Introduction

Atria and ventricles arise in a highly localized fashion from an initially linear tube in the embryonic heart. These embryonic chambers feature an increased cell proliferation and a working type of myocardial gene program. They are initially separated by the atrioventricular canal (AVC) that by virtue of its lower cell division rate provides a primitive morphological constriction. Because the myocardial lining of the AVC remains of a primitive type with low conductivity [1], it will delay the propagation of the electrical impulse between atria and ventricles. Most of the myocardial investment of the AVC will have disappeared by the end of gestation, but some nodal type of cells remain and form the AV junction and AV node, and continue to relay the impulse to the ventricles. Furthermore, the AV myocardium induces the overlying endocardium to undergo an epithelial–mesenchymal transition (EMT) and to invade the cardiac jelly, an extracellular matrix (ECM) that is deposited by the primary myocardium. These mesenchymal cushions are subsequently remodeled into thin valve leaflets, AV insulation, and components of the septa that ensure structural and functional compartmentalization of the heart [2–4].

Given the common developmental origin of AV nodal cells, insulation and valves, it may not be surprising that patients with congenital AVC defects often suffer from additional conduction problems, and that patients with Ebstein’s anomaly, where the tricuspid valve is incorrectly positioned and hypomorphic, frequently display ventricular pre-excitation [5]. These observations indicate a common regulatory pathway for the formation of the AVC/conduction system and the mesenchymal components of valves and septa.

BMP signaling plays a critical role in both AVC specification and the formation of cushion tissue [4]. Myocardial bone morphogenetic protein 2 (Bmp2) activates myocardial expression of the transcriptional repressor T-box 2 (Tbx2), and is thought to directly induce cushion formation in the adjacent endocardium [2, 3, 6–9]. Additionally, Tbx2 inhibits the chamber myocardial gene program in the AVC [10–12]. Tbx3 is genetically and functionally related to Tbx2, and suppresses chamber differentiation of the sinus node, the AV bundle and bundle branches [13, 14]. Tbx2 and Tbx3 expression overlaps in the AVC, suggesting that functional redundancy has prevented a full appreciation of their role in the development of this tissue to date [12]. Here, we present data in the mouse that implicate a feed-forward loop between Tbx2/Tbx3 and Bmp2 in this tissue.

Materials and methods

Mice and genotyping

Mouse work was performed in accordance with national and international guidelines

Mice carrying a null allele of Tbx2 (Tbx2 tm1.1(cre)Vmc, synonyms: Tbx2 −, Tbx2 Cre) [10] and/or Tbx3 (Tbx3 tm1.1(cre)Vmc, synonyms: Tbx3 −, Tbx3 Cre) [13] and mice for Cre-mediated misexpression of Tbx2 or Tbx3, NppaCre [15], Myh6Cre [16], Myh6-MerCreMer [17] and CAG-CAT-TBX3 (CT3) [13], and Nkx2-5 Cre/+ and Bmp2 floxneo/− [3] were previously described. For conditional misexpression of Tbx2, a Tbx2 expression cassette was introduced in the Hprt locus (HpCT2) (see online data supplement). All strains were maintained on outbred (NMRI or FVB/N) background.

Generation of the Hprt TBX2 allele

A ‘knock-in’ strategy into the X-chromosomal hypoxanthine guanine phosphoribosyl transferase (Hprt) gene locus was designed to replace mayor parts of the Hprt exon 1 (including the ATG) by a cassette suited for cre-mediated (mis-) expression described previously [18]. Homologous recombination results in a functional Hprt null allele, allowing direct selection of successfully targeted ES cells by 6-Thioguanine. The targeting vectors contained a 2.2-kbp 5′-homology region, followed by the ubiquitously expressed CMV early enhancer/chicken β-actin (CAG) promoter, the conditional expression cassette [18], and a 5.1-kbp 3′-homology region. The open reading frame (ORF) of human TBX2 (cDNA NM_005994.3) [19] was first subcloned in the vector pSL1180 (GE-healthcare), 5′ of an IRES-EGFP sequence, and then shuttled as 5′-NheI-ORF-IRES-EGFP-MluI-3′ fragment into the MluI and NheI sites of the targeting vector. This results in a reverse orientation of the ORF, relative to the CAG promoter, avoiding ‘leaky’ expression. After cre-mediated ‘flipping’- and excision events between pairs of loxP and loxM sequences, the ORF locates in sense direction, directly downstream of the CAG promoter. The targeting vector was verified by sequencing before linearization and electroporation in Hprt-positive SV129 ES cells (maintained beforehand in HAT medium). A two-step selection protocol was employed, starting 24 h after electroporation with the addition of 100 μg/ml G418, followed by the addition of 1.67 μg/ml 6-Thioguanine (Sigma) after an additional 5 days. Surviving colonies were expanded and genotyped by PCR (conditions are available upon request). To test the functionality of the expression cassette in candidate ES clones, the GFP epifluorescence was analyzed 6 days after electroporation with a cre-expression plasmid (pCAG::turbo-cre, kind gift from Achim Gossler). Verified ES clones were microinjected into CD1 mouse blastocysts. Chimeric males were obtained and mated to NMRI females to produce heterozygous F1 females.

Generation and isolation of transgenic embryos/mice

For the generation of Tbx2 or Tbx3 mutant embryos, mice heterozygous for Tbx2 or Tbx3 null alleles were intercrossed. For the generation of double-mutant embryos, double-heterozygous mice were intercrossed. Double-transgenic mice conditionally expressing TBX3 in the atria and the whole heart were generated by crossing CT3 mice with NppaCre or Myh6Cre mice, respectively. Double-transgenic mice conditionally expressing Tbx2 in the whole heart were generated by crossing HpCT2 mice with Myh6Cre or Mox2 Cre mice. For timed pregnancies, vaginal plugs were checked in the morning after mating, noon was taken as embryonic day (E) 0.5. Embryos were harvested in PBS, fixed in 4% paraformaldehyde overnight and stored in 100% methanol at −20°C before further use. Wild-type littermates were used as controls. Genomic DNA prepared from yolk sacs or tail biopsies was used for genotyping by PCR (see supplementary table 6).

Embryonic heart explant assay

Atria from Myh6-Cre;HpCT2 and control (Myh6-Cre) E10.5 embryos were dissected, individually placed on Transwell filters and cultured for 4 days in presence or absence of 30 μM Dorsomorphin in explant medium (Optimem supplemented with Penicillin/Streptomycin, Glutamax, FCS (2%), Insuline/Transferrin/Selenium). Total RNA was extracted with RNAPure reagent (Peqlab) and DNaseI (Roche) treated for 30 min at 37°C. RNA was reverse transcribed with RevertAid H-minus M-MuLV reverse transcriptase (Fermentas). For semi-quantitative PCR, the number of cycles was adjusted to the mid-logarithmic phase. Quantification was performed with Quantity One software (Bio-Rad). Normalization was against Gapdh. Assays were performed at least twice in duplicates. P values were calculated using the unpaired two-tailed Student’s t test. See supplementary table 6 for primer sequences and PCR conditions.

Sample preparation, RNA isolation, and gene expression analysis

Left atria of six NppaCre;CT3 mice and six control (NppaCre) mice (male, 6 weeks) were dissected and snap frozen in liquid nitrogen. Total RNA was isolated and purified using single prep nucleospin columns according to the manufacturer’s instructions (Macherey–Nagel). RNA quality was checked using a bioanalyzer (Agilent Technologies). A total of 250 ng of total RNA was used for biotin-16-UTP cRNA labeling and amplification using the Illumina RNA amplification kit (Ambion Inc., Austin, TX). Labeled RNA was hybridized to Illumina MouseRef-6 BeadChip following the manufacturer’s instructions (Illumina Inc., San Diego, CA). The arrays were scanned using an Illumina Bead array reader confocal scanner. BeadStudio software was used to assess the individual array quality. Unprocessed intensity values were averaged per bead type, exported from BeadStudio and subsequently normalized using VSN in R [20]. Genes were tested for significant differential expression using the empirical Bayes moderated t-statistics test in the R-Limma package [21] at a 5% Benjamini–Hochberg false discovery rate [22]. We found that the expression of 737 transcripts was significantly reduced in atria of NppaCre;CT3 mice, whereas 809 transcripts were significantly induced (threshold: p value <0.05).

Geneset analysis and GO term selection

The Global test [23] was used to test for significant association of specific GO terms with the differences in phenotypes between the NppaCre;CT3 and NppaCre mice. A series of selection criteria was applied to reduce the list of significant GO terms, i.e., Benjamini–Yekutieli FDR at 15% and the number of genes in the geneset between 10 and 75. The individual influence of each gene on the test statistics was calculated and used as an additional level of GO term selection, i.e., at least five genes per geneset with an influence above 5, representing at least 5% of the geneset.

Quantitative real-time PCR analysis

Quantitative real-time PCR analysis was performed as described previously [13]. In short, total RNA was isolated from left atrial appendices of 4-week-old adult mice using the RNeasy mini kit according to the manufacturer’s protocol (Qiagen). cDNA was reverse transcribed from 300 ng total RNA using the superscript II system (Invitrogen) and expression of different genes was assayed with quantitative real-time PCR using the Roche 480 Lightcycler. Relative start concentration (N(0)) was calculated as described previously [24]. Values were normalized to Gapdh expression levels. Primers sequences are provided in the online supplemental table 6.

In situ hybridization and immunohistochemistry

Non-radioactive in situ hybridization was performed as described previously [25]. Probes for Bmp10, Sox9, Fbln2, and Id3 were kindly provided by Herman Neuhaus, Benoit de Crombrugghe and Yvette Chin. IMAGE cDNA constructs were kindly provided by Fred van Ruissen. The following IMAGE cDNA constructs were digested with SalI and labeled with T7 RNA polymerase to generate DIG-labeled probes used for in situ hybridization: Tagln (BC003795), Fbln2 (BC005443), Lum (BC005550), Nkd2 (BC019952), Meox1 (BC011082), Ednra (BC008277), and Aldh1b1 (BC020001). Fgf12, Cacna2d2, and Hnt probe construct were generated by PCR amplification using gene-specific primers and were subsequently cloned into pBluescript SK vector for DIG labeling. For immunohistochemistry, the same fixation protocol as for the in situ hybridization analysis was used. Primary antibodies used were as follows: anti-cleaved caspase 3 (Cell Signaling, #9661 polyclonal) and anti-phosphohistone H3 (Cell Signaling, #9701 polyclonal).

ChIP data analysis

Conditional TBX3 over-expressing and cardiac specific tamoxifen inducible Cre (Mer-Cre-Mer) mice have been described before [13, 17]. Male mouse hearts were isolated 4 days after intra-peritoneal injections of tamoxifen, and TBX3 over-expression was confirmed by qRT-PCR, in situ hybridization and immunohistochemistry (not shown). ChIP was performed on mouse hearts using anti-TBX3 (A-20, Santa-Cruz). In this case, Mer-Cre-Mer mice, lacking the TBX3 expression construct, injected with tamoxifen served as ChIP control. Isolated DNA fragments were analyzed using high-throughput sequencing (data and analysis will be published elsewhere). Data significance of TBX3 binding peaks were analyzed using Fisher’s exact test with comparison to ChIP control data.

Transcription factor binding site prediction

To identify potential T-box binding sites, high-quality position weight matrices from Jaspar database were used (http://jaspar.genereg.net/; MA0009.1 for T-box binding sites). The relative score threshold was set to 70%.

Results

Tbx2 and Tbx3 are redundantly required for AVC patterning

Previous analyses indicated that Tbx2 is necessary and sufficient to suppress chamber gene expression in the AVC [10, 11, 26]. However, in our Tbx2 loss-of-function mouse AVC formation at E9.5 was grossly normal, and natriuretic peptide type A (Nppa/ANF) and other chamber markers were not ectopically expressed in the AVC [10]. Tbx2 is co-expressed in the AVC myocardium with the closely related family member Tbx3 arguing for functional redundancy of the two genes in this region. We wished to test this hypothesis by generating mice double mutant for Tbx2 and Tbx3. This effort was largely hampered by the fact that mice double heterozygous for Tbx2 and Tbx3 null alleles with a high penetrance exhibited postnatal lethality due to craniofacial defects [27]. Nonetheless, we managed to obtain some double-heterozygous animals for further interbreeding. Viable Tbx2 −/− ;Tbx3 −/− embryos from few litters at E9.5 appeared slightly retarded in their development, most likely due to arising hemodynamic failure. Morphologically, the constriction between the left ventricle and the atrium was largely absent (Fig. 1). Markers of chamber myocardium [Nppa, gap junction protein, alpha 5 (Gja5/Cx40)] [26] were expanded into this region whereas homozygous mutants for either Tbx2 or Tbx3 and double heterozygous mutants showed proper chamber gene repression in the AVC (Fig. 1; data not shown). Cre from the mutant alleles reflecting endogenous expression from either Tbx2 or Tbx3 or both in the absence of Tbx2 or Tbx3 protein, was present, although reduced in Tbx2 −/− ;Tbx3 −/− embryos (Fig. 1, arrows), indicating that early AVC specification had occurred.

Fig. 1
figure 1

Combined loss of Tbx2 and Tbx3 abrogates myocardial patterning of the atrioventricular canal. Comparative analysis of wild-type, Tbx2 −/−, Tbx2 −/− ;Tbx3 −/− and Tbx3 −/− embryos for cardiac morphology and molecular marker expression at E9.5. Left lateral views of whole E9.5 embryos and enlarged hearts (boxed regions in upper row shown in second row) reveal growth retardation and dilated avc phenotype in Tbx2/Tbx3 double mutant embryos. In situ hybridization analysis of marker gene expression in sagittal sections through the avc with probes as indicated. avc atrioventricular canal, la left atrium, lv left ventricle

Full size image

A Bmp2-Tbx2/3 feed-forward loop is required for EMT and formation of cushion mesenchyme in the AVC

Histological analysis of embryos compound mutant for null alleles of Tbx2 and Tbx3 revealed that loss of more than two functional alleles resulted in reduction of cardiac jelly and absence of AV cushion formation (Fig. 2). Bmp2 expression in the AVC myocardium is both required and sufficient to induce cushion formation and EMT of the adjacent endocardium [2, 3, 6]. Both Tbx2 and Tbx3 were downregulated in Nkx2-5 Cre ;Bmp2 floxneo/floxneo embryos, in which Bmp2 was inactivated in the heart (Fig. 3a), suggesting that the two genes are downstream mediators of Bmp signaling in this region. Expression of Bmp2 was normal in single mutants but reduced in compound mutants indicating the presence of a feed-forward loop between Bmp2 and Tbx2/Tbx3 (Fig. 3b). Consistently, expression of transforming growth factor, beta 2 (Tgfb2), a Bmp2 target in this tissue that is required for cushion formation and of hyaluronan synthase 2 (Has2), required for cardiac jelly formation and AV cushion development [2, 28] was downregulated in the AVC endocardium and myocardium of both Nkx2–5 Cre ;Bmp2 floxneo/floxneo embryos and compound Tbx2/3 mutants (Online Fig. 1). Hence, Tbx2 and Tbx3 are required in a redundant fashion to maintain the primary myocardium of the AVC, to maintain Bmp2 expression, and to induce the formation of cushion tissue from the endocardium in this region.

Fig. 2
figure 2

Combined loss of more than two alleles of Tbx2 and Tbx3 abrogates cushion formation in the AVC. Histological analysis of sagittal sections through the left atrium (la), atrioventricular canal (avc), and left ventricle (lv) by hematoxylin and eosin (H and E) staining shows normal cushion formation upon loss of one or two functional alleles of Tbx2 and Tbx3 (a, c, e), whereas loss of more than two functional alleles of the two genes results in partial loss of cardiac jelly and complete loss of AV cushion tissue (b, d, f). avc atrioventricular canal, a atrium, lv left ventricle

Full size image
Fig. 3
figure 3

A Bmp2-Tbx2/3 regulatory feed-forward loop for expression in the atrioventricular canal. a Inactivation of Bmp2 in cardioblasts or embryonic cardiomyocytes by Nkx2–5 Cre results in loss of Tbx2 and Tbx3 expression in the atrioventricular canal (asterisk). b In both compound Tbx2/Tbx3 mutants and Bmp2 conditional mutants, the atrioventricular constriction is largely absent. Bmp2 expression is lost from Tbx2/Tbx3 compound mutants, whereas reduced levels of Bmp2 transcripts (lacking exon 3; [3]) can still be found in the Bmp2 conditional mutants (asterisk). avc atrioventricular canal, a atrium, lv left ventricle

Full size image

Myocardial Tbx2 and Tbx3 induce mesenchymal cushion formation

To further elucidate the role of Tbx2 and Tbx3 in AVC and cushion development we used gain-of-function approaches to ectopically express Tbx2 and Tbx3 during early heart development. We crossed mice that harbored a Cre-activatable Tbx2 expression cassette in the transcriptionally competent Hprt locus (HpCT2) with Myh6Cre mice (Fig. 4a). Double-heterozygous Myh6Cre;HpCT2 embryos survived until E13.5–E14.5. Widespread activation of Tbx2 in the myocardium of the embryonic heart occurred relatively late at low levels, explaining the formation of chambers and the lack of early lethality observed previously using a constitutive over-expression approach [26]. Histological and in situ hybridization analyses revealed that ectopically expressed Tbx2 suppressed the expression of chamber markers Gja5 and Nppa at E11.5 as expected (Fig. 4b and not shown). Formation of ectopic sub-endocardial mesenchyme was observed in the atria, and to a lesser extent in the ventricles, which might be less susceptible to signals initiating cushion formation. Bmp2 was strongly induced in the atrial myocardium, whereas Notch gene homolog 1 (Notch1), a Bmp2 target which regulates EMT in the endocardium [2, 4], and Has2, were induced in the endocardium overlying the transgenic Tbx2 + myocardium (Fig. 4b).

Fig. 4
figure 4

Cardiac misexpression of Tbx2 or Tbx3 induces cardiac jelly and cushion formation in chamber myocardium. a Use of Myh6Cre driver to ectopically activate Tbx2 in the myocardium. b Robust activation of Tbx2 (Gfp) in Myh6Cre;HpCT2 leads to loss of Nppa and induction of Bmp2 in the myocardium, and Fbln2, Has2 and Notch1 in endocardium/mesenchyme (red arrows). c Transgenic constructs used to activate Tbx3 in the myocardium of the embryonic heart. d Immunohistochemical analysis of proliferation (PHH3) in E11.5 hearts of Myh6Cre;CT3 mice compared to control (CT3) mice. Black arrows depict Phospho-H3 positive cells in the ventricular myocardium of control mice and Myh6Cre;CT3 mice. Black bar, 100 μm. e In situ hybridization of serial sections in E11.5 hearts of Myh6Cre;CT3 mice compared to control mice. Snai1 and Has2 are induced in the endocardium and subendocardial mesenchyme formed in the atria of Myh6Cre;CT3 embryos (asterisk). avc atrioventricular canal, l/ra left/right atrium, l/rv left/right ventricle

Full size image

We next wished to test whether induction of endocardial cushion formation by Tbx2 depends on Bmp2 signaling. For this purpose, we cultured explanted E10.5 Myh6Cre;HpCT2 atria for 4 days in the absence or presence of 30 μM Dorsomorphin, an inhibitor of Bmp type 1 receptors. qRT-PCR assays revealed that all markers for EMT and cushion formation tested were down-regulated in the presence of Dorsomorphin (Online Fig. 2), providing further evidence that Tbx2-mediated EMT and cushion formation depends on Bmp-signaling.

To investigate whether Tbx3 is also sufficient to induce cushion formation, we ectopically activated Tbx3 expression in the developing heart using the Myh6Cre line and the Cre-activatable transgenic Tbx3 cassette reported on earlier CAG-CAT-TBX3 (CT3) [13] (Fig. 4c). Loss of Gja5 and Bmp10 expression confirmed functional TBX3 overexpression in double heterozygous (Myh6Cre;CT3) mice. Myh6Cre;CT3 embryos featured hypoplastic chamber walls, and the fraction of phospho-Histone 3-positive proliferating cells in the ventricular walls was reduced (Fig. 4d). In contrast, ectopic TBX3 expression did not result in induced apoptosis as defined by TUNEL assays (not shown). These results suggest that Tbx3 contributes to the low proliferation of the AVC myocardium. Bmp2 was ectopically activated in the myocardium of Myh6Cre;CT3 embryos, whereas snail homolog 1 (Snai1), a Notch1/Tgfb2 target required for AVC EMT[4] and Has2 were ectopically activated in the overlying endocardium (Fig. 4e). Moreover, a large subendocardial space formed that, closer to the AV cushions, was filled with mesenchymal cells expressing the cushion marker MAD homolog 6 (Smad6) (* in Fig. 4e; Online Fig. 3).

We next assessed the consequences of prolonged cardiac expression of Tbx3 using NppaCre to activate Tbx3 from E10 onwards in the atrial myocardium in the CT3 mice (Fig. 5a) [13]. E17.5 fetuses formed a thick sub-endocardial layer of cells in the atria. This tissue expressed mesenchymal marker genes including actin, alpha 2, smooth muscle, aorta (Acta2/α-SMA), fibulin 2 (Fbln2), versican (Cspg2/Vcan) and lumican (Lum) (Fig. 5b–d, Supplemental Table 1) that are associated with AV cushions. Furthermore, Tgfb2, Bmp6, inhibitor of DNA binding 3 (Id3) and SMAD family member 6 (Smad6) were induced in the endocardial layers (Fig. 5c, Supplemental Fig. 4, Supplemental Table 1). Taken together, both Tbx2 and Tbx3 in myocardium are sufficient to induce cushion formation and EMT in the adjacent endocardium.

Fig. 5
figure 5

Myocardial Tbx3 expression induces endocardial mesenchyme formation and nodal gene expression. a Transgenic constructs used to activate TBX3 in the developing atria. b, c Sections of E17.5 atria of CT3 and NppaCre;CT3 mice were probed for expression of indicated genes. Black arrowhead indicates the sinus node (san), white arrowhead the myocardium. Red arrowheads depict the thick endocardial mesenchymal layer that forms in NppaCre;CT3 atria. Black bar, 100 μm. d qRT-PCR analysis of left atria of NppaCre;CT3 double-transgenic mice compared to CT3 control mice. Expression levels in CT3 atria were set to 1. *p < 0.05. ra right atrium, san sinus node

Full size image

Identification of endocardial mesenchyme gene programs downstream of myocardial Tbx3

We used the NppaCre;CT3 transgenic model to gain deeper insight into the mechanism of gene regulation by Tbx3. Genome-wide microarray analyses (Illumina MouseRef-6 oligonucleotide BeadChips, 47,769 different oligonucleotides) were performed, comparing the atrial gene expression profile of six male double-transgenic NppaCre;CT3 mice and six male NppaCre control mice. Expression of 737 transcripts was significantly reduced in atria of NppaCre;CT3 mice, whereas 809 transcripts were significantly induced (threshold: p value <0.05). Components of the Tgfβ-, Bmp-, Fgf- and Wnt-signaling pathways that have been functionally implicated in AV cushion and valve formation [2] were highly up-regulated in atria of NppaCre;CT3 compared to NppaCre control mice (Supplemental Table 1). Furthermore, GO terms associated with EMT, such as TGF-β receptor signaling pathway, collagen and ECM were over represented in genes with overall higher expression in atria of NppaCre;CT3 (Supplemental Table 2). In addition, qRT-PCR and in situ hybridization analysis confirmed expression of genes in the subendocardial mesenchyme of NppaCre;CT3 mice (Twist1, Msx1, Meox1, Sox9, Id3, and Smad6) (Fig. 5d, Supplemental Fig. 4, Supplemental Table 1), whose expression and functional relevance in EMT and cushion formation in the AVC have been reported [2–4]. Furthermore, expression of Fgfr2, and of Wnt antagonists Frzb, Sfrp2 and Nkd2 was detected in the cushion mesenchyme and up-regulated in atria of NppaCre;CT3 mice, compatible with the established requirement for Fgf- and Wnt-signaling in cushion and valve formation (Fig. 5d, Supplemental Fig. 5, Supplemental Table 1) [2]. Pkd2 is normally expressed in the developing valves and was induced in atrial mesenchyme of NppaCre;CT3 mice (Fig. 5d). In human and mouse, mutations of Pkd2 result in valve abnormalities [29]. An additional 47 induced genes were identified in the microarray data whose specific expression in the fetal AV valvular mesenchyme was reported by Genepaint (http://www.genepaint.org/) (Supplemental Table 3). Furthermore, we provide a list of genes associated with cushion formation and EMT, as generated with the literature-base gene annotation tool Anni 2.0 [30] in Supplemental Table 4. This list contains established (e.g. Tgfb2, Gata4) as well as potential new players in cardiac cushion formation/EMT. Gli pathogenesis related protein-2 (Glipr-2), for instance, has been shown to be up-regulated during tissue fibrosis, a common pathway in the progression of many chronic disease states, and have the capacity to induce EMT in renal epithelial cells [31]. Further, both Tnc (Tenascin-C), Timp1 (Tissue inhibitor of matrix metalloproteinase 1) and Hgf have been linked to EMT via their involvement in the remodeling processes of the extracellular matrix and effects on cell motility during cellular migration [31–33].

Tbx3 imposes a nodal gene program on atrial myocardium

We observed down-regulation of known and novel chamber-specific genes in Tbx3-expressing atria of NppaCre;CT3 mice (Supplemental Table 1). qRT-PCR and in situ hybridization analysis confirmed normal chamber-restricted expression and TBX3-mediated atrial repression of Ckm [34], Nppb [35], Aldh1b1 and Ednra [36], and of Bmp10 [37] (Fig. 5d, Supplemental Fig. 4, 5). Furthermore, atria of NppaCre;CT3 mice had reduced expression of genes involved in muscle contractility, such as sarcomere complex genes and genes associated with mitochondria and their energy metabolism (Online Table 1). Consistently, compared to working myocytes, nodal (conduction system) myocytes feature a much poorer myofibril differentiation and sparser mitochondria [1].

Microarray analysis and subsequent validation by qRT-PCR and/or in situ hybridization revealed induction of genes in NppaCre;CT3 atria including Bmp2, Cx45/Gja7, Itpr1, Slco3A1, Id2, Cacna2d2, and Hnt (Fig. 5b–d, Supplemental Fig. 5, Supplemental Table 1), normally enriched in conduction system components, including the AV node. These genes are associated with (Slco3A1) or involved in the formation (Id2, Bmp2) or function (Cx45, Itpr1) of the conduction system [1, 3, 38–40]. Regulatory sequences of the Slco3A1 gene are thought to drive reporter gene expression of the cardiac conduction system reporter (CCS)-lacZ mouse strain [39]. Calcium channel subunit Cacna2d2 and Hnt, which are enriched in the sinus and AV node of adult mice [41] and in the developing cardiac conduction system, including the AVC (Online Fig. 5), were strongly induced in the atria of NppaCre;CT3 mice (Fig. 5, Supplemental Table 1).

We complemented the microarray analysis by performing ChIP-seq assays on hearts of mice in which Tbx3 was activated in the myocardium using tamoxifen-induced Cre recombination (due to technical limitations we were not able to obtain ChIP-seq data for endogenous Tbx3 from embryonic hearts). We observed several thousand peaks in the genome of Myh6-MerCreMer;CT3 mice, whereas tamoxifen-treated Myh6-MerCreMer mice did not yield specifically localized peaks. Interaction with T-box binding elements (TBEs) previously identified by mutational and in vitro binding analyses (Nppa, Myh6, Gjd3/Cx30.2, Id2) [38, 42–45], confirms the quality of the ChIP-seq data, and implies that Tbx3 directly regulates these genes in vivo (Fig. 6a, b, Supplemental Fig. 6). Moreover, binding of Tbx5 and Gata4, derived from recently published resource of ChIP-seq data from the heart derived cell line HL-1 (GEO: GSE21529) [46] coincided with Tbx3 peaks found in these loci (Fig. 6a, b). Analysis of the overlap between the microarray and the Tbx3 ChIP-seq data sets (online supplemental table 5) revealed a significant enrichment of Tbx3 binding peaks associated with both up- and down-regulated genes. Furthermore, down-regulated genes were associated with significantly more Tbx3 binding peaks than up-regulated genes, in keeping with the notion that Tbx3 acts as a strong transcriptional repressor in vitro and in cell culture. Overlap of the Tbx3-enriched datasets with the Tbx5 ChIP-seq data obtained in HL1 cardiac-like cells revealed an average association of 90% of these enriched genes with Tbx5 [46]. The observed interaction with the region of Nppb, Hcn4, Acta2, Cacna2d2, Bmp2 and Bmp4 suggests that T-box factors also directly regulate these genes (Fig. 6b, c; Online Fig. 7). Two genes that are not influenced by Tbx3 in the myocardium but indirectly in the endocardium, Has2 and Sox9, were indeed devoid of any Tbx3 peaks (Online Fig. 7).

Fig. 6
figure 6

Tbx3 ChIP-seq reveals interaction with known and novel binding sites in vivo. a Tbx3 binds to the region upstream of Cx30.2 (Gjd3), a region shown to function as a T-box responsive enhancer [44]. The relative position and sequence of the published T-box binding elements and conservation are shown at the bottom of the panel (red). Recently published Gata4 and Tbx5 ChIP-seq data from HL-1 cells [46] also shows specific binding within this region. b The TBE of Nppa [42, 43], is occupied by Tbx3 in atria and by Tbx5 and Gata4 in the HL-1 cell line [46]. Tbx3, Tbx5 and Gata4 binding peaks can also be observed upstream of Nppb, a gene showing a similar expression profile and response to Tbx3. c Tbx3 binding peaks surrounding the Bmp2 gene. The upstream element, shown to bind both Tbx3 and P300 [47], was cloned (underlined in green) upstream of a minimal promoter driving luciferase. Putative T-binding sequence is shown at the bottom of the panel. d Luciferase induction in H10 cells, relative to the minimal promoter only (Co), by the putative enhancer (TBE) described in c. The activity of this element can be modulated by the presence of Tbx3 and Tbx5

Full size image

Closer inspection of the Bmp2 locus revealed one particular genomic site with which Tbx3 interacts (Fig. 6c). In the developing heart, enhancer associated co-factor p300 was also found to interact with this region (Fig. 6c) [47]. When isolated and tested in transfection assays, this fragment stimulated reporter gene expression in cardiac-like H10 cells and responded to T-box factors (Fig. 6d). These data suggest that Bmp2 may be directly regulated by T-box factors including Tbx3 in the myocardium.

Discussion

Tbx2 and Tbx3 regulate chamber versus AVC development

Our study reveals that the T-box transcription factor Tbx2 together with Tbx3 locally repress chamber differentiation, stimulate development of the AVC myocardium and AV nodal phenotype, and induce AV cushion development, providing a mechanism for the co-localization and coordination of these important processes in heart development.

Tbx2 and Tbx3 are an evolutionary closely related pair of T-box proteins that share identical biochemical properties and are co-expressed in the AVC (reviewed in [48]). Hence, Tbx2 and Tbx3 may be functionally redundant in their requirement for AVC establishment. Our data confirmed this hypothesis. Individual loss of function of either Tbx2 or Tbx3 did not have a major impact on AVC development (Fig. 1) [10, 11, 14, 49, 50]. However, Tbx2/3 double mutant embryos largely failed to establish a morphological AVC. Noteworthy, the Cre inserts in the Tbx2 and Tbx3 loci are expressed in the putative AVC region of double mutants, indicating that the upstream regulatory pathway for AVC localized Tbx2/3 activation (most likely involving Bmp2) has been active. Bmp2 in the AVC myocardium induces Tbx2 through Smads in vivo [3, 7, 8], and was also required for Tbx3 expression (Fig. 3a). Both Tbx2 and Tbx3, in turn, were found to be required and sufficient to maintain expression of Bmp2. These findings indicate that a feed-forward loop activates and maintains Bmp2 and Tbx2/3 expression, respectively, in the AVC. We anticipate the presence of at least one more layer of regulation for AVC localized expression upstream of Bmp2, as Bmp2 (lacking exon 3) expression itself in the Bmp2 conditional KO is localized in the putative AVC region (Fig. 3) [3, 7, 8]. The redundancy of the paralogous T-box factor genes is limited to sites of co-expression. In the AVC, Tbx2 is expressed in a slightly broader and left-sided dominant manner, and indeed, Tbx2 mutants display a left-sided AVC malformation that functions as an accessory pathway causing ventricular pre-excitation [12]. Tbx3 expression is unique in the sinus node and AV bundle, which are severely affected in Tbx3 mutants [1]. Moreover, expression of Tbx3 is maintained in the AVC, unlike Tbx2 and Bmp2, and may be responsible for the maintenance of the pacemaker-like tissue of the AV conduction system. The double dose of the redundant T-box factors in the AVC may serve to firmly establish the AVC phenotype during early stages of cardiogenesis.

Gain-of-function scenarios revealed that both Tbx2 and Tbx3 are able to suppress expression of specific chamber marker genes and differentiation [13, 26, 51, 52]. Using expression profiling, we now identified of a broad spectrum of working myocardium associated genes, including sarcomere components and mitochondrial genes, suppressed by Tbx3, and induction of genes associated with the pacemaker/conduction system in the atrial working myocardium. These programs underlie the typical phenotype of AVC/conduction system myocardium [1], suggesting that Tbx3 to a large degree is responsible for this pacemaker phenotype. The ChIP-seq analysis identified binding sites of Tbx3 in genes both up- and down-regulated by Tbx3 in myocardium, implying that Tbx3 directly regulates these genes in vivo. Both Tbx2 and Tbx3 have context-dependent repression and activation domains, and can interact with multiple co-factors that may provide repression or activation activities to the protein complex [53]. The regulatory DNA regions that control expression of Bmp2 and other genes in the heart in vivo have not been established [54]. Whether the T-box factor-sensitive enhancer we identified regulates cardiac Bmp2 expression in vivo remains to be established.

Tbx2 and Tbx3 induce AV cushion development

Embryos with less than two alleles of Tbx2/Tbx3 failed to establish AV cushions, which are the precursors of the valves, and contribute to the septal structures and to the fibrous insulation. Conversely, ectopic expression of either Tbx2 or Tbx3 in the myocardium of the developing heart caused the initiation of cushion formation. These observations revealed that Tbx2 and Tbx3 in the AVC are redundantly required and sufficient to initiate cushion formation in the AVC. The mechanism of cushion formation has been extensively studied, and important roles for ligands and receptors of the Tgf-β-superfamily have been exposed [2]. Bmp2 expression in the AVC myocardium is both required and sufficient to induce cushion formation [3, 6, 7]. It activates Notch1, Twist1 and Tgfb2 in the endocardium, which subsequently regulate Snai1,2 and EMT [2, 4]. A previous report implicated Tbx2 in the direct activation of Tgfb2 and Has2 in myocardium, which subsequently induce cardiac jelly formation and cushion development. In that report, Tbx2 did not induce Bmp2 or Bmp4 [9]. Our data do not confirm this model, and indicate a different mechanism. Inactivation of Tbx2/3 caused loss of Bmp2 expression and vice versa, Tbx2 and Tbx3 in myocardium directly activate Bmp2 (and Bmp4). Further, inactivation of Tbx2/3 and of Bmp2 resulted in very similar phenotypes, which included failed AV cushion development and loss of expression of Tgfb2 and Has2. These data indicate that Tbx2/3 and Bmp2 expression are interdependent and part of the same regulatory network for AVC development. Furthermore, Tgfb2, Has2 and other critical components in the regulation of cushion development (e.g., Notch1, Snai1)[2–4] were activated indirectly and selectively in the endocardium by myocardial Tbx2/3. Consistently, our ChIP-seq data indicated that Tbx3 does not interact with the putative TBEs in Has2 or Tgfb2 [9], although the latter contain multiple other sites of Tbx3 interaction (Supplemental Fig. 7). Thus, endocardial activation by Tbx2/3 occurs via Bmp2 or another paracrine signal. Together, our data indicate a mechanism in which Tbx2/3 and Bmp2 maintain expression of each other in a feed-forward loop, and provide evidence that Bmp2 induces expression of genes involved in cushion development and EMT in the endocardium.