FNDC3B is a Potential Prognostic Biomarker and Correlated With Immune Inltrates in Glioma

Background: Recent discoveries have revealed that bronectin type III domain containing 3B (FNDC3B) acts as an oncogene in various cancers; however, its role in glioma remains unclear. Methods: In this study, we comprehensively investigated the expression, prognostic value, and immune signicance of FNDC3B in glioma using several databases and bioinformatics analysis. RNA expression data and clinical information of 529 patients from the Cancer Genome Atlas (TCGA) and 1319 patients from Chinese Glioma Genome Atlas (CGGA) databases were downloaded for further investigation. A nomogram model based on TCGA low-grade glioma (LGG) dataset was constructed to assess the prognosis value of FNDC3B in glioma by plotting calibration, K-M, and receiver operating characteristic (ROC) curves. The predicted nomogram was validated by CGGA cohorts. Differentially expressed genes (DEGs) were detected by the Wilcoxon test based on the TCGA-LGG dataset and the weighted gene co-expression network analysis (WGCNA) was implemented to identify the signicant module associated with the expression level of FNDC3B. Furthermore, we investigated the correlation between FNDC3B with cancer immune inltrates using TISIDB, ESTIMATE, and CIBERSORTx. Results: Higher FNDC3B expression displayed a remarkably worse overall survival and the expression level of FNDC3B was an independent prognostic indicator for patients with glioma. Based on TCGA LGG dataset, a co-expression network was established and the hub genes were identied. FNDC3B expression was positively correlated to the tumor-inltrating lymphocytes and immune inltration score, and high FNDC3B expression was accompanied by the increased expression of B7-H3, PD-L1, TIM-3, PD-1, and CTLA-4. Moreover, expression of FNDC3B was signicantly associated with inltrating levels of several types of immune cells and most of their gene markers in glioma. Conclusion: glioma and can be regarded as a promising prognostic and immunotherapeutic biomarker for the treatment of glioma.


Introduction
Glioma is the most common primary tumor of the central nervous system in adults and is characterized by high recurrence and mortality rates. According to the World Health Organization (WHO), glioma is typically divided into two principal subgroups: low-grade glioma (LGG; grade II and III) and glioblastoma multiforme (GBM; grade IV; most aggressive and lethal subtype) based on the malignant degree [1]. The median survival is less than two years, and the overall prognosis is poor for glioma patients even after surgical resection, chemotherapy, and radiation therapy (CRT), especially for those with GBM [2]. Even though many advances have been made in adjuvant therapy and surgery in the past few decades, the clinical outcomes have not been signi cantly improved for glioma patients. Apart from the traditional treatment, studies in recent years have revealed the use of novel and effective methods, such as immunotherapy to treat glioma owing to the success achieved from other solid tumors, including lung, bladder, and kidney cancers, and melanoma [3]. However, there is still an urgent need to identify additional immune biomarkers for combination therapy due to the resistance to monotherapy [4].
Furthermore, this may elucidate the mechanism of tumorigenesis and help to identify new molecular targets for treatment.
Fibronectin type III domain containing 3B (FNDC3B, also named FAD104), which belongs to the FNDC3 family, was initially identi ed as a regulator of adipocyte and osteoblast differentiation [5]. FNDC3B is an endoplasmic reticulum transmembrane protein with a single transmembrane domain at the C terminus preceded by nine repeated bronectin type III domains. Its biological function remains largely unknown [6]. Recent research demonstrated that FNDC3B plays a major role in cell adhesion, proliferation, and growth signaling due to the bronectin type III domain, which has the ability to combine with various proteins [7]. For the past few years, emerging evidence has demonstrated that FNDC3B was abnormally expressed in several types of human cancers, including hepatocellular carcinoma, acute myeloid leukemia, colorectal and cervical cancers [8][9][10][11]. For instance, Han reported that FNDC3B expression was correlated with a worse prognosis in cervical cancer, while its carcinogenic effects are still unclear [11]. Notably, a few studies have demonstrated that FNDC3B expression levels are correlated with glioblastoma. Wang and Xu reported that MiR-1225-5p and MiR-129-5p inhibit the malignant glioblastoma cells via targeting FNDC3B [12,13]. Furthermore, a newly integrated analysis of RNA binding proteins in glioma revealed that FNDC3B can not only serve as a useful prognostic biomarker but also promote glioma cell proliferation [14]. However, the overall expression pro le of FNDC3B and its potential role in the development and distinct clinical signi cance of glioma has not been fully elucidated.
Recent advancements in high-throughput sequencing technologies and large-scale cancer genomics databases have enabled a systematic and comprehensive analysis of genes from the perspective of bioinformatics. In the present study, we carried out an intensive analysis for the expression signature of FNDC3B using various publicly accessible databases, as well as applied data mining of the TCGA and CGGA datasets to explore its in-depth prognostic effect. We also investigated the correlations between FNDC3B expression and tumor immune microenvironment (TIM) in LGG patients in order to elucidate the underlying mechanisms and improve molecular diagnosis for glioma patients.

Methods
Gene expression pattern based on ONCOMINE and GEPIA2 FNDC3B expression levels in various cancers were rstly explored by ONCOMINE database (https://www.oncomine.org/) [15], which is currently the largest public cancer microarray database and integrated data-mining platform. The threshold for ONCOMINE was set according to the default settings of P < 0.0001, fold change > 2, and Gene Rank < Top 10%. Gene expression pro ling interactive analysis 2 (GEPIA2) is another useful web-based tool (http://gepia.cancer-pku.cn/) that contains RNA sequencing data based on 9,736 tumor and 8,587 normal control samples from The Cancer Genome Atlas (TCGA) Page 4/28 and The Genotype-Tissue Expression (GTEx) [16], providing differential expression analysis, correlation analysis, survival analysis, and custom data analysis. GEPIA2 contains 518 LGG samples, 163 GBM samples, and 207 normal brain samples. FNDC3B expression was compared between LGG or GBM and normal tissues by Student t-tests. Samples were considered to be signi cant with p < 0.05 and fold change > 2. Protein expression of FNDC3B in glioma and normal brain tissues were evaluated based on immunohistochemistry data from the Human Protein Atlas (HPA) (https://www.proteinatlas.org/). The mRNA expression of FNDC3B in various human cancer cell lines were obtained from Broad Institute Cancer Cell Line Encyclopedia (CCLE).

Data source and processing
Gene expression data and corresponding clinical information for glioma patients were downloaded from TCGA (LGG and GBM) and Chinese Glioma Genome Atlas (CGGA) (mRNAseq_325, mRNAseq_693 and mRNA_array_301) database. The data from the TCGA was applied to explore the prognostic role of FNDC3B in gliomas, and the three CGGA cohorts were used to validate the results. The DNA methylation along with transcriptional data of FNDC3B for patients with LGG from the TCGA database was downloaded via the cBio Cancer Genomics Portal (cBioPortal) website (http://www.cbioportal.org/). The detailed methodology of the study is shown in Figure 1.

Clinicopathological correlation and prognosis analysis
Coe cients of Cox regression were determined by data mining among 21 TCGA cancer types using OncoLnc (http://www.oncolnc.org/) to compare FNDC3B expression in different tumors. The gene expression data and corresponding clinical information from the TCGA and CGGA were used to evaluate the prognostic role of FNDC3B in gliomas. Patients with incomplete clinical information were eliminated. We explored the correlation between FNDC3B expression and various clinical features using the Wilcoxon test in R version 3.6.3. Kaplan-Meier analysis and log-rank test were implemented to evaluate the associations of FNDC3B and clinical variables with overall survival (OS). Univariate and multivariate Cox regression analyses were performed to further examine whether FNDC3B expression was a signi cant factor associated with OS when adjusted by clinical variables (age at initial pathological diagnosis, gender, neoplasm histologic grade, IDH mutation status, etc.). "Survival" and "survminer" packages in R software were used for the stepwise variable selection and Cox model construction.

Construction and evaluation of a nomogram
Nomogram is widely used as a predictive model for cancer patients and can provide prognostic risk individually and intuitively [17]. Based on the TCGA-LGG dataset, we constructed a prognostic nomogram model to predict the probability of 2-, 3-, and 5-year OS using "rms" package in R. The nomogram combined the expression level of FNDC3B with traditional clinical parameters (age, grade, and IDH status) and formulated the scoring criteria for all the parameters in the regression equation based on their regression coe cients. Then, the summed score for each patient was converted into the probability of the outcome time by using the nomogram. The performance and prediction e ciency of the nomogram were evaluated via plotting the calibration, K-M and receiver operating characteristic (ROC) curves of the three CGGA validation datasets.

Screening of differentially expressed genes (DEGs)
The median value was used to create a categorical dependent variable based on FNDC3B expression level. For the TCGA-LGG dataset, DEGs between the high and low FNDC3B groups were screened using Wilcoxon test (screening criteria: P < 0.01, FDR < 0.05, and |logFC| > 1). Volcano plot of all DEGs was generated using R. For accurate results, we set the average expression level to at least 1 for the raw DEGs. Then, the ltered DEGs were selected for further analysis. Heatmap was generated by "ComplexHeatmap" package in R. To identify biological pathway differences between the high and low FNDC3B groups, gene set enrichment analysis (GSEA) and kyoto encyclopedia of genes and genomes (KEGG) were performed on the screened DEGs using "org.Hs.eg.db", "clusterPro ler", "enrichplot" and "ggplot2" R packages. The correlations between FNDC3B with DEGs in pan-cancer were obtained by using Gene_Corr module of TIMER2.0 (tumor immune estimation resource, version 2) (http://timer.cistrome.org/). P < 0.05 was considered to be a signi cant enrichment.

Co-expression network creation and hub genes identi cation
To reveal the correlation between genes and identify the ones with signi cant relationships, 2,099 DEGs between the high and low FNDC3B expression groups were used to construct a weighted co-expression network. Patients with incomplete expression information were removed, and the remaining samples were implemented to construct the network. In order to ensure the reliability of the network structure, we set gene expression value larger than 1 in at least 10% of all the samples, and the average expression level was at least 0.5 for the raw DEGs. A scale-free gene co-expression network based on 673 ltered DEGs was constructed using the R package "WGCNA". Hierarchical clustering tree was created based on a dissimilarity measure (1-TOM), and genes with similar expression patterns were merged into the same module. The most relevant module was revealed by calculating the correlation between modules and the FNDC3B group; the genes in the most signi cant module were extracted to determine the target genes. For the selected module, Gene Ontology (GO) was performed in R using the packages "org.Hs.eg.db", "AnnotationDbi", "enrichplot" and "ggplot2" with q-values less than 0.0001. The Search Tool for the Retrieval of Interacting Genes Database (STRING) version 11 was applied to generate the PPI network and the combined score > 0.4 was used as the cut-off criterion. Cytoscape version 3.8.2 was employed to visualize the molecular interaction networks and biological pathways. Hub genes were identi ed via CytoHubba plugin with the top 10 MCC values. We used GeneMANIA database (http://genemania.org/) to build the gene-gene interaction network for hub genes in terms of physical interactions, co-expression, shared protein domains, pathways, predicted interactions, and colocalization, as well as to predict their biological functions. The relationships of the expression levels between the hub genes and FNDC3B were shown by scatterplots using "ggplot2" R package.

Analysis of FNDC3B-associated immunomodulators
To investigate the immune in ltration of FNDC3B in different cancers, TISIDB database (http://cis.hku.hk/TISIDB/) was applied to infer the correlations between 28 types of tumor-in ltrating lymphocytes (TILs) and FNDC3B expression. TISIDB integrates a variety of data sources in tumor immunology, including abundant human cancer datasets from TCGA and text mining results from PubMed. Spearman correlation test was implemented to estimate the association between FNDC3B and TILs. Furthermore, Estimation of Stromal and Immune cells in Malignant Tumor tissues using Expression (ESTIMATE) algorithm was employed to evaluate the stromal score, immune score, and ESTIMATE score for each sample using the downloaded data. Moreover, CIBERSORTx (https://cibersortx.stanford.edu/) was used to assess the relative variations of 22 types of tumor-in ltrating immune cells between the high and low FNDC3B expression groups in LGG. Packages "ggplot2", "ggpubr" and "ggExtra" in R were applied to investigate the correlation between FNDC3B and immune checkpoints based on TCGA-LGG dataset. The associations between gene markers of the signi cant immune cells and FNDC3B expression were also determined by the correlation analysis function in GEPIA2. Differences with a P-value < 0.05 were considered signi cant in all tests.

Statistical Analysis
All statistical analysis was performed in R software and P < 0.05 was considered statistically signi cant. Wilcoxon test was used to compare the differences for clinical characteristics or immune scores grouped by FNDC3B expression. Univariate (Log-rank test) and multivariate Cox regression (cox proportional hazard, coxph) analyses were performed to assess clinical traits associated with OS.

FNDC3B mRNA expression in various cancers
We rst assessed the expression of FNDC3B in different tumors and normal tissues of multiple cancer types using the Oncomine database. The results showed that abnormal FNDC3B expression was retrieved from a total of 368 datasets. Among them, FNDC3B expression levels were signi cantly upregulated in tumor tissues in 43 datasets, including brain and central nervous system (CNS), head and neck, esophageal, kidney, cervical, bladder, and colorectal cancers, etc. (Figure 2A, P < 0.0001, Fold Change > 2, and Gene Rank < Top 10%). In addition, its expression in leukemia, lymphoma, breast cancer, and sarcoma lymphoma was shown to be downregulated in multiple datasets. In summary, FNDC3B is generally upregulated in several tumors. In the brain and CNS cancers dataset, there were seven studies on the upregulation of FNDC3B and no studies on its downregulation. Comparison of FNDC3B across the seven studies showed a Median Rank = 302, suggesting that FNDC3B was highly expressed in glioma tissues and concentrated both in LGG and GBM ( Figure 2B, P <0.0001). As for the outlier analysis of FNDC3B, 822 different studies on FNDC3B have been included in Oncomine database. Among them, 11 studies indicated the upregulation of FNDC3B and four studies the downregulation in brain and CNS cancers ( Figure 2A). Then, we evaluated the differentially expressed level of FNDC3B in TCGA pan-cancer data using GEPIA2. Our ndings revealed that an elevated expression of FNDC3B was involved in a variety of tumors ( Figure 2C). In TCGA LGG and GBM cohorts, FNDC3B expression was signi cantly higher in tumors compared to matched normal tissues ( Figure 2D). Speci cally, FNDC3B expression was 2.71-fold in LGG and 7.16-fold in GBM vs. normal brain tissue. Moreover, the expression of FNDC3B was negatively correlated with FNDC3B DNA methylation; the methylation levels were reduced in the FNDC3B high group based on the TCGA-LGG dataset ( Figure 2E-2F). These results suggested that FNDC3B may be negatively regulated by methylation, which may lead to its high expression in glioma samples.
FNDC3B protein expression was explored using HPA database. The immunohistochemistry result revealed upregulated FNDC3B in glioma samples ( Figure 2G). In human cancer tissues, FNDC3B protein expression in glioma was ranked as the top 12 out of 20 distinct cancer types (Supplementary Figure 1A). Furthermore, we detected the FNDC3B mRNA level in human normal tissues using the GTEx database, and FNDC3B expression was mainly found in lung, adipose tissue, thyroid gland, endometrium, and ovary. It is worth noting that the brain displayed the lowest expression levels of global FNDC3B transcript across all normal tissues (Supplementary Figure 1B). In addition, we systematically elucidated the expression levels of the FNDC3B in different cancer cells by querying the CCLE database and found that FNDC3B was highly expressed in glioma cell lines (Supplementary Figure 1C).

High expression of FNDC3B predicts poor prognosis of glioma
OncoLnc and GEPIA2 are free online resources and databases for the analysis and visualization of datasets from the TCGA and GTEx projects. To examine the function of FNDC3B on OS in various cancers, we used OncoLnc online tool to perform Cox regression analysis. We found that FNDC3B expression in LGG was ranked rst among 21 different cancer types based on the FDR correction (Table  1). Moreover, we analyzed the relationships between FNDC3B expression and prognostic values in 33 types of cancer using GEPIA2 databases. As shown in Supplementary Figure 2, high FNDC3B expression levels were associated with poorer prognosis of OS and disease-free survival (DFS) in adrenocortical carcinoma (ACC), GBM, kidney Chromophobe (KICH), LGG, liver hepatocellular carcinoma (LIHC), mesothelioma (MESO), and pancreatic adenocarcinoma (PAAD); OS in cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC) and lung adenocarcinoma (LUAD); DFS in colon adenocarcinoma (COAD) and uveal Melanoma (UVM). A high FNDC3B expression was correlated with a better prognosis of OS and DFS in skin cutaneous melanoma (SKCM) as well as OS in acute myeloid leukemia (LAML).
To assess the prognostic signi cance of FNDC3B in glioma patients from TCGA and CGGA, samples were rst split into two groups according to the median expression of FNDC3B for each dataset. Among pan-glioma, LGG, and GBM in the TCGA datasets, patients with higher FNDC3B levels presented shorter OS and DFS ( Figure 2H) compared to patients expressing low levels of FNDC3B. Similarly, high FNDC3B expression was signi cantly associated with poor prognosis in all the three CGGA datasets (seq_325, seq_693, array_301) (Supplementary Figure 3). In particular, highly expressed FNDC3B was signi cantly related to reduced DFS in GBM (p<0.001), however, just marginally correlated with worse OS (P<0.05). We postulate that the less obvious but signi cant results of GBM may be due to the insu cient statistical power of a small sample size. Furthermore, we investigated the associations between FNDC3B expression and clinical characteristics, such as age, gender, grade, and isocitrate dehydrogenase (IDH) status. FNDC3B expression was higher in high-grade and IDH wildtype patients; there was no signi cant difference between age and gender based on the TCGA datasets ( Figure 2I). Our ndings revealed that higher FNDC3B expression is closely correlated with the malignant clinical characters of gliomas.

Construction and validation of a prognostic nomogram
A quantitative prognostic nomogram model to predict individual survival chances was established based on the TCGA-LGG dataset using Cox regression ( Figure 3A). According to the stepwise Cox multivariate regression analysis, age, grade, IDH status, and FNDC3B expression were features that were included in the nomogram, and the risk scores were calculated based on that model. The concordance index (Cindex) for OS prediction was 0.775, indicating high predictive performance of the model. A calibration curve was implemented to re ect the degree of consistency between the predicted risk and actual occurrence risk, and it can be used to estimate the accuracy of the model in predicting the probability of an individual outcome in the future [18]. In our study, the calibration curve showed acceptable agreement between nomogram-predicted and observed 2-, 3-and 5-year OS in the CGGA_325, CGGA_693, and CGGA_301 validation cohorts ( Figure 3B).
The patients in the training dataset were divided into high-risk and low-risk groups according to the positive and negative values of the risk score, respectively. The K-M survival curve showed a good discriminating ability of the nomogram (P = 1.4e-4) ( Figure 3C). Moreover, the area under the ROC curve for OS was 0.755, indicating a reliable predictive ability in the TCGA LGG dataset ( Figure 3D). The CGGA_325, CGGA_693, and CGGA_301 datasets were used to validate the performance of the nomogram. A risk score for each patient was generated by the same method. Consistently, the patients in the high-risk group had a notably poorer prognosis in all the three validation datasets (p < 0.0001) ( Figure 3C). The area under the curve (AUC) for CGGA_325, CGGA_693, and CGGA_301 OS was 0.8, 0.816, and 0.807, respectively ( Figure 3D). In conclusion, these results indicated that the nomogram had adequate performance in predicting the OS of glioma patients.

DEGs identi cation and weighted co-expression network construction based on FNDC3B expression
To further elucidate the role of FNDC3B expression in the glioma microenvironment, the median expression value was used to create a categorical variable for the TCGA-LGG cohort. The DEGs between the two groups were detected using Wilcoxon test. After setting FDR < 0.05 and fold change ≥ 2 in either direction, a total of 2,099 DEGs, including 1,631 upregulated and 468 downregulated genes were screened out in FNDC3B highly expressed group compared to the low expressed group ( Figure 4A). In order to obtain more reliable results, 670 DEGs were ltered after setting the mean expression value to 1 for the raw DEGs ( Figure 4B). GSEA was conducted to assess the potential functions of these DEGs. Our results suggested that various immune-related gene signatures were enriched in LGG samples, such as response to cytokine, cytokine secretion, immune system process, and in ammatory response ( Figure 4C). According to the KEGG analysis, DEGs were mainly concentrated in PI3K-Akt, p53, and MAPK signaling pathways ( Figure 4D). We used TIMER2.0 to explore the relationship between FNDC3B with the top 10 most signi cantly up or downregulated DEGs in pan-cancer. The corresponding heatmap showed that the correlation was consistent with the gene expression direction of the 20 DEGs only for LGG ( Figure 4E).
WGCNA was applied to build a co-expression network based on the 2,099 DEGs. Before constructing the co-expression network, we screened the DEGs by setting expression value larger than 1 in at least 10% of all the samples, and the average level at least 0.5. Finally, 673 DEGs were ltered and selected for subsequent analysis. Power eight was chosen as the appropriate soft threshold because it was the rst value to make the degree of independence reach 0.90 and the corresponding average connectivity was close to zero ( Figure 5A-5B). A total of ve gene modules were excavated (genes in the grey module that were not co-expressed; Figure 5C). As shown in Figure 5D, the turquoise module was the most relevant in FNDC3B expression level (R = 0.51, p = 7e-35). Therefore, the 282 hub genes included in the turquoise module were extracted for further analysis. GO enrichment analyses of these genes indicated that the response to interferon-gamma, neutrophil activation and degranulation, regulation of immune effector process, antigen processing and presentation, T cell activation, leukocyte migration, lymphocyte proliferation, mononuclear cell proliferation, interleukin-8 production and acute in ammatory response was related to FNDC3B-mediated immune events. The top 30 GO items, as ranked by their P-values, are shown in Figure 6A. A total of 210 nodes and 1,061 edges were mapped for the turquoise module genes in the PPI network (Supplementary Figure 4). Using CytoHubba in Cytoscape plug-in, we selected the top 10 genes ranked by the MCC method as hub genes, including TLR2, TLR7, PTPRC (CD45), CCR1, CCL5, TLR1, FN1, VCAM1 (CD106), CXCL10, and TLR6. They were immune-related genes and positively correlated with FNDC3B (R > 0.3 and P < 0.0001; Figure 6B).
A gene-gene interaction network for the 10 hub genes was built, and their functions were analyzed through the GeneMANIA database ( Figure 6C). We found that toll-like receptor signaling pathways, ERK1 and ERK2 cascade and NIK/NF-kappaB signaling, were enriched in LGG. Then, OncoLnc online tool was applied to investigate the function of these hub genes on OS of LGG. The results showed that all hub genes were independent risk factors for evaluating OS (Table 3), and the K-M curves based on GEPIA2 displayed that higher expression of these genes predicted shorter OS in LGG (Supplementary Figure  5). Furthermore, the gene expression analysis of the LGG samples from TCGA showed that the combined expression of the 10 hub genes had a signi cant effect on overall survival ( Figure 6D).

Relationship between FNDC3B expression and tumor immune in ltrates
Since previous studies have reported that TILs are independent predictors in cancers [19,20], we used TISIDB database to infer the correlations between the expression of FNDC3B and the abundance of 27 types of TILs across TCGA pan-cancers. As shown in Figure 7A, FNDC3B expression was positively correlated with TILs in several human cancer types, especially in LGG. Moreover, we investigated the associations between FNDC3B expression and immune subtypes across human cancers, and the landscape of correlations between FNDC3B expression and immune subtypes in different types of cancer ( Figure 7B). Among all cancer types, LGG showed the most signi cant results via Kruskal-Wallis test (p = 1.26e-23). In TISIDB, we further analyzed FNDC3B expression in different immune subtypes of LGG. We found FNDC3B was mainly expressed in three types, including C3 (in ammatory type), C4 (lymphocyte depleted type), and C5 (immunologically quiet type). FNDC3B expression was the highest in the C3 (in ammatory) type and the lowest in the C5 (immunologically quiet) type ( Figure 7C). These results  Figure 8A). Thus, the six genes were signi cantly upregulated in the FNDC3B high group compared with the low group ( Figure 8B). In summary, these results suggested that FNDC3B was correlated with clinically relevant immune checkpoint molecules in glioma.
Subsequently, to investigate whether FNDC3B expression was correlated with immune in ltration patterns in LGG, we compared the degree of immune cell in ltration between high and low expression groups using the ESTIMATE algorithm. The immune, stromal, and ESTIMATE scores were higher in the highexpression group than in the low-expression group ( Figure 8C). Furthermore, we explored the proportions of 22 types of immune cells for LGG using CIBERSORTx to acquire a deeper understanding of the relationship between FNDC3B expression and tumor immune in ltrates. Among the 529 TCGA-LGG samples, 265 samples were in the high expression group and 264 samples in the low expression group. Figure 8D shows the differences in the proportions of the 22 subpopulations of immune cells in these two groups. Naive B cells, plasma cells, naive CD4 T cells, resting memory CD4 T cells, activated memory CD4 T cells, follicular helper T cells (Tfh), M1 and M2 macrophages, activated dendritic cells, and neutrophils were the main immune cells affected by FNDC3B expression. Among them, there were more proportions of resting memory CD4 T cells (p < 0.001), activated memory CD4 T cells (p < 0.01), M1 macrophages (p < 0.01), M2 macrophages (p < 0.01), dendritic cells activated (p < 0.01), and neutrophils (p < 0.0001) in the high expression group. In contrast, the proportions of naive B cells (p < 0.01), plasma cells (p < 0.0001), naive CD4 T cells (p < 0.05), and follicular helper T cells (p < 0.05) were lower in the high expression group compared with the low expression group. The correlation matrix heatmap of 22 immune in ltration cells in LGG samples was shown in Figure 8E. Moreover, we analyzed the association between FNDC3B expression and gene markers of various TILs, including B cell, plasma cells, T cell, CD4+ T cell, Tfh, M1 and M2 macrophages, dendritic cells, and neutrophils (Table 4). Overall, FNDC3B expression was strongly positively correlated with gene markers of B cells, T cells, M1 and M2 macrophages, dendritic cells, and neutrophils for TCGA-LGG and TCGA-GBM.

Discussion
Glioma is the most common primary intracranial neoplasm, accounting for approximately 80% of malignant brain tumors. The limitations of classical treatments lead to a poor OS [21]. New progress in brain tumor research suggests that immunotherapy is a powerful tool for the treatment of gliomas [22][23][24]. Therefore, the identi cation of novel effective biomarkers for early diagnosis and promising immunerelated therapeutic targets for glioma patients has become imperative in clinical practice. Recently, several studies have reported that FNDC3B is an oncogene in various cancers, including glioma [14]. To our knowledge, the expression pattern and biological function of FNDC3B in glioma have not been studied in detail, and its possible prognostic value in glioma remains to be explored. Based on the integrated bioinformatics analysis, this is the rst report to comprehensively analyze FNDC3B expression pro les and its correlation with immune in ltrates in gliomas.
In the present study, we found that FNDC3B was highly expressed in glioma tissues by mining multiple databases, and the expression levels of FNDC3B increased with the level of the malignant degree, which was also con rmed in another study [12]. These results suggested that FNDC3B could serve as a promising molecular marker for predicting the degree of malignancy in brain glioma. K-M plots indicated that patients with high FNDC3B expression had worse OS and DFS than those with low expression in gliomas. Furthermore, univariate and multivariate Cox analysis demonstrated a positive correlation between FNDC3B expression and poor prognosis of patients with glioma. To further apply FNDC3B in clinical treatments, a prognostic nomogram for personalized prediction was constructed, integrating the FNDC3B expression level with signi cant clinical parameters (age, grade, and IDH status). The C-index, AUC value, and calibration curve for TCGA training and three CGGA validation datasets showed that our nomogram was reliable and performed adequately. In general, we identi ed and validated the expression level of FNDC3B as a useful and independent prognostic biomarker for glioma.
To further investigate the functions of FNDC3B in glioma, we performed GSEA analysis using DEGs based on TCGA LGG data. The results showed that multiple immune-related pathways were enriched in the FNDC3B high-expression group, such as cellular response to cytokine stimulus, in ammatory response, and immune system process. In the KEGG analysis, PI3K-Akt, p53 and MAPK signaling pathways participated in tumor development. We found that the expression of FNDC3B correlated with that of multiple T cell markers (Th1, Th2, and Th17) in LGG. This suggested that FNDC3B may be involved in the regulation of T cell response in glioma. In addition, WGCNA was performed on the DEGs to nd more valuable clues. Finally, 10 upregulated genes were identi ed as hub genes, including TLR2, TLR7, PTPRC, CCR1, CCL5, TLR1, FN1, VCAM1, CXCL10, and TLR6. Previous studies reported that T cells have the ability to directly recognize danger signals through the expression of toll-like receptors (TLRs) [25]; interactions between CCR1 and CCL5 contribute to T-cell activation [26], and CXCL10 is usually considered to be a pro-in ammatory chemokine that enhances recruitment of CD8+ and Th1-type CD4+ effector T cells to infected or in amed nonlymphoid tissues [27]. TILs are independent predictors in cancers [20]. Our ndings showed that FNDC3B was strongly positively correlated with immune in ltration in LGG and GBM among all cancer types in the database, especially in the cytotoxic T cells and anti-tumor associated immune cells, such as central memory CD8 T cell, effector memory CD8 T cell, central memory CD4 T cell, regulatory T cell (Treg) natural killer cell (NK), natural killer T cell (NKT), memory B cell, and macrophage. The comparative analysis of FNDC3B gene expression in different immune subtypes in LGG suggested that FNDC3B may be strongly linked to immunological properties in the tumor microenvironment. A previous study reported that the immune microenvironment affected the gene expression of tumor tissues, and the degree of stromal and immune cell in ltration in uenced prognosis [32]. In our study, the FNDC3B high-expression group displayed higher values for immune, stromal, and ESTIMATE scores than the FNDC3B low-expression group, indicating that the high expression level of FNDC3B is positively related to immune in ltration in gliomas. Moreover, consistent with the TISIDB results, we found that memory CD4 T cell, macrophages M1 and M2, and neutrophils were enriched in the FNDC3B high group based on CIBERSORT analysis. Additionally, a relatively strong correlation between FNDC3B expression and gene markers of T cells, CD4+ T cell, follicular helper T cells, and dendritic cells indicated the potential role of FNDC3B in regulating T cell function in LGG and GBM.
The FNDC3B expression was positively correlated with genes of immune checkpoint inhibitors, suggesting that FNDC3B could be a regulatory factor of various immune checkpoints in glioma. The correlation analysis showed that FNDC3B was mostly positively correlated with B7-H3, which was associated with a suppressive effect on T-cell activities in various tumors [33]. In recent years, studies have shown that B7-H3 is a promising novel target for glioma immunotherapy [34,35]. Nehama and colleagues reported that B7-H3 is highly expressed in more than 70% of GBM samples and that B7-H3redirected chimeric antigen receptor T (CAR-T) cells can effectively control tumor growth [22]. Currently, tumor immunotherapy has attracted great attention. Hence, our ndings indicated that whether we can participate in the immune checkpoint by inhibiting the expression of FNDC3B, and can FNDC3B be served as a powerful immune checkpoint blockade combination therapy to increase e cacy and reduce side effects?
There are several limitations in the current research. First, it was mainly based on online public databases and computational methods. Nevertheless, integrated bioinformatic analyses, such as multiple datasets and different databases strengthen the conclusion of this study. Second, more investigations are needed to identify the expression and function of FNDC3B as well as their correlations with immune cell in ltration; thus, further clinical and experimental studies in the laboratory are required for verifying its role in glioma.
In conclusion, our comprehensive analysis revealed that FNDC3B was upregulated in glioma, while increased FNDC3B expression predicted an unfavorable prognosis. Moreover, FNDC3B is associated with the in ltration of various immune cells, and it may play a vital role in the tumor immune microenvironment of glioma. Therefore, we reported that FNDC3B is a possible prognostic biomarker and an immune-related therapeutic target for glioma, which will be useful for clinical applications.       Supplementary Files