Transcriptional Impact of Rare and Private Copy Number Variants in Hypoplastic Left Heart Syndrome

Abstract

BACKGROUND

Hypoplastic left heart syndrome (HLHS) is a heterogeneous, lethal combination of congenital malformations characterized by severe underdevelopment of left heart structures, resulting in a univentricular circulation. The genetic determinants of this disorder are largely unknown. Evidence of copy number variants (CNVs) contributing to the genetic etiology of HLHS and other congenital heart defects has been mounting. However, the functional effects of such CNVs have not been examined, particularly in cases where the variant of interest is found in only a single patient.

METHODS AND RESULTS

Whole-genome SNP microarrays were employed to detect CNVs in two patient cohorts (N = 70 total) predominantly diagnosed with some form of nonsyndromic HLHS. We discovered 16 rare or private variants adjacent to or overlapping 20 genes associated with cardiovascular or premature lethality phenotypes in mouse knockout models. We evaluated the impact of selected variants on the expression of nine of these genes through quantitative PCR on cDNA derived from patient heart tissue. Four genes displayed significantly altered expression in patients with an overlapping or proximal CNV verses patients without such CNVs.

CONCLUSION

Rare and private genomic imbalances perturb transcription of genes that potentially affect cardiogenesis in a subset of nonsyndromic HLHS patients.

Introduction

Congenital heart defects (CHDs) not only comprise the single largest class of birth defects,¹ they account for 4.2% of all neonatal (i.e., less than 28 days) deaths and nearly one-quarter of all neonatal deaths attributable to birth defects in the United States.² The most commonly specified CHD causal for both neonatal deaths² and deaths during the first year of life³ is hypoplastic left heart syndrome (HLHS). HLHS encompasses a diverse spectrum of cardiac structural malformations of the left side,^4,5 with left ventricular hypoplasia as the hallmark feature. The resulting undersized chamber is insufficient for sustaining systemic circulation, making this condition universally fatal without surgical intervention. Even with advancements in staged surgical palliation, one year transplant-free survival remains poor at 68.7%.⁶

In addition to efforts to improve surgical outcomes, research on determining genetic causes and mechanisms of HLHS has been ongoing. A heritability analysis demonstrated that this disease is determined primarily by genetic factors.⁷ Genetic linkage studies have identified multiple significant loci in HLHS kindreds, including several that overlap loci found in kindreds with other left ventricular outflow tract malformations, providing evidence of genetic heterogeneity and shared etiology.^8–10 Association with several disorders caused by chromosomal aneuploidy (e.g., Jacobsen syndrome, Turner syndrome, and Down syndrome) is further evidence of the principal role genetics plays in the development of this condition. Although large-scale chromosomal abnormalities (i.e., greater than 1 Mb in length) are found in a high percentage of patients that also display extracardiac defects, the majority of HLHS cases do not harbor large-scale changes with additional phenotypes. The genetic alterations involved in the etiology of these nonsyndromic cases are still being sought.

A potential source of causal variants for nonsyndromic HLHS is single gene mutations. Mutations in the cardiac transcription factor NKX2-5¹¹ and the signaling receptor NOTCH1^12–14 have been identified in a handful of these cases. Similarly, a small number of patients have been found to harbor mutations in genes that control left-right embryonic axis patterning, but these patients also display the typical extracardiac defects observed in heterotaxy. Thus, the variants discovered in single genes so far account for a very small fraction of nonsyndromic HLHS cases, necessitating additional research to find other sources of genetic variation contributing to this ailment.

Recently, genomic structural variation, in the form of copy number variants (CNVs), has been explored for association with both syndromic and nonsyndromic CHDs.^13,15–27 Potentially etiologic CNV loci have been identified using various criteria, including size, rarity, gene content, and de novo status. Yet the phenotypic consequences of altering the dosage of genes within these loci were infrequently addressed. Additionally, transcriptional effects of these variants on overlapping genes of interest were not determined. In this report we detail our own search for CNVs in HLHS cases, as well as our efforts toward verifying their functional impact on transcription in order to demonstrate etiologic potential.

Methods

Patient Cohorts

Samples from two independent subject cohorts were obtained for this study. The first cohort consisted of 23 patients with CHDs, primarily HLHS (Supplemental Table 1). Most of the HLHS cases (14 out of 16) had no family history. Two of these cases displayed dysmorphia, indicative of a potential genetic syndrome, at the time of sample collection. Samples consisted of genomic DNA extracted from peripheral blood.

The second cohort, detailed in Supplemental Table 2, contained over 50 patients with non-familial, nonsyndromic HLHS or a variant thereof, with only two (HLHS30 and HLHS53) having a previously-diagnosed genetic syndrome (Turner syndrome). All of these patients underwent heart transplantation, at which time tissue from the explanted heart was collected, snap frozen in liquid nitrogen, and stored at −80°C. DNA was extracted from these tissues with either a commercially-available kit (DNeasy Blood & Tissue Kit, Qiagen, Valencia, CA) or by proteinase K digestion in lysis buffer (50 mM Tris, pH 8.0; 100 mM EDTA; 100 mM NaCl; 1% SDS), followed by ethanol precipitation.

Prior to SNP array genotyping, all DNA samples were diluted to 50 ng/μl, as measured either by the NanoDrop ND-3300 Fluorospectrometer (NanoDrop Technologies, Wilmington, DE) or the Wallac 1420 Multilabel Counter (PerkinElmer, Waltham, MA), with the Quant-iT PicoGreen dsDNA Assay Kit (Invitrogen/Molecular Probes, Eugene, OR). Samples were also verified to be of high molecular weight by agarose gel before submittal to the University of Colorado Anschutz Medical Campus Genomics and Microarray Core (Aurora, CO), for genotyping with the HumanOmni2.5 BeadChip (Illumina, San Diego, CA) SNP microarray platform. This study was approved by the Institutional Review Boards of the University of Colorado and the University of California San Diego; informed consent was obtained from the parents of all subjects.

CNV Detection and Selection

All samples from the first patient cohort, along with 52 samples from the second patient cohort, were genotyped by microarrays. Five samples from the second cohort yielded either unacceptably low genotype call rates or excessive segmentation, and were excluded from further analysis. Supplemental Table 2 displays the patient information for the 47 samples included from that cohort.

Figure 1 shows the analysis steps performed on the data generated by the Illumina arrays. These data were evaluated using two different workflows. Illumina’s GenomeStudio (version 2010.3) and the R/Bioconductor package crlmm²⁸ independently generated sample genotypes, normalized sample/reference probe intensity ratios (i.e., LogR Ratios), and SNP B allele frequencies (BAF) for evaluating zygosity. GenomeStudio utilized an in-house derived reference file created from the results of 99 control samples, whereas crlmm employed a reference annotation package created by the software developer. The data generated by GenomeStudio have been deposited in NCBI’s Gene Expression Omnibus (GEO), and can be accessed through GEO Series accession number GSE66032.

Figure 1.

Flow chart of data analysis steps. Raw data were imported into two independent genotyping and sample/reference ratio-calculating programs. Output from each of these programs was analyzed by software implementing two different classes of CNV detection algorithms, hidden Markov model (HMM) and circular binary segmentation (CBS). These programs generated four lists of putative CNVs. 645 CNV segments, identified in at least 3 of the 4 lists, were further examined.

The resulting genotyping and intensity data from each of these software packages were channeled into programs that implement two classes of CNV detection algorithms, the hidden Markov model (HMM) and circular binary segmentation (CBS). We employed the R/Bioconductor packages VanillaICE and DNAcopy to identify CNV segments within crlmm-derived data via HMM and CBS, respectively. Similarly, the Illumina HMM plug-in to GenomeStudio (cnvPartition, version 2.4.4) and the genomic segmentation (CBS) tool in Partek Genomics Suite version 6.6 (Partek Incorporated, St. Louis, MO) were executed on GenomeStudio-derived data. Unless noted, default settings were used for all programs, with requisite parameter specifications for applicable packages detailed in the Supplemental Methods.

After CNV detection, concordance between the four methods was evaluated to create a list of putative CNVs that could be prioritized into candidates. This strategy was employed to reduce the number of possible false positive segments. A variant was included if it was identified by at least three of the four detection algorithms. This condition narrowed the list to 645 autosomal CNVs (Supplemental Table 3).

Prioritization for the variant data was performed at the gene level (Figure 2). Specifically, lists of protein-coding genes overlapping detected CNV segments or within 50 kb of their breakpoints were compiled. Segment verification was performed by visualization of logR ratios and B allele frequencies in both Partek and GenomeStudio, via its Illumina Genome Viewer. List entries were evaluated for potential cardiovascular development roles by queries in the Mouse Genome Informatics Database (http://www.informatics.jax.org/allele), an integrated laboratory mouse database website. Genes whose knockout or transgenic mouse models showed either a cardiovascular phenotype or premature lethality (a possible sign of cardiovascular malformation) were considered to be prime candidates for further investigation. This process highlighted 59 entries (Supplemental Table 4). To distinguish verified CNVs that are likely benign polymorphisms from rare variants with etiologic potential, segments were assessed for overlap with entries from the Database of Genomic Variants (DGV) (http://dgv.tcag.ca/dgv/app/home) in March of 2014, using a track on the University of California Santa Cruz Genome Browser (http://genome.ucsc.edu/cgi-bin/hgGateway). Segments completely or partially overlapping fewer than 10 DGV entries or containing gene exons that overlap fewer than 10 DGV entries were considered rare, and their corresponding genes retained. This step generated a list of 20 genes for possible follow-up (Table 1).

Figure 2.

Flow chart of candidate prioritization steps. CNV data were first evaluated for protein-coding gene overlap/proximity, and then graded on aspects of these genes.

Table 1.

Candidate CNVs discovered in both patient cohorts with Illumina arrays

Sample	Cytoband	Coordinates (hg19)	Length (bp)	Estimated Copy Number	Candidate Genes Within or Near CNV	qPCR-Verified CNV?
PG05	5q33.3	158899789-159627399	727611	3	ADRA1B	yes
HLHS56	6p24.1	12306232-12320786	14555	1	EDN1	yes
HLHS24	11p15.4	10368614-10404266	35653	3	ADM	yes
PG07	11q13.2-q13.3	68100519-69469069	1368551	3	CPT1A, IGHMBP2	yes
HLHS27, HLHS39	16q23.3	83196349-83209154	12806	1	CDH13	yes (39 only)
HLHS08	1q24.2	170560604-170613602	52999	1	PRRX1	not tested
HLHS57	2p21	45408893-45974339	565447	3	PRKCE	not tested
HLHS22	3p22.3	32953920-32985486	31567	1	TRIM71	not tested
HLHS27	7q21.12	87524013-87656083	132071	3	ADAM22	not tested
HLHS44	9q21.32-q21.33	86881490-87359570	478081	3	NTRK2	not tested
PG16	15q26.1	90108613-90171306	62694	3	KIF7	not tested
HLHS23	16q12.2	56653745-56694878	41134	1	MT1E, MT1F	not tested
HLHS44	17q12	32441969-32566206	124238	3	ASIC2, CCL2	not tested
HLHS23	21q21.3	27513837-28194874	681038	3	APP	not tested
HLHS23	21q21.3	28202149-29000785	798637	3	ADAMTS1, ADAMTS5	not tested
HLHS41	21q22.2	41034215-41435172	400958	3	DSCAM	not tested

Note: The candidate gene(s) may not be the only gene(s) within or near the CNV.

Quantitative PCR (qPCR) Validation

Selected copy number variants were validated using pre-designed TaqMan Copy Number Assays (Applied Biosystems, Carlsbad, CA). Each sample reaction contained approximately 20 ng of genomic DNA, and four replicates per sample were assayed. All reactions were performed in an Applied Biosystems 7300 Real Time PCR System, with SDS v1.3.1 software applying manufacturer-recommended parameters. Copy number determinations were provided by Applied Biosystems CopyCaller Software v1.0, using analysis settings of no calibrator sample and a most frequent sample copy number of two.

In addition to patient samples, a portion of the DNA Polymorphism Discovery Resource Collection (Coriell Institute for Medical Research, Camden, NJ) was analyzed with the same assays for copy number. The first 90 samples, a manufacturer-defined subset of the panel, were tested in the same manner as the patient samples. This subset served as an additional gauge for the rarity of an assayed CNV within the entire human population, since the collection consists of samples representative of ancestries from the major geographic regions of the world.²⁹

Gene Expression Assays

Supplemental Table 5 displays the putative candidate genes assessed for expression in cardiac tissue with the listed Applied Biosystems TaqMan Gene Expression Assays. Most of these genes (ADAMTS1, ADM, EDN1, and NTRK2) were selected based on exhibition of a cardiac structural phenotype, in addition to premature lethality, in mouse knockout models.^30–33 The only gene tested lacking a cardiovascular phenotype in a mouse model was DSCAM, although the model does show perinatal lethality,³⁴ and the gene lies in a defined critical region of Down syndrome-associated congenital heart disease.^35,36 Thus, all of the presumptive candidates chosen for evaluation have plausible links to cardiac physiology or structural development.

Candidate expression was assessed in total RNA extracted from several sources. One source was pooled whole hearts, dissected from either E9.5 or E10.5 129S6 mouse embryos and placed into RNAlater Stabilization Solution (Ambion, Carlsbad, CA) prior to extraction. The tissue from six E10.5 and seven E9.5 hearts was lysed by vortexing in Buffer RLT (Qiagen, Valencia, CA), with 2-mercaptoethanol (Fluka Analytical, St. Louis, MO) added. Lysates were homogenized in a Qiagen QIAshredder spin column prior to RNA cleanup and concentration with the Qiagen RNeasy MinElute Kit, following the manufacturer’s protocol. Another source of RNA was explanted human heart ventricle tissue from the HLHS patients evaluated for copy number variation. Approximately 500 ng of extracted RNA was obtained from the seven patients harboring CNVs of interest, in addition to 13 gender-matched patients without these CNVs. Finally, we purchased pooled Human Heart Total RNA and pooled Human Fetal Heart Total RNA from Clontech (Mountain View, CA) to use as normal controls for comparison.

Prior to expression assessment, RNA samples were converted to cDNA using the Applied Biosystems High-Capacity cDNA Reverse Transcription Kit, according to the manufacturer’s protocol. Volumes were adjusted to create RNA reaction concentrations of approximately 10 ng/μl. One microliter from each of these reverse transcription reactions was used as the template for each gene expression reaction assembled; three replicates per sample were assayed. All expression experiments were performed in an Applied Biosystems 7300 Real Time PCR System, with SDS v1.3.1 software applying manufacturer-recommended parameters. In addition to the candidate genes, an endogenous reference gene, GAPDH, was evaluated and used to normalize the results.

Mouse assay data were analyzed for a relative change in gene expression between E9.5 and E10.5 by the 2^−ΔΔC_T method.³⁷ Alternatively, we compared normalized patient cycle threshold (C_T) values with an unpaired, two-tailed t-test using GraphPad Prism 6 (GraphPad Software, La Jolla, CA). To normalize, we computed the average C_T value of GAPDH for each patient sample and subtracted this value from each corresponding patient test gene C_T (ΔC_T). The ΔC_T values of a patient sample harboring the CNV of interest were compared to the values of the remaining patient samples without the CNV. The pooled adult and fetal heart samples were not included in these comparisons.

Results

Putative Candidate CNV Identification and Validation

We leveraged the availability of software for Illumina array results interpretation to create a consensus-based analysis strategy. Two genotyping and four segmentation algorithms generated lists of putative CNVs (see Methods and Figure 1). These lists were evaluated for concordant entries; 645 putative CNVs were identified by at least three of the four algorithms executed (Supplemental Table 3). The concordant entries were ranked by features of the genes contained within or near them. Prioritization of these entries via the scheme detailed in Figure 2 yielded 20 candidate genes potentially impacted by 16 rare or novel variants, 15 of which were found in separate patients (Table 1).

In addition to concordance, validation of aberrant copy number at several candidate loci was performed via qPCR. Having a greater confidence in what was identified due to the array’s high marker density and multiple algorithm detection, we only chose a subset of CNVs to verify. The shortest, a 12806 bp deletion covering a single exon of the cadherin 13 gene (CDH13) was detected in two patients, HLHS39 and HLHS27, by one and three packages, respectively. Unfortunately, only patient HLHS39 had a sufficient amount of sample remaining for confirmatory testing. The other four loci chosen for validation were discovered in individual patients.

Copy number variation was verified for all five selected loci in the predicted samples. Surprisingly, the amplification identified in sample HLHS24 near ADM displayed a calculated copy number of around four by qPCR (Figure 3A), whereas the array estimated a copy number of three. The remaining case cohort samples yielded concordant qPCR and array copy number results. Another unexpected finding was discovering a single PDR sample with a deletion at the candidate locus near EDN1 (Figure 3B). All of the other PDR samples displayed a normal copy number at each tested locus. These results corroborate the apparent private nature of most of the CNVs chosen for investigation.

Figure 3.

Bar graphs of calculated copy number for two candidate loci. A, Results for locus near ADM. Sample HLHS24 yielded a calculated copy number near four in two separate runs. B, Results for locus near EDN1. One case sample (HLHS56) and one control sample (PD0029, from PDR Collection) indicated a single-copy deletion; each result was confirmed by an additional run.

Expression of Putative Candidate Genes in Mouse Heart

To substantiate the potential of selected candidate genes to impact cardiac structural development, we verified their expression in embryonic mouse heart using TaqMan Assays (Supplemental Table 5). All of the putative candidate genes tested were expressed in the heart at the two time points studied, E9.5 and E10.5. Since two time points were examined, the relative change in expression of these genes from E9.5 to E10.5 could be determined. Figure 4 graphically displays the results of this comparison. Transcript levels of Dscam were virtually unchanged during the period. Only the transcript for Adm showed a slight, yet significant, decrease in expression, whereas the other genes studied showed significant increases. Two of them, Cdh13 and Ntrk2, were upregulated approximately 4-fold, a substantial amount that is suggestive of involvement with an emerging process during that time. Several steps of cardiogenesis commence during this time frame, including formation of endocardial cushions, the precursors to heart valves. It has been suggested that severe valve malformation underlies the development of HLHS.⁷ Thus, we contemplate that alterations in the dosage of these presumptive candidates may disrupt the normal formation of valves and other cardiac structures that are beginning to take shape during this period.

Figure 4.

Relative change in expression of putative candidate genes from E9.5 to E10.5. The change was significant for every gene except Dscam (based on unpaired, two-tailed t-test, assuming equal variances). Error bars denote maximum and minimum fold change values based on the highest and lowest ΔC_T values for each gene tested.

Expression of Putative Candidate Genes in Patient Hearts

Having established the expression of our candidate genes in the heart during a relevant period of structural development, we next wanted to determine if this expression was significantly altered in specific patients. The majority of the patient samples tested were derived from explanted cardiac ventricle tissue removed at the time of transplantation. This afforded the opportunity to directly measure candidate gene dosage effects on ventricular expression. Of the ten patient CNV-associated candidates chosen for expression studies, nine had corresponding patient RNA available to test, with ADRA1B being the lone exception. One of the candidates, CDH13, appeared to be disrupted by an identical deletion in two patients, although the CNV could only be verified in one patient due to the total consumption of the DNA sample from the other patient. Fortunately, RNA extracts from both patients were on hand. Two other candidates, ADAMTS1 and ADAMTS5, are adjacent to each other on chromosome 21 and overlapped by the same amplification. This circumstance can serve as a test of whether the expression of all genes within a structural variant are affected similarly (i.e., a strict dosage effect).

To explicitly determine possible effects of dosage on cardiac transcription, comparisons of candidate gene expression were made between samples from patients harboring CNVs that overlap or are proximal (i.e., within 50 kb) to the candidates, and samples from patients without these same CNVs. Using gender-matched patients from the same cohort should help control for possible effects from disease state and patient age, although both of these factors are fairly heterogeneous in our sample set. To see if age and disease state had a noticeable effect on expression of these candidates, pooled normal adult and fetal heart RNA samples were obtained and assayed along with the cohort samples. A qualitative inspection of the data from both of the pooled samples revealed similar values, and these values were well within the range of those exhibited by the patient samples (Supplemental Figure 1).

Having ruled out disease status and age as confounders, we proceeded with the comparisons. Four of the nine candidate genes demonstrated a significant difference in expression between patients with and without overlapping or proximal CNVs, at a level of significance of P<0.05 (Figure 5). ADAMTS1 was completely covered by a genomic amplification, whereas only the first 9–12 exons (out of 13–21, depending on transcript variant) of NTRK2 were overlapped by a different amplification. Both genes displayed an increase in expression, consistent with a copy number increase. Likewise, the qPCR-verified patient sample displaying a heterozygous deletion of a single exon in CDH13 showed decreased expression of the transcript specific to this exon. Another patient sample with a heterozygous deletion 8,806 bp from the 3′ end of EDN1 demonstrated reduced expression of this proximal gene, implying the disruption of a cis-regulatory element by the CNV. Therefore, in some cases alterations of gene or regulatory element dosage by overlapping CNVs appear to have a significant impact on cardiac gene expression. Interestingly, a strict dosage effect does not seem to exist, since expression of ADAMTS5, a gene adjacent to ADAMTS1 and completely overlapped by the same CNV in a patient, did not display the same significant increase (Supplemental Figure 2). Alternatively, other regulatory mechanisms may compensate for the altered dosage of particular genes.

Figure 5.

Plots of the ΔC_T mean and standard deviation for genes found to differ significantly between patients with and without overlapping or proximal CNVs. N.S. indicates not significant.

Discussion

In this report, we document our search for genomic imbalances that could play a role in the etiology of congenital heart defects, particularly nonsyndromic HLHS. Our strategy for identifying potentially deleterious CNVs was primarily based on gene content, similar to other reports.^15,20,26 Unlike other studies, we also considered genes that were near a CNV breakpoint (i.e., within 50 kb) to account for possible cis-regulatory element dosage effects. Our approach highlighted 20 putative candidate genes potentially altered by 16 variants, four of which had no gene overlap. Only one of the 16 CNVs was identified in more than a single patient, although this result could only be verified in one of the patients due to a lack of sample.

Hence, nearly all of our highlighted CNVs appear to be private; comparable observations have been noted elsewhere.^20,23,27 A private mutation model of disease would befit a heterogeneous, mostly sporadic condition such as HLHS. Yet previous studies, including our own, generally lacked the power to detect recurrent, yet extremely rare, causative CNVs due to insufficient sample sizes. Therefore, we adopted an alternative strategy: assessing the effects of some of our noted CNVs on the expression of putative candidate genes overlapped by or proximal to these variants.

First, we verified candidate gene expression in embryonic mouse heart during a period that corresponds to the start of valvulogenesis, a process that is often disrupted in HLHS patients. In addition to finding that every candidate was expressed, we discovered that several of these genes became substantially upregulated during the time frame. This result hints at the participation of these candidates in the formation of valves or other cardiac structures arising at that time.

Second, we measured nine of these genes for differential expression in human ventricular tissue between a patient harboring the gene-overlapping or proximal CNV and patients without the same CNV. Four of the nine, including one (EDN1) that was not overlapped by but proximal to a CNV, demonstrated significantly altered expression in the same direction as the copy number change, loss or gain (Figure 5). These findings underscore possible dosage sensitivity and involvement of several novel genes in the etiology of HLHS.

One of these genes, ADAMTS1, has also been identified by other groups within genomic imbalances of CHD patients that are absent in controls.^21,23,25 The gene encodes a metalloproteinase that regulates, through its proteolysis of the extracellular matrix proteoglycan versican, several vital cardiac developmental processes in mice, including remodeling of endocardial cushions³⁸ and the outflow tract,³⁹ as well as myocardial trabeculation.³² It is also located on human chromosome 21, whose trisomy results in Down syndrome, a condition associated with a high frequency of CHDs, though not typically HLHS. Thus, ADAMTS1 has multiple lines of evidence linking its aberrant dosage with cardiac structural defects, making it a strong candidate for involvement in the development of HLHS and other CHDs.

Unlike ADAMTS1, the other candidates we discovered, to our knowledge, have not been previously reported in human CHD studies. However, several of them have been implicated in normal mouse heart development, including EDN1 (endothelin 1)^31,40 and NTRK2 (neurotrophic tyrosine kinase, receptor, type 2; also known as TrkB).^33,41 Whether or not these candidates influence valvulogenesis or the formation of septal structures remains to be elucidated. Likewise, in some cases the determination of the precise mechanism of how specific patient CNVs are altering gene expression will require additional experiments. Yet the demonstration of their aberrant expression in human samples, combined with previous mouse data, supports their inclusion as valid candidate genes worthy of further investigation. We anticipate that as more candidate loci are identified, their functions and the cell types that they act upon will start to converge upon a limited number of cardiogenic networks or pathways. This could permit the development of biomarker screening panels to determine which processes are perturbed in a fetus, and perhaps direct appropriate intervention measures if available in the future.

Conclusion

We have identified rare and seemingly private copy number variants in HLHS patients that display a significant impact on the transcription of neighboring or overlapping cardiovascular genes. Our approach permits the discovery of credible candidates for involvement in the etiology of diseases with high heterogeneity and relatively low frequency.