Integrated analysis of C3AR1 and CD163 associated with immune infiltration in intracranial aneurysms pathogenesis

Abstract

Background

To identify potential immune-related biomarkers, molecular mechanism, and therapeutic agents of intracranial aneurysms (IAs).

Methods

We identified the differentially expressed genes (DEGs) between IAs and control samples from GSE75436, GSE26969, GSE6551, and GSE13353 datasets. We used weighted gene co-expression network analysis (WGCNA) and protein–protein interaction (PPI) analysis to identify immune-related hub genes. We evaluated the expression of hub genes by using qRT-PCR analysis. Using miRNet, NetworkAnalyst, and DGIdb databases, we analyzed the regulatory networks and potential therapeutic agents targeting hub genes. Least absolute shrinkage and selection operator (LASSO) logistic regression was performed to identify optimal biomarkers among hub genes. The diagnostic value was validated by external GSE15629 dataset.

Results

We identified 227 DEGs and 22 differentially infiltrating immune cells between IAs and control samples from GSE75436, GSE26969, GSE6551, and GSE13353 datasets. We further identified 41 differentially expressed immune-related genes (DEIRGs), which were primarily enriched in the chemokine-mediated signaling pathway, myeloid leukocyte migration, endocytic vesicle membrane, chemokine receptor binding, chemokine activity, and viral protein interactions with cytokines and their receptors. Among 41 DEIRGs, 10 hub genes including C3AR1, CD163, CCL4, CXCL8, CCL3, TLR2, TYROBP, C1QB, FCGR3A, and FCGR1A were identified with good diagnostic values (AUC >0.7). Hsa-mir-27a-3p and transcription factors, including YY1 and GATA2, were identified the primary regulators of hub genes. 92 potential therapeutic agents targeting hub genes were predicted. C3AR1 and CD163 were finally identified as the best diagnostic biomarkers using LASSO logistic regression (AUC = 0.994). The diagnostic value of C3AR1 and CD163 was validated by the external GSE15629 dataset (AUC = 0.914).

Conclusions

This study revealed the importance of C3AR1 and CD163 in immune infiltration in IAs pathogenesis. Our finding provided a valuable reference for subsequent research on the potential targets for molecular mechanisms and intervention of IAs.

1. Introduction

The prevalence of intracranial aneurysms (IAs) in the general population is approximately 1%–2%, and the incidence of IAs tends to increase with the advancement of modern diagnostic imaging technologies [1,15]. Rupture of intracranial aneurysms is the leading cause of subarachnoid hemorrhage (SAH) [17] and occurs more frequently in younger patients compared with other common types of strokes [20]. Early detection and treatment of IAs require multidisciplinary collaboration, particularly in patients with a sudden onset of SAH. Current medical intervention including endovascular coiling and neurosurgical clipping mainly focuses on prevention of rupture or re-rupture of IAs [17]. Despite improvements in management strategies, the mortality rate of aneurysmal SAH may reach as high as 25%–50% [26]. Given the non-negligible complications associated with invasive surgical procedures and the high morbidity and mortality rates associated with aneurysmal SAH, it is imperative to investigate the underlying mechanism and biomarkers of IAs to develop effective nonsurgical intervention options.

With multiple risk factors, vessel structures at intracranial arterial branch points undergo functional and morphological changes characterized by the impairment of endothelial function and vascular structure, chronic inflammation, and abnormal vascular remodeling [40]. Emerging evidence suggests that immune cell infiltration and inflammation play a crucial role in the degradation of vascular walls and formation, development, and rupture of IAs [51]. The pathogenesis of IAs is associated with various components of the inflammatory immune response, including leukocytes, cytokines, complement, and other humoral mediators [40]. It has been demonstrated that macrophage-mediated immune response contributes to vascular wall degradation, IAs development, and rupture [18,24]. As a source of macrophages in the vasculature, circulating monocytes with dysregulated subsets have been linked to proinflammatory response and development of IAs [41]. Hence, modulation of the vascular inflammatory process may be beneficial for the prevention of IAs development. However, there are no clinically validated drugs against the vascular inflammatory response and IAs development available. Preliminary studies have revealed a correlation between alterations in gene expression profiling and rupture of IAs [19,25]. The precise relationship between immune infiltration and gene expression alterations in IAs pathogenesis and the underlying regulatory mechanisms remain unknown.

In this study, we investigated the interrelationship between immune infiltration and gene expression alterations in IAs pathogenesis. We identified 22 significantly differentially infiltrating immune cells and 41 differentially expressed immune-related genes (DEIRGs) from four gene expression datasets (GSE75436, GSE26969, GSE6551, and GSE13353). Among these DEIRGs, 10 key genes (C3AR1, CD163, CCL4, CXCL8, CCL3, TLR2, TYROBP, C1QB, FCGR3A, and FCGR1A) were identified as hub genes with good diagnostic values for IAs. The expression of hub genes were evaluated by qRT-PCR analysis based on clinical subjects. We further explored the miRNA and transcription factor (TF) regulatory networks and the potential therapeutic agents of these 10 hub genes. Subsequently, using least absolute shrinkage and selection operator (LASSO) logistic regression, CD163 and C3AR1 were further identified as the optimal immune-related biomarkers which were closely associated with IAs pathogenesis. The diagnostic performance of CD163 and C3AR1 was validated by using the validation dataset (GSE15629). Our findings provided novel information on immune-related biomarkers and potential therapeutic targets of IAs.

2. Materials and methods

2.1. Data processing

GSE75436, GSE26969, GSE6551, and GSE13353 datasets were obtained from the Gene Expression Omnibus database (Supplementary Table 1). These four datasets contained 60 samples, including 39 IAs samples and 21 control samples. GSE75436, GSE26969, GSE6551, and GSE13353 datasets originated from the GPL570 platform. The probe IDs in each dataset were converted to gene symbols using the corresponding platform annotation files (GPL570). The expression matrix was then merged using the common gene symbols. The four datasets were used for the analysis of differential expression and as training sets in machine learning. The external GSE15629 dataset was used to validate the identified biomarkers. The GSE15629 dataset included 14 IAs samples and 5 control samples based on the GPL6244 platform chip data. The batch effects were eliminated by using the “sva” R package. Before and after “sva” processing, the batch impact was evaluated using principal component analysis. The research workflow was depicted in Supplementary Fig. S1.

2.2. Identification of differentially expressed genes (DEGs)

DEGs were screened between IAs samples and control samples by using limma package with the criteria |log₂fold change (FC)| > 2 and adj. P < 0.05 [34]. The “ggplot 2” and “pheatmap” R packages were used to visualize DEGs between IAs and control samples.

2.3. Single-sample gene set enrichment analysis (ssGSEA) of immune cell infiltration in IAs

The R package Gene Set Variation Analysis was employed to determine the levels of immune cell infiltration in IAs and control samples by using ssGSEA [10]. The levels of immune cell infiltration were compared by using Wilcoxon test. A p-value of <0.05 was considered statistically significant.

2.4. Weighted gene co-expression network analysis (WGCNA) of DEGs

The coexpression network was constructed using the R package “WGCNA” [16]. Before construction of a sample clustering tree, each sample was screened for outliers, which was eliminated based on cutHeight. Then, the optimal soft-thresholding power β was selected based on the standard scale-free networks. Next, we constructed WGCNA networks and examined various gene modules using a dynamic tree-cutting algorithm. After obtaining distinct gene modules using WGCNA, we performed module–trait associations analysis to investigate the correlations between gene modules and differentially infiltrating immune cells. Consequently, modules with a high related coefficient were considered as the potential candidates corresponding to differentially infiltrating immune cells and were chosen for further analysis. We performed the intersections analysis of DEGs and genes in curatorial modules by using the “VennDiagram” R package. The data obtained were applied to subsequent analysis as DEIRGs. The “OmicCircos” R package was used to visualize the chromosomal locations and expression patterns of DEIRGs [13].

2.5. Functional enrichment analysis of DEIRGs

The clusterProfiler package was employed to investigate the functions of DEIRGs through Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis [46]. GO analysis was used to determine the biological functions, molecular functions, and cellular components. An adj. p of <0.05 was considered statistically significant.

2.6. Protein–protein interaction (PPI) network construction and hub genes identification

The PPI interaction network was constructed using the String online tool (Search Tools for the Retrieval of Interacting Genes [STRING]) database [39]. Cytoscape's CytoHubba plug-in was used to analyze the hub genes in PPI network. According to the degree of nodes, the top 10 genes were regarded as hub genes. The GOSemSim package was used to measure semantic similarities among GO terms, gene clusters, and gene products [45]. The hub genes were evaluated using the geometric mean of these semantic similarities. The Corrplot package was used to analyze the association between hub genes.

2.7. Expression patterns and receiver operating characteristic (ROC) curve analysis of hub genes

The expression patterns of hub genes were compared by using boxplot and Wilcoxon test. A p-value of <0.05 was considered statistically significant. To evaluate the role of hub genes in the diagnosis of IAs, the “pROC” R package was used to analyze ROC curves. Genes with an area under the curve (AUC) > 0.70 were considered useful for the diagnosis of IAs [21].

2.8. qRT-PCR analysis of hub genes based on clinical samples

Peripheral blood from 8 patients with IAs and 8 normal controls was collected for qRT-PCR analysis. Total RNA was extracted by using Takara RNAiso Plus (9108) Trizol reagent. We performed reverse transcription analysis by using the Takara PrimeScript RT Master Mix (RR036A). The qRT-PCR analysis was then performed by using SYBR Green Premix (RR420A). The primer sets used were: forward 5′-ATGACTCTACCAGATGCCTCCCT-3′, reverse 5′-TGTGACATTCCGACACCGAGA-3′ for TLR2; forward 5′-AACTGCTCTAGGTGCTTCATTATGT-3′, reverse 5′-CCTCCATTTACCAGGCGAAGT-3′ for CD163; forward 5′-CTGGCTGTAAGTGGTCTCCGT-3′, reverse 5′-GCAATGAGCACTGTCAGCACC-3′ for TYROBP; forward 5′-CCAGGATAGTCTAAGGGAGGTGTTC-3′, reverse 5′-CCCACCGCAATCCTCAGTC-3′ for FCGR3A; forward 5′-CTGCCCTTGCTGTCCTCCT-3′, reverse 5′-CCGGCTTCGCTTGGTTAG-3′ for CCL3; forward 5′-GACCAGACCATCCGCTTCG-3′, reverse 5′-CGCATGAGGTTCACGCACA-3′ for C1QB; forward 5′-CCTCGCAACTTTGTGGTAGATT-3′, reverse 5′-GCTCAGTTCAGTTCCAGGTCATAC-3′ for CCL4; forward 5′-GAGGCCAAGGGCCAAGAG-3′, reverse 5′-CAAGGCACAGTGGAACAAGGA-3′ for CXCL8; forward 5′-CATTGCTCTAGCATCTGCCAATA-3′, reverse 5′-TTGAGAACCGCTGGATTGATT-3′ for C3AR1; forward 5′-CCAGCCTTCAAGAAGACAGACAT-3′, reverse 5′-TTGTACGCAGTGCTCACGG-3′ for FCGR1A. Results were calculated using the 2^−ΔΔCt method after normalization to the expression of actin housekeeping genes. We obtained all appropriate patient consent forms. Ethical approval for this study was obtained from the Ethics Committee of the First Affiliated Hospital of Shandong First Medical University & Shandong Provincial Qianfoshan Hospital.

2.9. Gene set enrichment analysis (GSEA) and correlation analysis between hub genes and infiltrating immune cells

The R package clusterprofile was used to perform GSEA. The chosen reference gene set was obtained from the Molecular Signature Database (MSigDB). An adj. p < 0.05 was considered statistically significant.

We used the “psych” package to calculate the Spearman correlation coefficient between the expression of hub genes and the differentially infiltrating immune cells [47]. The “ggpubr” package was used to illustrate their relationship [2].

2.10. Construction of hub gene-miRNA network and hub gene-TF network and prediction of potential therapeutic agents targeting hub genes

We used the miRNet database (https://www.mirnet.ca/) to predict the miRNAs targeting hub genes. The results were visualized using Cytoscape. In the network, hub genes and miRNA were represented by a red circle and a light orange triangle, respectively.

We used the NetworkAnalyst database (https://www.networkanalyst.ca/) to predict TFs targeting hub genes. The results were imported into the visualization software Cytoscape. In the network, hub genes and TFs were represented by a red circle node and blue diamond node, respectively.

We performed the drug-hub gene interactions analysis to identify candidate drugs for IAs by using the Drug–Gene Interaction Database (DGIdb) [6].

2.11. Identification and validation of the optimal immune-related biomarkers of IAs

LASSO logistic regression based on “glmnet” R package was used to identify the optimal biomarkers among screened hub genes. In the LASSO model, the diagnostic performance of biomarkers was evaluated using ROC curves, decision curve analysis (DCA), and P-R curves.

To perform validation analysis, we enrolled an external dataset (GSE15629). The diagnostic performance of the identified immune-related biomarkers was varified by using ROC curves.

2.12. Statistical analysis

The difference in the levels of immune cell infiltration was analyzed by the Wilcoxon rank-sum test with the threshold of P < 0.05. Spearman correlation coefficients were calculated to determine the correlation between hub genes and immune cell infiltration. The statistical analysis was carried out with R software (version v4.1.0). Student's t-test was used for measurement of qRT-PCR data which were presented as mean ± SD. A P value < 0.05 was considered statistically significant.

3. Results

3.1. Data preprocessing and normalization

There were 39 IAs samples and 21 control samples in the GSE75436, GSE26969, GSE6551, and GSE13353 datasets. After combining the datasets, the batch differences in the gene expression matrix were eliminated. Two expression boxplots demonstrated the elimination of batch differences (Supplementary Figs. S2A–B). Results of the principal component analysis demonstrated that the batch effects in various datasets were eliminated (Supplementary Figs. S2C–D).

3.2. Immune infiltrating cell analysis of IAs

Heatmap showed the enrichment fraction of 28 immune infiltrating cells in IAs and control samples (Fig. 1A). There were significant differences between IAs and control samples (p < 0.05) in the levels of 22 types of immune cells, including activated CD8 T cells, central memory CD8 T cells, type 1 T helper cells, effector memory CD8 T cells, activated CD4 T cells, effector memory CD4 T cells, T follicular helper cells, gamma delta T cells, regulatory T cells, type 2 T helper cells, type 17 T helper cells, immature B cells, natural killer cells, CD56bright natural killer cells, CD56dim natural killer cells, myeloid-derived suppressor cells, immature dendritic cells, natural killer T cells, plasmacytoid dendritic cells, and macrophage- and monocyte-activated dendritic cells (Fig. 1B).

Fig. 1.

Immune infiltrating cell analysis and identification of the key module using WGCNA analysis. (A) A heatmap revealed the enrichment fraction of 28 immune infiltrating cells. (B) A ssGSEA analysis revealed 22 significantly differentially infiltrated immune cells in IAs. (C) Scale-free network showing the scale-free fit index and mean connectivity. (D) Clustering dendrogram of WGCNA. (E) Visualization of the module-trait relationships and gene significances of each module. Among the identified 23 modules, the blue module including 2500 genes was the most relevant module associated with differentially infiltrating immune cells, particularly myeloid-derived suppressor cells. WGCNA, weighted gene co-expression network; ssGSEA, single sample gene set enrichment analysis; IAs, intracranial aneurysms.

3.3. Identification of the key module using WGCNA

WGCNA was used to identify the key gene module. By considering the soft-thresholding power as 18 (scale-free R2 = 0.90), we eventually identified 23 modules (Fig. 1C and D). The blue module was is most closely related to differentially infiltrating immune cells, particularly myeloid-derived suppressor cells (correlation coefficient = 0.94, P = 7e-29; Fig. 1E). The blue module comprised 2500 genes.

3.4. Identification of consensus DEGs/DEIRGs and functional enrichment analysis

We screened 227 DEGs, which included 84 upregulated and 143 downregulated genes (IAs vs. control). We used the volcano plots to visualize the differential gene expression between IAs and control samples (Fig. 2A). The heatmap revealed the expression patterns of DEGs in IAs and control samples (Fig. 2B).

Fig. 2.

Visualization of DEGs and identification, functional enrichment analysis, and chromosomal positions analysis of DEIRGs. (A) Volcano plots of the distributions of DEGs. The green dots represented 143 downregulated genes, and the red dots represented 84 upregulated genes. (B) A heatmap of 227 DEGs. (C) Identification of 41 DEIRGs by intersection of 227 DEGs and 2500 genes in the blue module. (D) GO analysis of DEIRGs. DEIRGs were primarily enriched in the chemokine-mediated signaling pathway, myeloid leukocyte migration, endocytic vesicle membrane, specific granule, chemokine activity, and chemokine receptor binding. (E) KEGG enrichment analysis of DEIRGs. DEIRGs were significantly involved in viral protein interactions with cytokines and their receptors, rheumatoid arthritis, chemokine signaling pathway, and TLR signaling pathway. (F) Circos plot showing the chromosomal mapping of 41 DEIRGs. The outer circle depicted chromosomes, and each one labeled with DEIRGs. DEGs, differentially expressed genes; DEIRGs, differentially expressed immune-related genes; GO, gene ontology; KEGG, kyoto encyclopedia of genes and genomes.

After obtaining the key module and DEGs, we further identified 41 DEIRGs after intersections analysis between 227 DEGs and 2500 genes in blue modules (Fig. 2C). As depicted in Fig. 2D, GO analysis revealed that DEIRGs were primarily enriched in the chemokine-mediated signaling pathway, myeloid leukocyte migration, endocytic vesicle membrane, specific granule, chemokine activity, and chemokine receptor binding. As presented in Fig. 2E, KEGG enrichment analysis revealed that these DEIRGs were involved in viral protein interactions with cytokines and their receptors, rheumatoid arthritis, chemokine signaling pathway, and TLR signaling pathway. These DEIRGs were distributed on chromosomes 1, 2, 3, 4, 5, 6, 7, 8, 11, 12, 14, 15, 17, 19, and 22 (Fig. 2F).

3.5. Construction of PPI network and identification of hub genes

Fig. 3A depicts the interactions between DEIRGs. We identified 10 hub genes (C3AR1, CD163, CCL4, CXCL8, CCL3, TLR2, TYROBP, C1QB, FCGR3A, and FCGR1A) by using CytoHubba plug-in Cytoscape (Fig. 3B). We analyzed the functional similarities of hub genes and found that FCGR1A (similarity score >0.5) had the greatest functional similarity (Fig. 3C). The correlation results demonstrated that the 10 hub genes were positively correlated with each other, and CCL3 and CCL4 had the strongest correlation (r = 0.94) (Fig. 3D).

Fig. 3.

PPI network construction and identification, expression evaluation, and diagnostic value evaluation of hub genes. (A) PPI network results of DEIRGs by STRING. (B) 10 hub genes (CD163, C3AR1, CCL4, CXCL8, CCL3, TLR2, TYROBP, C1QB, FCGR3A, and FCGR1A) were identified. (C) Functional similarity analysis showed that FCGR1A had the highest functional similarity score. (D) The correlations among the 10 hub genes were all positive. CCL3 and CCL4 had the strongest correlation. (E) Comparison of expression of hub genes between IAs and control samples. All 10 hub genes were upregulated in IAs samples. (F) ROC curve demonstrating the diagnostic values of hub genes for IAs. All 10 hub genes showed good diagnostic value (all AUC values > 0.7). PPI, protein–protein interaction; ROC, receiver operating characteristic; AUC, area under curve.

In addition, we evaluated the expression patterns and diagnostic value of hub genes. All of the hub genes were more highly expressed in IAs samples than in control samples (Fig. 3E). As illustrated in Fig. 3F, we performed ROC curve analysis to evaluate the specificity and sensitivity of hub genes for IAs diagnosis. The AUC values of TLR2, CD163, TYROBP, FCGR3A, CCL3, C1QB, CCL4, CXCL8, C3AR1, and FCGR1A were 0.8901, 0.7643, 0.8877, 0.8730, 0.8828, 0.8315, 0.9084, 0.7741, 0.9756, and 0.8095, respectively. The AUC values suggested that all these hub genes may have potential utility as diagnostic indicators in IAs.

3.6. Experimental validation of hub genes expression

Based on clinical samples, we performed qRT-PCR analysis to validate the expression levels of hub genes. The expression levels of C3AR1, CD163, CCL4, FCGR1A, FCGR3A, and TYROBP were upregulated in the blood of IAs patients (Fig. 4; all P < 0.05). However, there were no significant differences between the two groups in terms of C1QB, CCL3, CXCL8, and TLR2 expression.

Fig. 4.

Expression analysis of hub genes by qRT-PCR. The expression levels of CD163, C3AR1, CCL4, FCGR1A, FCGR3A, and TYROBP were upregulated in the peripheral blood of IAs patients. *P-value <0.05, **P-value <0.01, and ***P-value <0.001.

3.7. GSEA analysis of hub genes and their correlation with immune cells

To reveal the biological functions of 10 hub genes, we performed GSEA analysis (Supplementary Figs. S3A–J). Genes in cohorts with high expression of TLR2 (Supplementary Fig. S3A), CD163 (Supplementary Fig. S3B), TYROBP (Supplementary Fig. S3C), C1QB (Supplementary Fig. S3F), and C3AR1 (Supplementary Fig. S3I) were highly enriched in allograft rejection, antigen processing, and presentation. The cohorts with high expression of FCGR3A (Supplementary Fig. S3D) revealed gene enrichment in B cell receptor signaling pathway and chemokine signaling pathway. Leishmaniasis genes were enriched in cohorts with high expression of CCL3 (Supplementary Fig. S3E). Gene enrichment in the amebiasis chemokine signaling pathway was observed in cohorts with high expression of CCL4 (Supplementary Fig. S3G). Gene enrichment in Herpes simplex virus 1 was identified in cohorts with high expression of CXCL8 (Supplementary Fig. S3H). In cohorts with high expression of FCGR1A (Supplementary Fig. S3J), genes were enriched in antigen processing and presentation and chemokine signaling pathways. These findings suggested that hub genes were strongly associated with cell immunity and inflammation.

To determine whether the expression levels of hub genes were correlated with immune cell infiltration in IAs, we calculated the correlation coefficients between the expression levels of hub genes and the infiltration levels of immune cells. Results showed that the gene expression levels were significantly positively associated with the infiltrating levels of 22 immune cells (Fig. 5A). Of these, the hub gene TYROBP was significantly positively associated with macrophages (r = 0.929) and regulatory T cells (r = 0.928) (Fig. 5B).

Fig. 5.

Correlation analysis between hub genes and immune cells in IAs. (A) The results showed positive correlation among 10 hub genes and 22 immune cells. (B) TYROBP was significantly positively correlated with macrophages (r = 0.929) and regulatory T cells (r = 0.928).

3.8. Construction of target gene-miRNA network and target gene-TF network

We identified hub gene-miRNA pairs through network analysis of 10 hub genes by using miRNet database. As shown in Fig. 6A, the interaction network comprised 10 hub genes and 147 miRNAs. CXCL8 was notably regulated by 92 miRNAs, including has-let-7a-5hashsa-let-7b-5p. TLR2 was regulated by 15 miRNAs (e.g., hsa-mir-19a-3p and hsa-mir-19 b-3p). CCL4 was regulated by 12 miRNAs (e.g., hsa-mir-24–3p and hsa-mir-195–5p), and C3AR1 was regulated by nine miRNAs (e.g., hsa-mir-107 and hsa-mir-103a-3p). CCL3 was regulated by eight miRNAs (e.g., hsa-mir-24–3p and hsa-mir-223–3p). C1QB was regulated by four miRNAs (e.g., hsa-mir-26 b-5p and hsa-mir-124–3p). TYROBP was regulated by three miRNAs (e.g., hsa-mir-146a-5p and hsa-mir-20a-5p), and CD163 was regulated by two miRNAs (hsa-mir-27a-3p and hsa-mir-373–3p). FCGR1A was regulated by miRNA hsa-mir-27a-3p. The miRNA hsa-mir-449a regulated FCGR3A. In the target gene-miRNA network, hsa-mir-27a-3p was involved in the regulation of six hub genes (C3AR1, CD163, CCL4, CXCL8, FCGRA, and TYROBP), which suggested an important regulatory role of hsa-mir-27a-3p in IAs pathogenesis.

Fig. 6.

miRNAs and TFs regulatory network of hub genes and drug–hub gene interaction diagram. (A) Hub gene-miRNA regulatory network consisting of 10 hub genes and 147 miRNAs. The top three hub genes for miRNAs were CXCL8 (regulated by 92 miRNAs), TLR2 (regulated by 15 miRNAs), and CCL4 (regulated by 12 miRNAs). Hsa-mir-27a-3p regulated the highest number of hub genes (CD163, C3AR1, CCL4, CXCL8, FCGR1A, and TYROBP) in this network. (B) Hub gene-TF regulatory network consisting of 10 hub genes and 70 TFs. YYI and GATA2 were involved in regulating the highest number of hub genes. Both YY1 and GATA2 were found to regulate C3AR1, CD163, CCL4, CXCL8, and FCGR3A. (C) Orange ovals indicated related hub genes, and rectangle indicated the potential therapeutic agents. A total of 92 potential therapeutic agents of IAs were identified. Among them, 56 agents interacted with CXCL8, which had the biggest number of agents. No potential therapeutic agents targeting CD163, TYROBP, or C1QB were identified in the DGIdb database. DGIdb, Drug–Gene Interaction Database.

We constructed hub gene-TF regulatory network by using NetworkAnalyst database. The interaction network comprised 10 hub genes and 70 TFs (Fig. 6B). CD163 was regulated by 11 TFs (e.g., FOXI1 and NR2F1). A total of 11 TFs (e.g., SRY and FOS) regulated FCGR3A; 8 TFs (e.g., BRCA1 and FOXC1) regulated CCL4; 7 TFs (e.g., JUN and FOXC1) regulated TLR2; FCGR1A was regulated by 7 TFs (e.g., SPIB and FOS); 6 TFs (e.g., NR2F1 and ZNF354C) regulated TYROBP; C1QB was regulated by 6 TFs (e.g., NKX2-5 and GATA2); 6 TFs (e.g., JUND and YY1) regulated CCL3; 5 TFs (e.g., CREB1 and GATA2) regulated CXCL8, and 3 TFs (e.g., FOXC1 and GATA2) regulated C3AR1. Among these TFs, YY1 was involved in regulation of seven hub genes, including C3AR1, CD163, CCL4, CXCL8, FCGR3A, FCGR1A, and CCL3. Meanwhile, GATA2 regulated seven hub genes, including C3AR1, CD163, CCL4, CXCL8, FCGR3A, C1QB, and TLR2. The finding indicated that YYI and GATA2 appeared to be promising candidate regulators of IAs pathogenesis.

3.9. Identification of candidate therapeutic agents

We identified 92 potential agents for IAs treatment by using DGIdb (Supplementary Table 2). The drug–gene network was visualized using Cytoscape (Fig. 6C). The drug chembl 389348 was found to interact with C3AR1. Infliximab and nagrestipen interacted with CCL3, whereas clodronic acid and epoetin alfa interacted with CCL4. Three drugs (e.g., MDX-447 and MDX-210) interacted with FCGR1A, and four drugs (e.g., tomaralimab and DIAPEP-277) interacted with TLR2. 24 drugs (e.g., etanercept and cyclosporine) interacted with FCGR3A, and 56 drugs (e.g., ABX-IL8 and HUMAX-IL8) interacted with CXCL8. No agents targeting CD163, TYROBP, or C1QB were identified in this database.

3.10. Identification of the optimal immune-related biomarkers of IAs using LASSO logistic regression

We established a diagnostic model using the LASSO algorithm to further demonstrate the reliability of the diagnostic value of hub genes, followed by the identification of C3AR1 and CD163 (Fig. 7A and B). Then, we constructed a LASSO diagnostic model based on C3AR1 and CD163, and the ROC curve indicated that the LASSO model was highly accurate (AUC = 0.994; Fig. 7C). DCA showed that area under the LASSO logistic curve was greater than that under the gray line, suggesting that the diagnostic model of IAs performed well (Fig. 7D). The area under the PR curve (AUC-PR = 0.69) showed that this model performed well (Fig. 7E).

Fig. 7.

Establishment of a LASSO diagnostic model and validation analysis by external dataset. (A) LASSO coefficient profiles of 10 hub genes. (B) LASSO model. (C) ROC curve analysis. ROC curve analysis indicated that the LASSO model based on CD163 and C3AR1 had high accuracy (AUC value = 0.994). (D) DCA analysis showed that the diagnostic model of IAs performed well. (E) PR curve. PR curve indicated good diagnostic performance of the LASSO model. (F) ROC curve analysis of GSE15629 dataset verified the diagnostic performance of CD163 and C3AR1 (AUC = 0.914). LASSO, least absolute shrinkage and selection operator; DCA, decision curve analysis.

To validate the optimal immune-related biomarkers and achieve more reliable results, we used an external dataset GSE15629 to verify the diagnostic performance of C3AR1 and CD163. The ROC curve confirmed the high diagnostic accuracy of C3AR1 and CD163 (AUC = 0.914, Fig. 7F).

4. Discussion

Despite extensive studies on IAs pathogenesis, the detailed mechanism remains largely undetermined. The role of immune infiltration in IAs has received increased attention. Inflammatory immune response contributes to the abnormal vascular remodeling and development of IAs [15]. Immune activation was reported to be associated with Hunt–Hess grade and clinical outcomes of patients with aneurysmal SAH [23]. Animal studies showed that macrophages participated in IAs formation and macrophage depletion decreased the incidence of IAs in mice [15,27]. Clinical data showed that dysregulated M1/M2 polarization appeared to be involved in driving inflammatory responses in IAs [37]. Our findings revealed statistically significant differences in 22 types of infiltrating immune cells between IAs and control samples. To comprehensively investigate the interrelationship between genetic alterations and immune infiltration in IAs pathogenesis, we used four open public datasets to identify robust DEGs. Based on these DEGs, we identified 10 immune-related hub genes (C3AR1, CD163, CCL4, CXCL8, CCL3, TLR2, TYROBP, C1QB, FCGR3A, and FCGR1A) by constructing PPI networks. To evaluate and confirm the clinical utility of 10 hub genes, we investigated the diagnosis specificity and sensitivity of hub genes for IAs by using ROC curve analysis. All the AUC values for these genes were >0.7, indicating excellent diagnostic value. Further analysis of correlations revealed that expression levels of hub genes were significantly positively correlated with the infiltrating levels of 22 immune cells. Among the 10 hub genes, TYROBP was strongly positively associated with macrophages (R = 0.929) and regulatory T cells (R = 0.928), with the two highest R values of correlations. TYROBP, a 12-kDa DNAX activating protein, is expressed by natural killer cells and myeloid cells [7]. However, there are few studies on TYROBP in IAs. In the current study, qRT-PCR analysis showed a significant upregulation of expression level of TYROBP in the blood samples of IAs patients. Irene et al. reported similar findings in human abdominal aortic aneurysm (AAA) [11]. Besides TYROBP, the expression of CCL4, FCGR1A, FCGR3A, C3AR1, and CD163 in our study was also upregulated in IAs. Available data showed that CCL4 was involved in macrophage accumulation in the aneurysm wall of AAA [22]. There was lack of experimental research to reveal the function of FCGR1A and FCGR3A in IAs.

To investigate the candidate regulatory mechanism of 10 immune-related hub genes in IAs, we constructed the target gene-miRNA network and target gene-TF network. Our findings indicated that the potential regulatory network of 10 hub genes comprised 147 miRNAs. The hsa-mir-27a-3p was identified as the most promising miRNA, displaying close interactions with C3AR1, CD163, CCL4, CXCL8, FCGRA, and TYROBP in the miRNA-hub gene regulatory network. At present, there are still no studies focusing on the potential role of hsa-mir-27a-3p in IAs pathogenesis. Hsa-mir-27a-3p delivered by extracellular vesicles from glioblastoma cells may promote M2 macrophage polarization [50]. The level of hsa-miR-27a-3p was downregulated in cerebrospinal fluid of patients with Alzheimer's disease [35]. Additionally, miR-27a-3p may contribute to atherosclerosis by inducing vascular calcification [5]. Analysis of the TF-hub gene network revealed that the potential regulatory network of 10 hub genes comprised 70 TFs. Among them, YY1 and GATA2 showed the strongest interactions with hub genes. According to the available data, YY1, as an ubiquitously expressed GLI-Krüppel zinc finger-containing TF, participated in vascular development by modulating sprouting angiogenesis [49]. Following balloon injury, YY1 inhibited the proliferation and migration of vascular smooth muscle cells, as well as intimal hyperplasia [36]. To the best of our knowledge, except our preliminary study, no other studies have revealed the role of YY1 in IAs pathogenesis. GATA2 was a well-known zinc finger-containing TF involved in regulation of hematopoiesis. GATA2 was associated with complex congenital immunodeficiency, infection, and inherited and acquired hematologic disorders [12]. GATA2-deficient endothelial cells showed impaired NO production and angiogenesis [31]. According to a study on graft arteriosclerosis, GATA2 played a crucial role in promoting endothelial activation and vascular inflammatory response [32]. Further studies are required to unveil the regulatory mechanisms of YY1 and GATA2 in IAs pathogenesis.

Given the non-negligible difficulty and complications associated with invasive surgical procedures, we attempted to identify potential therapeutic agents targeting the 10 hub genes by using the DGIdb database. A total of 92 agents were screened as potential candidates for noninvasive treatment of IAs. Among them, 15 candidate therapeutic agents have been preliminarily reported in IAs via different mechanisms. As an antiplatelet and nonsteroidal anti-inflammatory drug, aspirin is commonly used to prevent and treat cardiovascular and cerebrovascular diseases [14]. In mice lacking microsomal prostaglandin E2 synthase type 1, low doses of aspirin (6 mg/kg/d) decreased the mortality rate and IAs rupture [29]. A prospective cohort study indicated that oral administration of aspirin reduced the risk of IAs development in patients with IAs of <7 mm [43]. Etanercept, a TNF-α inhibitor, decreased rat-induced vascular inflammation and IAs formation [44]. Pretreatment with the synthetic TNF-α inhibitor 3,6′-dithiothalidomide significantly decreased the incidence of IAs formation and rupture [38]. Future studies are warranted to validate the safety and practical application value of these screened agents in IAs treatment.

To construct a robust explanatory model with the lowest number of genes, we performed a LASSO regression analysis based on the 10 immune-related hub genes. C3AR1 and CD163 were identified as the optimal immune-related biomarkers with high AUC value (0.994). We further validated the biomarkers by using validation dataset (GSE15629). ROC curve analysis of GSE15629 dataset verified the diagnostic performance of C3AR1 and CD163. The current study confirmed the significant upregulation of C3AR1 and CD163 in the blood of patient with IAs by qRT-PCR analysis. CD163 was identified as a phenotype switcher of macrophage polarization to the anti-inflammatory M2 subtype [4]. Human data revealed a correlation between arterial CD163⁺ macrophage infiltration and remodeling of saccular IAs [9,28]. A decreased CD163⁺ macrophage accumulation was detrimental to arterial tissue repair and maintenance of aneurysm integrity following initial blood blister-like IAs rupture [42]. C3AR1 was responsible for the aggravation of white matter injury caused by abnormal microglial activation [48]. Endothelial C3AR mediated the vascular inflammation and dysfunction of blood–brain barrier [30]. C3AR knockout was reported inhibiting the development of thoracic aortic dissection in mice [33]. Despite previous reports of C3AR1 and CD163 related to IAs pathogenesis [3,8], there still existed some shortcomings in the two studies. These preliminary findings were only based on bioinformatic analysis without further verification by wet experiments, such as qRT-PCR analysis. Besides, neither of these studies had a verification analysis of the diagnostic performance of C3AR1 and CD163 by establishing a LASSO logistic regression.

Several limitations of this study deserve mention. First, the validation analysis of immune-related biomarkers by qRT-PCR is based on a small sample size and a larger study is required to generalize the expression trend of hub genes. Second, there is a lack of research on the various stages of IAs, which could provide more valuable evidence for predicting IAs development and rupture. Thirdly, further researches are needed to elucidate the clinical application value of biomarkers in IAs, and to determine the generalizability of our findings.

Author contribution statement

Shengjie Li: Conceived and designed the experiments; Performed the experiments; Contributed reagents, materials, analysis tools or data; Wrote the paper.

Jinting Xiao: Conceived and designed the experiments; Performed the experiments; Analyzed and interpreted the data; Contributed reagents, materials, analysis tools or data; Wrote the paper.

Zaiyang Yu, Junliang Li: Performed the experiments; Analyzed and interpreted the data.

Hao Shang, Lei Zhang: Analyzed and interpreted the data.

Funding statement

This study was supported by National Natural Science Foundation of China (Grant No. 82001318), Shandong Provincial Natural Science Foundation (Grant No. ZR2020QH119), Science and Technology Support Plan for Youth Innovation Teams of Colleges and Universities of Shandong Province of China (2021KJ095), Special Funding for Qilu Sanitation and Health Outstanding Young Talent Cultivation Project to Shengjie Li, and Academic Promotion Programme of Shandong First Medical University (2019LJ005).

Data availability statement

Data will be made available on request.

Declaration of interest's statement

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article.

figs1

Study workflow. IAs, intracranial aneurysms; ssGSEA, single sample gene set enrichment analysis; DEGs, differentially expressed genes; WGCNA, weighted gene co-expression network; DEIRGs, differentially expressed immune-related genes; GO, gene ontology; KEGG, kyoto encyclopedia of genes and genomes; GSEA, gene set enrichment analysis; TFs, transcription factors; ROC, receiver operating characteristic; LASSO, least absolute shrinkage and selection operator; qRT-PCR, quantitative real-time polymerase chain reaction; DCA, decision curve analysis.

figs2

Box plots of 60 samples from GSE13353, GSE26969, GSE6551, and GSE75436 datasets. (A) Before batch effect correction. (B) After applying batch effect correction. (C) PCA analysis of four datasets before batch effect correction. (D) PCA analysis of four datasets after batch effect correction. The batch effects of four datasets were basically eliminated after the correction. PCA, principal components analysis.

figs3

Analysis of biological functions of hub genes via GSEA. (A) TLR2. (B) CD163. (C) TYROBP. (D) FCGR3A. (E) CCL3. (F) C1QB. (G) CCL4. (H) CXCL8. (I) C3AR1. (J) FCGR1A. GSEA showed that the identified 10 hub genes were strongly associated with cell immunity and inflammation. Among them, genes in cohorts with high expression of CD163, C3AR1, TLR2, C1QB, and TYROBP were highly enriched during allograft rejection, antigen processing, and presentation.