The lncRNA Hand2os1/Uph locus orchestrates heart development through regulation of precise expression of Hand2

Long noncoding RNAs (lncRNAs) have been implicated as an important layer of regulatory information in fine-tuning the spatiotemporal expression of pleiotropic developmental loci in their chromatin neighborhood, thereby modulating cell fate determination in various biological processes (Han et al., 2018; Luo et al., 2016; Morris and Mattick, 2014; Pauli et al., 2011; Ponjavic et al., 2009; Yin et al., 2015). Heart formation is tightly regulated during mouse embryogenesis and involves restriction of mesodermal precursor cells to the cardiac lineage and the subsequent formation of a primitive heart tube, which, in turn, undergoes: looping; formation of the outflow tract and atrial and ventricular cavities; and septation to form the mature four-chambered heart (Bruneau, 2008; Olson and Schneider, 2003). Proper commitment of cardiac lineages during this complex process is required for normal development and function of the heart (Brade et al., 2013). Several lncRNAs have been reported to have roles in regulating heart development and function. For example, depletion of Chast/Wisper and overexpression of Mhrt, Tincr or Carel protected the heart from hypertrophy in response to pressure overload following transverse aortic constriction surgery (Cai et al., 2018; Han et al., 2014b; Micheletti et al., 2017; Shao et al., 2017; Viereck et al., 2016). Inhibition of Fendrr led to embryonic lethality around E13.5 with cardiac hypoplasia (Grote et al., 2013).

The lncRNA Hand2os1 (also named Uph or lncHand2) is divergently positioned at −123 bp upstream of the transcription start site (TSS) of HAND2 (Anderson et al., 2016; Wang et al., 2018). (Throughout this paper, we use the non-standard style of HAND2, etc. to distinguish it from the lncRNA.) HAND2, a transcription factor that promotes ventricular cardiomyocyte expansion and cardiac reprogramming, is a crucial regulator of embryonic heart development (McFadden et al., 2005; Song et al., 2012; Srivastava et al., 1997). Cardiac expression of HAND2 is initially detected in the cardiac crescent at E7.75, continues throughout the linear heart tube at E8.5, is specifically enhanced in the developing right ventricle (RV) and outflow tract (OFT) until E9.5-E10.0, and is downregulated in the cardiac mesoderm but maintained in the neural crest-derived aortic arch arteries (Srivastava et al., 1997; Tamura et al., 2014).

Precise expression of HAND2 is essential for normal heart morphogenesis and function. It is tightly regulated at the transcriptional level by a network of cardiac transcription factors and upstream enhancers, and at the post-transcriptional level by microRNAs (Bruneau, 2005; Dirkx et al., 2013; McFadden et al., 2005, 2000; Zhao et al., 2007, 2005). Constitutive HAND2 knockout (KO) in mice display right ventricle hypoplasia and embryonic lethality at E10.5 (Srivastava et al., 1997). Conditional ablation of HAND2 in specific sets of cardiac cells leads to embryonic lethality at various stages prior to embryonic day E15.5 (summarized in Table S1) (Holler et al., 2010; Morikawa and Cserjesi, 2008; Morikawa et al., 2007; Tsuchihashi et al., 2011; VanDusen et al., 2014). Overexpression of HAND2 in transgenic mouse models also leads to heart development defects and malfunctions (Table S1) (Dirkx et al., 2013; Togi et al., 2006).

In a polyA knock-in (KI) mouse model of Uph/Hand2os1 reported previously, termination of transcription by insertion of a triple polyadenylation (polyA) stop sequence into intron 1 of Uph (−644 bp upstream of the HAND2 TSS) abolished HAND2 expression and led to failed right ventricle formation and lethality at E10.5, partially phenocopying HAND2 KO mice (Anderson et al., 2016). It was concluded that transcription of Uph/Hand2os1 is an essential switch for the activation of HAND2 and the onset of heart morphogenesis (Anderson et al., 2016). However, the functional role of Hand2os1 transcripts and the Hand2os1 DNA sequences in the heart remains elusive.

To delineate the role of Hand2os1 in heart development and function, we generated three deletion alleles of Hand2os1 in mouse (Han et al., 2018). Full-length deletion of the entire Hand2os1 sequence (Hand2os1^F/F KO) led to dysregulated cardiac gene expression programs, septum lesion, heart hypoplasia and perinatal death, which are reminiscent of congenital heart diseases. A short distal deletion at the 3′ end of the Hand2os1 locus (Hand2os1^D/D KO) caused severe contraction defects in adult heart that progressively worsened with increasing age. By comparison, short deletion of the 5′ promoter and exons of Hand2os1 (Hand2os1^P/P KO) effectively diminished Hand2os1 expression, but failed to produce discernable heart phenotypes in either embryos or adults. These results indicate that the Hand2os1 DNA locus primarily controls heart development and function. To our surprise, cardiac expression of HAND2 was sustained in all three Hand2os1 KO mouse models we generated, in sharp contrast to the abolished expression of HAND2 in the Uph/Hand2os1 polyA KI embryos (Anderson et al., 2016). Importantly, single cell transcriptomic analysis revealed subtle yet prevalent upregulation of HAND2 and concordant global gene expression changes in subsets of cardiac cells of Hand2os1^F/F embryos. Altogether, these results illustrate a fine-tuning, yet crucial, role for the lncRNA Hand2os1 locus in restricting the precise spatial expression of HAND2, through which Hand2os1 modulates cardiac lineage development and heart function. This study reveals the unexpected complexity of lncRNA function in vivo, and also emphasizes the usage of complementary genetic and single cell approaches to delineate the primary molecular effects and elucidate physiological functions of an lncRNA in animals.

Hand2os1 and HAND2 are divergently transcribed from the shared core promoter sequences, and are highly enriched in the heart compared with other tissue types analyzed (Fig. 1A, Fig. S1A,B). Within the heart, Hand2os1 and HAND2 exhibit an inverse expression pattern during embryonic heart development and postnatal growth (Fig. S1C). We estimated that the abundance of Hand2os1 transcripts is ∼15-40 molecules per cell, relatively 7- to 23-fold lower than that of HAND2 transcripts (Fig. S1C). The DNA sequence of Hand2os1 is 17 kb in length and encompasses a super-enhancer element, and branchial arch (BA) and cardiac enhancers annotated previously (Fig. 1A) (McFadden et al., 2000; Yanagisawa et al., 2003). In E12.5 embryonic hearts, the Hand2os1 locus as well as HAND2 and its downstream regions harbor multiple DNase I hypersensitive sites (DHS, E11.5), and show strong binding signals of active histone H3K4me3 and H3K27ac marks, and RNA polymerase II and master transcription regulators of cardiac development, including GATA4, NKX2-5 and HAND2 (E10.5) itself (Fig. 1A) (He et al., 2014; Laurent et al., 2017; Ye et al., 2015; Yue et al., 2014). This suggests a possible involvement of multiple enhancers in regulating HAND2 expression.

To remove Hand2os1 transcription/transcripts with minimal manipulation of the genome, we first generated two mouse models carrying short genomic deletions at the 5′ or 3′ end of the Hand2os1 locus, keeping the cardiac and branchial arch enhancers intact (Fig. 1A) (Han et al., 2018). In the 5′ proximal knockout allele (Hand2os1^P KO), we deleted a 1 kb DNA sequence covering the core promoter and the first two exons of Hand2os1 (Fig. 1A) (Han et al., 2018). The deletion starts at −61 bp and −62 bp upstream of the TSSs of Hand2os1 and HAND2, respectively. To avoid any direct effect of promoter alteration on HAND2 expression, we generated a 3′ distal knockout allele (Hand2os1^D KO) by deleting a 2.7 kb DNA sequence that spans exons 4 and 5 of Hand2os1, and is located 13 kb upstream of the HAND2 TSS (Fig. 1A) (Han et al., 2018). Both Hand2os1^P/P and Hand2os1^D/D mice from heterozygotes crosses were born at the expected Mendelian ratio and had no overt morphological defects in the heart (Fig. 1B, Fig. S1A; data not shown).

Levels of truncated Hand2os1^P transcripts lacking exons 1 and 2 were substantially downregulated to ∼10-17% compared with heterozygous and wild-type littermates (Fig. 1C, Fig. S1A). Analysis of pre-mature transcripts using intronic primers suggested residual ∼8-32% of Hand2os1 transcription remaining in E12.5 embryonic hearts of Hand2os1^P/P mice (Fig. S1A,D). Hand2os1^P/P KO mice therefore provide a partial loss-of-function model. However, cardiac expression of HAND2 was not altered in Hand2os1^P/P mice (Fig. 1C, Fig. S1A).

Hand2os1^D/D mice expressed a mutant Hand2os1 RNA that lacks exons 4 and 5 with 67% of its sequence remaining (Fig. S1A). Embryonic expression of HAND2 in Hand2os1^D/D hearts was not affected (data not shown). However, Hand2os1^D/D adult cardiomyocytes showed moderate but significant increases of both Hand2os1 and HAND2 transcripts (∼53% and ∼34%, respectively) (Fig. 1D). It has been reported that aberrant upregulation of HAND2 in the postnatal heart contributes to pathological myocardial hypertrophy and heart failure (Dirkx et al., 2013). Interestingly, Hand2os1^D/D, but not Hand2os1^P/P, mice progressively developed heart contraction defects at 6-10 weeks, with a ∼10-30% decrease in fractional shortening (Fig. 1E; Table S2). Consistently, many genes involved in heart development, cardiac muscle contraction and the cell cycle were dysregulated in Hand2os1^D/D cardiomyocytes (Fig. S1E,F; Tables S3 and S4). These results suggest that the Hand2os1 locus may exert a complex, pleiotropic influence on HAND2 expression and heart physiology.

Next, to rule out the possibility that residual activities of Hand2os1 might promote HAND2 expression and heart morphogenesis in the promoter and distal KO mouse models, we deleted a 17 kb sequence covering the entire Hand2os1 genomic region to completely eliminate Hand2os1 expression (Fig. 1A, Fig. S2A) (Han et al., 2018). The deletion starts at −59 bp and −64 bp upstream of Hand2os1 and HAND2 TSSs, respectively, and encompasses the super-enhancer and two known enhancers of HAND2 expression (Fig. 1A). This mutant allele is designated as Hand2os1 full-length knockout (Hand2os1^F KO).

Heterozygous Hand2os1^F/+ intercrosses failed to produce viable homozygous offspring (0 out of 77 pups) at the weaning stage (Fig. 2A, Fig. S1A). Viable Hand2os1^F/F embryos were observed during mid-gestation at the expected Mendelian frequency until E16.5. Hand2os1^F/F newborns became cyanotic and invariably died shortly after birth (Fig. 2B). Gross morphological examination of hearts revealed abnormal blood coagulation and fatal thrombosis in Hand2os1^F/F newborns (Fig. 2C). Macroscopically, the most severe phenotype in homozygous embryonic hearts is the presence of a septum lesion (8 out of 12) (Fig. 2D). Right ventricular (RV) hypoplasia was also frequently observed with significantly decreased chamber volume (10 out of 12) and slightly reduced thickness of the compact myocardium of the right ventricle (3 out of 12) in Hand2os1^F/F mutant hearts (Fig. 2D). These defects are reminiscent of congenital heart diseases and, in combination, may provide a morphological explanation for the heart failure of Hand2os1^F/F mice in response to increased demand for cardiac output and stress at birth (see the Discussion).

Notably, Hand2os1^F/F newborns had cleft palate (Fig. S2B), resembling the craniofacial defects observed in branchial arch enhancer KO mice, which reportedly failed to suckle and died with an empty stomach 24 h after birth (Yanagisawa et al., 2003). As Hand2os1^F/F pups died much earlier, within hours of birth, we reasoned that the suckling defect is not the cause of death in these animals. It has been reported that loss of HAND2 in neural crest lineage mediated by WNT1-Cre led to misalignment of aortic arch arteries and embryonic lethality before E15.5 (Morikawa and Cserjesi, 2008). In comparison, Hand2os1^F/F embryos showed normally developed aortic arch arteries (Fig. S2C), and no gross abnormalities in other organs, including liver and lung (Fig. S2D). Moreover, Hand2os1^F/F pups showed normal floating lungs in a buoyancy test (data not shown), excluding the possibility of respiratory failure as the cause for their immediate death upon birth.

To reveal transcriptional changes that underlie the morphological defects and perinatal lethality, we performed RNA-seq analysis of E11.5 embryonic hearts and E16.5 ventricles isolated from littermates from Hand2os1^F/+ intercrosses. Interestingly, a subset of gene programs pertaining to cardiac muscle contraction, such as ACTA1, COX6C and MYL2, were upregulated in Hand2os1^F/F embryonic hearts at E11.5, implying abnormally increased cardiac myogenesis in the mutant heart (Fig. 2E,F; Tables S3 and S5). Further transcriptome analysis of E16.5 ventricles also revealed significant upregulation of genes related to hypertrophic cardiomyopathy in Hand2os1^F/F embryos (Fig. S2E; Tables S3 and S6). The transcriptomic defects may offer a molecular explanation for heart morphological defects and function failure of Hand2os1^F/F newborns.

To study the direct effect of Hand2os1 deletion on HAND2 expression, we first confirmed the complete absence of Hand2os1 transcripts in Hand2os1^F/F mutant embryos by RNA-seq and RT-qPCR (Figs S2A and S3A). No RNA signals were detected downstream of the Hand2os1 locus in Hand2os1^F/F embryos (Fig. S2A), ruling out the possibility that downstream transcription might compensate for the loss of Hand2os1. Thus, Hand2os1^F/F KO mice provided a complete loss-of-function model, in which the transcription, transcripts and DNA sequences of Hand2os1 were simultaneously removed. However, to our surprise, the coding sequence (CDS) of HAND2 showed comparable expression between homozygous and heterozygous littermates throughout heart morphogenesis from E9.5 to E16.5 (Fig. 3A,B). The levels of cardiac HAND2 transcripts were not lost in Hand2os1^F/F mutant embryos, in sharp contrast to the abolished expression of HAND2 in the Uph/Hand2os1 polyA KI embryos (Anderson et al., 2016) (see Discussion).

To confirm this finding, we performed RNA in situ hybridization analysis of E9.5 embryos and found similar distribution and expression patterns of HAND2 mRNA between homozygous and heterozygous littermates (Fig. 3C). Immunostaining analysis in transverse sections of E9.5 embryonic hearts also revealed comparable levels of HAND2 protein in mutant embryos (Fig. 3D, Fig. S3B). We noted a one- to twofold reduction in HAND2 RNA signals that fall into its 5′ untranslated region (UTR) in Hand2os1^F/F embryonic hearts, despite unchanged expression in the CDS of HAND2 (Fig. 3A, Fig. S3C). 5′ RACE analysis revealed an alternative TSS, 447 bp downstream of the annotated TSS of HAND2 (Fig. S3D). Shortening of the HAND2 5′ UTR might promote translation of HAND2 protein, as demonstrated in cultured cells (Fig. S3E) (Curtis et al., 1995; Leppek et al., 2018). Nevertheless, the overall expression of HAND2 at both the RNA and protein levels was sustained in complete absence of Hand2os1.

The mature four-chamber heart of an E16.5 embryo can be experimentally dissected into distinct compartments, in which HAND2 transcripts can then be analyzed by RT-qPCR. Interestingly, except for right ventricles, HAND2 expression showed a tendency to be upregulated in all other compartments of E16.5 Hand2os1^F/F hearts, compared with those of heterozygous littermates (Fig. 3E, Fig. S3F). In particular, HAND2 expression in the mutants was significantly increased by ∼40% in the septum and ∼24% in the right atrium (Fig. 3E).

Sustained and slightly upregulated expression of HAND2 in specific regions of mutant Hand2os1^F/F hearts led us to hypothesize that the Hand2os1 locus might be involved in tuning the spatial expression of HAND2 during heart formation. To reveal subtle alterations that are not readily detectable in population-based analysis due to the averaged expression of mixed cells, we performed high-throughput single cell RNA-seq analysis of E11.5 embryonic hearts isolated from Hand2os1^F/F and wild-type littermates (Fig. 4A). To delineate the primary transcriptional effects of Hand2os1 deletion, we chose to analyze embryos at E11.5, which is also the most convenient early time point that we could experimentally isolate enough cells for 10× Genomics cell sorting and library construction. After removing sequencing reads from hematopoietic cells, we obtained expression profiles for a total of 3600 cardiac cells, including 2108 for Hand2os1^F/F and 1492 for the wild type, with an average of 0.06 million reads/∼13,000 unique molecule identifier (UMI) counts per cell and a median level of 3492 expressed genes per cell [transcripts per million (TPM)>0] (Fig. 4A,B; Table S7).

Classification using the t-distributed stochastic neighbor embedding method (t-SNE) revealed four well-separated clusters of cells that differentially express distinct marker genes (Fig. 4B-D; Tables S7 and S8). About 43% (1550) of cells are cardiomyocytes (CMs) expressing NKX2-5 and MYH7, and ∼31% (1126) are mesenchymal cells (MCs) expressing CTHRC1 and POSTN. About 12% (416) of cells are epicardial cells (EPs) marked by UPK1B and UPK3B, and ∼14% (508) are endothelial cells (ECs) expressing CDH5 and PECAM1 (Fig. 4B,C; Tables S7 and S8) (Li et al., 2016). Interestingly, altered proportions of cardiomyocytes and mesenchymal cells were observed in Hand2os1^F/F embryonic hearts compared with the wild-type littermates (Fig. 4E). Mutant CMs and MCs increased ∼7% and decreased ∼8%, respectively, while ratios of mutant EPs and ECs remained similar (Fig. 4E). Hundreds of single cell profiles obtained for each cell type thus provided large numbers of cells for in-depth statistical comparisons of gene expression changes between mutant and wild-type hearts.

HAND2 is ubiquitously expressed (TPM>0) in 60% of cardiac cells at E11.5 (Fig. 5A). This observation was consistent with previous reports of robust expression of HAND2 detected in many types of cardiac cells using RNA in situ hybridization and immunostaining (Laurent et al., 2017; VanDusen and Firulli, 2012). Interestingly, the Hand2os1^F/F hearts appeared to have a 10% increase (from 60% to 70%) of HAND2-positive populations (Fig. 5A). The observed increase of mutant CMs mainly resulted from an increase in HAND2-positive cardiomyocytes (P=2E-04), while the observed decrease of mutant MCs mainly resulted from a decrease of HAND2-negative mesenchymal cells (P=6E-04) (Fig. 5B). Thus, the opposing changes in the percentages of mutant CMs and MCs appear to be positively correlated with changes of HAND2 expression in these populations.

Although the overall percentage of ECs and EPs remained the same, the percentages of HAND2-positive cells exhibited a significant increase. Thus, cell populations expressing HAND2 increased ∼8-24% in all types of cardiac cell in Hand2os1^F/F mutant hearts, with the most significant gain in the ECs (1.4-fold increase from 59% to 83%, P=4E-09) (Fig. 5A). Moreover, the median levels of HAND2 expression went up significantly by 8-12% across four cardiac cell types (Fig. 5A). In comparison, other master regulators of cardiac development, such as GATA6, GATA4 and NKX2-5, either showed unaltered or decreased expression in particular cardiac cell types (Fig. S4A). These results suggest that the subtle, yet global, upregulation of HAND2 is unlikely to result from sequencing variations between the wild-type and mutant samples.

Next, we sought to determine whether complete removal of Hand2os1 might also affect the expression of other nearby genes. High-order chromatin structure analysis by Hi-C in mouse embryonic stem cells (ESCs) (Bonev et al., 2017) showed that the HAND2 and Hand2os1 loci reside at the boundary of two topologically associating domains (TADs) (Fig. 5C). This boundary demarcates an upstream gene desert of ∼0.65 Mb in length from a downstream gene-rich region. Among 10 genes located within ±1 Mb genomic regions surrounding the TSSs of Hand2os1 and HAND2, we found that six were expressed in at least one type of cardiac cell; four of them exhibited very subtle but statistically significant upregulation in the mutant heart (Fig. 5D,E).

Of the four genes with significantly altered expression, one (Fbxo8) lies ∼0.75 Mb upstream of Hand2os1, and three (SAP30, 5033428I22Rik and HMGB2) lie within the same TAD immediately downstream of HAND2 (Fig. 5C). For SAP30 and 5033428I22Rik, the numbers of cells expressing these genes were significantly upregulated in subsets of mutant cardiac cells (∼9-12% increase of SAP30-positive CMs and ECs, P<0.002; and ∼21% increase of 5033428I22Rik-positive ECs, P=2E-06) (Fig. 5E). HMGB2 and FBXO8 were slightly upregulated in transcript abundance in subsets of cardiac cells (P≤0.01) (Fig. 5D). For comparison, two ubiquitously expressed genes, EZH2 and PSMD1, showed unaltered expression in both transcript abundance and expression frequency in the mutant hearts (Fig. S4B). The combined results demonstrate a cis-regulatory role for the Hand2os1 locus in dampening the expression of HAND2 and several neighboring genes in cardiac cells.

About 197 genes showed altered expression in Hand2os1^F/F cardiac cells (see the Materials and Methods; Table S9). Interestingly, gene ontology (GO) analysis of these dysregulated genes showed that non-CM cells (EP, MC and EC) are specifically enriched in functional terms related to cardiac muscle contraction and heart morphogenesis (Fig. 6A; Table S9). In addition, gene set enrichment analysis (GSEA) showed that non-CMs, but not CMs, of the Hand2os1^F/F mutant exhibit global upregulation of muscle contraction genes and downregulation of genes involved in cardiac septum development (Fig. 6B,C, Fig. S5; Table S3). To directly visualize expression changes, we plotted the ratio of gene expression of Hand2os1^F/F versus control embryos in averaged single cell and bulk RNA-seq of the E11.5 hearts. The heatmap revealed more dramatic expression alterations in mutant non-CMs than in mutant CMs, and the opposing changes in genes related to muscle contraction and cardiac septum development (Fig. 6D).

Many CM marker genes regulating muscle contraction and heart development, such as MYL4, ACTC1, TNNC1 and TNNT2, were significantly upregulated in their expression levels and frequencies (Fig. 6D,E). In contrast, marker genes enriched in non-CMs, such as HES1, SOX4 and FZD2, which are involved in septum development, were specifically downregulated (Fig. 6D,F). For example, ratios of MYL4-positive cells in mutant hearts were increased 18-33% in non-CMs (P<9E-06), and this was also accompanied by 17-22% increases (P<0.0001) of MYL4 transcript abundance (Fig. 6E). In comparison, ∼9-11% of non-CMs lost HES1 expression (P<0.05) and HES1 RNA abundance decreased by 12-18% (P<0.0001) in mutant hearts (Fig. 6F). These molecular changes corroborate the morphological defects of septum lesion and ventricle hypoplasia observed in Hand2os1^F/F embryonic hearts.

Next, we performed correlation analysis to compare the expression similarity of a panel of 1750 differentially expressed genes (Fig. 4D) in each of the four cardiac cell types with those of wild-type CMs. Mutant and wild-type CMs had the highest median levels of correlation coefficient compared with non-CM cells, indicating the robustness of this assay (Fig. 6G). Interestingly, compared with their wild-type counterparts, non-CMs of Hand2os1^F/F embryos showed significantly higher correlations with CMs (Fig. 6G), indicating the resemblance of gene expression programs in cardiomyocytes. Congruent with concurrent upregulation and downregulation of CM and non-CM marker genes, respectively, these results indicate that mutant non-CMs may be aberrantly reprogrammed towards cardiomyocytes, corroborating the global increase of HAND2 in Hand2os1^F/F embryonic hearts. It was reported that HAND2 facilitates cardiomyocyte proliferation and the reprogramming of fibroblasts into functional cardiac-like myocytes in vitro and in vivo (Song et al., 2012). We conjecture that broad changes in cardiac gene expression are likely to be secondary to altered expression of HAND2 in Hand2os1^F/F mutant hearts (see the Discussion).

Enhancer redundancy is commonly observed for the expression of essential developmental genes, and may provide phenotypic robustness in mammalian development (Osterwalder et al., 2018). Dysregulated expression of HAND2 in HAND2os1^F/F embryonic hearts led us to examine possible use of alternative enhancers to sustain HAND2 expression. We first analyzed the published data of HiCap, a genome-wide promoter-capture method to detect chromatin contacts generated in mouse ESCs (Sahlén et al., 2015). In ESCs, the HAND2 promoter interacts with two known upstream enhancers (BA and cardiac) in the Hand2os1 locus, and also with multiple downstream DNA elements embedded in the lncRNA 5033428I22Rik locus (a strong interaction within the first intron of 5033428I22Rik is marked by blue shading in Fig. 7).

Next, we performed chromosome conformation capture (3C) in hearts from E16.5 wild-type and mutant embryos. Using the TSS of HAND2 as the anchor, we analyzed chromatin interactions across its downstream 20 kb region to the first intron of 5033428I22Rik. Indeed, we observed significantly higher interaction frequencies of the HAND2 TSS with the promoter and 5′ sequences of 5033428I22Rik relative to the nearby regions in E16.5 embryonic hearts (Fig. 7). These chromatin contacts peak at a CTCF-binding site that is next to a strong HiCap signal detected in ESCs and overlaps with H3K27ac signals in the 5′ end of the first intron of 5033428I22Rik. Notably, levels of long-range DNA interactions between the HAND2 TSS and downstream regions are comparable in Hand2os1^F/F and wild-type embryos, which we reasoned might be due to subtle changes in chromatin structure in subpopulations of mutant cardiac cells or the limited sensitivity of 3C, or both. Nevertheless, analysis of chromatin contacts in ESCs and embryos suggested engagement of alternative downstream enhancers with the HAND2 promoter, in addition to regulatory elements embedded in the Hand2os1 locus.

Compared with extensive reports of lncRNA functions in cell lines, rigorous exploration and dissection of their potential actions and effects in animals are still limited (Li and Chang, 2014). Precise expression of HAND2 is crucial for heart formation. Transcription of the divergent lncRNA Hand2os1 is reportedly essential for HAND2 activation and heart morphogenesis (Anderson et al., 2016). Using three knockout mouse models, we demonstrate a key role of the Hand2os1 DNA locus in modulating HAND2 expression and normal heart development and function. Full-length deletion of Hand2os1 led to congenital heart defects and perinatal lethality. Importantly, in embryos lacking the entire Hand2os1 DNA sequence, single cell transcriptomic analysis of the heart revealed subtle yet prevalent upregulation of HAND2 and dysregulated cardiac gene expression programs. These results illustrate a crucial, fine-tuning function of the lncRNA Hand2os1 locus in accurately controlling the spatial expression of HAND2, through which Hand2os1 modulates cardiac lineage development and heart function.

Current lines of evidence indicate that disruption of the Hand2os1 DNA locus rather than Hand2os1 transcription/transcripts primarily contributes to dysregulated expression of HAND2 and heart morphological defects and lethality. The lack of discernable phenotypes in the heart and in animal survival in Hand2os1^P/P mice suggests that Hand2os1 transcripts and perhaps its transcription may be largely dispensable for cardiac development and expression of HAND2 in the heart. Consistent with this finding, a related report showed that deletion of the first two exons of Hand2os1 (lncHand2) did not cause apparent heart abnormality and failed to affect hepatic expression of HAND2 despite the absence of lncHand2 transcripts in mouse livers (Wang et al., 2018). Although residual transcripts were expressed in Hand2os1^P/P mice (∼2-7 molecules of truncated Hand2os1^P transcripts per cell versus 15∼40 molecules per wild-type cell) (Fig. 1C, Fig. S1C; see the Materials and Methods), the low abundance does not seem to justify a trans function of Hand2os1^P transcripts in a broader regulation of cardiac gene programs. Our genetic evidence based on Hand2os1^P/P and Hand2os1^F/F mice does not support a trans-acting mechanism of the Hand2os1 locus in regulating expression of cardiac genes, unless Hand2os1 and target gene loci are in spatially close proximity in the nucleus.

For the above reasons, we conjecture that broad alterations in cardiac gene programs observed in Hand2os1^F/F mutant hearts are likely to result from aberrant upregulation of HAND2 due to the loss of cis-regulatory DNA sequences embedded in the Hand2os1 locus. Hand2os1 transcripts may serve as proxy signals for the activity of important regulatory DNA elements embedded in its DNA locus in cardiac cells, in a manner similar to the lncRNA Rroid, the locus but not the RNA of which regulates the homoeostasis and function of specific innate lymphoid cells (Mowel et al., 2017). However, we could not absolutely exclude the possibility that the Hand2os1 RNA might be more specifically expressed in subsets of cardiac cells, in which it regulates the expression of key cardiac genes that are genomically distal but in the chromatin neighborhood of the Hand2os1/HAND2 locus. Other lncRNA molecules, such as Mhrt, Wisper and Fendrr, have been reported to have trans-acting effects on heart development and function (Grote et al., 2013; Han et al., 2014b; Micheletti et al., 2017). In addition, it remains possible that in other pathological or stress conditions yet to be revealed, Hand2os1 transcription and transcripts may play a role in defining the chromatin environment required for the precise regulation of HAND2 transcription.

Although the molecular effects of Hand2os1 on HAND2 expression in individual cells of E11.5 embryonic hearts are moderate, they can be robustly detected in hundreds of cells across all four cardiac cell types. Consistent with single cell analysis, subtle upregulation of HAND2 was also observed in the septum and the right atrium of E16.5 Hand2os1^F/F embryos. Notably, in bulk analysis of Hand2os1^F/F mutant hearts, expression changes of muscle contraction genes that are distal to the Hand2os1 locus appear to be more pronounced than changes in HAND2 expression. Subtle cis-effects could be easily missed in conventional ensemble analysis, which may confound the mechanistic interpretation of the function of lncRNA in cis or in trans. Through combinational analysis of multiple mouse models and single cell characterization, we conclude that HAND2 expression alteration is the primary molecular effect of inhibition of Hand2os1 in heterogeneous cell populations. HAND2 expression undergoes a decrease after E10.5 and then remains low throughout the remaining course of heart development (Srivastava et al., 1997; Tamura et al., 2014). The observed upregulation of HAND2 in E11.5 and E16.5 Hand2os1^F/F embryos may also reflect improper downregulation of HAND2 during cardiac development.

Like many lncRNAs, such as Flicr and Fendrr (Grote et al., 2013; Zemmour et al., 2017), Hand2os1 deletion had subtle molecular effects that nonetheless resulted in profound biological consequences in vivo. In Hand2os1^F/F embryos, the overall formation of an organized four-chamber heart is not affected, and the morphological defects in specific regions of the heart, including lesions in the interventricular septum and ventricle hypoplasia, are moderate. However, these abnormalities in combination may be sufficient to impair heart function, as indicated by apparent blood coagulation and fatal thrombosis, thereby leading to immediate death of mutant animals upon birth.

Interestingly, single cell and bulk analysis of gene expression revealed upregulation of HAND2 and muscle contraction genes but downregulation of septum genes in Hand2os1^F/F embryos. Multiple studies using various mouse models have demonstrated that aberrantly high levels of embryonic HAND2 led to septum lesion (Table S1). For example, MYH7-driven overexpression of HAND2 prevents the formation of the interventricular septum in embryos (Togi et al., 2006). In the Rim4 mouse model, which mimics a human chromosomal disorder caused by partial trisomy of distal 4q (4q+), a region containing 17 genes, including HAND2, overdose of HAND2 has been suggested as a major cause of severe ventricular septal defects (VSDs) and perinatal lethality (Tamura et al., 2013). In addition, mice deficient in miRNA-1-2 express more HAND2 protein, and exhibit a VSD and embryonic death from E15.5 to just after birth with 50% penetrance (Zhao et al., 2007). Moreover, MYH6-driven overexpression of HAND2 in adult cardiac muscle cells causes pathological heart hypertrophy (Dirkx et al., 2013). Comparisons of the above mouse models with the three Hand2os1 KO mice we generated suggest a correlation between the level of HAND2 overexpression and the severity of heart defects.

Based on these lines of evidence, we interpreted the observed molecular and morphological changes in mutant hearts to be a result of excess amounts of HAND2 transcripts in Hand2os1^F/F embryos. Our data suggest a model in which an apparently subtle but global increase in the HAND2 dose in all cardiac cell types may alter cardiac gene programs, eventually resulting in morphological and functional abnormalities. Yet a definitive demonstration of this model will require generation of additional mouse models that overexpress HAND2 at a level comparable with that in Hand2os1^F/F mutants.

Complete removal of the entire 17 kb Hand2os1 sequence abolishes Hand2os1 transcription and transcripts, but fails to attenuate HAND2 expression, leading to much weaker cardiac defects and delayed onset of death, compared with Hand2os1/Uph polyA KI mutant embryos, which show abolished expression of HAND2 and failed heart morphogenesis at E10.5 (Anderson et al., 2016). These discrepancies were unexpected, as polyA KI had been thought to minimally disrupt the genomic DNA compared with a large deletion. It is possible that residual low levels (∼8-32%) of nascent transcription/RNA of Hand2os1 in Hand2os1^P/P embryonic hearts might have a potential effect on HAND2 expression. Alternatively, one parsimonious explanation is that engagement of the HAND2 promoter with downstream enhancers embedded in the lncRNA 5033428I22Rik locus provides alternative use of enhancers that may sustain but imprecisely control HAND2 expression in Hand2os1^F/F embryonic hearts. The role of properly regulated promoter-enhancer interactions for precise gene expression has been demonstrated in the example of the lncRNA Pvt1 (Cho et al., 2018). Another trivial interpretation is that the Hand2os1 region might contain both negative and positive regulatory elements for HAND2 expression (Charite et al., 2001; McFadden et al., 2000; Voth et al., 2009), and the loss of Hand2os1 expression might be compensated by the loss of the negative element(s), of which function is normally de-repressed by the transcription of Hand2os1. Generation of an allelic series of deletion will define the cis-regulatory elements in the Hand2os1 locus in future studies.

On the other hand, one might consider the possibility that the severe phenotypes observed in the polyA KI mice may result from polyA-induced aberrant silencing, independently of the lncRNA function. The Hand2os1 and HAND2 locus is bivalently marked by H3K4me3 and H3K27me3 in mesodermal precursor cells (Almada et al., 2013); and the repressive H3K27me3 mark needs to be removed prior to transcription activation of HAND2 during the onset of cardiogenesis. Possibly, the proximal insertion of a transcription stop signal immediately upstream (−644 bp) of the HAND2 TSS may lead to artificial recruitment of the transcription termination machinery. Given the close juxtaposition (123 bp) between the TSSs of Hand2os1 and HAND2, abnormal occupancy of proteins involved in transcription termination at the proximal promoter of HAND2 might prevent the binding of transcription activators and/or chromatin remodelers, consequently blocking the removal of H3K27me3 and activation of HAND2 upon heart morphogenesis. One example to support this possibility is the study of the lncRNA ThymoD. Insertion of a polyA signal downstream of the TSS of ThymoD maintains the silenced local chromatin state and inhibits its repositioning into transcriptionally active domains in the nuclear interior (Isoda et al., 2017).

Nevertheless, together with the reported polyA KI mouse model, our study clearly demonstrates the complexity of the Hand2os1 locus in modulating HAND2 expression. Careful examinations using more mutant animals are necessary to dissect the precise function of Hand2os1 transcription, transcripts and its embedded DNA elements in future studies. The fact that different knockout strategies produce distinct phenotypes underlines the requirement to use complementary genetic approaches to study the physiological functions of an lncRNA in mouse models (Han et al., 2018; Luo et al., 2016; Yin et al., 2015). Last, we believe that careful genetic dissection coupled with single cell analysis will lead to in-depth understanding of the functions and mechanisms of action of lncRNAs, thus truly impacting on our understanding of the noncoding genome in animal development, fitness and disease.

All mice we used had a C57BL/6 background, with age described in the manuscript. Embryos were isolated at the developmental stages indicated in the manuscript. All animal experiments were conducted in accordance of institutional guidelines for animal welfare and approved by the Institutional Animal Care and Use Committee (IACUC) at Tsinghua University, China.

HEK 293T cells (CRL-3216, ATCC) were cultured in DMEM supplemented with 10% heat-inactivated fetal bovine serum (Hyclone) and 1% penicillin/streptomycin (Cellgro). ESCs (46C, Austin Smith Lab) were grown in DMEM (Cellgro) supplemented with 15% heat-inactivated fetal bovine serum (Hyclone), 1% glutamax (Gibco), 1% penicillin/streptomycin (Cellgro), 1% nucleoside (Millipore), 0.1 mM 2-mercaptoethanol (Gibco), 1% MEM nonessential amino acids (Cellgro) and 1000 U/ml recombinant LIF (leukemia inhibitory factor) (Millipore) on gelatin-coated plates. All cultured cells were maintained in a humidified incubator at 37°C with 5% CO₂ (Luo et al., 2016).

CRISPR/Cas9-meditated genetic deletion for lncRNA knockout mice generation was performed as previously described with minor modifications (Han et al., 2014a, 2018). Briefly, Cas9 mRNA and sgRNAs were co-injected into mouse zygotes. For each genetic deletion, we used two sgRNAs (Hand2os1^P, sg1 and sg2; Hand2os1^D, sg3 and sg4; Hand2os1^F, sg1 and sg5) (Table S10). For Hand2os1^P, we deleted a 1 kb DNA sequence covering the core promoter and the first two exons of Hand2os1, with 94% of Hand2os1 DNA sequences remaining intact. When genetic deletion was confirmed, the germline transmission was performed for two generations by mating with C57BL/6. F2 mice and later generations were used for heterozygote intercrosses.

Knockout primers (forward and reverse) were designed outside of the deleted region. For wild-type band amplification, we used one of the knockout primers together with a primer designed inside of the deleted region (Table S10). Knockout bands were confirmed by Sanger sequencing (Han et al., 2018).

Echocardiography was performed on Hand2os1^D/D, Hand2os1^P/P and their littermate control mice at 6-10 weeks. Briefly, mice were gently restrained in the investigator's hand during echocardiography detection. Two-dimensional, short-axis views of the left ventricle were obtained for guided M-mode measurements of the left ventricular (LV) internal diameter at end diastole (LVIDd) and end systole (LVIDs). LV internal diameter was measured in at least three beats from each projection and averaged. The fractional shortening (FS) was calculated by the following formula: FS (%)=[(LVIDd−LVIDs)/LVIDd]×100, which represents the relative change of left ventricular diameters during the cardiac cycle (Table S2).

For Hematoxylin and Eosin staining, E16.5 embryonic hearts were fixed in 4% PFA overnight at room temperature, dehydrated through a graded ethanol series (50%, 75%, 90%, 95%, 100%) and paraffin embedded. After sectioning (7 μm), the tissues were deparaffinized in xylene and rehydrated through a graded ethanol series (100%, 95%, 75%), then stained with Hematoxylin and Eosin. Ratio of RV area was calculated as the RV chamber area divided by whole ventricle area. Thickness of RV body wall was calculated as the ventricular compact myocardial area divided by its outer circumference. Relative thickness of RV body wall was calculated as thickness of RV body wall divided by whole ventricle circumference. These data were measured in Adobe Photoshop CC2014 after selection of the image areas with myocardial color range. For immunostaining, E9.5 embryos were fixed in 4% PFA at 4°C 2 h, dehydrated by 30% sucrose and embedded in OCT. Then frozen sections was cut at 7 μm on a cryostat set at −20 to −25°C. Immunostaining was performed with primary antibodies against TNNI3 (Abcam, ab56357, 1:200) and HAND2 (Abcam, ab200040, 1:100). Primary antibodies were visualized by staining with Alexa-conjugated secondary antibodies: Alexa Fluor 488 donkey anti-goat (Life Technologies, #A-11055) and Alexa Fluor 555 donkey anti-rabbit (Life Technologies, #A-31572) diluted 1:200. All the slides were mounted in Vectashield hardset antifade mounting medium (Vector Laboratories) and imaged on a Zeiss microscope.

Whole-mount in situ hybridization was carried out with digoxigenin-labelled antisense RNA probes as previously described with some modifications (Anderson et al., 2016; Wei et al., 2011). In brief, RNA probe for HAND2 were amplified from cDNA of mouse embryonic heart and transcribed in vitro using T7 RNA polymerase (Roche, 10881767001) with DIG RNA labeling mix (Roche, 11277073910) (Table S10). Embryos were fixed in 4% PFA at 4°C overnight, dehydrated through a graded methanol series (50%, 75%, 100%) and stored in 100% methanol at −20°C. The embryos were bleached in a solution containing 30% H₂O₂:methanol 1:5 for 2 h, then rinsed in methanol, rehydrated through a graded methanol series (100%, 75%, 50%) and then washed in PBS. The embryos were post-fixed for 20 min in 4% PFA. After washing in PBS, embryos were transferred to the hybridization buffer (50% formamide, 5× SSC, 500 µg/ml yeast RNA, 50 µg/ml heparin and 0.1% Tween-20) and pre-hybridized for 4 h at 65°C. Hybridizations were performed in fresh hybridization buffer containing 0.25 ng/µl digoxigenin-labelled antisense RNA probes overnight at 65°C. Post-hybridization washes were performed at 65°C in wash buffer 1 (50% formamide, 2× SSC), wash buffer 2 (2× SSC) and wash buffer 3 (0.2× SSC), then performed in MABT (100 mM maleic acid, 150 mM NaCl and 0.1% Tween-20) at room temperature. After 1 h of blocking at room temperature in 10% sheep serum, 2% blocking reagent (Roche, 11096176001) (diluted in MABT), embryos were incubated overnight at 4°C in blocking solution as above, with anti-DIG-AP antibody (Roche, 11093274910, 1:3000). Then mouse embryos were washed in MABT at room temperature. After the post-antibody washes, embryos were washed in NTMT [100 mM NaCl, 100 mM Tris-HCl (pH 9.5), 50 mM MgCl2 and 0.1% Tween-20]. Staining was visualized in BM Purple AP Substrate (Roche, 11442074001).

Chinese ink (Yidege) was injected into the left ventricles of E16.5 embryos to visualize the organization of the arteries.

Plasmids were transfected into 293T or ESCs using Lipofectamine 2000 (Life Technologies, #200059-61). For western blot, cells were harvested 24 h after transfection. HAND2 cDNA with different lengths of 5′ UTR or without 5′ UTR were cloned into a piggyBac vector. piggyBac-GFP were co-transfected with HAND2 cDNA as a control for transfection efficiency. For validation of HAND2 antibody, flag-tag was added to the N-terminal of HAND2.

Cultured cells were washed in PBS and boiled in 5× SDS sample buffer for 5 min at 95°C. After SDS-PAGE and transfer, membranes were blocked in 5% milk/TBS-Tween. Primary antibody was applied for 2 h and secondary HRP-conjugated antibodies was applied for 1 h at room temperature. Membranes were washed for 3×10 min in TBS-Tween after each antibody incubation, and incubated with ECL substrate before exposure to X-ray film. Primary antibodies against HAND2 (Abcam, ab200040, 1:1000), β-tubulin (Abmart, M30109, 1:2000), FLAG (EASYBIO, BE2005-100, 1:2000) and GFP (CWBIO, CW0086, 1:2000) were used. Secondary antibodies used included goat anti-mouse IgG (CWBIO, CW0102) and goat anti-rabbit IgG (CWBIO, CW0103) at 1:5000 dilution.

Tissue were washed by PBS and harvested in TRIzol reagent (Life Techonologies, #15596018). Adult cardiomyocytes were isolated using type II collagenase in the Langendorff retrograde perfusion mode (O'Connell et al., 2007). Total RNA was extracted according to the manufacture recommended procedure. Total RNA (0.5 to 2 µg) was reverse transcribed using a RevertAid First Strand cDNA Synthesis Kit (Fermentas, K1622) with random primers. RT-qPCR was performed using iTaq Universal SYBR Green Supermix (Bio-Rad, 1725121) on a Bio-Rad CFX384 or CFX96 RealTime System. RT-qPCR analysis data provide the expression relative to GAPDH or 18S expression ±s.e.m. (Table S10). For the estimation of RNA molecules per cell, we first drew a standard curve to normalize the PCR efficiency of Hand2os1 and HAND2 RT-qPCR primers. Next, we estimated the RNA molecules per cell by normalization of counted cell number used for PCR reaction. We estimated that the abundance of Hand2os1 transcripts is about ∼15-40 molecules per cell, relatively 7- to 23-fold lower than that of HAND2 transcripts. In E12.5 embryonic heart and adult heart, the Hand2os1 transcript is ∼15 and ∼40 molecules per cell, respectively (Fig. S1C). In Hand2os1^P/P embryonic heart at E12.5, only 10% of truncated Hand2os1^P transcripts remained, which is ∼2 molecules per cell. Similarly, in Hand2os1^P/P adult heart, only ∼7 molecules per cell are detected (17% of 40 molecules) (Fig. 1C).

Total RNA of E16.5 hearts of Hand2os1^F/+ and Hand2os1^F/F from heterozygotes intercrosses were used for 5′ RACE cDNA library construction. 5′ RACE was performed by following the manufacture of SMARTer RACE 5′ Kit (Clontech, 634858). One round of PCR amplification was performed with universal primer (forward) and gene-specific primer (reverse) (Table S10).

E16.5 hearts of Hand2os1^F/+ and Hand2os1^F/F from heterozygotes intercrosses were used for 3C. The 3C experiments were performed as described previously (Du et al., 2017) with minor changes. Each embryonic heart was freshly isolated and crosslinked using 1% formaldehyde. Then, the samples were lysed [10 mM Tris-HCl (pH 7.4), 10 mM NaCl, 0.1 mM EDTA, 0.5% NP-40 and proteinase inhibitor] on ice for 1 h, digested with MboI overnight, ligated at room temperature for 6 h, reverse crosslinked then their DNA was purified. The proximal ligation events were detected by RT-qPCR using the primers in Table S10. We amplified ten DNA fragments (44 kb in total) covering the Hand2os1/HAND2 DNA locus we investigated as naked DNA control for 3C experiments. The interaction frequencies in embryonic hearts were normalized to that of naked DNA control.

Adult (8 week) cardiomyocytes of CTRL (Hand2os1^+/+ and Hand2os1^D/+) and Hand2os1^D/D, and E11.5 hearts and E16.5 ventricles of Hand2os1^F/+ and Hand2os1^F/F from heterozygotes intercrosses were subjected to RNA-seq following polyA purification. The RNA libraries were constructed by following the Illumina library preparation protocols. High-throughput sequencing was performed on Illumina HiSeq 2500 or HiSeq X TEN. All RNA-seq data were mapped to the mouse reference genome (mm9) using TopHat (version 2.0.11) (Trapnell et al., 2012). Reads were assigned to their transcribed strand (Tophat parameter ‘--library-type=fr-firststrand’). The gene expression level was calculated using Cufflinks (version 2.0.2) (Trapnell et al., 2012) with the refFlat database from the UCSC genome browser. For visualization, the read counts were normalized by computing the numbers of reads per million of reads sequenced (RPM). Gene set enrichment analysis (GSEA) (version 2.2.4) was performed by comparing mutant samples with control samples (Subramanian et al., 2005). We used gene sets from KEGG V6.0 (Kanehisa and Goto, 2000) and gene ontology for GSEA (Table S3). First, genes with FPKM >10 and |log2 (fold change)|>0.2 [fold change=(FPKM of KO+0.1)/(FPKM of CTRL+0.1)] were selected as candidates. To exclude the inconsistently dysregulated candidates, we further filtered genes with t.test>0.1 or |log10 (t.test)×log2 (fold change)|<0.5 (dysregulation score). For 8-week-old cardiomyocytes of Hand2os1^D/D, 114 and 186 genes are upregulated and downregulated, respectively (Table S4). For E11.5 hearts of Hand2os1^F/F, 169 and 101 genes are upregulated and downregulated, respectively (Table S5). For E16.5 ventricles of Hand2os1^F/F, 31 and 19 genes are upregulated and downregulated, respectively (Table S6). Heatmaps were drawn using Cluster 3.0 and viewed by Treeview. The colors represent the fold change of gene expression, which is relative to average FPKM of each gene across all analyzed samples.

E11.5 hearts of wild-type and Hand2os1^F/F from heterozygotes intercrosses were subjected to single cell RNA-seq. We harvested six embryonic hearts for each genotype. Embryonic hearts were trypsinized (0.25%) for 5 min at 37°C individually and subjected to fluorescence-activated cell sorting (FACS) after 7-aminoactinomycin D (AAT Bioquest, 17501) staining for collection of living single cells. Next, six embryonic hearts were combined as a single sample for 10× Genomics Single Cell 3′ library construction (10× Genomics, PN-120237). The RNA libraries were constructed by following the manufacture recommended procedure to obtain ∼5000 cells barcoded per sample. Sequencing data analyses including sample demultiplexing, barcode processing and single cell 3′ gene counting were carried out using by the Cell Ranger Single-Cell Software Suite (software.10xgenomics.com/single-cell/overview/welcome) (Zheng et al., 2017). We obtained ∼190 million reads for 3469 detected Hand2os1^F/F cells and ∼181 million reads for 2563 detected wild-type cells, which indicate an average of 60,000 reads per cell. The cDNA insert was aligned to mouse reference genome (mm10). Cells with fewer than 1000 detected genes were removed. We used RandomForest approach to discriminate a population of hemopoietic cells and excluded them from downstream analysis, and finally obtained 1492 single cells from wild-type heart and 2108 single cells from Hand2os1^F/F heart. On the median, we detected 3492 genes per cardiac cell. As we compared regulatory expression differences between the mutant and wild-type cardiac cells in the same heart tissue, we reasoned that sample biases towards different cell populations with various sizes and fragilities would be less pronounced in our assay. For comparative analysis of wild-type and mutant single cell datasets, we took the union of the top 1000 genes with the highest dispersion (var/mean) from both datasets (‘WT’ object and ‘Mutant’ object) to perform the alignment procedure in the Seurat integration procedure (Butler et al., 2018). We ran a canonical correlation analysis (CCA) to identify common sources of variation between the two datasets. Then we aligned the top seven CCA subspaces (or dimensionalities) to generate a single new dimensional reduction integrated WT and mutant dataset that we used for subsequent analyses such as t-SNE visualization. Next, we used the ‘FindClusters’ function to identify four main cardiac cell types and verified them using known marker genes (Li et al., 2016). We achieved 43% cardiomycytes, 12% epicardial cells, 31% mesenchymal cells and 14% endothelial cells (Table S7). The dominant composition of CMs in our data is consistent with two previous single cell studies which profiled 96 cells at E11.5 and 1165 cells at E10.5 of embryonic hearts (Dong et al., 2018; Li et al., 2016). Increased percentages of MCs and EPs at E11.5 compared with those in the E10.5 heart reported previously are in accordance with increased proliferation of cushions and epicardium (Li et al., 2016). To identify unique cluster-specific marker genes and for heatmap plotting, we used the Seurat function ‘FindAllMarkers’ (thresh.test=0.5, test.use=’roc’) and define a group of differentially expressed genes (DEGs) containing 1750 genes (Table S8). Because the median UMI we detected in most of single cells did not reach one million UMIs, we used log2 (TPM/10+1) rather than log2 (TPM+1) to normalize the expression levels for following analysis. For expression analysis of HAND2-neighboring genes (Fig. 5D,E), a gene is defined as expressed if the corresponding transcripts are detected in more than 20% of cells in at least one cell type. Next, we identified dysregulated genes between wild-type and mutant hearts by averaged log2 (TPM/10+1) of each gene (>2) in each cell type. Genes with |log2 (mutant+1)/(wildtype+1)|>0.2 are defined as dysregulated genes, which were used for gene ontology analysis (Table S9). A total of 197 genes showed altered expression in Hand2os1^F/F cardiac cells, including 56 in CMs, 120 in EPs, 54 in MCs and 80 in ECs. Correlation coefficients of cardiac cells were analyzed by 1750 DEGs. We used average gene expression of all cardiomyocytes (591 single cells) from the wild-type sample as a standard gene expression profile, and calculated the Pearson correlation of each single cell compared with the standard. GSEA was performed by comparing all single cells of Hand2os1^F/F hearts with wild-type samples.

Published sequencing datasets used in this paper (Figs 1A and 7) were collected from Encyclopedia of DNA Elements (ENCODE) and Gene Expression Omnibus (GEO), including DHS-seq of E11.5 heart (ENCSR932SBO), CTCF ChIP-seq of postnatal day 0 (P0) heart (ENCSR491NUM), H3K27ac ChIP-seq of E12.5 heart (ENCSR123MLY), H3K4me3 ChIP-seq of E12.5 heart (ENCSR688ZOR), polyA RNA-seq of E12.5 heart (ENCSR150CUE), total RNA-seq of P0 heart (ENCSR035DLJ) (Yue et al., 2014), Pol II (8WG16) ChIP-seq of E12.5 heart (GSM1260035) (He et al., 2014), GATA4 ChIP-seq of E12.5 heart (GSM1260026) (He et al., 2014), HAND2 ChIP-seq of E10.5 heart (GSM1891956) (Laurent et al., 2017), NKX2-5 ChIP-seq of E12.5 heart (GSM1724109) (Ye et al., 2015) and HiCap in mESC (GSE60494) (Sahlén et al., 2015).

Results for RT-qPCR, echocardiography, ratio of RV area and relative thickness of RV body wall are shown as mean values with error bars representing the standard error (s.e.m.), except for Hand2os1/HAND2 RT-qPCR results in cardiomyocytes from CTRL (Hand2os1^{+/+, D/+}) and Hand2os1^D/D (shown as median with range). Replicates are indicated in the figure legends. For each comparison between two groups, statistical analysis was performed and P-values were calculated using an unpaired two-tailed Student's t-test and GraphPad Prism 5 software. Measurement of heart chamber area and circumference were performed using Adobe Photoshop CC2014. Imaging data analyses were carried out in Zen 2012. For single cell RNA-seq analyses, scatter plots (gene expression and correlation coefficient) are shown as median and interquartile range. We used a Mann–Whitney test for statistical analysis of gene expression and correlation coefficient for single cell RNA-seq results. Fisher's exact test was used for significance testing of gene expression frequency for single cell RNA-seq results. The P-values of Mendelian ratio are based on the χ² test.

We thank W. Pu, L. Yu, G. Ou, Y. Chen, Q. Xi, and Shen Laboratory members for insightful discussion and critical reading.