A cohort of familial and sporadic IA cases was recruited from the Department of Neurosurgery and Neurology in Xiangya Hospital and Hunan People's Hospital from January 2016 to December 2019. Both familial and sporadic cases were diagnosed by magnetic resonance angiography, computed tomography angiography, or digital subtraction angiography. Angiography image results were interpreted by at least a radiologist and neurosurgeon independently. Cases of disagreement were resolved by consensus or interpretation by a third physician. Several patients with SAH were confirmed by intracranial surgery performed as an emergency operation but with suspicious computed tomography angiography imaging. Patients with IA diagnosed with other vascular diseases such as cerebral arteriovenous malformation, moyamoya disease, or several hereditary diseases including hereditary hemorrhagic telangiectasia, Marfan's syndrome, or Sturge-Weber syndrome, were excluded. Individual demographic information and lifestyle data were collected through an interview conducted by the clinical reception staff. Clinical information was collected by consulting the medical record system. Peripheral blood samples were collected from all enrolled individuals. For familial IA, each family has two or more patients with IA in the first- to third-degree relatives. Three families with nine confirmed IA cases were selected for WES (Figure 1). A total of 384 sporadic IA cases and 384 controls was used for further association studies of the selected candidate variants. Control individuals were recruited from the community health service center in the same geographic region of Hunan Province, China, during the same period. The controls had undergone an annual health check-up and structured interview and showed no medical or family history of IA, SAH, or any other known cerebrovascular disease (13). This study was approved by the Medical Ethics Committee of Clinical Pharmacology Institute, Central South University, China (CTXY-150002-1), and written informed consent was obtained from all participants.

Genomic DNA was extracted from the peripheral blood of patients with IA and control individuals using TIANamp Blood DNA Extraction Kits (TIANGEN Biotech Co., Ltd., Beijing, China). WES was performed by Novel Bioinformatics Co., Ltd. (Shanghai, China). The exome was captured with an Agilent Sureselect All Exon V6 Kit (Agilent Technologies, Santa Clara, CA, USA) and sequenced on a HiSeq X Ten platform (Illumina, San Diego, CA, USA). Sequence mapping and variant detection were performed using the Burrows-Wheeler Aligner tool and Genome Analysis Toolkit software, respectively.

To prioritize the variants, a series of filters was used. Variants were given higher priority if they were (1) predicted to affect the protein-coding sequences including non-synonymous single-nucleotide variations such as missense, nonsense, or splice site variations, and insertions or deletions in the consensus coding sequencing region; (2) damaging or unknown as predicted by protein prediction programs [Sorting Intolerant From Tolerant (SIFT) and Polymorphism Phenotyping V2 (PolyPhen-2)]; (3) less common in reference databases (MAF < 0.05 in East Asian population according to the 1,000 genome database (http://www.1000genomes.org/) and gnomAD (https://gnomad.broadinstitute.org/); (4) high-quality variants (sequence read depth was more than 8x); (5) shared by all affected individuals in more than one family.

To confirm the variants discovered by WES, Sanger sequencing was conducted in the patients of three families. Primers (Supplementary Table 1) were designed according to NCBI (http://www.ncbi.nlm.nih.gov/). After polymerase chain reactions, the sequencing was performed using a BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Foster City, CA, USA) on an ABI GeneAmp 3730xl DNA Sequencer (Applied Biosystems).

For gene expression analysis, we downloaded five databases (Supplementary Table 2) from the GEO (https://www.ncbi.nlm.nih.gov/gds/) under the accession number GSE 13353 [11 ruptured IA (RIA) and 8 unruptured IA (UIA)], GSE 15629 (8 RIA and 6 UIA), GSE26969 (3 UIA), GSE54083 (8 RIA and 5 UIA) and GSE66238 (7 UIA) (14–18), and the superficial temporal arteries or meningeal media arteries were isolated as control arteries. We compared gene expression between ruptured and unruptured IA wall samples as well as unruptured IA and control samples. Gene expression studies were performed in Finnish, Polish, Chinese, Japanese, and Lebanese populations and reperformed using the limma package. The threshold for identification of differentially expressed genes was set as p < 0.05 and |log2 fold-change (FC)| > 1.0. Next, Venn analysis was carried out between the genes of the filtered variants and differentially expressed genes from the five online databases to explore the shared genes, which was checked in online Venn software (http://www.ehbio.com/test/venn/#/). Genes with log2FC < 0 were considered as down-regulated, whereas those with log2FC > 0 were considered as up-regulated.

Candidate variants for the association study were selected based on (1) the commonness of the variants: the allele frequency of a variant identified in familial patients; (2) the results of Venn analysis: the overlapped variants in WES filtered candidate list and differently expressed genes in IA specimens from the GEO databases, and (3) the functional relevance of IA: genes whose function was associated with current known IA pathology such as inflammation, extracellular matrix composition, angiogenesis, and vascular remodeling (19). For variants in overlapped genes in the WES filtered database and GEO differently expressed gene databasets as well as the functional relevance variants, more common variants were given more priority in the replicated association study. Genotyping of the candidate variants was performed using the Sequenom MassARRAY platform (Agena Bioscience, Inc., San Diego, CA, USA). Primers were designed using ASSAY DESIGN SUITE V2.0 based on a single-nucleotide polymorphism (SNP) locus (http://agenacx.com, Agena Bioscience, Inc). PCR amplification was performed on an ABI GeneAmp 9700 384-well Dual (Applied Biosystems), and SNP alleles were identified by the different masses of the extended prime primers using matrix-assisted laser desorption/ionization–time of flight mass spectrometry on MassARRAY Nanodispenser RS1000 (Sequenom, Inc., San Diego, CA, USA). The mass spectrometry peaks were detected using Typer 4.0 software (Agena Bioscience, Inc.) and the genotypes of each sample target locus were interpreted based on the mass spectrum peak map. Genotyping was performed by an operator blinded to the sample status (case or control subjects). The primers listed in Supplementary Table 3 were designed using Assay Design 3.1 (Sequenom, Inc.).

Homology of the candidate variants was determined using protein BLAST (http://blast.ncbi.nlm.nih.gov/Blast.cgi). Additional functional analysis of the candidate genes was performed using bioinformatic methods. STRING database (https://string-db.org/) was used to generate a protein–protein interaction network and a computational model was created using SWISS MODEL (https://swissmodel.expasy.org/) to predict the effects of variations on domain structure.

For mRNA expression analysis, GEO 2R was used to compare the differences in the mRNA levels between different groups. Genes with both p-value < 0.05 and |log2FC| > 1.0 were considered as potentially differentially expressed. For the selected candidate variants, their association with IA was analyzed using SPSS 21.0 software (SPSS, Inc., Chicago, IL, USA). Clinical characteristics of the participants are presented as the mean ± standard deviation for normally distributed continuous variables and proportions for categorical variables. Student's t-test, chi-square test, or Fisher's exact test were used to compare differences between the groups for continuous and categorical variables. Multivariate logistic regression analysis was used to estimate the odd ratio and 95% confidence interval of each SNP after adjusting for known risk factors. Bonferroni correction based on 20 independent effective tests was used to adjust the significance level of the association (0.05/20 = 0.0025). A p-value < 0.0025 was considered as a significant association.

This study included 9 familial IA participants, 384 sporadic IA participants, and 384 control individuals. Female sex was predominant in all three groups (88.89%, 69.53%, and 69.53%, respectively). The mean age of the familial IA cases was lower than that of the sporadic cases (50.50 ± 13.11 vs. and 57.10 ± 10.60 years) and the mean age of normal controls (66.54 ± 2.12 years) was higher than that of the other two groups. Four (44.44%) familial and 198 (51.56%) patients with sporadic IA suffered from a history of SAH. Smoking, drinking, medical history of diabetes, and hyperlipemia were more frequent in the control group. The detailed characteristics and clinical information of the study participants are summarized in Table 1.

WES generated 9.76–14.42 Gb bases of raw data for the nine familial participants, of which 9.47–13.23 Gb bases could be effectively mapped to the genome. The mean sequencing depth was between 90.3 and 125X. The average percentage of exomes covered into a read depth over 8X was 98.4%. Single-nucleotide variants were found between 653,710 and 811,492, whereas insertions and deletions were present between 82,537 and 105,085 (Supplementary Table 4).

We focused on non-synonymous single-nucleotide variants and frameshift insertions/deletions in exonic/splice regions, which were present in individuals with IA in more than one family. After filtering of low-frequency or rare variants (illustrated in Table 2), 22 variants remained.

Sanger sequencing was performed to confirm the exome findings. Except for c.1249C>T (p.R417Ter) in ANKRD36C (rs76474100) and c.268G>A (p.G90S) in PSPH (rs75395437), 20 variants were validated in the corresponding individuals. The detailed information is listed in Table 3. Among the variants, GUCY1B3 c.17+5T>G, ADAD2 c.1298+6C>G, and PNMT c.411-3C>T were in the splice region, and the remaining were missense and frameshift variants in the exonic region.

The IA wall samples and control STA samples in five GEO databases were divided into UIA vs. RIA and UIA vs. control groups. After analysis, 1,022, 223, and 198 genes showed significant differential expression within the UIA vs. RIA group in GSE13353, GSE15629, and GSE54083, respectively. A total of 588, 1,321, 1,533, and 1,026 genes was differentially expressed between UIA and control samples in GSE15629, GSE26969, GSE54083, and GSE66238, respectively. Venn analysis was performed to identify the shared candidate variants with the five GEO databases (Figures 2A,B), and five genes were confirmed to be differentially expressed among RIA and UIA tissues and control vessels. ANK3, GUCY1B3, and FNDC1 were shared in two databases (GSE13353&GSE54083, GSE13353&GSE15629, GSE15629&GSE66238, respectively), and were up-regulated in UIA compared to in RIA or controls. SLC38A6 and ART3 were in the GSE54083 and GSE66238 databases, respectively. The expression of SLC38A6 and ART3 was decreased in UIA compared to in controls, but increased expression of SLC38A6 was identified in UIA rather than in RIA. The details are listed in Table 4.

In the candidate list, variants in genes that were differentially expressed in IA tissues (ANK3, GUCY1B3, FNDC1, SLC38A6, and ART3) or those that potentially had relevance to IA biology (S100A1, PDSS1, and MEGF6) were prioritized in the replicate association studies. Previous studies showed that S100A1 may function in vascular endothelium maintenance (20, 21), and PDSS1 (22) and MEGF6 (23) play roles in extracellular matrix integrity. Thus, variants in these eight genes were genotyped, and the results of the association study are summarized in Table 5.

Our results showed that the variant of ANK3 c.4403G>A (p.R1468H) was significantly associated with the risk of sporadic IA after Bonferroni correction [odds ratio (OR), 4.77; 95% confidence interval (95% CI), 1.94–11.67; p-value, 0.00019]. The splice site variant of GUCY1B3 c.17+5T>G showed a lower MAF in sporadic cases compared to in the control population and may have protective effects in IA (OR, 0.46; 95% CI, 0.24–0.91; p-value, 0.023). However, this association did not reach the adjusted significance level (0.0025). No associations were observed for the other six variants (all p > 0.05).

The ANK3-centered protein-protein interaction network generated from the STRING database is shown in Figure 2C. Through literature survey, it was found that ANK3 encodes the protein ankyrin-G, which is expressed in various tissues including the kidney, brain, heart, and thyroid. The predicted domains of ankyrin-G contain a membrane-binding domain (MBD), spectrin-binding domain (SBD), death domain, and C-terminal domain (24, 25). The protein plays an essential role in many cellular processes such as promoting the assembly of intercalated discs and axon initial segments, regulating voltage-gated Na⁺ and K⁺ channels, participating in lateral membrane biogenesis, and regulating E-cadherin endocytosis (26–29). A computational model of the SBD domain (982–1,462 aa, sequence identity 77.28%) is shown in Figure 2D. It is predicted that ANK3 p.R1468H is at the end of the SBD (Figure 2F). This mutation affects evolutionarily conserved residues (Figure 2E). Although the variant is not in the highly conserved minimal region required for spectrin binding (surrounded by green lines in Figure 2D), Arg at the amino acid position 1,468 is altered to His, replacing a bulky side chain with a pentacyclic structure. This change may reduce clashes and affect the domain structure of the conserved sequence.