MEF2C controls segment-specific gene regulatory networks that direct heart tube morphogenesis

Abstract

The gene regulatory networks (GRNs) that control early heart formation are beginning to be understood, but lineage-specific GRNs remain largely undefined. We investigated networks controlled by the vital transcription factor MEF2C, with a time course of single-nucleus RNA- and ATAC-sequencing in wild-type and Mef2c-null embryos. We identified a “posteriorized” cardiac gene signature and chromatin landscape in the absence of MEF2C. Integrating our multiomics data in a deep learning-based model, we constructed developmental trajectories for each of the outflow tract, ventricular, and inflow tract segments, and alterations of these in Mef2c-null embryos. We computationally identified segment-specific MEF2C-dependent enhancers, with activity in the developing zebrafish heart. Finally, using inferred GRNs we discovered that the Mef2c-null heart malformations are partly driven by increased activity of the nuclear hormone receptor NR2F2. Our results delineate lineage-specific GRNs in the early heart tube and provide a generalizable framework for dissecting transcriptional networks governing developmental processes.

A recurrent observation in developmental biology is that the same transcription factor (TF) can play different roles depending on the cells or developmental stage in which it is expressed. The cell type- or stage-specific activity of a TF is dependent on a number of factors, including its own expression level, the availability of potential co-factors, the presence or absence of other TFs that may act co-operatively or antagonistically, and the dynamic state of the chromatin landscape in which the TF is acting (Hafner and Boettiger 2023; Lee and Young 2013; Spitz and Furlong 2012; Wagh et al. 2023). Together, these phenomena can be conceptualized as transcriptional or gene regulatory networks (GRNs) that can be used to understand the role of a particular TF within a given developmental context (Mercatelli et al. 2020). However, for most TFs, the precise GRNs underlying their pleiotropic activity in distinct cell types and at different developmental timepoints remain undefined.

The developing heart presents an excellent model for exploring this question. Several TFs critical for cardiac development have been identified and their expression patterns are well-defined (Bruneau 2013; Evans et al. 2010; Kelly et al. 2014). Moreover, lineage tracing studies have revealed distinct progenitor cell types that contribute to the various segments of the linear heart tube (Devine et al. 2014; Evans et al. 2010; Kelly et al. 2014; Lescroart et al. 2014; Meilhac et al. 2004) (Fig. 1A). To explore in detail how a specific TF can have distinct roles within different parts of a developing organ, we focused on a key cardiac transcription factor, MEF2C (Martin et al. 1993), and its role in each of the developing heart tube segments. In mice, loss of MEF2C leads to cardiac defects and embryonic lethality by mid-gestation at embryonic day (E) 10.5 (Lin et al. 1997). Interestingly, although MEF2C is expressed in both first heart field (FHF) and second heart field (SHF) progenitors in the cardiac crescent at E7.75, and throughout the developing heart tube from E8.5 to E9, the loss of MEF2C causes distinct defects in different segments of the heart tube (Fig. 1B-C, Supplemental Fig. S1A). At E9, the outflow tract, which gives rise to the aorta and pulmonary trunk, is severely hypoplastic and only a single hypoplastic ventricle has formed. By contrast, the inflow tract, precursor to the atria, appears to be expanded and improperly patterned (Fig. 1C). These segment-specific defects suggest that MEF2C plays distinct regulatory roles in the outflow tract, ventricles, and inflow tract.

Figure 1: MEF2C is expressed throughout the developing heart tube and its loss causes segment-specific defects.

A) Schematic of cardiac progenitors and their contributions to linear heart tube development from cardiac crescent (E7.75) to looped heart tube (E9) stage. B) Immunofluorescent staining of MEF2C (cyan) and cardiac Troponin T (cTnT, magenta) in E7.75, E8.5, and E9 WT embryos. C) Representative images of WT and Mef2c KO embryos at E7.75, E8.5 and E9. Cardiac progenitors are marked by the Smarcd3-F6-eGFP reporter transgene (green). BF, brightfield. D) Schematic of the methodology and biological insights presented in the current study. Elements of this panel were created in BioRender. B, B. (2024) https://BioRender.com/i72e213. Scale bars = 200 μm. FHF, first heart field; aSHF, anterior second heart field; pSHF, posterior second heart field; LV, left ventricle; RV, right ventricle; V, ventricle; IFT, inflow tract; OFT, outflow tract.

To dissect the segment-specific regulation of heart tube morphogenesis by MEF2C, we performed combined single-nucleus RNA-seq and ATAC-seq (snRNA-seq and snATAC-seq) on wild-type (WT) and Mef2c knockout (KO) mouse embryos at key stages throughout heart tube development. Taking advantage of the ability to match the transcriptomic and chromatin accessibility data for every cell in the dataset, we applied a suite of complementary bioinformatics tools to construct GRNs for cardiac progenitor lineages, precisely defined the role of MEF2C in regulating gene expression and chromatin accessibility in the developing heart tube, identified novel MEF2C-dependent enhancers, and discovered a genetic interaction that underlies the inflow tract malformation in Mef2c KO embryos (Fig. 1D).

RESULTS

Loss of MEF2C reduces expression of key cardiomyocyte genes across all heart tube segments and alters expression of anterior/posterior markers

To assess the role of MEF2C in controlling gene expression in the developing heart tube, we performed snRNA-seq on stage-matched WT and Mef2c KO embryos at E7.75, E8.5, and E9 (n = 2 embryos for each genotype at each stage, Supplemental Fig. S1B). All embryos carried the Smarcd3-F6-eGFP transgene reporter to mark cardiac progenitors (Krup et al. 2023). At E7.75, we harvested the entire embryos, whereas at E8.5 and E9 we removed the headfolds and posterior trunk via microdissection to increase the proportion of cardiac progenitors captured for sequencing. We created separate Seurat objects (Hao et al. 2021) for the embryos profiled at each timepoint (E7.75, E8.5, and E9). Following pre-processing, quality control, normalization with scTransform (Choudhary and Satija 2022), and dimensionality reduction, we performed unbiased Louvain clustering and plotted the data in uniform manifold approximation and projection (UMAP) space. We then used the expression of known marker genes to identify the cell types represented by each cluster (Supplemental Fig. S2A-F, Supplemental Data S1).

To resolve specific cardiomyocyte subtypes, we subset and re-clustered cell types of interest, including cardiac progenitors, cardiomyocytes, and related mesoderm cell types (Fig. 2A-C, Supplemental Fig. S2G-I, Supplemental Data S2). By doing so, we were able to identify clusters representing FHF and early differentiating cardiomyocytes (CMs/FHF), SHF, and juxtacardiac field (JCF) progenitors (Tyser et al. 2021) at E7.75 (Fig. 2A). At E8.5, we could distinguish inflow tract cardiomyocytes (IFT-CMs), ventricular cardiomyocytes (V-CMs), and outflow tract cardiomyocytes (OFT-CMs) (Fig. 2B). At E9 we identified the IFT-CMs more specifically as either atrial or atrioventricular canal (A- or AVC-CMs) (Fig. 2C). We next performed differential gene expression analysis between Mef2c KO and WT cells within each of these cell types (Fig. 2D-F, Supplemental Data S3). Overall, more genes were down-regulated than up-regulated in the Mef2c KO relative to WT embryos, consistent with the notion that MEF2C is an activator of the myogenic transcriptional program (Black and Olson 1998). Among the top differentially expressed genes (DEGs), we found several genes that have been associated with high confidence to congenital heart defects (Yang et al. 2022), including Myh6, Myh7, Actc1, and Foxp1 (Supplemental Fig. S2J). Furthermore, nearly all these top DEGs were identified as the nearest genes to MEF2C-occupied peaks in a previously published MEF2C ChIP-seq dataset of embryonic hearts at E14 (Akerberg et al. 2019), suggesting that these genes are direct targets of MEF2C in the developing heart.

Figure 2: Loss of MEF2C reduces expression of key cardiomyocyte genes across all heart tube segments and alters expression of anterior/posterior markers.

A-C) UMAPs of snRNA-seq data for cardiac progenitors, cardiomyocytes, and related mesoderm subtypes from E7.75 (A), E8.5 (B), and E9 (C) embryos labeled by cell type (left) and genotype/sample ID (right). D-F) Bar plots displaying the number of up-regulated and down-regulated genes in Mef2c KO relative to WT in cell types of interest at E7.75 (D), E8.5 (E), and E9 (F). G-I) Dot plots displaying expression of key CM genes and anterior/posterior (A/P) markers at E7.75 (G), E8.5 (H), and E9 (I). J) Fluorescence in situ hybridization of key CM genes and A/P markers in E8.5-E9 (5–10 somites) WT and Mef2c KO embryos. Note the reduced expression of CM genes Tnnt2, Ttn, Nppa, and Nkx2–5, the expanded expression of posterior markers Tbx5, Gata4, and Wnt2, and the loss of anterior OFT marker Tdgf1 in Mef2c KO embryos compared to WT. n = 5–8 WT embryos and n = 7–9 Mef2c KO embryos from 7 independent litters tested per probe. Scale bars = 200 μm. CMs, cardiomyocytes; FHF, first heart field; SHF, second heart field; JCF, juxtacardiac field; CrM, cranial mesoderm; PrxM, paraxial mesoderm; LPM, lateral plate mesoderm; SoM, somitic mesoderm; NMPs, neuromesodermal progenitors; KPs, kidney progenitors; ExM, extraembryonic mesoderm; HSCs, hematopoietic stem cells; V-CMs, ventricular cardiomyocytes; IFT-CMs, inflow tract cardiomyocytes; OFT-CMs, outflow tract cardiomyocytes; aSHF, anterior second heart field; pSHF, posterior second heart field; PostM, posterior mesoderm; PhM, pharyngeal mesoderm; MixM, mixed mesoderm; A-CMs, atrial cardiomyocytes; AVC-CMs, atrioventricular canal cardiomyocytes; Pe, proepicardium; VP, venous pole; *, Genes known to be associated with CHDs (Yang et al. 2022); #, Direct targets of MEF2C based on MEF2C ChIP-seq data (Akerberg et al. 2019).

Closer examination of our DEG data (Supplemental Data S3) revealed two categories of DEGs. The first category consisted of key cardiomyocyte genes that were substantially down-regulated across all three heart tube segments in the Mef2c KO embryos, including genes encoding sarcomeric proteins such as Tnnt2 and Ttn, the cardiac transcription factor Nkx2–5, and the natriuretic peptide Nppa (Fig. 2G-I). These data confirm the importance of MEF2C in early cardiomyocyte fate specification, as a direct activator of contractile protein gene expression and by regulating additional TFs and functional proteins necessary for proper cardiomyocyte development. The second category of DEGs consisted of genes involved in anterior/posterior patterning of the heart tube. Specifically, genes such as Tbx5, Gata4, and Wnt2, which are normally expressed only in the posterior IFT segment of the heart tube, were not only up-regulated in the IFT-CMs, but their expression was also expanded into the V-CMs in Mef2c KO (Fig. 2G-I). Additionally, expression of the anterior OFT-specific gene Tdgf1 was completely lost in Mef2c KO (Fig. 2H-I), as previously demonstrated (Barnes et al. 2016). Consistent with this apparent posteriorization of gene expression in the heart tube, we also observed a larger proportion of IFT-CMs at E8.5 and A-CMs and AVC-CMs at E9 in Mef2c KO embryos compared to WT, with a concomitant reduction in the proportion of V-CMs and OFT-CMs (Supplemental Fig. S2K). We validated the observed changes in gene expression with whole-mount embryo fluorescence in situ hybridization, confirming that Mef2c KO leads to broad reduction in the cardiomyocyte transcriptional program and a notable posteriorization of the developing heart tube (Fig. 2J).

MEF2C regulates chromatin accessibility broadly throughout the heart tube and in a segment-specific manner

To investigate how MEF2C regulates the observed segment-specific gene expression changes, we integrated our chromatin accessibility and gene expression datasets using ArchR (Granja et al. 2021). We performed dimensionality reduction using ArchR’s latent semantic index (LSI) implementation on the snRNA-seq and snATAC-seq modalities and then plotted the cells in UMAP space using the combined dimensions (Supplemental Fig. S3A-C). Matching cell barcodes in the snRNA-seq and snATAC-seq libraries allowed us to transfer the cell type labels from the Seurat objects to the integrated ArchR datasets. As we did with the gene expression data, we subset the cardiac cell types and associated mesoderm in order to better resolve cardiomyocyte subtypes. Finally, we created pseudobulk groups consisting of WT and Mef2c KO cells for each cell type label in the subset datasets, called peaks representing accessible regions using MACS2 (Zhang et al. 2008), and identified regions of chromatin that demonstrated MEF2C-dependent accessibility at each timepoint (Supplemental Data S4).

Our initial analyses revealed few changes to chromatin accessibility at E7.75 (Supplemental Fig. S3D). This was not entirely unexpected, given that there is no discernable phenotype in the Mef2c KO embryos at this timepoint, and agrees with the relatively few changes in gene expression we observed in CMs/FHF, SHF, and JCF cells at E7.75 (Fig. 2D). Thus, we focused on the E8.5 dataset to identify the earliest substantial changes in chromatin accessibility induced by the loss of MEF2C (Fig. 3A). Using our pseudobulked data, we identified differentially accessible regions (DARs) of chromatin between Mef2c KO and WT cells in each of the subset cell types and found that, as expected, cardiomyocytes displayed the largest numbers of DARs (Fig. 3B). We also found that in each of the CM subtypes (IFT-CMs, V-CMs, and OFT-CMs), Mef2c KO CMs lost more DARs than they gained (Fig. 3B), and these lost DARs were highly enriched for MEF2 binding motifs (Supplemental Fig. S3E). These data further reinforce MEF2C’s role as an activator of the contractile gene transcriptional program.

Figure 3: MEF2C regulates chromatin accessibility broadly throughout the heart tube and in a segment-specific manner.

A) UMAP of integrated snRNA-seq and snATAC-seq data for cardiac progenitors, cardiomyocytes, and related mesoderm subtypes at E8.5 labeled by cell types determined from snRNA-seq clustering (left) and genotype/sample ID (right). B) Bar plots displaying the number of gained and lost DARs in Mef2c KO relative to WT cell types of interest at E8.5. C) Venn diagrams displaying unique and overlapping DARs in the three heart tube segments at E8.5. D) Scatter plots displaying the relationship between the Log2 fold change (Log2FC) values of the DEG and DAR analyses for all identified Peak2Gene (P2G) links in the three heart tube segments at E8.5. Dots are colored by the P2G correlation score. E-F) Genome browser tracks displaying snATAC-seq accessibility profiles at the Myh6/Myh7 (E) and Wnt2 (F) loci for the indicated pseudobulked cell types at E8.5. DARs are highlighted and the peaks are indicated by red bars (Mef2c KO relative to WT IFT-CMs). Loops indicate P2G links, colored by the P2G correlation score. V-CMs, ventricular cardiomyocytes; IFT-CMs, inflow tract cardiomyocytes; OFT-CMs, outflow tract cardiomyocytes; aSHF, anterior second heart field; pSHF, posterior second heart field; LPM, lateral plate mesoderm; PostM, posterior mesoderm; PhM, pharyngeal mesoderm; NA, cells not available in the snRNA-seq dataset; DAR, differentially accessible region; DEG, differentially expressed gene.

Interestingly, when we examined the lost DARs in Mef2c KO embryos, we found 675 regions that demonstrated lost accessibility in two or more heart tube segments, and 2,549 regions that lost accessibility in a segment-specific manner (Fig. 3C). The former are likely regions involved in the regulation of general cardiomyocyte genes, while the latter may control segment-specific gene regulation. By contrast, the gained DARs in Mef2c KO embryos were almost exclusively unique to each heart tube segment (Fig. 3C). These segment-specific DARs are interesting because they provide evidence that distinct transcriptional networks exist for each of the three heart tube segments and that each segment-specific network is uniquely altered by the loss of MEF2C.

To identify the DARs most likely to represent bona fide regulatory elements, we performed a peak-to-gene linkage (P2G) analysis on our E8.5 dataset, which identifies statistically significant correlations between the accessibility of individual peaks and the expression of potential target genes within 250 kilobases of each peak. We found 817 P2G links in the OFT-CMs, 1,201 in the V-CMs, and 2,257 in the IFT-CMs. We then intersected the P2G links with the MEF2C-dependent DEGs at E8.5 (Supplemental Data S5) and plotted each P2G link according to the fold change of the peak’s accessibility, the fold change of linked gene’s differential expression, and the correlation strength of the P2G link (Fig. 3D).

These P2G DAR-DEG correlation plots reveal a number of interesting observations. Genes such as Nkx2–5, Myh6, and Myh7, whose expression decreased throughout the heart tube in Mef2c KO embryos, are linked to DARs in all three segments (Fig. 3D). In particular, the Myh6/Myh7 locus contains multiple regions that have P2G links and clearly altered accessibility in OFT-CMs, V-CMs, and IFT-CMs (Fig. 3E). This suggests that these DARs are likely to be regulatory elements that control expression of Myh6 and Myh7 and are sensitive to the expression of Mef2c in each segment of the heart tube. By contrast, there are also P2G links between peaks with lost accessibility and decreased gene expression that are unique to the individual segments, such as links to Wnt11 in the OFT-CMs, Nppa in the V-CMs, and Cacna1c in the IFT-CMs (Fig. 3D). Furthermore, examining the links between peaks with gained accessibility and genes with increased expression in Mef2c KO reveals additional evidence for the posteriorized transcriptional program we previously described. Namely, P2G links to Tbx5 in the V-CMs, as well as Gata4 and Wnt2 in the IFT-CMs (Fig. 3D and 3F). These P2G links represent potential regulatory elements that are activated or over-activated in response to the loss of MEF2C and, at least in part, promote the increased posteriorization of the heart tube.

Each heart tube segment exhibits a distinct MEF2C-depedent developmental trajectory

Our multiomic datasets containing precisely staged embryos at different developmental timepoints provided us the opportunity to dissect how lineage trajectories unfold in WT embryos and are altered by loss of MEF2C. To this end, we applied multimodal Models for Integrated Regulatory Analysis (MIRA) (Lynch et al. 2022), which uses deep learning and probabilistic graphical models to construct sets of “topics” that represent either gene expression or peak accessibility. These topics can then be used to plot cells in UMAP space, assign cell identities, perform pseudotime analyses, and construct cell state trees (Fig. 4A).

Figure 4: Each heart tube segment exhibits a distinct MEF2C-depedent developmental trajectory.

A) Schematic of multimodal Models for Integrated Regulatory Analysis (MIRA) pipeline. B) UMAP of outflow tract lineage cells plotted by MIRA topic models labeled by cell type (top) and genotype (bottom). C) Pseudotime plot of outflow tract lineage cells. D) Lineage trajectory stream plots for outflow tract lineage cells labeled by cell type (top) and genotype (bottom). E) Stream plots displaying the flow of gene expression (top) and chromatin accessibility (bottom) topics in the outflow tract lineage cells. Examples of the top genes and TF binding motifs for dynamic topics are labeled. F-I) Same as (B-E), but for ventricular lineage cells. J-M) Same as (B-E) and (F-I), but for inflow tract lineage cells.

We trained topic models using cardiac progenitors and cardiomyocyte subtypes from both genotypes at each of our three timepoints (E7.75: CMs/FHF, SHF, JCF; E8.5: aSHF, pSHF, OFT-CMs, V-CMs, IFT-CMs; E9: OFT-CMs, V-CMs, A-CMs, AVC-CMs) and then used those models to construct joint representation UMAPs of the outflow tract, ventricular, and inflow tract lineages (Fig. 4B, 4F, 4J). We next calculated pseudotime values for cells within the UMAPs and used the results to construct cell state trajectories for each of the three lineages (Fig. 4C-D, 4G-H, 4K-L).

We observed clear differences in the fate specification of Mef2c KO cells compared to WT cells, as represented by branch points in the trajectories (Fig. 4D, 4H, 4L). Most interestingly, we noticed that the structure of these trajectories, along with the manner and timing by which Mef2c KO cells diverged from WT cells within the trajectories, was different for all three lineages. For instance, within the outflow tract lineage, Mef2c KO and WT cells proceeded similarly up until E8.5, and then diverged as development proceeded to E9 (Fig. 4D). Within the ventricular lineage, there was no branch point. Instead, Mef2c KO V-CMs from both E8.5 and E9 were located earlier in the trajectory than even E8.5 WT V-CMs (Fig. 4H). This indicates that loss of MEF2C in V-CMs resulted in a marked delay or termination of normal development rather than a divergence to an alternative cell state. The situation was most complex for the inflow tract lineage, in which there were two points of divergence between Mef2c KO and WT cells (Fig. 4L). First, a subset of Mef2c KO cells branched away from the WT trajectory as they differentiated from posterior SHF (pSHF) progenitors to IFT-CMs at E8.5. A second set of Mef2c KO cells proceeded along the WT trajectory to E9, before ultimately diverging from WT cells at the terminal AVC-CM and A-CM states. This complex branching behavior may reflect the notion that both FHF and pSHF progenitors contribute to the IFT (Meilhac et al. 2004) and suggests that MEF2C likely controls distinct GRNs within these two IFT progenitor subpopulations. Together, these data reinforce our working hypothesis that MEF2C regulates development of the heart tube in a segment-specific manner via distinct regulatory networks.

We gained further insights regarding the segment-specific regulation of heart tube development by examining the top 200 genes and the top ranked TF binding motifs enriched in the gene expression and chromatin accessibility topics (Supplemental Data S6). As WT cells diverged from Mef2c KO cells in the outflow tract lineage, their gene expression topics became enriched for canonical outflow tract markers Tdgf1 and Wnt11 (Barnes et al. 2016; Zhou et al. 2007), and their accessibility topics for MEF2 and FOS binding motifs (Fig. 4E). By contrast, the Mef2c KO cells were marked by gene expression topics enriched for Notch and Wnt signaling components such as Dlk1 and Fzd1, and transcription factors Meis1 and Hand2, and exhibited enrichment of an accessibility topic containing ISL, NKX, and TEAD binding motifs (Fig. 4E), suggesting these cells may retain a more progenitor-like state in the absence of MEF2C.

The gene expression and accessibility topics enriched along the ventricular lineage trajectory supports our interpretation that the Mef2c KO cells experience a substantial delay or termination of their normal differentiation. Notably, topics that were enriched for CM genes (e.g. Myl2, Tnnt1) and cardiac TF binding motifs (e.g. TBX, MEF2, NKX) marked the terminal end of the differentiation trajectory consisting of only WT cells (Fig. 4I). Interestingly, the mid-point of the ventricular trajectory containing E8.5 and E9 Mef2c KO cells is marked by an accessibility topic enriched for the nuclear receptor NR2F2 binding motif, a transcription factor that is restricted to the IFT in WT embryos (Pereira et al. 1999) (Fig. 4I).

In the complex branching inflow tract trajectory, terminal WT cells were enriched for topics containing expected atrial genes, such as Angpt1 and Nr2f1, and binding motifs for TBX5 and SNAI, among others (Fig. 4M, Supplemental Fig. S4A-B). By contrast, the Mef2c KO cells streamed into one of two terminal branches enriched for genes such as Wnt6, Jarid2, Tbx2, and Zeb1 and binding motifs for ERF, GATA, and TEAD factors (Fig. 4M, Supplemental Fig. S4A-B). Together, these data illustrate the lineage-specific changes in gene expression and chromatin accessibility that accompany heart tube development and the distinct impact of the loss of MEF2C on these lineage-specific programs.

Candidate regulatory elements with MEF2C-dependent chromatin accessibility display enhancer activity in zebrafish

Our integrated chromatin accessibility data (Fig. 3) and developmental trajectory analyses with MIRA (Fig. 4) highlighted changes in chromatin accessibility during heart tube development that were MEF2C-dependent and segment-specific. We hypothesized that these regions contain regulatory elements that are necessary for driving segment-specific gene expression. To identify enhancers, we applied a series of filtering criteria to our integrated multiomic data and selected 12 candidates of the highest priority to screen based on the Log2FC value of peak accessibility and linkage to genes of interest (Fig. 5A, Supplemental Data S7). We enumerated these candidate regions with the prefix “MVEB,” after the initials of authors Muncie-Vasic and Brower. Overlaying our chromatin accessibility data at these regions with published ChIP-seq datasets (Akerberg et al. 2019; Nord et al. 2013) confirmed that each of our candidates had clear and strong occupancy of both MEF2C and H3K27ac, a histone post-translational modification that marks active enhancers (Fig. 5B). Of note, one of our ventricular-specific candidates, MVEB8, was located within the Nppa/Nppb super enhancer cluster (Man et al. 2021). Furthermore, two of the candidates we identified from differential accessibility in IFT-CMs overlapped with regions found in the VISTA enhancer database (Visel et al. 2007), with demonstrated enhancer activity in the hearts of E11.5 mouse embryos (Fig. 5C-D).

Figure 5: Candidate regulatory elements with MEF2C-dependent chromatin accessibility display enhancer activity in zebrafish.

A) Schematic of selection process to identify candidate MEFC-dependent enhancers from integrated snRNA-seq and snATAC-seq data. B) Genome browser tracks displaying snATAC-seq accessibility profiles, MEF2C ChIP-seq occupancy profiles (Akerberg et al. 2019), and H3K27ac ChIP-seq occupancy profiles (Nord et al. 2013) at example loci containing IFT-specific (left), V-specific (middle), and OFT-specific (right) candidate enhancers (yellow highlights). C) Genome browser tracks displaying snATAC-seq accessibility profiles, MEF2C ChIP-seq occupancy profiles (Akerberg et al. 2019), and H3K27ac ChIP-seq occupancy profiles (Nord et al. 2013) at the Myh6/Myh7 locus, which contains two candidate enhancers with IFT-specific altered accessibility (yellow highlights, MVEB1 and MVEB2) that overlap with regions found in the VISTA Enhancer Browser database (Visel et al. 2007). D) Images of E11.5 mouse embryos from the Vista Enhancer Browser database (Visel et al. 2007) demonstrating positive enhancer activity of regions that overlap with candidates MVEB1 and MVEB2. E) Schematic of the Tol2 transgenesis assay used to screen candidate enhancers in zebrafish. Elements of this panel were created in BioRender. B, B. (2024) https://BioRender.com/h83r503. F) Representative ventral view images of Tg(cmlc2:mCherry) zebrafish embryos at 72 hours post-fertilization (hpf) injected with candidate enhancers that demonstrated positive activity in the heart. Boxed area in the representative brightfield image (left) indicates the anatomical region of interest captured in the fluorescent images. G) Schematic representation of the observed onset of enhancer activity for candidate enhancers that demonstrated positive activity in the heart. H) Representative ventral view images of Tg(cmlc2:mCherry) zebrafish embryos at 24 and 72 hpf injected with MVEB2:eGFP or MVEB6:eGFP reporter constructs. Scale bars = 100 μm.

To determine whether our candidate regions are bona fide enhancers with activity in early heart tube development, we employed Tol2 transgenesis in zebrafish (Birnbaum et al. 2012; Fisher et al. 2006), a relatively high-throughput in vivo screening system (Fig. 5E). We reasoned that because Mef2c has orthologs in zebrafish (mef2ca and mef2cb), and because there is a high degree of overlap between the key cardiac transcriptional regulators in both species (Hinits et al. 2012), there should be concordance between the activity of our candidate regions in mouse and zebrafish. Indeed, specific regulatory elements have been shown to exhibit activity in the heart of both species (Kang et al. 2016; Yan et al. 2023; Zlatanova et al. 2023).

Upon screening, seven of eleven candidates displayed enhancer activity in developing zebrafish hearts (Fig. 5F, Supplemental Fig. S5A-B), with one of our candidates being excluded due to a highly repetitive “GT” sequence. The two candidates that overlapped with regions found in the VISTA enhancer database (MVEB1, MVEB2) exhibited activity in the heart by 72 hours post-fertilization (hpf) (Fig. 5F). Additionally, two candidates from our IFT-CM data (MVEB3 and MVEB6), two from our V-CM data (MVEB7 and MVEB8), and one from our OFT-CM data (MVEB12) demonstrated specific activity in the zebrafish heart at 72 hpf (Fig. 5F, Supplemental Fig. S5A-B). To our knowledge, four of these latter candidates (MVEB3, MVEB6, MVEB7, and MVEB12) have not been previously reported and represent novel, MEF2C-dependent enhancers with activity in the developing heart. For two of these lines (MVEB6 and MVEB7), we bred our injected F0 zebrafish and screened the F1 progeny embryos for germline transmission of the reporter construct. In these stably expressing lines, we observed robust expression of reporter activity throughout the ventricles at 72 hpf (Supplemental Fig. S5C), further validating these elements as active enhancers in the embryonic heart.

Interestingly, we also observed a range in the timing of enhancer activity in the embryonic zebrafish heart. Candidates MVEB2 and MVEB6 were clearly detectable as early as 24 hpf, while the others were not active until 72 hpf (Fig. 5G-H). This variation in the onset of enhancer activity could result from differences in the timing of accessibility of these regulatory elements, different sensitivity to MEF2C levels, or cooperation with other TFs or transcriptional regulators.

Loss of MEF2C induces an overactive posteriorized gene regulatory network driven by NR2F2 and GATA4

Our data indicate that loss of MEF2C causes a posteriorized gene signature (Fig. 2H-J) and chromatin accessibility landscape (Fig. 3D, 3F) in the developing heart tube. Moreover, as previously described (Verzi et al. 2005), the expanded but aberrantly formed inflow tract that forms in Mef2c KO embryos (Fig. 1C) indicates that loss of MEF2C leads to a more complex heart tube phenotype than a simple failure of cardiac differentiation and growth. To determine what transcriptional regulators may drive the inflow tract malformation, we performed TF binding motif enrichment analysis on the regions of chromatin that gain accessibility in the Mef2c KO IFT-CMs at E8.5. We found that Nuclear Receptor (NR) and GATA motifs were among the most enriched (Fig. 6A). From this observation, we hypothesized that MEF2C co-binds certain regulatory elements with NR or GATA factors in the WT condition, and that these NR and GATA factors become free to bind different regulatory elements with loss of MEF2C. Consistent with this hypothesis, we found that nearly half of the sites of accessible chromatin that mark WT IFT-CMs and contain an NR or GATA motif also contain a MEF2 motif, whereas the vast majority of sites that gain accessibility in Mef2c KO and contain an NR or GATA motif do not (Fig. 6B, Supplemental Fig. S6A).

Figure 6: Loss of MEF2C induces an overactive posteriorized gene regulatory network that is partially rescued by reduced NR2F2 dosage.

A) TF binding motif enrichment analysis for gained DARs in Mef2c KO relative to WT IFT-CMs at E8.5. B) Pie charts showing the proportion of lost or gained DARs (Mef2c KO relative to WT) containing NR and MEF2C motifs or only NR motifs. C) Odds ratio analysis for NR2F2 target genes (Rouillard et al. 2016) amongst DEGs up-regulated in Mef2c KO IFT-CMs. p-value calculated using Fisher’s exact test. D) Ridge plot displaying module scores for up-regulated NR2F2 targets in Mef2c KO and WT heart tube segments. E) Inferred GRNs constructed for Mef2c KO and WT IFT-CMs at E8.5 and E9. Boxed region of E8.5 WT GRN is shown at higher magnification in (F). F) Schematic of in silico simulated Mef2c KO in the E8.5 WT GRN. G) Results of GRN validation displaying the high accuracy (74%) of predicted relative to measured gene expression changes at E9. H) Visualizations of subnetworks consisting of 12 cardiac TFs and the top 100 DEGs within the WT and Mef2c KO E8.5 IFT-CM GRNs. I) Visualization of direct NR2F2 interactions in the WT and Mef2c KO E8.5 IFT-CM GRNs. Direct interactions that occur upon Mef2c KO are highlighted. #, mis-regulated DEG in Mef2c KO IFT-CMs at E8.5; *, Direct target of NR2F2 (Rouillard et al. 2016). J-K) Immunofluorescent staining of cardiac Troponin T (cTnT, green) in representative E9.5 embryos (18–24 somites) collected from Mef2c^+/−;Nr2f2^+/− to Mef2c^+/− crosses. Boxed regions in (J) are shown at higher magnification in (K). Arrows point to the ventricle, which is expanded in Mef2c^−/−;Nr2f2^+/− embryos compared to Mef2c^−/− embryos. Asterisks mark the atria, which are better developed and have undergone more looping in Mef2c^−/−;Nr2f2^+/− embryos compared to Mef2c^−/− embryos. n=5 Mef2c^−/− and n=7 Mef2c^−/−;Nr2f2^+/− embryos from 7 independent litters. Scale bars = 200 μm.

These data are particularly interesting given the importance of GATA factors, and in particular GATA4, in cardiac development (Tremblay et al. 2018) and the knowledge that nuclear receptor subfamily 2 group F member 2 (NR2F2, also known as COUP-TFII) is a necessary driver of atrial development (Pereira et al. 1999; Wu et al. 2013). Using publicly available ChIP-seq datasets (He et al. 2014; Wu et al. 2013), we found NR2F2 and GATA4 occupancy at a number of the regions of chromatin that gained accessibility in the Mef2c KO IFT-CMs at E8.5 (Supplemental Fig. S6B-C), suggesting that these DARs could indeed be sites of potential NR2F2 and GATA4 regulation. To determine whether altered NR2F2 and GATA4 activity may drive the posteriorization of the heart tube, we next asked if NR2F2 and GATA4 target genes were upregulated in the inflow tract of Mef2c KO embryos. Using publicly available datasets to identify the target genes of each TF (Liska et al. 2022; Rouillard et al. 2016), we calculated the odds ratio (OR) of finding NR2F2 or GATA4 target genes compared to non-target genes amongst our up-regulated DEGs at E8.5. We found strong evidence that NR2F2 (OR = 1.52, p = 1.79e-4) and GATA4 (OR = 4.23, p < 2.2e-16) targets were enriched amongst the upregulated genes (Fig. 6C, Supplemental Fig. S6D). Next, we took the list of NR2F2 and GATA4 target genes that were upregulated (Supplemental Data S8) and calculated a module score (Tirosh et al. 2016) for the expression of these genes in our dataset. In agreement with our working model that these NR2F2 and GATA4 targets are driving the posteriorized heart tube phenotype in Mef2c KO, we found the NR2F2 and GATA4 module scores were increased in all segments of the heart tube at E8.5 (Fig. 6D, Supplemental Fig. S6E).

To understand how altered NR2F2 and GATA4 activity upon loss of MEF2C affects the entire transcriptional network, we used CellOracle (Kamimoto et al. 2023) to construct GRNs for the WT and Mef2c KO IFT-CMs at E8.5 and E9 (Fig. 6E). In brief, CellOracle uses integrated gene expression and chromatin accessibility data to infer interactions between TFs and target genes based on the accessibility of a given TF’s binding motif near a potential target gene and the correlation of expression between the TF and the potential target. To validate that these inferred CellOracle networks effectively captured the GRNs that govern inflow tract development, we simulated the knockout of Mef2c in silico in the E8.5 WT IFT-CMs network and compared the predicted changes in gene expression to the measured changes captured in our E9 dataset (Fig. 6F-G). We found that the networks correctly predicted the gene expression changes with 74% accuracy, even with a relatively permissive DEGs cutoff of Log2FC >= 0.75 (Fig. 6G). As expected, the prediction accuracy increased with more stringent Log2FC cutoffs, and vice versa (Supplemental Data S9).

We next visualized the subnetworks of 12 core cardiac TFs and the top 100 DEGs in the WT and Mef2c KO inflow tracts (Fig. 6H). These subnetworks revealed that loss of MEF2C causes a network-wide reorganization of gene regulation. First, we observed that MEF2C may be critical for coordinating cooperative regulation by multiple cardiac TFs. For instance, in the WT network, MEF2C, MEF2A, and TBX20 all have an inferred activating effect on Ttn expression. However, in the Mef2c KO network, both MEF2A and TBX20 lose their activating interaction with Ttn. Similarly, in the WT network, MEF2C, MEF2A, and NKX2–5 are all inferred activators of Myh6, but in the Mef2c KO network, MEF2A and NKX2–5 are no longer predicted to activate Myh6 expression. These data suggest that MEF2C is required for cooperative activation with other cardiac TFs in the developing heart tube, similar to what was shown to occur during cardiomyocyte reprogramming in vitro (Stone et al. 2019). Additionally, we found that loss of MEF2C may alter the scope and strength of interactions between other cardiac TFs and their gene targets. For example, in the WT network, GATA4 and MEIS2 exhibit competitive regulation of Fbxo32, which encodes a muscle-specific ubiquitin-E3 ligase that is critical for sarcomeric function and associated with dilated cardiomyopathy (Al-Yacoub et al. 2016; Ghasemi et al. 2022). This competitive relationship is maintained in the Mef2c KO network, but the interactions are strengthened in both directions. Finally, we observed that the activity of NR2F2 is substantially altered in the Mef2c KO network, with a larger number of gene targets and apparent strengthening of both activating and inhibitory interactions. Most notably, the activating interaction between NR2F2 and Angpt1 expression is much stronger in the Mef2c KO network than in the WT network. Interestingly, the activation of Nr2f2 itself by TBX5 is lost upon Mef2c KO.

Next, we further narrowed in on the direct interactions between NR2F2 and GATA4 and their target genes in the WT and Mef2c KO inflow tract networks (Fig. 6I, Supplemental Fig. S6F). From these analyses we observe similar behavior for both TFs – approximately half of the WT interactions are maintained in the Mef2c KO, while nearly twice as many new direct interactions are established. This is consistent with our working model that in response to loss of MEF2C, aberrant NR2F2 and GATA4 activity regulate additional genes that drive the inflow tract phenotype in the posterior of the heart tube.

Reduction of NR2F2 dosage partially rescues the heart tube phenotype in Mef2c KO embryos

Given our observation that loss of MEF2C causes aberrant NR2F2 activity in the IFT-CM GRNs, we theorized that reducing the dosage of NR2F2 may rescue the inflow tract phenotype in Mef2c KO embryos. To test this possibility, we generated Mef2c^+/−;Nr2f2^+/− male mice and crossed them to Mef2c^+/− females. Although Nr2f2^+/− embryos have no apparent heart tube phenotype at E9.5 (Fig. 6J, Supplemental Fig. S7A), pups of this genotype were recovered at lower numbers than expected Mendelian ratios (Pereira et al. 1999), suggesting that Nr2f2 is a dosage-sensitive gene. We observed a notable partial rescue in ventricle and inflow tract development at E9.5 (18–24 somites) in the Mef2c^−/− ;Nr2f2^+/− embryos compared to Mef2c^−/− (Fig. 6J-K), although the extent of the rescue varied (Supplemental Fig. S7B-C). The Mef2c KO single-ventricle phenotype persisted in Mef2c^−/−;Nr2f2^+/− embryos, but these ventricles were larger and more expanded than in Mef2c^−/− embryos. Moreover, the posterior structures of the heart tube in Mef2c^−/−;Nr2f2^+/− embryos more closely resembled WT inflow tracts with expanded and looping prospective atria, compared to the completely disrupted inflow tract in Mef2c^−/− embryos (Fig. 6J-K). Finally, we asked whether any of the differentially expressed genes in Mef2c^−/− embryos (Fig. 2J) were rescued in the Mef2c^−/−;Nr2f2^+/− embryos. Using fluorescence in situ hybridization, we found that both Ttn and Myl2 exhibited partially rescued patterns of expression in Mef2c^−/−;Nr2f2^+/− embryos (Supplemental Fig. S7D-E). More than half (3/5) of the Mef2c^−/−;Nr2f2^+/− embryos exhibited increased expression of Ttn in the ventricle and inflow tract, as compared to Mef2c^−/− embryos (Supplemental Fig. S7D). Similarly, two of five Mef2c^−/−;Nr2f2^+/− embryos demonstrated rescued expression of Myl2 in the ventricle, whereas all Mef2c^−/− embryos had no detectable Myl2 expression in the heart tube (Supplemental Fig. S7E). Notably, the individual embryos that most clearly exhibited rescued gene expression also had the most rescued morphology, with more expanded ventricles and apparent looping of the heart tube. This partial rescue of the ventricle and inflow tract in Mef2c^−/−;Nr2f2^+/− embryos is evidence that the Mef2c KO phenotype in the posterior heart tube is driven, at least in part, by an altered gene regulatory network with increased and aberrant NR2F2 activity.

DISCUSSION

In this study we demonstrated that MEF2C controls complex, segment-specific GRNs that regulate gene expression and chromatin accessibility in the developing linear heart tube. We found that loss of MEF2C causes a posteriorization of the heart tube and revealed novel segment-specific regulatory elements with enhancer activity. These results address the long-standing question of how a given TF can play distinct regulatory roles in different cell types and at particular developmental stages. For instance, decades of work have established that FHF progenitors give rise to the left ventricle and atria, while the SHF is subdivided into the aSHF, which produces the outflow tract and right ventricle, and the pSHF, which contributes to the atria (Devine et al. 2014; Evans et al. 2010; Kelly et al. 2014; Lescroart et al. 2014; Meilhac et al. 2004). Although these progenitor cell types have different molecular signatures and give rise to individual structures with specific morphology and function, they largely rely on a shared set of cardiac TFs to regulate their development. Our data illustrate that these core cardiac TFs operate in unique GRNs for each of the inflow tract, ventricular, and outflow tract lineages in the developing heart tube. We found that distinct chromatin landscapes and regulatory elements are important determinants for setting up these GRNs, but other embryological phenomena are likely to be involved, such as morphogen signaling from differing neighbor cell types, heterochronic timing of differentiation (Rowton et al. 2022), and mechanical phenomena introduced by tissue morphogenesis and the initiation of cardiomyocyte beating (Barriga et al. 2018; Caldarelli et al. 2024; Tyser et al. 2016).

Using the GRNs inferred from our multiomic data, we identified a genetic interaction between Mef2c and Nr2f2 in the inflow tract of the developing heart tube. Previous work from our lab showed that TF occupancy at specific loci is highly dependent upon both TF-TF and TF-DNA affinities, and that TF-TF interactions not only serve to regulate gene expression, but prevent the redistribution of one or both TFs to exogenous loci (Luna-Zurita et al. 2016). The results we present here lend further support to those concepts. By removing MEF2C, the set of possible TF-TF interactions is limited and causes altered activity of partner TFs, including NR2F2 and GATA4. Our comparison of the WT and Mef2c KO GRNs revealed that the transcriptional networks regulating early heart development are highly sensitive to TF perturbations, with loss of MEF2C causing a reorganization of heart tube GRNs that is not limited to direct MEF2C connections, and reduction of NR2F2 dosage capable of partially rescuing the heart tube malformations. Thus, these GRNs may be implemented as a powerful tool for predicting the network-wide consequences of genetic mutations that underlie CHDs, potentially leading to unexpected molecular targets for intervention and correction.

Our approach provides a generalizable framework for integrating snRNA-seq and snATAC-seq data to build GRNs that provide specific and comprehensive models of gene regulation. By constructing these GRNs using both WT and mutant data, we were able to immediately test predictions made from the WT networks and validated the inferred regulatory interactions. These tools also allow specific networks to be built for particular cell or tissue types, provided there are enough cells in the group of interest to generate robust inferences. Moreover, the experimental mutant data revealed how the overall networks were altered – both in ways that would and would not be predicted from the WT networks. These unexpected differences help to illuminate indirect or transactivating gene regulation. Discoveries from these networks lead to many new hypotheses about regulatory elements, TF interactions, DNA binding, and chromatin organization. As these hypotheses are experimentally tested, the GRNs can be refined, leading to higher-fidelity models with greater predictive power. Such higher-fidelity networks will yield important new insights into the transcriptional regulation of developmental processes and will have the ability to reveal precisely how these processes are altered in the context of disease-causing mutations.

MATERIALS AND METHODS

Mouse Models

All mouse studies were performed in strict compliance with the UCSF Institutional Animal Care and Use Committee. Mice were housed in a standard 12-hour light/dark animal husbandry barrier facility at the Gladstone Institutes. The Mef2c^+/− allele used was the same as the original generated by Lin et al. (Lin et al. 1997). The Smarcd3-F6-eGFP reporter allele was generated in our lab and described previously (Krup et al. 2023). The COUP-TFII lacZ knock-in allele (abbreviated here as Nr2f2^+/−) was obtained from the Mutant Mouse Resource & Research Centers and was originally generated by Takamoto et al. (Takamoto et al. 2005). The Mef2c^+/− and Nr2f2^+/− lines were maintained in the C57BL6/J background (Jackson Laboratory #664). The Smarcd3-F6-eGFP allele was crossed into our Mef2c^+/− line from a mixed background of C57BL6/J (Jackson Laboratory #664) and CD-1 (Charles River #022).

Timed Embryo Dissections and Genotyping

To achieve timed matings, male and female mice were housed together in the evening and pregnancy was assessed by vaginal plug the following morning. Gestational stage was determined relative to noon on the day of plug detection, defined as day E0.5. Embryos were dissected and, at later stages when yolk was present, also de-yolked, in ice-cold PBS (Life Technologies, 14190250) with 1% fetal bovine serum (FBS; Thermo Fisher Scientific, 10439016) on ice. The posterior trunk was removed by microdissection for all embryos at E8.5 or later that were collected for imaging (IF or FISH via HCR) to ensure unobstructed views of the heart tube. PCR genotyping was performed on all embryos, details are included in the Supplemental Methods.

Whole-Mount Embryo Immunostaining and Imaging

Embryos collected for immunostaining were fixed for 1 h at room temperature in 4% paraformaldehyde (PFA) freshly diluted from 16% weight/volume PFA aqueous solution (Thermo Fisher Scientific, 043368–9M) in PBS (Life Technologies, 14190250). Embryos were stored at 4°C for up to 1 month prior to immunostaining. To permeabilize and block non-specific antigens, embryos were incubated at 37°C for 2 h with gentle rotation in a blocking buffer consisting of PBS (Life Technologies, 14190250), 5% normal donkey serum (Sigma Aldrich, S30-M), and 0.8% Triton X-100 (Sigma Aldrich, X-100–5ML). This was then removed and replaced with primary antibodies diluted in the same blocking buffer and incubated overnight at 37°C with gentle rotation. The next day, primary antibody solution was removed and embryos were washed 3× 30 min in the same blocking buffer at 37°C with gentle rotation. Secondary antibodies and DAPI (Abcam, ab228549, used at 1 μg/mL) were added in a blocking buffer consisting of PBS (Life Technologies, 14190250), 5% normal donkey serum (Sigma Aldrich, S30-M), and 0.4% Triton X-100 (Sigma Aldrich, X-100–5ML), and incubated for 3–4 h at 37°C with gentle rotation. Secondary antibody solution was removed and embryos were washed 3× 30 min in PBS (Life Technologies, 14190250) with 0.1% Tween-20 (Thermo Fisher Scientific, J20605.AP) at 37°C with gentle rotation. Embryos were imaged in PBS using an upright epifluorescence microscope (Leica MZFLIII, Leica DFC 3000G, Lumen Dynamics XCite 120LED) and acquisition software LASX (Leica). Primary antibodies used were sheep polyclonal MEF2C (R&D Systems, AF6786, used at 1:250) and rabbit polyclonal cardiac Troponin T (Thermo Fisher Scientific, 15513–1-AP, used at 1:250). Secondary antibodies used were Donkey anti-Sheep IgG (H+L) Cross-Adsorbed Secondary Antibody, Alexa Fluor^™ 647 (Thermo Fisher Scientific, A-21448, used at 1:1000) and Donkey anti-Rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor^™ 555 (Thermo Fisher Scientific, A-31572, used at 1:1000).

Whole-Mount Embryo Fluorescence In Situ Hybridization via Hybridization Chain Reaction and Imaging

The protocol for in situ hybridization via hybridization chain reaction (HCR^™) was adapted from the optimized manufacturer’s protocol for whole-mount mouse embryos (Molecular Instruments), and details can be found in the Supplemental Methods. HCR probes used in this study were ordered from the Molecular Instruments catalog, when possible, or custom-designed and manufactured by Molecular Instruments. Hybridization probes used were: Tnnt2-B3 (4 nM), Ttn-B2 (4 nM), Nppa-B2 (20 nM), Nkx2.5-B1 (20 nM), Tbx5-B1 (20 nM), Gata4-B1 (20 nM), Wnt2-B2 (20 nM), Tdgf1-B2 (20 nM), Myl2-B1 (40 nM). Amplification probes were all used at 60 nM: B1-Alexa Fluor^™ 647, B2-Alexa Fluor^™ 546, B2-Alexa Fluor^™ 647, B3-Alexa Fluor^™ 488.

Embryo Preparation for 10x Multiome snRNA-seq and snATAC-seq

For E7.75 embryos, whole embryos were dissected and harvested for single-nucleus library generation. For E8.5 and E9 embryos, the head folds and posterior trunk were removed by microdissection prior to harvesting for library generation to enrich the relative capture of cardiac cell types. Two embryos were collected per genotype per embryonic stage. Following dissection and rapid PCR genotyping, embryo samples selected for the multiome experiment were incubated in 200 μl TrypLE Select (Thermo Fisher Scientific, 12563–011) for 5 min at 37°C, triturated gently by pipetting up and down, and then incubated an additional 3 min at 37°C. The dissociated cell suspension was quenched with 600 μl of PBS (Life Technologies, 14190250) with 1% FBS (Thermo Fisher Scientific, 10439016), singularized by passage through the 35 μm mesh of a 5 mL Falcon^™ Round-Bottom Polystyrene Test Tube with Cell Strainer Snap Cap (Thermo Fisher Scientific, 08–771-23), pelleted by centrifugation at 300 g for 5 min at 4°C, and resuspended in 50 μl of PBS (Life Technologies, 14190250) with 0.04% bovine serum albumin (BSA, Sigma Aldrich, A1595). At this stage, the manufacturer’s protocol for Nuclei Isolation for Single Cell Multiome ATAC + Gene Expression Sequencing, Appendix, Low Cell Input Nuclei Isolation (10x Genomics, CG000365 Rev B) was followed exactly to prepare nuclei for library generation.

Generation of snRNA-seq and snATAC-seq Libraries and Sequencing

Libraries for snRNA-seq and snATAC-seq were prepared according to the manufacturer’s protocol for Chromium Next GEM Single Cell Multiome ATAC + Gene Expression (10x Genomics, CG000338 Rev B), using the 10x Genomics Chromium controller, Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Bundle (10x Genomics, PN-1000283), Chromium Next GEM Chip J Single Cell Kit (10x Genomics, PN-1000234), Single Index Kit N Set A (10x Genomics, PN-1000212), and Dual Index Kit TT Set A (10x Genomics, PN-1000215). Nuclei isolated from multiome samples were processed directly into the snATAC-seq transposition reaction and then captured in 10x Genomics GEMs via the Chromium Controller. GEMs were stored at −80°C for up to 4 weeks, allowing for collection of replicate embryos of each genotype at each development stage. All subsequent library preparation steps were performed together for all embryos of a given developmental stage to reduce the likelihood of batch artifacts. A targeted maximum recovery of 10,000 cells per sample were loaded onto the 10x Genomics Chromium controller instrument and each sample was indexed with a unique sample identifier (Single Index Kit N Set A for snATAC-seq libraries, Dual Index Kit TT Set A for snRNA-seq libraries). Final libraries were quality-controlled using an Agilent Bioanlyzer instrument with Agilent High Sensitivity DNA Kit (Agilent, 5067–4626). The DNA concentration of each library was measured using KAPA Library Quantification qPCR Kit (Thermo Fisher Scientific, 50–196-5234), and then libraries were pooled and sequenced on NovaSeq6000 S4 lanes (Illumina). All snRNA-seq libraries were sequenced to a depth of at least 20,000 mean raw aligned reads per cell and snATAC-seq libraries were sequenced to a depth of at least 35,000 mean raw aligned read pairs per cell (most >60,000).

Processing of Raw Sequencing Data

Raw sequencing reads were processed using the 10x Genomics Cell Ranger Arc v2.0.0 pipeline. Reads were demultiplexed using cellranger-arc mkfastq and aligned with cellranger-arc count to the mm10 reference genome containing an additional sequence for eGFP to map reads of the Smarcd3-F6-eGFP reporter transcript.

Analysis of snRNA-seq and snATAC-seq Data

Gene expression data were first analyzed using Seurat v4.3.0 in R (Hao et al. 2021). Chromatin accessibility and integrated multiome data were analyzed using the ArchR software package v1.0.2 in R (Granja et al. 2021). Multimodal models for Integrated Regulatory Analysis (MIRA) (Lynch et al. 2022) was used to examine dynamic changes in gene expression and chromatin accessibility in WT and Mef2c KO embryos across the time course of developmental stages that we collected (E7.75, E8.5, E9). Details for each of these bioinformatics pipelines can be found in the Supplemental Methods.

Cloning of Reporter Constructions for Zebrafish Transgenesis

To identify candidate enhancer regions to clone, we utilized our E8.5 ArchR dataset containing integrated chromatin accessibility and gene expression data. We identified all peaks that lost accessibility in the Mef2c KO (3,224), then filtered to the peaks that demonstrated altered accessibility in only one of the three heart tube segments (2,549), and that contained a MEF2 binding motif (1,059). Next, we utilized ArchR’s Peak2Gene functionality and filtered down to peaks correlated with DEGs in the Mef2c KOs (130) and were not located in promoter regions (119). Finally, from this list of 119 candidates, we selected 12 of the highest priority to screen based on the Log2FC value of the change in peak accessibility and linkage to genes of interest.

To generate the Tg(MVEB:egfp)^sfc26 zebrafish lines, 650–900 nucleotide fragments containing the candidate enhancers were cloned by PCR from mouse genomic DNA. Candidate nucleotide sequences and the primers used to amplify them are listed in Supplemental Data S7. The resulting fragments were cloned into the Tol2 transgenic vector E1b-eGFP-Tol2 (Birnbaum et al. 2012) (RRID: Addgene_37845) using the In-Fusion Cloning system (Takara Bio). All plasmid sequences were confirmed by long-read sequencing before zebrafish injection (Plasmidsaurus). We were unable to successfully clone one of our twelve candidates (MVEB10) due to a highly repetitive “GT” sequence, so our final screen consisted of eleven candidate enhancers.

Zebrafish Transgenesis Injections, Screening, and Imaging

All zebrafish experiments were reviewed and approved by the UCSF Institutional Animal Care and Use Committee and were performed in accordance with the Public Health Service Policy on the Humane Care and Use of Laboratory Animals. Zebrafish were raised under standard laboratory conditions at 28°C. The outbred wildtype zebrafish EKW (Danio rerio-Ekwill) and Tg(cmlc2:mCherry)^s890 lines have been described previously (Zhang et al. 2013). One-cell zebrafish embryos obtained from Tg(cmlc2:mCherry)^s890 and wildtype EKW (Ekwill strain) parents were injected with 1nl of injection mix containing 15pg of candidate-enhancer-eGFP and 15pg of ß-actin-BFP reporter plasmids along with 30pg of Tol2 mRNA. The ß -actin promoter drives expression of BFP strongly in somite muscles and was used as a positive injection control. As a negative control, we also injected and screened embryos with an empty reporter construct containing only the minimal E1b promoter and eGFP. Injected embryos were screened at 24 and 72 hours post-fertilization (hpf) for eGFP and BFP expression. Representative embryos were imaged for eGFP expression as a readout of candidate enhancer activity and for mCherry expression as a myocardial marker. Enhancer activity was scored as a percentage of embryos expressing eGFP in the heart compared to the total number of BFP-positive embryos. A minimum of at least 50 BFP+ embryos were screened per construct, except for MVEB1, for which only 17 BFP+ embryos were obtained. The exact numbers screened for each construct are displayed in Supplemental Fig. S5B.

Comparative Analyses with Published Datasets

NR2F2 and GATA4 target genes were downloaded from the Harmonizome (Rouillard et al. 2016) and TFLink (Liska et al. 2022) databases, respectively. Fisher’s exact test was used to determine the Odd’s Ratios and p-values for identifying NR2F2 and GATA4 targets among genes upregulated in Mef2c KO IFT-CMs at E8.5.

Publicly available ChIP-seq datasets were downloaded from the Gene Expression Omnibus (GEO) in wig or bigwig format for MEF2C (GSE124008) (Akerberg et al. 2019), H3K27ac (GSE52386) (Nord et al. 2013), GATA4 (GSE52123) (He et al. 2014), and NR2F2 (GSE46498) (Wu et al. 2013). Datasets provided in wig format were converted to bigwig format using the Galaxy (The Galaxy Community 2024) web tool wigToBigWig. CrossMap (Zhao et al. 2014) was used together with chain files from the UCSC Genome Browser (Raney et al. 2023) to lift genome annotations from mm9 to mm10 reference genome assemblies, and vice versa.

For the browser tracks in Fig. 5B-C, the MEF2C and H3K27ac ChIP-seq bigwigs were lifted from mm9 to mm10, the Galaxy (The Galaxy Community 2024) web tool bigwigAverage was used to generate a single average MEF2C and H3K27ac occupancy track from the individual replicate tracks, and the results were displayed along with our pseudobulked snATAC-seq data using the IGV software package (Robinson et al. 2011).

For the heat maps displayed in Supplemental Fig. S6B-C, we first created Browser Extensible Data (BED) format files for the gained DARs in Mef2c KO IFT-CMs containing NR or GATA motifs. We then lifted these BED regions from mm10 to mm9 and used the Galaxy (The Galaxy Community 2024) web tools computeMatrix and plotHeatmap with the downloaded NR2F2 and GATA4 ChIP-seq data to visualize NR2F2 and GATA4 occupancy at the DARs. At the computeMatrix step, missing data was converted to zeroes. At the plotHeatmap step, kmeans clustering was used to generate two clusters of DARs (effectively “occupied” and “not occupied”) and these were sorted by decreasing mean value prior to being plotted.

Gene Regulatory Network Inference and Analysis

We used CellOracle (Kamimoto et al. 2023) to infer GRNs for WT and Mef2c KO IFT-CMs at E8.5 and E9. Details of the GRN inference process and downstream analyses can be found in the Supplemental Methods.

Phenotypic Analysis of Mef2c^−/− and Mef2c^−/−; Nr2f2^+/− Embryos

We conducted comparative phenotypic analysis of Mef2c^−/− and Mef2c^−/−; Nr2f2^+/− embryos by performing immunofluorescent staining of cardiac Troponin T and analyzing the apparent heart tube phenotypes, initially unblinded to embryo genotype. We determined that the heart tube phenotypes ranged in apparent severity and fell into roughly three categories, with Category 1 being the most malformed (severely hypoplastic ventricle, no looping) and Category 3 the closest to normal morphology (more expanded ventricular chamber, some apparent looping). We then blinded ourselves to the embryo genotypes and re-examined the embryos, placing each into the appropriate category of severity.

Data and Code Availability

Raw and processed data for snRNA-seq and snATAC-seq datasets has been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE280587. Code used for multiomics analyses, GRNs, and plot generation is available in the GitHub repository associated with this manuscript: https://github.com/jmmuncie/Muncie-Vasic_2025.