Clinical samples and ethics of study<\/p>\n
Sample acquisition: (1) Glioma tissues: 30 cases (25 grade II\/III, 5 GBM; WHO 2021 criteria); (2) Normal brain controls: 5 cases obtained during neurosurgical resections for traumatic brain injury or intracerebral hemorrhage. Control tissues were sampled from morphologically normal cortex at least 2\u00a0cm away from the lesion margin, confirmed intraoperatively as non-pathological and verified by histopathological examination; (3) Peripheral blood: 25 glioma patients and 20 healthy donors. All glioma blood samples were collected preoperatively in the morning prior to anesthesia induction and before any adjuvant therapy, while control blood samples were obtained from age- and sex-matched healthy donors during routine health check-ups. Serum AFU levels were measured in all blood specimens.<\/p>\n
Ethical oversight: Approved by Guilin Medical University, The First Affiliated Hospital IRB (2023YJSLL-78) following Declaration of Helsinki guidelines. Written informed consent preceded sample collection.<\/p>\n
Data obtaining and processing<\/p>\n
Data Acquisition: (1) GWAS: Brain cancer datasets (ebi-a-GCST90018800, ieu-b-4875, ebi-a-GCST90018580) from OpenGWAS (https:\/\/gwas.mrcieu.ac.uk\/<\/a>); (2) TCGA: RNA-seq (TPM), methylation, and clinical data for GBM (n\u2009=\u2009168) and LGG (n\u2009=\u2009516) from GDC Portal (https:\/\/portal.gdc.cancer.gov\/<\/a>); (3) GTEx-TCGA Integration: Paired tumor\/normal TPM expression from UCSC Xena (tcga_RSEM_gene_tpm, gtex_RSEM_gene_tpm).<\/p>\n Validation Cohorts: (1) GEO: GSE16011 (n\u2009=\u2009284), GSE108474 (n\u2009=\u2009550), GSE15824 (n\u2009=\u200945); (2) CGGA: CGGA-301, CGGA-325, CGGA-693; (3) EMBL-EBI: E-MTAB-3892 (n\u2009=\u2009179); (4) Proteomics: CPTAC-GBM (PDC000204).<\/p>\n Data Processing: (1) Histological classification followed WHO guidelines; (2) RNA-seq data transformed as log\u2082(TPM\u2009+\u20091) or standardized per sample: Z\u2009=\u2009(x\u2014\u03bc)\/\u03c3; (3) Outlier removal: |Z|>\u20093.0.<\/p>\n Genetic co-localization and methylation-expression integration analysis<\/p>\n Bayesian colocalisation was performed using the R package coloc to assess whether glioma GWAS loci and brain eQTL signals share the same causal variants. We analyzed three independent glioma GWAS datasets obtained from OpenGWAS (ebi-a-GCST90018800, ieu-b-4875, ebi-a-GCST90018580; Table S1). eQTL summary statistics were derived from GTEx v8 brain tissues, including cortex, hippocampus, and cerebellum. For each locus, we applied a\u2009\u00b1\u2009500\u00a0kb window around the lead SNP and used default priors (p1\u2009=\u20091\u2009\u00d7\u200910\u207b4, p2\u2009=\u20091\u2009\u00d7\u200910\u207b4, p12\u2009=\u20091\u2009\u00d7\u200910\u207b5). Linkage disequilibrium (LD) was estimated using the 1000 Genomes Project Phase 3 European reference panel. Significant colocalisation was defined as posterior probability for hypothesis 4 (PP.H4.abf)\u2009>\u20090.8.<\/p>\n Analysis utilized preprocessed Level 3 methylation (450\u00a0K array) and RNA-seq data from TCGA. Additional probe filtering was applied: excluding probes with detection p-value\u2009>\u20090.01, on sex chromosomes, or containing SNPs\/cross-reactive sites. Batch effects from the plate variable were corrected using ComBat. Processed \u03b2-values were then median-aggregated across regulatory regions: promoters (TSS200: 0-200\u00a0bp; TSS1500: 200-1500\u00a0bp), 5’UTRs, first exons, and enhancers. Spearman correlations linked methylation to log\u2082(TPM\u2009+\u20091) expression. FUCA1 and FUCA2 expression profiles alongside promoter\/enhancer methylation values underwent cohort-wide Z-normalization.Samples were subsequently stratified into four molecularly defined subgroups based on directional deviation from population means: HyperMeth-HighExpr (methylation Z\u2009>\u20090 & expression Z\u2009>\u20090); HyperMeth-LowExpr (methylation Z\u2009>\u20090 & expression Z\u2009\u2264\u20090); HypoMeth-HighExpr (methylation Z\u2009\u2264\u20090 & expression Z\u2009>\u20090); and HypoMeth-LowExpr (methylation Z\u2009\u2264\u20090 & expression Z\u2009\u2264\u20090) subgroups for downstream analyses.<\/p>\n Spatial transcriptome gene localisation and expression<\/p>\n GBM WholeTranscriptomeAnalysis 10\u2009\u00d7\u2009and UKF241-C-ST datasets were analyzed using 10\u2009\u00d7\u2009Visium spatial transcriptomics. Cellular composition of each microregion was estimated by reference-based inverse convolution. A single-cell RNA-seq reference atlas was constructed from matched tumor samples after stringent quality control (genes per cell\u2009>\u2009500, unique molecular identifiers\u2009>\u20091,000, mitochondrial RNA fraction\u2009<\/p>\n Spatial deconvolution was performed using Cottrazm v1.2.0 (functions get_enrichment_matrix and enrichment_analysis) with the scRNA reference. To minimize batch effects across slides, data were log-normalized using Seurat v4.3.0 (NormalizeData function), and spatial alignment across tissue sections was harmonized according to the Sparkle database annotations. Cell-type distributions were visualized with Seurat\u2019s SpatialFeaturePlot, where gradient intensity reflects relative cellular abundance. Microregions were classified into: (1) Malignant (enrichment score\u2009=\u20091 for malignant cells), (2) Normal (enrichment score\u2009=\u20090 for malignant cells), and (3) Mixed (0\u2009\u20090.9 for enrichment scores), supporting the stability of our findings. Expression heatmaps were generated with pheatmap v1.0.12.<\/p>\n Single-cell transcriptomic profiling<\/p>\n Single-cell RNA-seq data from 28 glioblastoma samples [15<\/a>] (20 adult, 8 pediatric; GSE131928, Neftel et al., Cell 2019) were analyzed using TISCH2 [16<\/a>, 17<\/a>]. After quality control (7,930 cells retained; median 5,730 genes\/cell), FUCA1\/FUCA2 expression patterns were visualized via UMAP dimensionality reduction. Cells were defined as FUCA1\u207a or FUCA2\u207a if log-normalized expression values were greater than zero, and FUCA1\u207b or FUCA2\u207b otherwise. Cellular composition differences between subgroups were quantified across annotated cell types. Pathway activity was assessed using AUCell, scoring 6 functional categories: Immune response.<\/p>\n Metabolic processes; Signaling cascades; Proliferation pathways; Cell death mechanisms; Mitochondrial function. Differential pathway enrichment between positive or negative cells was evaluated via limma with Benjamini\u2013Hochberg correction (FDR\u200918].<\/p>\n Correlation analysis of clinical traits, prognosis and diagnosis<\/p>\n Clinical trait associations were interrogated using Biomarker Exploration for Solid Tumors (BEST) [19<\/a>]. We implemented a tiered survival analysis framework: first, univariate Cox regression identified candidate prognostic variables (retention threshold: P\u2009<\/p>\n After Z-normalization of FUCA1 and FUCA2 expression, samples were stratified into four combinatorial subgroups based on directional deviation from cohort means: FUCA1\u207a\/FUCA2\u207a (Z\u2009>\u20090 both), FUCA1\u207b\/FUCA2\u207a (FUCA1 Z\u2009\u2264\u20090 & FUCA2 Z\u2009>\u20090), FUCA1\u207b\/FUCA2\u207b (Z\u2009\u2264\u20090 both), and FUCA1\u207a\/FUCA2\u207b (FUCA1 Z\u2009>\u20090 & FUCA2 Z\u2009\u2264\u20090). Survival differences were evaluated through pairwise subgroup comparisons and omnibus testing.<\/p>\n Prognostic hazard ratios underwent inverse-variance meta-analysis (Meta package) using log(HR) as the effect measure. Diagnostic performance was quantified through receiver operating characteristic (ROC) analysis (pROC package), reporting area under the curve (AUC) with 95% confidence intervals. Smoothed ROC curves were generated to evaluate tumor versus normal classification accuracy.<\/p>\n Tumor immune microenvironment characterization<\/p>\n To delineate relationships between cellular composition and gene expression in LGG, we applied consensus deconvolution algorithms (xCell, TIMER, EPIC) to estimate immune cell abundances. Scatter plots visualized Spearman correlations between FUCA1\/FUCA2 expression and immune cell infiltration levels. Samples were dichotomized into high\/low expression groups using median FUCA1\/FUCA2 thresholds. Immune cell abundance differences between groups were assessed via Wilcoxon rank-sum tests across all algorithms, with results visualized in gradient heatmaps (red\u2009=\u2009high abundance; samples ordered by ascending gene expression).<\/p>\n Further stratification divided patients into expression quartiles (Q1: highest 25%; Q4: lowest 25%). Pathway activity scores were calculated using: Thorsson’s method (mean pathway scores per quartile): Heatmaps (pheatmap) compared pathway activities across quartiles. Wilcoxon tests identified differential pathway enrichment between median-based expression groups [20<\/a>, 21<\/a>]; PROGENy (easier package; 14 cancer pathways including MAPK\/NF-\u03baB) [22<\/a>].<\/p>\n To further evaluate tumor immune-related epigenetic features, methylation-derived tumor-infiltrating lymphocyte (MeTIL) scores were calculated [23<\/a>]. Briefly, principal component analysis (PCA) was applied to the methylation values of MeTIL markers, and the first principal component was extracted to generate MeTIL scores. Data were standardized into unit-free Z-scores using (x\u2009\u2212\u2009\u03bc)\/\u03c3 transformation. Samples were then divided into high and low FUCA1\/FUCA2 expression groups according to median values, and statistical differences in MeTIL scores between groups were assessed using Wilcoxon rank-sum tests.<\/p>\n \u03b1-L-fucosidase-related gene signature (AFURGS) enrichment analysis<\/p>\n Using TCGA-LGG data, we identified \u03b1-L-fucosidase-related genes (AFURGS) through integrated differential expression and co-expression profiling. Differentially expressed genes (DEGs) were defined by comparing FUCA1\/FUCA2 high- versus low-expression groups (thresholds: |log\u2082FC|>\u20091.5 and Benjamini-Hochberg (BH) adjusted P\u2009\u20090.6 and BH adjusted P\u2009P\u2009<\/p>\n \u03b1-L-fucosidase risk signature (AFURS) construction and validation<\/p>\n To establish a robust prognostic model for glioma, we developed a risk signature termed AFURS (\u03b1-L-Fucosidase Risk Score) through a structured feature selection pipeline [24<\/a>]. Candidate variables were first filtered by univariate Cox regression (p\u2009n\u2009=\u2009658;Exclude patients with an excessively short or absent survival time). The model’s performance was rigorously evaluated. We performed sufficient external validation in independent CGGA datasets (CGGA-325, CGGA-693) using Kaplan\u2013Meier analysis (log-rank test) and time-dependent ROC curves.<\/p>\n Immunotherapeutic relevance was evaluated using tumor immunophenotype scores from The Cancer Immunome Atlas (TCIA). Predictive efficacy was validated in two immunotherapy cohorts: IMvigor210 [25<\/a>]: advanced urothelial carcinoma treated with anti-PD-L1 (atezolizumab); GSE100797 [26<\/a>]: melanoma receiving adoptive T-cell therapy (ACT). Spatial transcriptomic validation was performed using 10\u2009\u00d7\u2009Visium data (GBM WholeTranscriptomeAnalysis). Area under the curve (AUC) scores for AFURS were calculated across microregions, with Spearman correlations assessing relationships between AUC and tumor microenvironment components.<\/p>\n Screening of potential targeted drugs<\/p>\n