A Multi-Institutional Experience in the Aortic and Arterial Pathology in Individuals with Genetically Conﬁrmed Vascular Ehlers-Danlos Syndrome

From the Society for Clinical Vascular Surgery

A multi-institutional experience in the aortic and arterial pathology in individuals with genetically conﬁrmed vascular Ehlers-Danlos syndrome

Sherene Shalhub, MD, MPH, FACS,a Peter H. Byers, MD,b Kelli L. Hicks, BS,a Kristofer Charlton-Ouw, MD,c Devin Zarkowsky, MD,d Dawn M. Coleman, MD,e Frank M. Davis, MD,e Ellen S. Regalado, MS, CGC,f Giovanni De Caridi, MD, PhD,g K. Nicole Weaver, MD,h Erin M. Miller, MS, LGC,i Marc L. Schermerhorn, MD,j Katie Shean, MD,j Gustavo Oderich, MD,k Mauricio Ribeiro, MD, PhD,l Cole Nishikawa, MD,m Christian-Alexander Behrendt, MD,n E. Sebastian Debus, MD, PhD,o Yskert von Kodolitsch, MD,o Richard J. Powell, MD,p Melanie Pepin, MS, CGC,b Dianna M. Milewicz, MD, PhD,f Peter F. Lawrence, MD,q and Karen Woo, MD, MS,q Seattle, Wash; Houston, Tex; San Francisco, Los Angeles, and Sacramento, Calif; Ann Arbor, Mich; Messina, Italy; Cincinnati, Ohio; Boston, Mass; Rochester, Minn; São Paulo, Brazil; Hamburg, Germany; and Lebanon, NH

ABSTRACT Objective: Vascular Ehlers-Danlos syndrome (vEDS) is a rare connective tissue disorder owing to pathogenic variants in COL3A1 that lead to impaired type III collagen production. We aim to describe the contemporary multi-institutional experience of aortic and arterial pathology in individuals with vEDS, to evaluate disease patterns and refine management recommendations. Methods: This cross-sectional, retrospective study of individuals with genetically confirmed vEDS was conducted between 2000 and 2015 at multiple institutions participating in the Vascular Low Frequency Disease Consortium. Aortic and arterial events including aneurysms, pseudoaneurysms, dissections, fistulae, or ruptures were studied. Demographics, COL3A1 variants, management, and outcomes data were collected and analyzed. Individuals with and without arterial events were compared. Results: Eleven institutions identified 86 individuals with pathogenic variants in COL3A1 (47.7% male, 86% Caucasian; median age, 41 years; interquartile range [IQR], 31.0-49.5 years; 65.1% missense COL3A1 variants). The median follow-up from the time of vEDS diagnosis was 7.5 years (IQR, 3.5-12.0 years). A total of 139 aortic/arterial pathologies were diagnosed in 53 individuals (61.6%; 50.9% male; 88.5% Caucasian; median age, 33 years; IQR, 25.0-42.3 years). The aortic/arterial events presented as an emergency in 52 cases (37.4%). The most commonly affected arteries were the mesenteric arteries (31.7%), followed by cerebrovascular (16.5%), iliac (16.5%), and renal arteries (12.2%). The most common management was medical management. When undertaken, the predominant endovascular interventions were arterial embolization of medium sized arteries (13.4%), followed by stenting (2.5%). Aortic pathology was noted in 17 individuals (32%; 58.8% male; 94.1% Caucasian; median age, 38.5 years; IQR, 30.8-44.7 years). Most notably, four individuals underwent successful abdominal aortic aneurysm repair with excellent results on follow-up. Individuals with missense mutations, in which

From the Division of Vascular Surgery, Department of Surgery,a and Depart- Lebanonp; and the Division of Vascular Surgery, University of California Los ments of Pathology and Medicine (Medical Genetics),b University of Washing- Angeles, Los Angeles.q ton School of Medicine, Seattle; the Department of Cardiothoracic and Author conflict of interest: none. Vascular Surgery,c and Division of Medical Genetics, Department of Internal Supported in part the National Center for Advancing Translational Sciences of Medicine,f University of Texas Health Science Center at Houston, Houston; the National Institutes of Health under Award Number UL1TR000423 (S.S.), in the Division of Vascular and Endovascular Surgery, Department of Surgery, part by funds from the Freudmann Fund for Translational Research in Ehlers- University of California San Francisco, San Franciscod; the Section of Vascular Danlos syndrome at the University of Washington (P.H.B.), and in part by the Surgery, Department of Surgery, University of Michigan, Ann Arbore; the National Institutes of Health (NIDDK 1K08DK107934) (K.W.). The content is Department of Cardiovascular and Thoracic Sciences, University of Messina, solely the responsibility of the authors and does not necessarily represent Messinag; the Division of Human Genetics, Cincinnati Children’s Hospital the official views of the National Institutes of Health. Medical Center,h and the Divisions of Cardiology and Human Genetics, Uni- Presented at the Forty-fifth Annual Symposium of the Society for Clinical versity of Cincinnati School of Medicine and Cincinnati Children’s Hospital Vascular Surgery, Las Vegas, Nev, March 18-22, 2017. Medical Center,i Cincinnati; the Division of Vascular and Endovascular Sur- Correspondence: Sherene Shalhub, MD, MPH, FACS, Division of Vascular Sur- gery, Beth Israel Deaconess Medical Center, Bostonj; the Division of Vascular gery, Department of General Surgery, University of Washington School of Surgery, Mayo Clinic, Rochesterk; the Division of Vascular and Endovascular Medicine, 1959 N.E. Pacific ST, Box 356410, Seattle, WA 98195 (e-mail: Surgery, Department of Surgery and Anatomy, Medical School of Ribeirão [email protected]). Preto, University of São Paulo, São Paulol; the Department of Surgery, Univer- The editors and reviewers of this article have no relevant financial relationships to sity of California, Davis Medical Center, Sacramentom; the Department of disclose per the JVS policy that requires reviewers to decline review of any Vascular Medicine,n and Department of Cardiology,o University Heart Center manuscript for which they may have a conflict of interest. Hamburg, University Medical Center Hamburg-Eppendorf, Hamburg; the 0741-5214 Division of Vascular Surgery, Dartmouth-Hitchcock Medical Center, Copyright Ó 2019 by the Society for Vascular Surgery. Published by Elsevier Inc. https://doi.org/10.1016/j.jvs.2019.01.069

1 2 Shalhub et al Journal of Vascular Surgery --- 2019

glycine was substituted with a large amino acid, had an earlier onset of aortic/arterial pathology (median age, 30 years; IQR, 23.5-37 years) compared with the other pathogenic COL3A1 variants (median age, 36 years; IQR, 29.5-44.8 years; P ¼ .065). There were 12 deaths (22.6%) at a median age of 36 years (IQR, 28-51 years). Conclusions: Most of the vEDS arterial manifestations were managed medically in this cohort. When intervention is required for an enlarging aneurysm or rupture, embolization, and less frequently stenting, seem to be well-tolerated. Open repair of abdominal aortic aneurysm seems to be as well-tolerated as in those without vEDS; vEDS should not be a deterrent to offering an operation. Future work to elucidate the role of surgical interventions and reﬁne management recommendations in the context of patient centered outcomes is warranted. (J Vasc Surg 2019;-:1-12.) Keywords: Vascular Ehlers-Danlos syndrome; COL3A1 mutation; Arterial dissection; Arterial aneurysm; Arterial rupture

Vascular Ehlers-Danlos syndrome (vEDS) is a rare syn- approval. The institutional review boards waived the drome in which type III collagen production is reduced patient consent process owing to minimal patient risk. or the collagen produced is defective because of Data were collected by each institute’s respective autosomal-dominant mutations in COL3A1.1-3 The investigator(s), deidentified, and then submitted and syndrome, which was previously called Ehlers-Danlos stored using a password-encrypted database main- syndrome type IV, is 1 of 13 subtypes of Ehlers-Danlos syn- tained by the University of Washington. drome.4 In addition to spontaneous intestinal perforation Identification of individuals with vEDS and inclusion or rupture of a gravid uterus, the hallmark of the disease and exclusion criteria. Individuals were initially identified is spontaneous arterial dissections, aneurysms, and with International Classification of Diseases-9-CM code rupture at a young age.1,2,5 756.83 or International Classification of Diseases-10-CM Up to 40% of individuals with vEDS experience their code Q79.6 for Ehlers-Danlos syndrome. Confirmatory first major arterial complication by the age of 40.2,6 The molecular testing results were then reviewed by a classically reported arteries include mesenteric, renal, geneticist (P.H.B.) to confirm that the COL3A1 variant is a iliac, femoral, and/or the abdominal aorta, followed by pathogenic variant in keeping with the ACMG guide- the carotid, subclavian, ulnar, popliteal, and tibial ar- lines.2,9,10 Individuals were included for analysis only if teries.7 A spontaneous carotid-cavernous fistula (CCF) is they had a pathogenic variant in COL3A1. The pathogenic pathognomonic of vEDS and estimated to occur in COL3A1 variants were grouped into missense mutations 9.8% of individuals with vEDS.8 (glycine substations), exon skip and splice site variants, The predominant challenges to studying vEDS are and haploinsufficiency (null mutations).6,9-11,14-16 The type driven by the rarity of the disease, the heterogeneous of amino acid substitution was noted as a large or small presentation of aortic and arterial pathology, the large amino acid.6,15 number of pathogenic variants in COL3A1 leading to Demographics, current age, age at diagnosis, family vEDS, a lack of robust longitudinal data, and underdiag- history, clinical diagnostic criteria,17 aortic and arterial nosis or misdiagnosis. Although we have a basic under- pathology, management of aortic and arterial pathology, standing of the natural history of the disease based on and outcomes were collected. Family history was seminal work2 and a limited understanding of the defined as a family history of vEDS, aortic or arterial aneu- genotype-phenotype correlation,6,9-11 we do not have a rysms and dissections, and/or sudden death. Arterial detailed natural history of aortic and arterial aneurysms pathology included aortic and arterial aneurysms, dissec- and dissections in this population, nor do we have a clear tions, pseudoaneurysms, fistulae, thrombosis, or ruptures. understanding of the risk of complications, once diag- Arterial pathology was noted as emergent if it was life nosed and treated.12 The aim of this study was to threating on presentation. This category included aortic describe the contemporary multi-institutional experi- or arterial rupture, symptomatic aortic/arterial dissec- ence of aortic and arterial pathology in individuals with tions, and CCF. vEDS, evaluate disease patterns, and refine management Given that subject data were collected locally at each recommendations to improve our understanding of participating institution and then submitted as genotype-phenotype correlations. de-identified data to the consortium, a comparison of METHODS all the presentations, COL3A1 variants, and demographics The Vascular Low Frequency Disease Consortium. This was performed to ensure that there were no duplicated multi-institutional retrospective cross-sectional cohort cases. Data were analyzed using Microsoft Excel 2007 study of individuals diagnosed with vEDS was conducted software (Microsoft, Redmond, Wash) and SPSS version between January 1, 2000, and December 31, 2015. The 11 19 for Windows (SPSS, Inc., Chicago, Ill). Continuous institutions were recruited through the Vascular Low data are presented as medians and interquartile ranges Frequency Disease Consortium (University of California- (IQRs).Continuous data were compared using Wilcoxon Los Angeles Division of Vascular Surgery).13 Each partici- rank-sum (Mann-Whitney) test. Categorical data were pating center obtained its own institutional review board compered by a c2 or Fisher’s exact test where Journal of Vascular Surgery Shalhub et al 3 Volume -, Number -

Table I. A comparison of individuals with vascular Ehlers- Danlos syndrome (vEDS) with and without a diagnosis of ARTICLE HIGHLIGHTS aortic/arterial pathology d Type of Research: Cross-sectional retrospective No aortic/ study of the Vascular Low Frequency Disease Aortic/Arterial arterial Consortium pathology pathology d (n ¼ 53) (n ¼ 33) P value Key Findings: In this group of 86 individuals with genetically confirmed vascular Ehlers-Danlos Current age 41 [31-49.5] 25 [15-41.5] <.001 syndrome, with pathogenic COL3A1 variants, most Age range 19-79 1-88 d patients were managed medically. For treatment of < Age at diagnosis 32 [23-43.3] 18 [10-31] .001 an enlarging aneurysm or rupture, embolization Male sex 27 (50.9) 14 (42.4) .442 and stenting were well-tolerated. Four patients Caucasian 47 (88.7) 27 (81.8) .491 underwent successful open abdominal aortic aneu- BMI 24.6 22.9 .068 rysm repair. [21.6-27.4] [17.1-25.3] d Take Home Message: Genetic confirmation of path- Hypertension 17 (32.1) 2 (6.1) .005 ogenic COL3A1 variants of vascular Ehlers-Danlos syn- Deep vein thrombosis 12 (22.6) 1 (3) .014 drome is essential for counseling affected individuals Intestinal perforation 9 (17) 5 (15.2) .823 on the effect of their variant type and directing care. Spontaneous PTX/HTX 7 (13.2) 5 (15.2) .800 When intervention is required, embolization and Current or past smoker 12 (22.6) 5 (15.2) .396 stenting are acceptable options. Open repair of Family history of VEDS 24 (45.3) 18 (54.5) .403 abdominal aortic aneurysms is also well-tolerated. Follow-up after vEDS 7.5 [3.5-12] 5 [1-11] .152 diagnosis included 19 individuals (22.1%) who were diagnosed as Died 13 (24.5) 0 .002 children (age <18 years old). The median follow-up Age at death 39 [27.8-51.3] dd from the time of vEDS diagnosis was 7. years 5 (IQR, BMI, Body mass index; HTX, hemothorax; PTX, pneumothorax; vEDS, vascular Ehlers-Danlos syndrome. 3.5-12.0 years). Values are number (%) or median [interquartile range]. A total of 139 aortic/arterial pathologies were diagnosed in 53 individuals (61.6%; 50.9% male; 88.5% Caucasian; median age 33, years; IQR, 25.0-42.3 years). The aortic/ arterial events presented as an emergency in 52 cases (37.4%). The diagnosis of vEDS was already established in 20 individuals (37.3%) before the diagnosis of aortic/arterial pathology. The diagnosis of vEDS was less likely be known in the emergent setting compared with the elective stetting (18.2% vs 55.2%; P ¼ .007). There were no differences between men and women in the age of the initial aortic/arterial pathology diagnosis (median, 32 years [IQR, 23-42 years] among men; median, 36 years [IQR, 26.8-43.3 years] among women; P ¼ .393), or the time of first aortic/arterial rupture (median, 33 years [IQR. 15.5-49.0 years] among men; median, ¼ Fig 1. Kaplan-Meier estimates of cumulative survival free 39 years [IQR, 30.0-46.7 years] among women; P .540). of any arterial pathology in a cohort of individuals with The individuals with aortic/arterial pathology were signif- vascular Ehlers-Danlos syndrome (vEDS). icantly older than those without (Table I). In addition, the group without aortic/arterial pathology included 14 children compared with only one child who presented appropriate. The comparisons of the onset of the arterial with type B aortic dissection at 12 years of age (had pathology and survival were performed using Kaplan- missense mutation, p.Gly244Arg). Those patients with Meier survival curves with the log-rank test. All statistical aortic/arterial pathology had more hypertension and tests were two-sided and a P value of less than .05 was considered statistically significant. deep vein thrombosis. There were no differences in the type of pathogenic COL3A1 variants or minor clinical RESULTS diagnostic criteria between those with and without Eighty-six individuals had molecular confirmation of aortic/arterial pathology, with the exception of a vEDS (47.7% male; 86% Caucasian; median age, 41 years; lower frequency of hypermobile small joints (32.1% vs IQR, 31.0-49.5 years; range, 1-88 years). The cohort 54.5%; P ¼ .039). 4 Shalhub et al Journal of Vascular Surgery --- 2019

Table II. Presentation and management of arterial pathology in patients with vascular Ehlers-Danlos syndrome (vEDS)by artery involved

Arteries, No. (%) Male, % Median age (IQR) Known vEDS diagnosis Missense mutation Dissection Carotid, 17 (12.2) 5 (41.7) 27 (24-36) 4 (33.3) 8 (66.7) 6 (50) Vertebral, 6 (4.3) 2 (33.3) 27.5 (22-32.5) 3 (50) 5 (83.3) 4 (66.6) Celiac, 14 (10.5) 6 (40) 44 (33-50) 12 (80) 11 (73.3) 7 (46.7) Gastric, 3 (2.2) 0 37 (36-37) 2 (66.7) 2 (66.7) 1 (33.3) Phrenic, 1 (0.7) 0 58 1 (100) 1 (100) 0 Splenic, 11 (7.9) 3 (27.3) 44 (31-51) 6 (54.5) 6 (54.5) 2 (18.2) Hepatic, 8 (5.8) 5 (62.5) 47 (38.5-50.5) 5 (62.5) 4 (50) 0 Superior mesenteric, 6 (4.3) 1 (16.7) 45.5 (39-54.4) 5 (83.3) 3 (50) 2 (33.3) Renal, 17 (12.2) 12 (70.6) 37 (30.5-41) 9 (52.9) 13 (76.5) 7 (41.2) Iliac, 23 (16.5) 18 (78.3) 41 (30-48) 12 (52.2) 19 (82.6) 7 (30.4) Femoral, 2 (1.4) 1 (50) 27 (33) 0 2 (100) 1 Popliteal, 1 (0.7) 0 26 0 1 (100) 0 Posterior tibial, 3 (2.2)b 2 (100) 19 (32) 1 (50) 1 (50) 0 IQR, Interquartile range. The percent given is the percentage of all arterial pathology. aSame individual, death owing to multiple mesenteric arterial ruptures. bOne individual had bilateral posterior tibial arteries aneurysms.

Arterial pathology. The most commonly affected ar- Aneurysm sizes were recorded in four cases (0.8, 0.8, teries were the mesenteric arteries (31.7%), followed by 1.2, and 3 cm). The predominant approach to treat- cerebrovascular and iliac arteries (16.5% each), and renal ment was medical management (Table II), with the arteries (12.2%). Arterial pathology included aneurysms exception of two patients requiring intervention: a (53.8%), dissections (35.3%), rupture (10.1%), pseudoaneur- 40-year-old man who underwent embolization (no additional detail) and a 29-year-old man (c.1124G>A, ysms (3.4%), thrombosis (2.5%), and CCF (4.2%). Fig 1 p.Gly375Glu), who presented with left renal artery shows the Kaplan-Meier estimates of cumulative sur- thrombosis that was treated with thrombolysis. This vival free of any arterial pathology. Table II summarizes approach was complicated by splenic hemorrhage the presentation and management of the arterial requiring a splenectomy. He presented 5 months later pathologies. with aneurysmal degeneration of the renal artery and d Carotid cavernous ﬁstulae. CCF occurred in four indi- underwent a successful nephrectomy (6 years of viduals (Table III). Management was predominantly follow-up). via embolization with satisfactory outcomes (Fig 2). d Iliac arteries. Twenty-three common, external, and None of the CCFs were associated with mortality. internal iliac arteries had pathology, which was diag- d Carotid and vertebral pathology other than CCF. These nosed as isolated iliac disease in nine (16.9%), and in patients presented with small aneurysms or dissec- association with abdominal aortic aneurysm (AAA) in tions. All were managed medically, as detailed in eight (15.1%) individuals (Table II). Most were managed Table II. Most had no complications, with the exception medically. Stenting was performed in two cases of one patient with a vertebral artery dissection leading (c.3847C>T, p.Gln1283Ter and c.3320G>A, Gly107Glu) to a lateral medullary infarct and death. for common iliac artery aneurysms, using femoral ar- d Mesenteric arteries. The celiac artery was the most tery exposure and stent repair, with satisfactory results. commonly affected mesenteric artery (Fig 3); however, Open iliac repair was performed in three cases. the splenic artery was most frequently affected by B For an iliac artery rupture in a 50-year-old man rupture (36.4%). Medical management was the most (c.1295G>A, p.Gly432Asp), which was complicated common approach (Table II), with the exception of with postoperative wound dehiscence with the cases in which rupture or pseudoaneurysms occurred. development of an enterocutaneous ﬁstula. These cases were treated most commonly with endo- B For an iliac artery dissection with thrombosis treated vascular embolization, and in one case with stenting with an uncomplicated iliofemoral bypass in a (Fig 4), with satisfactory results. There was one mortality 30-year-old man (c.674G>C, p.Gly225Ala). The patient in this group (a patient who presented with multiple has had 6 years of follow-up after the procedure mesenteric arterial ruptures). without complication. d Renal arteries. Renal arteries presented with nearly B For an iliac artery aneurysm treated with an uncom- equal frequency of dissections and aneurysms. plicated open bypass in a 41-year-old man Journal of Vascular Surgery Shalhub et al 5 Volume -, Number -

Table II. Continued. Aneurysm, pseudoaneurysm Rupture Medical management Endovascular embolization Endovascular stenting Open repair 5 (41.7), 1 (8.3) 0 100% 0 0 0 1 (16.7) 0 100% 0 0 0 7 (46.7), 2 (13.3) 2 (13.3) 13 (86.7) 1 (6.7)a 0 1 (6.7) 0 0 3 (100) 0 0 0 1 (100) 0 1 (100) 0 0 0 8 (72.7) 4 (36.4) 4 (36.4) 6 (54.5) 0 1 (9.4) 7 (87.5), 1 (12.5) 1 (12.5) 5 (62.5) 2 (25) 1 (12.5) (Fig 3)0 2 (33.3) 1 (16.7) 5 (83.3) 1 (16.7)a 00 8 (47.1), 1 (5.9) 0 14 (82.4) 1 (5.9) 0 2 (11.8) 16 (69.6) 1 (4.3) 17 (73.9) 1 (4.2) 2 (8.7) 3 (13) 10 1 0 0 1 1 0 1 (100) 0 0 0 30 2 0 0 1

Table III. Presentation and management of carotid cavernous ﬁstulae in patients with vascular Ehlers-Danlos syndrome (vEDS) Length of stay, days, and Age/Sex COL3A1 variant Presentation Management, access site disposition 49F c.2131G>A, p. Gly544Ser Rupture Embolization, Percutaneous, 7, home, alive at age 63; (missense) 5F, manual pressure closure 14 years follow-up 51M IVS8þ5G>A (Exon skip) N/A Embolization, percutaneous, 3, home, alive at age 70 5F, manual pressure closure 41F c.2337þ2T>C, Ipsilateral retro-orbital Embolization, percutaneous, 4, home; <1 year of p.Gly762_Lys779del pain, blurry vision, 6F, closure with angioseal follow-up; died from (splice site) and tinnitus (Fig 3) unrelated complications 42F Pathogenic Bilateral Nonoperative N/A F, Female; M, male; N/A, not applicable.

(c.2356G>A, p.Gly786Arg). Of note, this individual had middle cerebral artery dissection. The dissection was a prior open repair for a ruptured AAA repair at age managed medically and she is alive at age 43. 33. The patient has had 10 years of follow-up after d Coronary arteries: There were two cases (3.8%) of coro- the procedure without complication. nary artery dissection. One occurred in a 45-year-old d Posterior tibial arteries: Two patients had posterior woman (c.4360C>T, p.Gln1454Ter) who was initially tibial artery aneurysms (3.8%); one was managed medi- managed medically, but subsequently underwent coro- cally. The other individual was a 19-year-old man with nary artery bypass. She has had 1 year of follow-up bilateral posterior tibial artery aneurysms (1.5 and without complications. The second case was in a 2.0 cm). Open repair with saphenous vein bypass graft 41-year-old woman (c.2337þ2T>C, p.Gly762_Lys779del) was performed for the larger aneurysm. This procedure who was hospitalized for a spontaneous perforation of was complicated by hematoma owing to vein graft the colon. She developed a left anterior descending cor- disruption, requiring reoperation. onary artery dissection and subsequent cardiac arrest.

Other rare presentations included involvement of the Endovascular procedures for medium sized arteries. cerebral and coronary arteries as follows. Embolization of medium sized arteries was the predomi- d Cerebral arteries. Only one individual was affected nant endovascular procedure performed (n ¼ 16 [13.4%]), (1.9%). This was a 39-year-old woman (haploinsufﬁ- followed by stenting (n ¼ 3 [2.5%]) as described previously. ciency/null mutation) who presented with a left Percutaneous access was used in 14 cases for sheath sizes 6 Shalhub et al Journal of Vascular Surgery --- 2019

Fig 2. The angiographic ﬁndings of a carotid cavernous ﬁstula in a 41-year-old woman with vascular Ehlers-Danlos syndrome (vEDS) owing to a splice site mutation (c.2337þ2T>C/p.Gly762_Lys779del) presenting with sudden onset ipsilateral retro-orbital pain, blurry vision, and tinnitus. A, Before embolization. B, After coil embolization. RCCA, Right common carotid artery; RICA, right common internal artery.

Fig 3. Celiac artery dissection in 36-year-old woman with vascular Ehlers-Danlos syndrome (vEDS) owing to a splice site mutation (c.2337þ2T>C/p.Gly762_Lys779del) managed medically.A,Axial computed tomography imaging obtained 1 month before the dissection when she presented with a spontaneous hemoperitoneum managed medically. B, Axial computed tomography imaging demonstrating focal dissection and mild enlargement of the celiac artery. This remained unchanged on follow-up imaging over the next 5 years. C, Axial imaging demonstrating a large spontaneous subcapsular splenic hematoma and hemoperitoneum. This occurred after a complicated hospitalization for perforated sigmoid diverticulitis requiring sigmoid resection, transverse colostomy, and Hartmann’s pouch. of 4F to 7F. Manual pressure after the procedure was Tokyo, Japan; n ¼ 2), StarClose (Abbott Vascular, Santa used in seven cases (four in which the vEDS diagnosis Clara, Calif; n ¼ 1), Perclose ProGlide vascular closure was not known at the time); the closure was satisfactory device (Abbott Vascular; n ¼ 1). All were reported to be in all seven cases, with one (14.3%) hematoma reported. successful. Open femoral artery exposure and repair A closure device was used in ﬁve cases: Angio-Seal was performed in an additional three cases, with one vascular closure device (Terumo Medical Corporation, complicated by hematoma. Journal of Vascular Surgery Shalhub et al 7 Volume -, Number -

Aortic rupture occurred in seven individuals (41.4%) and was the cause of death in two individuals (11.7%): a ruptured type A aortic dissection and a ruptured type B aortic dissection complicated by infectious aortitis.

Mortality. The median follow-up after the ﬁrst arterial event was 5 years (IQR, 2.5-12.0 years). There were 12 deaths (22.6%) at a median age of 36 years (IQR, 28- 51 years; 58.3% male; 83.3% Caucasian). Table V summarizes the characteristics and causes of mortality and Fig 5 shows the Kaplan-Meier estimates of cumulative survival.

Genotype-phenotype correlation. The molecular confirmation in this cohort was performed by genetic testing, showing a pathogenic variant in COL3A1 (n ¼ 81) or by skin biopsy (n ¼ 5). Missense mutations with a substitution of glycine with a large amino acid were the most common type of pathogenic variant (n ¼ 44 [51.2%]) followed by missense mutations with a substitution of glycine residue with a small amino acid (n ¼ 12 [14%]). Null mutations occurred in only seven individuals (8.1%). Although not statistically significant, individuals with missense mutations in which glycine was substituted with a large amino acid had an earlier onset of aortic/ arterial pathology (median, 30 years; IQR, 23.5-37.0 years) compared with the other pathogenic COL3A1 variants Fig 4. Angiogram showing a proper hepatic artery pseu- (median, 36 years; IQR, 29.5-44.8 years; P ¼ .065). There doaneurysm in a 49-year-old man with vascular Ehlers- were no differences between the variant groups in terms Danlos syndrome (vEDS) owing to a haploinsufficiency of arterial/aortic pathology or rupture. mutation (A). The pseudoaneurysm was excluded with 5.0-mm 2.5-cm Viabahn stent (B and C). DISCUSSION A substantial challenge to evaluating treatment of arterial/aortic pathology in individuals with vEDS is the rare Aortic pathology. Aortic pathology was identified in 17 frequency of the disease (1:50,000), such that few centers individuals (32%; 58.8% male; 94.1% Caucasian; median have any significant experience.5,11,12,18 This multi- age, 38.5 years; IQR, 30.8-44.7 years). Hypertension was institutional study is a step in the direction of better un- noted in 52.9% (n ¼ 9) and smoking in 29.4% (n ¼ 5) of derstanding the natural history in vEDS and manage- the cases. The most common COL3A1 variant was a ment outcomes. missense mutation with glycine substitution with a large Several generalizations can be made based on this amino acid residue (n ¼ 11 [64.7%]). cohort. Most of the carotid (other than CCF), vertebral, Thoracic aortic pathology was reported in 10 individuals mesenteric, and renal manifestations of the syndrome and abdominal aortic pathology was reported in 7 can be managed medically. When management of arte- individuals (Table IV). Notably, there were two thoracic rial manifestations is required owing to spontaneous endovascular aneurysm repairs (TEVAR) performed for rupture, enlarging aneurysms, or pseudoaneurysms, an descending thoracic aortic aneurysm ruptures in which endovascular approach, mostly with coil embolization, the diagnosis of vEDS was unknown at the time. One and less frequently with stenting, seems to be well- individual died and the other had a successful repair tolerated and these results are consistent with previous (Table IV). In one case, an open thoracoabdominal aortic reports.18,19 Access site management in this series had aneurysm (TAAA extent II) was performed successfully in few complications, including those managed with a 19-year-old man with a missense mutation (Table IV). manual compression and closure devices. This finding Among the AAA cases, four individuals underwent suc- is to be interpreted with caution, because we usually cessful open AAA repair with excellent long-term follow- recommend open femoral artery exposure and primary up (three had a ruptured AAA; Table IV). In one case, an repair as previously described18,20 for access puncture endovascular aneurysm repair (EVAR) was performed for sites in patients with known vEDS diagnosis, because aortic dissection before diagnosing vEDS. This patient this method allows for the greatest control of the artery. had no short-term complications, with 1 year of follow-up. Although thrombolysis for arterial thrombosis was 8 Shalhub et al Journal of Vascular Surgery --- 2019

Table IV. Presentation and management of aortic pathology in 17 individuals with vascular Ehlers-Danlos syndrome (vEDS) Known Age/sex COL3A1 varianta Location/type Management vEDS diagnosis Outcomes Thoracic aorta (n ¼ 10) 33M c.601G>C, p.Gly201Arg Thoracic aneurysm Medical Yes Died at age 51 from a stroke 44F c.970G>A, p.Gly324Ser ATA aneurysm Medical Yes Alive at age 48 37F c.764G>A,p.Gly255Glu Ruptured type A Medical Yes Cause of death aortic dissection 79M c.3320G>A, p.Gly1107Glu Arch PAU Medical Yes d 40F c.2222G>A, p.Gly574Asp DTAA Medical Yes Diameter 4.2 cm; alive at age 41 53M c.3966delG, TBAD/rupture Medical Yes Cause of death, ruptured p.Lys1323Argfs*64 (null) owing to aortitis 40M c.926G>A, p.Gly142Glu TBAD, mild ATA Medical No Alive at age 49 dilation 12M c.547G>C, p.Gly183Arg DTAA rupture TEVAR No Discharged home after 2 days; died at age 27 from multiorgan failure 27F c.1024G>A, p.Gly342Arg DTAA rupture TEVAR No Cause of death 19M c.1231G>C, p.Gly244Arg TBAD Open repair Yes Extent II TAAA repair at age 21; discharged home after 10 days; alive at age 29 Abdominal aorta (n ¼ 7) 33F c.996þ2G>A, Infrarenal abdominal Medical No Aortic diameter 1.9 cm; alive p.Gly318_Pro332del aortic dissection at age 34 (splice site) 30F c.1330G>A, p.Gly444Arg Abdominal aortic Medical Yes d dissection 45M Pathogenic AAA/dissection EVAR No Alive at age 46 27M c.2113G>A, p.Gly705Arg AAA/dissection Open repair No Discharged home after 8 days; alive at age 31 33M c.2356G>A, p.Gly786Arg Ruptured AAA Open repair No Takeback for a ruptured gallbladder; alive at age 51 43F Pathogenic Ruptured AAA Open repair No Discharged to rehab after 24 days, wound complicated by infection; alive at age 48 48M c.3847C>T, p.Gln1283Ter Ruptured AAA Open repair Yes Take back right limb ischemia; discharged home after 7 days; alive at age 56 AAA, Abdominal aortic aneurysm; ATA, ascending thoracic aorta; DTAA, descending thoracic aortic aneurysm; EVAR, endovascular aneurysm repair; F, female; M, male; PAU, penetrating aortic ulcer; TBAD, type B aortic dissection; TEVAR, thoracic endovascular aneurysm repair. aAll are missense mutations unless specifically noted. performed in one case in this series, it was associated of an aneurysmal iliac artery has been reported to be with a spontaneous hemorrhagic complication. Given successful.23 the risk of spontaneous hemorrhage, we do not recom- Several observations can be made in relation to aortic mend thrombolysis in individuals with vEDS. pathology in this cohort. First, aortic disease seems to CCF embolization seems to offer satisfactory results. be more frequent in thoracic segment than in the This finding confirms previous reports; a referral to a neu- abdominal differently that the observed frequency in rointerventionalist is recommended in these cases.21,22 the general population. Given the small numbers, it is Although we did not find a consistent approach for difficult to draw any additional conclusions from this embolization in these cases, the transvenous approach observation, but it is worth noting with plans for further seems to offer a decreased risk of vascular injury.21,22 evaluation. Second, an open repair of ruptured AAA Iliac arteries dissections without thrombosis can be seems to be as well-tolerated as those without vEDS. managed medically, whereas aneurysm repairs are Therefore, vEDS should not be a deterrent to offering based on association with an AAA. Endovascular stenting an operation for a ruptured AAA. We recommend Teflon Journal of Vascular Surgery Shalhub et al 9 Volume -, Number -

Table V. Age and causes of death among 12 individuals with vascular Ehlers-Danlos syndrome (vEDS) Age Colon Arterial Arterial Age/sex Col3A1 variant Diagnosis Pneumothorax perforation pathology rupture Cause of death 19M Pathogenic 19 dd19 19 Myocardial infarction with associated ventricular rupture 27F c.1024G>A, p.Gly342Arg 17 23 d 27 27 Hemothorax 28F c.2069G>T, p.Gly690Val 23 dd23 26 Lateral medullary infarct owing to vertebral artery dissection 29M c.547G>C, p.Gly183Arg 12 dd12 12 Multiple organ failure 36F c.764G>A, p.Gly255Glu 36 dd36 36 Ruptured type A aortic dissection 41F c.2337þ2T>C, 24 d 36 36 41 Cardiac arrest owing to IVS34þ2T>C, left anterior p.Gly762_Lys779del descending coronary artery dissection 50F Pathogenic 32 33 46 38 46 Multiple visceral arteries ruptures 51M c.601G>C, p.Gly201Arg 29 25 d 33 d Stroke 53M c.3966delG, 43 dd36 53 Ruptured type B aortic p.Lys1323Argfs*64 dissection owing to infectious aortitis 28M c.2445þ2dupT, 24 22 d 22 d Unknown IVS37þ2dupT, p.Gly798_Pro815del 52M c.1295G>A, p.Gly432Asp 51 dd40 50 Unknown 70M IVS8þ5G>A, 58 d 50 51 d Unknown p.Gly195_Ser212del F, Female; M, male.

in an individual with vEDS. However, we recommend against the use of TEVAR or EVAR in individuals with vEDS, based on extrapolating from other connective tissues disorders such as Marfan syndrome. TEVAR and EVAR in this population carry a significant risk of use perforation and erosion at the fixation zones, owing to the fragility of the aortic wall, and, with TEVAR, the risk for retrograde aortic dissection.18,25 An interesting finding was the observation of a higher frequency of hypertension and deep vein thrombosis in the patients with vEDS and arterial pathology. We were not able to ascertain if these diagnoses occurred before or after the diagnosis of the aortic/arterial pathology. Fig 5. Kaplan-Meier estimates of cumulative survival in We also did not ascertain if the affected individuals individuals with vascular Ehlers-Danlos syndrome (vEDS). were taking an antihypertensive. Another interesting finding was the significantly higher frequency of small joint hypermobility in the individuals who did not have or felt reinforcements for the anastomoses in all aortic arterial pathology, but the significance of this finding is repairs. Third, two TEVARs were reported in this cohort unclear. (one associated with death) and one EVAR with a short We did not find sex differences at initial arterial presen- postoperative follow-up duration. These repairs were tation, similar to what has been previously reported.6 performed before knowing the vEDS diagnosis. There Other studies have shown an increased risk of sudden has been a single case report of successful of TEVAR24 death related to arterial ruptures in males younger 10 Shalhub et al Journal of Vascular Surgery --- 2019 than 20 years.10 A plausible explanation that unites these care, and pregnancy planning. These consequences disparate findings is the bias toward those who survived are relevant to family members as well. to have a diagnosis. Once the diagnosis is confirmed, we recommend Similar to prior studies, the missense mutations establishing a multidisciplinary care team locally and (glycine substitutions) in COL3A1 were over-represented at a tertiary care center to coordinate care including in this cohort. Missense mutations are the most com- diagnostic testing, surveillance, and surgical repairs.12 mon type of variant affecting COL3A1. The variant results The care team should include a vascular surgeon, a in a substitution of a glycine residue in the Gly-X-Y geneticist, and a primary care provider who organizes repeats of the triple helical domain, thus disrupting the the variable levels of care needed by the patient. Base- type III collagen folding. As a consequence a minimal line arterial imaging should be performed during the amount of normal collagen (10%-15%) is excreted from initial consultation visits. There are no trials delineating the cell into the extracellular matrix.14 Similarly, exon the frequency of surveillance imaging, nor did we skip and splice site variants create a frameshift that ascertain the frequency of imaging in this study. In results in exon(s) deletion and production of defective general, we recommend that surveillance imaging be collagen similar to the effect of pathogenic missense tailored to the symptoms and presentation of each variants.6,16 The biologic explanation is related to the individual. disruptive effect this type of variant has on the type III This study has several limitations. We did not evaluate collagen folding process.6,10 Although haploinsuffi- the role of medical management in mitigating the risk ciency/null mutations lead to the creation of a prema- of aortic/arterial pathology, using beta-blockers, losar- ture termination codon, the affected gene is essentially tan, vitamin C, or adjunctive measures during open silent, resulting in the production of one-half of the repairs such as the use of desmopressin.5,27-29 We amount of normal type III collagen. Thus, missense vari- were also unable to ascertain if the diagnosis of aneu- ants cause a more severe phenotype of vEDS,10 whereas rysms and dissections was incidental, made because haploinsufficiency mutations lead to milder pheno- of surveillance, or because of symptoms. Additionally, type.6,9-11 In our cohort, individuals with missense muta- the retrospective nature of this study limits our under- tions in which the substitution of glycine is with a large standing of the decision making used in management amino acid seemed to present with aortic/arterial or the indication for the elective operations beyond pathology an at earlier age, compared with the other what has been presented in the results. Last, the imag- types of variants, including missense mutations with ing for all the individuals in this cohort was not avail- substitution of glycine with a small amino acid. This able to review in a standardized manner. Review of finding is plausible; prior work demonstrated an imaging is highly relevant to understanding the natural increased disruption of the collagen triple helix when history of vEDS and should be included in future Gly is replaced by a large amino acid rather than by a study designs. smaller amino acid.6,15 However, our study was under- Future directions include work within the vEDS powered to detect a statistically significant difference Research Collaborative to enroll individuals with geneti- and, therefore, replication in a larger cohort is warranted. cally confirmed vEDS into a large natural history study.30 We recommend an increased index of suspicion at the This study is necessary to obtain detailed longitudinal time of initial presentation of aneurysms and dissec- data to better characterize the natural history and out- tions in young individuals, especially those with a family comes to guide future management recommendations. of aneurysms/dissections and the presence of clinical diagnostic features of vEDS. This finding, then, should CONCLUSIONS lead to confirmatory genetic testing rather than relying Most of the arterial manifestations of vEDS were on clinical criteria alone for a vEDS diagnosis, given the managed medically in this cohort. When intervention is overlap in clinical features with other forms of Ehlers- required owing to spontaneous rupture or enlarging Danlos syndrome and other genetically triggered aorto- aneurysms, embolization (and less frequently stenting) pathies.4,6 Additionally, understanding the type of seems to be well-tolerated. Open repair AAA was toler- COL3A1 variant allows for counseling on the effect of ated and as such, the diagnosis of vEDS should not be the variant type on an individual.6,9-11,26 Moreover, indi- a deterrent to offering an operation. Future directions viduals with an established diagnosis preoperatively include enrolling patients prospectively into the vEDS have improved outcomes when undergoing elective Research Collaborative to further ascertain the natural surgical repairs or interventions compared with those history, elucidate the role of surgical interventions, and who require emergency operative repair.11 Our work to refine management recommendations in the context and others substantiate that the genetic testing is no of patient-centered outcomes. longer experimental or investigational and has real consequences to the affected individual, in terms of The authors acknowledge Binod Shrestha (Department surveillance, the operative approaches, perioperative of Cardiovascular and Vascular Surgery, at the University Journal of Vascular Surgery Shalhub et al 11 Volume -, Number - of Texas Health Science Center at Houston) for assisting 12. Byers PH, Belmont J, Black J, De Backer J, Frank M, with data collection at the University of Texas Health Sci- Jeunemaitre X, et al. Diagnosis, natural history, and man- ence Center at Houston. agement in vascular Ehlers-Danlos syndrome. Am J Med Genet C Semin Med Genet 2017;175:40-7. 13. Harlander-Locke MP, Lawrence PF. The Current State of the AUTHOR CONTRIBUTIONS Vascular Low-Frequency Disease Consortium. Ann Vasc Surg Conception and design: SS, KH 2017;38:8-9. Analysis and interpretation: SS, PB 14. Smith LT, Schwarze U, Goldstein J, Byers PH. Mutations in Data collection: SS, KH, DC, FD, GC, KW, EM, MS, KS, GO, the COL3A1 gene result in the Ehlers-Danlos syndrome type IV and alterations in the size and distribution of the major MR, CN, KC, CB, ED, YK, DZ, RP, MP, DM collagen fibrils of the dermis. J Invest Dermatol 1997;108: Writing the article: SS, PB 241-7. Critical revision of the article: SS, PB, KH, DC, FD, GC, KW, 15. Mizuno K, Boudko S, Engel J, Bachinger HP. Vascular EM, MS, KS, GO, MR, CN, KC, CB, ED, YK, DZ, RP, MP, DM Ehlers-Danlos syndrome mutations in type III collagen Final approval of the article: SS, PB, KH, DC, FD, GC, KW, differently stall the triple helical folding. J Biol Chem 2013;288:19166-76. EM, MS, KS, GO, MR, CN, KC, CB, ED, YK, DZ, RP, MP, DM 16. Schwarze U, Goldstein JA, Byers PH. Splicing defects in the Statistical analysis: SS, PB, KH, DC, FD, GC, KW, EM, MS, KS, COL3A1 gene: marked preference for 5’ (donor) spice-site GO, MR, CN, KC, CB, ED, YK, DZ, RP, MP, DM mutations in patients with exon-skipping mutations and Obtained funding: SS, PB, KW Ehlers-Danlos syndrome type IV. Am J Hum Genet 1997;61: Overall responsibility: SS 1276-86. 17. Beighton P, De PA, Steinmann B, Tsipouras P, Wenstrup RJ. Ehlers-Danlos syndromes: revised nosology, Villefranche, 1997. Ehlers-Danlos National Foundation (USA) and Ehlers- REFERENCES Danlos Support Group (UK). Am J Med Genet 1998;77:31-7. 1. Pope FM, Martin GR, Lichtenstein JR, Penttinen R, Gerson B, 18. Brooke BS, Arnaoutakis G, McDonnell NB, Black JH 3rd. Rowe DW, et al. Patients with Ehlers-Danlos syndrome type Contemporary management of vascular complications IV lack type III collagen. Proc Natl Acad Sci U S A 1975;72: associated with Ehlers-Danlos syndrome. J Vasc Surg 1314-6. 2010;51:131-8. 2. Pepin M, Schwarze U, Superti-Furga A, Byers PH. Clinical and 19. Okada T, Frank M, Pellerin O, Primio MD, Angelopoulos G, genetic features of Ehlers-Danlos syndrome type IV, the Boughenou MF, et al. Embolization of life-threatening vascular type. N Engl J Med 2000;342:673-80. arterial rupture in patients with vascular Ehlers-Danlos syn- 3. Pyeritz RE. Ehlers-Danlos syndrome. N Engl J Med 2000;342: drome. Cardiovasc Intervent Radiol 2014;37:77-84. 730-2. 20. Lum YW, Brooke BS, Arnaoutakis GJ, Williams TK, 4. Malfait F, Francomano C, Byers P, Belmont J, Berglund B, Black JH 3rd. Endovascular procedures in patients with Black J, et al. The 2017 international classification of the Ehlers-Danlos syndrome: a review of clinical outcomes and Ehlers-Danlos syndromes. Am J Med Genet C Semin Med iatrogenic complications. Ann Vasc Surg 2012;26:25-33. Genet 2017;175:8-26. 21. Kanner AA, Maimon S, Rappaport ZH. Treatment of spon- 5. Eagleton MJ. Arterial complications of vascular Ehlers- taneous carotid-cavernous fistula in Ehlers-Danlos syn- Danlos syndrome. J Vasc Surg 2016;64:1869-80. drome by transvenous occlusion with Guglielmi detachable 6. Frank M, Albuisson J, Ranque B, Golmard L, Mazzella JM, coils. Case report and review of the literature. J Neurosurg Bal-Theoleyre L, et al. The type of variants at the COL3A1 2000;93:689-92. gene associates with the phenotype and severity of 22. Huynh TJ, Morton RP, Levitt MR, Ghodke BV, Wink O, vascular Ehlers-Danlos syndrome. Eur J Hum Genet 2015;23: Hallam DK. Successful treatment of direct carotid- 1657-64. cavernous fistula in a patient with Ehlers-Danlos syndrome 7. Oderich GS, Panneton JM, Bower TC, Lindor NM, Cherry KJ, type IV without arterial puncture: the transvenous triple- Noel AA, et al. The spectrum, management and clinical overlay embolization (TAILOREd) technique. BMJ Case Rep outcome of Ehlers-Danlos syndrome type IV: a 30-year 2017;2017. experience. J Vasc Surg 2005;42:98-106. 23. Tonnessen BH, Sternbergh WC 3rd, Mannava K, Money SR. 8. Adham S, Trystram D, Albuisson J, Domigo V, Legrand A, Endovascular repair of an iliac artery aneurysm in a patient Jeunemaitre X, et al. Pathophysiology of carotid-cavernous with Ehlers-Danlos syndrome type IV. J Vasc Surg 2007;45: fistulas in vascular Ehlers-Danlos syndrome: a retrospective 177-9. cohort and comprehensive review. Orphanet J Rare Dis 24. Khalique Z, Lyons OT, Clough RE, Bell RE, Reidy JF, 2018;13:100. Schwarze U, et al. Successful endovascular repair of acute 9. Leistritz DF, Pepin MG, Schwarze U, Byers PH. COL3A1 hap- type B aortic dissection in undiagnosed Ehlers-Danlos syn- loinsufficiency results in a variety of Ehlers-Danlos syndrome drome type IV. Eur J Vasc Endovasc Surg 2009;38:608-9. type IV with delayed onset of complications and longer life 25. Shalhub S, Eagle KA, Asch FM, LeMaire SA, Milewicz DM; Gen expectancy. Genet Med 2011;13:717-22. TAC Investigators. Endovascular thoracic aortic repair in 10. Pepin MG, Schwarze U, Rice KM, Liu M, Leistritz D, Byers PH. confirmed or suspected genetically triggered thoracic aortic Survival is affected by mutation type and molecular mech- dissection. J Vasc Surg 2018;68:364-71. anism in vascular Ehlers-Danlos syndrome (EDS type IV). 26. Schwarze U, Schievink WI, Petty E, Jaff MR, Babovic- Genet Med 2014;16:881-8. Vuksanovic D, Cherry KJ, et al. Haploinsufficiency for one 11. Shalhub S, Black JH 3rd, Cecchi AC, Xu Z, Griswold BF, COL3A1 allele of type III procollagen results in a phenotype Safi HJ, et al. Molecular diagnosis in vascular Ehlers-Danlos similar to the vascular form of Ehlers-Danlos syndrome, syndrome predicts pattern of arterial involvement and out- Ehlers-Danlos syndrome type IV. Am J Hum Genet 2001;69: comes. J Vasc Surg 2014;60:160-9. 989-1001. 12 Shalhub et al Journal of Vascular Surgery --- 2019

27. Ong KT, Perdu J, De Backer J, Bozec E, Collignon P, 29. Malfait F, De Paepe A. The Ehlers-Danlos syndrome. Adv Exp Emmerich J, et al. Effect of celiprolol on prevention of car- Med Biol 2014;802:129-43. diovascular events in vascular Ehlers-Danlos syndrome: a 30. Vascular Ehlers-Danlos Syndrome (vEDS) Collaborative. prospective randomised, open, blinded-endpoints trial. Available at: www.becertain.org/projects/patient-engagement/ Lancet 2010;376:1476-84. vascular-ehlers-danlos-syndrome-veds-collaborative.Accessed 28. Mast KJ, Nunes ME, Ruymann FB, Kerlin BA. Desmopressin May1,2018. responsiveness in children with Ehlers-Danlos syndrome associated bleeding symptoms. Br J Haematol 2009;144: 230-3. Submitted Sep 7, 2018; accepted Jan 23, 2019.