Population-Based Variation in Cardiomyopathy Genes

Abstract

Background

Hypertrophic cardiomyopathy and dilated cardiomyopathy arise from mutations in genes encoding sarcomere proteins including MYH7, MYBPC3, and TTN. Genetic diagnosis of cardiomyopathy relies on complete sequencing of the gene coding regions, and most pathogenic variation is rare. The 1000 Genomes project is an ongoing consortium designed to deliver whole genome sequence information from an ethnically diverse population and therefore is a rich source to determine both common and rare genetic variants.

Methods and Results

We queried the 1000 Genomes database of 1,092 individuals for exonic variants within three sarcomere genes MHY7, MYBPC3, and TTN. We focused our analysis on protein-altering variation, including nonsynonymous single nucleotide polymorphisms, insertion/deletion polymorphisms, or splice site altering variants. We identified known and predicted pathogenic variation in MYBPC3 and MYH7 at a higher frequency than what would be expected based on the known prevalence of cardiomyopathy. We also found substantial variation, including protein-disrupting sequences, in TTN.

Conclusions

Cardiomyopathy is a genetically heterogeneous disorder caused by mutations in multiple genes. The frequency of predicted pathogenic protein altering variation in cardiomyopathy genes suggests that many of these variants may be insufficient to cause disease on their own but may modify phenotype in a genetically susceptible host. This is suggested by the high prevalence of TTN insertion/deletions in the 1000 Genomes cohort. Given the possibility of additional genetic variants that modify the phenotype of a primary driver mutation, broad-based genetic testing should be employed.

Introduction

Cardiomyopathy is a major cause of heart failure, and hypertrophic cardiomyopathy (HCM) and dilated cardiomyopathy (DCM) each have a significant heritable component.¹ Although debated, the estimated prevalence of HCM in the general population is 1:500.^2–5 DCM, classified as left ventricular dilatation and systolic dysfunction in the absence of other causes of cardiomyopathy, has an estimated prevalence of 1:2500.^5–7 Mutations in MHY7, the gene encoding β-myosin heavy chain (MHC), and MYBPC3, the gene encoding myosin binding protein C, cause both HCM and DCM.^{1, 3, 5} Mutations in the TTN gene, encoding the giant protein titin, have been associated with multiple forms of cardiomyopathy, including DCM and arrhythmogenic right ventricular cardiomyopathy.^8–10 The number and size of cardiomyopathy genes have limited the sensitivity of genetic testing since most testing has not included TTN. Improvements in sequencing technology now facilitate access to cardiomyopathy genes, including TTN with its 363 exons.¹¹

The genetic diagnosis of cardiomyopathy relies on completely sequencing the coding regions of cardiomyopathy genes since most pathogenic variation is rare, or “private.” Common variation is atypical in inherited cardiomyopathy although common variants may modify disease phenotype. Genomic variants are defined as rare if present at a minor allele frequency (MAF) of less than 0.5% or as low frequency if present at a MAF of 0.5%–5%.¹² Since allele frequencies differ among ethnic and racial groups, interpretation of individual genetic variation is limited by the ethnic make up of available genome databases. The 1000 Genomes project is an ongoing consortium designed to deliver whole genome sequence from an ethnically varied population with the goal of making publically available the genomes of 2500 individuals. In 2010, the 1000 Genomes pilot project was published, providing data from low-coverage (~2–6 fold) whole-genome sequencing (WGS) of 179 individuals, high-coverage (average 42 fold) WGS of six individuals in two trios, and exon-targeted sequencing (more than 50 fold coverage) of 8,140 exons in 697 individuals.¹² Since that time, sequentially released data from each phase of the project has been made available. The most recent release in February, 2012 contained variant calls and phased genotypes across 1092 individuals, with an average coverage of over 4 fold per individual in those sequences released as part of the full project.¹²

Phenotypic data is unavailable from the subjects whose genomes were determined in the 1000 Genomes project. Thus, interrogation of genetic variation in this cohort should be considered to include normal individuals and those with disease, with an expected prevalence of cardiomyopathy mirroring that of the population at large. Analysis of the 1000 Genomes dataset now allows for estimates of variant prevalence. Unlike the HapMap project, which provided information on common variation, the 1000 Genomes project provides information on rare or private variation.¹³ Herein, we queried the 1000 Genomes database to interrogate genetic variation in three sarcomeric genes, MHY7, MYBPC3 and TTN, to demonstrate how this database can be used to estimate genetic variation in the population. We focused our analysis on protein-altering variation (PAV), or those DNA deviations that predict a distinct protein from the referent genome sequence. PAVs are mainly single nucleotide polymorphisms (SNPs) that are nonsynonymous, or missense, in nature. PAV also includes insertion/deletions (indel) in coding regions, splice site altering variants, and nonsense polymorphisms. We found substantial variation, including protein-disrupting sequences, in TTN. We also identified previously reported and predicted pathogenic variation in MYBPC3 and MYH7 in the 1000 Genomes database at a frequency substantially higher than what would be predicted based on population-based studies of DCM and HCM prevalence. This data can be used to inform estimates of baseline genetic variation and, importantly, suggests that “at-risk” genotypes are more common in the population at large than previously expected.

Methods

Variant discovery for sarcomeric and non-sarcomeric genes

Exonic boundaries for MYBPC3 (ENST00000545968), MYH7 (ENST00000355349), CAMK2D (ENST00000342666), KCNQ1 (ENST00000155840), KCNH2 (ENST00000262186) SGCD (ENST00000337851), SGCG (ENST00000218867), and LMNA (ENST00000368300) were downloaded from the Ensemble Genome Browser (www.ensemble.org). Exonic variants were extracted from the 1000 Genomes February 2012 updated genotypes for the integrated phase 1 release (#ICHG2011) using the online data slicer available through the 1000 Genomes browser (http://browser.1000genomes.org). Variants were filtered based on their Phred quality score discarding those below 29. Variants were subsequently submitted to the website's Variant Effect Predictor (VEP), allowing for the analysis of potentially splice site altering variation within exons. The default settings were adjusted so as to return NCBI terms for variant consequences and to yield SIFT (Sorts Intolerant From Tolerant), Polyphen 2(PP2), and Condel predictions and scores.^14–16 Variant predictions corresponding to the aforementioned transcripts were extracted. Variants were compared to those present in the NHLBI Exome Sequencing Project (ESP) (http://evs.gs.washington.edu/EVS/) using the Compare two Datasets tool available through the University of Pennsylvania Galaxy server (http://main.g2.bx.psu.edu/).

Variant discovery for TTN

The three primary isoforms of titin, N2-BA, N2-A, and N2-B, result from differential splicing of the I-band encoding region of titin. The N2-BA isoform encodes the full-length protein and contains blocks of sequence which are specific to either N2-B or N2-A titin.¹⁷ Both N2-BA and N2-B titin are expressed in the heart. The DNA sequence encoding the N2-BA isoform does not exist in the Ensemble or UCSC databases but instead must be deduced from the N2-B and N2-A isoforms.

The FASTA sequence for N2-A titin (NP_596869.4) was downloaded from the NCBI-Gene database and aligned with the Uniprot titin sequence Q8WZ42 using Uniprot's online sequence alignment tool (http://www.uniprot.org). The 927 amino acids absent from the N2-A isoform were extracted and run through BLAT, a sequence alignment tool publically available through the UCSC Genome Browser (http://genome.ucsc.edu/cgi-bin/hgBlat?command=start). The amino acids were mapped to positions 179,603,868–179,606,648 on chromosome 2 and were found to correspond to exon 45 in the N2-B isoform of titin (ENST00000460472) through cross-referencing of its exon boundaries on the Ensemble Genome Browser.

Exonic boundaries encoding the N2-A TTN isoform (ENST00000342992) and exon 45 of N2-B were downloaded from Ensemble. Exonic variants were extracted from the 1000 Genomes February 2012 updated genotypes for the integrated phase 1 release using the online data slicer. Variants were filtered based on their Phred-scaled quality score, retaining only those variants with a score 29 or greater.^{18, 19} Exonic variants were then submitted to the VEP using the previously described settings. Variant consequences corresponding to ENST00000342992 and exon 45 of ENST00000460472 were extracted to yield all variation within N2-BA TTN as defined by the Uniprot consensus sequence Q8WZ42. Amino acid positions from the N2-A transcript were shifted from position 4380 onward to give their amino acid positions within N2-BA titin. TTN variants were compared to those present in the NHLBI ESP as described above. Indels from the 1000 Genomes project were identified in the low-coverage, whole-genome project. Indels are not available from the NHLBI ESP.

Variant mapping to protein domains

Variants in MYBPC3 and TTN were mapped to their respective locations within the protein using Uniprot sequence identifiers Q14896 and Q8WZ42. Variants in MYH7 were mapped to their location within the protein using the boundaries outlined in Blair et al.^{20, 21} Pathogenic variants were identified through comparison with those listed on the online Human Gene Mutation Database (HGMD) Professional release 2011.3 (http://www.hgmd.org/).

Ethnic distribution of TTN variants

Subjects were matched to their corresponding ethnic groups (African, American, Asian, European) using the supplement provided with the integrated phase 1 release (ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/release/20110521/phase1_integrated_calls.20101123.A LL.panel). Variants were eliminated from this analysis in cases where the minor allele frequency exceeded that of the reference allele. The number of total PAVs per individual was assessed within each ethnic group for MYH7, MYBPC3, and TTN.

Statistical analysis

A Kruskal-Wallis one-way analysis of variance was performed in PRISM to test for differences in variant load between ethnic groups. Post-test analysis was performed using Dunn's multiple comparison test.

Results

Extracting protein-coding variants from cardiomyopathy-associated genes from the 1000 Genomes database

The variants used for this analysis were extracted from the 1000 Genomes February, 2012 updated genotypes for the integrated phase 1 release. This call set derives from low-coverage whole-genome and exome sequencing data and contains phased genotype calls across 1092 individuals for single nucleotide polymorphisms (SNPs), short insertion/deletions (indels), and larger deletions. We queried the dataset for SNPs in three genes associated with cardiomyopathy: MYH7 encoding the major myosin heavy chain in the heart; MYBPC3 encoding myosin binding protein C3, and TTN encoding the giant protein titin.²¹TTN and MYBPC3 SNPs were identified in data derived from low-coverage genome sequencing as they were not included in the targeted exome analysis performed as part of the pilot project. MYH7 SNPs were identified in both the low-coverage, whole-genome sequence data and the higher coverage targeted exome sequence. Of the SNPs in MYH7, 48.8% were found in both whole-genome and exome sequencing, 45.2% in exome sequencing alone, and 6.0% in WGS alone.

Variant discovery in three sarcomere genes

Of the 1153 total variants in MYH7, MYBPC3, and TTN, 1136 were SNPs and 17 were indels, the later of which were all found in TTN. Indels ranged in size from 1–3bps with the addition of one larger 18bp deletion. Of the 1133 coding SNPs, 386 were synonymous, 15 of which were exonic SNPs predicted to alter a splice site, and 726 resulted in missense changes. Three nonsense variants were detected, all of which were in TTN (Supplemental Table 1). Of these protein-altering variants (PAVs), 79.0% were defined as rare (MAF <0.5%), and 14.6% were low-frequency, defined as a MAF of 0.5%–5.0%. Thus, 93.6% of variants with the potential to alter protein structure or function are found in a small fraction of the population, underscoring the contribution of infrequent variation to genotype diversity.

Distribution of MYH7 and MYBPC3 variants

Figure 1 depicts the protein structure encoded by MYH7 and displays the position of PAV detected in 1092 individuals from the 1000 Genomes database. Twenty-one different PAVs were found in the 1000 Genomes cohort (Table 1). Eleven fall within the S1 and S2 domains of the β-myosin heavy chain protein, while the remaining 10 are within the rod domain. Of the 21 PAVs in MYH7, 8 were reported by both Polyphen2 (PP2) and SIFT (Sorts Intolerant From Tolerant) to be damaging (Supplemental Table 2).^{14, 16} All 21 variants were present in the 1000 Genomes database as rare or low-frequency variants.

Figure 1.

Missense variants identified in the MYH7 gene in 1000 Genomes. The position of 21 rare missense variants from the 1000 Genomes February 2012 updated genotypes for the integrated phase 1 release. Variants also reported as pathogenic in the Human Genome Mutation Database (HGMD) are underlined. ATP and actin refer to the ATP and Actin binding sites respectively.

Table 1.

MYH7 protein-altering variants found in 1000 Genomes.

AA Change	Domain	SIFT Prediction (score)	PP2 Prediction (score)	MAF (No. of occurrences from n=2184 alleles)
A26V *	S1 domain	T (0.25)	Benign (0.017)	0.0032 (7)
V133M	SI domain	D (0.04)	Probably damaging (0.998)	0.0005 (1)
N306D	S1 domain	T (0.52)	Possibly damaging (0.235)	0.0005 (1)
R442C *	S1 domain	D (0)	Probably damaging (1)	0.0005 (1)
L620Q	S1 domain	D (0.03)	Probably damaging (0.992)	0.0005 (1)
R858H *	S2 domain	D (0.02)	Benign (0.002)	0.0005 (1)
R869H *	S2 domain	T (0.07)	PosS1bly damaging (0.496)	0.0005 (1)
M982T	S2 domain	D (0)	Benign (0.06)	0.0005 (1)
R1045C *	S2 domain	D (0)	Benign (0.012)	0.0005 (1)
M1046I	S2 domain	T (0.2)	Benign (0.002)	0.0005 (1)
R1079W	S2 domain	D (0)	Possibly damaging (0.682)	0.0005 (1)
Q1267H	S2 domain	T (0.06)	Benign (0.001)	0.0009 (2)
R1289W	LMM	D (0)	Probably damaging (0.997)	0.0005 (1)
Q1458E	LMM	T (0.24)	Benign (0.037)	0.0005 (1)
N1486D	LMM	T (0.07)	Possibly damaging (0.539)	0 (0)
S1491C *	LMM	T (0.1)	Benign (0.001)	0.01 (17)
R1500L	LMM	D (0)	Probably damaging (0.982)	0 (0)
I1558F	LMM	D (0.02)	Probably damaging (0.99)	0.0005 (1)
I1558S	LMM	T (0.05)	Probably damaging (0.99)	0.0009 (2)
R1832G	LMM	D (0)	Possibly damaging (0.831)	0.005 (1)
E1902Q	LMM	T (0.16)	Benign (0.166)	0.0009 (2)

^*Reported in HGMD. T, tolerated in SIFT; D, deleterious in SIFT.

Figure 2 shows those PAVs detected in MYBPC3 in the 1000 Genomes database. Twenty-two different PAVs were found (Table 2). Eight PAVs were reported by both PP2 and SIFT to be damaging (Table 2). All PAVs were present at a rare or low-frequency.

Figure 2.

Missense variants identified in the MYBPC3 gene in 1000 Genomes. Cardiac myosin binding-protein C with depiction of 22 missense variants, one of which alters a splice site from the 1000 Genomes February 2012 updated genotypes for the integrated phase 1 release. Variants present in the HGMD are underlined.

Table 2.

MYBPC3 protein-altering variants in 1000 Genomes.

AA Change	SIFT Prediction (score)	PP2 Prediction (score)	MAF (No. of occurrences from n=2184 alleles)
V158M	D (0.01)	Possibly damaging (0.269)	0.05 (101)
R160W *	D (0)	Probably damaging (1)	0.0014 (3)
S217G *	D (0.01)	Benign (0.154)	0.0005 (1)
S236G *	T (1)	Benign (0)	0.09 (189)
A259V	T (0.3)	Benign (0.002)	0.0023 (5)
G278E *	T (0.14)	Possibly damaging (0.637)	0.0046 (10)
R326Q *	D (0.01)	Benign (0.02)	0.0018 (4)
R382W	D (0.02)	Probably damaging (0.936)	0.01 (31)
G440D	T (1)	Benign (0.021)	0.0005 (1)
G507R *	D (0)	Probably damaging (1)	0.0027 (6)
N515S	T (0.34)	Probably damaging (0.994)	0.0018 (4)
A522T *	T (0.2)	Benign (0.143)	0.0014 (3)
A833V *	T (0.52)	Probably damaging (0.888)	0.01 (14)
I867I g->a synSS	n/a	n/a	0.02 (34)
E872K	T (0.05)	Probably damaging (0.994)	0.0005 (1)
V896M *	T (0.05)	Benign (0.005)	0.0032 (7)
P952AfsX99	n/a	n/a	0.02 (45)
Q998E *	T (0.11)	Possibly damaging (0.799)	0.01 (17)
R1002W	D (0)	Probably damaging (1)	0.0027 (6)
R1036C	D (0)	Probably damaging (0.931)	0.0009 (2)
T1075S	T (0.05)	Benign (0.01)	0.0005 (1)
Q1103H	D (0)	Probably damaging(0.997)	0.0005 (1)
R1138H *	D (0)	Probably damaging(1)	0.0009 (2)
R1190H	T (0.05)	Probably damaging(1)	0.0005 (1)

^*Reported in HGMD. D, Deleterious in SIFT; T, tolerated in SIFT. synSS, synonymous splice site. The exonencoded predicted splice variant and the frameshift cannot be computed by SIFT or PP2.

Spatial distribution of TTN variants

Within TTN, 711 PAVs were detected in the 1000 Genomes database from 1092 individuals. Because of the large number of nsSNPs, we focused our analysis on indels. All indels had a Phred-scaled quality score of 29 or greater, indicating a base-call accuracy of almost 99.9%. Figure 3 shows the 17 indels detected, including one large 18bp deletion, I11443-E11449del, found in 62 individuals with a MAF of 0.03 (Table 3). This 18 bp deletion falls within the PEVK region that regulates extensibility of titin.²² Two individuals had multiple TTN indels, the implications of which depend on both phase and effect on protein frame. One had a frameshift combined with an in-frame deletion and the second had a frameshift and two in-frame deletions. One individual was homozygous for the 18bp deletion, and no individuals were homozygous for a frameshifting variant in TTN. Of the 17 indels in TTN, seven were found in the Ig-like domains and none within the fibronectin type III domains. Of these 17 indels, 6 are in-frame deletions and 11 are frameshifts that disrupt titin's carboxyl-terminus. Considering only the 11 indels that disrupt TTN, 3.2% of the population (35/1092) has a frameshift in TTN.

Figure 3.

Frameshifting mutations identified in TTN in 1000 Genomes. The titin isoform N2-BA is shown. Titin is linearly depicted with its 152 Ig-like domains in red and 132 fibronectin type III domains in blue. Depicted are the 17 indels identified the 1000 Genomes February 2012 updated genotypes for the integrated phase 1 release. Six of 17 indels do not disrupt reading frame. One (underlined) that deletes six amino acids in the PEVK region is found in 5% of the population. Variants are shown relative to the titin Uniprot Sequence identifier Q8WZ42. The dashed-lines below the protein schematic indicate the location of variants within the sarcomere.

Table 3.

TTN insertion/deletion polymorphisms

Location	Reference/Alternate Allele	AA Change	Domain	MAF (No. of occurrences in n=2184 alleles)
179666885–179666886	G/GT	Thr92fs	Ig-like 1	0.0037 (8)
179664626–179664628	ACTT/A	Glu198del		0.0009 (2)
179658211–179658212	C/CT	Ala486fs	Z-repeat 2	0.0027 (6)
179640461–179640462	C/CT	Glu2044fs		0.0009 (2)
179570049–179570051	CTCT/C	Glu9503del	Poly-Glu	0.0009 (2)
179554281	GT/G	Thr10313fs	PEVK	0.0005 (1)
179549673–179549674	A/AG	Ala10522fs	PEVK	0.0018 (4)
179548796	TC/T	Lys10535fs	PEVK	0.0005 (1)
179514942–179514959	TTTTCCTCTTCAGGAGCAA/T	Ile11443_Glu11449del	PEVK	0.03 (62)
179505311–179505312	A/AG	Thr11920fs	PEVK	0.0005 (1)
179501447–179501449	CAGG/C	Pro12089del	Ig-like 80	0.0018 (4)
179495075–179495076	TTC/T	Glu13084fs		0.0014 (3)
179428912–179428913	A/AG	Leu25675fs	Ig-like 125	0.0005 (1)
179439949	GC/G	Gly21996fs	Ig-like 115	0.0005 (1)
179414950–179414952	TTAA/T	Lys28898del	Ig-like 133	0.0005 (1)
179395288	GA/G	Ser33711fs	Ig-like 148	0.0032 (7)
179398424–1179398426	ACTT/A	Lys32665del	Ig-like 144	0.0005 (1)

Variation in sarcomere genes compared to nonsarcomere genes

The complexity of and requirements for sarcomere assembly are highly intricate. Because of the multiple protein-protein interactions required for proper assembly of the sarcomere, it has been suggested that sarcomere proteins, especially myosin, may be less tolerant of protein coding variation.²³ To assess this, we queried both common and rare variation in the three sarcomere genes, MYH7, MYBPC3 and TTN, and compared this to several genes encoding proteins important to cardiac function. We evaluated CAMK2D, encoding calcium/calmodulin-dependent protein kinase 2 subunit δ, KCNQ1 and KCNH2, encoding voltage gated potassium channels linked to hereditary long-QT syndrome, SGCD and SGCG, encoding the dystrophin associated proteins δ-sarcoglycan and γ-sarcoglycan, and LMNA, encoding the nuclear membrane protein Lamin A (Supplemental Table 1). The latter three genes have been linked to cardiomyopathy.²⁴ When considering PAVs/1 KB of coding sequencing, sarcomeric genes averaged 5.77 variants/KB compared to 6.35 variants/KB for non-sarcomeric genes (Supplemental Table 3). The range of PAV/1 KB of coding region varied considerably, from 2.29 to 14.84, suggesting that sarcomeric genes fall within this range. Among the three sarcomere genes, MYH7 had the greatest amount of total variation, but the least amount of PAV with only 26.6% of the total exonic SNPs altering protein in MYH7 compared to 73.3% and 68.2% PAV in MYBPC3 and TTN.

Known or Reported Disease-associated variants in 1000 Genomes

Twenty-two SNPs detected in the 1000 Genomes database were also found in the Human Genome Mutation Database (HGMD), a comprehensive online database of inherited disease mutations (www.hgmd.org) (Supplemental Table 4). Of the 22 SNPs in the 1000 Genomes project also identified in the HGMD, 11 were in MYBPC3, 7 in MYH7, and 4 in TTN. These SNPs represent a small portion of pathogenic variants reported in the HGMD, 3.6%, 3.2%, and 17.4% of published disease-causing mutations in MYBPC3, MYH7, and TTN respectively. Within the 1000 Genomes population, 18 of the 22 mutations also present in the HGMD appeared as rare variants while 3 others were present at a low frequency. One variant, S236G in MYBPC3, was present at a higher frequency (MAF=0.09; n=189), and therefore is not likely an HCM-associated mutation given its high prevalence.

To address the possibility of multiple variants within individuals, we evaluated the sarcomeric mutations found in both the HGMD and the 1000 Genomes project (Supplemental Table 4). MYBPC3 S236G was excluded from this analysis given its frequency (MAF 0.09, n=189). 99 individuals had at least one pathogenic variant, and of these, 93 individuals were heterozygous for a single variant. Three individuals from European and American cohorts were homozygous for MYBPC3 mutations (V896M, Q998E). One European individual was homozygous for the mutation S1491C in MYH7. Two individuals had two pathogenic variants each; one in MYH7 and one in MYBPC3 (R1138H MYBPC3 plus M982T MYH7, and S217G MYBPC3 plus S1491C MYH7). Thus, 5 of 1092 individuals in the 1000 Genomes project carry multiple reported pathogenic mutations in sarcomeric genes. This number also exceeds population estimates of cardiomyopathy prevalence, suggesting that these variants may be insufficient to cause disease but may modify phenotype in the context of a primary driver mutation.

We employed three protein prediction algorithms as part of our analysis: SIFT, PP2, and Condel, a normalized composite of the prior two programs.¹² We used Condel to analyze previously reported mutations now identified in the 1000 Genomes dataset and found that it predicted 15 of the 17 possible variants to be pathogenic. Using this same approach, Condel predicted that 83.7% (36/43) of the missense variants detected in MYH7 and MYBPC3 would be pathogenic. Comparison to the NHLBI Exome Sequencing Project (ESP) We queried the NHLBI ESP²⁵ for total SNPs in MYH7, MYBPC3, and TTN. This is a database of high coverage exonic sequence that includes 5379 samples drawn from African American and European American individuals. Average coverage in the NHLBI ESP across the three sarcomeric genes MYH7, MYBPC3, and TTN is higher (114.32, 55.42, and 108.73 fold, respectively) than the lower coverage 1000 Genomes project, but the ESP lacks indel calls at this time. 51.2%, 60.6%, and 63.5% of all SNPs identified in the 1000 Genomes project in MYH7, MYBPC3, and TTN respectively were also identified in the NHLBI ESP. Those SNPs in the 1000 Genomes Project, but not in the NHLBI ESP, were rare (present at an average frequency of 0.0020, 0.0037, and 0.0010 for MHY7, MYBPC3, and TTN, respectively.) Those SNPs found in the NHLBI ESP but not the 1000 Genomes project were similarly rare (present at an average frequency of 0.00018, 0.00020, and 0.0049 respectively). The similar frequency of rare alleles in the low coverage 1000 Genomes project and the high coverage NHLBI ESP argues that these findings are not false positives related to low coverage sequence. The different ethnic groups that constitute each sequenced cohort likely account for the difference in observed rare alleles.

Distribution of variants by race and ethnicity

A Kruskal-Wallis one-way analysis of variance was used to test for differences in PAV in MYH7, MYBPC3 and TTN among ethnic groups (African, American, Asian, European) in the 1000 Genomes cohort. Five variants were removed prior to this analysis as their MAF exceeded that of the reference allele. The frequency of variants differed significantly across the four groups in MYH7 [H(3)=148.1, p<0.0001], MYBPC3 [H(3)=10.05, p=0.0181], and TTN [H(3)=262.9, p<0.0001], (Figure 4). On post-test comparisons, Africans were shown to have significantly more variation in MYH7 than all other ethnic groups at 0.68 variants/individual while Asians significantly more variation in TTN at 28.76 variants/individual. The later differed from the results of the 1000 Genomes pilot project in which African individuals were reported to have the greatest variation across the genome as measured by the number of total and novel variants within each ethnic group.¹² Thus variant frequencies differ between ethnic groups at the level of individual genes and, notably, the extent and direction of these ethnic differences changes between genes.

Figure 4.

Protein altering variation (PAV) within the 1000Genomes cohort differs by cohort. The histograms above show the percentage of individuals within each cohort (AFR=African; AMR=American; ASN=ASN; EUR=European) having a specified number of PAVs per gene. This analysis included all PAV, except those 5 instances in which the MAF exceeded that of the reference allele. The amount of PAV in specified genes differed by racial groups and this differences was significant (MYH7 [H(3)=148.1, p<0.0001], MYBPC3 [H(3)=10.05, p=0.0181], and TTN [H(3)=262.9, p<0.0001], p values are from non-parametric ANOVA). On post-test comparisons, African individuals had more variation in MYH7 than all other groups while Asians significantly more variation in TTN.

Discussion

Pathogenic variation in 1000 Genomes is higher than expected; implications for testing

The 1000 Genomes project launched in 2008 with the goal of creating an ethnically diverse public reference database for DNA polymorphisms. Unprecedented in its scope, the project is the first to approximate the breadth of human genetic variation and provide a quantitative framework in which to predict potentially relevant variants. We queried the database of 1,092 individuals for three sarcomeric genes as they were thought to contain little sequence variation.^{21, 23, 26} This cohort, in the absence of phenotypic data, should be viewed as a representative sample of the population at large, containing both normal and abnormal individuals. Within the 1000 Genomes database, we identified variants that were previously reported as pathogenic in the HGMD. Although reported as mutations, some of these polymorphisms are not sufficient to cause cardiomyopathy as several have been reported in normal populations. We also found variants not reported in the HGMD that are predicted to be pathogenic using computational algorithms. Summing these with the observed TTN frameshifts yields a frequency that substantially exceeds population estimates of the combined HCM and DCM prevalence. It is likely that some of these variants are positioned to modify disease outcome and favors a comprehensive sequence analysis. Supporting this idea, compound and double mutations have been noted in younger more affected individuals.²⁷ Therefore, we favor comprehensive sequencing analysis especially in younger individuals.

Currently, assessing gene variants involves the use of prediction tools, segregation studies within families and animal/cellular models to predict pathogenicity. As cardiomyopathy genetic testing panels become more expansive, and the number of variants increases, costly and labor-intensive tools for analysis will prove challenging. Improved prediction tools will be necessary to better anticipate the effect of a given PAV. Better algorithms, along with comprehensive testing, should improve the information obtained from genetic analysis.

Indels in TTN

We also uncovered a high frequency of indels in TTN, all of which have a base call accuracy of almost 99.9%. In light of the high cumulative frequency of TTN indels in the general population (9%), TTN may be positioned to modify phenotype. Just over 5% of the general population has an 18bp in frame deletion in the proline rich region of titin. This region of titin has been closely linked to the elastic properties of titin and its ability to sense and respond to sarcomere length.²² This allele will be important to track in studies of cardiac modifiers given its prevalence and potential functional role. Notably, 3% of the population has a protein truncating TTN allele. This estimate is similar to the recent findings from Herman et al. where protein truncating TTN mutations were found in 25% of DCM subjects and in 3% of a normal control population.²⁸ Our analysis, unlike Herman et al., focused only on indels as protein truncating mutations in TTN, and did not include predicted TTN intronic splice site mutations. Thus, our analysis is expected to underestimate the population prevalence of protein truncating mutations in TTN.

Likely underestimates of pathogenic and predicted pathogenic variation

With increasing sequencing coverage, the ability to detect private variation increases.²⁹ High coverage sequence of MYH7 yielded additional variants that were not detected in the low coverage sequencing. The variation described in TTN and MYBPC3 derives only from low coverage WGS, and therefore our study probably underestimates variation. Overall, it is unlikely that these variants represent false positives SNPs. As of last year's reporting, the pilot project low coverage genome sequence is estimated to have less than a 5% false discovery rate for SNPs and a 1–3% genotype error rate.¹² Our analysis also did not query variation in intronic sequence, including potentially pathogenic splice-site altering sequence, thus providing an additional source for pathogenic variation which is not represented in our baseline estimates of genetic variation. At present, the current 1000 Genomes database contains data on 1,092 individuals and will extend sequence coverage to 2,500 individuals to provide further information on rare and low-frequency variation.

Interpreting rare variation in cardiomyopathy genes

Given the extent of rare PAV in both sarcomeric and non-sarcomeric genes, the mere presence of rare variation does not imply pathogenicity. Segregation of a rare variant within a family or demonstration of impaired protein function within a biological context is often used to interpret the findings of genetic testing. However, these in vitro studies are often impractical and not all family structures support this analysis, especially in families with diseases that limit lifespan. The continued development of analyses that better predict variant pathogenicity will be required. This analysis highlights the challenges that accompany analyzing population-based genetic information and applying findings to the individual patient with cardiomyopathy. The 1000 Genomes dataset will aid in interpreting genetic testing information, although its use is limited by lack of phenotypic correlates. In case-control studies of genotype and phenotype, control subjects should be carefully considered if the subjects have not been clinically evaluated or especially if they are too young to yet manifest disease.

The 1000 Genomes Project is an international consortium designed to provide full genomic sequence information from an ethnically varied population. The clinical history of 1000 Genomes participants is unknown, so information from the project is best used to determine population estimates of sequence variation. We queried the 1000 Genomes database for genetic variation in three cardiomyopathy genes, MYH7 (β myosin heavy chain), MYBPC3 (myosin binding protein C), and TTN (titin). We focused our analysis on rare variation that was either reported previously as pathogenic or was predicted to be pathogenic by computational algorithms. These three genes encode sarcomere proteins, and mutations in these genes cause familial dilated and hypertrophic cardiomyopathy. We identified predicted and previously reported pathogenic variation at a much higher frequency that the incidence of both dilated and hypertrophic cardiomyopathy. Specifically, we found 9% of genomes had insertion/deletion polymorphisms in TTN. The functionality of this genetic variation is not known given the absence of clinical information from 1000 Genomes participants. However, these variants may be important genetic modifiers that increase the risk for developing heart failure in the general population. For patients with a history of familial cardiomyopathy, these findings favor comprehensive genetic testing to identify both primary mutations and secondary disease modifiers.