Drug metabolism-related gene ABCA1 augments temozolomide chemoresistance and immune infiltration abundance of M2 macrophages in glioma

We were directed at identifying the drug metabolism-related genes potentially involved in the in vivo process of medications administered for glioma treatment. The Gene Expression Omnibus database (GEO database) [20] is an internationally free public repository that embodies plentiful genomics data and comprehensive clinical information on various diseases. As shown in Table 1, three GEO datasets of glioma, including GSE4290 [21], GSE15824 [22], and GSE2223 [23‐25], were extracted under the following conditions: ⑴ GEO datasets: glioma, ⑵ organism: homo sapiens, ⑶ study type: expression profiling by arrays. Then, samples in the above datasets were analyzed with GEO2R to obtain the differentially expressed genes (DEGs) in glioma and normal brain tissues. The screening criteria of the DEGs were defined as |logFC |> 1 and p < 0.01. Subsequently, we applied the Omicstudio [26] tool for Venn analysis to screen out one co-differentially expressed gene, ABCA1, which was the overlapping area of drug metabolism-related genes and DEGs in glioma.

Several online databases and tools were utilized to assess whether ABCA1 had potential diagnostic and prognostic value in glioma. The Xiantao tool (https://​www.​xiantaozi.​com/​) is an online platform for data analysis and graph drawing that can access and analyze detailed RNA-sequencing data and clinical information on various diseases from the Cancer Genome Atlas (TCGA) database. Hence, we used the Xiantao tool for prognostic analyses. Meantime, we used the BEST database (https://​rookieutopia.​com/​app_​direct/​BEST/​) and the PanCanSurvPlot [27] tool to confirm intensively. All samples involved in the prognosis analysis were categorized into low-expression and high-expression groups by the median expression of ABCA1. The log-rank significance test was used for the statistical methods to analyze the relationship between ABCA1 expression level and the survival rate of patients, with a 95% confidence interval (CI).

Moreover, receiver operation characteristic (ROC) curves plotted by the Xiantao tool were utilized to predict the diagnostic value and effectiveness of ABCA1 in glioma. A total of 1846 samples, 1152 normal samples from the GTEx project (Genotype-Tissue Expression) and 694 tumor samples from the TCGA platform, were included. The area under the curve (AUC) reflects the predictive power of ABCA1, whose numerical value closer to 1 indicates better diagnostic performance, significantly when exceeding 0.9.

Considering the differential expression of ABCA1 showing statistical and clinical significance in the prognosis of glioma, we further explored its expression level in tumor and normal tissues via multiple databases and statistical software. The expression data from GEO datasets, including GSE4290, GSE15824, and GSE2223, were analyzed by GraphPad Prism (version 8.0.2). In addition, the UCSC database [28] had integrated abundant gene expression data from cancer research projects and individual labs, through which we downloaded the RNA-sequencing data (TCGA TARGET GTEx: RSEM tpm, n = 19,131) of pan-cancer and extracted ABCA1 expression data of 662 glioma samples and 1157 normal brain samples. We then compared the expression level of ABCA1 in the glioma tissues and normal brain tissues.

Subsequently, we successively applied three large-scale web databases for further verification. The GEPIA2 [29], a comprehensive web database for large-scale gene expression analysis, was exploited to verify the mRNA expression difference. The UALCAN database [30], an interactive web resource providing high-quality gene and protein expression analysis of various tumors, was utilized to compare the mRNA and protein expression levels. The HPA database [31], aiming at showing all proteins in the human body at the level of cells and tissues, was used to obtain the Immunohistochemical staining pictures of glioma and normal brain tissues.

TMZ is an indispensable drug prescribed for the standard chemotherapy of glioma, but TMZ resistance appears in many cases [32]. Thereby, we explored the correlation between ABCA1 overexpression and the drug activity of TMZ, intending to investigate whether ABCA1 enhanced the chemoresistance of TMZ in glioma treatment.

Gene correlation and enrichment analysis were performed to understand the role of ABCA1 and the biological processes in glioma. The LinkedOmics database [38] integrates, analyzes, and compares the multi-omics data and clinical information across all TCGA tumors for genomics researchers. We conducted a gene correlation analysis associated with ABCA1 through the LinkFinder analytical module. Volcano plots and heat maps were employed to display the co-expressed genes of ABCA1. We simultaneously analyzed survival heat maps of the top 50 genes associated with ABCA1 via the GEPIA2 database. Moreover, another analytical module, the LinkInterpreter, carried out the gene set enrichment analysis, including gene ontology cellular component (GO-CC), gene ontology biological process (GO-BP), gene ontology molecular function (GO-MF) and Kyoto encyclopedia of genes and Genomes (KEGG) pathway.

The Pearson correlation between ABCA1 expression and the abundance of immune cells in glioma tissues from the TCGA database was evaluated using the Xiantao tool, which applied the ssGSEA algorithm to estimate immune infiltration [39, 40]. Subsequently, the TISIDB [41] and HPA databases were utilized to verify the above immunological results. The Tumor Immune Single-cell Hub2 database (TISCH2) [42], providing a platform for investigating the tumor microenvironment, was also applied to further validate at the single-cell level. Moreover, we investigated the immune molecules correlated to ABCA1 expression using the Xiantao tool and TISIDB database. Additionally, TIMER2.0 [43], an integrated web database aiming to estimate the immune infiltration across various TCGA tumors systematically, was adopted to ascertain whether ABCA1 expression and immune infiltrates correlated with the clinical outcomes of glioma patients.

GraphPad Prism (version 8.0.2) was used for statistical analysis. Kaplan–Meier analysis was applied to assess survival rate or probability for glioma patients. The unpaired T-test was adopted for the statistical hypothesis test to compare the differential expression level of ABCA1 in the glioma tissues and normal brain tissues. Correlation analysis included Pearson correlation test and Spearman correlation test. All experiments were conducted no less than three times with mean ± standard deviations (SD) for analysis. The p-value less than 0.05 was statistically significant (*p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001).

To ferret out the potential drug metabolism-related genes in glioma, we retrieved and analyzed three gene expression profiles of glioma and normal brain samples through the GEO database to acquire the differential expression genes with the screening criteria set as |logFC|> 1 and p < 0.01. As shown in Table 1, GSE15824 totally contains 30 glioma samples and five normal brain samples, in which 872 down-regulated and 750 up-regulated genes were screened out. GSE4290 embodies 157 glioma samples and 23 normal brain samples, with 1938 down-regulated and 1801 up-regulated genes screened out. GSE2223 includes 50 glioma and four normal brain samples, with 602 down-regulated and 343 up-regulated genes screened out. In the meantime, we collected certain drug metabolism-related genes reported to be involved in drug disposition, metabolism, and resistance to cancer chemotherapeutic agents [44]. Then, we conducted a Venn analysis to identify the co-DEGs from the above differentially expressed genes and the drug metabolism-related genes. One co-differentially expressed gene, ABCA1, was finally discovered and pended for further analysis (Fig. 1a, b).

We explored the ABCA1 associated clinical outcomes for glioma patients through several web databases and bioinformatics tools. We started with using the ABCA1 gene expression data and corresponding clinical information to assess the overall survival (OS), disease-specific survival (DSS), and progression-free interval (PFI) of glioma patients in the TCGA database. The result displays that glioma patients with higher ABCA1 expression levels have encountered worse OS, DSS, and PFI (Fig. 1c–e). Further investigation into OS, DSS, and disease-free interval (DFI) through the PanCanSurvPlot tool demonstrates the same result (Fig. 1l–n). To be more convincing, the BEST database was employed to examine the survival of glioma patients in the TCGA, GEO, and CGGA databases. The survival plots reveal lower expression levels of ABCA1 associated with better OS for glioma patients (Fig. 1f–k). Subsequently, we evaluated the diagnostic performance of ABCA1 in distinguishing glioma tissues from normal brain tissues through the ROC curve. The numerical value of AUC (0.957, 95% CI 0.861–0.923) indicates that ABCA1 has a superb capability in glioma diagnosis (Fig. 1o). These results initially imply that ABCA1 has important clinical significance for glioma patients and is worth profound investigation.

According to the analysis of ABCA1 expression data in three GEO datasets and the TCGA database, we discovered that the expression level of ABCA1 in glioma was significantly elevated in comparison to the normal brain tissues (Fig. 2a–f). The analysis results based on 681 glioma tissues and 207 normal tissues from the GEPIA2 database display higher ABCA1 expression levels in glioma than in normal tissues (Fig. 2g). Additionally, data from the UALCAN database indicate that ABCA1 is over-expressed in glioma samples at the level of mRNA and protein (Fig. 2h, i). The immunohistochemical staining of ABCA1 shows intense staining in glioma tissues and weak in normal tissues, which reveals the highly expressed level of ABCA1 in glioma patients (Fig. 2j).

The CellMinerCDB platform conducted a correlativity analysis of ABCA1 expression and TMZ activity in glioma cell lines to investigate whether ABCA1 augmented TMZ resistance in the treatment of glioma patients. The diagram exhibits that ABCA1 expression remarkably negatively associates with TMZ resistance in central nervous system tumors, especially in diffuse glioma (Pearson r = − 0.41, p = 0.038) and astrocytoma (Pearson r = − 0.55, p = 0.0043) (Fig. 3a–c). Cell experiments were followed for further exploration and verification of the above analysis. The mRNA expression data measured by real-time PCR revealed that the expression level of ABCA1 in TMZ-resistant glioma cells (U118-R and T98G-R) was significantly higher than that in non-resistant glioma cells (U118 and T98G) (Fig. 3d). The results of analyzing the protein expression levels of ABCA1 with Western blotting showed that ABCA1 in TMZ-resistant glioma cells was significantly higher than that in non-resistant glioma cells (Fig. 3e). Reduced expression levels of ABCA1 (siABCA1-1 and siABCA1-2) in TMZ-resistant glioma cells U118-R and T98G-R were detected by Western blotting, which indicated that ABCA1 expression successfully suffered from interference and knockdown (Fig. 3f). Cell death of U118-R and T98G-R cells increased after ABCA1 expression had been interfered by siRNA. After ABCA1 got knocked down, treatment with TMZ at 100 μM could prominently facilitate the death rate of U118-R and T98G-R cells (Fig. 3g, h). The relative cell survival of U118-R and T98G-R cells significantly reduced with the treatment higher level of TMZ, and declined more sharply after ABCA1 knockdown (Fig. 3i, j). Moreover, the cell clone forming rate of U118-R and T98G-R cells decreased after ABCA1 knockdown. Based on ABCA1 knockdown, treatment with TMZ at 100 μM strikingly inhibited the clone forming of U118-R and T98G-R cells (Fig. 3k, l). The data of colony formation analysis and CCK-8 showed ABCA1 knockdown significantly reduced proliferation and colony formation ability of glioma cells.

Subsequently, MOE software was employed to predict the affinity of ABCA1 and TMZ. HSA was selected as a positive control to assess the binding intensity of TMZ to ABCA1. The negative number of S was negatively correlated to the binding affinity between TMZ and proteins. The results of molecular docking between TMZ and different protein structures of ABCA1 displayed that the smallest negative numbers of S were − 7.2662 (PBD_5XJY, Fig. 4a) and − 5.5759 (PBD_7ROQ, Fig. 4b). As to HSA in different protein conformations, the smallest negative numbers of S were − 4.9861 (PBD_3B9L, Fig. 4c) and − 4.7540 (PBD_3JRY, Fig. 4d). Other docking results are shown in Table 2. The docking scores of ABCA1 and TMZ were relatively high compared to HSA and TMZ. These results indicated that the interaction strength of TMZ and ABCA1 is greater than that of TMZ and HSA.

The LinkedOmics database was adopted for gene correlation and enrichment analysis of the TCGA glioma to investigate the biological roles of ABCA1 in glioma. Figure 5a shows that 6785 genes were positively associated with ABCA1, while 6381 genes were negatively correlated with ABCA1 (p < 0.05). Additionally, the heatmaps presented the top 50 genes associated with ABCA1 negatively and positively (Fig. 5b, c). The results of survival heatmaps showed that positive co-expressed genes with a risky hazard ratio (25/50) acted as risk factors. In contrast, negative co-expressed genes with a protective ratio (7/50) were more likely to be protective factors (Fig. 5d). GO annotations of the biological process described that ABCA1 was chiefly involved in the process of immune regulation and mitochondrial metabolism, such as cytokine metabolism, cytokine production, immune cell activation, complex assembly, and so on (Fig. 5e). Results of KEGG pathways depicted the top pathways that ABCA1 participated in, including oxidative phosphorylation, Parkinson’s disease, and infection (Fig. 5f). Moreover, GO annotations of cellular components and molecular function indicated that ABCA1 mainly functioned in immune regulation and mitochondrial metabolism (Additional file 1: Fig. S1a, b).

A chart displayed the correlation analysis results of ABCA1 and immune cell infiltration. It revealed that ABCA1 expression had a positive correlation with the abundance of T helper cells, T gamma delta, macrophages, T effector memory cells, T central memory cells, neutrophils, activated dendritic cells (aDC), eosinophils and Th17 cells in glioma (p < 0.05) (Fig. 6a). Then, the above results were verified by the TISIDB database. The consistent correlation of ABCA1 expression was discovered with the infiltration level of macrophages, neutrophils, aDC, and Th17 cells (Fig. 6b, c, Additional file 2: Fig. S2a–f). Notably, a difference existed in that CD163 (CD163 molecule; M2 macrophages), HLA-DPB1 (major histocompatibility complex, class II, DP beta 1; aDC), and STAT (signal transducer and activator of transcription; Th17 cell) were intensely stained in the immunohistochemistry images of glioma tissues, while NOS2 (nitric oxide synthase 2; M1 macrophages) and CD66b (CEACAM8, CEA cell adhesion molecule 8; neutrophil) were nearly not detected (Fig. 6d, e, Additional file 2: Fig. S2g–i). Hence, we used the TISCH2 database to intensively investigate the cell types of immune infiltration in glioma samples from the datasets of GSE148842 and GSE162631 at the single-cell level. The diagram depicted that across the immune cells infiltrating the glioma tissues, the preponderant ones were monocytes/macrophages (mainly M2 macrophages), and other immune cells (CD8 T cells and M1 macrophages) were few or even undetectable (Fig. 6f–o). All the results revealed that a mounting infiltration abundance of M2 macrophages in glioma was primarily related to the highly expressed level of ABCA1.

Based on the notable correlation of ABCA1 with immune infiltration, we explored the association between ABCA1 and immune molecules related to macrophages in GBM and LGG [45]. Immunoinhibitors (IL-10, interleukin 10; CSF1R, colony stimulating factor 1 receptor; TGFB1, transforming growth factor beta 1; TGFBR1, transforming growth factor beta receptor 1), immunostimulators (CD86, CD86 molecule; IL-6, interleukin 6; IL-6R, interleukin 6 receptor), and chemokines (CCL18, C–C motif chemokine ligand 18; CCL22, C–C motif chemokine ligand 22) own positive correlations with ABCA1 expression (Fig. 7). Additionally, we explored whether immune infiltrates owned a potential correlation with the clinical outcomes of glioma patients. The results displayed that at a certain level of gene expression, macrophages with high infiltrating abundance acted as an adverse hazard ratio and were associated with poor cumulative survival in patients with glioma (Additional file 2: Fig. S2j, k).