Sequencing of Linkage Region on Chromosome 12p11 Identifies PKP2 as a Candidate Gene for Left Ventricular Mass in Dominican Families.

Increased left ventricular mass (LVM) is an intermediate phenotype for cardiovascular disease (CVD) and a predictor of stroke. Using families from the Dominican Republic, we have previously shown LVM to be heritable and found evidence for linkage to chromosome 12p11. Our current study aimed to further characterize the QTL by sequencing the 1 LOD unit down region in 10 families from the Dominican Republic with evidence for linkage to LVM. Within this region, we tested 5477 common variants [CVs; minor allele frequency (MAF) ≥5%] using the Quantitative Transmission-Disequilibrium Test (QTDT). Gene-based analyses were performed to test rare variants (RVs; MAF < 5%) in 181 genes using the family-based sequence kernel association test. A sample of 618 unrelated Dominicans from the Northern Manhattan Study (NOMAS) and 12 Dominican families with Exome Array data were used for replication analyses. The most strongly associated CV with evidence for replication was rs1046116 (Discovery families P = 9.0 × 10-4; NOMAS P = 0.03; replication families P = 0.46), a missense variant in PKP2 In nonsynonymous RV analyses, PKP2 was one of the most strongly associated genes (P = 0.05) with suggestive evidence for replication in NOMAS (P = 0.05). PKP2 encodes the plakophilin 2 protein and is a desmosomal gene implicated in arrythmogenic right ventricular cardiomyopathy and recently in arrhythmogenic left ventricular cardiomyopathy, which makes PKP2 an excellent candidate gene for LVM. In conclusion, sequencing of our previously reported QTL identified common and rare variants within PKP2 to be associated with LVM. Future studies are necessary to elucidate the role these variants play in influencing LVM.

Cardiovascular disease (CVD) is a significant public health burden, affecting more than one in three American adults ≥20 yr old and resulting in an estimated $316.6 billion in medical costs in 2012 (Mozaffarian 2016). Increased left ventricular mass (LVM), an intermediate phenotype for CVD, is predictive of stroke and CVD events (Levy ; Devereux ; Rodriguez ). Traditional factors, including elevated blood pressure, body mass index, and weight, are known to influence LVM (Savage ; Garner ). Heritability studies suggest that LVM may be genetically controlled, with reported heritability estimates ranging from 0.24 to 0.70 in various populations (Post ; Swan ; Juo ; Sharma ; Assimes ; de Simone ; Wang ). In addition, genome-wide association studies, primarily performed in samples of European ancestry, have identified significant associations between LVM and common variants within several loci (Vasan ; Fox ). A large meta-analysis of individuals of European ancestry found variants within 14q12 and 2p21 to be associated with decreased LVM and variants within 15q14 to be associated with increased LVM (Vasan ). In a meta-analysis of African American cohorts, variants within 8q11 were associated with increased LVM (Fox ). Though these loci were significantly associated with LVM in the discovery cohorts of each study, none achieved replication in their respective validation cohorts. Additional variants have been implicated in LVM through other genome-wide association studies, although none achieved statistical significance (Vasan ; Arnett ; Harper ).

In a previous genome-wide linkage study, we identified a novel region on chromosome 12p11 linked to LVM using extended families from the Dominican Republic (MLOD = 3.11), particularly in families with higher waist circumference (MLOD = 4.45) (Wang ). We further characterized this region by investigating common variants in a population-based cohort of Caribbean Hispanics and identified several candidates, including SOX5 (Della-Morte ). For this study we aimed to identify additional common and rare variants contributing to the 12p11 linkage signal by sequencing the 1 LOD down region on 12p11 in 10 extended Dominican families with strong evidence for linkage to this region (LOD ≥ 0.1). We validated the observed associations using both genome-wide and exome-wide genotype data from unrelated Dominicans in the Northern Manhattan Study (NOMAS) as well as exome-wide genotype data from 12 additional extended Dominican families with suggestive evidence for linkage to the 12p11 region.

Materials and Methods

Study samples

Individuals from both NOMAS, a population-based cohort, and the Family Study of Stroke Risk and Carotid Atherosclerosis, a family study consisting of select probands from NOMAS and their family members (Sacco ), were used for the current study. Details of NOMAS and the Family Study have been published previously (Sacco , 2007). Briefly, a total of 3298 stroke-free population-based participants were enrolled in NOMAS from 1993 to 2001. The family study enrolled a subset of Caribbean Hispanic probands from NOMAS with a high risk of cardiovascular disease (Sacco ) who could provide a family history, obtain family members’ permission for research staff to contact them, and have at least three first-degree relatives able to participate. All probands were identified in northern Manhattan and family members were enrolled in New York at Columbia University and in the Dominican Republic (DR) at the Clinicas Corazones Unidos in Santo Domingo. All participants provided written informed consent and the study was approved by the Institutional Review Boards of Columbia University, University of Miami, the National Bioethics Committee, and the Independent Ethics Committee of Instituto Oncologico Regional del Cibao in the DR (20070478 and 20072012).

Twenty-two families with a family-specific LOD score >0.1 at the chromosome 12p11 QTL for LVM were selected for our current study; 10 families were sequenced as part of our discovery analyses (n = 180 individuals) (Table 1) and 12 families were genotyped on the Exome Array for our replication analyses (n = 143 individuals) (Supplemental Material, Table S1). Details of the family selection have been published previously (Dueker ). Additional replication analyses were performed using a sample of 618 unrelated Dominicans from NOMAS (Sacco ) who were not enrolled in the family study.

Table 1

Characteristics of Dominican families included in analyses for chr12p11 analyses

Family ID	Individuals per Family	Family-Specific LOD Score	LV Mass Residual (g)				Age (yr) µ ± SD	BMI (kg/m2) µ ± SD	Waist Circumference (inch) µ ± SD	% Female	% Diabetes	% Smoker	% Hypertension
			Mean	SD	Min	Max
5275	24	0.94	−0.10	2.71	−4.77	6.16	44.0 ± 15.7	33.3	31.2 ± 6.3	38.0 ± 6.2	12.5	33.3	33.3
3719	12	0.92	0.43	2.64	−3.04	5.36	49.3 ± 16.2	33.3	32.3 ± 7.8	37.8 ± 5.4	8.3	33.3	41.7
5103	35	0.44	1.31	2.34	−3.30	6.89	42.2 ± 17.8	37.1	26.9 ± 5.3	36.3 ± 5.8	9.6	37.1	42.9
2235	15	0.37	−0.46	3.40	−5.62	5.54	49.3 ± 16.9	60.0	29.1 ± 3.4	39.0 ± 5.6	20.0	60.0	66.7
4641	13	0.30	−0.44	2.44	−4.37	4.04	42.9 ± 18.0	53.8	31.3 ± 8.5	37.3 ± 6.1	53.8	53.8	23.1
2783	15	0.21	−0.57	3.00	−3.97	6.66	40.7 ± 15.4	26.7	29.5 ± 6.0	36.9 ± 5.2	13.3	26.7	53.3
6081	27	0.20	−0.55	1.45	−3.13	2.47	46.3 ± 16.9	14.8	27.7 ± 5.4	36.0 ± 5.3	3.7	14.8	40.7
3561	17	0.16	−0.06	3.18	−4.99	6.70	47.1 ± 13.8	41.2	35.3 ± 7.5	42.2 ± 6.9	29.4	41.2	52.9
1917	10	0.15	0.20	2.84	−4.06	4.01	45.8 ± 16.8	80.0	33.8 ± 7.2	41.0 ± 6.0	20.0	80.0	60.0
803	12	0.14	−1.16	2.10	−4.25	3.11	54.7 ± 24.2	41.7	30.3 ± 5.2	38.8 ± 5.1	16.7	41.7	41.7

Echocardiographic evaluation and risk factor measurements

As detailed previously (Wang ; Della-Morte ), standard two-dimensional echocardiography, including color-Doppler flow study was performed according to the guidelines of the American Society of Echocardiography (Sahn ). High quality parasternal long axis views of the left ventricle were obtained, from which left ventricular end-diastolic diameter (LVDD), left ventricular end-systolic diameter (LVSD), interventricular septum (IVS), and posterior wall thickness (PWT) were derived (Di Tullio ). LVM was calculated according to the modified American Society of Echocardiography formula: LVM = 0.8 [1.04 (LVDD + IVS + PWT)3 − (LVDD)3] + 0.6 (Devereux ).

Vascular risk factors, including body mass index (BMI), systolic blood pressure, and smoking status, were collected during a standardized interview (Elkind ). Smoking status was defined as never vs. ever, systolic blood pressure was defined as the average of two separate systolic blood pressure measurements taken after rest, and presence of diabetes was defined as fasting blood glucose level ≥126 mg/dl or self-reported history of diabetes. All data collection procedures were standardized and identical across NOMAS and the Family Study.

Discovery sample sequencing and quality control

In the family sample, genomic DNA was isolated from whole blood. Targeted sequencing of the exons in 181 genes within the 1 LOD unit down region on 12p11 (chr12:24–51 Mb), as well as sequencing beyond the exons for candidate genes identified previously (ARID2, BICD1, BIN2, c12orf68, FAR2, RACGAP1, SLC38A1, SOX5) (Della-Morte ), was performed using a customized Agilent SureSelect Enrichment kit. A detailed description of sequencing methods has been previously published (Wang ). Briefly, DNA libraries were sequenced on an Illumina HiSeq2000 and the raw sequencing reads were aligned to the human reference sequence hg19 with the Burrows-Wheeler Aligner (Li and Durbin 2010). Variant calling was performed using the Genome Analysis ToolKit and potential functions of variants were annotated using ANNOVAR v. 2016Feb01 (Wang ) and SeattleSequation 138.

Quality control was conducted at both variant and sample levels, as described previously (Wang ; Dueker ). Within each individual sample, variants with a depth <4 or Phred-Like score <100 were set as missing. Variants with VQSLOD <−4 and variants with call rate <75% were removed from further analysis. Individuals with low concordance (<95%) between the sequencing data and available genotype data were removed (n = 3). Additionally, individuals missing LVM measures (n = 9) and/or covariate values (n = 1) were removed. For the remaining family study samples, pedigree structure was confirmed using the Graphical Relationship Representation software v. 1.2.1.41. Mendelian error checking was performed, and Mendelian errors were set to missing for all the variants called using PLATO v. 0.84 (Grady ).

Replication sample genotyping and quality control

Exome Array:

NOMAS participants and our 12 replication families were genotyped using the Illumina HumanExome-24v1_B Beadchip, at the Hussman Institute for Human Genomics in the Center for Genome Technology (Miami, FL). Our Exome Array included custom exonic variants selected on the basis of sequencing data obtained in the discovery family data set. Details of the variant selection have been described previously (Wang ). A total of 4128 single nucleotide variants (SNVs) within our 12p11 region were available for quality control analyses.

Of the 659 NOMAS participants and 150 replication family members genotyped on the Exome Array, 99.8% had a genotype call rate >98%. A subset of NOMAS participants genotyped on the Exome Array also had Affymetrix 6.0 whole-genome genotype data available and all participants had high concordance with this additional data set (≥96%). We removed seven NOMAS individuals and five replication family members due to unexpected duplication or relatedness, gender discrepancy, and low call rate (<98%). For our current study, individuals missing LVM measures and/or covariate values were removed (n = 38 NOMAS participants; n = 2 replication family members), leaving us with a final sample of 618 Dominican NOMAS participants and 143 replication family members. At the variant level, we removed SNVs with call rate <95% (n = 3 in NOMAS; n = 7 in the replication families) and monomorphic SNVs, leaving us with 1419 exonic rare single nucleotide variants (RVs) in our region for Exome Array analysis in NOMAS and 842 exonic RVs in the replication families. Mendelian error checking was performed in the replication families and Mendelian errors were set to missing for all the variants called using PLATO v. 0.84 (Grady ).

Affymetrix 6.0 genotyping chip:

In addition to Exome Array data, many of our NOMAS participants also had Affymetrix 6.0 whole-genome genotyping data available. These data were used for our common single nucleotide variant (CV) analyses since only 10.5% of CVs identified in the discovery families were available on the Exome Array (n = 574). Details of our genotyping and QC have been reported previously (Della-Morte ). These data were imputed using the 1000 Genomes phase 1, version 3 reference panel with IMPUTE2 v.2.2.2 (Howie ). Variants with INFO ≤0.4 were removed from analyses.

Statistical methods

Family-based discovery analyses:

As in previous analyses, LVM was natural log transformed and multiplied by 10 to ensure it was normally distributed and properly scaled for analyses in SOLAR 6.6.2 (Wang ; Della-Morte ). Common and rare SNVs were defined based on frequencies from Dominican NOMAS participants, as described previously (Wang ). SNVs were classified as common if they had MAF ≥5% and rare if they had MAF <5% or could not be imputed efficiently (INFO ≤0.4) in NOMAS Dominicans. Analyses in the families were performed using the sequencing data. Single-variant analyses were performed for CVs using the Quantitative Transmission-Disequilibrium Test (QTDT), implemented in SOLAR. Adjustment was made for sex, BMI, systolic blood pressure, and smoking status. Covariates were identified using a polygenic screen implemented in SOLAR, with covariates having P < 0.1 included in analyses. CVs with P < 9.13 × 10−6 were considered significant based on a Bonferroni correction of 5477 tests.

To evaluate the contribution of our most strongly associated CV to our linkage results, linkage analyses were performed in our combined sample of sequenced and replication families in SOLAR following our previously detailed protocol (Wang ). Briefly, linkage analyses were run with and without CV genotype as a covariate and a likelihood ratio test was performed to determine if the LOD score significantly decreased after conditioning on the CV.

Gene-based analyses were performed for RVs using the Family SNP-set (Sequence) Kernel Association Test (Fam-SKAT) v. 1.8 (Chen ), adjusting for the same covariates included in our single-variant analyses. We employed two different gene-based analyses based on annotation from ANNOVAR v. 2016Feb01 (Wang ) and SeattleSequation 138: exonic RVs (UTR3, UTR5, synonymous, missense, nonsense, or splice-site variants) and a subset of nonsynonymous RVs only (missense, nonsense, or splice-site variants). These analyses were restricted to genes with ≥2 polymorphic variants and a P < 1.83 × 10−4 was considered significant based on a Bonferroni correction of 273 tests (181 exonic RV genes and 92 nonsynonymous RV genes).

Using SAS v. 9.3, we computed the residual LVM value after adjusting for the associated risk factors to better visualize the distribution of RVs in relation to LVM.

Replication analyses:

LVM was natural log transformed and multiplied by 10 to ensure a normal distribution. We additionally removed outliers falling 3 SD above or below the mean LVM value. Single-variant CV analyses were performed in the NOMAS sample using linear regression, implemented in PLINK v. 1.7. Variants were coded additively. Replication was defined as CVs with P < 0.05 and having the same direction of effect. Affy 6.0 genotyping data were used for these analyses.

Gene-based RV association analyses using Exome Array data were performed using SKAT-O v. 1.1.2 in the NOMAS sample and Fam-SKAT v. 1.8 in the replication family sample. These analyses were performed using the same two filtering algorithms employed in the discovery analyses; all exonic RVs and then a subset of nonsynonymous RVs only. Analyses were restricted to genes with ≥2 polymorphic variants. Replication was defined as genes with P < 0.05. All association analyses were adjusted for the same covariates included in our discovery analyses. NOMAS analyses were additionally adjusted for factors that were associated with LVM (P < 0.1) in this sample: diabetes, time between LVM measurement and baseline patient assessment, waist hip ratio, and the first principal component obtained via principal components analyses implemented in Eigenstrat to account for population substructure. Details of our Eigenstrat analysis and resulting principal components can be found in Figure S1 and File S1.

Data availability

The data that support the findings of this study are currently being uploaded to dbGaP.

Results

Participant and SNV characteristics

A total of 180 individuals in 10 Dominican families were included in our discovery analyses. Characteristics of these 10 families are summarized in Table 1. Family size ranged from 10 to 35 individuals and family-specific LOD scores for the 12p11 region ranged from 0.14 to 0.94. LVM residual values ranged from −5.62 to 6.89. Mean age was between 40 and 54 yr and mean BMI was similar across the 10 families. Variability was seen with respect to percent of participants with diabetes and hypertension in each family (Table 1).

Within the 10 families, sequencing identified 5473 CVs and 10,167 RVs. Among RVs, 24.6% (n = 2503) were novel and 22.8% (n = 2318) were classified as exonic. A total of 20.2% of exonic RVs (n = 468 variants) were either missense, nonsense, or splice-site variants (Table 2). In the NOMAS sample, 96.5% of the CVs and 44.4% of the exonic RVs identified through sequencing were available for replication analyses. An additional 572 exonic RVs on the Exome Array were polymorphic in the NOMAS sample but were monomorphic in the discovery family sample and were included in gene-based replication analyses. In the replication families, 29.6% of the exonic RVs identified through sequencing were available for analysis in addition to 153 exonic RVs that were on the Exome Array and polymorphic in the replication families but monomorphic in the discovery family sample.

Table 2

Polymorphic variants identified by resequencing of chr12p11 region in 10 Dominican families

		MAFa
Type of Variantb	Total	Novel	<5%	≥5%
Missense	608	85	372	151
Nonsense	7	1	5	1
Synonymous	598	67	316	215
Splice site	6	1	4	1
UTR 3′ or UTR 5′	2042	374	995	673
ncRNA exonic	127	32	66	29
Intronic	11,380	1839	5627	3914
ncRNA intronic	251	43	116	92
Upstream or downstream	315	61	158	96
ncRNA Other	8	0	5	3
Intergenic	298	0	0	298
Total	15,640	2503	7664	5473

MAF based on NOMAS DR frequencies.

Function based on ANNOVAR(2016Feb01) annotation and dbSNP138/GVS 138 annotation.

CV association results

Results from the CV analyses are shown in Figure 1 and our most strongly associated variants with suggestive evidence for replication are listed in Table 3. While no CV reached peak-wide significance (P < 9.2 × 10−6), rs1046116 achieved a P-value of 9.0 × 10−4 and showed evidence for replication in NOMAS (P = 0.03). This variant was associated with decreased LVM and is a missense variant located within PKP2. To investigate this association further, we tested rs1046116 for association with LVM in our replication families and found no evidence for replication (P = 0.46). To evaluate the contribution of rs1046116 to our linkage signal, we performed linkage analyses in our combined discovery and replication family sample and observed a change in LOD score from 9.09 to 8.30 (P < 0.0001) with rs1046116 adjustment, indicating that this CV accounted for some of the linkage signal at the 12p11 QTL. Eight additional variants showed suggestive evidence for association in both the families and NOMAS (P < 0.05). The minor allele of these variants, except for rs11168985 (located in CPNE8), was associated with increased LVM.

Figure 1

Peak-wide common variant association results for the chr12p11 region. The multipoint LOD score over the region is shown as a solid line. Red circles represent Quantitative Transmission-Disequilibrium P-values in the Dominican families and black circles represent linear regression P-values in NOMAS. The blue dot above the multipoint LOD score line indicates the position of PKP2. The dashed line indicates suggestive association at P < 0.001.

Table 3

Common variants with P < 0.05 in the discovery families and evidence for replication in NOMAS

				Discovery Families			NOMAS
Gene	SNP	BP	Effect Allele/Reference Allele	EAF	Direction of Effecta	P-valueb	MAF	Direction of Effecta	P-valueb	Function
PKP2	rs1046116	33,021,934	G/A	0.15	−	9.04 × 10−4	0.20	−	0.03	Missense
ERGIC2	rs1035607	29,509,513	C/A	0.39	+	4.72 × 10−3	0.42	+	0.05	Intronic
OR10AD1	rs11168459	48,596,241	G/A	0.26	+	0.01	0.24	+	0.02	Missense
SLC38A4	rs2191162	47,197,648	A/G	0.27	+	0.01	0.34	+	0.03	Intronic
VDR	rs731236	48,238,757	G/A	0.29	+	0.02	0.34	+	0.03	Synonymous
ANO6	rs74081827	45,833,755	A/G	0.08	+	0.03	0.09	+	1.99 × 10−3	3′ UTR
TSPAN11	rs35989439	31,145,916	T/A	0.21	+	0.03	0.13	+	0.01	3′ UTR
CPNE8	rs11168985	39,045,983	A/C	0.25	−	0.04	0.22	−	0.01	Downstream
RPAP3	rs7311790	48,061,435	A/G	0.04	+	0.04	0.07	+	0.02	Intronic

+ indicates effect allele associated with increased LVM; − indicates effect allele associated with decreased LVM.

QTDT P-value for families, linear regression P-value for NOMAS. Bold typeface indicates p-value <0.05.

RV association results

Gene-based RV analyses were performed using two filtering algorithms; first, analyzing all exonic RVs (n = 181 genes) and second, restricting analyses to nonsynonymous RVs (n = 92 genes). A total of 54 genes showed suggestive evidence for association in the discovery families under at least one filtering algorithm (P < 0.05). Results for these genes are shown in Table S2. Among these top genes, we observed at least suggestive evidence for replication of five genes in NOMAS and three genes in the replication families (P < 0.10). Summary results for these genes are shown in Table 4. In exonic RV analyses, NELL2 was our most strongly associated gene (P = 2.2 × 10−4), although this finding did not replicate in NOMAS (P = 0.53) or the replication families (P = 0.18) (Table S2). While no gene achieved replication in both replication samples, our most strongly associated gene with evidence for replication in NOMAS was ALG10B (P = 0.02 in the discovery families; P = 0.04 in NOMAS; P = 0.86 in the replication families). When analyses were restricted to nonsynonymous RVs, this association became slightly attenuated in the discovery families (P = 0.06), and was even stronger in NOMAS (P = 0.03). GXYLT1 was our most strongly associated gene with evidence for replication in the replication families (P = 0.04 in the discovery families; P = 0.004 in the replication families; P = 0.36 in NOMAS).

Table 4

Gene-based association results for rare variant analysis in the chr12p11 region, for genes with P < 0.05 in the Dominican families and P < 0.10 in the replication families or NOMAS

		Exonic RVs						Nonsynonymous RVs
		Discovery Families		Replication Families		NOMAS		Discovery Families		Replication Families		NOMAS
Gene	MB Start	#SNVs	Pval	#SNVs	Pval	#SNVs	Pval	#SNVs	Pval	#SNVs	Pval	#SNVs	Pval
NELL2	44.90	24	2.2 × 10−4	5	0.18	8	0.53	7	0.006	2	0.02	5	1.00
PKP2	32.94	16	4.68 × 10−3	6	0.98	14	0.13	6	0.05	6	0.98	11	0.05
ALG10B	38.71	42	0.02	9	0.86	22	0.04	7	0.06	2	0.81	7	0.03
FAM186A	50.72	26	0.02	15	0.16	21	0.04	18	0.06	12	0.20	18	0.12
ZNF641	48.73	17	0.03	–	–	8	0.11	3	0.36	–	–	5	0.15
PPFIBP1	27.67	23	0.05	7	0.33	12	2.54 × 10−3	6	0.12	5	0.39	9	9.53 × 10−3
SLC2A13	40.14	19	0.06	7	0.39	9	0.16	2	0.02	3	0.53	3	0.04
GXYLT1	42.48	24	0.04	2	4.27 × 10−3	9	0.36	–	–	–	–	3	0.85
SLC38A4	47.16	21	0.02	6	0.07	12	0.89	3	0.29	–	–	5	0.88

In nonsynonymous RV analyses, IFLTD1 showed the strongest evidence for association in the discovery families (P = 0.003). However, since no variants met our inclusion criteria in NOMAS or the replication families, replication was not possible. NELL2 was our second most strongly associated gene (P = 0.006), and replicated in the replication families (P = 0.02) but not NOMAS (P = 1.00) (Table S2). Two genes, PKP2 and SLC2A13, were moderately associated in the discovery families and showed at least suggestive evidence for replication in NOMAS (P < 0.10), although they did not replicate in the replication families. Details of the nonsynonymous RVs in PKP2, SLC2A13, and NELL2 are shown in Table 5. Within the discovery families, six nonsynonymous variants in PKP2 were observed, all of which were missense variants. These variants were carried by individuals within five families: F2783, F3561, F3719, F5103, and F5275. In families F2783 and F3719, carriers of a rare PKP2 allele had higher average LVM residual compared to noncarriers. In the remaining families, carriers of a rare PKP2 allele had lower average LVM residual compared to noncarriers (Figure 2). When examining each missense variant individually, we observed that carriers of the rare allele of rs143004808, rs146882581, rs151264959, and rs62001016 had higher average LVM residual compared to noncarriers. In contrast, carriers of the rare allele of rs200069860 and rs75909145 had lower average LVM residual compared to noncarriers (Figure 3).

Table 5

Annotations for nonsynonymous RVs in PKP2, SLC2A13, and NELL2

					MAF
Gene	Variant	Positiona	Minor/Major Allele	Discovery Families with Variant	Discovery Families	Replication Families	NOMAS	Function	CADD Score	GERP Score	Amino Acids	PolyPhenb
PKP2	rs200069860	3,3030,850	A/T	F3561	0.008	–	–	Missense	18.04	5.37	GLY/CYS	Probably-damaging
PKP2	rs151264959	32,949,047	T/C	F2783	0.003	0.007	7.6 × 10−4	Missense	17.76	5.06	ASP/ASN	Probably-damaging
PKP2	rs146882581	32,994,073	A/G	F5103	0.003	0.007	0.008	Missense	3.33	2.01	THR/MET	Benign
PKP2	rs62001016	33,031,023	A/G	F5103, F5275	0.02	0.01	0.01	Missense	1.72	3.84	ALA/VAL	Benign
PKP2	rs75909145	33,049,457	A/C	F3561, F5275	0.03	–	–	Missense	14.02	3.17	SER/ILE	Benign
PKP2	rs143004808	33,049,590	T/C	F3719	0.008	–	0.005	Missense	33.00	4.07	ASP/ASN	Probably-damaging
PKP2	rs112592855	32,949,140	C/T	–	–	0.003	0.002	Missense	12.39	5.06	THR/ALA	Benign
PKP2	rs140852019	32,974,348	C/T	–	–	–	7.6 × 10−4	Missense	12.98	2.55	ASN/SER	Benign
PKP2	rs139159464	32,996,248	T/C	–	–	–	7.6 × 10−4	Splice site	6.18	1.10	–	Unknown
PKP2	rs201803918	33,030,840	A/G	–	–	–	7.6 × 10−4	Missense	11.25	5.37	ALA/VAL	Benign
PKP2	rs149542398	33,031,888	T/C	–	–	–	7.6 × 10−4	Missense	15.64	0.71	ARG/HIS	Benign
PKP2	12_32975421	32,975,421	A/G	–	–	–	0.01	Nonsense	39.00	4.16	ARG/Stop	Unknown
PKP2	12_33021968	33,021,968	A/G	–	–	–	0.005	Nonsense	38.00	2.05	ARG/Stop	Unknown
PKP2	rs146102241	32,977,026	T/C	–	–	0.007	–	Missense	23.60	5.32	VAL/ILE	Probably-damaging
PKP2	rs139734328	32,949,101	T/G	–	–	0.007	–	Missense	17.80	5.06	ARG/SER	Benign
SLC2A13	rs139518863	40,499,594	T/C	F3719	0.005	0.020	0.01	Missense	19.66	4.00	SER/ASN	Possibly-damaging
SLC2A13	rs186341127	40,499,132	A/G	F5103	0.008	0.010	0.01	Missense	7.55	1.54	ALA/VAL	Benign
SLC2A13	rs146020551	40,265,659	G/A	–	–	0.02	0.002	Missense	12.44	4.57	VAL/ALA	Benign
NELL2	rs367712742	44,902,736	C/T	F5103	0.01	–	–	Missense	11.99	5.25	GLN/ARG	Possibly-damaging
NELL2	12_45059356	45,059,356	T/C	F6081	0.003	–	–	Missense	9.88	4.38	ARG/HIS	Probably-damaging
NELL2	rs144730385	45,105,152	T/C	F3719	0.01	–	0.002	Missense	6.04	1.40	SER/ASN	Benign
NELL2	rs17574839	45,108,480	C/T	F5103, F5275	0.03	–	–	Missense	9.57	4.62	ASN/ASP	Benign
NELL2	rs201652982	45,171,085	T/C	F6081	0.003	–	–	Missense	32.00	5.62	ASP/ASN	Probably-damaging
NELL2	rs372522341	45,269,034	C/T	F6081	0.006	–	–	Missense	16.52	5.14	ASN/SER	Possibly-damaging
NELL2	rs2658973	45,269,640	T/C	F2235, F6081	0.008	0.05	0.02	Missense	13.58	3.08	VAL/ILE	Benign
NELL2	rs138454729	45,059,310	C/G	–	–	0.003	0.0008	Missense	18.80	4.39	ILE/MET	Possibly-damaging
NELL2	12_44926372	44,926,372	A/G	–	–	–	0.0008	Missense	27.50	5.72	SER/LEU	Probably-damaging
NELL2	rs146936717	44,915,791	T/G	–	–	–	0.002	Missense	17.79	4.71	ARG/SER	Probably-damaging

Position based on hg19.

PPH HumanDiv.

Figure 2

Box and whisker plots showing the distribution of LVM residual by PKP2 nonsynonymous rare variant carrier status, stratified by family. LVM residuals were calculated by adjusting for sex, body mass index (BMI), systolic blood pressure, and smoking status. Carriers of a PKP2 nonsynonymous rare variant are shown in gray and noncarriers are shown in white.

Figure 3

Box and whisker plots showing the distribution of LVM residual within the 10 sequenced families by PKP2 nonsynonymous rare variant carrier status, stratified by variant. LVM residuals were calculated by adjusting for sex, body mass index (BMI), systolic blood pressure, and smoking status. Carriers of a rare allele of a PKP2 variant are shown in gray and noncarriers are shown in white.

Since our most strongly associated CV was a missense variant in PKP2, we performed an additional gene-based analysis on all nonsynonymous variants in PKP2 (common and rare) which resulted in a slightly stronger association in the discovery families (P = 0.04, seven variants) and no association in NOMAS (P = 0.41, 12 variants) or the replication families (P = 0.98, seven variants).

Candidate variant and gene association results

We additionally tested whether candidate CVs identified in our previous study of LVM (Della-Morte ) (n = 33 CVs in 14 genes), a study which primarily investigated CV associations with LVM in all Caribbean Hispanics, were associated in our current study which focused specifically on Dominicans. Association results for these CVs are summarized in Table S3. Eleven of the previously reported CVs, located in SLC38A1, BICD1, RACGAP1, ARID2, FAR2, and c12orf68, were found in our discovery families. These CVs included three intronic variants in SLC38A1 which showed suggestive evidence for association in the discovery families and NOMAS, although the direction of effect differed (rs11183394, rs6582621, rs7133522). One other CV in SLC38A1, rs7956629, was moderately associated in the discovery families (P = 0.02), but did not replicate in NOMAS (P = 0.29). No other CVs were associated with LVM in the families.

We also performed gene-based RV analyses on 14 genes containing the previously reported candidate CVs. Results are summarized in Table S4. Exonic RV analyses identified PDZRN4 (P = 0.03) and ARID2 (P = 0.04) to be moderately associated in the discovery families, although neither replicated in NOMAS or the replication families (P > 0.05). One additional gene, c12orf68, was moderately associated in the discovery families (P = 0.03); however, replication was not possible as no variants met our inclusion criteria in either replication sample. In nonsynonymous RV analyses, SLC2A13 was the most strongly associated gene and was a top gene in our peak-wide nonsynonymous RV analyses.

Discussion

Building upon our previous studies which identified a region on 12p11 to be linked with LVM in Dominicans, we performed targeted resequencing in 10 extended families with evidence for linkage to refine the region and identify LVM susceptibility genes. Through these resequencing efforts, we found suggestive evidence for both common and rare variants within 12p11 to influence LVM in Dominican families. Common variant analyses revealed rs1046116, a missense variant in PKP2, to be the most strongly associated variant with evidence for replication in our population-based sample of Dominicans. When investigating this association further by testing rs1046116 for association in our 12 additional families with suggestive evidence for linkage to the region, we observed no association; however, these families generally showed weaker evidence for linkage to this region than the 10 families sequenced for our discovery analyses. Further, although this association could not be replicated in our replication families, linkage analyses in our combined family sample revealed that rs1046116 significantly contributed to the original linkage signal as evidenced by a decrease in LOD score from 9.09 to 8.30 in these families. This variant was associated with decreased LVM, suggesting that the G allele of rs1046116 may be protective against the development of left ventricular hypertrophy.

Interestingly, we also found suggestive evidence that RVs within PKP2 influence LVM. In nonsynonymous RV analyses, PKP2 was one of the most strongly associated genes with suggestive evidence for replication, thus suggesting that both common and rare nonsynonymous variants, particularly missense variants, may influence LVM. Indeed, when a gene-based analysis was performed on all nonsynonymous variants, common and rare, the association between PKP2 and LVM became slightly stronger in the families, although the association diminished in NOMAS.

To explore the associations underlying PKP2 further, we performed an in-depth characterization of the individual rare nonsynonymous variants within this gene. Three missense variants were predicted to be probably-damaging by PolyPhen and showed strong evidence of being conserved (rs200069860, rs151264959, rs143004808). Of particular interest was rs143004808, which was observed in F3719 and had a Combined Annotation Dependent Depletion (CADD) score of 33, indicating that this variant is in the top 0.1% of variants with respect to deleteriousness. Two nonsense variants (located at base pair positions 32,975,421 and 33,021,968) also had similar CADD scores and were observed in the NOMAS sample.

Missense variants in PKP2, including several of the ones identified in our study, have been previously implicated in cardiac phenotypes (Cerrone ; Peters 2014; van der Zwaag ; Saguner ), making PKP2 an excellent candidate gene for LVM. PKP2 encodes plakophilin 2, a member of one of three major protein families found in desmosomes (Getsios ). Desmosomes are protein structures in cell membranes that maintain adhesion between neighboring cells and serve as anchoring sites for intermediate filaments. They are found in tissues that experience mechanical stress, including the myocardium (Getsios ; Desai ). Mutations in PKP2 are known to play a role in arrythmogenic cardiomyopathies (AC), most notably arrythmogenic right ventricular dysplasia/cardiomyopathy (ARD/C) (Gerull ; Fernandez-Rosado ; Groeneweg ; Trenkwalder ). Several studies have shown that PKP2 haploinsufficiency contributes to pathogenesis in AC (Joshi-Mukherjee ; Hall ; Kirchner ; Rasmussen ).

In characterizing the clinical course of ARD/C, one study found that 9% of ARD/C patients with a PKP2 mutation had left ventricular dysfunction and patients carrying more than one mutation were three times more likely to have left ventricular dysfunction compared to patients with one mutation (Bhonsale ). Studies have also provided further evidence suggesting that PKP2 mutations may impact left ventricular cardiomyopathies (Horimoto ; Hamid ; Sen-Chowdhry ; Saguner ), including a recent case report that identified a pathogenic PKP2 deletion in two siblings with left ventricular noncompaction cardiomyopathy (Ramond ). Together, these studies support the biologic plausibility for a role of PKP2 in LVM.

In addition to PKP2, our analyses revealed nonsynonymous variants within SLC2A13 to be suggestively associated in both our discovery families and the NOMAS sample. Included in our nonsynonymous analyses were two missense variants observed in both the families and NOMAS (rs139518863 and rs186341127), as well as one missense variant observed in NOMAS only (rs146020551). Rs139518863 has strong evidence of being evolutionarily conserved (Genomic Evolutionary Rate Profiling (GERP) score = 4), is predicted to be possibly-damaging according to PolyPhen, and has a CADD score of 19.66 indicating that this variant is in the top 10–1% of all variants with respect to deleteriousness. Similar characteristics were observed for rs146020551. Interestingly, our previous association study found CVs in SLC2A13 to be associated with LVM in Caribbean Hispanics (Della-Morte ). These findings, in combination with results from our current study, suggest that variants in SLC2A13 may be involved in LVM in Dominicans and, more broadly, Caribbean Hispanics.

Additional candidate genes identified through our resequencing study were GXYLT1 and NELL2. GXYLT1 was identified in our exonic RV analyses and encodes an enzyme that adds xylose to O-glucose residues bound to epidermal growth factor repeats of Notch proteins (Sethi ). Notch signaling plays a role in cardiac development and mutations within Notch signaling genes have been associated with cardiac structural abnormalities including left ventricular outflow tract abnormalities, making GXYLT1 an excellent candidate gene for LVM (Nemir and Pedrazzini 2008; Padang ; Penton ). NELL2 was one of our most strongly associated genes in RV analyses and encodes the neural epidermal growth factor-like 2 protein which is largely expressed in brain but is also expressed in hematopoietic cells (Luce and Burrows 1999). Its role in LVM is unknown.

When evaluating the results from our study, there are several limitations which should be noted. First, only a subset of families was sequenced; however, these were the families with the strongest evidence for linkage to the 12p11 region. Second, sequencing was performed primarily on exons, therefore missing noncoding regions affecting LVM. However, for candidate genes previously implicated in LVM, sequencing beyond the exons was performed to allow for identification of noncoding variants. Third, the RV replication analyses were performed using Exome Array data, thereby limiting our analyses to only those variants included on the Exome Array. However, variants on the Exome Array were selected to be functional (primarily missense variants) and had to be observed at least three times in at least two people (http://genome.sph.umich.edu/wiki/Exome_Chip_Design), and we added custom content within the 12p11 region to our Exome Array. Fourth, our study was limited to individuals from the Dominican Republic and may not be generalizable to other Hispanic or non-Hispanic populations. Fifth, due to our sample sizes, we had limited power to detect RVs in our study.

In conclusion, our current targeted resequencing study, in combination with our previous studies, shows evidence of a role for both common and rare variants within the 12p11 region in LVM pathogenesis, particularly missense variants within PKP2. Functional studies are needed to elucidate the mechanism underlying the association of the implicated genes with LVM.

Supplementary Material

Supplemental material is available online at www.g3journal.org/lookup/suppl/doi:10.1534/g3.117.300358/-/DC1.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.