Systematic large-scale assessment of the genetic architecture of left ventricular noncompaction reveals diverse etiologies

Purpose To characterize the genetic architecture of left ventricular noncompaction (LVNC) and investigate the extent to which it may represent a distinct pathology or a secondary phenotype associated with other cardiac diseases. Methods We performed rare variant association analysis with 840 LVNC cases and 125,748 gnomAD population controls, and compared results to similar analyses on dilated cardiomyopathy (DCM) and hypertrophic cardiomyopathy (HCM). Results We observed substantial genetic overlap indicating that LVNC often represents a phenotypic variation of DCM or HCM. In contrast, truncating variants in MYH7, ACTN2, and PRDM16 were uniquely associated with LVNC and may reflect a distinct LVNC etiology. In particular, MYH7 truncating variants (MYH7tv), generally considered nonpathogenic for cardiomyopathies, were 20-fold enriched in LVNC cases over controls. MYH7tv heterozygotes identified in the UK Biobank and healthy volunteer cohorts also displayed significantly greater noncompaction compared with matched controls. RYR2 exon deletions and HCN4 transmembrane variants were also enriched in LVNC, supporting prior reports of association with arrhythmogenic LVNC phenotypes. Conclusion LVNC is characterized by substantial genetic overlap with DCM/HCM but is also associated with distinct noncompaction and arrhythmia etiologies. These results will enable enhanced application of LVNC genetic testing and help to distinguish pathological from physiological noncompaction.

defined as lacking mutations in sarcomeric genes, the denominator used for PRDM16 in this cohort was 93 cases (to include the 18 genotype positive cases from the sarcomeric gene studies from this cohort and avoid inflating the frequency of PRDM16 variants).
See Table S1 for the number of patients sequenced per gene and per cohort.

Defining rare variants
Rare variants were defined as having a filtering allele frequency (FAF) in gnomAD exomes (v 2.1) less than 0.0001. FAFs are defined as the highest disease-specific maximum credible population allele frequency for which the observed allele count is not compatible with pathogenicity. The threshold of 0.0001 was applied in our previous study on rare variant association analysis in cardiomyopathies 7 and has been demonstrated to be an appropriate level based on the specific characteristics (prevalence, genetic heterogeneity and penetrance) in cardiomyopathies 8 . To be defined as rare, variants had to be below this threshold for both the popmax FAF (highest allele frequency of the major subpopulations of gnomAD -African, East Asian, Latino, non-Finnish European and South Asian) and the overall FAF (to filter variants that may be enriched in the gnomAD founder populations -Ashkenazi Jewish and Finnish).

Calculation of denominator for gnomAD
To account for variable coverage in exome-sequenced gnomAD samples, the number of individuals deemed sequenced per gene in gnomAD was calculated using the mean number of alleles across all rare protein-altering variants detected in gnomAD for the gene in question, and then converted from allele number to individuals (divided by 2 for autosomal genes and 1.46 for X chromosome genes based on the male-female ratio in gnomAD).

Specific additional analysis for RBM20, HCN4 and RYR2
In addition to assessing overall gene level rare variant frequencies for truncating and non-truncating variants, further analysis of specific variant classes was performed for three genes (RBM20, HCN4 and RYR2) based on previously published reports on the pathogenicity of such variants (described in detail in Table S11). For RBM20, non-truncating variants in the established pathogenic hotspot between residues 634 and 638 inclusive were assessed 9,10 . For HCN4, non-truncating variants in the transmembrane region of the channel were assessed (amino acid residues 267 to 517 as defined by Uniprot entry Q9Y3Q4).
For RYR2, we additionally assessed the frequency of structural variants (SVs) in LVNC cases and controls. As controls, we used the recently added SV dataset in gnomAD, derived from whole genome sequencing of 14,891 individuals 11 . Of note, the detection of SVs requires particular assays and/or analysis pipelines, and it is uncertain whether all of the constituent studies in this analysis were designed to detect such variants. However, we have decided to use the total number of cases where RYR2 is sequenced (429) to provide a conservative estimate of the contribution of these variants in LVNC.

Cardiomyopathy cohorts
For comparison to the burden of rare variation in dilated cardiomyopathy (DCM) and hypertrophic cardiomyopathy (HCM), variants from cardiomyopathy cases from previously published studies or unpublished data from cardiomyopathy cohorts at the Royal Brompton Hospital were used and reanalysed using the same methods as with the LVNC data. Specifically, previously published data from clinical genetics cohorts from the Oxford Molecular Genetics Laboratory (OMGL), UK, and the Laboratory of Molecular Medicine (LMM), USA was used for ACTC1, ACTN2, MYBPC3, MYH7, TNNT2, TPM1 and TTN (for DCM) and ACTC1, ACTN2, MYBPC3, MYH7, TNNT2 and TPM1 (for HCM) 7 . For the remaining genes -HCN4, PRDM16 and RBM20 (for DCM and HCM) and TTN (for HCM), previously unpublished data from cardiomyopathy outpatient clinic cohorts from the Royal Brompton Hospital, UK, was used, comprising up to 863 unrelated DCM patients and up to 685 unrelated HCM patients. See Table S4 for details of the number of DCM and HCM cases analysed for each gene.

Association analysis between genotype and age of onset in LVNC
To assess if there is any significant association between the enriched LVNC variant classes and the age at diagnosis or presentation, we analysed a subset of the case cohort for which this information was available (LMM and Dutch cohorts). Patient ages were dichotomised into adult (≥18 years) and child (<18 years), and the potential over-representation of specific variant classes in either age group was assessed by means of two-sided binomial exact tests, performed with the R function binom.test(). Note that for the variants detected in the Dutch cohort, patient age information was only available for those classified as (likely) pathogenic by van Waning et al. All VUS from that study were classified as having an unknown age at presentation and excluded from this analysis.

Analysis of the pedigree with the c.732+1G>A variant (index case from the Italian cohort)
The 31-year-old male proband presented at echocardiographic examination with apical and lateral non-compaction of the left ventricle, and marked hypertrabeculation. Of note, he had undergone cardiac ablation at 15 years of age to treat ventricular pre-excitation, and was the only carrier of the p.Pro83Ser variant in PRKAG2, which he likely inherited from the father, and which was previously proposed as a genetic modifier predisposing to ventricular pre-excitation 12 . Sequencing was performed on the proband and his family members using the Illumina TruSight Cardio panel.

MYH7 truncating variants in population cohorts
Of the UK Biobank individuals with exome sequencing data at the time of analysis (n=49,960 participants), those with cardiac magnetic resonance (CMR) imaging (n=12,447) were included in this study. Individuals with truncating variants (frameshift, nonsense, splice acceptor, splice donor) in MYH7 were analysed, alongside controls without MYH7 truncating variants that were matched for year-of-birth, sex, and ethnicity. All individuals were flagged as genetically Caucasian (UK Biobank phenotype ID 22006) and were not related (based on UK Biobank phenotype ID 22021 and KING kinship coefficient estimates from the genotype relatedness file with a cut-off of 0.0884 to include pairs of individuals with greater than 3rd-degree relatedness 13 through ukbtools 14 ). MYH7 truncating variant carriers and year-of-birth-, sex-, and ethnicity-matched non-carrier controls were also identified from healthy control cohorts from the UK Digital Heart Project (n=912) 15 and the Egyptian Collaborative Cardiac Genomics (ECCO-GEN) Project (n=400) 16 , all of which had CMR imaging and TruSight Cardio gene panel sequencing 17 . The CMR data of all paired participants was randomised and analysed blindly by two cardiologists who calculated the max NC/C ratio for each individual, and the mean of the ratio from the two separate analyses was assessed. This research has been conducted using the UK Biobank Resource under Application Number 47602.

Power calculations
Prior to the main analysis, we performed power calculations to assess the statistical power provided by the comparison of 840 patients with 120915 population individuals (average number of sequenced gnomAD individuals over the genes we tested). At alpha=0.05, the comparison of 840 cases with 120915 controls provides 80% power to detect an enrichment of 1% against a background variation rate of 0.91% for a single gene tested.
After the analysis, we have re-ran specific power calculations to assess the statistical power we had in those comparisons where -in spite of the enrichment being significant -burdens were representative of extreme cases: • lowest burden in LVNC cases (non-truncating variants in the RBM20 hotspot, 3 Figure S1:

List of Supplemental Tables
Supplemental Table 1: List of genes and Ensembl transcripts analysed in this study (i.e. those analysed in ≥50% of the constituent cohorts). The number of LVNC probands sequenced for each gene in each cohort is shown, along with the total number of cases analysed for each gene in this meta-analysis.
Supplemental Table 2: List of all rare, protein-altering variants detected in the LVNC cases analysed in this meta-analysis for the genes in Table S1. Rarity is defined as a gnomAD exomes (v2.1) filtering allele frequency (overall and popmax) < 0.0001. The variant classes included are frameshift, nonsense and essential splice site (truncating), missense and inframe insertions/deletions (non-truncating) and exon deletions (structural variants). The number of cases in which the variant is detected (per cohort) is also shown. For variants of those classes that are enriched in LVNC, clinical classification information (based on ClinVar v201909) is provided.
Supplemental Table 3  Supplemental Table 11: Summary of published studies supporting a role in the non-compaction phenotype for PRDM16 and ACTN2 and in arrhythmogenic and non-compaction phenotypes for RYR2 and HCN4.