The RNA-sequencing data of 504 patients with primary LGG from the TCGA as the training cohort and 725 normal brain samples from GTEx were obtained from the University of California Santa Cruz (UCSC) Xena website (https://xena.ucsc.edu). The RNA-sequencing data of 407 patients (including 270 patients in CGGA cohort 1 and 137 patients in CGGA cohort 2) were downloaded from the CGGA (http://www.cgga.org.cn) as the validation cohort. Corresponding clinical information of patients in the training cohort and validation cohort was acquired from the TCGA and CGGA, respectively.

Gene expression was quantified by normalized estimation of fragments per thousand base transcripts per million mapped reads (FPKM) and log2-based transformation. The ComBat method was performed to remove the batch effects using the R package “sva”. Next, DEGs were identified by the “limma” package in R software using the absolute value of the log2-transformed fold change (FC) > 2 and the adjusted P value (adj. P) < 0.05 as the threshold.

Univariate Cox regression analysis was performed to analyze the prognostic significance of overlapping genes between the DEGs and IRGs. Twenty-seven genes correlated with overall survival (OS) (P < 0.05) were screened out. Furthermore, nine genes and their regression coefficients obtained by least absolute shrinkage and selection operator (LASSO, R package: glmnet) regression analysis were applied to construct the inflammatory signature. A 10-fold cross-validation was performed to select the optimal lambda (penalty for the number of characteristics), which determined the performance of the lasso-cox model (number of features included in the model and predictive deviations). The inflammatory signature score of each patient in the training cohort and validation cohort was calculated by the following formula:

β_n is the coefficient of each gene derived from the LASSO regression, and x _n is the expression level of each gene. Primary LGG patients were divided into high- and low-inflammatory signature groups in the training cohort (TCGA) and validation cohort (CGGA) according to the median inflammatory signature score.

Gene set enrichment analysis (GSEA), gene set variation analysis (GSVA), gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses were applied to explore the biological functions and signaling pathways related to the high- and low-inflammatory signature groups using the MSigDB database. Gene sets with adjusted P values <0.05 were included in the analysis. The STRING database (https://string-db.org/) was used to construct a PPI network of overlapping genes between the DEGs and IRGs. The PPI network was visualized using the Cytoscape software.

We integrated the TCGA-LGG gene expression profile and computationally reconstructed a brain-specific transcriptional network using the SJARACNe algorithm (25). The network included potential master regulators (1899 transcription factors and 8403 signaling proteins) with their transcriptionally predicted target genes. Then, we used the network-based Bayesian inference of drivers (NetBID) (26) (https://github.com/jyyulab/NetBID) algorithm to infer the regulatory activity of the master regulators in each sample based on their target gene expression value and the regulatory relationship. We hypothesized that if a transcription factor/signaling protein is a “hidden” driver between LGG subsets, its regulons in the network should be enriched in the differentially expressed genes, although the driver itself is not necessarily differentially expressed.

Human glioma cell lines (U87-MG and U251) were acquired from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China), cultured in DMEM (Gibco, USA) supplemented with 10% fetal bovine serum (Gibco, USA) and penicillin-streptomycin-glutamine (Gibco, USA) and incubated at 37°C in a 5% CO₂ humidified atmosphere.

The viability of human glioma cell lines (U87-MG and U251) treated with BRPF1-specific inhibitor (GSK6853, Selleck) or vehicle (DMSO, Sigma) was measured by Cell Counting Kit-8 (CCK-8) according to the manufacturer’s protocol. Briefly, U87-MG and U251 cells (2 × 10³/well) were seeded into 96-well cell culture plates and incubated with a BRPF1-specific inhibitor (GSK6853, Selleck) or vehicle (DMSO, Sigma) for 24-96 h at 37°C in a 5% CO₂ humidified atmosphere. CCK-8 reagent was added to the medium and incubated for 1-2 h at 37°C in a 5% CO₂ humidified atmosphere. Cell culture plates were read at 450 nm, and the OD value was obtained. Cell proliferation curves were produced using GraphPad Prism 8.

Total protein of human glioma cell lines (U87-MG and U251) was extracted with RIPA lysis buffer (Millipore, Cat. No: 20-188) containing protease inhibitor (Roche, Cat. No: 11873580001) and quantified via BCA assay (Beyotime, Cat. No: P0012S). 10% SDS-gel were applied to separate proteins. Primary antibodies against BRPF1 (Abcam, Cat. No: ab251669,1:1000 dilution) and β-actin (ORIGENE, Cat. No: TA811000, 1:2000 dilution) were used for immunoblotting. Enhanced chemiluminescence was used to detect protein bands, and the intensity of the protein bands was determined with an Image software (Bio-Rad).

The pGreenPuro (CMV) vector was used to construct a BRPF1 knockdown plasmid. The shRNA sequences were as follows: shBRPF1-1#, AGGACTACATCTGGCTGGATATCAT, and shBRPF1-2#, CCGCATCAGCATCTTTGACAA.

One milliliter of DMEM containing 10% FBS with 0.6% agarose (Sangon Biotech) was added to a 12-well cell culture plate as a base support. Human glioma cell lines (U87-MG and U251) were seeded in 1 ml of DMEM containing 10% FBS with 0.35% agar at 1 × 10⁴ cells/well and layered onto the base support. Then, 0.5 ml of DMEM containing 10% FBS with BRPF1-specific inhibitor (GSK6853, Selleck) or vehicle (DMSO, Sigma) was layered on top of the agar gel. Three weeks later, the number of colonies in each well was counted under a microscope (Leica DMIL LED).

All statistical data are shown as the mean ± standard deviation (SD). The Student’s t-test or analysis of variance (ANOVA) was used for statistical analyses. Survival analyses were compared between the high- and low-inflammatory signature groups via Kaplan-Meier analysis methods using the ‘survival’ and “survminer” packages in R. Univariate Cox regression analysis was applied to identify potential prognostic genes, and Lasso regression was performed to screen out gene sets to construct an inflammatory signature. Multivariate Cox regression analysis was used to determine clinical factors (including the inflammatory signature) as independent risk factors for OS in primary LGG. The methodology of Grambsch and Therneau was used to verify the proportional hazards assumption in the Cox proportional hazards model. A P value < 0.05 was defined as statistically significant.

The workflow of our study is presented in Figure 1 . A total of 911 patients with primary LGG met the inclusion criteria, including 504 patients from the TCGA as the training cohort and 407 patients from the CGGA as the validation cohort (270 patients in CGGA cohort 1 and 137 patients in CGGA cohort 2). The clinical characteristics of the primary LGG patients from the TCGA and CGGA are listed in Table 1 .

To identify inflammation-related genes with prognostic significance, we first downloaded RNA-seq data of normal brain tissue from GTEx and primary LGG from the TCGA. After normalization and batch effect removal, a total of 6,089 DEGs were selected using the absolute value of the log2-transformed fold change (FC) > 2 and the adjusted P value (adj. P) < 0.05 as the threshold ( Supplementary Table S1 ). Among them, 3734 genes were upregulated and 2355 genes were downregulated, as shown in Figure 2A . Two hundred IRGs were downloaded from MSigDB. After overlapping the DEGs and IRGs, thirty-five genes were obtained ( Figure 2B ), and 71.34% of these genes (25/35) were upregulated in the LGG group ( Figure 2C ). STRING was used to construct the PPI network of overlapping genes. After visualization using the Cytoscape software, we discovered that some proteins were closely related to other proteins, such as LPAR1, IL1β, CCL2, MYC, and IL1α ( Figure 2D ).

Next, thirty-five overlapping genes were further analyzed by univariate Cox regression analysis, and twenty-seven genes significantly associated with OS in primary LGG (TCGA) were obtained ( Figure 2E ). Considering collinearity, twenty-seven inflammatory prognostic genes were subjected to Lasso Cox regression, and nine genes, including CSF, SELL, TACR1, ICAM4, ITGB8, LPAR1, MSR1, TLR3 and TIMP1, were obtained to construct an inflammatory signature ( Figure 2F ). Nine pivotal genes illustrated the differences in the survival time of patients with primary LGG (TCGA). CSF, SELL and TACR1 were correlated with an adverse prognosis, while the other six genes were correlated with a good prognosis ( Figure S1 ).

An inflammatory signature (IFS) was constructed with nine genes,and their coefficients were previously identified. The gene expression profile of the nine genes in the training cohort (TCGA) is shown in Figure 3A . The primary LGG patients in the training cohort (TCGA) were divided into high- and low-inflammatory signature groups using the median inflammatory signature score as the cutoff value. The inflammatory signature score and survival status distribution of the primary LGG patients in the training cohort (TCGA) are shown in Figures 3C, E . The Kaplan-Meier survival curve showed that patients in the high-inflammatory signature group (high IFS group) had shorter survival time (P< 0.0001; Figure 3G ).

The 270 primary LGG patients from the CGGA were used as a validation cohort 1 to verify the performance of the inflammatory signature. The inflammatory signature score for each patient in the validation cohort 1 (CGGA cohort 1) was calculated using the same method. The primary LGG patients in the CGGA cohort 1 were also divided into high- and low-inflammatory signature groups using the median inflammatory signature score. The gene expression profile of the nine genes, inflammatory signature score and survival status distribution of the primary LGG patients in the CGGA cohort 1 are shown ( Figures 3B, D , F ). We found that primary LGG patients in the high-inflammatory signature group had a worse prognosis (P=0.0015; Figure 3H ). Consistently, the similar results were discovered in validation cohort 2 (CGGA cohort 2) ( Figure S2 ). These results suggested that the inflammatory signature was a good predictor of the OS of patients with primary LGG.

To assess the independent prognostic role of the inflammatory signature, univariate and multivariate Cox regression analyses were applied to confirm its performance in the training and validation cohorts. The results from univariate Cox regression analysis showed that IFS_group, grade, histology subtype, age in the training cohort and grade and histology subtype in the validation cohort 1 and IFS_group, grade, histology subtype, age in the validation cohort 2 were significantly associated with patient survival ( Figures S3 and S4A ). Multivariate Cox regression analysis showed that the high-inflammatory signature group was independently associated with a worse OS of primary LGG patients in both the training and validation cohort 1 (P<0.001; Figures 4A, B ). These results indicated that the inflammatory signature was a strong independent prognostic factor for patients with primary lower-grade gliomas.

Based on the multivariate Cox regression analysis, a nomogram was constructed for predicting primary LGG 1-, 3- and 5-year OS time, which integrated both the inflammatory signature group and clinicopathologic variables, including age, sex, histological subtype, and grade ( Figure 4C ). The data analysis conformed to the proportional hazards assumption (TCGA: P = 0.062, CGGA cohort 1: P = 0.099, CGGA cohort 2: P =0.621). The C-index of the nomogram in the training cohort (TCGA) was 0.826 (95% CI; 0.787–0.865). The areas under the curve (AUC) of the 1-, 3- and 5-year OS predictions for the constructed nomogram were 0.901, 0.887 and 0.801 in the training cohort, respectively ( Figure 4D ). Meanwhile, calibration curve for this nomogram were developed and plotted, which showed that this nomogram model had good accuracy ( Figures 4F–H ). In the validation cohort (CGGA cohort 1 and CGGA cohort 2), we found consistent results. The areas under the curve (AUC) and calibration curve were also plotted ( Figures 4E, I – K and S4C – F ). These results demonstrated that the nomogram had good accuracy in predicting the 1-, 3- and 5-year survival of patients with primary LGG in both the training cohort (TCGA) and validation cohort (CGGA cohort 1 and CGGA cohort 2).

In the disease development process, changes in inflammation levels in patients are often accompanied by an immune response (18). Previous studies have reported that alterations in immune cells in the tumor microenvironment are related to tumorigenesis and progression (27–29). To explore the differences in immune cells in LGG patients with high- and low-inflammatory signature, we adopted the CIBERSORT method to analyze the distribution of immune cells in LGG tissues in the training cohort (TCGA). After deconvolution, M2 macrophages were the most abundant immune cells, followed by monocytes and activated mast cells ( Figure 5A ). Then, we confirmed similar results in the CGGA cohort 1. M2 macrophages were also the most abundant immune cells ( Figure S3A ). The proportions of M2 macrophages and resting CD4 memory T cells in the high-inflammatory signature group were significantly higher than those in the low-inflammatory signature group in the TCGA and CGGA cohort 1 ( Figures 5A , S5A ). These results showed the heterogeneity of the immune cells of the TME in primary LGG and M2 macrophages and resting CD4 memory T cells that demonstrated high activity in the TME during the primary LGG development process.

Correlation analysis based on the training cohort (TCGA) suggested that the number of M2 macrophage cells was inversely related to the number of activated mast cells (r²= -0.70). Activated NK cells were negatively correlated with resting NK cells (r²= -0.45), and naïve CD4 T cells were negatively correlated with resting memory CD4 T cells (r²= -0.41) ( Figure 5B ). These results indicated that antagonistic functions might exist between these cells in the LGG development process. In contrast, we found a positive correlation between plasma cells and naïve B cells or activated memory CD4 T cells (r^{2 =} 0.52 or 0.52) in the TCGA cohort and a highly positive correlation between regulatory T cells (Tregs) and resting NK cells or resting mast cells (r^{2 =} 0.54 or 0.42) in the CGGA cohort 1( Figures 5B , S5B ). These results suggest that these cells might have synergistic functions in the tumor microenvironment.

To depict transcriptional characteristics between the high- and low-inflammatory signature groups, the DEGs were further screened by comparing the gene expression profiles. A total of 2123 upregulated genes and 1690 downregulated genes were selected in the high-inflammatory signature group compared with the low-inflammatory signature group in the TCGA cohort with the absolute value of fold change >1 and adjusted P value (adj. P) <0.05 as the threshold ( Supplementary Table S2 ). The GO analysis results showed that the differentially expressed genes were mainly enriched in biological processes linked to inflammatory response and immunity, such as adaptive immune response, leukocyte-mediated immunity and immune effector process ( Figure 5C ). The bar plot of the KEGG analysis revealed that immune-related signaling pathways were enriched, such as neuroactive ligand receptor interaction, systemic lupus erythematosus and primary immunodeficiency pathways ( Figure 5D ). GSEA and GSVA were also performed to decipher the difference between the high- and low- inflammatory signature groups. The GSEA and GSVA results were also involved in the immune process ( Figures 5E – I ). Interestingly, similar results were found in the CGGA cohort 1. The results of the GO, KEGG, GSEA and GSVA analyses were also enriched for immunity ( Figures S5C – I ).

Maximum surgical resection combined with radiotherapy and chemotherapy is the main treatment protocol for gliomas (7). However, some patients still suffer from surgical sequelae and tumor recurrence. In addition, targeted therapy is relatively rare due to the lack of effective drug targets (30). In our study, a three-step approach was developed to screen drug targets for primary lower-grade gliomas based on the inflammatory signature. First, combined with transcriptomic data in the TCGA cohort, the NetBID algorithm was applied to screen out “hidden” driver genes between the normal brain and high- or low-inflammatory signature groups, and 1210 overlapping genes were obtained ( Supplementary Table S3 ). Second, the DEGs between the normal brain and the high- or low-inflammatory signature group were identified by bioinformatics methods, and 7102 overlapping upregulated genes were obtained ( Supplementary Table S4 ). Finally, the information of 3522 drug targets in the therapeutic target database (TTD) was integrated, and 100 overlapping genes were identified that could be potential drug targets for lower grade gliomas with high- and low-inflammatory signatures ( Figure 6A and Supplementary Table S5 ). After gene annotation, the 100 overlapping genes could be mainly classified into the following four categories: transcription factors, epigenetic molecules, protein kinases, and cell surface proteins ( Figure 6B ). In the validation cohort (CGGA cohort 1 and CGGA cohort 2), we obtained 218 and 475 overlapping genes using same screening method, respectively ( Supplementary Tables S6 , S7 ). Finally, we identified 22 common genes among TCGA cohort, CGGA cohort 1 and CGGA cohort 2 ( Supplementary Table S8 ).

Epigenetics plays an important role in the pathogenesis of nervous system diseases (including tumors), and several epigenetic regulatory molecules have been considered potential drug targets (31–34). Some studies have reported that epigenetic molecules containing bromodomains can be used as drug targets, such as BRD4 (35), p300/CBP (36), and TRIM24 (37). Drugs or inhibitors that target these molecules, such as JQ1, have good antitumor effects (38, 39). Among the 22 common genes, epigenetic molecule BRPF1 contains bromodomain and has been identified as a therapeutic target for liver cancer (40). Therefore, we selected BRPF1 for further research in this study. According to the Human Protein Atlas database, the protein level of BRPF1 was higher in low- and high-grade gliomas than in normal brain tissue, and the highest expression level of BRPF1 was found in high-grade gliomas ( Figure 6C ).

To investigate the drug target potential of BRPF1 in glioma, we selected the BRPF1-specific inhibitor GSK6853 to treat U87-MG and U251 glioma cell lines and determined the IC50 value of GSK6853 by CCK-8 assay. The results showed that GSK6853 exhibited excellent inhibitory activity against U87-MG and U251 cell lines with IC50 values of 26.47 μM and 35.55 μM, respectively ( Figure 7A ). Next, we treated U87-MG and U251 cell lines with three concentrations of GSK6853 (20 μM, 40 μM, and 80 μM) to inhibit BRPF1 function. The CCK-8 assay showed that inhibition of BRPF1 function suppressed glioma cell proliferation ( Figures 7B, C ). The number of U87-MG and U251 cell clones decreased after inhibiting BRPF1 function ( Figures 7D – F ). To further clarify the effect of BRPF1 expression on glioma cell proliferation, we first interfered with BRPF1 expression using shRNA in the U87-MG and U251 cell lines ( Figure 8A ). Knockdown of BRPF1 expression also attenuated the growth of U87-MG and U251 cell lines, as determined by CCK-8 assay ( Figures 8B, C ). Moreover, knockdown of BRPF1 expression reduced the clone number of U87-MG and U251 cells in the plate clone formation assay and soft agar colony formation assay ( Figures 8D – F ). These experiments indicated that BRPF1 is involved in glioma cell proliferation and is a potential drug target for the treatment of gliomas.