A total of 72 members from 40 families at six participating hospitals were included in this study. Serum samples from the patients were transferred to Gachon HHT center (Incheon, Korea) for genetic screening of ENG and ACVRL1. The study group included four previously described families (family No. 1, 2, 5, 6) including a total of 27 individuals with known pathogenic variants and clinical symptoms [12,13]. The genetic variant analyses were performed after obtaining informed consent from patients and their family members. All experimental procedures were reviewed and approved by the Institutional Review Board of the Lee Gil Ya Cancer and Diabetes Institute, Gachon University (Incheon, Korea; IRB No. GU1102-001).

Blood samples were collected into vacuum blood collection tubes containing ethylenediaminetetraacetic acid (EDTA). Genomic DNA was isolated from 200 μL of the collected whole blood samples using the Exgene Blood SV mini kit (GeneAll, Seoul, Korea) according to the manufacturer’s protocol.

All exons of ENG and ACVRL1 and their flanking sequences were amplified by polymerase chain reaction (PCR) using PrimeSTAR HS DNA Polymerase (Takara Bio, Shiga, Japan) using appropriate primer sets as previously described [12]. The PCR products, which were separated on a 1%–2% agarose gel, were purified using the QIAquick Gel Extraction Kit (QIAGEN, Hilden, Germany) or the Expin PCR SV Kit (GeneAll). Then, the purified PCR products were subjected to direct sequencing. The GenBank accession numbers for the reference sequences of ENG and ACVRL1 are followings: NC_000009.12 (ENG genomic DNA), NM_000118.3 (ENG cDNA), NP_000109.1 (ENG protein), NC_000012.12 (ACVRL1 genomic DNA), NM_000020.3 (ACVRL1 cDNA), and NP_000011.2 (ACVRL1 protein). Missense variants were analyzed using PolyPhen-2 (http://genetics. bwh.harvard.edu/pph2/) [14], SIFT (https://sift.bii.a-star.edu.sg/) [15], and PROVEAN (http://provean.jcvi.org/index.php) [16], which are in silico predictive programs for predicting the possible impact of an amino acid substitution on a human protein. PolyPhen-2 predicts the substitution impact using straightforward physical and comparative considerations [14]; SIFT uses sequence homology and the physical properties of amino acids [15]; PROVEAN uses an alignment-based score approach [16]. The genome aggregation database (v3.1) (https://gnomad.broadinstitute.org/) was used to obtain the frequencies of variants in large populations. Variants were classified as pathogenic very strong (PVS), pathogenic strong (PS), pathogenic moderate (PM), pathogenic supporting (PP), benign stand-alone (BA), benign strong (BS), and benign supporting (BP), and then the selected criteria were combined according to the scoring rules to identify a pathogenic classification from five categories (pathogenic, likely pathogenic, benign, likely benign, or uncertain significance) based on American College of Medical Genetics and Genomics (ACMG) Standards and Guidelines [17].

We retrospectively reviewed the clinical symptoms, physical examinations, family histories, and clinical histories of 42 patients from 24 families who were positive in the HHT genetic screening (ENG1 or ACVRL1). Patients meeting more than three of the Curaçao criteria were classified as definite HHT, patients meeting two criteria were classified as probable HHT, and patients meeting less than two criteria were classified as unlikely. We analyzed the clinical manifestations according to the Curaçao criteria and the genetic variants in the 42 patients. The clinical manifestations of 30 patients from 16 families who were negative for genetic screening were also reviewed (Fig. 1).

Statistical analyses were carried out using GraphPad Prism ver. 7.03 (GraphPad, San Diego, CA, USA). The clinical presentations of patients between the ENG and ACVRL1 pathogenic variant groups were compared using Fisher’s exact test.

We performed genetic analyses for all family members who agreed to undergo testing. The analysis showed that all symptomatic members in each family had the same genetic variant. Among 72 individuals in 40 families who were referred for genetic screening, a total of 42 had genetic variants that met the pathogenic criteria (PVS, PS, PM, or PP) based on the ACMG Standards and Guidelines [17]. Twenty-six patients in 12 families (50%) had genetic variants in ENG, and 16 patients in 12 families (50%) had genetic variants in ACVRL1 (Tables 1-3). Among the 12 ENG genetic variants, 11 were classified as pathogenic or likely pathogenic, and one was classified as of uncertain significance according to the ACMG Standards and Guidelines [17]. These variants included splicing defects by substitution (3/11), frameshifts by deletion (3/11), frameshifts by duplication (2/11), a frameshift by deletion-insertion (1/11), a nonsense variant by deletion-insertion (1/11), and a translation defect by substitution (1/11) (Table 2). One missense variant by substitution (family 21 c.821C>T; p.Thr274Ile) was found in a definite HHT patient and her daughter who exhibited epistaxis. This variant was also reported in two members of a Danish family who suffered from epistaxis, telangiectasia, and CAVMs [18]. In silico prediction results of this variant were inconsistent. The PolyPhen-2 [14] and PROVEAN [16] programs predicted the variant as being provably damaging/deleterious, while the SIFT [15] program predicted it as being tolerated (Supplementary Table 1). Although the variant was classified as being of uncertain significance according to ACMG Standards and Guidelines [17], we suspect that this variant was pathogenic because (1) at least four definite HHT patients had this variant (PP4), (2) the variant was not detected in the genome aggregation database (PM2), (3) two out of three in silico programs predicted this variant as to be deleterious, and (4) the p.Thr274Ile variant was not detected in ENG homologs. It is of interest to note that the same splicing defect variant (c.360+1G>A) was detected in three unrelated families (1, 20, and 26) (Table 2). There were 12 families with ACVRL1 genetic variants, which were classified as pathogenic (5/12), likely pathogenic (3/12), or of uncertain significance (4/12) [17], including frameshifts by deletion (2/12), frameshifts by duplication (2/12), missense variants by substitution (6/12), a splicing defect by substitution (1/12), and a nonsense variant by substitution (1/12) (Table 3). Among six missense variants, two variants (family 6c.199C>T; p.Arg67Trp and family 10 c.925G>A; p.Gly309Ser) were previously reported in several HHT families in more than five different studies [10] and classified as likely pathogenic [17]. One missense variant (family 32 c.1124A>G; p.Tyr375Cys) that was previously reported in a Chinese patient [19] and three novel missense variants (family 12 c.781G>C; p.Ala261Pro, family 30 c.605T>G; p.Val202Gly, and family 38 c.1005T>G; p.Asn335Lys) were classified as being of uncertain significance [17]. We suspected that the four missense variants were pathogenic because (1) these variants were not detected in the genome aggregation database (PM 2) and ENG homologs and (2) three in silico programs (PolyPhen-2 [14], SIFT [15], and PROVEAN [16]) predicted these variants to be deleterious (PM3) (Supplementary Table 1). A substitution variant (c.1048+5G>A) found in family 40 was reported to cause a splicing defect based on a functional test [20].

Of the 42 patients with HHT genetic variants, 55% were male and their mean age was 53.9 years. Twenty-six (62%) and 16 (38%) patients had ENG and ACVRL1 genetic variants, respectively. Epistaxis was the most common clinical presentation (38/42, 90%), followed by visceral AVMs (24/42, 57%) and telangiectasia (21/42, 50%). Thirty-seven patients (88%) had a family history of HHT. Among the patients with AVMs, 12 patients (29%) had PAVMs, six patients (14%) had CAVMs, four patients (10%) had hepatic AVMs (HAVMs), and two patients (5%) had gastrointestinal tract lesions. Based on the Curaçao criteria, 24 patients (57%) had definite HHT, 17 (41%) had probable HHT, and 1 (2%) were unlikely to have HHT (Table 4).

A comparison between the clinical presentations of the ENG and ACVRL1 genetic variant groups revealed that more patients with ENG genetic variants (65%) met the definite HHT Curaçao criteria than patients with ACVRL1 genetic variants (44%) (Table 5). PAVMs were more frequent in ENG patients than in ACVRL1 patients. Among the 30 patients who were negative in the genetic variant screening, six showed definite clinical manifestations of HHT (Table 6).

A proband was a 49-year-old man (family 16), diagnosed with HHT, who presented with recurrent epistaxis, PAVM, and telangiectasia lesions; he also had a family history of epistaxis and telangiectasia. Telangiectasia lesions were detected on the tongue, lips, and nasal mucosa. He underwent surgery for a brain abscess that was suspected to be a complication associated with PAVM (Fig. 2). His two sons also had epistaxis and telangiectasia lesions and had a history of syncope.