targets in cancer. Int J Biol Sci. 2017;13:815-827. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Lee S, Dong HH. FoxO integration of insulin signaling with glucose and lipid metabolism. J Endocrinol. 2017;233:R67-R79. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Gomes AR, Brosens JJ, Lam EW. Resist or die: FOXO transcription factors determine the cellular response to chemotherapy. Cell Cycle. 2008;7:3133-3136. [DOI] [PubMed] [Google Scholar]
- 8. Vurusaner B, Poli G, Basaga H. Tumor suppressor genes and ROS: complex networks of interactions. Free Radic Biol Med. 2012;52:7-18. [DOI] [PubMed] [Google Scholar]
- 9. Yang JB, Zhao ZB, Liu QZ, et al. FoxO1 is a regulator of MHC-II expression and anti-tumor effect of tumor-associated macrophages. Oncogene. 2018;37:1192-1204. [DOI] [PubMed] [Google Scholar]
- 10. Kerdiles YM, Beisner DR, Tinoco R, et al. FoxO1 links homing and survival of naive T cells by regulating L-selectin, CCR7 and interleukin 7 receptor. Nat Immunol. 2009;10:176-184. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Ouyang W, Beckett O, Flavell RA, et al. An essential role of the Forkhead-box transcription factor FoxO1 in control of T cell homeostasis and tolerance. Immunity. 2009;30:358-371. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Gubbels Bupp MR, Edwards B, Guo C, et al. T cells require FoxO1 to populate the peripheral lymphoid organs. Eur J Immunol. 2009;39:2991-2999. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Stockmann C, Schadendorf D, Klose R, et al. The impact of the immune system on tumor: angiogenesis and vascular remodeling. Front Oncol. 2014;4:69. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Xiufeng G, Zhiping H, Jingzhi W. Advantages and possibility of transcription factors FOXOs in HCC. Cell Biochem Funct. 2025;43:e70126. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Wang K, Chen Q, Liu N, et al. Recent advances in, and challenges of, anti-angiogenesis agents for tumor chemotherapy based on vascular normalization. Drug Discov Today. 2021;26:2743-2753. [DOI] [PubMed] [Google Scholar]
- 16. Thorsson V, Gibbs DL, Brown SD, et al. ; Cancer Genome Atlas Research Network. The Immune Landscape of Cancer. Immunity. 2018;48:812-830.e14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Mariathasan S, Turley SJ, Nickles D, et al. TGFβ attenuates tumor response to PD-L1 blockade by contributing to exclusion of T cells. Nature. 2018;554:544-548. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Zeng D, Li M, Zhou R, et al. Tumor microenvironment characterization in gastric cancer identifies prognostic and immunotherapeutically relevant gene signatures. Cancer Immunol Res. 2019;7:737-750. [DOI] [PubMed] [Google Scholar]
- 19. Liu D, Lin JR, Robitschek EJ, et al. Evolution of delayed resistance to immunotherapy in a melanoma responder. Nat Med. 2021;27:985-992. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Zhang X, Zhang X, Li G, et al. A novel necroptosis-associated lncRNA signature can impact the immune status and predict the outcome of breast cancer. J Immunol Res. 2022;2022:3143511. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Ho KK, Myatt SS, Lam EW. Many forks in the path: cycling with FoxO. Oncogene. 2008;27:2300-2311. [DOI] [PubMed] [Google Scholar]
- 22. Myatt SS, Lam EW. The emerging roles of forkhead box (Fox) proteins in cancer. Nat Rev Cancer. 2007;7:847-859. [DOI] [PubMed] [Google Scholar]
- 23. Takano M, Lu Z, Goto T, et al. Transcriptional cross talk between the forkhead transcription factor forkhead box O1A and the progesterone receptor coordinates cell cycle regulation and differentiation in human endometrial stromal cells. Mol Endocrinol. 2007;21:2334-2349. [DOI] [PubMed] [Google Scholar]
- 24. Bleve A, Durante B, Sica A, et al. Lipid metabolism and cancer immunotherapy: immunosuppressive myeloid cells at the crossroad. Int J Mol Sci. 2020;21:5845. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Joyce JA, Fearon DT. T cell exclusion, immune privilege, and the tumor microenvironment. Science. 2015;348:74-80. [DOI] [PubMed] [Google Scholar]
- 26. Rodriguez-Garcia A, Lynn RC, Poussin M, et al. CAR-T cell-mediated depletion of immunosuppressive tumor-associated macrophages promotes endogenous antitumor immunity and augments adoptive immunotherapy. Nat Commun. 2021;12:877. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Liu L, Bai X, Wang J, et al. Combination of TMB and CNA stratifies prognostic and predictive responses to immunotherapy across metastatic cancer. Clin Cancer Res. 2019;25:7413-7423. [DOI] [PubMed] [Google Scholar]
- 28. Pinato DJ, Howlett S, Ottaviani D, et al. Association of prior antibiotic treatment with survival and response to immune checkpoint inhibitor therapy in patients with cancer. JAMA Oncol. 2019;5:1774-1778. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data and materials generated that are relevant to the results are included in this article. Other data are available from the corresponding author upon reasonable request.