Introduction

In the past several years, the quality of variant classification for Mendelian disorders has been improved through data sharing which has been facilitated, in part, through databases such as ClinVar, an NIH open-access database of clinically observed variants and their classifications. However, one of the most recognized challenges with databases such as ClinVar is the number of discordantly or improperly classified variants, which has been attributed to factors such as the age of the entries, nonclinical submissions, and differences in application of variant interpretation guidelines [1,2,3,4,5].

The contribution of gene-based knowledge to the process of variant classification is beginning to be more widely recognized. The American College of Medical Genetics and Genomics (ACMG) and the Association for Molecular Pathology (AMP) 2015 variant interpretation guidelines included some criteria (PM1, PVS1, and PP2) which allowed for incorporation of gene-based knowledge into variant classification [6]. In addition, ClinGen (the Clinical Genome Resource) has established working groups to develop consensus methods for variant interpretation in specific diseases and genes, producing, for example, gene-specific guidance for MYH7-associated inherited cardiomyopathies and RASopathies [7, 8]. Furthermore, Muiño-Mosquera et al. [9] recently published pilot disease- and gene-specific guidelines for reclassification of FBN1 variants identified in their own laboratory. Gene-specific knowledge can improve variant classification in the following ways: (1) knowledge of critical residues for which the PM1 criteria can be applied allows for variants at these residues that would otherwise be classified as variants of uncertain significance (VUS) to be classified as likely pathogenic (LP) or pathogenic (P); (2) appropriate MAF cutoffs based on disease incidence allow for variants that would otherwise be classified as VUS (or LP) to be classified as likely benign or benign (or VUS) by using BS1 and BA1; and (3) improved inter- and intra-laboratory consistency in variant classification. Together, this leads to a reduction in the number of VUSs reported, and ultimately results in better service to practitioners and improved patient care.

Marfan syndrome (MFS), due to variants in FBN1, is an autosomal dominant connective tissue multisystem disorder [10]. There are over 2000 published FBN1 variants, and many are unique to individual families. Missense variants, especially cysteine (Cys) substitutions, are the most common type of FBN1 variant [11, 12]. The type of FBN1 variant identified and the likelihood of that variant being pathogenic are recognized as important factors when making a diagnosis of MFS, with de novo (in the absence of family history), nonsense, frameshift, splicing, and missense substitutions of conserved residues considered most likely to be pathogenic [13].

In order to optimize classification of FBN1 variants, it is important to have a thorough understanding of fibrillin-1 function and structure. Fibrillin-1 is a 350 kDa calcium-binding glycoprotein with a highly modular organization [14, 15]. Fibrillin-1 is composed of a repetition of two disulfide-rich modules: 47 epidermal growth factor (EGF)-like domains and seven transforming growth factor β-binding protein-like (TB) domains. The majority of fibrillin-1 consists of EGF-like domains, and 43/47 EGF-like domains are calcium-binding (cbEGF-like domains). There are also two hybrid domains, which contain properties of both cbEGF-like and TB domains. Cys residues play an important and well-studied role in fibrillin-1 structure, and occur in sets of six and eight in each EGF-like (calcium- and noncalcium-binding) and TB domain, respectively. FBN1 variants that disrupt or create new Cys in these domains are known to lead to the MFS phenotype and other fibrillinopathies [13, 16,17,18]. Muiño-Mosquera et al. [9] highlighted the importance of these Cys residues in their suggested FBN1 guidelines. They suggested that disruption of a Cys in the cbEGF-like domains should qualify as a strong criteria (PS3), and disruption of all other Cys residues in FBN1, creation of a new Cys in the cbEGF-like domains, and disruption of calcium-binding residues in the cbEGF-like domains should all qualify as a moderate criteria (PM1) [13]. However, Muiño-Mosquera et al. did not provide guidance on numerous other critical residues of fibrillin-1 that allow for further refinement of the application of the PM1 criterion (detailed in materials and methods and Fig. 1), nor did they discuss application of the PP2 criterion (missense variant in a gene which has a low rate of benign missense variation and in which missense variants are a well-known disease mechanism).

Fig. 1
figure 1

FBN1 domain alignment and critical residues. Amino acid residues known to be critical for function are either bolded or highlighted. a EGF/cbEGF-like domain alignment. Cysteines (C) bolded in red; calcium-binding (D/N-x-D/N -E/Q-Xm-D/N-Xn-Y/F) highlighted in yellow [note: an asparagine (N) to serine(S) at the second D/N position maybe tolerated; see text]; Gly (G)/Ala (A) at residue 3 between obligatory Cys2 and Cys3 in a subtype of cbEGF-like domains highlighted in blue; cbEGF interdomain packing (Gly between Cys3 and Cys4 and aromatic residue Y/F/H/W between Cys5 and Cys6) highlighted in green (note: the Gly interacts with the Y/F/H/W in the upstream cbEGF domain); “EGF core” which comprises the major beta-sheet of the EGF fold is underlined (note: only highlighted/bolded amino acids in the EGF core are currently known to be critical). b TB domain alignment. 5′ domain = 5′-adjacent domain; 3′ domain = 3′-adjacent domain; AA = amino acid; cbEGF = calcium-binding EGF-like domain; EGF = noncalcium-binding EGF-like domain; TB = transforming growth factor beta domain; HYB = hybrid domain

Full size image

While there are many readily accessible tools and variant databases, such as ClinVar, that provide clinical laboratories access to previously reported genetic variants, these tools may not provide sufficient evidence to perform an accurate independent variant classification. Furthermore, clinical laboratories may lack the resources to gather the necessary knowledge regarding critical residues in their gene of interest in order to develop gene-specific criteria for variant classification.

In order to better understand how clinical laboratories are classifying variants affecting critical and noncritical residues in FBN1, we analyzed FBN1 variant classifications and associated evidence submitted to the ClinVar database. After this analysis, we applied gene-specific guidelines, including refined criteria for PM1 based on known critical residues, to reanalyze the variants and determine if application of these guidelines and knowledge help to reclassify or resolve discrepant classifications of variants listed in ClinVar.

Materials and methods

Data retrieval

ClinVar data were downloaded on Dec 12, 2017 for a total of 1405 unique FBN1 variants with 1963 classifications from 18 different submitters. On February 13, 2019, downloaded data were reanalyzed and updated, and 21 additional missense variants were added from another search of ClinVar. Names of submitting groups were blinded for the analysis and variant submission evidence provided was reviewed manually.

Categorization of FBN1 critical residue variants

Missense variants were categorized by domain and then investigated as to whether they impacted consensus/critical amino acid residues or otherwise were likely to impact fibrillin function (Table 1). Such missense variants that occurred in each domain (Fig. 1) included those that: (1) created or destroyed a Cys residue in a cbEGF-like domain [14, 16,17,18,19] (Fig 1a); (2) altered any of the five residues in the calcium-binding consensus sequence D/N-x-D/N-E/Q-Xm-D/N-Xn-Y/F in a cbEGF-like domain [note: an asparagine (N) to serine (S) at the second D/N position maybe tolerated] [18, 20,21,22,23,24,25] (Fig 1a); (3) altered the critical glycine (Gly)/alanine (Ala) residue between Cys2–Cys3 in a subset of cbEGF-like domains [26,27,28,29,30] (Fig 1a); (4) altered interdomain packing residues in a subset of cbEGF-like domains [20, 29, 31,32,33]; (5) created or altered a Cys residue in a TB domain (Fig. 1b) [18, 34]; (6) created or altered a Cys residue in an EGF domain [35, 36] (Fig. 1a); (7) created or altered a Cys residue in the second hybrid (hybrid2) domain [37] [note: we did not consider Cys in the hybrid1 domain to meet criteria for PM1 because this domain has an extra Cys of unclear function and studies have shown that hybrid1 domain-deleted mice have a normal lifespan [38]]; (8) impacted a critical residue in the fibrillin-unique N terminus (FUN) site [39,40,41]; or (9) impacted a residue in the Arg-Gly-Asp (RGD) integrin binding site in the TB domain [42, 43]. However, ClinVar did not contain entries for variants in the RGD region, so knowledge of these critical residues could not be assessed.

Table 1 FBN1 critical missense residue type, frequency, and ClinVar clinical classification
Full size table

FBN1 variant reanalysis and classification

FBN1 variant nomenclature was based on RefSeq NM_000138.4. Variants were classified in ClinVar based on a five-level classification system as per the ACMG/AMP 2015 variant classification guidelines: P, LP, VUS, likely benign (LB), or benign (B) (Supplemental Table 1) [6]. A single variant’s classification submitted by multiple submitters was considered discrepant if the classifications disagreed between (L)B versus VUS, (L)B versus (L)P, or VUS versus (L)P.

Variant classifications were reanalyzed by our group for pathogenicity based on FBN1-specific modified 2015 ACMG/AMP/CAP criteria [6]. These modifications included: application of PM1 for variants that affect any of the nine types of residues discussed above and an automatic minimum classification of (L)P for any variant impacting a Cys residue in a cbEGF-like or TB domain. In addition, we applied PP2 for those variants that met additional pro-pathogenicity criteria and for which we were unable to apply BS1, BP4, or other pro-benign criteria. The PP2 criterion is applicable for FBN1 since it has an ExAC missense constraint z-score of 5.33, which is above the ClinGen recommended z-score of 3.09 for application of PP2 [44]. In addition, the minor allele frequency (MAF) was analyzed using gnomAD population data (https://gnomad.broadinstitute.org/) [44]. A cutoff for applying the BS1 criteria (allele frequency greater than expected for the disorder) was set conservatively at 0.01%, based on using the 1/5000 as the population frequency of MFS (commonly stated to range from 1/5000 to 1/10,000) [45,46,47,48], and also considering that no single FBN1 variant is responsible for >50% of all of MFS. As an added measure of stringency, we only applied BS1 if a variant was present at a frequency of >0.01% and occurred in at least three alleles in the specific gnomAD population it was detected in. We used the HGVS nomenclature Version 15.11 for functional and clinical classification to describe variant effects (https://varnomen.hgvs.org/). Classified variants (n = 674) were uploaded to the FBN1-specific Global Variome shared Leiden Open Variation Database (LOVD) (https://databases.lovd.nl/shared/genes/variants 0000459080_0000459753).

Results

There were 1426 FBN1 variants with 1999 total variant classifications from 18 submitters (mostly clinical laboratories) in ClinVar. There were 11 submitters with 45 or more FBN1 variant submissions. There were 266 variants that had multiple (>1) submissions, and 80 of these (30.1%) had discrepant variant classifications.

Overview of missense variants

Since FBN1 missense variants are generally more challenging to classify compared to nonsense, splicing, or frameshift variants, we focused our analysis on missense variants. Of the 1426 FBN1 variants listed in ClinVar, 674 (47.3%) were unique missense variants with a total of 973 variant calls. Of the 674 missense variants, 140 of them (20.8%) had more than one submitter and 43/140 (30.7%) had discrepant classifications.

Variants that occur at critical residues or otherwise impact fibrillin-1 function and are not present in population databases would be expected to be functionally classified as, at a minimum, probably affects function (PAF), or clinically classified as LP due to their functional significance [13]. Of the 674 missense variants in ClinVar, 361 (53.6%) occurred at a known critical residue or created a Cys in relevant domains of fibrillin-1 (see “categorization of FBN1 critical residue variants” in methods section), for a total of 451 calls (Table 1; Fig. 2; supplemental table). Of these 451 calls, 361 (80.0%) gave a (L)P clinical classification; however (L)P classification frequency was dependent on which type of critical residue the variant impacted (Fig. 3a). For the well-known cbEGF-like domain Cys residues, 90.0% (208/231) of the variant classifications were correctly clinically classified as (L)P. In fact, overall, missense variants that impacted a Cys residue were highly recognized as (L)P variants (283/310 calls or 91.3%) (Fig. 3a). On the other hand, for non-Cys critical residues, only 55.3% (78/141) of clinical classifications were (L)P. Variants impacting the critical glycine/alanine and interdomain packing residues (not taking into account aromatic residue substitutions; see below) in the cbEGF-like domain were the least likely critical residue variants to be classified as (L)P (39.3% or 11/28 variant calls). Overall, we were able to resolve discrepancies in all nine variants with conflicting multisubmitter classifications and reclassify 56 of the remaining 352 (15.9%) of these functionally important variants, as described in more detail below. Overall this led to a revision of 17.2% (77/451) of variant classifications for functionally important variants.

Fig. 2
figure 2

FBN1 critical region missense variant frequency by location. cbEGF = calcium binding EGF-like domain; Cys = cysteine residue; cbEGF C2–C3 Gly/Ala = Gly/Ala at residue 3 between obligatory Cys2 and Cys3 in a subtype of cbEGF-like domains; TB = TGF-beta-binding protein domain; Hybrid2 = second hybrid domain; FUN = fibrillin unique N-terminal domain

Full size image
Fig. 3
figure 3

a Percent of critical region missense variants classified as (L)P. b Percent of provided evidence stating critical residue as factor supporting (L)P classification. cbEGF = calcium binding EGF-like domain; Cys = cysteine residue; cbEGF C2–C3 Gly/Ala = Gly/Ala at residue 3 between obligatory Cys2 and Cys3 in a subtype of cbEGF-like domains; TB = TGF-beta binding protein domain; Hyb2 = Hybrid2 domain; FUN = fibrillin unique N-terminal domain

Full size image

Missense variants impacting Cys residues in the cbEGF-like domain

As mentioned above, 90% (208/231) of cbEGF-like Cys clinical classifications were (L)P (Table 1; Fig. 3a). Of these, 130 (62.5%) provided evidence regarding variant classification. Justification for 74.6% of these calls (97/130) included verbiage related to the importance of cbEGF-like domain Cys (Fig. 3b). For the remaining 25.4% of calls, the majority (25/33) was from 2014 or earlier.

There were 20 cbEGF-like Cys missense variants that were classified 23 times as VUS by five different submitters. One variant with a dual submission had a VUS and an LP call, and the rest had two VUS calls. Most (18/20) of the VUS variants were submitted to ClinVar between 2013 and 2017, with two variants submitted in 2009. All variant calls that were not classified as (L)P were reclassified by us to (L)P or PAF (supplemental table).

Missense variants impacting calcium-binding residues in the cbEGF-like domain

Variants impacting the D/N-x-D/N-E/Q-Xm-D/N-Xn-Y/F calcium-binding residues in the cbEGF-like domains comprised the next largest grouping of critical residue FBN1 missense variants in ClinVar with a total of 110 variant calls for 80 variants. As described below, some of the missense changes in this region involved the substitution of a known or potentially tolerated amino acid and we did not consider these variants to be L(P). This situation impacted 9 variants with 13 clinical classifications [11 VUS and 2 L(B)]. Of the remaining 71 variants with 97 variant calls, 68.0% (66/97) were called (L)P by 9 different submitters (Fig. 3a). Of these, 39 (59.1%) classifications had meaningful submitter evidence provided. However, only 4/39 (10.2%) submissions highlighted the importance of the calcium-binding residues (Fig. 3b). Twenty-two (30.6%) of the (L)P variants were at the first D/N of the calcium-binding consensus region. However, not all submitters recognized the relevance of this first D/N, since a variant at this position was also clinically classified as VUS eight other times. In fact one submitter initially classified a “first D/N” variant as (L)P and then later classified a different “first D/N” variant as VUS.

Not including variants that we considered to be VUS (as described above and below), the majority (25/31; 80.1%) of LB or VUS clinical classifications at calcium-binding residues was submitted between 2015 and 2018, with one submission in 2008 and five submissions from 2011 to 2014. For the VUS and LB submissions, 18/25 submitters provided variant classification evidence, with a variety of supportive evidence statements, including in some cases that the variant was not a Cys-based variant in a cbEGF-like domain.

There were only four discrepant multisubmitter clinical classifications (one occurrence of LB versus VUS and three occurrences of VUS versus LP) of the calcium-binding residues. For the variant that was classified as LB and VUS, neither submission included meaningful evidence to support classification. For the VUS versus LP discrepant variant calls (n = 13), 8/13 calls had associated evidence in their submission, and all of these were from more recent years (2016–2018), but only one mentioned the calcium-binding critical residue.

Nine variants with 13 calls were kept at or revised to VUS (or functional effect unknown). Three of these variants were substitutions of a D for N or vice versa at a D/N position: p.Asp1197Asn, p.Asp1240Asn, and p.Asn1907Asp. The other six variants were Asn > Ser changes at the second D/N in the D/N-x-D/N-E/Q-Xm-D/N-Xn-Y/F consensus sequence: p.Asn615Ser, p.Asn1030Ser, p.Asn1282Ser, p.Asn1489Ser, p.Asn2526Ser, and p.Asn2650Ser (Table 2). During the course of this study, we observed that some of these Asn > Ser variants had relatively high minor allele frequencies in gnomAD. For example, p.Asn1030Ser is observed in 0.22% (68/30616 alleles, including two homozygotes) of the South Asian population. We examined all asparagine residues in the second D/N position across all cbEGF-like domains to ascertain the presence and frequency of serine substitutions in the gnomAD population. Of the 43 cbEGF-like domains, 18 (41.9%) had an asparagine (versus aspartate) at that position, and 9/18 (50%) had an Asn > Ser change described in ClinVar and/or present in gnomAD (Table 2). Given that seven of the nine p.Asn > Ser variants are present in gnomAD and since Asn and Ser are both similar polar uncharged amino acids with a Grantham distance of 46 (range 0–215), it is possible that a serine at this position is tolerated (i.e., D/N-x-D/N/S-E/Q-Xm-D/N-Xn-Y/F) and should be categorized as VUS or of uncertain functional effect. However, functional studies are needed to fully assess the impact of a serine at this position. Overall, of the 44 calcium-binding variant calls that were (L)B or VUS in ClinVar, 75% (33/44) were reclassified to (L)P.

Table 2 FBN1 asparagine (Asn) to serine (Ser) changes at the second D/N of the cbEGF-like calcium-binding region (D/N-x-D/N-E/Q-Xm-D/N-Xn-Y/F) observed in Clinvar and/or gnomAD
Full size table

Missense variants impacting Cys residues in the TB domain

The third largest group of critical residue missense variants impacted Cys residues in the 8-Cys TB domain. There were 34/37 (91.9%) L(P) variant clinical classifications for 30 TB domain Cys variants (Fig. 3a) and three VUS classifications. Of the 34 (L)P classifications, 22 provided evidence, but only four of these acknowledged the importance of the 8-Cys motif of the TB domain in their classification evidence (Fig. 3b). For the VUS classifications, one of the submitters provided evidence supporting their VUS classification (submitted to ClinVar in 2017), but did not acknowledge the importance of the 8-Cys motif of the TB domain.

Critical glycine (Gly)/alanine (Ala) between Cys2–Cys3 in a subset of cbEGF-like domains

There were 21 clinical classifications of 19 different variants that impacted the Gly/Ala in the third position between Cys2 and Cys3 in a subset of cbEGF-like domains. Overall only 9/21 (42.9%) classifications were (L)P (Fig. 3a). The majority (17/19) of the variants had a single submission, whereas two variants had two submissions [VUS and (L)P]. Of the nine (L)P submissions at this position, eight (88.9%) contained evidence, but only four of these mentioned the importance of this residue (Fig. 3b). Other reasons provided for the (L)P call included that the variant was previously reported in MFS patients, absence from databases, in silico predictions, conserved residue, nonconservative change, and the presence of nearby pathogenic missense variants (including Cys variants). For VUS submitters that provided evidence, evidence supporting the call included that the variant had been reported before in association with MFS, was not observed in databases, was a nonconservative amino acid substitution, occurred at a highly conserved residue, and had deleterious in silico predictions, but did not affect a Cys residue.

EGF-like domain cysteine missense variants

There were 26 variant calls for 20 Cys missense variants in the EGF-like domain, and all but one of them were clinically classified as (L)P (Fig. 3a). The single VUS classification was a submission classified in 2009 and did not contain evidence. For the 25 (L)P calls, 14 contained evidence for classification and 57.1% (8/14) mentioned that EGF-like domain Cys variants factored into the classification (Fig. 3b).

Other critical residues

There were seven interdomain packing residues in a cbEGF-like domain with nine calls, and 2/9 clinical classifications were (L)P. Two of the interdomain packing residues classified as VUS (p.Tyr1839Phe and p.Tyr2236Phe) involved the change from one aromatic residue (tyrosine) to another (phenylalanine), and thus we did not change their classification from VUS. One variant had multiple calls [two VUS and one (L)P]. None of the (L)P calls provided evidence stating a critical residue as a factor supporting the classification (Fig. 3b).

There were nine variants with nine (L)P variant calls that impacted Cys residues in the second hybrid (hybrid2) domain. Of the (L)P calls, four provided evidence, and two described the importance of Cys variants in general, with one mentioning the importance of Cys variants in hybrid domains.

Finally, there were eight variants with eight calls that occurred in the FUN domain. All eight of these calls were (L)P, but only one of the calls had evidence and this evidence did not state the importance of the FUN domain (Fig. 3a, b). Notably, seven of the eight calls involved disruption of a Cys, which may have influenced the (L)P call.

Missense variants at residues not known to be critical

There were 313 other missense variants with 522 clinical classifications that did not occur at a previously described critical residue (supplemental table). Of these, 224 were classified by a single submitter. For the 89 variants with multiple submitters, there were 298 variant calls. Conflicting classifications occurred with 34/89 (38.2%) variants with multiple submitters.

For these noncritical residue missense variants, 105/522 (20.1%) were classified as (L)P, 318/522 (60.9%) were VUS, and 99/522 (19.0%) were (L)B. Applying review of the evidence, when available, and our refined MAF cutoff, 30.5% (97/318) of the VUS calls were revised to (L)B and 63.8% (67/105) of the L(P) calls were revised to (L)B (4/105) or VUS (63/105), respectively (Fig. 4). Overall, we were able to resolve variant discrepancies in all 34 variants with conflicting classifications and reclassify 94 (33.7%) of the remaining 279 missense variants not known to be critical or impact fibrillin-1 function.

Fig. 4
figure 4

Reclassification of ClinVar-classified noncritical FBN1 missense variants. The original classification in ClinVar is represented by the “ClinVar-based” bars. The reclassifications are represented by the “Gene and MAF-guided” bars

Full size image

Discussion

Through this analysis, we demonstrate the importance of applying gene-based knowledge to variant classification. Although our scope was different, we observed similarities to that of Muiño-Mosquera et al. [9], in that the application of gene-based guidelines changed the interpretation of previously classified FBN1 variants. Unlike Muiño-Mosquera et al., which looked at FBN1 variants of all types observed in their own laboratory, our study was focused specifically on FBN1 missense variants submitted to ClinVar, giving a more global perspective of the impact of gene-based guidelines to FBN1 variant classification and some of the more widespread inadequacies of FBN1 variant assessment. We further enhanced the Muiño-Mosquera study by including additional categories of FBN1 critical residues, incorporation of the PP2 criterion when appropriate, and a refinement of the MAF cut point to use for BS1. Furthermore, while their method resulted in an increased number of VUSs (59.7% of those outside of well-established functional sites), our approach actually led to an overall decrease in the number of VUS/effect unknown calls outside of well-established functional sites. Overall, we were able to resolve all 43 missense variants that had ClinVar discrepant multisubmitter classifications and revise variant calls in 150 (23.8%) of the remaining 631 FBN1 missense variants. Taken together, we revised 24.8% (241/973) of all FBN1 missense variant classifications in ClinVar.

While the majority (80.0%) of ClinVar classifications for critical residue missense variants were (L)P, there were specific types of critical residue variants that were more often misclassified. Because of the well-known association between FBN1 Cys variants and MFS, it is not surprising that most submitters correctly classified Cys variants as (L)P. However, when considering evidence provided supporting (L)P calls for Cys variants, there seemed to be a lowered awareness regarding the importance of Cys variants in non-cbEGF-like domains. Much of the submitter evidence supporting (L)P classification had to do with other valid reasons supporting pathogenicity, rather than the functional relevance of the residue. As noted, non-Cys critical residue variants had a lower likelihood of being classified as (L)P compared to Cys variants. Also, there was less relevant evidence highlighting the pathogenicity of these functional regions. In fact, some submitters classified these variants as VUS in part because they did not occur at a Cys. Thus, there appears to be a decreased awareness regarding the functional importance of some calcium-binding, glycine/alanine, and interdomain packing residues, as well as critical variants in the FUN domain. A caveat to this analysis is that this data is somewhat limited in its analytical power since only 59% of submitters included substantive evidence with their variant submissions, thus it is unclear what criteria were used to classify the other 41% of variants.

In addition to the Cys versus non-Cys variant classification challenges, we observed several major themes that occurred with misclassification of critical FBN1 residues. First, misclassification occurred more frequently in variants that were last evaluated several years ago (e.g., pre-2015). For these variants, reevaluation is likely needed by the submitter. Second, there were not only the expected interlaboratory inconsistencies in variant classification and recognition of critical region variants, but also intralaboratory inconsistencies. The age of the classifications sometimes appeared to play a role in intralaboratory variability of classification or evidence, but was not always a factor.

There were several noncritical residue variants that had conflicting classifications and/or were deemed to be misclassified largely because of the MAF of the variant. In fact, based largely on the utilization of an appropriate MAF cutoff, we were able to greatly reduce both the number of (L)P calls (from 105 to 38; 63.8% decrease of noncritical missense variants), as well as decrease the number of reportable noncritical variants (from 423 to 322; 23.9% decrease) that might otherwise have cluttered a report. It has been pointed out by others that there is a critical need to provide guidance surrounding MAF thresholds [7, 49]. Our results demonstrate this same phenomenon, as well as the need to reevaluate variants when updated MAF data and thresholds are made available. It is also worthwhile to point out that the evaluation of updated MAF data was helpful in determining that a serine substitution at the second D/N of the calcium-binding region in cbEGF-like domains may be tolerated.

Although the main focus of this study was to evaluate classification (and classification evidence) of FBN1 missense variants, we noted that many ClinVar entries did not contain substantive evidence to support their classification. In fact, only 59.3% (267/450) of variant submissions from reportable variants at critical residues contained evidence. This percentage is similar to what was reported by Groth et al. [3], and was also a criticism in their report. We agree with their assessment that “variants without supporting data are of very little value”. Indeed, it was impossible for us to assess the validity of a submitter’s variant classification without any context with regards to criteria utilized for classification. When supporting data such as patient phenotype, case counts, literature citations, gene-specific details (e.g., critical region), and ACMG/AMP criteria used to classify the variant are included in a ClinVar entry, this adds value to clinicians and laboratorians alike for the process of variant curation and clinical decision-making. However, this information must also be used with discretion, since the NIH does not independently verify information submitted to ClinVar.

A recent study of a subset of FBN1 variants in four variant databases (HGMD, UMD-FBN1, ClinVar, and UniProt) concluded that current variant databases could not be relied upon for a genetic diagnosis of MFS due to incorrectly interpreted conclusions on variants [3]. As discussed earlier, Muiño-Mosquera et al. [9] also has published on the importance of applying disease- and gene-specific guidelines for a more uniform interpretation of FBN1 variants. Pepin et al. [2] likewise highlighted that, in the context of variant curation and interpretation, there should be collaboration between clinical laboratories, disease-, and gene-specific experts. With expanding genetic analyses becoming increasingly available and offered by multiple laboratories, our results demonstrate the importance of utilizing gene-based knowledge and applying appropriate ACMG/AMP criteria for variant classification, as well as updating classifications as more evidence becomes available, not just for FBN1, but other genes as well. Laboratories that perform genetic testing, including exome and genome sequencing, that includes FBN1 analysis can use the supplemental table and Fig. 1 published here to aid in classification of previously described and novel FBN1 variants, respectively.

While having phenotypic data available while performing variant evaluations is ideal, the reality is that this is oftentimes not the case for many clinical laboratories. Thus, our approach demonstrates that FBN1 variant classification can be refined without always having the context of phenotype available, a situation that most clinical laboratories can relate to. Our results further emphasize, as others have previously shown, that while many variants are correctly classified in ClinVar, there is room for improvement. Collecting gene-based evidence to apply to variant classification is not a trivial task, and clinical laboratories may lack the resources to compile such evidence for each gene they encounter. Thus, articles such as this may serve as a useful resource for clinical laboratories performing FBN1 testing as well as provide a perspective to the clinical community regarding limitations in inter- (and in some cases, intra-) laboratory variant classification.