A systems genetics approach identifies roles for proteasome factors in heart development and congenital heart defects

Abstract

Congenital heart defects (CHDs) occur in about 1% of live births and are the leading cause of infant death due to birth defects. While there have been remarkable efforts to pursue large-scale whole-exome and genome sequencing studies on CHD patient cohorts, it is estimated that these approaches have thus far accounted for only about 50% of the genetic contribution to CHDs. We sought to take a new approach to identify genetic causes of CHDs. By combining analyses of genes that are under strong selective constraint along with published embryonic heart transcriptomes, we identified over 200 new candidate genes for CHDs. We utilized protein-protein interaction (PPI) network analysis to identify a functionally-related subnetwork consisting of known CHD genes as well as genes encoding proteasome factors, in particular POMP, PSMA6, PSMA7, PSMD3, and PSMD6. We used CRISPR targeting in zebrafish embryos to preliminarily identify roles for the PPI subnetwork genes in heart development. We then used CRISPR to create new mutant zebrafish strains for two of the proteasome genes in the subnetwork: pomp and psmd6. We show that loss of proteasome gene functions leads to defects in zebrafish heart development, including dysmorphic hearts, myocardial cell blebbing, and reduced outflow tracts. We also identified deficits in cardiac function in pomp and psmd6 mutants. These heart defects resemble those seen in zebrafish mutants for known CHD genes and other critical heart development genes. Our study provides a novel systems genetics approach to further our understanding of the genetic causes of human CHDs.

Author summary

Congenital heart defects (CHDs) are birth defects resulting in the abnormal structure and function of the heart. Genetic mutations are a significant cause of CHDs. Many studies have used genome sequencing of CHD patients and their families to gain knowledge of the mutations that cause CHDs. However, these studies have only found about 50 percent of the genes involved in CHDs. Here, we take a new approach to identifying genes that are required for heart development and that may cause CHDs, generating a list of over 200 candidate genes. Using multiple data systems, including human exome sequences, mouse transcriptomes, and protein-protein interactions, we identify a small group of related potential CHD genes that includes multiple genes encoding proteasome factors. These factors are known to be important for assembling the proteasome, a large molecular machine that eliminates unneeded or damaged proteins from the cell, but which has not been shown to contribute to CHD. We use a CRISPR-based approach in zebrafish to specifically eliminate some of these candidate genes and reveal new roles for proteasome genes in heart development. We show that loss of proteasome gene functions leads to zebrafish heart defects that resemble those seen in other zebrafish CHD-gene mutants. This study shows that a proteasome gene family contributes to heart development, advancing our understanding of the causes of CHDs. By increasing our understanding of the genetic causes of CHDs, our work should lead to better screening, more accurate diagnoses, and, ultimately, better treatments for these disorders.

Introduction

Congenital heart defects (CHDs) occur in about 1% of live births and are the leading cause of infant death due to birth defects [1–4]. CHDs are structural malformations of the heart that result from disruptions in cardiac development. There is a broad spectrum of CHDs, ranging from those affecting a particular valve or chamber to more severe and complex abnormalities involving multiple heart chambers and vessels [5]. CHDs may occur as isolated malformations or in combination with extracardiac defects [5–7]. Although environmental causes can contribute to CHDs, numerous studies point to a strong genetic component and high heritability for many forms of CHDs [8–12]. Identifying the genetic causes of CHD can not only provide better understanding of heart development but also can inform genetic counseling and clinical care.

Several types of genetic alterations have been shown to contribute to CHDs, including chromosomal aneuploidies, copy number variants, small insertions or deletions, as well as de novo and inherited single-nucleotide variants [7,12,13]. Large-scale whole-exome sequencing studies find that CHD cases show an excess of damaging coding de novo variants (DNVs) as well as rare inherited loss-of-function coding mutations [14–19]. These studies estimate that about 10% of CHD cases may be caused by coding DNVs, while whole genome sequencing studies estimate that noncoding DNVs may also confer a substantial contribution to CHDs [20,21]. Taken together, it is estimated that these CHD patient cohort sequencing studies have thus far accounted for about 50% of the genetic contribution to CHDs [12].

Many approaches have been taken to define CHD-causing genes [13]. One study has defined high-confidence CHD genes as genes in which variants have been reported as the monogenic cause of CHD in at least 3 independent familial or sporadic cases, in one or more separate publications (132 genes at the time of this study; [22]). In this study, we refer to these 132 genes as “known” CHD genes. However, it is estimated that there are over 440 risk genes for CHDs [18]. Therefore, a major hurdle that remains for understanding the causes of CHDs is the identification and validation of human CHD genes that are as yet unknown.

As a complementary approach to sequencing patient cohorts for human disease-gene discovery, larger-scale human population genome data has been analyzed for essential genes [23–26]. Such studies find that genes that are essential for mammalian embryonic development are strongly correlated with human disease genes, especially for developmental disorders [25]. An analysis of the ExAC human exome sequencing database identified a set of genes for which heterozygosity for nonsense mutations is rare or absent in the normal adult population [23]. These genes are thus predicted to be haploinsufficient in humans and are strong candidates for contributing to birth defects. From this analysis, Cassa et al. assigned a heterozygote selection value (s_het score) to each human gene in the genome [23]. The heterozygote selection (s_het) statistic is calculated from the measured frequency of null mutations in genes in a normal human adult population, relative to expectation. The s_het score correlates with known human autosomal dominant disorders, indicating the relevance of genes with high heterozygote selection values to human disease [23]. Many genes known to be associated with congenital disorders, including cardiac defects, have high s_het values [23]. The s_het score also correlates with developmental lethality in homozygous mouse knockout strains generated by the International Mouse Phenotyping Consortium (IMPC) [23,27]. Thus, these analyses implicate high s_het genes in mammalian development and disease.

As a complement to mouse models, zebrafish offer many advantages for screening and characterizing new CHD genes. Zebrafish provide the ability to examine the earliest stages of heart development in live externally-developing embryos, and many studies have documented that genes needed for proper heart development in zebrafish also contribute to CHDs in humans [28–30]. Another major advantage of the zebrafish model is the availability of efficient CRISPR screening and mutagenesis approaches, which have shown success for candidate CHD gene discovery [29,31,32].

Here we take a novel approach to identify candidate human CHD genes. Using human exome sequence data, mouse transcriptome data, and protein-protein interaction network analyses, we identify a subnetwork of potential CHD genes that includes multiple proteasome factor genes. We use CRISPR screening in zebrafish to identify roles for these proteosome factor genes in zebrafish heart development. Finally, we generate stable zebrafish mutant lines for two of the identified proteasome genes and use them to demonstrate novel functions for proteosome genes in heart development.

Results

Identification of 245 candidate CHD genes

To take a new approach to identifying candidate CHD genes, we turned to the high s_het genes, which are associated with Mendelian developmental disorders as well as cellular and embryonic viability [23]. With specific respect to heart development, we find that genes with damaging de novo variants in patients with CHDs are strongly correlated with high s_het values, whereas no correlation is observed between s_het values and genes with damaging de novo variants in control subjects (Fig 1A). In addition, most of the “known” CHD genes (132 genes, defined by Yang et al.; S1 Table; [22]), are found in the top s_het deciles (Fig 1B), further implicating high s_het genes in CHDs.

Fig 1. Identifying new candidate CHD genes.

A. Correlation of high s_het genes with genes exhibiting damaging de novo variants in CHD cases (black bars; P = 1.09E-19). No correlation is observed between s_het deciles and damaging de novo variants observed in controls (grey bars; P = 0.7). s_het deciles are from Cassa et al.’s Supplementary Table 1 [23]. de novo variant data are from Jin et al.’s Supplementary Data Set 9 (CHD cases) and Supplementary Data Set 10 (controls) [18]. B. Correlation of high s_het genes with known CHD genes, defined by [22] (S1 Table). s_het deciles are from [23]. C. Venn diagram showing overlap of genes with high s_het values in pink (3190 genes; S2 Table; top 2 deciles in dataset from [23]) and genes expressed in the mouse cardiac muscle cell lineage in green (1119 genes; S3 Table; datasets from [33] and [34]). The intersection is 245 candidate genes for CHDs (S4 Table). D. Venn diagram showing overlap of our Candidate CHD gene set in maroon (245 genes; S4 Table) and known CHD genes in purple (132 genes; S1 Table; [22]). The intersection is 11 genes, shown in the box. E. Gene Ontology (GO) term enrichment for candidate gene set, obtained using DAVID. The number of genes (out of 245) represented by each term is shown on top of each bar. Bonferroni-adjusted P values are shown.

We selected the genes with the highest s_het values (in the top two deciles, 3190 genes; S2 Table; [23]). Over half of known CHD genes are represented within these highest s_het gene cohorts (Fig 1B). We asked which of the 3190 high s_het genes are likely to play a role in heart development by identifying those that are also expressed in the cardiac muscle cell lineage during embryonic development. For this analysis, we used cardiac gene expression datasets from two studies: the cardiac muscle cell lineage, obtained from single-cell RNA-seq of whole mouse embryos (cell cluster 34, stages E9.5-E13.5; [33]), and the differentially-expressed gene sets from single-cell RNA-seq of Nkx2–5-positive and Isl1-positive cells from E7.5-E9.5 mouse embryos [34]. We chose these datasets because they encompass early stages of mammalian heart development, they are comprised of early myocardial cell types, and genes with high levels of embryonic heart expression are enriched for mutations in CHD patients in exome studies [14,15,18,19]. These combined mouse gene expression datasets provide a cardiac muscle cell lineage gene set (1119 genes; S3 Table). The intersection of the early cardiac muscle-expression genes with the high s_het genes identified 245 genes that we termed our candidate CHD genes (Fig 1C; S4 Table).

Of our 245 candidate genes, 11 overlap with the set of 132 known human CHD genes (Fig 1D). Among these 11 genes are well-characterized genes with respect to mammalian heart development and CHDs, including GATA4, NKX2–5, and TBX5 [35,36]. Thus, the majority of candidate genes identified through our approach (234/245 genes) are not considered “known” CHD genes, suggesting that our approach has identified new CHD genes. In support of our candidate genes being associated with CHDs, we find that 44 of our 245 genes have had potential damaging mutations identified in CHD cases [18] (S5 Table). We next employed a Gene Ontology (GO)-term analysis of our candidate gene list. Many of the top Biological Process GO terms enriched in our candidate gene list include general terms such as positive regulation of cellular and metabolic processes (Fig 1E). We also find enrichment for several cardiac development-related terms, and these terms are represented by more than the 11 known CHD genes in our candidate gene set (Fig 1D - 1E). Furthermore, using published transcriptome data [37], we determined that 154 of our 245 genes are expressed in human embryonic heart tissues (S6 Table). Thus, the intersection of high s_het genes and cardiac cell-lineage gene expression identifies known as well as potential new players in heart development and CHDs, without reference to their known phenotype.

Protein-protein interaction networks reveal potential new CHD gene modules

To further explore the relationships between our 245 candidate CHD genes and known CHD genes, we turned to the STRING online database [38] to generate a Protein-Protein Interaction (PPI) network. We created a large PPI network (Fig 2A) consisting of 366 proteins that correspond to our 245 candidate genes (S4 Table) and 132 known CHD genes (S1 Table; [22]), including the 11 overlapping genes from Fig 1D. This large network shows that proteins encoded by the candidate CHD genes share many functional and/or physical associations with known CHD proteins.

Fig 2. PPI networks reveal connections between known and candidate CHD genes.

A. A Protein-Protein Interaction (PPI) network showing the evidence-based interactions among the proteins encoded by our 245 candidate CHD genes (yellow halos; S4 Table), 132 known CHD genes (green halos; S1 Table), and candidate genes that are also known CHD genes (blue halos, 11 genes from Fig 1D). Interactions between proteins/nodes are indicated by a single grey line, whose thickness corresponds to the strength of the supporting data. This network consists of 365 proteins/nodes, contains 980 edges, and is associated with a PPI enrichment P value < 1.0E-16. Purple nodes (Proteasome cluster genes) indicate the 8 factors (NF1, ODC1, POMP, PSMA6, PSMA7, PSMD3, PSMD6, and UBQLN2;) identified in the “Regulation of ornithine decarboxylase, and proteasome assembly” local cluster analysis (S7 Table). B. PPI subnetwork showing the interactions among the Proteasome cluster genes (purple nodes) and their direct connections from the larger PPI network in A. The node color code scheme is the same as in A. The lines between nodes are color coded to indicate the type of evidence supporting the protein-protein interaction. This subnetwork consists of 23 nodes, contains 58 edges, and has a PPI enrichment P value < 1.0E-16.

To further assess the interactions embedded within our large network, we utilized STRING’s local network cluster analysis feature for functional enrichment. STRING identified 26 clusters, or functionally-enriched subnetworks (S7 Table). Most of these clusters consist mainly of proteins encoded by known CHD genes (S7 Table). The cluster with the highest percentage and number of candidate, but unproven, CHD genes is the “Regulation of ornithine decarboxylase, and proteasome assembly” cluster, or Proteasome cluster (S7 Table). The eight network genes in this cluster are NF1, ODC1, POMP, PSMA6, PSMA7, PSMD3, PSMD6, and UBQLN2, and these are highlighted in purple in the PPI network (Fig 2A and S7 Table). NF1 is a known CHD gene, while the remaining seven genes are candidate CHD genes. POMP, PSMA6, PSMA7, PSMD3, PSMD6, and UBQLN2 all encode subunits of the proteasome complex or proteasome-interacting factors [39], and these factors have not previously been shown to cause CHDs.

Because the Proteasome cluster, out of our 26 subnetworks, contains the most candidate genes with unknown roles in CHDs, we decided to focus our efforts on this cluster. To more closely examine the relationships of the Proteasome cluster genes with other known CHD genes, we used STRING to generate a new subnetwork consisting of the 8 Proteasome cluster factors along with their direct connections from within our larger PPI network. The resulting subnetwork consists of 23 proteins encoded by 11 known CHD genes (green halos in Fig 2B) and 13 candidate CHD genes (yellow halos in Fig 2B), with 1 overlapping gene (GLI3; blue halo in Fig 2B). This subnetwork illustrates the close connections of the proteasome factors PSMA7, PSMA6, PSMD6, and PSMD3 with known CHD factors, in particular NF1 and GLI3 (Fig 2B). NF1 and GLI3 are both linked with multiple forms of CHDs, including atrial and ventricular septal defects, which are common forms of CHDs that are also associated with CHD genes such as NKX2–5 and TBX5 [13,22]. Together, these findings support potential roles for our candidate CHD genes, in particular the Proteasome cluster genes, in heart development and CHDs.

G0 CRISPR targeting in zebrafish embryos reveals roles for Proteasome cluster genes in heart development

We next wanted to test for roles for the Proteasome cluster genes in heart development by using G0 CRISPR targeting in zebrafish embryos. Six of these eight human genes, NF1, ODC1, POMP, PSMA6, PSMD3, and PSMD6, have orthologs in the zebrafish genome, with NF1 and PSMA6 having duplicate zebrafish orthologs: nf1a and nf1b, and psma6a and psma6b [40]. To test the functions of zebrafish nf1a, nf1b, odc1, pomp, psma6a, psma6b, psmd3, and psmd6, we turned to an efficient G0 CRISPR targeting approach, using a pool of four CRISPR guide RNAs per gene [31]. As a positive control for using this approach to detect heart defects, we generated G0 CRISPR embryos for the zebrafish hand2 gene, using a pool of four hand2 guide RNAs. HAND2 is a known CHD gene [41,42] (S1 Table), and zebrafish hand2 mutants show severe defects in early myocardial precursor cell migration, leading to two separate myocardial domains, or cardia bifida [43]. Zebrafish hand2 G0 CRISPR embryos show severe cardia bifida, observed using expression of the pan-myocardial transgene myl7:EGFP (Fig 3A - 3B). Almost 100% of hand2 G0 CRISPR embryos show severe bifida at 2 days post fertilization (dpf) (Fig 3C), thus closely resembling null hand2^-/- mutant embryos [43]. These findings support the efficacy of the four-guide approach in attaining gene knockout and high frequency G0 CRISPR heart defects.

Fig 3. G0 CRISPR targeting of Proteasome cluster genes leads to heart defects in zebrafish embryos and larvae.

A.-C. hand2 G0 CRISPR leads to cardia bifida. A. 2 dpf control embryo showing myl7:EGFP heart. V, ventricle. A, atrium. Scale bar = 200 μm. B. 2 dpf hand2 G0 CRISPR embryo showing two domains (cardia bifida) of myl7:EGFP (arrows). C. Graph showing frequencies of different heart defect categories observed in G0 CRISPR embryos. Cardia bifida and heart looping defects were scored at 2 dpf. Hearts were again scored at 4 dpf for defects arising between 2 dpf and 4 dpf and were divided into two classes: lateral chamber arrangement and elongated chamber arrangement. Examples of the elongated chamber arrangement are illustrated in panels Q, S, T and V, and an example of the lateral chamber arrangement is shown in panel U. D. Graph showing frequencies of embryos with heart defects (at 4 dpf; all categories combined). Each dot represents an experimental replicate batch of embryos. Bars represent mean +/- SEM. N = 4-6 replicates per gene of interest; N = 24 control replicates. Each replicate batch consists of 7-75 embryos (mean = 36±16). The total number of embryos scored is given above the x-axis. These total numbers also apply to the corresponding columns in Panel C. ** P < 0.01. **** P < 0.0001. E.-O. Lateral views of 4 dpf larvae, showing representative G0 CRISPR phenotypes. The bubble (asterisk in E) is the swim bladder, an indicator of healthy larvae [40]. The arrowhead in I points to severe heart cavity swelling/edema. Scale bar = 500 μm. P.-V. Ventral views of hearts labeled with myl7:EGFP in 4 dpf larvae with pigment inhibited. Representative G0 CRISPR heart phenotypes are shown for each gene. Scale bar = 200 μm. V, ventricle. A, atrium.

We then used this G0 CRISPR approach to target the Proteasome cluster genes. At 2 dpf, the hearts were scored for cardia bifida and other heart defects, and at 4 dpf they were scored again to identify defects that had arisen since 2 dpf. After targeting the Proteasome cluster genes, only rare cases of cardia bifida were observed at 2 dpf (Fig 3C), but we did observe other forms of heart defects from 2-4 dpf. Of the eight Proteasome cluster genes, G0 CRISPR targeting of odc1 led to the most severe heart defects, with severe edema, or swelling around the heart cavity, and hearts that are reduced and elongated by 2 dpf (Fig 3C, 3D, 3I, 3R). G0 CRISPR targeting of individual nf1a, nf1b, psma6a, or psma6b genes did not lead to a high frequency of heart defects (Fig 3C - 3D). However, G0 CRISPR targeting of nf1a+nf1b, pomp, psma6a+psma6b, psmd3, and psmd6 all led to a high frequency of defects in heart development, arising between 2 dpf and 4 dpf (Fig 3C-3D). These larvae show edema and hearts in which the chambers appear malformed and not properly looped (Fig 3E - 3V). We classified the 4 dpf heart defects that we observed for these Proteasome cluster genes as “lateral” or “elongated” (Fig 3C). In control 4 dpf larvae in ventral view, the atrium appears behind (dorsal to) and to the right (anatomical left) of the ventricle (Fig 3P). In the case of the “lateral” looping defect, the atrium is displaced ventrally so that the two chambers are in nearly the same dorsal-ventral plane and appear side-by-side (as for psmd3 in Fig 3U). In the case of the “elongated” looping defect, the ventricle is displaced anteriorly in addition to the atrium being displaced ventrally (as for nf1a+nf1b in Fig 3Q). The high frequency of similar heart defects in these G0 CRISPR embryos suggest that these Proteasome cluster genes all play similar roles in heart development.

pomp and psmd6 mutant zebrafish embryos exhibit heart defects and other phenotypes

To further examine the roles for these proteasome factors in heart development, we generated stable mutant strains for pomp and psmd6. We chose these two genes because one encodes a 19S proteasome subunit (psmd6), one encodes a chaperone needed for 20S proteasome assembly (pomp), and both genes have had de novo missense variants identified in single CHD cases [18,39]. In addition, while pomp and psmd6 are both expressed broadly in zebrafish embryos [40], single-cell RNA analyses have identified pomp and psmd6 expression in zebrafish embryo heart cells starting at 14–21 hours post fertilization and in early fetal human heart cardiomyocytes [44,45]. For both genes, we used CRISPR to generate alleles that delete the transcription start site (TSS) and 5’end of each gene (Fig 4A - 4B). These TSS deletion alleles are pomp^scm41 and psmd6^scm40, hereafter referred to as pomp and psmd6 mutants.

Fig 4. pomp and psmd6 mutant zebrafish larvae show cardiac and extracardiac phenotypes.

A. Schematic illustrating pomp gene structure and CRISPR deletion. pomp exons are shown as boxes and the start codon is labeled. Guide RNA target sites are labeled in green and orange. DNA sequencing of the pomp^scm41 strain showed the pomp gene sequence between the two guide RNA sites has been deleted. B. Schematic illustrating psmd6 gene structure and CRISPR deletion. psmd6 exons are shown as boxes and the start codon is labeled. Guide RNA target sites are labeled in blue and red. DNA sequencing of the psmd6^scm40 strain showed the psmd6 gene sequence between the two guide RNA sites has been deleted. C.-D. qRT-PCR analysis of (C) pomp and (D) psmd6 expression in 4 dpf phenotyped pomp and psmd6 control (presumed +/+ and +/-) or mutant (mut, presumed -/-) larvae. Each dot represents a replicate batch of embryos. Bars represent mean +/- SEM. N = 5 (control) or 4 (mutant) replicates. Each replicate consists of 10-20 embryos. **** P < 0.0001. E. Western analysis using anti-Ubiquitin and anti-beta Tubulin (as a loading control). Larvae from pomp and psmd6 clutches were phenotyped as control (presumed +/+ and +/-) or mutant (presumed -/-) larvae at 4 dpf and collected for lysates. Lysates were prepared from 3 replicate pools of animals for each condition, with n = 20 animals per replicate. Larvae from bortezomib or DMSO control treatments were collected at 4 dpf, with n = 20 animals per lysate. Molecular weight markers are shown. F.-K. Images of live 4 dpf larvae. The asterisk in F marks the swim bladder. Arrowheads in H, K point to heart cavity swelling/edema. Black arrows in H, K point to reduced craniofacial structures. White arrows in H, K point to areas of blood pooling. Embryos were genotyped. N = 3 pomp^+/+, 12 pomp^+/-, 9 pomp^-/-, 4 psmd6^+/+, 8 psmd6^+/-, 6 psmd6^-/-. Scale bars = 500 μm. L.-Q. Ventral views of hearts labeled with myl7:EGFP in live 4 dpf larvae with pigment inhibited. R.-S. Ventral views of hearts labeled with myl7:EGFP in formaldehyde-fixed 4 dpf larvae with pigment bleached. Scale bars = 100 μm. V, ventricle. A, atrium. T. Graph showing frequencies of heart defect categories observed in 4 dpf larvae. The total number of larvae scored is given at the base of the columns. Embryos were genotyped. U. Survival plot of two independent clutches each of pomp and psmd6. Dead embryos were collected and fixed over the course of the experiment; all embryos were genotyped at the conclusion of the experiment. Control (+/+ and +/-) curves are shown for only one clutch of each line; similar results were seen with the second clutch of controls for each line. Numbers of animals per clutch: pomp clutch 1: 40 +/+ , 100 +/-, 52 -/-; pomp clutch 2: 44 +/+ , 93 +/-, 61 -/-; psmd6 clutch 1: 60 +/+ , 107 +/-, 55 -/-; psmd6 clutch 2: 27 +/+ , 61 +/-, 22 -/-.

To confirm that these deletion alleles cause loss of expression of their respective genes, we used qRT-PCR on whole larvae at 4 dpf to examine pomp and psmd6 expression. As expected, pomp expression is lost in pomp mutant larvae, and psmd6 expression is lost in psmd6 mutant larvae (Fig 4C - 4D). We observe that pomp expression is upregulated in psmd6 mutant larvae, and psmd6 is upregulated in pomp mutant larvae (Fig 4C - 4D), indicating a compensation effect on gene expression for these proteasome factors. A compensatory upregulation of proteasome gene transcription has been shown to occur in response to proteasome inhibition [46]. To confirm that loss of pomp and psmd6 functions lead to a defect in proteasome function, we examined levels of ubiquitinated proteins. We observe an increase in ubiquitinated proteins in both pomp and psmd6 mutant larvae (Fig 4E). This increase is similar to that observed when wild-type embryos are treated with the proteasome inhibitor bortezomib (Fig 4E). These results show that we have generated transcription-null alleles of pomp and psmd6 and that loss of these genes leads to the expected defects in proteasome function.

To examine the phenotypes of pomp and psmd6 mutant embryos, we used incrosses of heterozygous pomp and psmd6 fish. pomp^+/+, pomp^+/-, psmd6^+/+, and psmd6^+/- larvae exhibit normal body and heart morphology at 4 dpf (Fig 4F, 4G, 4I, 4J, 4L, 4M, 4O, 4P). For both pomp and psmd6 homozygous mutants, we observed heart edema and heart morphology defects by 4 dpf. Both pomp and psmd6 mutant hearts showed elongated and lateral heart looping defects, similar to those seen in the G0 CRISPR targeted embryos, although the two mutant lines differed in the frequency with which these phenotypes occurred (Fig 4H, 4K, 4N, 4Q, 4T). We also observed similar elongated heart looping defects in bortezomib-treated embryos (Fig 4R - 4T). In addition to the heart looping defects, pomp and psmd6 mutant larvae exhibit reduced craniofacial structures and blood pooling in the head (Fig 4H, 4K). Mutants of both lines are non-viable, with 50% of pomp mutants dying by 7 dpf and 100% dead by 8 dpf, and 50% of psmd6 mutants dying between 4-5 dpf and 100% dead by 7 dpf (Fig 4U). These results support our findings from the G0 CRISPR targeting and provide further support for the Proteasome cluster genes playing similar roles in heart development.

pomp and psmd6 mutant hearts exhibit cellular blebbing

We next investigated the timing of the appearance of heart development defects and the morphology of the hearts in pomp and psmd6 mutant embryos. We collected embryos at a series of time points during embryonic development and examined expression of the pan-cardiomyocyte and myocardial marker myl7, using RNA in situ hybridization and myl7:EGFP [47,48]. At 18 hours post fertilization (hpf), myl7 is expressed in cardiomyocyte precursors in the anterior lateral plate mesoderm (ALPM) in control embryos, and this expression appears normal in pomp and psmd6 mutants (Fig 5A - 5C). At the time of myocardial tube formation at 24 hpf, myl7 expression and tube formation appear normal in pomp and psmd6 mutants (Fig 5D - 5F). At the stage of early chamber formation and heart looping (48 hpf), and continuing to about 72 hpf, myl7 expression and heart morphology continue to appear largely normal in pomp and psmd6 mutants (Fig 5G - 5L). These findings suggest that initial heart development occurs normally in pomp and psmd6 mutants.

Fig 5. Early heart development appears normal in pomp and psmd6 mutants.

A.-C. Dorsal views of ALPM labeled for myl7 in blue in 18 hpf embryos. Normal heart morphology: 6/6 pomp^+/+, 7/7 psmd6^+/+, 11/11 pomp^-/-, 8/8 psmd6^-/-. Scale bar = 100 μm. D.-F. Dorsal views of hearts labeled for myl7 in blue in 24 hpf larvae. Normal heart morphology: 8/8 pomp^+/+, 8/8 psmd6^+/+, 6/6 pomp^-/-, 10/10 psmd6^-/-. Scale bar = 100 μm. G.-I. Ventral views of hearts labeled with myl7:EGFP in 48 hpf embryos. Normal heart morphology: 13/16 pomp^+/+, 11/12 psmd6^+/+, 14/14 pomp^-/-, 11/12 psmd6^-/-. Scale bar = 50 μm. V, ventricle. A, atrium. J.-L. Ventral views of hearts labeled with myl7:EGFP in 72 hpf larvae. Normal heart morphology: 35/38 pomp^+/+, 22/24 psmd6^+/+, 39/45 pomp^-/-, 25/29 psmd6^-/-. All embryos were genotyped. Scale bar = 50 μm. V, ventricle. A, atrium.

We then examined 4 dpf hearts using confocal imaging of myl7:EGFP. Maximum intensity projections of the entire heart made at lower magnification have the effect of highlighting the ventral surface morphology due to its greater brightness relative to the deeper tissue. By examining these images, we observed abnormal arrangement of the atrium and ventricle and rounding-up of the cells on the myocardial surface in both pomp and psmd6 mutants (Fig 6A - 6C). By examining projections of a few slices obtained at higher magnification through the ventral ventricular wall, we observed that the trabecular myocardial cells also appeared rounded (Fig 6D - 6F). To quantify this cellular blebbing phenotype, we counted the number of rounded cells protruding outward from the heart wall. While wild-type hearts occasionally showed such rounded cells, pomp and psmd6 mutant hearts showed about 5X more rounded cells (Fig 6G).

Fig 6. pomp and psmd6 mutants exhibit myocardial cell blebbing.

A.-C. Ventral views of hearts as maximum intensity projections (MIPs) of the entire myocardium, labeled with myl7:EGFP in 4 dpf larvae. Scale bar = 100 μm. V, ventricle. A, atrium. D.-F. Ventral views of hearts as MIPs of 2-7 slices (1.6-5.6 μm total section thickness) through the ventral ventricular wall, labeled with myl7:EGFP in 4 dpf larvae. Arrows in E and F indicate examples of rounded cardiomyocytes in the interior wall of the ventricle. Hearts in D - F are from different larvae than those in A - C. Presence of rounded trabecular cardiomyocytes: 0/17 pomp^+/+, 1/15 psmd6^+/+, 25/29 pomp^-/-, 24/27 psmd6^-/-. All embryos were genotyped. Scale bar = 50 μm. G. Graph showing numbers of protruding rounded cardiomyocytes per heart at 4 dpf. Only cells clearly projecting outward from the surface of the heart were counted. Each dot represents a heart/larva. Bars represent mean +/- SD. N = 9-13 per genotype. **** P < 0.0001.

pomp and psmd6 mutant hearts exhibit reduced cardiac function

We then asked whether pomp and psmd6 mutant embryos show defects in cardiac function. We made videos of control and mutant hearts at 3 dpf and 4 dpf over successive cardiac cycles to quantify heart rate and ventricle areas at systole and diastole, using established approaches [49,50]. To estimate cardiac function, area measurements were used to calculate percent ventricular fractional area change (FAC). At 3 dpf, pomp and psmd6 mutant hearts still appear morphologically normal (Fig 7A - 7C), and heart rates, ventricle areas, and FAC are comparable to control values for pomp and psmd6 mutants (Fig 7G - 7I). At 3 dpf, the only significant difference observed is ventricular area at systole in psmd6 mutants (Fig 7G). At 4 dpf, pomp and psmd6 mutants exhibit obvious edema and malformed hearts (Fig 7D - 7F), and both pomp and psmd6 mutants show significantly reduced heart rates, reduced ventricle areas at systole and diastole, and reduced ventricular FAC (Fig 7J - 7L). These results show that pomp and psmd6 are both needed for proper cardiac function at 4 dpf. These results also suggest that the appearance of the cardiac morphological defects observed at 4 dpf, such as edema and cellular blebbing, arise at a similar time as the functional deficits.

Fig 7. Cardiac function deficits are observed by 4 dpf in pomp and psmd6 mutants.

A.-F. Brightfield views of hearts taken from videos of 3 dpf and 4 dpf larvae. A’-F’ panels show ventricles outlined in red. Arrowheads in E’ and F’ point to edema. Scale bars = 100 μm. G.-I. Graphs showing ventricular area, heart rate, and ventricular fractional area change (FAC) measurements at 3 dpf. Each dot represents the average of 3 measurements from a single heart/larva. Bars represent mean +/- SEM. Animals were genotyped after video recordings. N = 13 for pomp control (+/+). N = 7 for psmd6 control (+/+). N = 7 for pomp^-/-. N = 8 for psmd6^-/-. * P < 0.05. J.-L. Graphs showing ventricular area, heart rate, and ventricular FAC measurements at 4 dpf. Each dot represents the average of 3 measurements from a single heart/larva. Bars represent mean +/- SEM. Animals were genotyped after video recordings. N = 12 for all conditions: pomp control (+/+ and +/-), psmd6 control (+/+ and +/-), pomp^-/-, and psmd6^-/-. ** P < 0.01; **** P < 0.0001.

pomp and psmd6 mutant hearts exhibit reduced outflow tracts

We then examined additional markers associated with myocardial differentiation and heart development. To assess myocardial sarcomere formation in pomp and psmd6 mutants, we examined immunostainingοfα-actinin, which localizes to Z-discs in striated cardiac and skeletal muscle [51,52]. We observe that α-actinin is organized in its periodic pattern in wild-type and in pomp and psmd6 mutant hearts at 4 dpf (Fig 8A - 8E). No significant differences in sarcomere lengths or myofibril widths were observed between wild-type and mutant hearts (Fig 8F - 8G), suggesting that pomp and psmd6 are not required for sarcomere or myofibril formation.

Fig 8. Myocardial sarcomere and myofibril formation appear normal in pomp and psmd6 mutants.

A.-C. Ventral views of ventricles labeled with α-actinin (A-C) in 4 dpf larvae. Scale bar = 25 μm. D.-E. Magnified views of α-actinin stain. Encircled green lines illustrate sarcomere lengths, with corresponding measurements shown. Encircled magenta lines illustrate myofibril widths, with corresponding measurements shown. Scale bar = 10 μm. F.-G. Graphs showing (F) sarcomere length and (G) myofibril width measurements at 4 dpf. Each dot represents the average of 9-14 measurements (one per myofiber) from a single heart/larva. Bars represent mean +/- SEM. All animals were genotyped. N = 7 pomp^+/+ and psmd6^+/+. N = 8 pomp^-/- and psmd6^-/-. No significant differences were observed between wild-type and mutant hearts.

To assess formation of the outflow tract in pomp and psmd6 mutants, we examined expression of the outflow tract smooth muscle marker elastinb (elnb) [53]. Early expression of elnb appears reduced in pomp and psmd6 mutant hearts at 72 hpf (Fig 9A - 9C). At 96 hpf, the elnb expression domain appears strongly reduced and dysmorphic (Fig 9D - 9F). We measured both the width of the lumen of the outflow tract elnb domain (“internal width”) and the length of the elnb domain and find that both measurements are reduced at both time points for pomp and psmd6 mutants (Fig 9G - 9H). These results show that pomp and psmd6 are needed for proper outflow tract formation. The outflow tract defects observed at 72 hpf in pomp and psmd6 mutants appear to arise prior to the other cardiac structural and functional deficits observed at 4 dpf.

Fig 9. pomp and psmd6 mutants exhibit reduced outflow tracts.

A.-F. Ventral views of outflow tracts, labeled with elnb (green), in (A-C) 72 hpf larvae and (D-F) 96 hpf larvae. Myocardium is labeled with myl7 (purple). G.-H. Graphs showing (G) the internal width of the elnb domain, measured at the maximum width position, and (H) the length of the elnb domain. Each dot represents an individual heart/larva. Bars represent mean +/- SD. All animals were genotyped. N = 7-9 per condition. * P < 0.05; ** P < 0.01; *** P < 0.001; **** P < 0.0001. Scale bar = 25 μm.

Discussion

In this study, we take a novel approach to identify candidate human CHD genes. Using multiple data systems, including human exome sequences, mouse transcriptomes, and protein-protein interactions, we identify a subnetwork of potential CHD genes that includes multiple genes encoding proteasome factors. We use G0 CRISPR targeting in zebrafish to demonstrate roles for these proteosome factor genes in zebrafish heart development. Furthermore, our analyses of stable mutant zebrafish strains for two of these genes, pomp and psmd6, reveal novel roles for proteosome genes in heart development. Our work addresses a major hurdle for understanding the causes of CHD, through the identification and validation of a new set of candidate CHD genes.

A major innovative aspect of this study is our identification of a set of 245 human genes with presumptive roles in heart development and CHDs. The standard approach to identify CHD genes is through using exome and genome sequencing of CHD patients and their families to identify deleterious mutations. In contrast, our approach identifies candidate CHD genes through their lack of deleterious mutations in the normal human population. While our list of 245 candidate CHD genes includes several known CHD genes, most of the 245 genes have unknown requirements in early heart development. There are about 130 known CHD genes, but it is estimated that there are over 400 genes that contribute to CHDs [18,22]. Thus, our list of candidate CHD genes could represent a majority of unknown CHD genes.

We use PPI network analyses to identify a functionally-related subnetwork that includes five genes encoding proteasome factors: POMP, which encodes Proteasome maturation protein, and PSMA6, PSMA7, PSMD3, and PSMD6, which encode proteasome subunits. The proteasome is a multi-subunit complex that is part of the ubiquitin-proteasome system (UPS), which is responsible for carrying out the majority of normal cellular protein degradation [39,54]. PSMA6 and PSMA7 both encode α subunits of the 20S core particle of the proteasome. PSMD3 and PSMD6 both encode non-ATPase subunits of the proteasome’s 19S regulatory particle. POMP (also known as Ump1) functions as an assembling chaperone for the 20S proteasome [39]. The proteasome and the UPS have well-established roles in adult cardiac diseases [55]. However, proteasome factors have thus far only been indirectly linked with CHDs. In human CHD exome studies, only single, likely deleterious de novo variants have been identified in PSMD6 (a case with a conotruncal defect) and POMP (a case with an undefined CHD) [18]. Mutations in PSMD12, which encodes a 19S non-ATPase subunit, have been linked with a human neurodevelopmental syndrome that in some cases involve CHDs of variable forms and severities [56]. In mice, Psmd6 homozygous mutants have complete pre-weaning lethality [27], but heart defects have not been assessed in these mutants, and none of our five other proteasome genes have had phenotypic characterization in mice. By demonstrating roles for proteasome factors in zebrafish heart development, our study supports investigating further links between proteasome genes and CHDs in humans.

To demonstrate the functions of proteasome genes in early heart development, we use zebrafish G0 CRISPR targeting and mutant strains. Zebrafish have many advantages for discovering and characterizing new CHD genes [28–32]. In our study, we take advantage of a previously-established CRISPR screening approach [31] to show that our subnetwork genes are all required for zebrafish heart development. Although some of the genes we test have duplicate copies in the zebrafish genome, we were able to use co-CRISPR-targeting of gene duplicates to demonstrate their functions in heart development. Furthermore, using stable mutant strains, we show that pomp and psmd6 mutant embryos have phenotypes that closely resemble, and thus validate, the G0 CRISPR embryos.

A key finding from our study is that zebrafish pomp and psmd6 mutant embryos share heart phenotypes that are observed in other zebrafish mutants for heart development and CHD genes. The cardiac edema, dysmorphic heart chambers, and impaired cardiac looping phenotypes that we observe in pomp and psmd6 mutant embryos are commonly seen, with varying severities, in zebrafish mutants for known CHD genes [28,40,57]. In many of the well-characterized examples, such as nkx2.5, tbx1, and tbx5 mutants, such phenotypes are observed earlier in development than what we have described for pomp and psmd6 mutants [58–61]. We also observed diminished cardiac function in pomp and psmd6 mutants, with defects comparable to those that have been reported in zebrafish models of hypoplastic left heart syndrome [49,62]. A distinctive phenotype we observe in pomp and psmd6 mutants is myocardial cell blebbing. This phenotype has also been observed in other zebrafish mutants for critical heart development genes, including snai1b, tcf21, wt1a. flii, and klf2a;klf2b double mutants [63–66]. In these examples, myocardial cell blebbing has been shown to be due to cell adhesion and cell polarity defects but is not associated with increased cell death [63–66]. Although beyond the scope of this study, more in-depth analyses are needed to understand how proteasome factors might interact with known CHD factors in heart development and myocardial cell adhesion.

The earliest defect that we have been able to define in pomp and psmd6 mutants is reduced outflow tracts, with measurable defects observed at 72 hpf for both mutants. Defective outflow tract formation has also been described in zebrafish mutants for known CHD genes, including tbx1 and rbfox2 [49,59,67]. The zebrafish outflow tract shares evolutionary conservation with the mammalian outflow tract [68–70], and the outflow tract is a major site of human CHDs [71,72]. Defects in cardiac function have been shown to cause reduced outflow tract development in zebrafish embryos [73]. However, we do not see significant heart function defects in both pomp and psmd6 mutants until 96 hpf, when we also observe cardiac edema and blood pooling in the head in both mutants. Vascular and other pleiotropic defects are often observed in zebrafish mutants with cardiac defects [57]. Zebrafish jag1b;jag2b mutants, a model for Alagille syndrome, exhibit cardiac edema along with cranial hemorrhaging and craniofacial defects [74]. CRISPR targeting of the known CHD gene mib1 in zebrafish embryos also leads to cranial hemorrhaging and cardiac edema [75]. While pericardial edema frequently accompanies defects in cardiac development in zebrafish, non-cardiac defects can also lead to edema [57,76]. Our studies here on pomp and psmd6 mutants have not yet defined whether there are causative links between the cardiac and vascular or craniofacial defects observed.

A potential limitation of our study is the selection of the embryonic cardiac muscle cell lineage gene set that we used to intersect with high s_het genes to identify our candidate CHD gene list (S3 Table and Fig 1C). We focused on the cardiac muscle cell lineage because genes with high levels of embryonic heart expression are highly enriched for damaging de novo mutations in CHD patients in exome studies [14,15,18,19]. We used gene expression data from two mouse studies because of their use of single-cell RNA-seq and because they addressed very early stages of heart development [33,34]. Recently, human heart atlases have become more refined and are capturing earlier developmental stages [37,45,77–81]. In future studies, it should be possible to examine how high s_het genes and known CHD genes intersect with different cell lineages that are relevant to heart development, such as endocardial and neural crest cell lineages.

Another limitation of our study is that it is unclear how directly the proteasome factor genes are functioning in heart development. pomp, psmd6, and other genes encoding proteasome factors are very broadly expressed throughout zebrafish embryogenesis [40,82]. pomp and psmd6 mutant embryos exhibit extracardiac defects, including reduced craniofacial development. These findings are consistent with the craniofacial defects observed upon loss of other zebrafish proteasome factors [56,82]. In humans, mutations in PSMD12 are linked to a syndrome characterized by neurodevelopmental and craniofacial issues as well as, in some cases, cardiovascular defects [56]. These issues make it challenging to identify the primary mechanism through which proteasome factors are needed for heart development. Future studies employing zebrafish genetic mosaics or mouse conditional mutant strains could help address when, and in which tissue(s), proteasome factors influence heart development.

Materials and methods

Ethics statement

All experiments involving live zebrafish (Danio rerio) were carried out under the approval of Seattle Children’s Research Institute’s Institutional Animal Care and Use Committee.

Identifying high s_het genes expressed in early heart development

s_het values and deciles are from Cassa et al.’s Supplementary Table 1 [23]. We binned s_het values by decile and used ANOVA to calculate the correlation of high s_het genes with genes exhibiting damaging de novo variants in CHD cases (Fig 1A). de novo variant data are from Jin et al.’s Supplementary Data Set 9 (CHD cases) and Supplementary Data Set 10 (controls) [18].

To select high s_het genes, we took the top 2 deciles of s_het genes (3199 genes) from Cassa et al.’s Supplementary Table 1 [23]. We entered these 3199 gene symbols into the Multi-symbol checker in HGNC (HUGO Gene Nomenclature Committee site; https://www.genenames.org/tools/multi-symbol-checker/) [83]. We collected the updated approved gene symbols to generate a revised list of 3190 high s_het genes (S2 Table).

Mouse cardiac muscle cell lineage datasets were generated from [33,34]. Cao et al. [33] performed single-cell RNA sequencing on whole mouse embryos from E9.5-E13.5. We took all genes from Cao et al.’s Cluster 34 Cardiac Muscle Lineages (gene list obtained from the file DE_gene_main_cluster.csv, downloaded from https://oncoscape.v3.sttrcancer.org/atlas.gs.washington.edu.mouse.rna/downloads). We entered this list into the HGNC HCOP (Comparison of Orthology Predictions) tool to obtain 610 human genes. Jia et al. [34] performed single-cell RNA sequencing on isolated cardiac progenitor cells from E7.5-E9.5 mouse embryos. We combined the gene lists from Jia et al.’s Supplementary Data 1 (list of genes that are differentially expressed in 3 Nkx2.5-positive cardiac progenitor cell clusters) and Supplementary Data 2 (list of genes that are differentially expressed in 5 Isl1-positive cardiac progenitor cell clusters). We entered these combined lists into the HGNC HCOP tool to obtain 651 human genes. These two gene sets were then combined, with duplicates removed, to identify 1119 genes (S3 Table). We used BioVenn (https://www.biovenn.nl/) to identify the overlap between the high s_het genes (3190 genes) and genes expressed in the mouse cardiac muscle cell lineage (1119 genes) (Fig 1C and S4 Table). Venn diagrams were generated using BioVenn.

BioVenn was used to determine the overlap between our set of 245 candidate CHD genes and 132 known CHD genes (Fig 1D). BioVenn was also used to determine the overlap between the 245 Candidate CHD genes and genes exhibiting damaging de novo mutations in CHD cases (Supplementary Data Set 9 from [18]; S5 Table).

A list of early human cardiac-expressed genes was generated from [37], which performed spatial transcriptomics on hearts from 4.5-9 week-old human embryos and single-cell RNA sequencing of hearts from 6.5 week-old human embryos. We combined all differential expression gene lists from their S2 Table and S3 Table to generate a list of 4670 genes. BioVenn was used to determine the overlap between these 4670 genes and the 245 candidate CHD genes (S6 Table).

For functional annotation, the 245 candidate CHD gene list was loaded into the Functional Annotation tool in DAVID ( [84]; DAVID Knowledgebase version 2023q2; originally DAVID v6.7 2020), selecting functional annotation category GOTERMS to specifically focus on GO terms and not additional redundant annotation terms.

Constructing protein-protein interaction networks

To construct Protein-Protein Interaction (PPI) networks, we used STRING [38], initially using v11.0 but also using versions up to v12.0. To generate the large network, we created a color-coding effect utilizing the “Payload datasets” feature within STRING. To generate a new payload dataset, we selected Homo sapiens and entered our list of known, candidate, and overlapping (candidate/known) gene symbol names. We made three different color categorizations and assigned a color code to each gene/node within each category. We used the following settings for the large network: We selected high confidence 0.7 for the minimum required interaction score. We selected the “max number of interactions to show” to be “none” for both the 1^st and 2^nd shell, to ensure the only nodes in the network are those from our candidate and known CHD gene lists. We selected “confidence” for “meaning of network edges”. We selected all active interaction data sources. We selected “hide disconnected nodes in the network”.

We utilized STRING’s local network cluster analysis for functional enrichment. We selected cluster CL:2692 “Regulation of ornithine decarboxylase, and proteasome assembly” within the Analysis tool in order to highlight the nodes/genes associated with that cluster in our network.

To generate a subnetwork based on the network nodes in CL:2692, we generated a new gene list. We started with the 8 nodes identified in CL:2692, and we then used the node interaction data from our large network to identify all proteins within our large network that share a direct connection with any of the eight nodes. A subnetwork was generated with these genes as inputs, with nodes color-coded as for the large network. We again used a minimum required interaction score of 0.7, utilized all active interaction data sources, and limited the nodes included within our subnetwork to only those specifically inputted by setting the max number of interactions to show to none for both the 1^st and 2^nd shell. We selected “evidence” for “meaning of network edges”, with the line colors indicating the different types of evidence supporting the interactions between two given proteins.

Zebrafish husbandry

Zebrafish were raised and staged as previously described [85,86]. Time indicated as hpf or dpf refers to hours or days post-fertilization at 28.5°C, respectively.

The wild-type stock and genetic background used was AB (ZFIN: ZDB-GENO-960809-7). The Tg(myl7:EGFP)^twu34 line has been previously described (ZFIN: ZDB-ALT-050809-20) [45]. For fish stock maintenance, eggs were collected from 20-30 min spawning periods and raised in Petri dishes in ICS water [87] in a dark 28.5°C incubator, up to 5 dpf. After 5 dpf, the fish were maintained on a recirculating water system (Aquaneering) under a 14 h on, 10 h off light cycle. From 6-30 dpf, the fish were raised in 2.8 L tanks with a density of no more than 50 fish per tank and were fed a standard diet of paramecia (Carolina) one time per day and Zeigler AP100 dry larval diet two times per day. From 30 dpf onwards, the fish were raised in 6 L tanks with a density of no more than 50 fish per tank and were fed a standard diet of Artemia nauplii (Brine Shrimp Direct) and Zeigler adult zebrafish feed, each two times per day.

G0 CRISPR targeting in zebrafish embryos

The sequences for the oligonucleotides used to synthesize the single-guide RNAs for G0 CRISPR targeting were taken from the published genome-scale Lookup Table [31] and are provided in S8 Table. For a negative control 4-guide set, we used the “Genetic Screen Scramble1 Control” guides [31] (S8 Table). sgRNAs were synthesized as described [31]. Pools of four gene-specific (or control) oligos, each incorporating a T7 RNA polymerase site, were annealed to a common scaffold oligo and transcribed in vitro to generate pools of four sgRNAs, as described [31]. For G0 CRISPR phenotype analysis, 2 µL of a 4-guide cocktail of sgRNA at 2 µg/µL was combined with 2 µL of Cas9 protein (IDT Alt-R S.p. Cas9 Nuclease V3; 1081058) at 10 µM (diluted as in [31]) and incubated at 37°C for 5 minutes. In cases where two 4-guide cocktails were combined, 1 µL of each cocktail was used. 1 µL of phenol red injection solution (0.1% phenol red and 0.2M KCl in water) was added to generate the working solution for embryo injections. 1-cell stage embryos, collected from Tg(myl7:EGFP)^twu34 fish, were injected in the yolk with 2 nL of the RNP working solution. Heart morphology was scored in the live embryos using the myl7:EGFP transgene. At 2 dpf, the hearts were scored for cardia bifida and other heart defects (e.g., incomplete looping, dysmorphic chambers), and at 4 dpf they were scored again to identify defects that had arisen since 2 dpf.

To assess CRISPR guide efficiencies, we assayed G0 embryos injected with the 4-guide cocktail for psmd6. 4 dpf G0 CRISPR embryos were lysed and screened for Cas9-generated mutations (insertions or deletions, “indels”) by PCR amplification, using pairs of primers (S8 Table) flanking each of the four psmd6 target sites, followed by a restriction digest to test for loss of a restriction site just upstream of or overlapping the PAM site. We detected indels in 12/12 G0 embryos for psmd6 site 1, 0/12 embryos for psmd6 site 2, 11/12 embryos for psmd6 site 3, and 7/12 embryos for psmd6 site 4. Thus, for psmd6, over 90% of G0 embryos were targeted by more than one guide RNA. This efficiency is comparable to what was previously observed for the hand2 4-guide set, in which each of the four targeted hand2 sites was shown to be disrupted in at least 86% of sequenced alleles [31].

Generation of zebrafish mutant strains for pomp and psmd6

To generate deletions encompassing the transcriptional start site and 5’ end of pomp and psmd6, we used a protocol based on that previously described [88]. CRISPR target sequences were selected using the Integrated DNA Technologies Alt-R HDR design tool (https://www.idtdna.com/site/order/designtool/index/HDRDESIGN). For each gene, 2 sites were chosen upstream and two downstream of the transcriptional start site, such that deletions of approximately 500–2500 base pairs would be generated. The Alt-R crRNAs were annealed with Alt-R tracrRNA to yield functional gRNA duplexes as in [88]. The four combinations of one upstream plus one downstream gRNA were tested to find the pair with the highest efficiency in generating the desired deletion. The sequences of the crRNAs used to generate the pomp and psmd6 deletions are given in S8 Table. RNP complexes of gRNA + Cas9 protein were assembled as in [88], except that 0.5 µL each of one upstream and one downstream gRNA (each at 25 µM) were combined with 1 µL of 25 µM Cas9 plus 2 µL of H₂O. After a 5-minute incubation at 37°C, 1 µL of phenol red injection solution was added. 1-cell stage embryos, collected from Tg(myl7:EGFP)^twu34 fish, were injected in the yolk with 1-2 nL doses of the RNP complexes. Injected embryos were raised to adulthood. To identify F0 fish with germline transmission of pomp or psmd6 deletions, F0 adults were crossed with Tg(myl7:EGFP)^twu34 zebrafish. A subset of each of the resulting F1 clutches was screened by PCR analysis using primers flanking the expected deletions, where amplicons were only generated in animals carrying a deletion allele. The remaining F1 animals from positive F0 fish were raised to adulthood. Sanger sequencing was performed on the PCR-amplified deletion allele from F1 heterozygous animals and confirmed the deletion of the genomic sequence between the two CRISPR target sites. To identify heterozygous mutant carriers for the F1 and subsequent generations, fin-clippings from adults were collected, and PCR analysis was performed using a cocktail of three primers that generate different sized amplicons from the wild-type and deletion alleles. Primer sequences are provided in S8 Table. The deletion allele-specific primer pair spans the deletion but does not generate an amplicon from the wild-type allele due to the length of the intervening sequence, and the wild-type allele-specific primer pair includes one primer within the deletion. For pomp^scm41, primers T1F1 + T1R1 produce a 225 bp product from the wild-type allele and T3F3 + T1R1 produce a 190 bp product from the deletion allele. For psmd6^scm40, T3F1 + T3R1 produce a 298 bp product from the wild-type allele and T3F1 + T1R1 produce a 266 bp product from the deletion allele. The same PCR-based assays were used to genotype immunostained and RNA in situ-stained embryos. Heterozygous carriers were outcrossed with Tg(myl7:EGFP)^twu34 fish each generation. Mutant embryos used in experiments were F3 generation or later.

Quantitative reverse transcription PCR (qRT-PCR)

Embryos were obtained from incrosses of heterozygous pomp^scm41 or psmd6^scm40 fish. At 4 dpf, replicate groups of 10–20 clearly identifiable phenotypically mutant (pericardial and peri-ocular edema, reduced heads) and phenotypically wild-type larvae (normal appearance with swim bladder present) were flash-frozen in liquid nitrogen and then homogenized in TriZol (Invitrogen, ThermoFisher Scientific; 15596026) by trituration through a 27G needle. After phase separation, the aqueous portion was extracted with 24:1 chloroform:iso-amyl alcohol. One-half volume of 100% ethanol was added to the resulting aqueous phase and the samples were then further purified using the RNAqueous Micro Kit (Invitrogen, ThermoFisher Scientific; AM1931). Total RNA was reverse transcribed with the SensiFAST cDNA Synthesis kit (Bioline, Meridian Life Science; BIO-65054). Primer pairs were designed using Primer-BLAST such that they either span an intron (pomp) or one of the primers crosses an exon-exon boundary (psmd6). Primers for rpl13a were previously described [89]. Primers used are listed in S8 Table. qPCR was performed using the KAPA SYBR FAST kit (Roche; 07959567001) on a Bio-Rad CFX96 machine. Ct values for the genes of interest were normalized to rpl13a, and then ΔΔCts were calculated by normalizing each sample ΔCt to the average ΔCt of the replicate wild-type samples. The ΔΔCt values were then log transformed to give fold change vs the average transcript level of the wild-type replicates.

Bortezomib treatments

Embryos from spawnings of Tg(myl7:EGFP)^twu34 fish were enzymatically dechorionated with Pronase (Sigma; 10165921001) at 24 hpf as described [86]. The embryos were arrayed in the wells of a 12-well plate at 25 animals per well in 3 mL of embryo medium (EM; 14.97 mM NaCl, 0.50 mM KCl, 0.98 mM CaCl₂.2H₂O, 0.15 mM KH₂PO₄, 0.99 mM MgSO₄.7H2O, 0.05 mM Na₂HPO₄, 0.83 mM NaHCO₃) containing bortezomib (Sigma; 5043140001) or 0.5% DMSO as a vehicle control. Treatments were done in triplicate wells. The drug treatments began at approximately 24 hpf, and the drug-containing media were replaced each day until 4 dpf, at which point the larvae were collected as 3 replicates per treatment condition and lysed as described below for immunoblotting.

Immunoblotting

Embryos were obtained from incrosses of heterozygous pomp^scm41 or psmd6^scm40 fish. At 4 dpf, replicate groups of 20 clearly identifiable phenotypically mutant (pericardial and peri-ocular edema, reduced heads) and 20 phenotypically wild-type (normal appearance with swim bladder present) larvae were de-yolked in normal Ringer’s and lysed in 12 µL per embryo of 1.5x NuPAGE LDS sample buffer (Invitrogen, ThermoScientific; NP0007). Approximately one embryo equivalent per sample was separated by reducing SDS-PAGE on 4-12% Bis-Tris NuPAGE gels (Invitrogen, ThermoScientific; NP0322BOX) in MOPS buffer (Invitrogen, ThermoScientific; NP0001) and transferred to nitrocellulose. Blots were blocked with Intercept TBS (LI-COR Biosciences; 927–60001) and probed with anti-ubiquitinated proteins (clone FK2; 1:1000; EMD Millipore; 04–263) followed by goat anti-mouse IR Dye 800CW secondary (1:10,000; LI-COR Biosciences; 926–32210). Blots were imaged, then additionally probed with anti-alpha actin (CloneC4; 1:1000; MP Biomedical; 0869100) followed by goat anti-mouse DyLight 680 secondary (1:20,000; ThermoScientific; 35518) and were then re-imaged. Immunoblots were visualized on a LI-COR Odyssey infrared scanner, and images were generated using Image Studio Lite v5.2 (LI-COR Biosciences).

Zebrafish whole-mount RNA in situ hybridization and immunostaining

The following cDNA probes were used for RNA in situ hybridization: myl7 [47] and elnb [53]. Whole-mount in situ hybridization colorimetric and fluorescent in situ staining was performed as previously described [90,91], except that hybridizations for both experiments were performed in hybridization buffer that included 5% dextran sulfate. Following staining, tail clips from post-in situ hybridized embryos were lysed and genotyped.

The primary antibodies used for immunostaining were anti-α-actinin (ACTN2) (clone EA-53, 1:200; Sigma; A7811) and anti-GFP (1:200; Torrey Pines Biolabs; TP401). The secondary antibodies used were goat anti-rabbit IgG (H + L) Alexa Fluor 488 (Invitrogen, ThermoScientific; A-11008) and goat anti-mouse IgG (H + L) Alexa Fluor 568 (Invitrogen, ThermoScientific; A-11004). 96 hpf larvae were fixed in fresh 4% paraformaldehyde in PBS for 4 hours at room temperature, washed out of fixative, and stored in PBS containing 0.02% sodium azide at 4°C. The larvae were permeabilized with 5 ug/mL Proteinase K (Sigma; 3115836001) for 90 minutes, washed in PBS plus 0.1% Tween-20 (PBTw), treated with acetone for 20 minutes at -20°C, washed with PBTw, permeabilized with 1% sodium dodecyl sulfate in PBS for 15 minutes, and washed again with PBTw before being blocked in 2% heat-inactivated normal goat serum, 2% bovine serum albumin in PBTw (Fish Block) for 16 hours at 4°C. Primary antibodies were diluted in Fish Block and incubated with the embryos for 16-20 hours at 4°C. The embryos were then washed in PBTw, re-blocked in Fish Block for 2 hours at room temperature, and incubated with secondary antibodies plus 5 µg/mL DAPI diluted in Fish Block for 16-20 hours at 4°C. Finally, the embryos were washed in PBTw and stored in 4% PFA at least overnight before tail tips were dissected for genotyping.

Microscopic imaging of zebrafish embryos and larvae

For myl7:EGFP imaging in live embryos, embryos were raised in 0.003% N-phenylthiourea (Sigma; P7629) in ICS water beginning at 6 hpf/shield stage and were anaesthetized in tricaine (Sigma; A5040) just prior to imaging. For the bortezomib-treated embryos and the DMSO-treated sibling controls, melanin was bleached from pigmented embryos post-fixation by treatment with a solution of 1 part PBTw: 1 part 0.1% KOH: 0.1 part 30% H₂O₂. For imaging of myl7:EGFP and whole-mount RNA colorimetric in situs, embryos were imaged in 2.5% methyl cellulose (Sigma; M0387) in ICS water. Images were captured using an Olympus SZX16 stereomicroscope with an Olympus DP74 camera and cellSens Dimension v4.1 imaging software. Sequential brightfield and GFP fluorescence images were captured and later merged using Adobe Photoshop.

Imaging of beating hearts was performed similarly to approaches previously described [49,50]. Animals were brought to room temperature (about 24°C) before imaging. Three larvae at a time were anesthetized in 5 mL of ICS water containing 0.008% tricaine for 3 dpf animals and 0.016% tricaine for 4 dpf animals. Animals were anaesthetized for 1 minute and were then transferred to 2.5% methyl cellulose for imaging. Videos of beating hearts were acquired using an Olympus SZX16 stereomicroscope, a Basler ace 1.3MP 200 camera, and Basler pylon Viewer 7.5. Images were collected at 6 ms intervals (167 fps) for 8 sec. After imaging, the individual larvae were collected for genotyping. Videos were analyzed in ImageJ. Heart rate was measured by recording the time from end systole to end diastole over 5 cardiac cycles, then dividing 5 by that time interval and multiplying the quotient by 60 to give beats per minute. At both end diastole and end systole, the border of the ventricle was outlined, using the ImageJ polygon tool, to measure cross-sectional area. Three separate measurements at different cardiac cycles were made for each animal to calculate an average end diastolic area and end systolic area. Ventricular fractional area change was calculated using the following formula: [(area at diastole−area at systole)/area at diastole]×100.

For whole-mount immunostaining and fluorescent in situs, embryos were partially cleared in 80% glycerol. The trunk and tail were removed to facilitate mounting of the head and pharynx in 4% propyl gallate (Sigma; P3130) in 80% glycerol. Hearts were imaged on a Leica TCS SP5 confocal with a 20x air or 40x oil immersion objective. Maximum intensity projections were made in ImageJ (https://fiji.sc/). Rounded cardiomyocytes were manually counted using the “Multi-point” tool in ImageJ by scrolling through Z-stacks captured with the 40x objective of merged GFP and DAPI channel images of individual hearts and marking the double-positive nuclei protruding outward from the myocardial wall. Cardiac sarcomere lengths and myofibril widths were measured using the “Straight line” tool and Measure function in ImageJ on the α-actinin channel of Z-stacks captured with the 40x objective. The measurements were made only where an individual myofiber could be clearly distinguished. Only one of each measurement was made per myofiber to ensure that a representative sample of myofibers per heart were scored.

Statistical analysis and data visualization

For Fig 3D, where the incidence of heart defects observed with CRISPR injections of each gene of interest was compared to the control guide injections, significance was determined using one-way ANOVA with Dunnett’s correction for multiple comparisons. For all other comparisons between wild-type controls and mutant siblings, unpaired two-tailed Student’s t tests were used. Statistics were performed in GraphPad Prism 10. All plots were made in GraphPad Prism 10, except for Figs 3C and 4T, which were made in MS Excel. The original data underlying Figs 3-9 is provided at the online repository Dryad [92].

Supporting information

S1 Table. Known CHD gene list.

(XLSX)

S2 Table. s_het top two deciles gene list.

(XLSX)

S3 Table. Mouse embryonic heart-expressed gene list.

(XLSX)

S4 Table. Candidate CHD gene list.

(XLSX)

S5 Table. Candidate CHD genes with CHD de novo variants.

(XLSX)

S6 Table. Candidate CHD genes with human embryonic heart expression.

(XLSX)

S7 Table. Local network clusters.

(XLSX)

S8 Table. Sequences of oligonucleotides used.

(XLSX)

Data Availability

The Original Data file containing the data that support the findings of this study is publicly available in the open online repository Dryad at the link: https://doi.org/10.5061/dryad.hx3ffbgs1.

Funding Statement

This work was supported by the Department of Defense Office of the Congressionally Directed Medical Research Program (Award #W81XWH-20-1-0433) (PI LM), the Additional Ventures Single Ventricle Research Fund (PI LM), the Additional Ventures Tools and Technology Expansion Award (PI LM), and Seattle Children’s Hospital Heart Center Research Initiative. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.