Almost half of the patients with TS have CHD (Silberbach et al. 2018), demonstrating that TS is a significant risk factor for CHD. However, the existence of individuals with TS and structurally normal hearts demonstrates that an absent or structurally abnormal second Xchr is not sufficient to cause abnormal heart development. Increased, but variable penetrance may be a sign of interactions between the sensitized genome and modifier genes. Genetic modification of CHD is supported by observations in mice, as CHD incidence in mutant mouse models depends on their genetic background. For example, Gata4± mice have a reported CHD incidence ranging from 12 to 76% (Rajagopal et al. 2007), Nkx2-± between 5 and 50% (Winston et al. 2010), and Tbx5± between 40 and 80% (Bruneau et al. 2001). In addition to incidence, CHD phenotype also varies with respect to mouse genetic background. For example, heterozygous Gata4 mutations in mice can cause endocardial cushion defects, atrial or ventricular septal defects, hypoplastic right ventricle, or cardiomyopathy depending on genetic background (Rajagopal et al. 2007). Similarly, the phenotypic spectrum of GATA4 mutations in humans is broad, indicating that genetic background (i.e., modifiers) may influence both the penetrance and expressivity of CHD risk genes in humans (Rajagopal et al. 2007).

Familial, non-syndromic BAV is phenotypically heterogenous and is inherited in an autosomal dominant pattern with incomplete penetrance (Huntington et al. 1997). Multiple genes/loci have been implicated in familial BAV, including TGFBR2, 5q15-21, TGFBR1, NOTCH1, SMAD6, ACTA2, 13q33-qter, 15q25-q26.1, KCNJ2, and 18q (Martin et al. 2007; Arrington et al. 2008; Gillis et al. 2017). Interestingly, Xchr loci are not strongly implicated in non-syndromic BAV despite the fact that TS has the highest prevalence of BAV among all genetic syndromes. This is also interesting given that non-syndromic BAV is more common in 46XY men (3:1) (Kong et al. 2017), which implies that absent Xchr genes may raise the risk for BAV but does not fully explain the phenotype. It also makes TS an ideal syndrome in which to look for genetic modifiers of BAV given the significantly enriched phenotype and thus increased power to detect rare modifiers.

In this study, we utilized VAAST 2.0, SKAT, and SKAT-O to identify rare autosomal candidate genetic modifiers of BAV in TS. We chose VAAST 2.0 for our bioinformatic analysis given its native integration of amino acid substitution, allele frequency, and phylogenetic conservation data to its variant association testing which improves power and variant prioritization accuracy (Hu et al. 2013). Because VAAST 2.0 was designed to be a general-purpose disease-gene finder capable of identifying both rare and common alleles responsible for both rare and common diseases, we chose to complement the VAAST 2.0 analysis with that of two alternative aggregative variant association tests (i.e., SKAT and SKAT-O).

CRELD1, which was the top-scoring gene in VAAST 2.0 and highly ranked in all 3 gene burden tests, has been implicated in both syndromic and non-syndromic CHD (Ackerman et al. 2012; Robinson et al. 2003; Zatyka et al. 2005), specifically atrioventricular septal defects. Creld1 controls the formation of the atrioventricular cushion and is required for the vascular endothelial growth factor (VEGF)-dependent proliferation of endocardial cells by promoting calcineurin-dependent nuclear translocation of nuclear factor of activated T cells 1 (NFATc1), and thereby, the expression of Nfatc1 target genes (Mass et al. 2014). Nfatc1 null mice and mice lacking NF-ATc1 exclusively in endocardial cells fail to develop mature heart valves and is embryonic lethal. Conditional Creld1 knockouts demonstrate that Creld1 controls the formation of the septa and valves but also the maturation and function of the myocardium at later developmental stages (Beckert et al. 2021). There is no association of CRELD1, NF-ATc, or calcineurin signaling with non-syndromic BAV. However, the aortic valve develops from the endocardial cushions in the outflow tract of the primitive heart tube (Martin et al. 2015). Since Nfatc1 has been implicated in the transcriptional regulation of valvulogenesis, it is plausible that on a distinct genetic background, variants in CRELD1 could yield predominantly a BAV phenotype. Additionally, a set of 4 miRNA (miR-130a, miR-122, miR-486, and miR-718) have been correlated with BAV and aortic dilation (Abu-Halima et al. 2020) in TS and play a role in activating the TGF-β1 pathway and vascular remodeling mediated through vascular endothelial growth factor (VEGF-A) signaling pathway. CRELD1 lies in the VEGF-A pathway, and rare variants in CRELD1 and other genes in the VEGF-A pathway modify the risk of CHD in patients with trisomy 21 (Ackerman et al. 2012). This has been functionally validated in a mouse model of trisomy 21, whereby a null allele of Creld1, which has no heart or other phenotype on a disomic mouse model, greatly increases the occurrence of CHD in a mouse model of trisomy 21 (Li et al. 2012).

Interestingly, the duplication CNV associated with BAV in TS is also associated with conotruncal defects and left-sided lesions in 22q11.2 deletion syndrome suggesting that genetic modifiers of CHD may manifest differently on different high-risk backgrounds (Prakash et al. 2016). CRELD1 has not been implicated in the pathogenesis of non-syndromic, non-familial BAV via GWAS (Helgadottir et al. 2018; Fulmer et al. 2019; Bjornsson et al. 2018; Yang et al. 2017; Gehlen et al. 2022; Gago-Díaz et al. 2017). Thus, the effect of CRELD1 variants may be more penetrant in the setting of Xchr haploinsufficiency, representing an as-of-yet unidentified mechanism contributing to the polygenic heritability of BAV.

CRELD1 is alternatively spliced. The 4 main transcripts expressed in the left ventricle and aorta are ENST00000452070.5, ENST00000397170.7, ENST00000435417.1, and ENST00000383811.7 (https://gtexportal.org/home/gene/CRELD1). The 3 variants in CRELD1 enriched in our TS cases are interesting as the coding consequences for the variants are transcript-dependent. The first variant (c.9943412, G > A) would result in a synonymous proline-proline change in most CRELD1 isoforms, however, in a specific isoform that codes for a shortened protein (337aa vs. 420aa) this change creates a nonsynonymous change of an arginine to a glutamine. The second variant (c.9943399, C > T) results in a nonsense mutation in both the canonical isoform and the cardiac isoforms. Many isoforms end prior to the third variant; in those transcripts it lies in the 3’ untranslated region. Interestingly, it results in missense variation (aspartate to an asparagine) in one isoform that is much larger than the canonical CRELD1 protein (537aa vs. 420aa). The significance of these isoform-specific consequences to BAV is not currently known and is the subject of further study.

Strengths of this study include a large sample size, and while 2 cohorts were assembled, significant quality measures were taken to minimize bias introduced by differences in study design. Weaknesses of this study include a lack of information regarding additional known CHD risk factors, including maternal age and maternal diabetes status. An additional weakness includes the potential to have misclassified BAV status in those we consider controls. Whereas misclassification would have decreased our study power and as such, our robust association of CRELD1 is likely a valid finding. We opted to include all individuals with a karyotypically-confirmed diagnosis of TS given the lack of robust karyotype-phenotype correlation in TS, and the possibility of somatic mosaicism in those with only a 45,X cell line (Gravholt et al. 2022; Youssoufian and Pyeritz 2002).

Next steps include obtaining phenotypic data and DNA from biological parents and siblings of the Iowa cohort to determine whether the variants were inherited or occurred spontaneously and if there is an association with CHD in family members. Recurrence of damaging mutations would be consistent with the premise that genetic risk factors for CHD are present at low frequency in the general population. Unfortunately, it is difficult to recapitulate TS in mice given that a large portion of the human Xchr maps outside the mouse Xchr, the human pseudoautosomal region contains multiple genes that are autosomal in mice (Perry et al. 2001), and substantially more genes escape Xchr inactivation than in mice (15% versus roughly 3%) (Berletch et al. 2010). As BAV is also enriched in other genetic syndromes, such as Loeys-Dietz syndrome, validation could include a similar study investigating rare CRELD1 variation in similarly sensitized BAV populations that have animal models (MacFarlane et al. 2019) available for later functional validation.

In conclusion, we propose that CRELD1 variants show additive effects with haploinsufficiency for Xchr genes to interfere with valvulogenesis and increase the risk for BAV. The sensitized TS background aided the identification of the CRELD1 modifiers, which have not previously been identified as associated with non-syndromic BAV. Further study of the VEGF-A pathway in syndromic and non-syndromic BAV may provide additional insight into the role of CRELD1 in the pathogenesis of BAV.