3.1 Differential expression analysis
As shown in Fig. 1A, EFNA4 was overexpressed in most cancer types, including bladder urothelial carcinoma, breast invasive carcinoma, cholangiocarcinoma, colon adenocarcinoma, oesophageal carcinoma, glioblastoma multiforme, head and neck squamous cell carcinoma, kidney renal papillary cell carcinoma, liver hepatocellular carcinoma, lung adenocarcinoma, lung squamous cell carcinoma, prostate adenocarcinoma, rectum adenocarcinoma, stomach adenocarcinoma, thyroid carcinoma, uterine corpus endometrial carcinoma (p < 0.001 for all), cervical squamous cell carcinoma, endocervical adenocarcinoma (p < 0.01 for both) and kidney renal clear cell carcinoma (p < 0.05). However, EFNA4 expression was significantly downregulated in kidney chromophobe (p < 0.001) and pheochromocytoma and paraganglioma (p < 0.01). Subsequently, we compared the expression levels of EFNA4 between paired tumors and adjacent normal samples. EFNA4 expression was upregulated in bladder urothelial carcinoma, breast invasive carcinoma, colon adenocarcinoma, head and neck squamous cell carcinoma, liver hepatocellular carcinoma, lung adenocarcinoma, lung squamous cell carcinoma, prostate adenocarcinoma, stomach adenocarcinoma, thyroid carcinoma, uterine corpus endometrial carcinoma (p < 0.001 for all), cholangiocarcinoma, kidney renal clear cell carcinoma (p < 0.01 for both), kidney renal papillary cell carcinoma and rectum adenocarcinoma (p < 0.05 for both) and downregulated in kidney chromophobe (p < 0.001) (Fig. 1B).
3.2 Survival analysis
Log-ranked Kaplan-Meier analyses of low- and high-EFNA4-expression groups showed the prognostic potential of EFNA4 in various types of cancer (Fig. 2A). High EFNA4 expression was significantly correlated with poor OS in adrenocortical carcinoma (p = 0.003), glioblastoma multiforme (p = 0.024), GBMLGG (p < 0.001), kidney renal clear cell carcinoma (p = 0.002), low-grade glioma (p < 0.001), liver hepatocellular carcinoma (p < 0.001), mesothelioma (p = 0.006), sarcoma (p = 0.046), skin cutaneous melanoma (p = 0.016) and thymoma (p = 0.042).
As shown in Fig. 2B, upregulated EFNA4 expression was significantly correlated with poor DSS in adrenocortical carcinoma (p = 0.003), GBMLGG (p < 0.001), kidney renal clear cell carcinoma (p < 0.001), low-grade glioma (p < 0.001), liver hepatocellular carcinoma (p = 0.003) and mesothelioma (p < 0.001) and poor PFI in adrenocortical carcinoma (p < 0.001), GBMLGG (p < 0.001), kidney renal clear cell carcinoma (p < 0.001) and low-grade glioma (p < 0.001).
3.3 Genetic alterations
As demonstrated in Fig. 3A, we observed the highest alteration frequency of EFNA4 (> 13%) in cholangiocarcinoma, with the major gene alteration type being copy number amplification. Copy number amplification had the highest frequency of all alteration types of EFNA4 in all cancer types. The loci and types of genetic alterations of EFNA4 are depicted in Fig. 3B. Genetic alterations were primarily missense mutations. In addition, methylation significantly inhibited the mRNA expression of EFNA4 in GBMLGG.
3.4 Immune infiltration analysis and prediction of immunotherapeutic efficacy
The progress of tumors has been demonstrated to be significantly influenced by tumor microenvironment in numerous studies (21). In this study, the results of the Xcell, Immunome Characterization, Tumor Immune Dysfunction and Exclusion, and Microenvironment Cell Populations-counter algorithms revealed that the infiltration levels of cancer-associated fibroblasts in the tumor microenvironment of most cancers were positively correlated with EFNA4 expression (Figs. 4A-B). For instance, the infiltration levels of cancer-associated fibroblasts had a strong connection with EFNA4 expression in glioblastoma multiforme (p < 0.001, ρ = 0.224).
In GBMLGG, significantly positive correlations were observed between EFNA4 expression and the levels of infiltration of macrophages (R = 0.638), neutrophils (R = 0.542), eosinophils (R = 0.508), aDCs (R = 0.471), iDCs (R = 0.423), T cells (R = 0.351), cytotoxic cells (R = 0.334), CD56dim NK cells (R = 0.300), Th2 cells (R = 0.285), Th17 cells (R = 0.268), NK cells (R = 0.251) and T helper cells (R = 0.136) (p < 0.001), while negative correlations were found with the infiltration of pDCs (R = -0.326), Tcm (R = -0.311), TFH cells (R = -0.290), CD56bright NK cells (R = -0.242), Tgd cells (R = -0.212), regulatory T cells (R = -0.184), CD8 cells (R = -172) and mast cells (R = -0.161) (p < 0.001 for all) (Figs. 5A-B).
Furthermore, a significant positive correlation was found between the expression of EFNA4 and that of all 8 immunomodulatory genes in GBMLGG (Fig. 5C). To understand the possible impact of EFNA4 on the efficacy of ICB therapy, we assessed whether the expression of EFNA4 was related to TMB or MSI, and found that EFNA4 expression was positively correlated with TMB scores (Fig. 5D) but negatively correlated with MSI (Fig. 5E) in GBMLGG. These results suggest that EFNA4 can predict the efficacy of ICB therapy in GBMLGG.
3.5 Construction of a nomogram for predicting OS
Tumor tissues had higher expression of EFNA4 than normal tissues (p < 0.01) (Fig. 6A). The median value of EFNA4 expression (median = 2.694) was set as a cut-off value to separate patients with GBMLGG into low- and high-EFNA4-expression groups. The high-EFNA4-expression group had a significantly poorer OS than the low-EFNA4-expression group (p < 0.001). The ROC curve (AUC = 0.895) and time-dependent ROC curves demonstrated that EFNA4 expression had satisfactory efficacy when predicting the survival conditions of patients with GBMLGG in 1, 3, and 5 years (Figs. 6B-G).
EFNA4 expression was strongly correlated with the IDH status, 1p/19q codeletion frequency, primary therapy outcome, and histological type of GBMLGG (Figs. 7A-D). Baseline data were displayed in Table 1. Univariate Cox regression analysis revealed that EFNA4 expression (p < 0.0001), age (p < 0.0001), grade (p < 0.0001), and radiation therapy (p < 0.01) were important risk factors affecting the OS of patients with GBMLGG, whereas sex and race were not obvious risk factors (Fig. 7E). In multivariate Cox regression analysis, EFNA4 expression (p < 0.01), age (p < 0.01), and tumor grade (p < 0.01) were identified as independent risk factors (Fig. 7F). These results exhibited good efficacy of our prediction model based on EFNA4 expression.
A nomogram integrating risk factors we found in univariate and multivariate analyses (Fig. 7G), including EFNA4 expression, age, grade, and radiation therapy, was established to predict the OS of patients with GBMLGG in 1, 3, and 5 years. The nomogram (C index = 0.711) demonstrated moderate accuracy. The calibration curve demonstrated good conformity and accuracy of the predictive model (Fig. 7H).
Table 1
Baseline data sheet for patients with GBMLGG
Characteristics | Low expression of EFNA4 | High expression of EFNA4 | P value |
n | 349 | 350 | |
Primary therapy outcome, n (%) | | | 0.005 |
PD | 54 (11.6%) | 58 (12.5%) | |
SD | 96 (20.6%) | 52 (11.2%) | |
CR | 97 (20.9%) | 43 (9.2%) | |
PR | 41 (8.8%) | 24 (5.2%) | |
WHO grade, n (%) | | | < 0.001 |
G2 | 162 (25.4%) | 62 (9.7%) | |
G3 | 136 (21.4%) | 109 (17.1%) | |
G4 | 19 (3%) | 149 (23.4%) | |
IDH status, n (%) | | | < 0.001 |
WT | 46 (6.7%) | 200 (29%) | |
Mut | 301 (43.7%) | 142 (20.6%) | |
1p/19q codeletion, n (%) | | | < 0.001 |
Non-codel | 188 (27.2%) | 332 (48%) | |
Codel | 161 (23.3%) | 11 (1.6%) | |
Gender, n (%) | | | 0.853 |
Female | 150 (21.5%) | 148 (21.2%) | |
Male | 199 (28.5%) | 202 (28.9%) | |
Race, n (%) | | | 0.030 |
Asian | 9 (1.3%) | 4 (0.6%) | |
Black or African American | 10 (1.5%) | 23 (3.4%) | |
White | 322 (46.9%) | 318 (46.4%) | |
Age, n (%) | | | < 0.001 |
<= 60 | 304 (43.5%) | 252 (36.1%) | |
> 60 | 45 (6.4%) | 98 (14%) | |
Histological type, n (%) | | | < 0.001 |
Astrocytoma | 88 (12.6%) | 108 (15.5%) | |
Oligoastrocytoma | 82 (11.7%) | 53 (7.6%) | |
Oligodendroglioma | 160 (22.9%) | 40 (5.7%) | |
Glioblastoma | 19 (2.7%) | 149 (21.3%) | |
OS event, n (%) | | | < 0.001 |
Alive | 270 (38.6%) | 157 (22.5%) | |
Dead | 79 (11.3%) | 193 (27.6%) | |
DSS event, n (%) | | | < 0.001 |
No | 277 (40.9%) | 157 (23.2%) | |
Yes | 67 (9.9%) | 177 (26.1%) | |
PFI event, n (%) | | | < 0.001 |
No | 223 (31.9%) | 130 (18.6%) | |
Yes | 126 (18%) | 220 (31.5%) | |
3.6 Functional enrichment analysis and GSEA
To examine molecular mechanisms underlying the involvement of EFNA4 in the development of GBMLGG, co-expressed genes between EFNA4-binding proteins and EFNA4 were identified in the TCGA-GBMLGG dataset. Protein-protein interaction networks based on the top 50 and 10 EFNA4-binding proteins demonstrated a close relationship between EFNA4 and the EPHA and EPHB families, which was consistent with the results of a previous study (Figs. 8A-B)(22).
We performed GO and KEGG analyses on the top 50 genes identified in the protein-protein interaction network and those identified via single-gene differential analysis of EFNA4 in GBMLGG. GO analysis revealed an enrichment of EFNA4-binding proteins in biological processes related to peptidyl-tyrosine modification (GO:0018212), peptidyl-tyrosine phosphorylation (GO:0018108), positive regulation of kinase activity (GO:0033674), protein autophosphorylation (GO:0046777) and ephrin receptor signaling (GO:0048013) in GBMLGG (Fig. 8C). The proteins were enriched in cellular components such as neuron-to-neuron synapse (GO:0098984), postsynaptic specialization (GO:0099572), asymmetric synapse (GO:0032279), postsynaptic density (GO:0014069) and early endosome (GO:0005769) (Fig. 8D). The proteins were enriched in molecular functions such as transmembrane receptor protein tyrosine kinase activity (GO:0004714), transmembrane receptor protein kinase activity (GO:0019199), protein tyrosine kinase activity (GO:0004713), protein serine/threonine/tyrosine kinase activity (GO:0004712) and ephrin receptor activity (GO:0005003) (Fig. 8E). In addition, EFNA4-binding proteins were enriched in KEGG pathways including the MAPK signaling (hsa04010), Ras signaling (hsa04014), PI3K-Akt signaling (hsa04151), axon guidance (hsa04360), and Rap1 signaling (hsa04015) pathways (Fig. 8F).
Drug susceptibility analysis revealed that EFNA4 expression affected the IC50 values of tinib-type drugs, such as dasatinib, lapatinib, linsitinib, and sunitinib, which validated that the function of EFNA4 was associated with the tyrosine kinase signaling pathway in GBMLGG (Figs. 8G-J). EFNA4 expression was positively correlated with the IC50 scores of dasatinib, lapatinib, and sunitinib and negatively correlated with the IC50 score of linsitinib.
Genes identified through a single-gene differential analysis were demonstrated on a volcano plot, and co-expressed genes were demonstrated on heat maps (Figs. 9A-C). GO analysis demonstrated that the gene set via single-gene differential was strongly associated with immune-related biological processes related to EFNA4 (Fig. 9D). The gene set was enriched in cellular components such as immunoglobulin complex (GO:0019814) and immunoglobulin complex, circulating (GO:0042571), which were involved in the progression of GBMLGG (Fig. 9E). The gene set was enriched in molecular functions such as antigen binding (GO:0003823) and immunoglobulin receptor binding (GO:0034987) (Fig. 9F). KEGG analysis revealed that the gene set was enriched in neuroactive ligand-receptor interaction (hsa04080), cytokine-cytokine receptor interaction (hsa04060), and viral protein interaction with cytokine and cytokine receptor (hsa04061) (Fig. 9G).
For complementary validation of the biological significance of EFNA4 in GBMLGG, GSEA was performed to analyze genes identified via single-gene differential analysis. EFNA4 was found to act as a negative regulator of processes such as vesicle fusion, neurotransmitter delivery, and GABA signaling pathway (Fig. 9H) and a positive regulator of inflammation, interferon-γ response, and epithelial-mesenchymal transition in GBMLGG.