Genome-wide Association Meta-analysis Supports Genes Involved in Valve and Cardiac Development to Associate with Mitral Valve Prolapse

Abstract

Background.

Mitral valve prolapse (MVP) is a common cardiac valve disease, which affects 1 in 40 in the general population. Previous GWAS have identified six risk loci for MVP. But these loci explained only partially the genetic risk for MVP. We aim to identify additional risk loci for MVP by adding dataset from the UKBiobank.

Methods.

We reanalyzed 1007 cases and 1469 controls from the MVP-France study and 479 cases and 862 controls from the MVP-Nantes study for genotype reimputation using HRC and TOPMed panels. We also incorporated 434 MVP cases and 4,527 controls from the UKBiobank for discovery analyses. Genetic association was conducted using SNPTEST and meta-analyses using METAL. We used FUMA for post-GWAS annotations and MAGMA for gene-based and gene-set analyses.

Results.

We found TOPMed imputation to perform better in terms of accuracy in the lower ranges of minor allele frequency (MAF) below 0.1. Our updated meta-analysis included UKBiobank study for ~8 million common SNPs (MAF>0.01) and replicated the association on Chr2 as the top association signal near TNS1. We identified an additional risk locus on Chr1 (SYT2) and two suggestive risk loci on chr8 (MSRA), and chr19 (FBXO46), all driven by common variants. Gene-based association using MAGMA revealed 6 risk genes for MVP with pronounced expression levels in cardiovascular tissues, especially heart and globally part of enriched GO terms related to cardiac development.

Conclusions.

We report an updated meta-analysis GWAS for MVP using dense imputation coverage and an improved case control sample. We describe several loci and genes with MVP spanning biological mechanisms highly relevant to MVP, especially during valve and heart development.

Introduction

Heart valves are key functional elements of the heart that display specific biological mechanisms in health and disease. During the heart morphogenesis, the mitral valve develops soon after cardiac looping. The complex shape of the mitral valve, with two leaves, allows a very precise balance of force to maintain unidirectional blood flow through the mitral orifice. The importance of valve development in the origin of mitral valve disease has been established through the discovery of the genetic causes of rare syndromes such as Marfan, Loeys-Dietz and Ehlers-Danlos and familial non-syndromic cases.^1,2 The causal genes play key roles in extracellular matrix deposition and organization, being influenced by TGF beta and/or ciliogenic signaling nodes.

The adult valve can lose its flexibility under the action of permanent mechanical stress and a degenerative process gradually takes place leading to prolapse (MVP), and in many patients, the incapacity of the valve to close. The resulting mitral regurgitation requires surgery repair or replacement, as it greatly increases the risk of heart failure, arrhythmia and even sudden death.³ MVP is common, affecting approximately 1 in 40 individuals in the general population.³ As for many heart diseases, the establishment of MVP occurs as a result of mild dysfunctions of the many complex biological mechanisms required during development and/or valve function. We have shown through family studies the requirement of DCHS1, a member of the cadherin super family, for cell alignment during valve development.¹ More recently, we have shown that loss of primary cilia during development leads to progressive myxomatous degeneration of the mitral valve in mice and humans.² Using GWAS, we have identified predisposition loci,^{1, 4} particularly those near TNS1, are involved in cell adhesion. This finding further supported the importance of cytoskeleton organization revealed by the study of the polyvalvulopathy syndrome caused by FLNA mutations.⁵ However, taken together, these loci explain only a small fraction of the genetic contribution to MVP. For instance, the relevance of fine regulation of valve and cardiac development mechanisms is established to be at play for early onset syndromic and non-syndromic valve disease. However, this is unknown for late onset and aging-related valve disease. Identification of additional risk loci will likely provide a better and more complete understanding of the genetic and biological basis of MVP.

One of the limitations of the genetic study of MVP is the lack of large cohorts with genome-wide genotyping that would allow increasing the power to discover new predisposition genes of the current GWAS involving only 1,500 patients.¹ In addition, since our initial study, it has become possible to achieve much denser and more accurate GWAS through the high-density genetic imputation panels provided by HRC,⁶ and more recently TopMED⁷ consortia. These panels offer a theoretical increase in power, genomic coverage, and fine mapping, notably through the study of low-frequency variants.^{7, 8}

In this work, we first compare the imputation performance of the newly generated HRC and TOPMed panels in the context of our cohorts. We thus describe comparable results using these two panels, in favor of a better coverage for low-frequency variants for the TOPMed panel. Then, we performed a GWAS meta-analysis including a new case-control study defined using the UKbiobank resource.⁹ We replicate the TNS1 locus, and described four new suggestive loci. The gene-based association analysis identifies several new loci that inform established and original mechanisms for the biology underlying the genetic risk for MVP.

Materials and Methods

Details on the populations and methods are available in the Supplemental methods. The experimental data that support the findings of this study are available from the corresponding author upon request. Summary statistics of the meta-analysis will be publicly available soon after article acceptance. MVP-France approvals were obtained from CPP Ile-de-France VI, (approval n° 60–08, June 25th, 2008), the “Commission Informatique et Libertés” (CNIL) (approval n° 908359, October 14th, 2008) and the French Ministry of Health (ID-RCB: 2008-A00568–47) and was registered on the ClinicalTrial.Gov website (protocol ID: 2008–01). MVP-Nantes was approved by CPP Ouest IV-Nantes (N°215/2013, du 06/03/2013), by: 913630, of 16/10/2014) and was registered on the ClinircalTrials.gov website (protocol ID: NCT03884426).

Results

Imputation accuracy between HRC and TopMED in MVP GWAS datasets

We first aimed to assess the usefulness of using larger imputation reference panels in the two French MVP case controls studies. MVP-France study (MVP-F) included 1,007 cases and 1,469 controls with 492,438 genotyped variants. MVP-Nantes study (MVP-N) included 479 cases and 862 controls with 370,697 genotyped variants. We generated 229,973,672 and 230,053,813 imputed variants in MVP-F and MVP-N, respectively, from the TOPMed reference panel, which represent 5.8-fold more variants compared to HRC (39,117,105 in MVP-F and MVP-N) (Figure 1A). However, most of these variants had an imputation score Rsq smaller than 0.3: 84% (MVP-F-TOPMed), 88% (MVP-N-TOPMed), 51% (MVP-F-HRC), and 60% (MVP-N- HRC) (Figure 1A). Compared to MVP-N, the relatively larger sample size of MVP-F study generates a slightly larger number of well-imputed variants (Figure 1A).

Figure 1. Comparison of the imputation quality of the results for MVP-F and MVP-N using HRC and TOPMed as reference panel, respectively.

(A). The numbers of variants generated using different imputation accuracy thresholds; (B). The imputation quality at different MAF region using all the cleaned SNPs (Rsq > 0.3) in cohorts MVP-Paris and MVP-Nantes. Rsq: imputation accuracy; MVP-F: MVP-France cases-control study. MVP-N: MVP-Nantes case-control study.

We then classified MAF into five regions ((0,0.01], (0.01,0.05], (0.05,0.1], (0.1,0.2] and (0.2,0.5]). Most of the variants with Rsq < 0.3 are rare variants which minor allele frequency (MAF) is smaller than 0.01, as illustrated for all SNPs generated on chromosome 22 (Supplementary figure I A,B). We also found that TOPMed panel imputed variants present, on average, a better accuracy for low frequency ranges (0,0.01], (0.01,0.05] and (0.05,0.1], while relatively stable accuracy at MAF ranges (0.1,0.2] and (0.2,0.5] compared to HRC panel in both studies (Figure 1B). Consistent results were also found at the variant level (Supplementary figure I C,D). We randomly selected 173804 variants (approximately 8000 for each chromosome) to compare the imputation accuracy between the two panels. The low frequency ranges (0,0.01] represented 75% of SNPs, compared to other MAF categories (Supplementary Figure I C). The ratios of the number of variant where TOPMed Rsq was higher than HRC Rsq divided by the number of variants where HRC Rsq was higher than TOPMed Rsq were nearly 1.5 for all MAF categories (Supplementary Figure I D). Taken together, we use TOPMed as the imputation panel to allow analyzing more well imputed variants, especially in the low frequency (0.01<MAF<0.1) category.

Genetic association analyses in French and UKBiobank case control studies

We meta-analyzed three GWAS involving a total of 1,920 MVP cases and 6,858 controls and ~8 million (8,021,974) genotyped or imputed common SNPs (MAF>0.01). We confirmed a deviation from the expected levels of significance in this updated meta-analysis (Figure 2A).

Figure 2. SNP-based association results with MVP in a GWAS meta-analysis involving MVP-F, MVP-N and MVP-UKB case-control studies (lambda GC = 1.08).

(A) The Q-Q plot represents the expected (x-axis) versus the observed (y-axis) P-values; (B) Manhattan plot summarizes the -log10 (P) of each SNP by chromosome obtained from the GWAS meta-analysis. The blue line indicates the threshold for suggestive association (P-value < 1 × 10⁻⁶) and the red line indicates the genome-wide significance threshold (P < 5 × 10⁻⁸). Both plots represent the association of ~ 8 million SNPs genotyped and imputed using TOPMed panel.

We first looked-up the association with MVP in the UKBB dataset of the previously reported 6 loci in Dina et al.,¹ We found nominal association for 2 loci and consistent direction of effects of 5 out of 6 loci in the UKBB case control study (Supplementary Table I).

The strongest association signal was observed on chromosome 2 at the TNS1 locus, where we report 46 variants with a genome-wide significant P-value < 5×10⁻⁸ (Figure 2B). In addition to successfully replicating the previously reported top associated SNP rs12465515 (P-value = 8.61×10⁻¹⁰; OR = 1.30[1.19–1.42]), we found a new top associated variant on chromosome 2 (lead SNP: rs7595393; P-value = 4.68×10⁻¹⁰; OR = 1.31[1.20–1.42], Table 1, Supplementary figure II). Both SNPs are highly correlated (r²=0.92, 1000G Phase 3: CEU). As expected, conditional analyses both on rs12465515 and rs7595393 at the TNS1 locus resulted in disappearance of the association signal (Figure 3). Functional annotation of 107 SNPs in LD (r2>0.6) with 4 independent significant SNPs (ISS) at this locus indicated the existence of 4 SNPs predicted to be deleterious alleles (CADD score > 12) and 7 SNPs likely to lie within regulatory elements according to the RDB score (RDB score > 3a) (Supplementary Table II).

Table 1.

Associations of top SNPs with MVP obtained in the GWAS meta-analysis.

					MVP-France	MVP-Nantes	MVP-UKB	ALL
RSID	CHR	PosB38	RA*	Freq	P-value OR † [95%CI ‡]	P-value OR [95%CI]	P-value OR [95%CI]	P-value OR [95%CI]	Heterogeneity P-value
rs199723025	1	202712600	A	0.05	1.40×10−4 1.61 [1.25–2.08]	0.31 1.26[0.88–1.83]	1.92×10−5 1.72[1.34–2.22]	4.55×10−8 1.69[1.40–2.04]	0.26
rs7595393	2	217006531	G	0.38	1.40×10−6 1.34 [1.19–1.52]	9.69×10−3 1.44 [1.22–1.70]	7.40×10−4 1.27 [1.10–1.48]	4.68×10−10 1.31 [1.20–1.42]	0.73
rs56028519	8	10341263	A	0.74	2.94×10−3 1.19 [1.04–1.35]	1.21×10−2 1.31 [1.10–1.57]	1.11×10−4 1.34 [1.16–1.56]	1.04×10−7 1.29 [1.17–1.41]	0.67
rs4802272	19	45732692	A	0.55	2.10×10−3 1.18 [1.06–1.33]	3.39×10−4 1.25 [1.07–1.47]	1.97×10−2 1.18 [1.03–1.36]	6.54×10−7 1.24 [1.14–1.34]	0.39

^*RA: risk allele

^†OR: odds ratio

^‡CI: confidence interval.

Figure 3. Conditional Analysis of the TNS1 locus.

Locus Zoom plots of the TNS1 locus demonstrate conditional analysis of Dina reported lead SNP rs12465515, the top SNP rs7595393 using GCTA-COJO. Linkage disequilibrium of SNPs with the conditioned SNPs is based on data from MVP-F cohort and is shown by the color of the points. The best signals in each plot are marked as a purple rhombus. (A) Association of SNP with MVP; (B) MVP association conditioned on Dina reported lead SNP rs12465515 (rs12465515 added as an additional covariate); (C) MVP association conditioned on the top SNP rs7595393.

We also report a new genome-wide significant locus and two suggestive and original association signals (Figure 2, Table 1). On chromosome 1, the lead SNP (rs199723025, effect allele frequent (EAF) = 0.05, P-value = 4.55×10⁻⁸, OR = 1.69[1.40–2.04]) is an intergenic deletion variant of Synaptotagmin 2 gene (SYT2) (Table 1, Supplementary figure III). We found that the lead variant is a significant eQTL in atrial appendage and artery aorta (P-value: 2.4×10⁻⁵ and 1×10⁻⁸) for the lysine demethylase 5B gene (KDM5B) involved in DNA stability and repair. The second locus on chromosome 8 (rs56028519, EAF = 0.74, P-value = 1.04×10⁻⁸, OR = 1.29[1.17–1.41]) is an intronic variant of the methionine sulfoxide reductase A gene (MSRA) (Table 1, Supplementary figure IV). The lead SNP rs56028519 is an eQTL of a LncRNA gene (AF131215.2) in heart atrial appendage (P-value=4.5×10⁻⁷) and a LncRNA gene (RP11–981G7.6) in left ventricle (P-value=1.3×10⁻⁷). We found that several SNPs in LD (r²>0.6) with the 2 ISS are located near the regulatory elements, especially rs11249991 and rs11781529 that showed high potential to be regulatory (RDB score > 3a). Four SNPs in total are likely to be deleterious, including rs11783281 (CADD score = 19.34, Supplementary Table II). The third suggestive association signal was located in a particularly gene-rich region on chromosome 19 (rs4802272, EAF = 0.55, P-value = 6.54×10⁻⁷, OR = 1.24[1.14–1.34]) (Table 1, Supplementary figure V). Although the lead SNPs mapped to FBXO46, the association signal spans five genes in total, including, SIX5, FOXA3, RSPH6A, DMPK and DMWD (Supplementary figure V). The lead SNP is an eQTLs of DMPK and DMWD in artery aorta (P-value: 2.1×10⁻⁹, 3.4×10⁻¹¹). Two SNPs in high LD with the lead SNP showed high deleteriousness scores (rs62111759, CADD score = 13.28 and rs672348, CADD score = 15.42) (Supplementary Table II).

Genome-wide gene-based association and pathway analyses

We performed a genome-wide gene-based association analysis using MAGMA (v1.08)¹⁰ to estimate the gene level association on the basis of all SNPs in a gene. This method is a complementary approach to single SNP GWAS analyses and usually reveals genes with locally consistent associated SNPs that individually may do not reach genome-wide significance.

In total, we highlight 18 suggestively associated genes (P-value < 10⁻⁴), including 6 at the genome-wide level (P-value < 2.61×10⁻⁶) (Supplementary Table III, Supplementary Figure VI). We highlight GLIS1 (P=1.86×10⁻⁵) that we have previously identified to associate with MVP using single SNP and pathway analyses,⁴ TGFB2 (P=1.00×10⁻⁵), a high-profile candidate gene for myxomatous valve disease, MSRA (P= 6.31×10⁻⁶), one of the suggestive loci in the SNP GWAS, and TBX5 (P = 1.62×10⁻⁵), a key regulator of heart development (Supplementary Table III). The ten remaining genes mapped into two gene-rich loci on chromosome 17 and 19. On chromosome 17, 3 genes reached gene-wide significance including SRR (P-value = 1.07×10⁻⁷), in addition to TSR1 (P-value = 3.22×10⁻⁷), and SGSM2 (P-value = 2.05×10⁻⁶) (Supplementary Table III). On chromosome 19, we report significant association SIX5 (P-value = 4.30×10⁻⁷), DMPK (P-value = 6.23×10⁻⁷), and DMWD (P-value = 2.43×10⁻⁶) (Supplementary Table III). Tissue expression analyses using the GTEx data resource showed that DMPK and DMWD are ubiquitously expressed, including in the heart atrial appendage and heart left ventricle (Supplementary Figure VI). We note that low frequency variant contributed little to the association of the genes above mentioned, where the associations were driven by common variants (Supplementary Figure VII). Section in situ hybridization lookups validated developmental expression of many of these gene within the mitral valves, including. Syt2, candidate gene at the new locus we describe in the SNP-GWAS (Supplementary Figure VIII).

MAGMA pathway analysis using 9,996 gene sets (GO terms obtained from MsigDB) indicated that several enriched gene sets for MVP associated genes are related to cardiac biology. Among top 10 enriched GO terms we found cardiac ventricle formation (P-value = 1.42×10⁻⁵), cardiac chamber formation (P-value = 3.01×10⁻⁵), cardiac right ventricle morphogenesis (P-value = 2.12×10⁻⁴), cardiac atrium development (P-value = 3.00×10⁻⁴) and cardioblast differentiation (P-value = 6.78×10⁻⁴) (Supplementary table IV).

Discussion

Here we describe a high genetic coverage meta-analysis of GWAS based on TOPMed imputation involving ∼8 million common variants in ∼2000 MVP patients and ∼6800 controls. In addition to replicate the association at the TNS1 locus, we identified several associated variants and genes, involving established and original mechanisms for the biology underlying the genetic risk for MVP.

Our results using the TOPMed imputation panel provided higher resolution association map for the risk of MVP in a reasonably well powered dataset. However, from our previous GWAS, only TNS1, LMCD1 and SMG6, showed association at the genome-wide level of significance when adding the UKBB dataset. The differences in MVP definition (electronic health records in UKBB versus predominantly surgery reports) may have influenced the estimates of effects and their significance at the non-replicating loci. Larger datasets with more homogeneous definition of phenotype are needed to definitely conclude about their role in genetic susceptibility to MVP.

We provide confirmatory results and fine mapping at the TNS1 locus where we now report rs7595393 as the new lead associated SNP. The association signal on Chr2 is located in a gene-desert genomic region. We have previously provided solid biological evidence, which included a myxomatous valve phenotype in the heterozygote knockout mouse supporting the gene encoding Tensin 1, a focal adhesion protein, to be causal.¹ Functional annotations at this locus describe several SNPs belonging to the association block of rs7595393 as plausible causal variants. In the absence of eQTL, functional annotation for open chromatin and enhancer marks specifically in the mitral valve, we acknowledge that in silico annotation has limited ability to point at the causal variants. In parallel to this work, we have recently explored the functional properties of potential causal SNPs in this locus.¹¹ Using open chromatin maps that we generated in mitral valve tissue we found that one SNP (rs6723013) from the same association block than rs7595393 is located in open chromatin region specific to mitral valve tissues. Interestingly, the deletion using CRISPR-Cas9 of the sequence surrounding rs6723013 caused a significant change of the expression of TNS1 specifically, confirming this sequence to harbor a long-range regulator and the most likely causal variant in this locus.¹¹

One important addition in the current study is the improved coverage of low frequency variants (0.01<MAF<0.10). Our current data do not allow us to firmly conclude about the putative role of low frequency variants in MVP genetic risk, which needs to be explored further using larger datasets, given the limited power of our study for this category of variants (e.g at MAF=0.05, for a significance level of 5×10⁻⁸, and a genetic relative risk of 1.50, the power of our current sample is 0.39). Nonetheless, one unprecedented genome-wide significant (SYT2, Chr1) association signal that we report here is driven by low frequency (0.02 ≤ MAF ≤ 0.05) variants, although none the genes in the vicinity of this association signal present biological link to valve development or biology. Follow-up validation in larger studies is needed to confirm this association signal.

We also report two suggestive association signals in MSRA on Chr8 and FBXO46 on Chr19. Interestingly, we also report MSRA suggestive association at the gene-level with MVP. MSRA encodes methionine sulfoxide reductase A, a ubiquitously expressed and conserved enzyme, including in the heart, with the highest levels in the kidney and the nervous system. Several GWAS association signals near MSRA were reported for blood pressure,¹² neuroticism,¹³ and glomerular filtration.¹⁴ The biological implication of MSRA in the degenerative process of the valve is unclear, and could be through this enzyme protective role against oxidative stress during aging.¹⁵

As for FBXO46 locus on Chr 19, we found evidence for association of several genes with MVP including suggestive association with FBXO46, and significant association for SIX5, DMPK and DMWD. This locus overlaps a complex genomic region where CTG repeats in the 3’UTR of DMPK are involved in myotonic dystrophy, an autosomal dominant disorder characterized by myotonia, muscular dystrophy, cataracts hypogonadism and cardiac disorders including arrhythmia and mitral valve prolapse.^{16, 17} We found that SNPs in this locus are eQTLs of DMPK and DMWD cardiovascular tissues in GTEx, although genotype correlation with expression in mitral valve is not known.

We report several genes involved in cardiac development among the top associated genes and enriched Go Terms. This applies to GLIS1 that we have previously implicated in MVP risk,⁴ the TGF-beta family member TGFB2, an essential growth factor for myocardial cells endothelial to mesenchymal transition and valve elongation during valve development,¹⁸ the inhibitor of DNA binding 2 gene (ID2) regulated by TGF-beta1 and BMP-7¹⁹ and whose expression is lost in valve forming regions of Smad4-deficient endocardium mice,²⁰ and TBX5, a key regulator of cardiac development.²¹ TBX5 expression is not detected in human heart valves,²² which may suggest the valve prolapse phenotype to result from defaults in the interplay between myocardium, conduction tissue and mitral valve apparatus during heart development or later as a consequence of valve aging. We also report evidence of valve expression in mouse developing hearts for less obvious associated genes including SMG6, SRR and ABCC3 whose function in mitral valve myxomatous process may deserve future investigation.

Our study presents several limitations. The inclusion of a new dataset from the UK Biobank and the generation of a dense association map did not compensate the limited power of our study, especially for rare and low frequency variants, and the original suggestive loci described will need to be replicated in future studies. The gene-based results are not able to detect association signals involving gene-desert genomic regions with long-range enhancers, as the one we observe on Chr2 upstream TNS1. Another limitation of this method is that gene-rich genomic regions provides redundant association signals involving the same sets of variants and do not allow to point specifically at a potential causal gene. Functional annotation is based on existing eQTLs and gene expression pattern in cardiovascular tissues, especially heart atrial appendage and left ventricle, where cell composition and gene expression may differ from gene expression in the mitral valve.

Conclusions

To summarize, we report an updated meta-analysis GWAS for MVP using dense imputation coverage and an increased case control sample. We describe several established and original associated loci and genes with MVP spanning biological mechanisms highly relevant to heart valve disease. Follow-up biological studies in cell and animal models are needed to better understand their direct effect on the valve degenerative process.