A strategy to discover new organizers identifies a putative heart organizer

Organizers are regions of the embryo that can both induce new fates and impart pattern on other regions. So far, surprisingly few organizers have been discovered, considering the number of patterned tissue types generated during development. This may be because their discovery has relied on transplantation and ablation experiments. Here we describe a new approach, using chick embryos, to discover organizers based on a common gene expression signature, and use it to uncover the anterior intestinal portal (AIP) endoderm as a putative heart organizer. We show that the AIP can induce cardiac identity from non-cardiac mesoderm and that it can pattern this by specifying ventricular and suppressing atrial regional identity. We also uncover some of the signals responsible. The method holds promise as a tool to discover other novel organizers acting during development.

O rganizers are formally defined as signalling regions, unique in being able both to induce and to pattern adjacent tissue 1 . The dorsal lip of the blastopore, which can induce a complete secondary axis in amphibian embryos, was the first organizer to be discovered, in 1924 (ref. 2). Given the complexity of the vertebrate embryo, one might expect that many organizers should exist, but only a few others have been described 3 : Hensen's node (the amniote equivalent of the dorsal lip of the blastopore, which can induce and pattern the central nervous system) 4 , the notochord/floor-plate (which can induce and organize different sets of neurons in the neural tube) 5 , the zone of polarizing activity (ZPA, which can induce a patterned set of limb elements) 6 and the midbrain-hindbrain boundary (isthmus, which can specify and pattern the adjacent regions of the midbrain/tectum and hindbrain/cerebellum) 7 . One reason why so few organizers have been found could be that their discovery requires grafting appropriate tissues at the right time and place, and there are too many possible combinations.
It has been shown that in some cases organizers can substitute for each other. Most strikingly, a graft of Hensen's node from an early (primitive streak stage) embryo into the anterior limb bud of a much later embryo can mimic the action of the limb organizer, the ZPA, by inducing and patterning a full set of skeletal elements including the digits 8 . This raises the intriguing possibility that organizers may share a genetic signature ('synexpression') that confers them with inducing and patterning properties. If this is the case, we should be able to use this genetic signature to point to potential new organizing regions. In this study, we tested this idea by comparing the transcriptomes of three known amniote organizers (Hensen's node, the notochord/floor-plate and the ZPA). This defines a synexpression set of 48 transcripts that are either enriched or depleted in organizers, which we then used to explore the embryo for other regions of synexpression. This suggested that the endoderm of the anterior intestinal portal (AIP) might be a new organizer. The AIP is an endodermal invagination appearing with the head-fold, which moves caudally down the embryo to form the foregut 9 . As it does so, it is closely associated with the developing heart tube. We therefore performed a series of experiments to test whether the AIP can act as an organizer of the heart. Ablation and transplantation experiments in vivo, along with co-culture in vitro, revealed that the AIP endoderm can induce heart fate from non-cardiac mesoderm as well as pattern the heart field by specifying ventricular and suppressing atrial regional identity.

Results
Defining a gene signature for organizers. To investigate whether organizers have a common genetic signature of enriched and depleted transcripts, we designed a differential microarray screen in chick embryos, comparing three known organizers to their most similar non-organizer tissue. First, we chose stage (HH 10 ) 3 þ /4 Hensen's node, which induces and patterns the neural plate 11 , compared with the posterior primitive streak, which cannot (Fig. 1a). We also sampled the later node (HH6), which has lost its ability to induce a full neural plate 11 . This three-way comparison was designed to reveal transcripts associated with full organizer function. Second, we selected the notochord and floor-plate at HH10-11, responsible for organizing the dorso-ventral axis of the neural tube 5 , and compared this to the dorsal neural tube (Fig. 1b). Finally, we compared the posterior wing bud, containing the ZPA 6 , to the anterior wing bud at HH20-21 and HH24 (Fig. 1c). Each comparison generated a list of transcripts significantly enriched or depleted in that organizer (Z1.2-fold the log 2 of the expression level, and a false discovery rate r0.05. The dataset was submitted to ArrayExpress with the title 'Microarray analysis of chick embryonic tissues: gastrulation, neural tube/notochord and limb development' and given accession number E-MTAB-4048). These lists were then combined using a Boolean algorithm to find enriched or depleted transcripts common to all three organizers. This approach uncovers relative changes in expression between  Fig. 1c). One possible interpretation of this is that the synexpression set is enriched for signalling molecules emitted by the organizer and depleted of transcriptional repressors that suppress the organizer state.
The AIP endoderm as a candidate organizer. Next, we used this set to verify the expression of the selected genes ( Fig. 1d- There is substantial evidence that the early endoderm adjacent to the bilateral cardiac mesoderm is required for normal heart formation in chick, amphibian and fish embryos 13,14 , but less is known about the AIP endoderm of later stages. Genetic or manual ablation of the AIP endoderm in mouse and chick embryos results in cardia bifida [15][16][17][18][19] (failure of the bilateral heart progenitors to fuse in the midline), rotation defects 16 and downregulation of early cardiac genes 15 , indicating that a functional AIP endoderm is required for normal heart formation, but its inducing and patterning abilities have not been tested. Could the AIP be an organizer of the heart?
The AIP can induce cardiac identity from non-cardiac mesoderm. We first confirmed that ablation of the AIP at HH8 does indeed cause cardia bifida and abnormal heart rotation ( Supplementary  Fig. 5). Next, to test if the AIP can induce cardiac identity, a necessary prerequisite is to identify a suitable responding tissue. Fate and specification maps indicate that the paraxial mesoderm adjacent to Hensen's node at HH5 (prospective head mesoderm) is neither fated nor specified as heart [20][21][22][23][24] , but is competent to respond to cardiac inducing signals 22 . We compared this mesoderm (named #3) with four other regions of HH5 mesoderm ( Supplementary Figs 6a and 7a): #3 does not contribute to the heart when grafted homotopically ( Supplementary Fig. 7b-h), and is the only region of mesoderm that does not express cardiac markers when explanted either in vitro (Fig. 2 Table 2). Mesoderm #3 is therefore suitable for testing whether the early AIP can induce heart fate. The AIP induces early cardiac markers, MYOCD, NKX2.5  Tables 2 and 3). Strikingly, spontaneous beating is observed after 48 h of in vitro co-culture of #3-mesoderm with AIP (Supplementary Movie 1), but not with control, non-AIP endoderm. Together these results show that the AIP endoderm can induce cardiac identity in mesoderm that is not otherwise fated to become part of the heart.
Patterning by inducing ventricular identity and suppressing atrial character. Many of the known early heart markers are normally expressed in the bilateral cardiac mesoderm before the AIP forms (around stage HH5; Supplementary Fig. 6a-f), whereas regional markers of anterior-posterior heart tube patterning are observed later (after HH9; Supplementary Fig. 6v-n 0 ). Ventricular markers (VMHC1/MYH15, IRX4 and NPPB) are initially expressed in the medial splanchnic mesoderm [25][26][27] immediately adjacent to the AIP (Supplementary Fig. 6d 0 ,f 0 ,g 0 ), and the atrial marker AMHC1 is in the lateral splanchnic mesoderm 28 ( Supplementary Fig. 6c 0 ). In #3-mesoderm explants cultured either in a host embryo or in vitro, AIP induces VMHC1/MYH15, IRX4 and patchy NPPB expression ( Fig. 2b-d Previous tissue recombination experiments using HH5-6 chick explants have demonstrated that anterior-lateral endoderm adjacent to the early cardiogenic mesoderm can induce heart gene expression in posterior primitive streak 29 . Are the induction properties of the HH8 AIP endoderm distinct to the early anterior-lateral endoderm? To test this, we co-cultured HH5 anterior-lateral endoderm with non-cardiac #3-mesoderm (Supplementary Fig. 9a and Supplementary Table 4). Unlike AIP endoderm, anterior-lateral endoderm does not induce ventricular markers (Supplementary Fig. 9b-e) nor does it induce AMHC1 ( Supplementary Fig. 9f). These data show that the anterior-lateral endoderm is different to the AIP endoderm and that the ability to induce ventricular character is a property specific to the AIP endoderm.
To test if the AIP patterns the heart tube by specifying ventricular identity, AIP from transgenic GFP donors was grafted over the cardiogenic mesoderm of a host embryo (Fig. 3a). Strikingly, this expands the expression of VMHC1/MYH15 and IRX4 posteriorly into the prospective atrial region ( Fig. 3c-l), while expression of the atrial marker AMHC1 is abolished ( Fig. 3m-q). Moreover, the AIP represses AMHC1 in #1-mesoderm which would normally express it ( Fig. 3r-t and Supplementary Table 3); since IRX4 is known to repress AMHC1 (refs 26,30), this repression by the AIP may be either direct or indirect, due to prior AIP-mediated induction of IRX4. Thus, the AIP not only induces ventricular identity, but also actively blocks atrial character. The ability to induce a new fate in responding tissues and to impart a spatial pattern are properties shared by other known organizers 3 . Taken together, these results show that the AIP endoderm is distinct from other endodermal regions, in that it can both induce cardiac identity in mesoderm not destined for this fate and also pattern heart mesoderm by promoting ventricular and suppressing atrial regional character.
Common signals between organizers. Grafts of either Hensen's node 8 , notochord or floor-plate 31 into the anterior limb bud can induce extra digits, suggesting that at least some signals (an obvious one being SHH) may be common to different organizers. We therefore tested whether the heart-inducing or ventricle-specifying functions of the AIP can be mimicked by other organizers. We find that Hensen's node induces patchy expression of VMHC1/MYH15 but not MYOCD in #3-mesoderm cultured in a host ( Supplementary Fig. 10a,b and Supplementary  Table 5), as well as in explants ( Supplementary Fig. 10c,d and Supplementary Table 5). Hensen's node has also been shown to induce ectopic VMHC1, but not AMHC1 in the germinal crescent and the lateral plate mesoderm 32 , comparable to the ability of the AIP to induce ventricular markers. Conversely, to test whether the AIP can mimic the neural inducing activity of Hensen's node, we grafted AIP endoderm into the area opaca 33 at HH3 þ : this induces the early neural marker SOX3, but not the later neural plate marker SOX2 ( Supplementary Fig. 10e,f). Finally, we tested whether the AIP can mimic a ZPA graft into the anterior wing bud: strikingly, the AIP causes duplication of digit-2 ( Supplementary Fig. 10h). Therefore, the AIP is partly interchangeable with other organizers at earlier and later stages of development, some of which could be related to its expression of SHH (Fig. 1l) Figure 2 | AIP induces cardiac and ventricular identity in non-cardiac mesoderm. (a) AIP or control (Cont.), lateral endoderm from a HH8 quail co-cultured with anterior-medial mesoderm (#3) from a HH5 chick embryo overnight in the anterior area opaca of a host chick. (b-g) Regional myocardium markers VMHC1/MYH15 (b), IRX4 (c), NPPB (d) and GJA5 (e) are induced in #3-mesoderm by quail (brown) AIP (red arrowheads), but not by control endoderm (Cont., black arrows), whereas AMHC1 (f) and SHOX2 (g) are not induced (black arrows). (h) In vitro co-culture of #3-mesoderm with GFP-AIP (j,m,p,s,v,y,b',e') or -Cont. (k,n,q,t,w,z,c',f') for 24 (j,k,m,n) or 48 h (p,q,s,t,v,w,y,z,b',c',e',f'). No expression is seen in #3-mesoderm cultured alone (i,l,o,r,u,x,a',d'). AIP (brown) induces MYOCD (j), NKX2.5 (m), VMHC1/MYH15 (p), IRX4 (s), NPPB (v) and GJA5 (y) in #3-mesoderm (red arrowheads), but Cont. does not (brown, k,n,q,t,w,z, black arrows). Neither AMHC1 nor SHOX2 are induced (black arrows) by AIP (b',e') or by Cont. (c',f'). *Pr0.05, **Pr0.005, ***Pr0.0005 using two-tailed Fisher's exact test. Scale bars, 0.5 mm in whole-mounts and 0.1 mm in insets, explants and sections. acting together and/or sequentially [35][36][37]39) and SHH (refs 40,41) play important roles in other organizers; both have been implicated in cardiogenesis 34,42 and both are expressed in the AIP. However, our synexpression set contains many other secreted or membrane-associated factors that have not hitherto been implicated either in the activity of other organizers or in cardiac gene induction or heart patterning. Four of these are expressed in the early AIP: NRP1, FBLN7, KIRREL3 and VTN (Fig. 1j,k,m,n). To investigate their activity, we co-cultured #3-mesoderm with pellets of cells transfected with expression plasmids encoding these four factors. This combination partly mimics the AIP, sometimes inducing patchy expression of the ventricular markers VMHC1/MYH15 (Fig. 4a) and NPPB (Fig. 4b), but not IRX4 (Supplementary Fig. 11f) or other early cardiac markers (Fig. 4c, Supplementary Fig. 11a-e and Supplementary Table 6).
These results suggest that other signals are required, possibly SHH and FGF8, but also considerable complexity, not unlike other organizers, which are believed to emit different signals at different times to account for their various inducing and patterning functions [35][36][37] . Accordingly, some gene expression profiles change over time as the organizer changes its properties. Although HH8 AIP endoderm was used in co-cultures, the AIP continues its developmental programme, expressing transcripts that are detected in the later AIP ( Supplementary Fig. 12). HH8 AIP cultured either alone or with #3-mesoderm expresses genes, including those that encode secreted or membrane-associated molecules normally observed in the HH10 AIP ( Supplementary  Fig. 4j,k), after 24 h (Supplementary Fig. 12a-d) and those expressed in the HH12-13 AIP (Supplementary Fig. 4q,s) after 48 h (Supplementary Fig. 12e-h). These results suggest that the signalling properties of the AIP endoderm mature over the culture period as they do in the embryo.
Taken together, our results implicate the AIP as an organizer of the heart, and suggest that a complex combination of signals account for its heart inducing and patterning functions at early stages of heart tube formation.

Discussion
Our results suggest that organizers do share a common genetic signature, and that this property can be used to identify putative new organizers acting during development. The choice of the three organizers to compare was relatively arbitrary and perhaps we should have used a different well-characterized organizer, the mid-hindbrain boundary (MHB) or isthmic organizer 3,7 instead of the comparison between notochord/floor-plate with dorsal neural tube (particularly as the latter is also a signalling region). The main reason why we did not use the MHB is that it is difficult  to identify a non-organizer tissue that is similar enough to this boundary in every other respect. The organizer gene set could be refined further using this tissue in future. Nevertheless, this study is an important proof of principle and, based on the synexpression of 48 genes, we uncover the AIP endoderm as a putative organizer because it can both induce cardiac identity and pattern the heart field by specifying regional characteristics (inducing ventricle and suppressing atrial character).
Previous studies have concluded that the endoderm is required for heart development (reviewed in ref. 13). Many of these studies investigated the early anterior-lateral endoderm adjacent to the cardiac mesoderm at HH4-6 (refs 34, [43][44][45][46][47][48]. Tissue recombination experiments have shown that this endoderm can reprogram non-cardiac tissues to heart identity 29 and BMPs, FGFs and inhibition of WNT signalling have been implicated in this 13,22,32,34,[49][50][51][52][53] (Fig. 5a). Some recombination experiments have used posterior primitive streak or posterior lateral plate (#4-5; Supplementary Figs 6a and 7a) explants as a responding tissue to test for cardiac induction 29,49,50,53 . However, it has since been shown that these tissues express MYOCD (ref.  Table 1) leaving open the possibility that these experiments do not demonstrate a true change of fate in this tissue. Widespread expression of some early cardiac markers makes it difficult to find a responding tissue suitable for an induction assay (neither being fated to give rise to heart nor express any of the markers being assessed as a result of the induction). Here we use anterior-medial mesoderm (#3), which is neither fated nor specified as heart [20][21][22][23][24] and can be used to assess both induction and patterning rigorously.
The expression patterns of early cardiac markers, like NKX2.5 (ref. 29), and the signalling molecules previously implicated in their induction (BMP2/4/7 (ref. 22), FGF8 (ref. 34) and CRESCENT (ref. 49)) indicate that initial signals act at an early stage of cardiogenesis (HH4-6, Fig. 5a) before the AIP forms. Our results show that the AIP from a later stage embryo can also induce heart identity (reprogramming non-heart #3-mesoderm from an earlier stage donor) as well as pattern it by specifying ventricular and suppressing atrial character. During normal development, the AIP is likely to be responsible for the later patterning functions (Fig. 5b), but our results show that it can also mimic the activity of earlier endoderm. This could be due to the presence of AIP precursors within this early endoderm 21,55,56 and the continued expression of BMP2 (ref. 22) and FGF8 (ref. 34). This is reminiscent of other organizers, for example, Hensen's node can initiate the entire process of neural induction, even though the first neural inducing signals begin before gastrulation, long before Hensen's node can be defined 38 . This could be a general feature of organizers, perhaps because they produce many signals. However, the signals emitted by organizers also change over time as their patterning properties evolve, which is seen in organizers like Hensen's node 11,36 , as well as in the AIP.
Heart progenitor cells become determined to either the ventricular or atrial compartment at around HH8 (refs 28,57,58), consistent with the timing of signals from the AIP endoderm that induce ventricular and repress atrial character (Fig. 5b). Retinoic acid signalling from the posterior lateral plate mesoderm is known to control the allocation of cardiac progenitor cells to the atrial lineage 28,57,59 . We have shown that signals from the AIP endoderm, NRP1, FBLN7, KIRREL3 and VTN when combined can sometimes induce VMHC1/MYH15 and NPPB (Fig. 5c), but whether these signals also repress atrial identity remains to be tested. BMP2/4, FGF4 and Wnt inhibitors have also been shown to induce VMHC1 (refs 22,32,49), although the results of these studies do not rule out induction of heart identity rather than regionalization. Additional regional markers need to be tested to resolve this. Interestingly, both VTN and NRP1 have other, much later roles in heart patterning.
There is conflicting evidence about the extent to which cells that make up the AIP roll around it (involution) and turn over rapidly, or whether they remain in place as it advances caudally   28,57,58 . The AIP endoderm induces ventricular character and represses atrial identity. Signals from the AIP endoderm, including NRP1, FBLN7, KIRREL3 and VTN described here (red text and arrows) and previously reported BMP2, FGF4 and Wnt inhibitors 22,32,49 (black text and arrows), induce some ventricular markers, which are integrated with atrial-inducing retinoic acid signalling from the posterior lateral plate mesoderm 28,57,59 . The signals from the AIP endoderm that repress atrial identity are unknown. (c) At HH9, when regional markers of anterior-posterior heart tube patterning begin to be expressed, AIP endoderm induces ventricular markers VMHC1/MYH15, IRX4, NPPB and GJA5 (the latter expressed from HH12) and represses the atrial marker AMHC1. Four secreted molecules from the AIP endoderm, NRP1, FBLN7, KIRREL3 and VTN can induce VMHC1/MYH15 and NPPB.
down the embryo 19,55,60 . Neither possibility is incompatible with the AIP being an organizer. The cellular composition of Hensen's node constantly changes as cells move in and out of it 61 but it also contains resident cells 62 and similar findings have been made for the ZPA of the limb 63 . In all these cases the organizer property, as well as molecular markers for it, remain in the region rather than moving with the cells: cells regulate their expression and functional properties according to their current position 61 . These findings raise the general principle that organizers define a position within the embryo where a set of properties come together. Using a 'synexpression' gene set for this state should not only allow us to seek new organizers, but also to investigate the clues that instruct cells to have such properties in particular places in the embryo. Our novel approach of identifying genes common to different organizers offers great potential not only for uncovering new organizing centres during development, but also as a tool to identify new signalling molecules underlying organizer function, as well as transcription factors that confer cells with this property. Interestingly, we find that the synexpression set is enriched for signalling molecules (15 of the 31 enriched genes), whereas transcription factors are more abundant among the depleted genes (6/17), suggesting that some transcriptional repressors may function to suppress the organizer state in non-organizer tissues.
Hybridization to Affymetrix microarrays was conducted by ARK-Genomics at the Roslin Institute, University of Edinburgh, as described in the Affymetrix GeneChip Expression Analysis Technical Manual (Affymetrix, Inc.). Briefly, 15 mg of total RNA from each sample was reverse-transcribed using a T7-oligo (dT) primer in the first-strand complementary DNA (cDNA) synthesis. After RNAase-H treatment, second-strand cDNA synthesis was carried out. The cDNA was then purified and used as a template in the subsequent in vitro transcription reaction for linear amplification of each transcript, and incorporation of biotinylated CTP and UTP using T7 RNA polymerase. The biotinylated cRNA targets were fragmented and hybridized to the Affymetrix GeneChip Chicken Genome Array, which provides good coverage of the chicken genome (32,773 transcripts corresponding to over 28,000 genes). Twenty-seven arrays were used in total (samples from 9 tissues, each in triplicate). Hybridization was performed at 45°C for 16 h with constant rotation (60 rpm). The microarrays were then automatically washed and stained with Streptavidin-phycoerythrin conjugate (SAPE; Invitrogen) in a GeneChip fluidics station (Affymetrix). Fluorescence intensities were measured with a GeneArray scanner 3,000 (Affymetrix). The scanned images were analysed as described in the Affymetrix GeneChip Command Console 3.0 User Manual.
Gene expression data generated from the GeneChip software (GCOS) were normalized using the probe logarithmic intensity error (PLIER) method 65 within the Affymetrix Expression Console software package. The normalized data were then analysed using the Limma and FARMS 66 packages within R in Bioconductor. Venn diagrams were generated using the R package Vennerable with dependencies, graph, RBGL, gplots, gtools and reshape. Weights used for scaling the Venn diagrams were based on a priori count data reflecting intersections of genes between different comparisons of datasets. Probes with a fold change (FC) Z1.2 (log 2 ) and a false discovery rate r0.05 were deemed significant. The results of these individual comparisons (using significant enrichment or depletion as the indicator, rather than absolute amounts) were then combined using the Boolean algorithm: ((HH4 HN versus HH4 PS AND (HH4 HN versus HH6 HN) OR HH6 HN versus HH4 PS)) AND VNT versus DNT AND (HH20 PL versus HH20 AL OR HH24 PL versus HH24 AL). Probes that are differentially enriched or depleted significantly in this algorithm comprise the organizer gene set. The first set of comparisons selects genes enriched or depleted in the early (HH3 þ /4) Hensen's node compared with the posterior streak, some of which remain expressed the same way at the later (HH5/6) stage. The comparisons for the limb select genes enriched or depleted in the posterior limb bud (containing the ZPA) at either HH20/21 or HH24.
The dataset was submitted to ArrayExpress with the title 'Microarray analysis of chick embryonic tissues: gastrulation, neural tube/notochord and limb development' and given accession number E-MTAB-4048.
Embryo manipulation. Fertilized Brown Bovan Gold hens' eggs (Henry Stewart, UK), Coturnix coturnix japonica quails' eggs (B.C. Potter, Rosedean Farm, UK, and Blue Bridge Engineering Limited, Essex, UK) and transgenic GFP chicken eggs 67 (Roslin Institute, Edinburgh, UK) were used. Embryo manipulations were performed in Pannett-Compton saline and modified New culture 68,69 . Tissues were excised from donor embryos using 0.12% w/v trypsin (porcine trypsin 1:250, Sigma) in Pannett-Compton saline and 30 G syringe needles, washed with saline and kept on ice until needed. Mesoderm (#1-5, from both the left and the right sides and used randomly) was harvested from HH5 À to HH5 þ donor embryos. Early AIP endoderm (AIP) and control (Cont.) lateral endoderm (the endoderm underlying the somites and proximal lateral plate mesoderm) were harvested from HH8 À to HH8 þ embryos. Early AIP endoderm was used because cardiac progenitors are closely associated with the AIP at this time, and because it is easier to isolate the AIP at this stage than later, when the heart is beating. Anterior-lateral endoderm (#1-2E) and anterior-medial endoderm (#3E) were harvested from HH5 donor embryos. Host embryos ranged from HH4 þ to HH6. Grafts were placed on the left or the right side of host embryos; no differences were noted in the results. After grafting, the embryos were incubated for either 6-9 h or overnight (12-18 h). Limb grafts were performed in ovo at HH20-21 and incubated for 5 days 70 .
In vitro culture of explants. Tissues were excised from chicken and quail embryos as described above, and cultured in vitro for 24 or 48 h at 37°C with 5% CO 2 on a 0.4 mm Nucleopore filter (Whatman, WHA110407) floating on DMEM (Gibco) with 10% FBS, 5% chicken embryo extract (US Biological), 1% GlutaMAX (Gibco) and 1% Penicillin/Streptomycin (Gibco) 29 . Tissues were arranged such that the basal surface of the AIP faced the mesoderm explants. Spontaneous beating in 48 h explant co-cultures was recorded with a QImaging Retiga 2000R camera and QCapture Pro software.
Probes and antibodies. Whole-mount in situ hybridization and immunohistochemistry were performed as described 73,74 . Probes used were as follows:  86), kindly provided by the originating laboratories. AMHC1 and NKX2.5 were kindly provided by T. Brand and GATA4 by B. Pain. The cDNA available for VMHC1 (MYH15) 27 is a 2.5 kb cDNA, which also has significant (73%) homology to MYH7b (Slow Myosin 2, SM2); therefore, a short 0.6 kb probe (VMHC1/MYH15, generated after cutting with PstI, where the two sequences differ most) was used instead (see VMHC1/MYH15 short in Supplementary Fig. 6). Other probes, including MEF2C (ChEST776g19), NPPB (ANF; ChEST509m15) and GJA5 (CX40; ChEST304a4), were transcribed from EST clones (Supplementary Data 1) 87 . Following whole-mount in situ hybridization, embryos were incubated in QCPN (Developmental Studies Hybridoma Bank, diluted 1:5) and/or anti-GFP antibody (Life Technologies, A11122, diluted 1:2,000); secondary antibodies used were goat anti-mouse IgG peroxidase (Jackson, 115-035-003), goat anti-rabbit IgG peroxidase (Santa Cruz, sc-2004) or goat anti-rabbit IgG alkaline phosphatase (Upstate, 12-448) diluted 1:1,000. Cryosections of 15 mm or paraffin sections of 10 mm were prepared. For skeletal staining, embryos were stained with Alcian Blue 88 .  Tables 1-6) were considered 'positive' if high-level expression was observed throughout the majority of the explant. The n-numbers in the figures represent those that were scored as positive. The n-numbers that include positive and patchy expression are indicated. Explants with no detectable expression or very weak expression (perhaps due to non-specific staining, constitutive expression and/or marginal induction) were scored as negative. Explants with high-level expression in a small patch (some of which could be due to contamination) were excluded from P-value calculations, unless indicated. To calculate P values, positive (plus patchy where specified) and negative counts were compared across two conditions (#3 alone versus #3 þ AIP or #3 þ Cont.) using two-tailed Fisher's exact test with a 2 Â 2 contingency table. P values r0.05 were deemed significant.
Data availability. The microarray data have been deposited in the ArrayExpress database under accession code E-MTAB-4048. All other data supporting the findings of this study are available within the article and its Supplementary information files or from the corresponding author upon reasonable request.