Low-Allele-Frequency Somatic Variants in a Cohort of Sporadic Saccular "Berry" Cerebral Aneurysms.

Institutional review board authorization was obtained through the University of Washington. Informed consent was obtained from all subjects under this Institutional review board (STUDY00001213: Tissue Bank for the Investigation of the Genetics and Basic Biology of Human Vascular Malformations and STUDY00010829: Clinical Phenotype–Genotype Relationships in Vascular Malformations of the Intracranial and Extracranial Vascular Tree). Anonymized data have been made publicly available at Dryad and can be accessed at https://doi.org/10.5061/dryad.cnp5hqcgk . All individuals underwent surgery at the University of Washington hospitals between January 1, 2010, and December 31, 2016, for microsurgical clipping of IAs. Locations of IAs in this study are shown in Figure  1A and 1B . The aneurysm wall distal to the clip was harvested and immediately placed into liquid nitrogen (Figure  1C and 1D ). Peripheral blood samples were drawn at the time of surgery for harvesting germline DNA (Figure  1E ). Samples were stored in a −80 °C freezer. In some cases, IAs were sent to pathology for normal processing, fixation, blocking, and slide analysis. Tissue punches or slices were obtained for DNA retrieval from archived tissue. Clinical data were collected regarding history, individual demographics, imaging studies, neuropathology reports, operative information, and individual outcomes. Data were collected regarding pathological assessment. All clinical assessments of patients were performed by a neurosurgeon. Location of the aneurysm and size were confirmed by reviewing the computed tomography angiograms or conventional angiograms. A and B , Coronal and sagittal depictions of the locations of exome (red) and validation (blue) aneurysms. The aneurysm wall distal to the clip was harvested ( C and D ). Both the aneurysmal tissue and peripheral blood were collected for isolating germline DNA ( E ). The DNA extraction protocol was the same as previously described by this group. 37 DNA extraction from IAs and their paired blood samples were conducted with the QIAamp DNA mini kit (Qiagen, Hilden, Germany) and QIAamp DNA Blood Mini Kit (Qiagen) treatment protocol. A DNA repair step using the NEBNext FFPE DNA Repair Mix (New England Biolabs, Ipswich, MA) was performed to ensure the integrity of both DNA and RNA available from formalin‐fixed paraffin‐embedded tissue. Ten nanograms of DNA were used per digital droplet polymerase chain reaction, using previously described methods. 38 , 39 Whole‐exome sequencing (WES) was carried out on Illumina HiSeq 2500 instruments (Illumina, San Diego, CA) following exome capture (SeqCap EZ Exome version 3; Roche, Basel, Switzerland) and library generation (KAPA Hyper Prep; Kapa Biosystems, Boston, MA). Aneurysm samples were sequenced to a target coverage of 200× over captured regions with the lowest covered sample reaching an average coverage of ≈145×. Paired blood samples were sequenced to a target coverage of 100× with the lowest covered sample reaching an average coverage of ≈121×. Following sequencing, reads were mapped against the human genome (GRCh37) using the BWA‐MEM aligner version 0.7.5. 40 Reads then underwent postprocessing using the Genome Analysis Toolkit version 3.3 (Broad Institute, Cambridge, MA) 41 using the developer's recommended best‐practices. Potential somatic mutations were then identified using an ensemble approach with multiple callers for single‐nucleotide polymorphisms (MuTect version 1.1.4, Varscan version 2.3.7, JointSNVMix version 0.7.5 and SomaticSniper version 1.0.4) and indels (Varscan and SomaticIndelDetector). All potential variants were evaluated by manual inspection using an integrative genomics viewer ( https://software.broadinstitute.org/software/igv/ ). The resultant variant calls were then annotated using Ensembl Variant Effect Predictor GRCh37 release 110 42 and Funcotator version 4.2.4.0 41 to allow for variant impact assessment. Variants were compared against Genome Aggregation Database version 4.0.0 43 and Catalogue of Somatic Mutations in Cancer release version 92 42 databases to allow for assessment of previous identification and prevalence in the population. To validate the presence of suspected somatic variants in the samples submitted for exome sequencing, we designed a custom capture set of 82 xGen Lockdown Probes (Integrated DNA Technologies, Coralville, IA; https://www.idtdna.com/ ) meant to target the sites of putative variants. Samples were prepared similarly to samples undergoing exome sequencing with the exception that samples were pooled before target capture. Then, samples were sequenced to a target depth of >100× and reads were mapped and postprocessed identically to exome samples. For variant identification we used the Platypus variant caller version 0.7.9 for expedited analysis as compared with tools used in the exome calling. Platypus was run using the default parameters except for the minimum variant frequency, which was set to 0, and the minimum posterior probability, which was set to 0, to allow identification of low‐level variants. A panel of 80 genes was designed on the basis of the results of exome sequencing consisting of 1400 xGen Lockdown Probes. Samples were prepared similarly to the procedure used for the small exome validation probe set (standard exome preparation with pooling before target capture). Samples were sequenced to a target depth of 100×, and reads were mapped and postprocessed identically to exome samples. For variant identification, we again used the Platypus variant caller version 0.8.1, this time as part of a 2‐step variant identification process. Initially, each sample was singly called for variants using Platypus (minimum posterior probability=0, minimum reads for a call=5, and minimum variant frequency = 0.01). Resulting variant calls for all samples were then merged and used to guide batch variant calling against all samples with Platypus in the second step using identical parameters to the first step. Variants were annotated, analyzed, and confirmed with manual inspection. A second custom panel of 206 genes (list of genes available in Table  S1 ) was designed on the basis of evidence of involvement in pathogenesis of IA or other genetic vascular disease (eg, PKD1/2 , COL3A1 , PDGFRB ). Sequencing and analysis were performed with the exact same methods as described above. Hematoxylin and eosin and Verhoeff's‐Van Gieson staining was completed for all samples as previously described. 37

Ten patients underwent clipping for 11 anterior circulation IAs (Table  1 ). Eight IAs were ruptured, and 3 were unruptured. There were 4 anterior communicating artery, 2 pericallosal anterior cerebral artery (A2), 2 middle cerebral artery (MCA) bifurcation (M2), 2 anterior temporal artery aneurysms and 1 MCA distal (M4) fusiform dissecting IA. One of these patients had both a ruptured anterior communicating artery IA and a large unruptured MCA bifurcation IA treated at the same surgery (Figure  2 ). Histologic analysis of ruptured IAs showed characteristic intimal proliferation (Figure  3 ). WES and VAL Data Set Patient Demographics ACA indicates anterior cerebral artery; Acomm anterior communicating; H&H, Hunt and Hess; IA, intracerebral aneurysm; ICA, internal cerebral artery; MCA, middle cerebral artery; PCA, posterior cerebral artery; PComm, posterior communicating; PICA, posterior inferior cerebellar artery; VAL, validation data set; and WES, whole‐exome sequencing. The patient was found to have a ruptured 7‐mm anterior communicating aneurysm (WES‐3A) as well as a 4‐mm right M4 aneurysm (WES‐3B). A indicates anterior; L, left; P, posterior; R, right; and WES, whole‐exome sequencing. While portions of adventitia (***) and tunica media (**) are noted the endothelial layer and the internal elastic lamina are largely absent, the latter only seen on a small focus (not shown). There is intimal proliferation and an atheroma present (*). Verhoeff–van Gieson stains confirm lack of internal elastic lamina in most of the arterial wall but present in an adventitial artery (arrowhead). Scale bars: left 1 mm, right 200 μm. Patient characteristics are shown in Table  1 . The average patient age was 49 years. There were 3 male and 7 female patients. Four patients had a history of smoking, and 5 had hypertension. The average size of the largest IA diameter was 6 mm. No patients had syndromic abnormalities or known germline genetic disorders predisposing to IAs. Two individuals had a family history of IAs, including 1 patient with IA history in a sister (age 19 years) and father (age 60 years) and another patient with IA history in the father (age 80 years) and a sister (age 60 years). These patients had no evidence of a known germline genetic disorder; thus, IAs in this series were thought to reflect common sporadic anterior circulation IA pathology. In all samples, DNA quality and quantity passed internal quality controls. Comparing the germline and IA wall exomes revealed no evidence of copy number alterations (data not shown). A total of 224 single‐nucleotide variants were identified using WES. Of the 224 single‐nucleotide variants, 74 were found within coding regions (Table  S2 ). Allele frequencies ranged between 3% and 24% (Figure  4 and Table  S3 ). Four IAs had no coding variants identified in the exome (WES‐1, WES‐3A, WES‐4, and WES‐9). Variants from WES were validated with targeted next‐generation sequencing of the same IA walls after a separate DNA extraction, in 8 of the 11 IAs with additional available tissue. All variants within those samples were confirmed to exist with very similar allele fractions, validating our method of variant filtering. Of the coding variants, most were missense (89.3%). There were 5 (4.9%) stop‐gain variants. Six variants (5.8%) were found in splice sites, including 4 splice site donor variants (3.9%) and 2 splice site acceptor variants (1.9%). One hundred three somatic coding variants were identified in 7 of the 11 aneurysms harvested. Y ‐axis labels each gene names, and x axis indicated the allele frequencies of each variant. Average allele frequencies ranged from 3.85 to 18.23. Allele frequencies were in a similar range in each sample, supporting a common hypermutated cell population or lineage. WES indicates whole‐exome sequencing. Many of the detected single‐nucleotide variants corresponded to genes known to be important in cancer pathogenesis or vascular tissue development. A significant number of these somatic variants were in highly conserved amino acids, predicted to alter the function of the protein by in silico prediction tools. In the patient with 2 simultaneous IAs, each had a completely different genotype with no common variants (WES‐3A, anterior communicating artery location and WES‐3B, M1 location). Notably, the anterior communicating artery IA, which had ruptured, did not have pathogenic somatic variants identified, whereas the unruptured MCA IA had 20 variants identified around the 8% AF range, including 2 well‐known oncogenes, ABCB4 and RELA . Two genes, ABCB4 and RAB11FIP1 , had unique variants, in 2 independent patients. Both samples with ABCB4 variants were unruptured right anterior temporal artery IAs. A targeted next‐generation sequencing panel was designed for sequencing of our validation cohort of 68 independent sporadic saccular or fusiform IA walls (Table  S2 ). The 68 IAs in this cohort were from 68 different patients. The panel included 80 genes, altered in the 11 WES samples (Table  S1 ). When available, a matching peripheral blood sample was also included to confirm the somatic nature of any detected variants. Thirteen additional single‐nucleotide variants were detected in unrelated IAs in the following genes: DCC , ATXN1L , CCDC178 , GAD2 , IARS2 , ERBb4 , COL4A5 , FAM122C , ANK2 , SF3B3 , and ADCY7 (Table  2 ). These variants were also in the same AF range, 2.1% to 18.6%, and most were novel and predicted to be pathogenic by various prediction algorithms (Table  S3 ). There were no exact repeated variants as would be seen with focal hotspots. Whole‐Exome Sequencing Suspected Pathogenic Variants AF indicates allele frequency; CADD, combined annotation dependent depletion; gnomAD, Genome Aggregation Database; PolyPhen, polymorphism phenotyping; and SIFT, sorting intolerant from tolerant. The second custom panel included sequencing of 206 additional genes in a subset of 31 samples of our validation cohort (due to limited tissue availability, detailed in Table  S2 ). Among these were genes that cause syndromes with high risk of IA or aortic aneurysms; genes in the same family as some of the targets identified with the discovery WES, such as the collagen genes; and genes that are downstream mediators in pathways, such as transforming growth factor‐β and platelet‐derived growth factor, that were shown to be effectors in aneurysm pathogenesis (Table  S1 ). This experiment yielded 7 variants within the following genes: COL4A1 , COL5A2 , FBN1 , MYLK , PKD1 , SERPINH1 , TYK2 (Table  3 ). Allele frequencies ranged between 2.07% and 18.89%. Variants in Same Genes ACA indicates anterior cerebral artery; AComm, anterior communicating artery; ATA, anterior temporal artery; MCA, middle cerebral artery; PICA, posterior inferior cerebellar artery; VAL, validation data set; and WES, whole‐exome sequencing. Combined, these 2 validation panels identified low AF somatic mutations in 13 of the 68 IAs in our discovery cohort (19%). The 68 IAs in this cohort were from 67 different patients. Two IA samples were harvested from the same patient (validation data set [VAL]‐15A and VAL‐15B). In addition, 1 of the IA samples in this cohort was also harvested from the same patient as 1 of the IA samples in the discovery exome cohort (VAL‐50 [WES‐7B]). Five of these 13 IAs carried 2 different variants. No specific variant was detected in more than a single sample. The ATXN1L deletion (ENST00000427980.2: c.1008delG; p.G239fs) and TYK2 insertion (ENST00000525621.1: c.1304‐1305insACGCTCA; p.V275*) were the only indel variants detected in these panels, causing frame shifts, predicted to be deleterious. There was an additional premature “stop” variant in SF3B3 , similar to the WES variant, as well as a splice acceptor mutation in GAD2 that is predicted to alter the splicing. Twenty‐nine distinct variants in 22 genes with potential to be pathogenic driving mutations were identified on the basis of the known functions of the genes. Table  4 highlights 7 of these variants. Most of these variants are novel and not previously reported in large population databases. Population AFs, when available, are reported from the Genome Aggregation Database version 4.0.0 ( https://gnomad.broadinstitute.org/ ; containing genomic data from 807 162 individuals). Interestingly, some of them were reported as somatic mutations in cancer samples in the COSMIC database. 42 Both the polymorphism phenotyping and sorting intolerant from tolerant tools are computational tools used to predict pathogenicity of missense variants. 43 , 44 , 45 The Combined Annotation‐Dependent Depletion score was described to assess measure of variant deleteriousness that can effectively prioritize causal variants in genetic analyses, particularly highly penetrant contributors to severe Mendelian disorders. 46 Somatic Variants Detected in Genes Known to Be Associated With IA Pathogenesis AF indicates allele frequency; CADD, combined annotation dependent depletion; gnomAD, Genome Aggregation Database; IA, intracerebral aneurysm; PolyPhen, polymorphism phenotyping; and SIFT, sorting intolerant from tolerant. Two patients with ruptured MCA IAs harbored variants in both COL4A5 and PABIR3 (WES‐5 and VAL‐50). One patient had a right anterior cerebral artery IA included in the WES cohort (WES‐7) but also had another IA of the right MCA bifurcation that was included in the validation set (VAL‐50). The anterior cerebral artery IA (WES‐7) had a total of 12 variants detected in the WES, and 7 of the genes carrying these variants were selected for the validation experiment. Sequencing of the MCA IA (VAL‐50) revealed no reads with any of these 7 variants but instead presented 2 unique variants in COL4A5 and PABIR3 genes, again demonstrating unique somatic genotypes in each IA.

This study was funded by the US National Institutes of Health under National Heart, Lung, and Blood Institute grant 1R01HL103996 (to Dr. Ferreira). Dr. Ferreira was supported, in part, by the Chap and Eve Alvord and Elias Alvord Chair in Neuro‐Oncology in Honor of Dr and Mrs Ellsworth C Alvord, Jr. Dr Kim was supported, in part, by the Mark and Sheri Robison Family. The funding sources had no role in the design and conduct of the study, collection, management, analysis and interpretation of the data, preparation, review, or approval of the manuscript or decision to submit the manuscript for publication.

Here, we demonstrate the role of somatic mosaicism in the development of saccular IAs. In human fusiform IAs, we were the first to describe driving pathogenic gain of function mutations in PDGFRB to be in the vascular smooth muscle layer; however, this may not be the case for saccular IAs, which are histologically different (Figure  3 ). Somatic gain‐of‐function PDGFRB variants in the kinase and juxta membrane domain are found in mosaic and sporadic human fusiform IAs, but not in saccular IAs. The variants found were able to cause auto‐hyperphosphorylation of PDGFRB, when overexpressed in cell lines. This suggests a novel mechanism of somatic variants, for the formation of vascular pathology. Interestingly, many of the variants identified in saccular IAs have been identified in oncogenesis, consistent with a shared disruption in regulation of cell proliferation. Although variants were detected at very low allele frequencies, we were able to consistently validate all variants in similar fractions. We also saw consistency in the allele frequencies within a sample, suggesting a possible subpopulation of cells within the aneurysm wall with the described genotypes. The low range of AF of the somatic variants could be a possible explanation for why coding mutations in 4 of the 13 IAs were not detected, as potential variants in those IAs could be below the range we can distinguish from sequencing artifacts or level of detection with WES. Saccular IAs harbored genetic variants in several different pathways important in cancer development including oncogenes, extracellular matrix organization, DNA repair mechanisms and angiogenesis. One of the identified oncogenes with variants was cell surface tyrosine kinase ERBb4 , also known as HER4, which has been implicated in cancers both via overactivating and loss‐of‐function mutations and downstream effects via mitogen‐activated protein kinases and phosphatidylinositol 3‐kinase/protein kinase B pathways. 47 Both of the ERBb4 variants we observed (I436L and H647Y) are on the extracellular domain and are not reported in cancers, although an adjacent amino acid, Q646 mutation in human mammary tumor cell lines, was shown to have constitutively active ErbB4. 48 ErbB4 also plays a role in preserving the blood brain barrier integrity and was recently implicated in the pathogenesis of early brain injury in rat subarachnoid hemorrhage models. 49 Multiple genetic variants involved in extracellular matrix remodeling were identified among saccular IAs. COL4A1 encodes the − 1 chain of collagen IV, an integral part of the basement membrane. Monogenic germline mutations of COL4A1 are associated with various vascular syndromes, including HANAC (hereditary angiopathy with nephropathy, aneurysms, and muscle cramps) which causes IA. 50 The variant we observed also affects a glycine residue in the triple helix region, like the variants reported in HANAC syndrome; however, it has not been reported before in these patients or healthy controls. A systemic review of patients with all COL4A1 related conditions found that 44.4% had asymptomatic IA and half of the cases had multiple IA. 50 COL4A5 encodes the − 5 chain of collagen IV and have been implicated in X‐linked Alport syndrome. 51 , 52 There have been case reports of significant vascular disease including early presentation of aortic diseases such as dissection and aneurysm among male patients with Alport's syndrome. 53 These clinical observations, along with the evidence shown in this series, may suggest that a defective form of collagen may be a critical factor contributing to IA development. The concept of the same genes causing familial cancers with germline mutations and sporadic tumors of the same kind via somatic mutations is now common knowledge. Our discovery of somatic variants in FBN1 and PKD1 , genes causing Marfan syndrome and autosomal dominant polycystic kidney disease (2 of the syndromes with highest risk of developing IA) may suggest a similar concept for IA. Neither of the somatic variants in these genes have been reported to cause the associated syndromes before. The FBN1 variant is novel, whereas the PKD1 variant has been observed in only 8 individuals in the Genome Aggregation Database and is extremely rare. In summary, we show that IA saccular aneurysms harbor somatic low allele fraction variants in coding genes predicted to alter protein function. Additional studies are needed to further define the relationship between the variants identified in this series and mechanisms of saccular IA development. Many of the variants identified here do not have a clear role in IA pathogenesis, such as the tumor suppressors, but these may be important in creating an advantage for the mosaic clones of cells to accumulate other crucial mutations. There are several limitations of this study including the small sample size, especially in the WES discovery cohort, and the lack of IA diversity in this series. All aneurysms in this series that underwent WES were in the anterior circulation. It is noteworthy that some gene variants identified in the WES cohort were altered in posterior circulation IA within the validation cohort (Table  2 ). IAs of the posterior circulation are also at higher risk of rupture and increased growth rate. 54 Some of the identified variants were predicted to be benign/tolerated or exist commonly in the healthy population. We did not perform any in vivo studies to establish the pathogenicity of these variants in the IA tissue or the causality of the hypermutated phenotype we have observed. Further research is required to understand the role of the variants described and how mosaicism influences the penetrance of saccular IAs and their origins. The incompletely understood heritability of IA could be due to complex interactions of germline and somatic vessel tissue variants. Perhaps the identified familial risk loci determine the propensity of acquiring somatic variants within the target tissue, rather than playing a direct role in pathogenesis. These new minimally invasive molecular methods may quantitate the extent of mosaicism within these IAs and help us understand how IAs develop and determine new therapeutic targets, just as these methods have revolutionized the field of oncology.

None.

Tables S1–S3