Good performance of the criteria of American College of Medical Genetics and Genomics/Association for Molecular Pathology in prediction of pathogenicity of genetic variants causing thoracic aortic aneurysms and dissections

There is a growing interest in better recognition of genetic factors leading to thoracic aortic aneurysms and dissections (TAAD) [1, 2]. Diagnosis of TAAD leads to important clinical decisions. Although aortic size remains the main criterion for prophylactic surgical intervention, recent evidence indicates that aortic dissection may occur in nondilated or mildly dilated aorta, so aortic size loses its predictive ability [3‐5]. Early diagnosis and subsequent prophylactic management prior to dissection significantly improves survival [6].

Previously, we described 51 TAAD patients studied with the use of whole exome sequencing (WES) or panel analysis (TruSight One, Illumina). We have reported a significant difference in event-free survival between ‘genotype-positive’ group consisting of patients with variants considered as pathogenic/likely pathogenic (P/LP) on account of either literature reports, protein disruption, de novo occurrence, segregation analysis or strong pathogenicity predictions as compared to ‘genotype-negative’ patients [7].

In the present work we analyzed an independent cohort of 132 patients tested in the routine clinical setting for variants in 30 genes associated with TAAD or 174 genes included in TruSight Cardio panel (Illumina). Considering the recent progress in development of free online databases collecting knowledge on human genetic variation we aimed to evaluate the performance of variant classification based on recommendations of American College of Medical Genetics and Genomics and the Association for Molecular Pathology (ACMG-AMP) [8]. The evaluation was based on a comparison of the event-free survival in patients classified as genotype-positive vs. genotype-negative by the ACMG-AMP criteria or by the expert curated information from the ClinVar database.

ClinVar being an expert knowledge-based database enables only the classification of variants with the existing evidence regarding observed health status. Novel variants and those lacking sufficient evidence add to the growing burden of VUS. In silico algorithms [13] add valuable supporting evidence regarding pathogenicity of given variant. The development of ACMG-AMP criteria which combine both lines of evidence with many levels of supporting information into an integrated prediction was a cornerstone in unifying interpretation of clinical significance of genetic variants. Since their initial development the criteria were evaluated by multiple laboratories, several areas for improvement were identified and there is an agreement on the need of further validation [14, 15]. In the present study we have tested and found good performance of an online tool available at varsome.com which allows classification based on ACMG-AMP criteria. Although the tool doesn’t cover all the criteria and is the subject to constant change in the algorithms implementing the ACMG-AMP criteria, it is free and convenient to use (https://​varsome.​com/​about/​resources/​acmg-implementation/​). The constantly improving functionalities of the AMG-AMP tool result in the growing number of more definitive predictions (B, LB, LP and P) and decreasing VUS classifications. Indeed, after application of ACMG-AMP criteria out of 72 variants having no ClinVar record or classified as VUS only 22 variants kept their VUS status (26 were classified as P/LP and 24 as B/LB). The goal of our study was to test the accuracy of these predictions using clinical data and ClinVar database as a reference. Similar to our previous findings [7], patients carrying known pathogenic variants (P/LP by ClinVar), had AAD or were referred for elective aortic surgery significantly earlier than patients in whom no rare variants in genes of interest or only variants classified as B/LB by ClinVar were identified. Importantly, the genotype-positive group according to the ACMG verdict alone had similar (significant) difference in the event-free survival vs. the reference group as the group defined by ClinVar alone. Furthermore, there was no significant difference in event-free survival when we compared genotype-positive group according to ClinVar to patients carrying variants predicted to be P/LP by ACMG criteria alone. In the similar way, there was no significant difference between patients carrying only B/LB variants according to ACMG alone vs. the reference group. These findings indicate that ACMG predictions in regard to TAAD-susceptibility genes are of similar high clinical significance to information provided by ClinVar.

We also observed that among variants classified as pathogenic, those predicted to cause LOF were associated with higher severity then the remaining variants. Thus, variant type (LOF vs. non LOF) could provide additional clinically useful information. It is likely that in the future even better predictions will be possible based on knowledge collected for individual variants in sufficiently powered studies.

Interestingly, in our cohort of TAAD patients there was no significant difference in disease progression between patients carrying VUS by either ACMG or ClinVar vs. the reference group suggesting that variants lacking any sort of evidence for being P/LP are more likely to be B/LB, at least regarding being the cause of monogenic disease. However, it should be noted that this does not preclude their role as risk factors in multifactorial form of disorder. Interestingly, our data are consistent with the findings from the study of 1005 patients diagnosed with nonischemic dilated cardiomyopathy which analysed prognostic impact of disease-causing variants classified based on the ACMG-AMP criteria. The rate of major adverse cardiovascular events, end stage heart failure and malignant ventricular arrhythmia in 10-year follow-up was significantly higher in the genotype-positive group compared to the genotype-negative but showed no trend towards higher rate of these conditions among patients carrying VUS [16].

Our results also indicate that the TAAD specific panel performs similar to the larger universal cardiac panel. Out of 22 variants identified using TAAD-specific panel and predicted to be P/LP by ACMG and/or VarSome only one variant (in the HCN4 gene) wouldn’t have been found if universal cardiac panel was used. This observation supports previous reports suggesting that use of smaller gene panels consisting of the most frequently mutated genes may be a cost-effective strategy as a first step of the genetic diagnosis for TAAD patients [17].

Furthermore, based on our analysis of TAAD patients, the P/LP variants we found by either ClinVar or ACMG resided mainly within the relatively few genes (FBN1, SMAD3, TGFB2, TGFBR1, COL3A1, MYH11, MYLK) generally agreed to be “definitively” associated with hereditary TAAD [2] and the highest risk of early fatal complications [18]. Patients with genetic defects in TGFBR1, TGFBR2, SMAD3, TGFB2 (affecting TGF-β signalling pathway), COL3A1 (extracellular matrix component), FBN1 (TGF-β signalling affecting extracellular matrix component), MYH11, ACTA2, MYLK and PRKG1 (structural components and modifiers of smooth muscle layer of the aorta) are at risk of dissection or rupture at aortic diameters less than 5.0 cm recommended for surgical intervention, and therefore sequencing is of key importance for decision-making regarding the time of prophylactic surgery [1].

Gene panel containing the most established genes associated with the highest risk of early fatal complications in patients with hereditary TAAD (ACTA1, COL3A1, FBN1, MYH11, SMAD3, TGFB2, TGFBR1, TGFBR2, MYLK) was sufficient to identify prevailing majority of variants most likely to be causative of the disease.

ACMG classification tool available at varsome.com is useful in assessing pathogenicity of novel genetic variants in TAAD patients. Variant pathogenicity as assessed by ACMG criteria is a strong predictor of earlier AAD or need for surgical intervention. Hopefully, the progress in combining empiric data with the further development of bioinformatic tools will lead to the reduction in the number of VUS allowing even more reliable predictions that could be successfully used for clinical decisions.