Whole Exome Sequencing for Familial Bicuspid Aortic Valve Identifies Putative VariantsCLINICAL PERSPECTIVE
Background—Bicuspid aortic valve (BAV) is the most common congenital cardiovascular malformation. Although highly heritable, few causal variants have been identified. The purpose of this study was to identify genetic variants underlying BAV by whole exome sequencing a multiplex BAV kindred.
Methods and Results—Whole exome sequencing was performed on 17 individuals from a single family (BAV=3; other cardiovascular malformation, 3). Postvariant calling error control metrics were established after examining the relationship between Mendelian inheritance error rate and coverage, quality score, and call rate. To determine the most effective approach to identifying susceptibility variants from among 54 674 variants passing error control metrics, we evaluated 3 variant selection strategies frequently used in whole exome sequencing studies plus extended family linkage. No putative rare, high-effect variants were identified in all affected but no unaffected individuals. Eight high-effect variants were identified by ≥2 of the commonly used selection strategies; however, these were either common in the general population (>10%) or present in the majority of the unaffected family members. However, using extended family linkage, 3 synonymous variants were identified; all 3 variants were identified by at least one other strategy.
Conclusions—These results suggest that traditional whole exome sequencing approaches, which assume causal variants alter coding sense, may be insufficient for BAV and other complex traits. Identification of disease-associated variants is facilitated by the use of segregation within families.
Bicuspid aortic valve (BAV), the most common cardiovascular malformation (CVM), occurs in ≈2% of the population.1–4 BAV (Online Mendelian Inheritance in Man 109730) describes a common, congenital malformation in which the aortic valve has 2 rather than 3 semilunar cusps. Several BAV-associated phenotypes have been described with clinical importance. For example, BAV is frequently found in subjects with aortic valve disease.5–8 Furthermore, BAV is associated with specific CVMs (eg, coarctation of the aorta [50%–80%] and ventricular septal defect [20%]).9–11 Finally, aortic root dilation is common in BAV even in hemodynamically normal valves.12–14
Clinical Perspective on p 683
On the basis of large multiplex families with BAV, autosomal dominant inheritance with reduced penetrance has been proposed,15–17 but only 2 genes, KCNJ218 and NOTCH1,19 that account for <5% of BAV cases have been identified. Reasons to explain this apparent lack of success include that there are many genes involved in cardiac valve development (genetic heterogeneity), a focus on protein-changing variants in known candidate genes, and mounting evidence that BAV exhibits complex inheritance (ie, may not be because of a single genetic variant). In previous studies, we identified BAV as heritable20 but genetically heterogeneous.21,22 Our results21,22 and those of others23 question simple Mendelian inheritance and implicate BAV as a complex trait.
Despite strong evidence for a genetic basis for BAV, the genetic origins remain largely unknown. Thus, our objective was to identify genetic variants underlying BAV by whole exome sequencing (WES) a multiplex BAV kindred. Because linkage studies21–23 have identified loci that segregate with BAV, we hypothesize that a family-based approach can identify variants that segregate with BAV within a family.
The proband was a pediatric patient with a diagnosis of BAV, self-identified as white. The proband’s family members were invited to participate in a family-based genetic study, and a detailed 3-generation family history was obtained (Figure 1). A complete medical history and blood sample were obtained from each participant. Medical records, including echocardiograms, were obtained for individuals who had previously received cardiology evaluation. For individuals without previous study, echocardiograms were performed.20 For every new affected individual identified, all first-degree relatives were evaluated.20 The study was approved by the Institutional Review Board of Cincinnati Children’s Hospital, and subjects or their parents gave informed consent as appropriate.
Affected status of family members was defined using strict and broad phenotype criteria. Family members without a history of cardiovascular disease and no echocardiographic evidence of a CVM were considered unaffected. In the strict phenotype analysis, family members with BAV were considered affected; individuals with other CVM that was not BAV were considered to have unknown status and were excluded from relevant parts of the analysis. However, because other CVM in family members could be considered to have a developmental origin, and thus, potentially share a genetic cause with BAV, a broadly defined affected status included family members with any type of CVM (eg, BAV and dilated aorta).
Whole Exome Sequencing
WES was performed on all 17 family members at the Genetic Variation and Gene Discovery Core of Cincinnati Children’s Hospital. One microgram of double-stranded DNA (blood) was used. Quantity was determined by Invitrogen Qubit (Life Technologies, Grand Island, NY) high sensitivity spectro-fluorometric measurement. DNA was sheared by sonication to an average size of 200 bp on a Diagenode Bioruptor (Diagenode Inc, Denville, NJ). Library construction was performed using Illumina TruSeq DNA Sample Preparation kit (Illumina Inc, San Diego, CA) with a size selection at 350 bp after adapter ligation. One microgram of genomic library was recovered for exome enrichment using Nimblegen SeqCap Exome kit (Roche Nimblegen, Inc, Madison, WI). Enriched libraries were sequenced on an Illumina HiSeq2000 (Illumina Inc), generating ≈30 million paired end reads of 100 bases each per sample, corresponding to an average coverage of ×30.
Variant calling was performed with Genome Analysis Toolkit (GATK) version 1.3-2.24 Samples were individually preprocessed by realigning reads around putative indels using GATK’s IndelRealigner tool, marking putative polymerase chain reaction duplicate reads with Picard’s MarkDuplicates tool and by recalibrating base quality scores and calculating Base Alignment Quality scores with GATK’s CountCovariates and TableRecalibration tools. After preprocessing, samples were jointly processed with UnifiedGenotyper to generate initial variant calls. Variants were then filtered using GATK Variant Quality Score Recalibration.25
Empirical Error Control
Mendelian Inheritance Error Detection
Mendelian inheritance errors (MIEs) describe alleles in an individual, which could not have been received from either biological parent by Mendelian inheritance. MIEs were identified using R package trio applied to trio subfamilies of the extended family (Figure 1). Presumably, MIEs are genotyping errors. However using the MIE filter eliminates de novo variants. Because this family exhibits disease segregation across generations, a de novo variant contributing to disease is not expected.
Various metrics have been proposed to improve WES data quality. Specifically, sufficient sequencing coverage is recognized to affect sequence quality; however, there is no consensus threshold.25,26 Previous sequencing studies have highlighted the importance of quality score that accounts for sequencing technologies, read length, machine cycle, and sequence context.25,27 From high throughput genotyping data, it is clear that call rates provide an important quality control metric.28
Bioinformatic Prediction of Putative Functionality
Variants were annotated using SnpEff version 2.0.5,29 the GRCh37.64 database, and the –no-upstream and –no-downstream options. Accordingly, variant effect was considered to be high, moderate, low, or modifier. High effect variants are those that drastically alter protein sequence (eg, stop-gain and splice site variants). A moderate effect is assigned to all other variants that change the protein sequence (eg, missense), a low effect to synonymous variants, and a modifier effect to variants that lie outside coding regions.
Variant Selection Strategies
Given the large amount of sequence data, we sought options that would streamline variant selection. Although BAV is a complex trait, this multiplex family has a strong autosomal dominant segregation. Thus, we expect that the most promising variants will segregate with disease. Furthermore, given the low frequency of BAV in the general population, variants that are common (minor allele frequency [MAF], ≥10%) in a public data set (GRCh37.64) are of lower priority. Variants not in the public databases were assumed to have MAF<10%. Among segregating variants, the most promising candidates are those predicted to have high functional effect and MAF<10%. Lower priority variants may have lower predicted functional effect and MAF<10%. Finally, variants that do not segregate with disease are low priority regardless of MAF.
Three selection strategies frequently used in WES studies plus extended family linkage were compared among variants passing empirical error control. For each of the 4 strategies, primary screens focused on the strict phenotype, which designated BAV as affected and any other CVM as unknown. Secondarily, to understand the effect of using a broad phenotype definition, variant selection was also performed using the broad phenotype, which designated both BAV and other CVM as affected in linkage to compare consistency.
As previous WES have successfully identified variants predicted to have high functional effect, we initially restricted the analysis to coding sense variants. First, we sought to identify variants present in affected but not in unaffected individuals. As reduced penetrance may limit the success, we also compared affected individuals (II-5, III-3, and III-6) with unaffected individuals who married (outbloods, II-4, II-12) into a family with affected individuals, thus representing a different lineage.
Affected First Cousins
To mimic a distantly related family members search strategy, we performed variant filtering based on affected first cousins both having BAV (III-3 and III-6; Figure 1). In this approach, variants shared between cousins were selected without regard for functionality.
Selection of variants was based on segregation in a nuclear subfamily. For the strict phenotype, there was a subfamily with parents and 3 children (II-4, II-5, III-1, III-2, and III-3; Figure 1). Both the father (II-5) and a daughter (III-3) had BAV. Variants selected by this strategy were present in affected individuals and absent in unaffected (ie, affected and unaffected individuals could not carry the same variant).
Extended Family Linkage
Because BAV is a complex trait, we followed the basic principle that identified variants as potentially causal if they were present in more affected individuals than in unaffected individuals. Using our data, an alternate variant frequency in affected minus alternate variant frequency in unaffected (ALTDiff) individuals was estimated. ALTDiff has the beneficial property of not relying on inheritance models and requires no assumptions on penetrance. Two-point parametric linkage analysis was performed using Superlink30 on variants for which ALTDiff≥0.10. BAV was modeled as 80% penetrant autosomal dominant disease with disease-allele frequency of 0.01. For chromosomes that exhibited evidence of linkage, we examined all variants to determine whether an ALTDiff of 0.10 points was reasonable.
Validation of Genetic Variants
Confirmation of genetic variants segregating with disease was obtained by genotyping using the TaqMan 5′ nuclease assays. Single nucleotide variation probes were ordered from Applied Biosystems, and 15 ng of genomic DNA was amplified on an ABI 7900HT real-time polymerase chain reaction machine for allelic discrimination. Genotype calling was performed using SDS2.3 (Applied Biosystems, Foster City, CA).
We recruited 17 members of single kindred (Figure 1). Among 3 generations of participants, there were 3 individuals with BAV (II-5, III-3, and III-6). In addition to BAV, subject II-5 also had a ventricular septal defect and had undergone aortic valve replacement at the age of 35 years for aortic stenosis. Three family members had other CVM, including abnormal aortic valve that did not meet criteria for BAV (II-11), dilated aortic root (II-7), and AV septal defect (III-4). BAV was designated as a strict phenotype; individuals with other types of CVM were considered affected using the broad phenotype.
Results of WES
The Variant Call Format file generated by GATK listed 154 442 variants identified in 17 individuals, and 128 272 variants passed GATK quality control. Among variants passing GATK quality control, 2393 variants were located on X or Y chromosomes (2378 and 15, respectively); we eliminated these variants from further analysis because our phenotype of interest did not exhibit sex-linked inheritance. In addition 15 178 variants were deletions or insertions (indels). Indels were not analyzed further given the challenges of genotype calling.31
Mean sequence coverage was 40.5, but coverage was highly skewed (Figure 2A). Median coverage was 26 with an interquartile range of 11 to 54. The quality score from GATK indicated overall good quality (median, 96.2; interquartile range, 36.1–99.0; Figure 2B). The mean individual call rate was 97.7% (median, 97.7; interquartile range, 97.1–98.3). The mean variant call rate was 97.7% (median, 100; interquartile range, 100–100; Figure 2C). Overall, these results support a good quality sequencing experiment. However, 1 individual (III-4) had lower coverage and quality before quality control (Figure I in the Data Supplement); this individual did not have BAV thus was not used in the primary discovery set.
Development of Metrics for Error Control
To identify metrics and thresholds, which would be useful for WES data, 110 701 autosomal binary single nucleotide variations were evaluated in 5 trio subfamilies of the extended family. Using the GATK genotype calls, the MIE rate was 16.8%. MIEs were distributed evenly across the subfamilies. MIEs were increased with both low and extremely high coverage, low quality score, and intermediate call rates (Figure 2).
By examining the overlap between MIE and the error control metrics, we identified the following thresholds that provide a balance between MIE and single nucleotide variation removal: coverage ≥20 and ≤250; quality score ≥20; and call rates >80%. After applying these postgenotype error control filters, MIE was 2.8%. Thus, error control metrics were defined by the relationship between MIE and coverage, quality score, and call rate. After applying error control metrics, 54 674 variants remained that equates to over a 50% reduction in the number of markers. Of note, the majority of markers were removed because of low coverage; indeed 25% of the markers had coverage <11 reads even though the mean coverage was 40. This skewed coverage was inherent in the exome enrichment kit used.
Variant Selection Strategies
Because WES has demonstrated notable success identifying deleterious protein-coding variants, the initial analysis was restricted to variants altering coding sense unique to affected individuals. Of 10 351 variants with altered coding sense, none were present in all affecteds but no unaffecteds. To account for incomplete penetrance, the analysis was restricted to 3 individuals with BAV (II-5, III-3, and III-6) and 2 outbloods (II-4 and II-12). There were 266 variants altering coding sense shared in common between the 3 individuals with BAV but not present in the outbloods (Table I in the Data Supplement). Only 12 of these variants were present in <20% of the unaffecteds Of these, only 1 (rs12139100) was predicted to be high effect but this variant had MAF>10% in the general population, thus making it an unlikely candidate. Only 4 of the variants (rs150868809 in MUL1 and rs12118933 in ASTN1 and two novel loci at chromosome 16 position 55735817 in SLC6A6 and chromosome 3 position 187451356 in RP11-211G3.3) had MAF<10% in the general population. Importantly, using linkage analysis, there was only nominal support (logarithm of odds [LOD]=1.6 for each). Given the nominal support from linkage and modest bioinformatic effect, these variants were not considered further.
When comparing selection strategies unrestricted with respect to coding sense, affected cousins identified the largest number of variants and linkage identified the least (Table). Overall, the majority of variants were identified by a single selection strategy. Eight high-effect variants overlapped between ≥2 variant selection strategies. All 8 were identified using the affected cousins; 6 of these were also found in affected lineage and 3 were found in nuclear subfamily (1 overlapped all 3; rs12139100). Because 6 of the 8 variants were common in the general population (MAF>10%) and the other 2 were found in a majority of the unaffected relatives, the enthusiasm for these variants was diminished. Although no high-effect variants were identified using linkage (Table), all linkage variants were identified by at least one other strategy (Table I in the Data Supplement). Given the consistency of linkage with the other strategies, we focused on the variants identified from linkage with a general population MAF<10%.
Variants Identified From Linkage
The extended family exhibited suggestive evidence of linkage (LOD=2.3) at 3 single nucleotide variation: rs148192660 (9q34), rs45450992 (9q34), and rs2288474 (5p15.2; Figure 3). Although none of these variants reached the threshold for significance (LOD≥3.0), given the family structure and the marker informativeness, reaching a LOD=3.0 was highly improbable. Variant rs2288474 (MAFdbSNP=0.087) is a synonymous variant in ANKH (inorganic pyrophosphate transport regulator [Homo sapiens]). Variant rs148192660 (MAFdbSNP=0.0023) is a synonymous variant in TRAF2 (tumor necrosis factor receptor–associated factor 2). Variant rs45450992 (MAFdbSNP=0.056) is a synonymous variant in EHMT1 (Euchromatic histone-lysine N-methyltransferase). These genotypes were confirmed using TaqMan with 100% concordance.
Effect of Phenotype Definition on Variant Discovery
When using the strict phenotype, all 3 affected individuals carried the 3 variants from linkage. To determine the extent to which these findings held up using the broad phenotype, we evaluated the additional 3 individuals. Individual II-11 has an abnormal aortic valve that did not meet BAV criteria but was heterozygous at the same three loci (Figure 4) as individuals with BAV. Individual II-7 who had dilated aorta, which is frequently associated with BAV,14 carried the minor allele for the 2 variants on chromosome 9. However, individual III-4, who has an atrioventricular septal defect did not carry any of the minor alleles. In addition, an unaffected child (III-1) carries the minor allele for the 2 variants on chromosome 9, which suggests possible reduced or age-dependent penetrance or that only when an individual has all 3 variants will BAV result.
Using a multiplex BAV kindred and WES, we failed to identify high-effect coding sense variants underlying BAV but identified 3 synonymous variants, which segregated with BAV. Importantly, each of these was also identified using an alternate search strategy. Although these variants are unlikely candidates (synonymous variants, not in candidate genes), they are consistent with the inheritance of BAV. Surprisingly, broadening the phenotype definition weakened support for these variants. Thus for complex traits, such as BAV, using large families and a strict phenotype definition can be useful for identification of noncoding sense variants in WES data.
Effect of Variant Selection Strategy and Phenotype Definition
Strongest Evidence for Effect is in Synonymous Variants
We failed to find high-effect variants, which were identified consistently across several selection strategies that met our assumptions of not common in the general population (MAF<10%) and not present in a high frequency in unaffected family members. Rather our strongest evidence was for 3 synonymous variants. These 3 variants were not common (MAF<10%) and were present in <20% of our unaffecteds. Each of these variants was identified using linkage (LOD>2.0), as well as at least one other selection strategy. The failure to identify high-effect variants is in contrast to previous WES studies of Mendelian traits, which have identified protein-changing variants.32–36 However, it is important to note that clinical exome sequencing currently identifies such variants for only 25% of the cases,37 raising the possibility that the remaining 75% of cases include noncoding sense variants, as exemplified by our results.
Implications for Gene Discovery in BAV
On the basis of our results, we found that bioinformatic filtering approaches commonly used in WES may not be sufficient for complex traits, such as BAV. Indeed, there was no overlap in the putative high-effect variants identified between the commonly used strategies, when underlying assumptions were met (not common in the general population and not frequent in unaffected family members). However, there was overlap between synonymous variants. The importance of noncoding sense variants, such as synonymous variants, has been appreciated since the completion of ENCODE (Encyclopedia of DNA Elements).38 Unfortunately, traditional bioinformatic filtering does not yet use this information. Without the reliance of bioinformatic filtering, both the affected cousins and nuclear family approaches yielded too many variants for practical follow-up. As such, the use of multiplex families may help narrow WES data to relevant variants segregating with disease to overcome concerns of discovery of novel variants.
Putative BAV Variants: Biological Plausibility of Involvement in Valve Development
Heart valves develop from embryonic endocardial cushions located at the AV canal and cardiac outflow tract. Cushions are derived from endothelium and interstitial tissue, which is remodeled into mature valves.39 This remodeling is prominent during late gestation and continues into postnatal life.40 Previously, variants in NOTCH1 have been implicated in BAV19; however, in our study, linkage did not support a role of NOTCH1 variants in BAV. Indeed, previously we reported that this family linked to chromosome 9 in the region of NOTCH1 but that Sanger sequencing failed to identify any missense variants of interest.21 Nonetheless, NOTCH1 is located 340 727 bases from TRAF2 and 1 073 206 bases from EHMT1. TRAF2 encodes a protein belonging to the tumor necrosis factor receptor–associated factor protein family, which functions as a mediator of antiapoptotic signals. Although there is no evidence that TRAF2 is involved in heart development, TRAF2 interacts with SMAD441 and endothelial-specific knockout of SMAD4 affects valve development.42 In addition, based on the UCSC Genome Browser HMR Conserved Transcription Factor Binding Sites track,43 rs148192660 (TRAF2) is located directly adjacent to a conserved predicted binding site for MEF2A, a transcription factor expressed in the heart valve.44 Defects in EHMT1 occur in a chromosome 9q34.3 microdeletion syndrome, which is associated with CVM.45 Furthermore, rs45450992 (EHMT1) is predicted to alter a Msx-1 binding motif in H1 embryonic stem cells (haploreg).46 Importantly, Msx-1 is expressed in heart valve and mediates epithelial mesenchymal interactions in AV cushions.44,47
On chromosome 5, an ANKH variant (rs2288474) was suggestive of linkage. ANKH encodes a protein that is a growth factor-regulated delayed–early response gene in mammalian cells.48 Although ANKH has not been associated with cardiac disease, mutations in ANKH have been shown to cause familial calcium pyrophosphate crystal deposition49 and craniometaphyseal dysplasia.50 Furthermore, ANKH is expressed in heart and is stimulated by EGF,48 which is known to inhibit development of heart valve primordia. In addition, rs2288474 was part of a haplotype associated with osteoprotegerin levels51 and osteoprotegerin is expressed in valve leaflets.52 However, these results must be interpreted with caution because biological links could be made with many different genes based on limited evidence. Furthermore, as exome sequencing captures <1% of the variation of the genome, it is possible that regions not covered may harbor causal variants. Thus, additional studies are necessary.
Broadening of the Phenotype Alters Findings
These results also demonstrate that incorporating phenotypic plasticity into WES may impair the ability to detect variants. When using the strict phenotype, all 3 affected individuals carried the 3 variants. To determine the extent to which these findings held up using the broader phenotype classification, we evaluated the 3 individuals with other CVM. Evaluating the co-occurrence of all 3 variants indicated some genotype–phenotype overlap (eg, an individual with abnormal aortic valve that did not meet BAV criteria had the same 3 variants as individuals with BAV). However, an individual with atrioventricular septal defect did not carry any of the minor alleles. Given the known differences between development of the AV canal and the formation of semilunar valves, this is not surprising. However, an individual with dilated aorta, which is frequently associated with BAV,14 carried the minor allele for 2 of the 3 variants. These results in combination with the success of WES for Mendelian traits32–36 suggest that stringent criteria for phenotype definition is essential.
Although BAV has strong evidence of a genetic basis, successful identification of disease-causing variants has been limited. Part of the challenge with identifying genetic underpinnings of BAV is that it is a complex trait, implicating involvement of >1 genetic variant. To take a different approach than previous studies, we performed WES on a multiplex kindred with multiple individuals with BAV. We evaluated 3 variant selection strategies successfully used in WES studies but identified no putative high-effect variants, which segregated with disease. Through a combination of error control and a strict phenotype definition, we were able to identify 3 variants using extended family linkage. Although none of the 3 variants were predicted to be high effect, they are in genes belonging to networks implicated in relevant developmental pathways. Additional studies are necessary to replicate the genetic findings and identify potential developmental mechanisms for these results, the approach provides a framework for analysis of WES family data in the context of BAV and other complex traits.
We thank the study participants.
Sources of Funding
This study was funded, in part, by P50 HL74728 and P30 DK078392.
The Data Supplement is available at http://circgenetics.ahajournals.org/lookup/suppl/doi:10.1161/CIRCGENETICS.114.000526/-/DC1.
- Received September 27, 2013.
- Accepted June 16, 2014.
- © 2014 American Heart Association, Inc.
- Ward C
- Roberts WC,
- Ko JM
- Rosamond W,
- Flegal K,
- Friday G,
- Furie K,
- Go A,
- Greenlund K,
- et al
- Hor KN,
- Border WL,
- Cripe LH,
- Benson DW,
- Hinton RB
- Mack G,
- Silberbach M
- Emanuel R,
- Withers R,
- O’Brien K,
- Ross P,
- Feizi O
- Wessels MW,
- Berger RM,
- Frohn-Mulder IM,
- Roos-Hesselink JW,
- Hoogeboom JJ,
- Mancini GS,
- et al
- McBride KL,
- Pignatelli R,
- Lewin M,
- Ho T,
- Fernbach S,
- Menesses A,
- et al
- McKenna A,
- Hanna M,
- Banks E,
- Sivachenko A,
- Cibulskis K,
- Kernytsky A,
- et al
- Fishelson M,
- Geiger D
- Aikawa E,
- Whittaker P,
- Farber M,
- Mendelson K,
- Padera RF,
- Aikawa M,
- et al
- Moskowitz IP,
- Wang J,
- Peterson MA,
- Pu WT,
- Mackinnon AC,
- Oxburgh L,
- et al
- Karolchik D,
- Barber GP,
- Casper J,
- Clawson H,
- Cline MS,
- Diekhans M,
- et al
- Chakraborty S,
- Cheek J,
- Sakthivel B,
- Aronow BJ,
- Yutzey KE
- Stewart DR,
- Kleefstra T
- Ward LD,
- Kellis M
- Shetty R,
- Pepin A,
- Charest A,
- Perron J,
- Doyle D,
- Voisine P,
- et al
Bicuspid aortic valve (BAV) is the most common congenital cardiovascular malformation and is an important clinical problem largely because of great heterogeneity of outcomes. For example, BAV frequently underlies aortic valve disease, aortic root dilation, and is often associated with other cardiovascular malformation (eg, coarctation of the aorta and ventricular septal defect). However, some individuals have minimal or no clinical implications from BAV. Knowledge of the cause of BAV promises to provide insight into improved management and risk stratification strategies. Although highly heritable, the use of simple inheritance models and mutation analysis of candidate genes have identified causal variants in <5% of BAV cases. Thus, in most cases, the genetic cause of BAV is unknown. Recent studies have suggested that BAV is a complex trait. To identify genetic variants underlying BAV, we performed whole exome sequencing in a multiplex BAV kindred. However, we failed to identify high-effect coding sense variants underlying BAV but identified 3 synonymous variants segregating with BAV. Although these variants may seem unlikely candidates (synonymous variants, not in candidate genes), they segregate with the BAV phenotype and are supported by recent studies of the human genome, which highlight the importance of noncoding sense variants especially for complex traits. Surprisingly, broadening the phenotype definition weakened support for these variants. Thus for complex traits, such as BAV, using large families and a strict phenotype definition can be useful for identification of noncoding sense variants in WES data.