The dural angioleiomyoma harbors frequent GJA4 mutation and a distinct DNA methylation profile

The International Society for the Study of Vascular Anomalies (ISSVA) has defined four vascular lesions in the central nervous system (CNS): arteriovenous malformations, cavernous angiomas (also known as cerebral cavernous malformations), venous malformations, and telangiectasias. From a retrospective central radiological and histopathological review of 202 CNS vascular lesions, we identified three cases of unclassified vascular lesions. Interestingly, they shared the same radiological and histopathological features evoking the cavernous subtype of angioleiomyomas described in the soft tissue. We grouped them together with four additional similar cases from our clinicopathological network and performed combined molecular analyses. In addition, cases were compared with a cohort of 5 soft tissue angioleiomyomas. Three out 6 CNS lesions presented the same p.Gly41Cys GJA4 mutation recently reported in hepatic hemangiomas and cutaneous venous malformations and found in 4/5 soft tissue angioleiomyomas of our cohort with available data. Most DNA methylation profiles were not classifiable using the CNS brain tumor (version 12.5), and sarcoma (version 12.2) classifiers. However, using unsupervised t-SNE analysis and hierarchical clustering analysis, 5 of the 6 lesions grouped together and formed a distinct epigenetic group, separated from the clusters of soft tissue angioleiomyomas, other vascular tumors, inflammatory myofibroblastic tumors and meningiomas. Our extensive literature review identified several cases similar to these lesions, with a wide variety of denominations. Based on radiological and histomolecular findings, we suggest the new terminology of “dural angioleiomyomas” (DALM) to designate these lesions characterized by a distinct DNA methylation pattern and frequent GJA4 mutations.


Introduction
Intracranial vascular lesions encompass a broad spectrum of entities which differ in hemodynamic physiology, structure and prognosis [1][2][3]. They have been increasingly seen in clinical practice primarily because of new developments in imaging technology [4]. In recent years, an effort has been made to categorize these vascular anomalies, classified over the years in a variety of ways by many authors, often on the basis of blood flow patterns, amplified magnetic resonance imaging (MRI), angiography, histopathological features or demographics [5]. In 1996, the International Society for the Study of Vascular Anomalies (ISSVA) developed a classification scheme with two main groups: proliferative vascular lesions (vascular tumors) and vascular malformations. In 2014, the ISSVA refined its classification system, in order to establish a consistent terminology to serve as a guide for all specialized medical personal: clinicians, radiologists, pathologists and to improve patient management and treatment options [6,7]. The four main categories of commonly encountered vascular malformations found in the Central Nervous System (CNS) are, as described by McCormick et al. [8], arteriovenous malformations, cavernous angiomas (also known as cerebral cavernous malformations (CCM)), venous malformations, and telangiectasias [9,10]. In addition, mixed malformations occasionally occur [11,12]. Molecular alterations are now well-characterized in association with those lesions: CCM1/2/3, MAP3K3 and PIK3CA gene mutations in CNS cavernomas [13][14][15][16][17][18][19], and KRAS mutations in arteriovenous brain malformations [20]. Moreover, recent studies have shown that using DNA-methylation classification, most histopathological CNS and sarcoma tumors cluster into corresponding methylation classes and are stratified into clinically relevant, molecularly distinct subgroups [21,22]. Here, we introduce a novel CNS tumor type with recurrent GJA4 mutation and a distinct DNA-methylation profile, for which we suggest the term 'dural angioleiomyomas' (DALM), which is not currently included in the World Health Organization (WHO) Classification of CNS tumors [23].

Cohort
The database from the neuropathological department at Sainte-Anne Hospital in Paris was retrospectively searched for CNS vascular lesions, referred to in reports as vascular lesions, vascular malformations, cavernomas, angiomas, arteriovenous lesions/malformations, or telangiectasias between 1996 and 2017 ( Fig. 1). Two hundred and twenty-three cases were retrieved and reviewed by two pathologists (PV and ATE). Twenty-one cases were excluded based on: small sample size (n = 11), an extracranial location (n = 6) or the unavailability of  Lastly, three lesions remained unclassified which presented similar clinical, imaging, and histopathological features evoking the cavernous subtype of angioleiomyoma. Thereafter, we screened our national French neuropathological network database and found four other cases presenting the same histopathology (one of them was previously described in [24]). We then compared histopathological and molecular features from our 7 CNS cases with a cohort of 5 soft tissue angioleiomyomas. This study was approved by our local ethics committee. Written informed consent for clinical and imaging investigations and molecular analysis was obtained from all patients enrolled in this study.

Central radiological review
The central radiological review of preoperative MRI and computed tomodensitometry (CT), when available, was performed by a senior neuroradiologist (TP).

Central histopathological review and immunohistochemistry
The central pathology review was performed conjointly by 3 neuropathologists (ATE, PV, EL), a pathologist expert in soft tissue tumors (FL), and a pathologist expert in vascular lesions (MW). For the CNS cases, additional immunohistochemical stainings were performed on paraffin-embedded sections including: CD34 (1:40,  clone QBEnd10

Molecular analyses
Blood samples were drawn and genomic DNA was extracted using the WIZARD Genomic DNA Purification Kit (Promega, Madison, USA) according to the manufacturer's protocol for whole blood. DNA was extracted from the tissues by the WIZARD Genomic DNA Purification Kit. When available, the frozen samples were homogenized in a lysis solution, incubated 1 h at 55 °C with proteinase K and DNA isolation was conducted according to the manufacturer's protocol for Tissue DNA. Sequencing was done with a custom SureSelect XT DNA target enrichment panel designed with SureDesign tools (Agilent Technologies, Santa Clara, USA). All coding and non-coding exons from the three CCM genes, and 50 base pairs of intronic flanking sequence were tested both on DNA extracted from peripheral leucocyte and on DNA extracted from frozen tissues. The libraries were prepared according to the SureSelect QXT target enrichment protocol (Agilent Technologies). Sequencing was performed on a MiSeq next generation sequencer (Illumina, San Diego, USA). Alignment of raw data and variant calling and CNV detection was performed using the SeqPilot SeqNext software version 4.0 (JSI Medical Systems). Whole exome sequencing (WES) and bioinformatic analyses were also performed when cryopreserved tissue was available. Library preparation, exome capture, sequencing and raw data analysis was performed by IntegraGen SA (Evry, France). WES was performed on genomic DNA, and matched constitutional DNA extracted from blood samples at IntegraGen (Evry, France). Libraries were prepared from 150 ng of fragmented genomic DNA using the NEBNext Ultra DNA Library Prep Kit for Illumina (New-England Biolabs) and sequences captured using the SureSelect XT Human All Exon CRE-V2 kit (Agilent) followed by paired-end 75 bases massively parallel sequencing on Illumina NextSeq 500. Following base calling using the Real-Time Analysis (RTA2) software sequence pipeline, sequence reads were mapped to the human genome build (hg19) using the Burrows-Wheeler Aligner (BWA) tool. Duplicate readings were removed and variant calling allowing for the identification of genetic alterations as well as SNV (Single Nucleotide Variation) small insertions/deletions (up to 20 bp) were performed via the Broad Institute's GATK Haplotype Caller GVCF tool (3.7) for constitutional DNA and via the Broad Institute's MuTect tool (2.0, -max_alt_ alleles_in_normal_count = 2; -max_alt_allele_in_nor-mal_fraction = 0.04) for somatic DNA. An in-house post-processing in order to filter out candidate somatic mutations that are more consistent with artifacts or germline mutations was applied. Ensembl's VEP (Variant Effect Predictor, release 90, GENCODE 27) program was used to process variants for further annotation. Taking into account data available from the dbSNP (dbSNP150), the 1000 Genomes Project (1000G_phase3), the Exome Variant Server (ESP6500SI-V2-SSA137), and the Exome Aggregation Consortium (ExAC r3.0) and in-house databases. Regarding missense changes, two bioinformatics predictions for pathogenicity were available: SIFT

DNA methylation array processing and copy number profiling
Genomic DNA was extracted from formalin-fixed and paraffin-embedded (FFPE) tissue of the three undetermined vascular lesions. DNA methylation profiling of all samples was performed using the Infinium Methyla-tionEPIC (850 k) BeadChip (Illumina, San Diego, CA, USA) or Infinium HumanMethylation450 (450 k) Bead-Chip array (Illumina) as previously described [21]. All computational analyses were performed in R version 3.3.1 (R Development Core Team, 2016; https:// www.Rproje ct. org). Copy-number variation analyses from 450 k and EPIC methylation array data was performed using the conumee Bioconductor package version 1.12.0. Raw signal intensities were obtained from IDAT-files using the minfi Bioconductor package version 1.21.4 [21]. Illumina EPIC samples and 450 k samples were merged to a combined data set by selecting the intersection of probes present on both arrays (combineArrays function, minfi). Each sample was individually normalized by performing a background correction (shifting of the 5% percentile of negative control probe intensities to 0) and a dye-bias correction (scaling of the mean of normalization control probe intensities to 10,000) for both color channels. Subsequently, a correction for the type of material tissue (FFPE/frozen) and array type (450 k/ EPIC) was performed by fitting univariable, linear models to the log2-transformed intensity values (remove-BatchEffect function, limma package version 3.30.11). The methylated and unmethylated signals were corrected individually. Beta-values were calculated from the retransformed intensities using an offset of 100 (as recommended by Illumina). All samples were checked for duplicates by pairwise correlation of the genotyping probes on the 450 k/850 k array. To perform unsupervised non-linear dimension reduction, the remaining probes after standard filtering [21] were used to calculate the 1-variance weighted Pearson correlation between samples. The resulting distance matrix was used as input for t-SNE analysis (t-distributed stochastic neighbor embedding; Rtsne package version 0.13). The following non-default parameters were applied: theta = 0, pca = F, max_iter = 30,000 perplexity = 10.

Clinical and imaging findings
The clinical data concerning the 7 patients with undetermined vascular lesions are detailed in Table 1. The median age of presentation in our cohort was 52 years (ranging from [46][47][48][49][50][51][52][53][54][55][56][57][58][59]. The male-to-female ratio was 1.3 (4 males and 3 females). All tumors were extra-parenchymal, mostly supratentorial (4/7 cases) and three cases were intraorbital (developed from the dura mater of the optic nerve) as confirmed by surgical findings. Data from treatment and follow-up were available for all patients (for details see Table 1). A gross total resection was achieved for all patients with no residual disease. All patients were free of disease at the end of follow-up (mean, 86 months; median, 58 months) including one having survived more than 20 years after resection. Preoperative imaging was available for 6/7 patients. Using neuroimaging, all lesions presented the same features on preoperative MRI: solitary extra-axial, well-circumscribed, attached to the dura mater, and located near the left parietal hemisphere (case #1) ( Fig. 2A-C), the right temporal hemisphere (case #2) (Fig. 2D-F), the right cavernous sinus (case #5) the right occipital hemisphere (case #6) and the intraorbital portion of the optic nerve (cases #3, 4, and 7) (Fig. 2G-I). Axial CT scan without a contrast medium was available for only one patient (case #2) and showed a round, well-circumscribed, hyperdense mass with regular margins arising from the cranial dura mater in the right temporal hemisphere (Fig. 2D). No change in the adjacent skull, dural calcification or intralesional calcification was noted (Fig. 2D). They measured in their largest axial diameter between 16 mm (case #1) and 27 mm (case #3) (median: 25 mm, mean: 22.8 mm). The lesions appeared as isointense relative to grey matter on unenhanced T1-weighted MRI and hyperintense on FLAIR and T2-weighted images. FLAIR sequencing revealed an absence of parenchymal edema for intracerebral lesions. Fat-suppressed contrast-enhanced T1-weighted MRI demonstrated an intense and irregular inhomogeneous enhancement after gadolinium administration (Fig. 2C,F,H). No dural tail or bone abnormalities (bony erosion, thinning, or hyperostosis) were noted. No characteristic perilesional hemosiderin deposition, low signal rim or concurrent vascular malformation was observed (Fig. 2E). For the three intraorbital cases, the masses were intraconal, well-circumscribed, encapsulated, hyperintense in T2 compared to muscles, lying close to the rectus muscles, abutting the globe or located in the orbital apex, pushing the optic nerve and in contact with the dura mater of the optic nerve ( Fig. 2G-I).
No angiographic study was performed for any of the patients because the lesions were misdiagnosed as atypical meningiomas or other dural-based lesions on presurgical MRI. After surgery, imaging confirmed total resection of the lesions.

Histopathological and immunohistochemical characterization
The seven CNS lesions had the same histopathological features (Fig. 3). They consisted of aggregates of abnormally enlarged vascular cavities (Fig. 3A,D,G), separated by thick uneven fibrous septa and lined by a single layer of endothelial cells stained by the CD34 antibody (Fig. 3B,E,H). There was no intervening brain parenchyma or identifiable mature vessel wall structures. There were not any recent or organized thrombi in the vascular lumens. There were no hemosiderin deposits peripheral to the lesion. The endothelial cells did not present atypia, mitotic activity, or plump epithelioid cytology. We did not identify any elastic lamina by orcein staining. There was no lymphocytic infiltration or calcification. All cases presented myxoid changes. Immunostainings for SMA and h-caldesmon (Fig. 3C,F,I) showed muscular layers of varying thickness surrounding the vascular cavities, including in perivascular concentric arrangements, whereas desmin was only focally expressed. We did not identify nervous fibers or brain parenchyma with PS100 and GFAP immunostainings. These lesions lacked SSTR2a (apart from the endothelial layer), and EMA immunopositivities which are the classical markers of meningioma. STAT6 was not expressed in any of the cases.

GJA4 mutation is a frequent event in dural angioleiomyoma
To gain insight concerning the genomic abnormalities underlying these lesion, targeted sequencing for CCM1/2/3 was performed. The whole genes were fully covered. The mean coverage depth of the targeted regions in panel sequencing data for blood DNA was between 770 and 830X and the mean coverage for tissue DNA was over 6000X. No variant was identified in the exons and flanking introns nor in the blood DNA or the tissue DNA for all three patients. There was no evidence of a large rearrangement of the CCM1/2/3 genes or of MAP3K3, PIK3CA or KRAS. Of the seven patients, only two had lesional frozen tissue samples available allowing for a genetic evaluation. WES was performed on tumor genomic DNA (gDNA) and corresponding constitutional gDNA was extracted from blood. The bioinformatic analysis identified 34 and 26 somatic coding single nucleotide variants (cSNV) in the tumors. An identical variant in GJA4 (NM_002060.3 c.121G > T; p.Gly41Cys) was identified for the two patients. The substitution was reported as probably damaging by PolyPhen-2 (score = 1), and accordingly as damaging in SIFT prediction (score = 0). For the other cases, we tried to detect this mutation using targeted Sanger sequencing after PCR amplification of the locus on DNA extracted from fixed-formalin paraffin embedded tissues. One of the four tested cases presented the mutation, with valid positive controls. Thereafter, we searched the same mutation by Sanger sequencing in a series of 5 soft tissue angioleiomyomas. All cases tested (4/5) presented the same mutation GJA4 (p.Gly41Cys), for the last case, the technique was not contributive due to insufficient DNA quality.

DNA methylation profiling suggest a distinct epigenetic profile
Using DNA methylation-based classification and the Brain Tumor and Sarcoma Classifiers (version 12.5/12.2; www. molec ularn europ athol ogy. org), only 1/7 tumor was classifiable (calibrated scores for DNA methylation class > 0.9) (see details in Table 2). Next, a t-Distributed Stochastic Neighbor Embedding (t-SNE) analysis was performed alongside the genome-wide DNA methylation profiles from the sarcoma reference cohort [22] as well as a more focused analysis with selected reference groups. Five of the six cases grouped together and showed no obvious relation to any of the other reference classes such as angioleiomyomas/myopericytomas of the soft tissue, angiosarcomas, epithelioid hemangioendotheliomas, solitary fibrous tumors/hemangiopericytomas, inflammatory myofibroblastic tumors or meningiomas (Fig. 4). No significant copy number alterations were observed in our cases on the copy number variation profiles. The MGMT  promotor was unmethylated in all cases and there was a global hypomethylation.

Literature review
Our extensive review of cases from the English-language literature, describing only cases with similar radiological and histopathological features to our cases, found 74 similar radiologically and histologically described cases of DALM . The median age of patients was 46   [26][27][28][29], with only one pediatric case [61]. The male/female ratio was 1.4 (43 males and 31 females). The symptoms depended on tumor location, but in 10% of cases the tumors were incidental. Different tumor locations were described, but most of cases were found in the cavernous sinus (43% of cases) and were radiologically interpreted as meningiomas (78% of cases). After resection, only one case recurred (192 months after the initial surgery) [44] and all patients were alive at the end of follow-up. No genetic or epigenetic characterizations were available for any of these cases.

Discussion
Vascular lesions of the CNS are frequent and represented by intraparenchymal CCM, arteriovenous malformations, venous malformations and telangiectasias. Here, we identified a putative new tumor type characterized by the same dural location, a similar histopathological pattern and a frequent GJA4 p.Gly41Cys mutation (3/6 cases), distinct from genes implicated in CCM (CCM1/2/3, MAP3K3, PIK3CA) and arteriovenous malformations (KRAS) of the brain. Particularly, GJA4 mutations were not found using WES in cavernous malformation cohorts [19]. The GJA4 gene encodes the Gap Junction Protein Alpha 4 (or Con-nexin37), a protein from the gap junctions of endothelial cells [72]. Interestingly, a previous study evidenced that connexin37 knockout mice displayed severe vascular abnormalities (particularly in the testis and intestine), looking like "cavernous hemangiomas" and died perinatally from haemorrhage [73]. More recently, the same GJA4 (p.Gly41Cys) mutation was reported in a subset (as in our work) of a series of vascular lesions located in the liver and the skin [74]. In this study, the vascular lesions were denominated as "hepatic hemangiomas" and "cutaneous venous malformations" but the histopathological data are limited, impeding us from performing a detailed comparison [74]. This recent study showed that this mutation leads to a non-canonic activation of SGK1 (serum/ glucocorticoid-regulated kinase 1) which is implicated in various neoplasms (as breast cancer, hepatocellular carcinoma, glioblastoma, colorectal cancer and non-small cell lung cancer) [75][76][77][78][79]. During embryogenesis, SGK1 is necessary for normal angiogenesis with a prosurvival role in endothelial and vascular smooth muscle cells [80]. In our study, we evidenced that dural cases shared morphological similarities with the cavernous variant of angioleiomyomas. They presented well-differentiated smooth muscle cells stained by h-caldesmon and actin with perivascular concentric arrangements of myoid cells intervening dilated vascular channels of variable thickness [81,82]. The presence of myoid cells arranged circumferentially in layers around the vascular lumina and the absence of expression of desmin may also evoke the diagnosis of myopericytoma, but those features are not specific to this tumor and may be encountered in a subset of angioleiomyomas [82]. Moreover, angioleiomyomas and myopericytomas are now classified in the latest WHO classification, and fall within the same morphological spectrum of the perivascular tumor type [83]. Our dural cases present clinical (affecting adults between the fourth and the sixth decades) [82], and radiological similarities with soft tissue angioleiomyomas (such as hyperintensity on T2-weighted imaging and a strong enhancement after contrast injection) [84]. Furthermore, we showed that the same recurrent substitution (GJA4 p.Gly41Cys) was shared by soft tissue and intracranial ALM, reinforcing their close vicinity. This mutation was probably not described before, due to the lack of comprehensive molecular studies of this frequent (in soft tissue) and benign tumor. Here, we did not find any copy number variations, contrary to other previous soft tissue studies (showing cytogenetic abnormalities including monsonosmy of chromosome 13, and loss of 6p, 21q, and 13q), which however, are lacking a histopathological characterization [85][86][87]. Based on epigenetic profiling, all intracranial tumors clearly clustered together but were different from all other established groups, particularly soft tissue angioleiomyomas and meningiomas. Since DNA methylation profiles are thought to represent a combination of both somatically acquired DNA methylation changes and a signature reflecting the cell of origin [88], it is reasonable to assume that tumors from the dura mater represent a distinct tumor type from soft tissue angioleiomyomas. Based on our center's experience, the prevalence of DALMs seems to be low, representing only 1.5% (3/202 cases) of CNS vascular and perivascular lesions resected over the past 20 years. However, this tumor type is probably misdiagnosed, and variably called "angioleiomyoma", "myopericytoma", "venous hemangioma" and "cavernous hemangioma". Moreover, the terminology of "angioleiomyomas" is probably not widely known and consequently rarely used by neuropathologists. Finally, because the natural history of DALM seems very favourable, a subset of patients is probably not resected, these lesions being radiologically interpreted as meningiomas. In summary, we performed for the first time a comprehensive analysis of a distinctive intradural perivascular tumor type presenting histopathological similarities with soft tissue angioleiomyomas, frequently having GJA4 mutations. Because of its dural location and distinct DNA methylation profile, we suggest the term "dural angioleiomyoma" for this benign tumor.