RNA-seq data of CFI in multiple tumor tissues and corresponding normal tissues were obtained from The Cancer Genome Atlas (TCGA; http://cancergenome.nih.gov/) and Genotype-Tissue Expression (GTEx; https://www.gtexportal.org/home/datasets) and analyzed using R programming language. Gene Expression Profiling Interactive Analysis (GEPIA; http://gepia.cancer-pku.cn/index.html) was used to analyze survival differences between the CFI^high and CFI^low glioblastoma multiform (GBM) and low-grade glioma (LGG) patients. The glioma datasets with mRNA expression and clinical data, including PRS_type, WHO grade, age, chemotherapy status, IDH_mutation_status, 1p/19q_codeletion_status, and histology grade, were acquired from the Chinese Glioma Genome Atlas (CGGA; http://www.cgga.org.cn/) (Dataset ID: mRNAseq_693 and mRNAseq_325). The correlation between CFI expression and clinical features was evaluated by Cox regression analyses. Variables with P < 0.05 in the univariate analysis were further confirmed in the multivariate analysis.

The gene set enrichment analysis (GSEA) tool (Han et al., 2019) was used to identify enriched gene sets according to their association with CFI expression level. Gene set permutations were performed 1,000 times to screen for the CFI-related significant biological pathways. A normalized P-value <0.05 and FDR q-value <0.05 were considered statistically significant.

The prognostic value of each clinical parameter listed above was determined by the receiver operating characteristic (ROC) analysis using the R “survivalROC” package. The sensitivity and specificity of prognostic signatures in predicting the overall survival (OS) were evaluated in terms of the area under the curve (AUC) of ROC. A prognostic nomogram was then constructed using the rms package of R in order to predict 1-, 2-, and 3-year OS of glioma patients based on the CGGA dataset. Pearson's correlation analysis was performed to identify CFI-related genes using the CGGA dataset. P < 0.05 was considered as statistically significant.

The normal human astrocyte cell line NHA was purchased from Lonza, and human glioma cell lines A172, U118, LN229, U251, T98G, and U87 were obtained from the American Type Culture Collection (ATCC). All cell lines were cultured in as previously described (Cai et al., 2020). A total of 134 GBM tissues, 30 glioma tissues (grades I–III), and 9 paired normal brain tissues were collected from the First Affiliated Hospitals of Nanjing Medical University (Nanjing, China) and 904th Hospital of Joint Logistic Support Force of People's Liberation Army (Wuxi, China), from 2014 to 2019. The tissues were snap frozen in liquid nitrogen after storing at −80°C overnight. The clinical and pathological characteristics for 134 GBM patients are shown in Supplementary Table 2. The tissue microarray (TMA) for immunohistochemical staining was constructed based on 9 NBTs, 7 grade I, 8 grade II, 15 grade III, and 18 grade IV glioma specimens. The glioma tissues were confirmed histologically by two independent neuropathologists according to the WHO criteria. The study protocol was reviewed and approved by The Institutional Review Board of Nanjing Medical University and Anhui Medical University. All patients had signed written informed consent before the surgery.

Quantitative real-time PCR (qRT-PCR) and western blotting were performed as described previously (Cai et al., 2020). The primers for CFI were as follows: forward: 5′-CTCAGCAGAGACAAAGAC-3′, reverse: 5′-GTGGAAGCACAGAAATAAC-3′. GAPDH served as the internal control and the primers were as follows: forward 5′-GAAGGTGAAGGTCGGAGTC-3′, reverse: 5′-GAAGATGGTGATGGGATTTC-3′. Antibody against CFI was obtained from Novus Biologicals. Anti-VEGFR2, anti-AKT, anti-phospho-AKT (Thr308), anti-p38 MAPK, and anti-phospho-p38 (Thr180/Thr182) primary antibodies were obtained from Cell Signaling Technology. Anti-FAK and anti-phospho-FAK (Tyr576/577) primary antibodies were purchased from Abcam. Anti-β-actin antibody, which was purchased from Sigma-Aldrich, acted as the negative control. All experiments were conducted in triplicate and repeated at least three times.

Briefly, paraffin-embedded tissue sections (4-μm-thick) were baked at 60°C for 1 h, deparaffinized in xylene, and rehydrated in graded concentrations of ethanol (100, 95, and 85% for 5 min each). The sections were placed in pH 6.0 citric buffer for 20 min, treated with 3% hydrogen peroxide in phosphate-buffered saline, and incubated with goat serum. Each section was incubated with primary and secondary antibodies separately. Chromogen was added and the specimens were counterstained with hematoxylin. Immunohistochemical images were taken using a Leica DM IRE2 microscope (Leica Microsystems Imaging Solutions Ltd., Cambridge, United Kingdom) and analyzed by Image-Pro Plus v6.0 software (Media Cybernetics Inc., Bethesda, MD, USA). The integrated optical density (IOD) was calculated according to the following grading system: staining extensity was categorized as 0 (≤5% positive cells), 1 (>5% and ≤25% positive cells), 2 (>25% and ≤50% positive cells), or 3 (>50% positive cells), and staining intensity was categorized as 0 (negative), 1 (weak), 2 (moderate), or 3 (strong). All the samples were examined and scored by two experienced pathologists. Cases with discrepancies in the scores were discussed to reach a consensus.

The pCDNA3.1-CFI recombinant plasmid was constructed by cloning the CFI cDNA into the pCDNA3.1 vector. Two GBM cell lines stably expressing shCFI or shCtrl were established using a lentiviral packaging kit (Santa Cruz Biotechnology) and selected with 5 mg/ml puromycin 48 h after transfection. The cells were transfected using Lipofectamine 3000 (Invitrogen) according to the manufacturer's protocol.

The proliferation rate of glioma cells was examined by the CCK-8 kit (Beyotime) following the manufacturer's instructions. Briefly, the suitably transfected cells were seeded into 96-well plates at the density of 5 × 10³ cells/well, and 10 μl CCK-8 reagent was added per well at 24, 48, 72, and 96 h of culture. The absorbance at 450 nm was measured to calculate the percentage of viable cells.

The cells were seeded into a six-well plate at the density of 5 × 10² cells/well and cultured for 14 days. The colonies were washed thrice with PBS, fixed with 100% methanol for 20 min, and stained with 1% crystal violet in 20% methanol for 15 min. After washing thrice with PBS, the number of colonies was counted and the averages were calculated.

The suitably treated glioma cells were seeded onto a coverslip in six-well plates at the density of 3 × 10⁵ cells/well. The monolayer was scratched using a sterile 20-μl pipette tip, and the wound area was photographed at 0 and 24 h under a light microscope (Leica, Wetzlar, Germany). The migration index was calculated in terms of wound coverage.

The invasion assay was performed as described previously (Cai et al., 2020) using 24-well BD Matrigel Invasion Chambers (BD Biosciences). Briefly, 5 × 10⁴ suitably treated glioma cells were seeded in the upper wells of the chamber in serum-free media. The lower chamber wells contained DMEM supplemented with 20% FBS. After incubation for 24 h, non-invading cells were scraped off and invading cells on the bottom cells were fixed and stained. The number of invading cells was counted in three representative fields per sample.

An orthotopic glioma model was established by intracranially injecting 5 × 10⁶ U251 cells into 4−6-week-old immunodeficient mice. Tumor growth was monitored by in vivo bioluminescence imaging. The tumor length (L) and width (W) were measured with calipers, and the volume (V) was calculated as (L × W²)/2. Mice were sacrificed when they showed signs of ill health such as weight loss, rough coat, etc. Brains were harvested and further fixed with 4% paraformaldehyde at 4°C overnight. All animal studies were approved by the Animal Welfare Ethical Review Committee of Nanjing Medical University and Anhui Medical University and conducted in accordance with the National Guidelines for the Care and Use of Laboratory Animals.

SPSS 24.0 (IBM, Chicago, USA), R3.3.1 (https://www.r-project.org/), and GraphPad Prism 7.0 programs were used for statistical analyses. The association between clinicopathologic features and CFI was evaluated with the Wilcoxon signed rank test and logistic regression. Pearson's correlation analysis was used to identify CFI-related genes in the CGGA database. Univariate and multivariate Cox regression analyses were used to evaluate the relationship between the clinical variables and OS. The nomogram model was created using the rms package of R software. ROC and AUC were calculated using the package of “survivalROC” in R. All experimental data was expressed in the format of mean ± standard deviation. Student's t test was used for pairwise comparison, and one-way analysis of variance and Cox regression analysis were used for multiple group comparison. Survival was analyzed using the Kaplan–Meier method with GraphPad Prism 7. P < 0.05 was considered statistically significant.

We identified CFI as a potential biomarker of glioma on the basis of bioinformatics analysis and the differential expression between gliomas and corresponding normal tissue samples (data not shown). We analyzed the CFI expression profile across diverse tumors in TCGA database and found that CFI was significantly overexpressed in COAD (colon adenocarcinoma), GBM, LGG, READ (rectum adenocarcinoma), STAD (stomach adenocarcinoma), and THCA (thyroid carcinoma) relative to the corresponding normal samples (Figure 1A). Analysis of both TCGA and GTEx datasets indicated significantly higher levels of CFI in 10 tumor types compared to the normal specimens (Figure 1B). Furthermore, higher CFI expression was associated with significantly shorter OS and disease-free survival (DFS) in both GBM (Figure 1C) and LGG (Figure 1D) patients of TCGA cohort. We then stratified all glioma patients in TCGA database into the CFI^high and CFI^low groups and found that CFI overexpression correlated to shorter OS in all gliomas (Figure 1E). Consistent with this, high CFI expression level also predicted worse prognosis for all glioma patients in the CGGA database (Figure 1F). Taken together, CFI was overexpressed in gliomas and associated with poor prognosis.

Furthermore, we tried to investigate the expression and prognostic value of CFI in glioma subgroups in the CGGA dataset according to WHO 2016 classification (Louis et al., 2016; Jiang et al., 2020). As shown in Supplementary Figures 1A–C, the CFI expression was significantly elevated in IDH wildtype grade II glioma, IDH mutant and 1p/19q non-codeleted grade III glioma, and IDH wildtype GBM compared to corresponding control groups. Then, we evaluated the prognosis stratification ability of the CFI in all six glioma subgroups. We divided patients into high-risk and low-risk groups in various groups of gliomas by their respective CFI expression. We found that patients in high-risk groups had shorter OS than low-risk groups in IDH mutant grade II glioma and IDH mutant GBM subgroups (Supplementary Figures 1D,H). But, we observed no statistically significant difference in patients' OS between high-risk and low-risk groups in IDH wildtype grade II glioma, IDH mutant and 1p/19q codeleted or non-codeleted grade III glioma, and IDH wildtype GBM (Supplementary Figures 1E–G,I). The limited number of cases may lead to insignificant p-value.

The correlation between CFI expression in gliomas and the clinicopathological features were next analyzed in the CGGA dataset. As shown in Figure 2A, patients with recurrent or secondary gliomas expressed higher levels of CFI compared to those with primary gliomas. In addition, CFI expression was positively correlated with the WHO tumor grade (Figure 2B), age (≥41 years), and chemotherapy status (Figures 2C,D). Consistent with this, CFI expression was significantly higher in the GBM patients compared to LGG patients (Figure 2G). Furthermore, glioma samples with IDH mutation or 1p/19q co-deletion expressed lower levels of CFI compared to the corresponding non-mutated samples (Figures 2E,F). Collectively, CFI expression is strongly associated with the clinicopathological features of glioma, including PRS type (primary/recurrent/secondary), WHO grade, age, chemotherapy status, IDH mutation status, 1p/19q co-deletion status, and histology grade, and is therefore a promising biomarker for prognosis as well as a therapeutic target.

In a previous study, we found that glioma patients expressing high levels of CFI had significantly worse OS and DFS compared to the CFI^low patients in TCGA and CGGA databases. In this study, we performed univariate and multivariate Cox regression analyses on the CGGA dataset to determine the prognostic value of CFI expression and other clinical variables in gliomas (Table 1). Univariate Cox regression analysis indicated that CFI expression, PRS type, histology, WHO grade, age, chemotherapy status, IDH mutation, and 1p19q co-deletion were significantly correlated with the OS of glioma patients, whereas gender and radiotherapy status did not show any correlation (Figure 3A). Seven parameters were further identified as independent prognostic factors of OS by multivariate analysis, including CFI expression [P < 0.001, hazard ratio (HR) = 1.149, 95% confidence interval (CI) = 1.079–1.222], PRS type (P < 0.001, HR = 1.930, 95% CI = 1.643–2.268), WHO grade (P < 0.001, HR = 2.676, 95% CI = 1.959–3.655), age (P = 0.013, HR = 1.290, 95% CI = 1.056–1.576), chemotherapy status (P = 0.001, HR = 0.672, 95% CI = 0.529–0.855), IDH mutation (P = 0.002, HR = 1.467, 95% CI = 1.156–1.861), and 1p19q co-deletion (P < 0.001, HR = 2.299, 95% CI = 1.640–3.223) (Figure 3B).

To further elucidate the predictive accuracy of the above parameters for OS in glioma patients, we next conducted an ROC analysis. The AUC of CFI was 0.734 compared to 0.681, 0.733, 0.499, 0.593, 0.457, 0.514, 0.69, and 0.615 calculated for PRS type, histology, gender, age, radiotherapy status, chemotherapy status, IDH mutation status, and 1p19q co-deletion status, respectively. Thus, CFI can predict the OS of glioma patients with moderately high sensitivity and specificity (Figure 3C). However, since CFI expression alone is insufficient to evaluate survival, we constructed a nomogram incorporating the seven independent prognostic factors to quantitatively estimate the 1-, 3-, and 5-year OS possibility of patients with glioma. The survival probability of each patient can be calculated by adding up the scores of each parameter. As shown in Figure 3D, WHO grade and CFI expression contributed the most to prognosis, followed by PRS type, 1p19q co-deletion status, chemotherapy status, IDH status, and age. The total score can be easily calculated by adding up the scores for every parameter to evaluate the survival probability of each patient. The calibration curve for the 3-year OS showed an optimal concordance between the prediction by nomogram and actual observation in CGGA dataset (Figure 3E). However, the calibration curves for the 1- and 5-year OS were not satisfactory due to some limitations of nomogram itself (data not shown). The AUC of the 3-year OS prediction for the nomogram was 0.875 (Figure 3F). The C-index of the risk score was 0.769. Thus, the nomogram showed moderate sensitivity and specificity for predicting 3-year OS of glioma patients.

To further investigate the potential function of CFI in glioma, we conducted Gene Set Enrichment Analysis (GSEA) using the Kyoto Encyclopedia of Genes and Genomes (KEGG) sets (c2.cp.kegg.v6.2.symbols). The gene sets with normalized P-value <0.05 and FDR q-value <0.05 were considered significant. As shown in Supplementary Table 1, JAK/STAT, NOD-like receptor, pathways in cancer, T cell receptor, and vascular endothelial growth factor (VEGF) were significantly enriched in the CFI^high gliomas (Supplementary Figure 2) and therefore may be involved in mechanisms underlying CFI upregulation in glioma. We next analyzed the correlation between the expression levels of CFI and other genes using the CGGA dataset. Ten genes showed the maximum positive correlation with the expression level of CFI, including ANXA1 (Pearson's r = 0.845, P < 0.001), ANXA2 (Pearson's r = 0.822, P < 0.001), C1R (Pearson's r = 0.849, P < 0.001), C1S (Pearson's r = 0.838, P < 0.001), CASP4 (Pearson's r = 0.805, P < 0.001), CP (Pearson's r = 0.801, P < 0.001), PLAU (Pearson's r = 0.805, P < 0.001), PROS1 (Pearson's r = 0.795, P < 0.001), SERPINA3 (Pearson's r = 0.802, P < 0.001), and TNFRSF12A (Pearson's r = 0.795, P < 0.001) (Supplementary Figures 3A–J). Furthermore, these 11 genes (CFI, ANXA1, ANXA2, C1R, C1S, CASP4, CP, PLAU, PROS1, SERPINA3, and TNFRSF12A) were also positively correlated with each other (Supplementary Figure 3K). These findings suggest that the co-expressing genes may be involved in CFI-induced tumorigenesis.

Recent studies suggested that VEGF signaling is a complex process involving GBM growth stimulation or inhibition (Kil et al., 2012; Lu et al., 2012; Szabo et al., 2016). Additionally, GBM inevitably progressed during anti-VEGF therapy, such as bevacizumab, through adapting and using alternative signaling pathways to sustain tumor growth (Lu et al., 2012). In this study, we sought to gain insight into the role of CFI in VEGF pathway by immunoblot and the results showed that phosphorylation level of focal adhesion kinase (FAK) at tyrosine sites (Y576 and Y577) was markedly decreased after CFI depletion among various receptor molecules including VEGF receptor 2 (VEGFR2), FAK, protein kinase B (AKT), and p38 mitogen-activated protein kinase (p38 MAPK) (Supplementary Figure 4). Whereas, the total protein level of FAK was not influenced, suggesting that CFI may contribute, at least in part, to FAK activation in whole VEGF signaling pathway.

We validated the bioinformatics data on CFI expression level with six paired GBM and NBT specimens, as well as in glioma specimens of different grades. As shown in Figure 4A, CFI protein level increased through glioma grades I–II, III, and IV compared to the NBTs. Furthermore, the expression level of CFI was markedly elevated in the GBM specimens relative to the NBTs (Figure 4B). The expression level of CFI mRNA was also detected in normal human astrocytes (NHAs) and six GBM cell lines (A172, U118, LN229, U251, T98G, and U87), and all GBM cells except A172 expressed markedly higher levels of CFI mRNA compared to the NHAs (Figure 4C). CFI protein expression levels also showed similar trends (Figure 4D). The western blotting results were confirmed by immunohistochemistry, and the glioma tissues showed significantly higher in situ expression of CFI compared to the NBTs. Notably, grade IV gliomas had the strongest CFI expression (Figures 4E,F). Consistent with the bioinformatics results, Kaplan–Meier survival analysis suggested that patients with high CFI expression level had shorter overall and progression-free survival compared to those with low CFI expression (Figures 4G,H). Taken together, CFI was significantly upregulated in glioma tissues and cell lines and correlated with worse prognosis in GBM patients, indicating its potential as a prognostic biomarker.

To gain further insights into the role of CFI in gliomas, we knocked down its expression in U251 and LN229 glioma cell lines using three independent lentiviral shRNAs. As shown in Figure 5A, the cells transfected with shCFI constructs showed a significant reduction in CFI protein level compared to the control shRNA group. Furthermore, shCFI-1 showed the highest silencing efficiency (Figure 5A) and was used for the subsequent experiments. To determine whether inhibition of CFI can affect glioma development, we analyzed the proliferation, migration, and invasion capacities of the CFI-knockdown cells in vitro. The growth rates of CFI-depleted U251 and LN229 cells were significantly lower compared to the control cells (Figure 5B). Consistent with this, knocking down CFI also markedly reduced the number of colonies formed by the glioma cells (Figures 5C,D). Accumulating evidences showed that the invasion and migration of tumor cells to adjacent and distant tissues are the major hindrances in the treatment of glioma patients (Sanai and Berger, 2018). Therefore, we next evaluated the effect of CFI knockdown on the invasion and migration of glioma cells through transwell and wound healing assays, respectively. As shown in Figures 5E,F, CFI depletion significantly decreased the number of invading cells compared to the controls. Likewise, the wound coverage of the shCFI-transfected glioma cells was also significantly lower 24 h after plating, which was indicative of decreased migration potential compared to the control cells (Figures 5G,H). Collectively, these results suggest that downregulation of CFI effectively inhibits cell proliferation, invasion, and migration of glioma cells.

To further establish an oncogenic role of CFI in glioma, we overexpressed the CFI gene in A172 cells (Figure 6A) and performed the same functional assays. As shown in Figure 6B, the A172 cells overexpressing CFI displayed a marked increase in growth rate compared to the control cells, as well as a significant increase in the number of colonies (Figure 6C). Consistent with these properties, ectopic expression of CFI in A172 cells also enhanced their invasiveness (Figure 6D) and the migration index in the wound healing assay (Figure 6E) compared to the control cells. Together, CFI overexpression induced proliferation, invasion, and migration in glioma cells, indicating that the gain of function of CFI likely enhances their malignant potential.

To further explore the role of CFI in glioma progression in vivo, we injected luciferase-labeled control or shCFI U251 cells intracranially into 4-−6-week-old nude mice. Tumor growth was monitored using in vivo bioluminescence imaging system. The control mice exhibited tumor growth over 21 days after inoculation, whereas the CFI-depleted glioma cells formed significantly smaller tumors (Figure 7A). Representative H&E stainings for intracranial xenografts are shown in Figure 7B. Immunohistochemical staining showed decreased expression of CFI and Ki67 in CFI-knockdown tumors compared to control tumors in mice (Figure 7B). Quantification of tumor volume also indicated a marked deceleration in xenograft growth in the CFI-knockdown vs. control groups (Figure 7C). Consistent with these findings, Kaplan–Meier analysis demonstrated that the mice injected with shCFI-U251 cells survived considerably longer than those injected with the control U251 cells (Figure 7D). Finally, CFI protein levels were significantly lower in the xenografts derived from shCFI-U251 cells compared to control cells (Figure 7E). Together, inhibition of CFI impedes glioma growth in vivo, indicating its therapeutic potential.