Exome sequencing in 38 patients with intracranial aneurysms and subarachnoid hemorrhage

Individuals of Northwestern German descent (genetically belonging to the group of non-Finnish, Northwestern Europeans) were recruited in the Interdisciplinary Neurovascular Center of the University Medical Center Hamburg-Eppendorf, Hamburg, Germany; approximately 70 patients with aSAH and 200 patients with intracranial aneurysms in total are treated there per year. Inclusion–exclusion criteria were as follows: (i) Only individuals with a family history of UIA and/or aSAH and substantially affected sporadic patients with ≥ 2 UIA or ≥ 1 UIA and aSAH were included. To that, we selected patients based on age, number of aneurysms and family history, resulting in a collective of relatively young patients with an increased number of aneurysms and family history above average compared to data provided in the literature (Table 1) [15, 16]. Experienced physicians assessed the patients; familial history, clinical appearance, primary risk conditions (e.g. hypertension) and life-style risk factors (e.g. smoking) were recorded (Table 1). Missing clinical data of patients previously treated in a different institution were gathered as far as possible. Phenotype definitions following the FIA study protocol [9]: “UIA and/or aSAH” (individuals having a documented UIA based on intra-arterial angiogram or operative report; individuals having an aSAH documented in medical records based on CT/MRI-scan, intra-arterial angiogram or operative report); “probable aSAH” (aSAH was stated in the medical history with supporting documentation such as severe headache or altered level of consciousness). (ii) Informed consents for blood collection, DNA storage and genetic analyses were obtained from all study participants. (iii) Patients with a familial or personal history of disorders associated with IA (i.e. polycystic kidney disease, Ehlers–Danlos syndrome, Loeys–Dietz syndrome and Marfan syndrome) were excluded based on clinical findings. In total, we screened 101 individuals, of which 75 met the described inclusion criteria. To assess the putative risk of rupture in patients with UIA in comparison to those with a proven aneurysmal bleeding, we calculated the PHASES score (Table 1) [17]. In patients with multiple UIAs, the score was calculated for each aneurysm site and size, and the highest score was documented (Table 1). In the aSAH group, the score for the ruptured aneurysm was calculated retrospectively (Table 1). Finally, we assessed clinical characteristics for following subgroups (Table 1): exome sequenced individuals (ES IND), individuals with positive family history (FAM IND) and substantially affected sporadic individuals (SPO IND).

Exome sequencing (ES) is a genomic technique for sequencing all coding regions (= exome) of a genome: after capturing only the subset of DNA that encodes proteins, the exome is sequenced using a high-throughput DNA sequencing technology. We performed ES in 35 unrelated individuals and 3 affected members of a family. Selection criteria and constraints of subjects for exome and Sanger sequencing are explained in detail in Supplementary Methods and Table S1. Although genome-wide association studies (GWAS) that focus on the identification of common variants (minor allele frequency, MAF > 5%) revealed a large number of risk loci associated with IA/SAH, the accounted heritability in these loci was low [18]. The prevalence of IA is reported to be approximately 5% in average [2]; for genetic disorders the phenotype prevalence and penetrance should be used to define the allele frequency cutoff [19]. Therefore, for the validation of reported UIA/SAH risk genes, we screened for low-frequency (MAF > 1 to 5%), rare (MAF > 0.1 to 1%), very rare (MAF ≤ 0.1%) and unknown variants (MAF = 0) in these genes [20, 21], thereby including those variants with strong, intermediate as well as moderate effect size and penetrance [22]. For the identification of novel susceptibility genes in the family-based approach we searched for unknown variants (MAF = 0; with anticipated significant effect size/penetrance) shared by the affected family members. Subsequent variant prioritization, the top familial candidate disease gene was examined in the remaining 37 patients from our cohort.

Detailed information on exome sequencing, variant calling, filtering strategy, variant prioritization, pathogenicity assignment, variant classification, mutation analysis, presentation of clinical data, statistical analysis (gene association tests), molecular modeling as well as databases and repositories is outlined in Supplementary Methods.

Our study cohort comprises 75 subjects (53 female; 70.7%) including 48 individuals (64.0%) with aneurysmal subarachnoid bleeding (aSAH group, Table 1) and 27 patients (36.0%) with UIA (UIA group, Table 1). 31 patients (41.3%) showed multiple IA (including ruptured and unruptured), the mean count of aneurysms was 2.3 in the UIA group and 1.8 in the aSAH group. In the latter group, 5 patients (10.4%) suffered from a previous aSAH. In addition to UIA and/or aSAH, three patients (4.0%) had an aortic aneurysm. 30 individuals (40.0%) reported a family history of IA and/or SAH. The mean PHASES score was 4.8. Clinical data are summarized in Table 1. Taken together, this cohort consists of individuals with a family history of UIA and/or aSAH and substantially affected patients with UIA and/or aSAH.

38 patients underwent ES. The presence of possibly pathogenic variants in known disease genes for vascular/connective tissue disorders was excluded (details are given in Supplementary Methods and Table S2). Subsequently, exome data were analyzed for variants in the reported risk genes ADAMTS15, ANGPTL6, ARHGEF17, LOXL2, PCNT, RNF213, THSD1 and TMEM132B. In total we found 20, exclusively heterozygous missense variants with MAF ≤ 0.05 (5%) in 6 genes (Table 2): 9 PCNT variants were identified in 9 patients, 4 RNF213 variants in 3 patients, 3 THSD1 variants in 6 patients, 2 ANGPTL6 variants in 3 patients, and 1 variant in ADAMTS15 and TMEM132B in 1 patient each. No variants were detected in ARHGEF17 and LOXL2. According to ACMG/AMP guidelines [23], we defined 17 variants of uncertain significance (VUS) and 3 likely benign variants (LBV) (Table 2). Details on criteria for classifying variants are given in Table 2 and in Supplementary Methods. All detected variants were annotated in dbSNP (www.​ncbi.​nlm.​nih.​gov/​SNP/​) and gnomAD [24]; we did not find undescribed variants in the reported risk genes.

In total, 18 unrelated individuals carry variants in reported risk genes (Table 2). Eleven patients (IA24, IA49, IA57, IA59, IA64, IA72, IA75, IA76, IA83, IA84 and IA90) had one variant. Seven patients had concomitant variants: IA63 and IA85 had two variants in PCNT; for IA63 these variants were in trans. IA79 showed one variant each in PCNT and RNF213, IA54 in PCNT and THSD1, IA61 in ANGPTL6 and RNF213, and IA17 in THSD1 and TMEM132B. Notably, the number of variants had no impact on clinical manifestation. Finally, we detected 3 variants (1 in ANGPTL6 and 2 in RNF213) in patient IA78 (Table 2). Within the analyzed patient cohort, we found 4 recurrent VUS (Table 2): ANGPTL6 c.287A>G p.(Leu96Pro), PCNT c.4354G>A p.(Gly1452Arg) and PCNT c.7652C>T p.(Ala2551Val) in 2 patients each; and THSD1 c.871C>T p.(Glu291Lys) in 4 patients. In case of DNA of further affected and unaffected family members was available, cosegregation with disease was checked. The PCNT c.6739C>T p.(His2247Tyr) variant in IA24 did not cosegregate with the disease in the family (Figure S1A), indicating that this variant is not pathogenic. THSD1 c.871C>T p.(Glu291Lys) in IA90, however, was also identified in 2 affected siblings (Figure S1A), thereby supporting a clinical relevance for this variant.

We performed ES in a family with three affected family members (Fig. 1a). The three siblings share no reportable sequence alteration in known genes associated with vascular/connective tissue disorders (Table S2). From ES data, we extracted potentially pathogenic sequence variants by bioinformatics stratification, biological filter steps and variant prioritization (details are outlined in Supplementary Methods). Briefly, our filtering steps retained homozygous and heterozygous variants if they (i) were present in all affected family members; (ii) were listed neither in the database gnomAD [24], nor in a combined exome dataset of population-based controls from PopGen [25, 26] and of control samples collected in the Institute of Clinical Molecular Biology in Kiel (IKMB-controls); (iii) were predicted to be synonymous variants, missense variants, nonsense variants, frameshift indels or intronic alterations at exon–intron boundaries ranging from − 2 to + 2; and (iv) were predicted to be damaging by CADD, REVEL, M-CAP and/or ClinPred; missense variants with less than two scores above the respective pathogenicity thresholds were excluded; nonsense variants and indels causing premature termination were retained if CADD score was above recommended pathogenicity threshold; synonymous variants were retained if consequences on splicing was suggested by > 1 in silico predictor (Table 3). This stringent filtering pipeline revealed five candidate risk variants for UIA/aSAH in NEK4, EDIL3, EDNRB, DNAH9 and GGA3 in this family (Table 3). We classified these variants according to ACMG/AMP guidelines [23], and we defined five VUS (Table 3) including EDIL3 c.383G>A p.(Cys128Tyr) (Fig. 1b).

Based on pathogenicity predictions and a strong functional relevance of EDIL3 protein (see “Discussion”, Supplementary Methods and Table S3), we analyzed EDIL3 coding region in available exome data and by Sanger sequencing in 35 and 37 patients, respectively. We detected a very rare variant classified to be a VUS_PM2+PP3 in EDIL3 (c.1159A>C, p.(Lys387Gln)) (Fig. 1b) in two further unrelated patients (IA10 and IA13) both with no reported family history. IA10 showed two intracranial, one carotid and two aortic aneurysms; IA13 had two intracranial aneurysms and aSAH at the age of 35 years. The allele frequency of EDIL3 c.1159A>C p.(Lys387Gln) is 0.000131 in gnomAD and similarly small with 0.000101 within 5814 healthy control individuals from the German population-based datasets IKMB-controls and PopGen [24‐26]. Notably, gnomAD may contain data from cerebrovascular patients and thus the presence of specific variants does not exclude pathogenicity. The variant is predicted to be deleterious by three ensemble pathogenicity classifiers including CADD (score 24.5), REVEL (score 0.723) and M-CAP (score 0.266). We also identified additional variants with MAF ≤ 0.05 in the remaining four candidate risk genes in available exomes of our cohort. However, relatively high MAFs and/or low pathogenicity prediction scores do not support a role for NEK4, EDNRB, DNAH9 and GGA3 variants in the etiology of UIA/aSAH (Table S4).

To test the association of unknown and very rare EDIL3 variants (MAF ≤ 0.001) with susceptibility to UIA/aSAH, we compared the numbers of individuals carrying at least one missense variant in this gene among the 73 index case subjects in the present study (3/73 carrier; 4.11%) versus 100 healthy/control individuals with North German ancestry (Hamburg controls: 0/100 carrier; 0.0%) by gene burden and Sequence Kernel Association Tests (SKAT) [27]. We found a significant enrichment in individuals carrying EDIL3 missense variants with an MAF below 0.001 (0.1%) in our study cohort versus healthy/control individuals (SKAT, p = 0.0298; burden test, p = 0.0152; Table S5). We obtained even stronger results when comparing allele counts among the 73 index case subjects versus 64,603 non-Finnish European individuals with exome/genome sequences derived from the gnomAD database v2 (326 variants in 64,603 individuals; SKAT, p = 2.53e−33; burden test, p = 0.0000150; Table S5) [24]. Finally, the cumulated allele frequency of very rare (allele frequencies ≤ 0.001) EDIL3 missense variants in non-finish Europeans listed in gnomAD (326/64,603) is 0.00505. Compared to this, the cumulative frequency of unknown/very rare variants (i.e. c.383G>A and c.1159A>C) in EDIL3 in our cohort (3/73; 0.0411) is significantly higher (Fisher’s exact p = 0.00624; Table S5).

We explored the structural impacts of the p.(Cys128Tyr) and p.(Lys387Gln) changes using a model for EDIL3 EGF-like domains and discoidin domains, respectively [28, 29]. Molecular replacement of the amino acid Cys¹²⁸ for a tyrosine resulted in (i) loss of the disulfide bridge between amino acid 128 and Cys¹⁴³, and (ii) aberrant intramolecular interactions of Tyr¹²⁸ with cysteines 143, 145 and 154 (Fig. 1c). Replacement of the amino acid Lys³⁸⁷ for a glutamine resulted in loss of six intramolecular contacts with Tyr⁴⁵² (Fig. 1c). These data suggest that both p.(Cys128Tyr) and p.(Lys387Gln) induce structural alterations which likely interfere with EDIL3 function.

Taken together, we identified two different EDIL3 missense variants, c.383G>A p.(Cys128Tyr) and c.1159A>C p.(Lys387Gln), in three affected individuals from one family and in two unrelated patients, respectively. The p.(Cys128Tyr) variant is absent from databases, the p.(Lys387Gln) variant is very rare; this supports the hypothesis that the two variants might be involved in the pathogenesis of the disease. Both alterations are predicted to be damaging by combinational pathogenicity predictors as well as by molecular replacement, and the functional relevance of EDIL3 for vascular biology is strong (see “Discussion” section). Therefore, EDIL3 is a valid candidate disease gene for UIA/aSAH.

For establishing our study cohort, we focused on patients with a severe clinical phenotype, a family history of UIA and/or SAH and few or no exogenous risk factors [9, 10]. This resulted in a relatively young cohort consisting of patients with proven aSAH or a high number of UIA. As a clinical tool, the PHASES score assists physicians to assess the risk of aneurysm rupture; however, this score does not consider family history and genetic factors. Notably, the scores in our aSAH- and UIA-groups were similar and moderately high. This indicates a moderate sensitivity in a selected cohort of strongly affected patients with a high percentage of positive family history. Thus, incorporating genetic/familial aspects might improve the sensitivity of PHASES scoring system. We found EDIL3 variants in strongly affected patients with up to six IA. The two sporadic patients had no familial history and few exogenous risk factors for IA formation. One of these patients was diagnosed with three additional extracranial aneurysms and died subsequent to aortic surgery. The co-occurrence of intracranial and extracranial aneurysms in subject IA10 may suggest an overlapping pathogenesis of intracranial and extracranial vascular diseases, as already reported by others [30, 31], and may reflect the general importance of EDIL3 for vascular integrity.

Continual evaluation of putative risk genes and refinement of reported risk alleles is essential for shaping the genetic etiology of UIA/aSAH [9, 10]. We examined reported risk genes ADAMTS15, ANGPTL6, ARHGEF17, LOXL2, PCNT, RNF213, THSD1 and TMEM132B via exome sequencing in 38 individuals with UIA/aSAH and identified variants with MAF ≤ 0.05 (5%) in 18 of them. Previously, four rare missense variants in PCNT exon 38 encoding a coiled coil motif have been identified in affected members of 2 different families and in 3 out of 161 unrelated patients with nontraumatic SAH or UIA (Figure S2A); thus a specific role for this domain in the etiology of IA/SAH has been suggested [8]. We identified PCNT variants in 9 patients, which do not cluster in a specific region of the gene (Figure S2A). Two of these variants were predicted to alter splicing, three of these variants are likely benign (Table 2). Neither previously reported PCNT missense variants nor those identified in our study localize within a constrained coding region [8, 32]. PCNT is a highly mutable gene which is reflected by a very low misZ score (Table 2) [24]. Taken together, PCNT remains a good candidate gene for IA/SAH, however, final verification of PCNT as a risk gene is still pending. Variants in RNF213 predispose to Moyamoya disease in diverse ethnicities [33, 34], and rare variants have been recently identified in 25 from 249 patients with IA/SAH in a French Canadian population [4]. Moreover, functional and genetic data strongly support that RNF variants are associated with a variety of vascular disorders, as it has been discussed by Zhou et al. (2018) [4]. It was suggested that RNF213 exon 29 encoding AAA + ATPase domains might be the risk region for IA/SAH (Figure S2B) [4]. We identified four VUS in three patients with IA/SAH; one patient with a giant IA carries two rare RNF213 variants (Table 2), and the variant c.8084C>T p.(Ala2695Val) in IA61 localizes in the linker region between the two AAA + domains (Figure S3). Moreover, we found a further VUS in RNF213 (c.10450G>A p.(Gly3484Ser)) in two siblings with SAH but no detectable IAs (Supplementary Results, Table S6 and Figure S1B). Taken together, the plurality of independent evidences and RNF213 intolerance of variation (misZ = 2.3) indicate that RNF213 sequence alterations are risk factors for IA/SAH and related disorders. Genetic evidence suggested a role of THSD1 variants in the pathogenesis of IA/SAH, previously, and missense variants identified in affected individuals clustered in the intracellular protein portion (Figure S2C) [5]. We observed two and one variants in the extracellular and intracellular THSD1 protein portion, respectively (Figure S2C). The THSD1 c.871C>T p.(Glu291Lys) variant, which has a relatively high MAF (0.01864) was found in four unrelated individuals and it co-segregated with the disease in the family of IA90 (Figure S1A). Taken together, our data do not oppose but rather support a role of TSHD variation in the etiology of IA/SAH. For the discussion on variants in ANGPTL6, ADAMTS15, TMEM132B, ARHGEF17 and LOXL2 see Supplementary Discussion; overall evidence for an overlap of our data with previous studies on these five genes is very limited [this study, 6, 7, 9‐11].

There is no significant overlap in putative susceptibility genes between various ES-based studies [35], which may reflect a pronounced heterogeneity of IA/SAH pathogenesis. We also aimed to identify genetic risk factors for UIA/aSAH by a family-based ES approach, and we propose EDIL3 as new risk gene. From the genetic perspective, EDIL3 is a very good candidate disease gene for UIA and aSAH because we identified likely deleterious variants in affected relatives in a family (Fig. 1a) and, additionally, in two unrelated individuals with UIA/aSAH. Moreover, association tests showed a statistically significant gene burden for EDIL3. Functional assessment of EDIL3 protein and the predicted amino acid substitutions (p.(Cys128Tyr), p.(Lys387Gln)) strongly supports this hypothesis (Supplementary Methods, Tables S3 and S4): The mouse homologue Edil3 was identified as a matrix-associated matricellular protein expressed by endothelial cells during embryonic vascular development [36]. Through its interaction with integrin receptors, Edil3 promotes adhesion of endothelial and vascular smooth muscle cells [36, 37]. Both cell types are crucial for wall integrity of blood vessels and angiogenesis [38]. Edil3 is regulated by hypoxia or vascular injury and has been implicated in vascular remodeling during angiogenesis [39‐41]. EDIL3 is highly expressed in blood vessels and in brain (Figure S3). Summarized, EDIL3 belongs to a number of epidermal growth factor (EGF) repeat containing genes in the vascular wall, which are essential for wall integrity and remodeling [37]. Amino acid changes p.(Cys128Tyr) and p.(Lys387Gln) affect highly conserved domains. EDIL3 has 3 N-terminal EGF-like domains, followed by two discoidin domains (Fig. 1b) [36]. The EGF modules are stabilized by six conserved cysteines forming 3 disulfide bonds in the pattern of 1–3, 2–4, and 5–6 (Fig. 1c) [28]. The p.(Cys128Tyr) change affects the 2nd Cysteine in the 3rd calcium-binding EGF-like domain, which likely compromises the shape and function of the small domain (Fig. 1c). Accordingly, mutations of highly conserved cysteines in calcium-binding EGF-like domains play a seminal role in several diseases such as the Marfan syndrome [42]. The p.(Lys387Gln) change alters the 2nd discoidin domain (Fig. 1c), a structural entity that is found in many extracellular and membrane proteins involved in cellular adhesion, migration, or aggregation events, and associated with organogenesis (vasculogenesis and angiogenesis) [43]. The exact role of Lysine 387 that lies within a β-sheet of the discoidin domain has not yet been documented, however modelling of p.Lys387Gln suggested strong effects on the domain structure (Fig. 1c). Notably, it has been shown that depletion of EDIL3 disrupts TGFβ signaling pathways [44]. Dysregulation of this signaling pathway is widely accepted to be central in the pathogenesis of hereditary aortopathies [45]. Numerous familial linkage and GWAS reporting on potential gene loci for IAs have been summarized elsewhere [18, 21]. A genome-wide-linkage/haplotype-association analysis revealed a strong evidence for a linkage of IAs to the chromosomal region 5q14.3 with a p value of 0.001 for marker D5S428; this DNA marker localizes ca. 1 Mb distal from EDIL3 [46].