ROBO2 Variants Associated with Atrial Septal Defect Define a Novel Regulatory Element

Abstract

Background:

Atrial septal defects (ASDs) are a prevalent types of congenital heart disease (CHD). Previous genome-wide association studies (GWAS) have identified common variants associated with ASDs, though their mechanisms remain unknown. We aimed to expand insights into the architecture of common variants associated with ASD risks and elucidate functional mechanisms.

Methods:

We conducted a GWAS using isolated ASD cases and healthy controls and replicated findings in an independent cohort. We examined epigenetic marks within this ASD locus in human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) and fetal human hearts. We characterized the consequences of deletions introduced by CRISPR-Cas9 mutagenesis of hiPSCs to assess the effect on downstream gene expression. Additionally, we investigated the 3D genome architecture of the locus using chromosome conformation capture sequencing.

Results:

We identified a novel ASD locus on chromosome 3p12.3 encompassing the ROBO2 gene, which encodes the Roundabout guidance receptor-2 for Slit ligands. This locus includes fifteen common single nucleotide polymorphisms (SNPs), an enhancer, and a CTCF binding site. Deletions of varying lengths within the ASD-associated locus in hiPSC-CMs reduced ROBO2 expression and dysregulated the expression of extracellular matrix genes. Chromosome conformation capture sequencing indicated this region physically interacts with the ROBO2 promoter and demonstrates that the CTCF binding site is essential for this contact.

Conclusions:

Novel common SNPs in regulatory elements controlling ROBO2 transcription contribute to risks for ASDs. These data infer key roles for the Roundabout guidance receptor-2 and Slit ligands in embryogenic development and maturation of the atrial septa.

Introduction

Atrial septal defects (ASDs) are congenital heart malformations that occur in approximately 2 of 1000 live births¹ and cause shunting of oxygenated blood into the pulmonary circulation. Previous genetic studies have identified familial dominant, pathogenic variants that cause isolated ASDs^2–5 and ASDs that occur as components of congenital syndromes^6,7. However, the causal genetic etiology is undefined among the majority of ASD patients, who have with sporadic occurrence or familial clustering in the absence of a genetic syndrome⁸. Better understanding of the genetic etiology of ASD can improve patient health outcomes as well as our understanding of complex developmental processes. As ASDs usually have little impact on reproductive or general health until middle age when arrhythmias and pulmonary hypertension cause symptoms, we hypothesized that common variants implicated in ASD can be inherited without negative selection. Consistent with this model, previous genome-wide association studies have defined multiple ASD susceptibility loci that function via unknown mechanisms^9,10.

To better understand the contribution of common variants to ASD, we conducted a genome-wide association study (GWAS) on subjects with unexplained, isolated ASDs that occur without additional cardiac or extracardiac birth defects or developmental syndromes. Studied subjects had no familial history of CHD and prior genetic analyses failed to identify causal variants. We identified and validated novel ASD-associated SNPs located in the second intron of Roundabout guidance receptor 2 (ROBO2), a gene that encodes a cell membrane receptor interacting with the Slit guidance ligand (SLIT).¹¹ Within this ASD locus we identified 15 common SNPs, a CTCF binding site, and a putative cardiac enhancer. By genetic manipulation of sequences encompassing ASD-associated SNPs, we demonstrate these critical sequences for ROBO2 expression. These findings highlight the role of common genetic variants in the development of ASDs and reveal a novel regulatory element for ROBO2 transcription.

Methods

Genotyping data for GWAS is available through dbGaP (phs000571.v1.p1). ATAC-seq results from hiPSC-CMs during differentiation have been published^12,13 and are available through GEO (GSE283866). The data, analytic methods, and study materials that support the findings of this study are available upon reasonable request.

The protocol was approved by the Institutional Review Boards of Boston Children’s Hospital, Brigham and Women’s Hospital, Great Ormond St. Hospital, Children’s Hospital of Los Angeles, Children’s Hospital of Philadelphia, Columbia University Medical Center, Icahn School of Medicine at Mt. Sinai, Rochester School of Medicine and Dentistry, Steven and Alexandra Cohen Children’s Medical Center of New York, and Yale School of Medicine. Written informed consent was obtained from each participating subject or their parent/guardian.

Most ASD probands and parents were recruited into the CHD Genes Study of the Pediatric Cardiac Genomics Consortium (CHD genes: ClinicalTrials.gov identifier NCT01196182) as previously described¹⁴ using protocols approved by institutional review boards of each institution. 88 ASD probands were ascertained in the Cardiac Center at The Children’s Hospital of Philadelphia in accordance with an IRB approved protocol prior to the initial of the CHD genes study. Probands selected for this study had no history of CHD in first-degree relatives. ASD diagnoses were obtained from echocardiograms, catheterization, and operative reports. The pathogeneses for ASD were unknown; patients with previously identified cytogenetic anomalies or pathogenic CNVs identified through routine clinical evaluation were excluded. After obtaining signed informed consent, case and parental (when available) whole blood samples were collected, processed for DNA and stored using standard techniques.

Control cohorts were datasets stored in dbGaP. NHLBI Framingham SNP Health Association Resource (SHARe) genotyped on the Illumina HumanOmni5M-4v1 array, dbGaP accession number phs000342.v16.p10 and the Health and Retirement Study (HRS) genotyped on the Illumina HumanOmni2.5–8v1_A array, dbGaP accession number phs000428.v1.p1.

Results

Genome-wide association studies of Atrial Septal Defect patients identified a novel association with 3p12.3.

We compared genotypes from 296 ASD subjects (obtained on Illumina HumanOmni2.5–8v1_A platform) with genotypes from 2473 control participants (Framingham Heart Study offspring cohort; (Illumina HumanOmni5M-4v1 genotypes accessed via dbGaP phs000342.v16.p10). After removing individuals of divergent ancestry by mapping both disease and control cohorts with the four HapMap populations and applying other quality control procedures¹⁵ (see Methods), we compared the allele frequencies in 1,377,257 SNPs from 177 ASD cases and 2416 controls. Two SNP clusters achieved genome-wide significance (> 5E-8): four SNPs on chromosome 3 (including rs34785004: OR=3.4, P=3.1E-11; Table S1) and three SNPs on chromosome 6 (Figure 1A, Figure S1, and Table S1).

Figure 1. Identification of ROBO2 SNPs associated with ASDs.

(A) Manhattan plot showing the statistical association of SNPs with ASDs in the discovery cohort. Four SNPs on chromosome 3 within ROBO2 that achieve genome-wide significance (P ≤ 5 × 10⁻⁸, red line) are shown. (B) Locus zoom plot of the GWAS results. (C) Schematic of the ASD haplotype, indicating the locations for the CTCF binding motif (dark blue triangle), putative enhancer (dark blue box), ROBO2 exons (light grey box), and SNPs (red, GWAS SNPs and other SNPS in LD; grey, SNPs in LD with GWAS SNPs with higher MAF in subjects with non-EU ancestry). (D) CTCF ChIPseq of hiPSCs. (E) Histone modification ChIP-seq of D30 hiPSC-CMs (F) Fetal heart tissue histone modification ChIP-seq. In D to F, ChIP-seq peaks are shown as a box. (G) ATAC-seq of hiPSC-CMs from undifferentiated hiPSCs (D0) to hiPSC-CMs (D30), with peak height indicating normalized read counts.^12,13 CPC: Cardiac progenitor cells. (H) DNase-seq of fetal heart tissue, with genomic locations at the bottom. Panels D to H are drawn on the same genomic coordinates. F and H were adapted from ENCODE.¹⁶

Analysis of the quantile-quantile plot (genomic inflation factor = 1.03, Figure S1A) indicated minimal population stratification. In replication analysis, we studied 144 additional ASD cases and 13,256 controls (Health and Retirement Study (HRS) participants; dbGaP phs000428.v1.p1) that were both genotyped on the Illumina HumanOmni2.5–8v1_A array. After applying quality control filters, the allele frequencies for 1,430,752 SNPs in 95 ASD and 8,794 control subjects confirmed the association of ASD with chromosome 3 SNPs (p=0.002 for the haplotype, OR=2.1), but they did not validate the chromosome 6 SNPs (Table S1). Combining data from the discovery and replication cohort, we found that 18% of ASD subjects carried the chromosome 3p12.3 haplotype compared to 6.7% of control subjects (P=2.5E-9; OR=2.8; Table S1).

3p12.3 carries 15 common SNPs and putative regulatory elements.

The four ASD-associated SNPs (rs113434908, rs34785004, rs34915655, and rs10511057) reside within an 11-kb region within the 379-kb intron 2 of ROBO2 (Figure 1B, C). While ROBO2 is mainly recognized for its role in axon guidance¹¹, it also participates in other biological processes¹⁷, including cardiac development¹⁸. The Robo-Slit family regulates cell adhesion and migration of heart tube formation in Drosophila.^19–21 In mice, Robo2 is expressed in both atria at E10.5 and in the endocardium of the great arteries at E12.5.²² Although Robo2 knockout mice do not show any cardiac abnormalities. Robo1 null mice display various septal defects, including atrioventricular septal defect (AVSD), ventricular septal defects (VSD) and ASDs,^8,19 and VSD are increased in Robo1/Robo2 knockout mice compared to Robo1 knockout mice.²³ Rare human loss of function variants in ROBO1 cause VSDs or tetralogy of Fallot (ToF).²⁴ Together these data indicate cardiac developmental roles for the ROBO pathway as well as functional redundancy between Robo1 and Robo2.

We sequenced the entire ROBO2 gene in 36 ASD cases and 34 unaffected parents with the risk haplotype (Methods; Supplemental File 1). No damaging coding variants or rare promoter variants were identified in linkage disequilibrium (LD) with the ASD-associated SNPs (Supplemental File 2). Consistent with the HaploReg database²⁵, we found 11 SNPs in LD with the four GWAS SNPs (Figure 1C, labeled as SNP1 through 15) that form a 29-kb ASD risk haplotype within intron 2. Given higher minor allele frequencies (MAF ≥ 0.07) for five SNPs among African and Asian populations and the location of two SNPs in repeat sequences, we suggest that these are less likely contributors to ASD than the other eight SNPs (Table S2).

In addition, we examined whole genome sequencing (WGS) data (N=5706)¹² of CHD trios and surveyed additional variants located within the haplotype. Among 640 individuals carrying the ASD risk SNPs, we identified five simple tandem repeat (STR) alleles within the ASD locus enriched among the ASD haplotype carriers (Table S3) compared to others studied by WGS. Further examination of 5 ASD trios showed that two STR alleles segregated with the ASD-associated haplotype (Figure S2, 3 from mother and 2 from father) suggesting that these alleles are likely in the same haplotype. The functional significance of these alleles needs further investigation.

Within a 29 kb region of the second intron of ROBO2 containing ASD-associated SNPs we identified two potential regulatory elements: a CTCF-binding site (CBS) (Figure 1C) that pairs with another CTCF site near the ROBO2 promoter to form a chromatin loop²⁶, and a putative fetal human cardiac enhancer (hg19: chr3:77,320,232–77,321,301; hg38: chr3:77,271,081–77,272,150)²⁷ that includes the ASD-risk SNP, rs116127393. To verify the CBS and the putative enhancer, we surveyed chromatin immunoprecipitation sequencing (ChIPseq) data from hiPSC-CMs¹², which are developmentally immature and express ROBO2. We identified a CTCF peak overlapping with the CBS in hiPSCs (Figure 1D) and H3K4me1, H3K4me3, and H3K27ac peaks overlapping with the enhancer in mature hiPSC-CMs and fetal human heart tissue (Figure 1E, F).

Next, we compared assay for transposase-accessible chromatin with sequencing (ATAC-seq) results from hiPSC-CMs during differentiation^12,13 and DNase I hypersensitive sites sequencing (DNase-seq) data from fetal hearts.¹⁶ Two open chromatin peaks overlapped with the CBS and the putative enhancer (Figure 1G, H) in both hiPSC-CMs and fetal heart tissues, highlighting the potential regulatory role of the ASD locus. To identify transcription factors regulating the putative enhancer, we examined published ChIPseq data for cardiac transcription factors^28,29 and found that NKX2–5 binds the enhancer in both hiPSC-CMs and fetal mouse hearts (Figure S3). Notably, mutations in NKX2–5 are a known genetic etiology for ASD.³⁰

ROBO2 haplotype is implicated in familial CHD.

We previously reported³¹ two families (MAR and MBE) with inherited ASDs and other malformations due to unknown genetic variant(s). Using WGS and/or Illumina OMNI5 data we found that the ROBO2 haplotype was absent in all MAR relatives (not shown) but present in 10 MBE family members: five with ASDs, two with bicuspid aortic valve (BAV) alone or with aortic stenosis (AS), and three unaffected individuals (Figure S4). Two MBE relatives with patent ductus arteriosus (PDA) did not carry the ROBO2 haplotype. Given the allele frequency of the risk-haplotype of 0.03, the likelihood is low that either family would carry the ROBO2 haplotype (p=0.008; chi-square test). The odds are 1 in 50 (LOD score =1.7, θ=0) that five affected MBE individuals would randomly inherit the ROBO2 risk haplotype from an ancestor. While a pathogenic ROBO2 variant has been excluded in this family, we suggest that the risk haplotype contributes to prevalent ASDs in affected family members, perhaps through interactions with another variant.

ASD-associated ROBO2 locus regulates ROBO2 expression in hiPSC-CMs.

To gain further insight into the potential involvement of ROBO2 in ASD, we examined its expression in published single-nuclei RNA-seq data from fetal hearts at post-conception weeks 9, 11, 13, and 15³². ROBO2 was expressed within fibroblast, ventricular cardiomyocytes, and atrial cardiomyocytes at a low level (Figure S5). ASD risk SNPs did not overlap with known cardiovascular eQTLs³³, possibly due to predominant fetal expression of ROBO2. As we had no cardiac tissue that expressed detectable levels of ROBO2 from ASD subjects with risk alleles, we employed hiPSC-CMs for further studies.

To determine if the identified regulatory elements within the ASD locus are involved in regulating ROBO2 expression, we employed CRISPR-Cas9 guides to target a 25kb region encompassing 13 of the 15 ASD-risk SNPs, the CBS, and the putative enhancer (Figure 2A). We obtained cell lines with three different genotypes with mutations spanning both the CBS and enhancer: heterozygous deletion (n=2; designated 25 kb Δ/+), homozygous deletion (n=2; designated 25 kb Δ/Δ), and compound heterozygous lines carrying inversion of the 25kb sequences and a 14bp deletion on the CBS (n=3; designated 25kb inv or 25 kb inv/14bp Δ). (Figure S6A-D) Additionally, we targeted the CBS and nearby genomic sequences to assess the roles of CTCF and obtained four clones designated as 9kb Δ/+, 9kb Δ/Δ, CTCF Δ/+, and CTCF Δ/Δ (Figure 2A, Figure S6E-I). All hiPSC lines differentiated and matured into beating cardiomyocytes. We also verified that the cells have normal karyotypes and can differentiate into three germ layers. (Figure S6J, K)

Figure 2. Mutations in the ASD-associated locus significantly reduce ROBO2 expression hiPSC-derived cardiomyocytes.

(A) Schematic of the ASD locus genotypes. Mutants were generated using CRISPR/Cas9. Risk SNPs (SNP1-SNP15) are represented as circles (red: GWAS SNPs and nearby SNPS in LD; grey, SNPs in LD with GWAS SNPs with higher MAF in subjects with non-EU ancestry). The dark blue triangle indicates the CTCF binding site, and the blue box indicates the putative enhancer. The black vertical line indicates the start or end of the mutation, while the black dashed line represents the deleted region. The thick grey line indicates the intron. The red arrow represents the inverted haplotype. (B) Comparison of ROBO2 expression, normalized to GAPDH expression, assessed by qPCR at day 8 post hiPSC differentiation. The number of replicates analyzed is denoted as N. Fold change was analyzed by ddCt method. P-values were calculated with a t-test of dCt values. **** P < 0.0001, *** P < 0.001, ** P < 0.01, * P < 0.05. (C) Normalized ROBO2 expression of the mutants from RNA-seq. P values were calculated by DESeq2 and corrected for multiple testing with the Benjamini-Hochberg method. (D) Normalized expression of 338 common DEGs in the mutants. Known CHD genes are highlighted. (E) Gene Ontology analysis results of the 338 common DEGs. P values were adjusted with the Bonferroni method. Gene count indicates the number of the tested genes that fall into the given GO term.

On differentiation day 8 (D8), when hiPSC-CMs expressed cardiac transcripts, ROBO2 expression was heightened (Figure S7A). We measured ROBO2 expression in D8 WT and mutant hiPSC-CMs with qPCR. Homozygous deletion of the CBS or combined deletion of the CBS and enhancer significantly lowered ROBO2 expression, whereas heterozygous deletion of the CBS alone did not affect ROBO2 expression (WT vs. CTCF Δ/+; P=0.0825, vs. CTCF Δ/Δ; P=0.00886, vs. 9kb Δ/+; P=0.244, vs. 9kb Δ/Δ; P=0.0424, vs. 25kb Δ/+; P=9.48E-6, vs. 25kb Δ/Δ; P=1.64E-10, vs. 25kb inv; P=1.43E-07, Figure 2B). Inversion of the ASD locus with deletion of the CBS also decreased ROBO2 expression, indicating that proper looping between the ROBO2 promoter and the enhancer is necessary for ROBO2 expression. To address the high variance in ROBO2 expression observed in the 9kb mutants, we quantified ROBO2 expression in 9kb Δ/+ and 9kb Δ/Δ using digital droplet PCR (ddPCR), which confirmed the initial results (Figure S7B).

Lower expression level of ROBO2 leads to mis-regulation of extracellular matrix genes

We then conducted RNAseq on D8 hiPSC-CMs from WT, 9kb Δ/Δ, 25kb Δ/+, 25kb Δ/Δ, and 25kb inv (N=5 for WT, N=3 for 9kb Δ/Δ, N=2 for 25kb Δ/+, 25kb Δ/Δ, and 25kb inv; Figure S7C) to gain further insights into the role of ROBO2 in immature hiPSC-CMs. Consistent with our earlier results, all four mutants exhibited significantly reduced ROBO2 transcript levels (Figure 2C). No consistent changes were observed in the expression of other ROBO or SLIT genes. (Figure S7D) All mutant lines expressed normally spliced ROBO2 transcripts (Figure S7G).

In total, 338 genes were commonly up- or down-regulated across all four mutants (Figure S7E, Figure 2D). Gene ontology analysis revealed enrichment in cell division and extracellular matrix (ECM) genes among the 338 genes (Figure 2E). Reflecting these alterations in ECM gene expression, at differentiation day 30, the 25 kb Δ/Δ lines were morphologically distinct (Figure S7F). Unlike the isogenic wildtype, 25 kb Δ/+, and 25 kb inv lines, cardiomyocytes derived from both homozygous deletion iPSC lines had irregular, elongated inter-cellular connections that outlined acellular spaces between cardiomyocyte clusters. Several CHD genes³⁴ such as SALL1, CFC1, and SOX9 were also differentially expressed across all mutants. Notably, SOX9 is a regulator of ECM composition and is essential for cardiac development in zebrafish³⁵ and mice³⁶. Mutations in SOX9 have been associated with various CHDs, including septal defects³⁷. While ROBO2 is known to regulate the SOX2/SOX9 balance in fetal rat lung development,³⁸ it is unclear if ROBO2 similarly regulates SOX9 in other cell types including cardiomyocytes.

ASD locus sequences can act as weak enhancers.

To test the functional activities of each ASD-associated SNP and the ASD locus, we employed lentivirus enhancer reporter assay³⁴. We cloned 870–1443 bp sequences carrying reference or alternative alleles upstream of the minimal promoter and GFP (Table S5, Figure S8A). We then delivered plasmids using lentivirus and assessed mean GFP fluorescence intensity (MFI) in hiPSC-CMs (N=4 per construct). We observed weak but increased GFP MFI in the constructs carrying sequences flanking SNP2, 5, and 9. The construct flanking SNP9 encompasses the putative fetal cardiac enhancer sequences, and the GFP MFI from this construct significantly decreased with the alternative allele for SNP9 (Figure S8B). This suggests that the ASD locus may encompass a weak enhancer in hiPSC-CMs, and the ASD-associated alleles may reduce the enhancer activity.

ASD-associated ROBO2 locus physically interacts with the ROBO2 promoter.

To better understand how the ASD-associated locus regulates ROBO2 transcription, we examined whether the CTCF plays a role in promoter-enhancer looping with 4Cseq and Hi-C. We found that the ASD locus is in contact with the ROBO2 promoter, located 250kb away, in both undifferentiated hiPSCs and D8 hiPSC-CMs (Figure 3A, B)³⁹.

Figure 3. The physical interaction between the ASD-associated locus and the ROBO2 promoter in hiPSCs and hiPSC-CMs.

4Cseq and Hi-Cseq of (A) D0 WT hiPSCs (B) D8 WT hiPSC-CMs (C) D8 9kb Δ/Δ hiPSC-CMs (D) D8 25kb inv/14bp Δ hiPSC-CMs. The green boxes and arrows represent ROBO2 exons and introns, respectively. A grey box indicates the ASD-associated locus. Vertical green boxes indicate the CTCF ChIP peaks from WT hiPSCs. The results from 4Cseq with the ROBO2 promoter as the viewpoint are shown. The peak height from 4Cseq corresponds to the read counts spanning the given base pair. The box depicts the Hi-C results. The intensity of red reflects with observed reads. All reads were normalized with Knight-Ruiz (KR) algorithm⁴⁰. Black boxes show a 10kb window that has significant interaction with the ROBO2 promoter. The deleted region and the inverted region are highlighted with yellow boxes. Figures were adapted from Juicebox⁴¹.

We compared the results with the previous Hi-C data⁴² and observed that this interaction persists in more mature hiPSC-CMs and fetal heart tissue as well (Figure S9). The interaction between the promoter and the ASD locus diminished in 9kb Δ/Δ and 25kb inv/14bp Δ mutants, while the interaction between the promoter and the CTCF binding site downstream of the ASD locus increased (Figure 3C, D) in the mutants. With ATACseq, we observed that the open chromatin peak overlapping with the CTCF binding site disappeared in both 9kb Δ/Δ and 25kb inv/14bp Δ mutants (Figure S10), further suggesting that the CTCF binding in this locus is disrupted by mutations. These findings indicate that the ASD locus physically interacts with the ROBO2 promoter and potentially regulates the ROBO2 promoter through promoter-enhancer looping throughout cardiac development. The CTCF binding site within the ASD locus is necessary for this interaction, and the disruption of the looping leads to a decrease in ROBO2 transcription.

Discussion

Our studies define a common haplotype found in ~18% of ASD patients that is associated with a 2.8-fold risk for ASDs, likely by altering an ROBO2 expression. The ROBO2 ASD locus was not identified in two previous CHD GWAS.^9,10 Differences in phenotypes, genotyping arrays, study cohort designs, and small sample sizes may account for our novel findings. Risk alleles span a 29-kb fragment in intron 2 of the ROBO2 gene that contains orientation-specific regulatory elements, including a CTCF binding sequence and enhancer. We identify novel regulatory elements that are likely regulating ROBO2 transcription in fetal cardiomyocytes. Mutation of these elements in cardiomyocytes reduces and alters the expression of ROBO2 alleles. Consistent with the functions of Robo-Slit signaling in cell adhesion and migration, we observed mis-regulation of various extracellular matrix and cell adhesion-related genes in the mutants. Together with experimental model systems, these data help elucidate the role of ROBO2 expression in the development of the human heart. In addition, our findings provide mechanistic insights into how common variants could increase risk for ASDs.

Precise orchestration of the migration by cardiovascular progenitor cells is essential for cardiac morphogenesis. Early committed progenitor cells exit the anterior lateral plate mesoderm and migrate toward the midline cardiogenic area to form a linear heart tube. After the heart tube loops, intracardiac cell migration partitions the common atria and ventricle and forms cardiac valves, while migratory cardiac neural crest cells remodel the outflow tract, dividing the aorto-pulmonary trunk and shaping the pericardium. In zebrafish and mice, Robo-Slit signals play critical roles in controlling cardiac cell polarity ¹⁹, guiding the adhesion or migration of cardiovascular precursor cells throughout these developmental processes ^24,43, and demonstrate diverse cardiovascular malformations when Robo receptors and/or Slit ligands are disrupted or mis-expressed ^18,23,24,44. Consistent with these critical embryonic functions, the expression of Robo-Slit genes is precisely regulated, not only spatio-temporally by cardiac transcription factors (e.g., Nkx2.5, Tbx1, Chd7) but also within specific regions of polarized primordial cardiovascular cells.⁴⁵

Our genetic and functional analyses demonstrate the importance of cis-acting ROBO2 sequences in the critical temporal-spatial expression of the Robo-slit signaling pathway during heart development. The ROBO2 locus contains two CTCF binding motifs, oriented in convergent directions within the promoter and second intron of ROBO2 that are positioned to anchor a chromatin loop that brings the fetal heart enhancer in proximity to the ROBO2 promoter (Figure 3A, B). Loss of the CTCF binding motif and the enhancer from both ROBO2 genes would be expected to result in marked transcription reduction, as observed in the 25kb Δ/Δ hiPSC-CMs (Figures 2B, C). By contrast, inverting a fragment containing these regulatory sequences (25kb inv) could disrupt CTCF binding motif interactions and loop formation and further diminish the enhancer activity on the ROBO2 promoter (Figure 2B, C and Figure 3D).

Analysis of RNAseq of immature hiPSC-CMs revealed reduced expression of several transcription factors, such as SOX9 and SALL1. Loss of function mutations in SALL1 cause Townes-Brocks syndrome, which is characterized by malformations in renal, ear, anal, and limb.⁴⁶ Homozygous truncating mutations in SALL1 caused tetralogy of Fallot and VSD.⁴⁶ During kidney development, knock out of Sall1 diminished Robo2 expression^47,48, and Robo2 is a direct target of Sall1.⁴⁸ Our results suggest that ROBO2 may directly or indirectly regulate SOX9 and SALL1 in other cell types. How ROBO2 expression modulates SOX9 and SALL1 in D8 hiPSC-CMs needs further investigation.

Our studies indicate low penetrance of the ASD-associated haplotype at the population level. Because our association analyses did not adjust for covariates that may differ between cases and controls, residual confounding cannot be excluded and should be considered when interpreting the estimated effect size. Based on a haplotype population prevalence of 6% (Table S1) and the overall prevalence of ASDs (2 per 1000 live births) we estimate penetrance of ~3% for this ASD locus. However, as 50% of haplotype carriers in family MBE exhibit ASDs (and 70% with CHD), ROBO2 expression might have greater influence in the context of other ASD and/or CHD loci, potentially contributing to the diversity of heart malformations.

The functional mechanisms of the ASD SNPs and its role in fetal heart development require further investigation. Our enhancer reporter assay suggests that this ASD locus may contain sequences acting as transcriptional enhancers, and that ASD risk SNPs modulate this enhancer activity (Figure S7). Further characterization of the fetal cardiac enhancer and how single nucleotide polymorphisms within this locus modulate enhancer function in vitro and in vivo could elucidate the functional mechanisms of these variants. The locus also contains eleven simple tandem repeats, with more in the vicinity. Some simple tandem repeats are known to alter gene expression in cis ⁴⁹, and whether the identified repeat alleles can modulate ROBO2 expression needs further investigation. As ROBO2 is expressed in multiple cell types within the fetal heart.³² D8 hiPSC-CMs, prior to glucose selection, could consist of immature cardiomyocytes and other cell types such as fibroblasts and endothelial cells. Whether SNPs associated with ASD influences ROBO2 expression in these cell types or their interaction with cardiomyocytes needs clarification. Further exploration of the genetic architecture of CHD holds considerable potential to answer these questions and provide new insights into the developmental regulation of gene transcription.

Supplementary Material

Contributing Members of the PCGC Investigators

Supplemental Methods

Tables S1–S7

Figure S1–10

Files S1-S2 (Separate Excel)

References 50–74

Nonstandard Abbreviations and Acronyms:

ASD

Atrial Septal Defect

CHD

Congenital Heart Disease

AVSD

atrioventricular septal defect (AVSD)

VSD

ventricular septal defects

ToF

tetralogy of Fallot

STR

simple tandem repeat

CBS

CTCF-binding site

BAV

bicuspid aortic valve

AS

aortic stenosis

PDA

patent ductus arteriosus

ddPCR

digital droplet PCR

MFI

mean fluorescence intensity