The clinical, RNA sequencing data and single cell sequencing data of all GBMLGG patients in this study were obtained from the CGGA website (http://www.cgga.org.cn/) (20). Validation set data downloaded from TCGA database (https://portal.gdc.cancer.gov) (21) and collates TCGA GBM and TCGA LGG project STAR process of RNAseq data and extract the TPM format of the data. This study adhered to the Declaration of Helsinki (2013 revision) and conformed to the publication guidelines provided by CCGA and TCGA. No studies on humans or animals by the authors were included in this study. The analysis flow chart of this study is shown in Figure 1 .

The clinical data and survival data of CGGA dataset and TCGA dataset were visualized by R software (R version 4.4.1) with R package “pheatmap[1.0.12]” and “ggplot2[3.4.4]”. GraphPad Prism 10 for macOS (version 10.4.0) software was used for statistical analysis of sequencing data differences between CGGA dataset and TCGA dataset. Statistical methods: Ordinary one-way ANOVA was used for WHO Grade, Unpaired t test was used for IDH and 1p/19q. The expression differences between groups of sequencing results were completed by R software (R version 4.4.1), and the R package was “ggplot2[3.4.4]” for graphical visualization. The ROC curve of sequencing data was analyzed and visualized by R software (R version 4.4.1), R package: pROC[1.18.0], ggplot2[3.4.4].

In the analysis of CGGA and TCGA databases, the researchers used R language (version 4.4.1) to perform Pearson correlation calculation, and set the significance threshold (P<0.05) and the absolute value of correlation coefficient criterion (>0.65) for the selection of relevant genes. The gene sets that met the criteria were imported into the DAVID bioinformatics platform (version 6.8) for functional annotation, which selected standard gene nomenclature and limited the research object to the human genome. Through this process, gene ontology (GO) functional annotation and KEGG pathway enrichment data were successfully obtained, and the top six items with significance level (P<0.05) were selected for display. In terms of data presentation, the GO analysis results of cell component (CC), molecular function (MF) and biological process (BP) were visualized by R software (R version 4.4.1) with R package “ggplot2[3.4.4]”. KEGG pathway data were visualized using Microsoft Excel (version 16.75.2) under Mac system.

CCGA data set by TIMER2.0 (http://timer.cistrome.org/) will be graded to immune cells infiltrating CGGA expression data, and then use the Python (Python version number 3.10.13) extraction data LOXL2 expression and merged with immune score, Spearman correlation coefficient was used to calculate the correlation coefficient, and significant results were screened and ranked for visualization of the results. The immune infiltration analysis of TCGA database was performed in R language environment (version 4.4.1) using the ssGSEA calculation method embedded in GSVA software package (version 1.46.0) (Hanzelmann et al., 2013). The 24 immune cell marker gene sets published by Bindea in the journal Immunity were integrated to systematically evaluate the immune microenvironment characteristics of samples (22, 23). Then the ggplot2 toolkit (version 3.4.4) was used to complete the chart visualization.

CCGA single-cell sequencing data sets were analyzed with R software (R version 4.4.1), cell clustering was performed with the Seurat package (version 4.0), and the results were visualized with the R package “ggplot2[3.4.4]”. Single-celled sequencing results of pretreatment process in accordance with the standards of previously reported methods, cell cluster similarity is through the same trait genes in clusters to define the number (24), first we define by CellMarker 2.0 (http://117.50.127.228/CellMarker/) cells, Oligodendrocyte:” OLIG2, FA2H, UGT8, CNP “; Tumor Cell: “PARP1, CD44”; Macrophage: “TSPO, CD63”; T Cell: “CD3, CXCR3, CCR6”. By UMAP dimensionality reduction, CGGA glioma single cells were divided into 14 cell clusters with clear transcriptomic characteristics. The 14 cell clusters were divided into four functional subgroups by unsupervised clustering analysis: Tumor Cell Area: cluster (2,3,5,6,7,10,17); Oligodendrocyte Area (0); Macrophage Area (1,4,8); T Cell Area (9,12), and the enrichment degree of LOXL2 in each cell cluster was observed by color depth on the UMAP map. Validation set us through Single Cell (https://singlecell.broadinstitute.org/single_cell). Enter the main page, The search strategy of single cell sequencing data set was “Metadata contains (species: Homo sapiens) AND (disease: glioma) AND (organ: Brain OR brain)”, and the target gene was selected as LOXL2. Cell cluster was defined as “Malignant, Stroma, T Cellsl, Other Immune”, which also had four functional cell subsets.

First, proportional hazards hypothesis test and Cox regression analysis were performed on CCGA and TCGA clinical survival data using R (4.2.1) version survival package [3.3.1], rms[6.3-0]. Variable screening strategy: if the sample in univariate satisfies the set P-value threshold, it will enter the multivariate Cox to construct the model. The P-value threshold is 0.1, and the univariate and multivariate COX analysis will be performed. Then, CGGA survival data were analyzed by GraphPad Prism 10 for macOS (version 10.4.0) software, logrank was used for statistical analysis, and the results were visualized by Python (Python version 3.10.13). TCGA survival data were completed using R software (R version 4.4.1), we used the R package “survival[3.3.1], survminer[0.4.9], ggplot2[3.4.4]”, and used the survival package to test the proportional hazards hypothesis and perform the fitting survival regression. The survminer package and ggplot2 package were used to visualize the results, and Cox regression was used for statistical analysis. In terms of prognostic model construction, time-dependent ROC analysis was performed by integrating CCGA and TCGA survival data and LOXL2 expression data by timeROC package pair, and the analysis results were visualized by ggplot2 to construct 1-year, 3-year and 5-year time-dependent ROC plots. The survival package was used to test the proportional hazards hypothesis and perform Cox regression analysis. The rms package was used to construct a nomogram related model and visualize it to construct a nomogram. The survival package was used for proportional hazard hypothesis testing and Cox regression analysis, and the rms package [6.3-0] was used for Calibration analysis and visualization. The survival package was used to fit the prognostic model, and the stdca R file was used for DCA analysis to construct the 1-year, 3-year, and 5-year prognosis DCA maps. The time-dependent Cindex was constructed by analyzing the data through the pec package. The C-index value of the survival model at different times was shown in the form of broken lines.

P < 0.05 was considered statistically significant. The following P values were considered: *P < 0.05, **P < 0.01, and ***P < 0.001, with statistical methods as described above.

The baseline data of glioma patients divided into groups based on the expression level of LOXL2 in CCGA ( Table 1 ) showed significant differences in key clinicopathological parameters: age distribution: patients in the LOXL2 high expression group were significantly older (45.7 ± 13.3 years vs 40.8 ± 10.0 years, p<0.001); Tumor grade: there was a significant difference in the distribution between groups (p<0.001); The patients with high LOXL2 expression were mainly WHO grade IV (35.5% vs 8.6%). The low LOXL2 expression group contained 29.6% WHO grade II tumors, which was significantly higher than 2.3% of the high LOXL2 expression group. The 1p/19q co-deletion was rare in the high expression group (0.7% vs 19.7%, p<0.001). IDH wild type was predominant in the high expression group (35.5% vs 10.9%, p<0.001). There was no significant difference in gender distribution between the two groups (p=0.193).

The analysis of baseline data of glioma patients based on TCGA database ( Table 2 ) showed that the expression level of LOXL2 was significantly correlated with clinicopathological characteristics: patients in the high LOXL2 expression group were significantly older (51.7 ± 15.8 years vs 42.5 ± 13.3 years, p<0.001); There was no significant difference in gender distribution between groups (p=0.449). The difference in WHO grade distribution was extremely significant (p<0.001). The high expression group was mainly WHO grade IV; The 1p/19q co-deletion rate in the low expression group was 20.6%, which was 4.9 times higher than that in the high expression group (4.2%) (p<0.001). The proportion of IDH mutant in the low expression group was 41.7%, which was significantly higher than that in the high expression group (20.6%, p<0.001). The proportion of IDH wild type in the high expression group was 30.6%, which was 4.3 times that in the low expression group (7.1%).

The clinical data and survival data of patients in CCGA and TCGA datasets are shown in Figure 2 . The gene expression data are shown by grouping differential expression ( Figures 3A, C, E, G, I, K ). There were significant differences in the expression of LOXL2 in WHO Grade, isocitrate dehydrogenase (IDH) mutation and 1p/19q deletion groups between CCGA and TCGA datasets (P<0.0001). ROC curve showed that WHO Grade G2G3&G4 (CGGA: 0.813; TCGA: 0.879), IDH Mutant& Wildtype (CCGA: 0.750; TCGA: 0.800), 1p/19q Codel& non-codel (CGGA: 0.848; TCGA: 0.787) ( Figures 3B, D, F, H, J, L ). This analysis revealed that high expression of LOXL2 was strongly correlated with advanced pathological characteristics of glioma, IDH wild type and 1p/19q non-codeletion, suggesting its potential value as a biomarker for malignant progression of glioma.

In order to explore the biological functions related to LOXL2, Pearson correlation analysis (|R| > 0.65, P < 0.05) was used to screen the genes with the highest correlation with LOXL2 in TCGA and CGGA databases. GO and KEGG analyses were performed based on the above gene sets. Through functional enrichment analysis of the cross-dataset (CCGA and TCGA) of LOXL2 in glioma, we found that its core functions were significantly focused on extracellular matrix (ECM) dynamic remodeling and tumor invasion microenvironment regulation. Among the biological processes (BP) ( Figures 4A, D ), LOXL2 strongly drives angiogenesis, collagen fiber assembly and ECM remodeling. Cell localization (CC) showed that LOXL2 was mainly enriched in collagenetic ECM structures, focal adhesion and ER lumen ( Figures 4B, E ). At the molecular function (MF) level ( Figures 4C, F ), LOXL2 specifically binds to collagen and integrin, and endows the ECM with tensile strength, providing a mechanical shelter microenvironment for glioma stem cells. KEGG pathway analysis further revealed that LOXL2 may activate the FAK/PI3K-Akt signaling axis through ECM-receptor interaction ( Figures 4G, H ), suggesting its potential role in glioma immune escape.

In order to explore the immune infiltration function related to LOXL2, the data analysis results were visualized ( Figures 5A, B ). The size of each circle represents the correlation between the enrichment degree of immune cells and LOXL2, and the color of the circle represents the P value. The distance between the circle and the baseline also indicates the correlation between the enrichment of the immune cell and LOXL2, which is the same as the significance of the size of the circle. Subsequently, the top 3 immune cells with the strongest correlations and significant differences in p-values were selected ( Figures 5C–N ). The expression of LOXL2 was strongly positively correlated with the infiltration of macrophages (CGGA: R=0.69, TCGA: R=0.61) and neutrophils (CGGA: R=0.61, TCGA: R=0.55) (all P<0.001). There was a significant correlation between CCGA-NK cell infiltration (R=0.64, P<0.001), suggesting that LOXL2 may affect immune surveillance through NK cell activation pathway. Tcga-nk cell correlation (R=0.52, P<0.001), suggesting that LOXL2 is involved in the regulation of Th2-type immunosuppressive microenvironment. Through immune infiltration analysis, we found that LOXL2 may regulate the immune function of glioma through the cooperation of macrophages and neutrophils.

We through the single-cell transcriptome analysis reveals LOXL2 based in glioma regulatory role in the immune microenvironment CCGA glioma single-celled sequencing data, through the UMAP dimension reduction could be divided into 14 CGGA glioma single transcriptome study characteristics of clear cell clusters. The 14 cell clusters were divided into four functional subgroups by unsupervised clustering analysis: Tumor Cell Area: cluster (2,3,5,6,7,10,17); Oligodendrocyte Area (0); Macrophage Area (1,4,8); T Cell Areas (9,12) ( Figure 6A ). The expression of LOXL2 in the target Cluster was 3.55 times higher than that in the control group. 28.7% of the cells in Cluster 7 expressed LOXL2, while only 13.6% of the cells in the control group expressed LOXL2, with a 2.1-fold difference. LOXL2 was specifically highly expressed in cluster 7, which was located in the Tumor Cell Area defined by us and may be a key marker, suggesting that LOXL2 is closely related to tumors ( Figure 6B ). According to the grouping difference ( Figures 6C–H ), most of LOXL2 was enriched in the Aacrophage Area and Tumor Area. By single database SingleCell (https://singlecell.broadinstitute.org) analysis of the independent validation set ( Figures 6I, J ), we observed a similar population distribution characteristics, four types of cells Malignant Cells, Stromal cells, Other Immune cells and T Cells. Across data set analysis shows that LOXL2 genes show significant cell type specific expression pattern: in the CCGA data mainly occurred in macrophages and tumor cells, while in validation specificity high expression in malignant cells and stromal cells. This cross-platform and repeatable expression pattern suggests that LOXL2 may affect the tumor microenvironment through dual mechanisms, participate in the regulation of tumor cell malignant phenotype and affect the function of immune cells through stromal cell remodeling.

Univariate and multivariate Cox regression analysis of overall survival (OS) showed that the univariate and multivariate Cox regression analysis of Age, Grade, IDH and 1p/19q in CCGA and TCGA datasets were p<0.001, suggesting that in Cox regression analysis, LOXL2 expression was an independent prognostic factor independent of known prognostic factors, including WHO grade, age at diagnosis, IDH mutation and 1p/19q deletion. These findings suggest that LOXL2 in CGGA and TCGA database are independent prognostic factors ( Tables 3 , 4 ).

To explore the prognostic predictive value of LOXL2 in glioma patients, Kaplan-Meier analysis and Cox proportional hazard model analysis were performed based on CGGA and TCGA database ( Figures 6A, B ). In CGGA database, the overall survival of patients with high expression of LOXL2 was significantly shorter than that of patients with low expression of LOXL2, HR=4.60 (Cl: 3.42-6.83), with statistical significance (P<0.001). In addition, the prognostic value of LOXL2 was also verified in TCGA database, with HR=4.06 (Cl: 3.08-5.87), which was statistically significant (P<0.001). The time-dependent ROC curve showed that CCGA 1-year AUC=0.817, TCGA 1-year AUC=0.793, CCGA 3-year AUC=0.897, TCGA-3 year AUC=0.776, TCGA 5-year AUC=0.730, CCGA 5-year AUC=0.925 ( Figures 7 C, D ).

We combined CCGA clinical survival data and LOXL2 expression data. On the basis of multivariate regression analysis, a prognostic Nomogram was established, and multiple predictors were integrated by using graduated line segments and plotted on the same plane according to a certain proportion to express the relationship between each predictor variable in the prediction model. A prognostic nomogram and a prognostic calibration curve were constructed ( Figures 7E, F ). In order to benefit patients with clinical utility or evaluation model, we also built a 1 year, 3 years, 5 years of decision-making curve analysis (DCA)( Figures 7G–I ). the value of the note is LOXL2 DCA curves in the threshold probability (Pt) is about 0.25 (1 year), 0.50 (3 years), There was a significant peak of receiving treatment benefit at 0.62(at 5 years). The time-dependent C-index was balanced around 0.7 at 500, 1000, 2000, 3000, and 4000 days, indicating that LOXL2 has important translational value in predicting the prognosis of glioma.