Identification of ACBD3 as a new molecular biomarker in pan-cancers through bioinformatic analysis: a preclinical study

We used the “TISSUE” module of Human Protein Atlas (THPA) database (https://​www.​proteinatlas.​org) in December 2022, which works to provide information on the tissue and cellular distribution of all 24,000 human proteins, to investigate ACBD3 expression from the normal tissue atlas and tumor cell lines. And then explored ACBD3 expression in cells using the “SUBCELL” module of THPA database. We downloaded the relevant clinical and RNA-sequencing (RNA-seq) data for normal tissues and 33 tumors from The Cancer Genome Atlas (TCGA) database on 23 February 2023 [14]. R software (version 4.2.1) and ggplot2 package (version 3.3.6) were performed to generate statistical analysis and visualizations, respectively. The Wilcoxon rank-sum test and T test were performed to compare the differences in ACBD3 expression levels between cancer and normal tissues, and between cancer and paracancerous tissues. The Cancer Cell Line Encyclopedia (CCLE) database was used to verify gene expression in pan-cancers on 14 May 2023 [15]. Immunohistochemistry images of different cancers and normal tissues were obtained from the “TISSUE” and “PATHOLOGY” module of THPA database [16]. The data on the relationship between ACBD3 expression and immune infiltrates among various tumors were obtained from the “Immune-Gene” module of TIMER2.0 database in December 2022, which used six state-of-the-art algorithms to provide more reliable estimates of immune infiltration levels for tumor profiles [17].

The ACBD3 expression among various tumor pathological stages was visualized using the “Stage Plot” module of the gene expression profiling interactive analysis (GEPIA2) database (Used on 28 February 2023). ACBD3 phosphoprotein levels in normal tissues and different cancers were obtained by searching for “PhosphoProtein” module of ACBD3 from the Clinical Proteomic Tumor Analysis Consortium (CPTAC) database in March 2023 [18]. Furthermore, we entered ACBD3 in the “Gene Symbol” module of the TISIDB (an integrated repository portal for tumor-immune system interactions) database [19], a website for analyzing interactions between tumors and the immune system, to observe the correlation between ACBD3 expression and various molecular and immune subtypes among different cancers.

We obtained 50 available experimentally determined ACBD3-binding proteins from the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database in February 2023 by searching ACBD3 and Homo sapiens organism. We regulated the parameters as follows: “full STRING network”, “evidence”, “Experiments, Textmining, Databases”, “medium confidence (0.400)”, and “no more than 50 interactors” in the 1st shell. Subsequently, the protein − protein interaction (PPI) network was visualized using Cytoscape (version 3.9.1). Next, we retrieved the top 100 ACBD3-related target genes from the “Similar Gene Detection” module of the GEPIA2 database. A Venn diagram was constructed to compare ACBD3-binding and ACBD3-related target genes. Moreover, the possible functions of ACBD3-binding genes were explored using Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses by the R package “ClusterProfiler” (version 4.4.4).

We investigated the diagnostic and prognostic values of ACBD3 using relevant pan-cancer clinical data from TCGA database. The diagnostic performance of ACBD3 in pan-cancers was assessed using the receiver-operating characteristic (ROC) curve and the area under the ROC curve (AUC) applied by the “pROC” package (version 1.18.0). The included AUC values ranged from 0.7 to 1. As a measure of diagnostic accuracy, when the value of the AUC is closer to 1, the diagnostic value is higher. Furthermore, we constructed Kaplan–Meier (K-M) plots using the “survival” and “survminer” package (version 3.3.1) to demonstrate the relationship between ACBD3 expression level and the prognosis [OS (overall survival), DSS (disease-specific survival), and PFI (progress-free interval)] of various tumors [20], the Gene Expression Omnibus (GEO) database was used to verify this relationship [21].

The cBioPortal website was used to investigate genetic alterations in ACBD3 on February 2023. Next, we searched out the mutation type, the alteration frequency, and the copy number alteration (CNA) of all TCGA tumors from the “Cancer Types Summary” module. Furthermore, we obtained the three-dimensional (3D) diagram of ACBD3 structure by using the “Mutations” module. In addition, we also accessed the correlation between ACBD3 alternation and clinical outcomes in different tumors by performing the “Comparison/Survival” module. “Meth-Exp correlation” module of DNA Methylation Interactive Visualization Database (DNMIVD) (http://​www.​unimd.​org/​dnmivd/​) was used to assess the relationship between promoter methylation and ACBD3 expression, and “MethSurv” (https://​biit.​cs.​ut.​ee/​methsurv) was used to explore the effect of DNA methylation on tumor prognosis and the relationship between ACBD3 expression and methylation levels [22, 23].

These results suggested that ACBD3 was moderately expressed in the majority of normal tissues. ACBD3 was highly expressed in the cerebral cortex, hippocampus, duodenum, small intestine, colon, gallbladder, pancreas, prostate, placenta, appendix, and bone marrow. ACBD3 was expressed at low levels in the oral mucosa, liver, ovary, soft tissue, and adipose tissue (Additional file 1: Fig. S1A). The expression of ACBD3 ranked high in breast cancer, kidney cancer, and myeloma tumor cell lines (Additional file 1: Fig. S1B). Intracellular ACBD3 was mainly distributed in the Golgi apparatus (Additional file 1: Fig. S1C, D).

Furthermore, we found that ACBD3 expression was statistically upregulated in eleven cancer types as compared to normal tissues, including stomach adenocarcinoma (STAD), lung squamous cell carcinoma (LUSC), lung adenocarcinoma (LUAD), liver hepatocellular carcinoma (LIHC), kidney renal clear cell carcinoma (KIRC), head and neck squamous cell carcinoma (HNSC), glioblastoma multiforme (GBM), esophageal carcinoma (ESCA), colon adenocarcinoma (COAD), cholangiocarcinoma (CHOL), and breast invasive carcinoma (BRCA), and downregulated in uterine corpus endometrial carcinoma (UCEC), kidney chromophobe (KICH), and thyroid carcinoma (THCA) (Fig. 1A). The CCLE database revealed that ACBD3 was highly expressed in skin cutaneous melanoma (SKCM), BRCA, GBM, KIRC, brain lower grade glioma (LGG), and THCA (Additional file 1: Fig. S1E). In comparison with the adjacent normal tissues, ACBD3 expression was statistically upregulated in thirteen cancers, including STAD, prostate adenocarcinoma (PRAD), pancreatic adenocarcinoma (PAAD), LUSC, LUAD, LIHC, kidney renal papillary cell carcinoma (KIRP), KIRC, HNSC, ESCA, CHOL, BRCA, and bladder urothelial carcinoma (BLCA), and downregulated in KICH and THCA (Fig. 1B). We also constructed a violin diagram of the relationship between various pathological stages of tumors and ACBD3 expression (Fig. 1C). Immunohistochemistry images of normal breast tissue, liver tissue, lung tissue, kidney tissue, BRCA, LIHC, LUAD, and KICH were displayed (Fig. 2).

We assumed that the different expression levels of ACBD3 might affect the immune infiltration of various tumors. Therefore, we explored the correlation between ACBD3 expression and immune infiltration in different cancers using the TIMER2.0 database. Remarkably, the expression of ACBD3 was actively correlated with cancer-associated fibroblast (CAF) infiltration in HNSC (Additional file 1: Fig. S2A) and positively related to neutrophil infiltration in BLCA, COAD, and THCA (Additional file 1: Fig. S2B). ACBD3 expression also positively related to endothelial cell infiltration in COAD, HNSC, and KIRC (Additional file 1: Fig. S2C).

Figure 3 showed that ACBD3 expression was statistically different among the immune subtypes of nine tumors, including BLCA (Fig. 3A), HNSC (Fig. 3B), STAD (Fig. 3C), SKCM (Fig. 3D), sarcoma (SARC) (Fig. 3E), ovarian serous cystadenocarcinoma (OV) (Fig. 3F), LUSC (Fig. 3G), LIHC (Fig. 3H), and GBM (Fig. 3I). Moreover, for immune subtype C1 (wound healing), ACBD3 expressed high in BLCA, SKCM, SARC, and LUSC. For immune subtype C2 (IFN-gamma dominant), ACBD3 expressed high in HNSC, STAD, and OV. For immune subtype C4 (lymphocyte depleted), ACBD3 expressed high in LIHC, and GBM.

Furthermore, we identified twelve tumors with molecular subtypes associated with ACBD3 expression. Among the molecular subtypes of G-CIMP-high in LGG, ACBD3 showed the highest expression (Fig. 4A). Among the molecular subtypes of LunA in BRCA, ACBD3 showed the highest expression (Fig. 4B). Among the molecular subtypes of 1-ERG in PRAD, ACBD3 showed the highest expression (Fig. 4C). Among the molecular subtypes of Kinasesignaling in pheochromocytoma & paraganglioma (PCPG), ACBD3 showed the highest expression (Fig. 4D). Among the molecular subtype of CIN in COAD, ACBD3 showed the highest expression (Fig. 4E). Among the molecular subtypes of Immunoreactive and Proliferative in OV, ACBD3 showed the highest expression (Fig. 4F). Among the molecular subtypes of classical in LUSC, ACBD3 showed the highest expression (Fig. 4G). Among the molecular subtypes of iCluster:1 in LIHC, ACBD3 showed the highest expression (Fig. 4H). Among the molecular subtypes of CIN in STAD, ACBD3 showed the highest expression (Fig. 4I). Among the molecular subtypes of BRAF_Hotspot_Mutants in SKCM, ACBD3 showed the highest expression (Fig. 4J). Among the molecular subtypes of C1in KIRP, ACBD3 showed the highest expression (Fig. 4K). Among the molecular subtypes of Basal in HNSC, ACBD3 showed the highest expression (Fig. 4L).

We identified 50 ACBD3-binding proteins from the STRING database (Fig. 5A). Next, GO|KEGG enrichment analysis was conducted on the 50 proteins, revealing that the biological processes (BP) of the 50 proteins included Golgi organization, intermembrane lipid transfer, and Golgi vesicle budding. The cellular components (CC) of the 50 proteins were mainly involved in the Golgi apparatus subcompartment, endoplasmic reticulum-Golgi intermediate compartment, and trans-Golgi network. The molecular function (MF) of 50 proteins were primarily enriched in sterol binding, protein kinase A binding, and lipid transfer activity. The KEGG pathway enrichment of the 50 proteins was primarily related to endocytosis and cholesterol metabolism (Fig. 5B).

In addition, the top 100 ACBD3-related target genes were retrieved, and a Venn diagram identified the four genes at the intersection of the ACBD3-binding and ACBD3-related target genes as: ARF1, BLZF1, BPNT1, and GORASP2 (Fig. 5C). The expression levels of these four genes were closely related to those of ACBD3: ARF1 (R = 0.65), BLZF1 (R = 0.70), BPNT1 (R = 0.67), and GORASP2 (R = 0.62) (Fig. 5D–G).

The ROC curve was drawn to explore the diagnostic value of ACBD3 in different cancers, and an AUC value > 0.7 was considered to have a diagnostic value. As shown, ACBD3 had an accurate diagnostic value for 17 kinds of cancers, including CHOL (AUC = 0.990), LIHC (AUC = 0.863), STAD (AUC = 0.902), PAAD (AUC = 0.756), ESCA (AUC = 0.883), esophagus adenocarcinoma (ESAD) (AUC = 0.901), esophagus squamous cell carcinoma (ESCC) (AUC = 0.841), BRCA (AUC = 0.805), KICH (AUC = 0.895), Lung cancer (LUADLUSC) (AUC = 0.726), LUSC (AUC = 0.717), LUAD (AUC = 0.741), SARC (AUC = 0.930), PCPG (AUC = 0.739), GBM (AUC = 0.787), Glioma (GBMLGG) (AUC = 0.708), and Oral squamous cell carcinoma (OSCC) (AUC = 0.701) (Fig. 6).

Figure 7 showed that the OS, DSS, and PFI of four tumors were closely correlated with the expression levels of ACBD3, including PAAD, adrenocortical carcinoma (ACC), SARC, and GBMLGG. As for PAAD, the prognosis was negatively related to ACBD3 expression, including OS [hazard ratio (HR) = 1.63, 95% confidence interval (CI): 1.07–2.46, p = 0.022], DSS (HR = 1.67, 95% CI: 1.05–2.67, p = 0.032), and PFI (HR = 1.32, 95% CI: 0.90–1.94, p = 0.155). As for ACC, the prognosis was negatively related to ACBD3 expression, including OS (HR = 2.56, 95% CI: 1.17–5.61, p = 0.018), DSS (HR = 2.60, 95% CI: 1.15–5.88, p = 0.022), and PFI (HR = 4.02, 95% CI: 2.02–8.03, p < 0.001). Also, for SARC, the prognosis was negatively related to ACBD3 expression, including OS (HR = 1.66, 95% CI: 1.10–2.48, p = 0.015), DSS (HR = 1.86, 95% CI: 1.18–2.91, p = 0.007), and PFI (HR = 1.60, 95% CI: 1.14–2.24, p = 0.006). And for GBMLGG, the prognosis was negatively related to ACBD3 expression, including OS (HR = 1.47, 95% CI: 1.15–1.87, p = 0.002), DSS (HR = 1.47, 95% CI: 1.14–1.90, p = 0.003), and PFI (HR = 1.30, 95% CI: 1.05–1.61, p = 0.014).

We used the GSE57495, GSE83300, and GSE19750 in the GEO database for validation and found that the OS prognosis was negatively related to ACBD3 expression in GBMLGG (HR = 2.30, 95% CI: 1.19–4.43, p = 0.013). However, there was no statistically significant relationship between ACBD3 expression and prognosis in PAAD and ACC (Additional file 1: Fig. S3).

The gene alteration characteristics of ACBD3 in TCGA pan-cancer atlas were obtained. We found that among the many alteration types of ACBD3, “Amplification” was the most common type in patients with BRCA (9.22% alteration frequency) and LIHC (5.38% alteration frequency) (Fig. 8A). It is worth noting that, “Amplification” was the only type of genetic alteration in ovarian serous cystadenocarcinoma, pheochromocytoma, and paraganglioma (Fig. 8A). “Mutation” is the only type of genetic alteration in kidney chromophobe and adrenocortical carcinoma cases (Fig. 8A). “Deep Deletion” is the only type of genetic alteration in diffuse large B-cell lymphoma and prostate adenocarcinoma (Fig. 8A). Figure 8B displayed the 3D structure of ACBD3 protein. Figure 8C displayed the number, sites, and types of the ACBD3 genetic alteration. As shown in Fig. 8C, “Missense Mutation” was the most dominant type of mutation. Alteration in R484Q/ * and R516W in GOLD-2 could have led to the missense mutations in ACBD3.

In addition, we determined the relationship between ACBD3 alterations and clinical outcomes in BRCA and LIHC. For BRCA, patients with ACBD3 alterations had a worse prognosis in terms of progression-free survival (Fig. 8D) and disease-specific survival (Fig. 8E) than those patients without ACBD3 alterations. Patients with LIHC and ACBD3 alteration had a better prognosis in overall survival than those patients without ACBD3 alterations (Fig. 8F).

Furthermore, we accessed the relationship between DNA methylation and ACBD3 expression and found that promoter methylation was positively correlated with ACBD3 expression in PAAD (Pearson r = 0.2) (Fig. 9A). In addition, higher methylation β values of ACBD3-Body-N_Shelf-cg15084160 led to a worse OS prognosis in PAAD [HR = 1.52, CI(1.014;2.281), P < 0.05] (Fig. 9B and Table 1).

Figure 10A summarized the correlation between various cancers and ACBD3 phosphorylation sites. The S316 locus showed an increased phosphorylation level in BRCA when compared with normal tissues (Fig. 10B). Compared with normal tissues, the S20 locus exhibited a lower phosphorylation level in HNSC (Fig. 10C), with LUAD (Fig. 10D) exhibited the opposite trend. The phosphorylation levels at S43 were higher in KIRC (Fig. 10E), LUAD (Fig. 10D), and GBM (Fig. 10F) and decreased in HNSC (Fig. 10C).