2.2. Construction of the LCDGs risk signature

To identify the hub LCDGs, differential expression analysis, univariate Cox regression analysis, weighted gene co-expression network analysis (WGCNA), and least absolute shrinkage and selection operator (LASSO) analysis were performed on TCGA-glioma samples. The univariate Cox regression analysis utilized a cutoff criterion of P < .001. Differential expression analysis was conducted using the R package “limma,” with criteria of |log fold change (FC)| > 1 and P < .05. The “WGCNA” package was employed to establish a co-expression network linking LCDGs with sample modules, determining the best soft threshold based on the scale-free topological criterion (>0.86) and a minimum power value. Hierarchical clustering was then used to construct dendrograms based on the topological overlap matrix (TOM). The number of generated modules was controlled by defining main parameters with a minModuleSize of 10. LCDGs displaying significant interconnections were assigned to various patterns. Gene significance and module eigengenes were calculated for all modules. The “VennDiagram” package facilitated visualization of the overlap of LCDGs among WGCNA, differential expression, and univariate Cox regression analyses. These genes were further integrated into the LASSO analysis to identify the hub LCDGs in glioma for further investigation.

LCDGs risk signatures were generated based on the expression of 14 identified hub LCDGs. The risk score for the signatures was calculated using the formula: risk score = Σexpi × βi (expi represents the relative expression of LCDGs i, and β is the regression coefficient). Glioma patients were then divided into low- and high-risk subgroups based on the median value of the calculated risk score. Cox regression and Kaplan–Meier (KM) survival analyses were conducted to assess and compare the prognostic ability using the “survival” package. The accuracy of the risk signature for prediction was calculated using the “timeROC” and “cliROC” packages. Finally, we utilized the “rms” package to construct a nomogram that incorporated the risk scores of the gene signature for predicting the outcomes of glioma patients.

2.3. Immune infiltration analysis

Firstly, CIBERSORT analysis was conducted to assess immune cell infiltration in glioma and normal samples. Additionally, the relationship between the expression of hub LCDGs and immune cell infiltration was also assessed by CIBERSORT analysis. Subsequently, single-sample gene set enrichment analysis (GSEA) was performed to explore discernible differences in immune cell infiltration and immune functions among different risk subgroups of glioma. The relationship between the scores of immune-associated factors (including stromal and immune scores) and the risk signature was evaluated through Spearman correlation analysis. The connection between the risk signature and immune-associated genes was also investigated by referencing previously identified potential immune checkpoints.^[16]

2.4. Machine learning model construction

The LCDGs identified from LASSO analysis were used to construct 4 different machine learning models: Extreme Gradient Boosting, Support Vector Machine, Random Forest, and Generalized Linear Models. These models were constructed using R packages such as “xgboost,” “randomForest,” “dalex,” and “caret,” respectively. Residual analysis was applied to the samples to evaluate the accuracy of these models, followed by plotting the reverse cumulative distribution of residuals. Receiver operating characteristic (ROC) curves for the 4 machine learning models were plotted using the “pROC” R package. Each LCDG in the models was assigned an importance score to filter out the key genes for glioma patients. Additionally, differential expression analysis and ROC analysis were conducted for each LCDG in glioma. Based on the results of the machine learning models, differential expression analysis, and ROC curves, the best feature among the LCDGs was selected for further analyses.

2.5. Clinical value of LAPTM4A in glioma, GBM, and LGG

Differential expression analysis was performed to evaluate the expression level of the LAPTM4A gene in glioma, GBM, and LGG samples. To compare the overall survival (OS) of LAPTM4A, KM survival analysis, and Cox regression analysis were conducted. The predictive accuracy of LAPTM4A was verified using the R package “timeROC.” A nomogram was constructed to predict the OS of glioma patients based on the expression level of LAPTM4A and other clinical features, using the “rms” package in R. Furthermore, GSEA and GSVA analyses were carried out to investigate the potential mechanism of LAPTM4A. GSEA was performed on the gene expression matrix using the “c2.cp.kegg.symbols.gmt” sets to detect Kyoto Encyclopedia of Genes and Genomes pathways. The potential biological functions of LAPTM4A were evaluated using the GSVA score matrix and the “GSVA” R package. Statistical significance was defined as P values < .05.

2.6. Pan-cancer analysis of LAPTM4A

RNA sequencing data from the TCGA database, comprising 33 distinct types of cancers, were utilized to visualize the differential gene expression of LAPTM4A between normal and cancerous tissues. KM analysis was employed to evaluate the prognostic value of LAPTM4A in different cancer types.

2.7. Chemotherapy sensitivity analysis

Chemotherapy drugs certified by the FDA or verified by clinical laboratory were obtained from the CellMiner database (https://discover.nci.nih.gov/cellminer). Pearson correlation analyses were conducted to assess the sensitivity of LAPTM4A to these drugs. The results were visualized using R packages such as “ggplot2,” “limma,” “ggpubr,” and “impute.”

2.8. Cell culture

The glioma cell line U373 and the normal control NHA cells were purchased from BNCC. They were cultured in Dulbecco Modified Eagle’s medium (Gibco, Thermo Fisher Scientific, Waltham, MA) supplemented with 10% fetal bovine serum (Hyclone, Logan, UT) and 1% penicillin/streptomycin (Solarbio, Beijing, China). The cells were maintained at 37°C in a 5% CO₂ incubator.

2.9. Quantitative real-time PCR (qRT-PCR)

RNA isolation was performed using Trizol reagent (Invitrogen, CA), followed by cDNA synthesis using a Transcriptor First-strand cDNA synthesis kit (Roche, Basel, Switzerland). Subsequently, qRT-PCR was conducted using SYBR green master mix. Each sample was analyzed in triplicate, and gene expression was quantified using the 2^−ΔΔCT method. The primer sequences are listed in Supplementary Table 1, http://links.lww.com/MD/L392.

3.2. Clinical value of risk signature in glioma and LGG

Based on the calculated median value of risk scores, glioma patients from the TCGA database were classified into high- or low-risk subgroups (Fig. ​(Fig.3A).3A). Investigating the association between LCDGs and glioma risk subgroups revealed higher expression of genes AP3S2, BORCS6, GATA2, HPS6, KIF1B, NCOA4, PSAP, and SPAG9 in the high-risk subgroup, while ACP2, AGA, CTNS, GNS, LAPTM4A, and SNAPIN showed lower expression (Fig. ​(Fig.3B).3B). Furthermore, the risk signature of LCDGs independently predicted outcomes for glioma patients (Fig. ​(Fig.3C-D).3C-D). KM survival analysis demonstrated a negative correlation between risk scores and OS of glioma patients (Fig. ​(Fig.3E).3E). The risk signature exhibited high predictive accuracy with AUC values of 0.867, 0.875, 0.845, 0.867, and 0.875 at 1-year, 3-year, 5-year, 7-year, and 9-year OS follow-up, respectively (Fig. ​(Fig.3F).3F). Compared to other clinical features such as gender, age, and histological type, the risk signature showed higher predictive accuracy for the OS of glioma patients (Fig. ​(Fig.3G).3G). Notably, patients aged > 60 (Fig. ​(Fig.3H),3H), with G3 tumor grade (Fig. ​(Fig.3I),3I), GBM histological type (Fig. ​(Fig.3J),3J), or progressive disease (Fig. ​(Fig.3K)3K) exhibited significantly higher risk scores compared to patients aged ≤ 60, G2 tumor grade, LGG histological types, or primary therapies. Evaluation of the risk signature’s predictive role in LGG prognosis revealed a significant negative correlation between risk scores and OS of LGG patients (Fig. ​(Fig.3L).3L). The LCDGs risk signature exhibited moderate predictive accuracy at 1-year, 3-year, 5-year, and 7-year follow-up with AUC values all above 0.7 (Fig. ​(Fig.3M).3M). Additionally, LGG patients aged > 60 showed significantly higher risk scores compared to those aged ≤ 60 (Fig. ​(Fig.3N).3N). However, due to the separation of all GBM patients into the high-risk subgroup in our LCDGs risk signature, this model could not be applied to analyze the prognosis of GBM patients separately. Finally, we created a nomogram plot for glioma patients incorporating clinical characteristics and the risk score (Fig. ​(Fig.4A),4A), demonstrating substantial agreement (Fig. ​(Fig.44B).

Open in a separate window

Figure 3.

The Role of LCDGs risk signature. (A) The risk score distributions in the TCGA-glioma patients. (B) The differential expression of LCDGs in the risk subgroups of TCGA-glioma patients. (C-E) The survival status of risk signature in the TCGA-glioma patients. Survival (F) and clinical (G) ROC analysis of risk signature in the TCGA-glioma patients. Correlations between risk scores and ages (H), tumor grades (I), histological types (J), and primary therapy outcomes (K) in glioma patients. (L-M) The survival status of risk signature in the GSE4290-LGG patients. (N) Correlations between risk scores and ages in LGG patients.

Open in a separate window

Figure 4.

Associations between risk signature and immune characteristics. (A-B) Nomogram for predicting the OS of risk scores. (C) The proportion of immune cells between normal control and glioma samples. Boxplots of scores of immune-associated functions (D) and immune cells (E) in risk signature. Associations between risk signature, stromal scores (F) and immune scores (G). (H) Expression of immune checkpoints among 2 risk subgroups in glioma patients.

To investigate immune infiltration in glioma, a CIBERSORT analysis was conducted. Figure ​Figure4C4C shows significant differences in the infiltration levels of several immune cell types between glioma and control samples, including memory B cells, CD8 T cells, CD4 memory resting T cells, regulatory T cells, resting/activated NK cells, M0/M1/M2 macrophages, resting dendritic cells, activated mast cells, and eosinophils. Patients with high-risk scores exhibited significantly increased immune cell functions and subpopulations, except for aDCs, DCs, mast cells, neutrophils, and NK cells (Fig. ​(Fig.4D-E).4D-E). Moreover, the risk signature showed a positive correlation with stromal (Fig. ​(Fig.4F)4F) and immune (Fig. ​(Fig.4G)4G) scores, indicating its association with the immune microenvironment. Differential expression of various immune checkpoints was observed within the 2 risk subgroups. Patients with high-risk scores showed elevated expression of almost all immune biomarkers, such as TNFRSF18, TNFRSF4, and TNFRSF14, except for VTCN1, TNFSF18, HHLA2, CD200, TMIGD2, and ADORA2A (Fig. ​(Fig.44H).

3.3. Identification the core LCDG LAPTM4A in glioma, GBM, and LGG

The analyses of 4 different algorithms consistently demonstrated small residuals ranging from 0 to 0.1 for all 4 algorithms employed, as represented by box plots (Fig. ​(Fig.5A)5A) and reverse cumulative distribution (Fig. ​(Fig.5B).5B). The diagnostic models’ high accuracy was confirmed by the results of ROC curves, with all 4 algorithms exhibiting AUC values close to 1 (Fig. ​(Fig.5C).5C). Evaluating the importance scores of the 4 algorithms, LCDGs LAPTM4A and HPS6 demonstrated the highest score (Fig. ​(Fig.5D).5D). To further validate if the hub LCDGs significantly predict glioma occurrence, we conducted a disease-specific association ROC curve for 14 identified LCDGs. The AUC values for LAPTM4A and HPS6 were 1.000 and 0.999, respectively, indicating their specific predictive ability for glioma (Fig. ​(Fig.5E).5E). Through differential expression analysis, both HPS6 (Fig. ​(Fig.5F)5F) and LAPTM4A (Fig. ​(Fig.5G)5G) showed significant upregulation in TCGA-glioma patients. However, in the GSE4290 validation cohort, only LAPTM4A exhibited significant differential expression in glioma tissues (Fig. ​(Fig.5I),5I), whereas there was no significant value for gene HPS6 (Fig. ​(Fig.5H).5H). The significant upregulation of LAPTM4A gene and protein expression levels were also confirmed in U373 glioma cells through qRT-PCR (Fig. ​(Fig.5J)5J) and immunofluorescence analysis (Fig. ​(Fig.5K).5K). Consequently, LAPTM4A was identified as the core LCDG in glioma and employed for subsequent prognostic analysis.

Open in a separate window

Figure 5.

Identification of LAPTM4A in glioma. (A) Box plots of sample residuals of the 4 algorithms. (B) The reverse cumulative distribution map of model residuals. (C) ROC analysis of algorithms. (D) Importance score of feature genes in the model. (E) The ROC curve of LCDGs in glioma patients. Gene expression level of HPS6 (F) and LAPTM4A (G) in TCGA-glioma patients. Gene expression level of HPS6 (H) and LAPTM4A (I) in GSE4290-glioma patients. qRT-PCR (J) and immunofluorescence (K) analysis of LAPTM4A.

To further confirm the role of LAPTM4A in glioma prognosis, KM analysis indicated a significant negative association between LAPTM4A expression and OS of glioma (Fig. ​(Fig.6A)6A) and LGG (Fig. ​(Fig.6Q)6Q) patients, while no significant relationship was observed for GBM patients’ OS (Fig. ​(Fig.6M).6M). The AUC values for LAPTM4A in glioma (Fig. ​(Fig.6B)6B) and LGG (Fig. ​(Fig.6R)6R) OS ranged from 0.60 to 0.75, indicating moderate accuracy in predicting the prognosis of glioma and LGG. Figure ​Figure6C6C and D demonstrated that LAPTM4A served as an independent prognostic factor for OS in glioma patients. Interestingly, we observed significantly higher levels of LAPTM4A expression among glioma patients aged > 60 compared to those aged ≤60 (Fig. ​(Fig.6E).6E). Additionally, patients with G3 tumor grade (Fig. ​(Fig.6F),6F), GBM histological type (Fig. ​(Fig.6G),6G), or progressive tumor status (Fig. ​(Fig.6H)6H) also exhibited significantly higher LAPTM4A expression, suggesting a potential correlation between LAPTM4A and glioma progression. When exploring LAPTM4A gene expression level in GBM, TCGA (Fig. ​(Fig.6I),6I), GSE4290 (Fig. ​(Fig.6J),6J), and GSE22866 (Fig. ​(Fig.6K)6K) GBM patients all showed a significant upregulation compared to normal controls. LAPTM4A exhibited perfect predictive accuracy for the occurrence of GBM (Fig. ​(Fig.6L).6L). Similar results were found in LGG patients, with significant upregulation of LAPTM4A expression in TCGA (Fig. ​(Fig.6N)6N) and GSE4290 (Fig. ​(Fig.6O)6O) tumor samples. In LGG, LAPTM4A achieved an AUC of 1.00 (Fig. ​(Fig.6P).6P). To predict glioma patients’ OS, a nomogram plot incorporating LAPTM4A expression and relevant clinical characteristics was developed (Fig. ​(Fig.7A-B).7A-B). Moreover, tight associations were observed between LAPTM4A and several pathways, such as allograft rejection, asthma, primary immunodeficiency, and natural killer cell-mediated cytotoxicity, as revealed by GSVA and GSEA enrichment analyses (Fig. ​(Fig.7C,7C, D). Finally, LAPTM4A exhibited significant and distinct correlations with various chemotherapy drugs, indicating a negative association between elevated LAPTM4A expression and sensitivity to drugs like 6-mercaptopurine, tamoxifen, and imatinib (Fig. ​(Fig.77E).

Open in a separate window

Figure 6.

The role of LAPTM4A in glioma, LGG, and GBM. The OS of LAPTM4A in glioma were analyzed by KM (A), timeROC curve (B), univariate (C) and multivariate Cox (D) regression analyses. Correlations between LAPTM4A expression level and ages (E), tumor grades (F), histological types (G), and primary therapy outcomes (H) in glioma patients. Gene expression level of LAPTM4A in TCGA (I), GSE4290 (J), and GSE22866 (K) GBM patients. ROC curve (L) and KM (M) analyses of LAPTM4A in GBM. Gene expression level of LAPTM4A in TCGA (N) and GSE4290 (O) LGG patients. ROC curve (P), KM (Q), and time survival ROC curve (R) of LAPTM4A in LGG.

Open in a separate window

Figure 7.

Functional analyses and immune therapy of LAPTM4A. (A-B) Nomogram of LAPTM4A for predicting the OS of glioma. (C) GSEA analysis. (D) GSVA analysis. (E) Drug sensitivity analysis of LAPTM4A.