The ventral to dorsal BMP activity gradient in the early zebrafish embryo is determined by graded expression of BMP ligands.

In the early zebrafish embryo, a ventral to dorsal gradient of bone morphogenetic protein (BMP) activity is established, which is essential for the specification of cell fates along this axis. To visualise and mechanistically determine how this BMP activity gradient forms, we have used a transgenic zebrafish line that expresses monomeric red fluorescent protein (mRFP) under the control of well-characterised BMP responsive elements. We demonstrate that mRFP expression in this line faithfully reports BMP and GDF signalling at both early and late stages of development. Taking advantage of the unstable nature of mRFP transcripts, we use in situ hybridisation to reveal the dynamic spatio-temporal pattern of BMP activity and establish the timing and sequence of events that lead to the formation of the BMP activity gradient. We show that the BMP transcriptional activity gradient is established between 30% and 40% epiboly stages and that it is preceded by graded mRNA expression of the BMP ligands. Both Dharma and FGF signalling contribute to graded bmp transcription during these early stages and it is subsequently maintained through autocrine BMP signalling. We show that BMP2B protein is also expressed in a gradient as early as blastula stages, but do not find any evidence of diffusion of this BMP to generate the BMP transcriptional activity gradient. Thus, in contrast to diffusion/transport-based models of BMP gradient formation in Drosophila, our results indicate that the establishment of the BMP activity gradient in early zebrafish embryos is determined by graded expression of the BMP ligands.

Introduction

During the development of multicellular organisms, a recurrent mechanism for tissue patterning is the formation of morphogen gradients. The original definition of a morphogen is a molecule produced at a localised source which then diffuses into the surrounding tissue to provide positional information and specify different cell fates in a dose-dependent manner (Wolpert, 2011). Bone morphogenetic proteins (BMPs) have been described as morphogens and BMP gradients have been documented in various developing organisms such as the sea urchin, Drosophila, Xenopus, zebrafish, and mouse (Ramel and Hill, 2012). The best understood models of BMP gradient formation are the Drosophila BMP activity gradients. In the Drosophila embryo, the BMP gradient that is required for the specification of dorsal and lateral tissues involves the redistribution of the BMP ligands within a uniform expression domain (O'Connor et al., 2006). In the developing Drosophila wing, a BMP gradient forms across the Anterior/Posterior (A/P) axis which extends beyond the source of the BMP ligand Decapentaplegic (Dpp; BMP4 orthologue). It is still debated whether this involves free diffusion, restricted diffusion, and/or transcytosis of Dpp (Erickson, 2011; Kicheva and Gonzalez-Gaitan, 2008; Schwank et al., 2011; Yan and Lin, 2009). Much less is understood about how BMP gradients are formed in vertebrate embryos. There is increasing evidence that the establishment of these gradients is highly complex and finely regulated (Wolpert, 2011), and that mechanisms operating in Drosophila are not sufficient to explain them. A prerequisite for studying gradients in early vertebrate embryos is a sensitive molecular tool for directly visualising them.

The BMPs, along with the related growth and differentiation factors (GDFs), constitute a subfamily of the transforming growth factor β (TGF-β) superfamily (Schmierer and Hill, 2007). BMPs and GDFs signal through heteromeric serine/threonine kinase receptor complexes comprising type II and type I receptors (also called ALKs). Ligand binding promotes phosphorylation and activation of the type I receptor by the type II receptor, leading to phosphorylation of a subset of receptor-activated Smads (R-Smads) (Moustakas and Heldin, 2009). Once activated, the R-Smads form heteromeric complexes with Smad4, which accumulate in the nucleus and regulate target gene transcription. Importantly, the Smads constantly shuttle between the cytoplasm and the nucleus in both the presence and absence of signalling and the levels of activated complexes in the nucleus are determined by the relative activities of the receptor kinases in the cytoplasm and a nuclear phosphatase. In the presence of signal the Smad nucleocytoplasmic shuttling provides a sensing mechanism for receptor activity (Schmierer and Hill, 2007). For many years, TGF-β superfamily signalling was divided into two distinct branches with respect to the R-Smads: BMPs and GDFs were thought to signal exclusively through Smad1, Smad5 and Smad8, whilst TGF-β, Activin and Nodals signalled through Smad2 and Smad3 (Schmierer and Hill, 2007). However, it is now established that in most cell types TGF-β additionally robustly activates Smad1 and Smad5 (Bharathy et al., 2008; Daly et al., 2008; Goumans et al., 2002; Liu et al., 2009; Wrighton et al., 2009). Hence, monitoring the phosphorylation status of a particular R-Smad is not a reliable way to discriminate signalling by different ligands. A more specific method is to exploit reporter plasmids with binding sites for transcription factors specifically activated by the pathway in question. For BMP/GDF pathways, such a reporter was generated in which BMP responsive elements (BREs) from the mouse Id1 enhancer, which bind phosphorylated Smad1/5–Smad4 complexes, drive luciferase expression (Korchynskyi and ten Dijke, 2002). Most importantly, this BRE-luciferase reporter exclusively responds to BMP/GDF signals and not to TGF-β signals, despite the latter's ability to phosphorylate and activate Smad1/5 (Daly et al., 2008; Gronroos et al., 2012; Liu et al., 2009). We have recently generated a fluorescent reporter transgenic zebrafish line using a modified version of this BRE-luciferase reporter (BRE-mRFP) (Wu et al., 2011).

In zebrafish embryos, a BMP activity gradient that is stronger ventrally during the blastula and gastrula stages is required to pattern the embryo along its dorsal/ventral (D/V) axis. The existence of a BMP activity gradient is evident when monitoring the expression of BMP target genes during gastrulation, such as the bmp ligands themselves or the ventral ectoderm markers gata2 and ΔNP63 (Bakkers et al., 2002; Nguyen et al., 1998). The first direct visualisation of the zebrafish BMP activity gradient was obtained using anti-phosphorylated Smad1/5 (p-Smad1/5) immunostaining (Tucker et al., 2008). This study concluded that a ventral to dorsal gradient of nuclear p-Smad1/5 is set up between 4.75 h and 5 h post-fertilisation (hpf; 30–40% epiboly stages). However, it is not known if the newly established BMP activity gradient (as detected with p-Smad1/5) is also reflected in transcriptional output as early as 30–40% epiboly, nor how it is formed.

Here we exploit our BRE-mRFP transgenic reporter line to determine the mechanism by which the ventral to dorsal BMP activity gradient is established. We first definitively prove that mRFP expression faithfully reports BMP and GDF signalling at both early and late stages of development. Then, taking advantage of the unstable nature of mRFP mRNA and the sensitivity of its detection, we readily visualise the ventral to dorsal BMP transcriptional activity gradient that is generated during blastula stages, showing it is established by the 40% epiboly stage. We demonstrate using single and double in situ hybridisation (ISH) that the graded BMP transcriptional activity occurs in a pre-established domain of graded bmp ligand transcripts. Using anti-BMP2B antibody staining, we also show that graded bmp ligand transcription results in graded protein expression. Moreover, in contrast to the mechanisms of BMP gradient formation in Drosophila, our results provide no evidence for diffusion or re-distribution of the BMP ligands during BMP gradient formation. Finally, we define the sequence of signalling events that lead to graded bmp transcription to generate a robust BMP activity gradient.

Materials and methods

Fish maintenance, screening, strains and crosses

The zebrafish colony was maintained as described (Westerfield, 2000). Founders for the BRE-mRFP transgene (20% of F0 screened, n=49) were identified by crossing injected adult zebrafish to wild-type and screening their progeny for mRFP fluorescence at 24 hpf. Experiments described here used F2, F3, or F4 progeny of five different founders. The swr TA72A allele is a strong loss-of-function allele corresponding to an ENU-induced point mutation in the zebrafish bmp2b termination codon (Kishimoto et al., 1997).

Plasmids, mRNAs, morpholinos and injections

The mRFP ORF was amplified from pcDNA-mRFP-N and cloned into pGL3-BRE-luciferase (Korchynskyi and ten Dijke, 2002). BRE-mRFP was then amplified from pGL3-BRE-mRFP and cloned into pT2KXIGΔin. To generate the BRE-mRFP transgenic line the pT2KXIGΔin-BRE-mRFP plasmid DNA (20 ng/μl) was co-injected with transposase mRNA (25 ng/μl) as described (Kawakami et al., 2004; Urasaki et al., 2006). The noggin1 and chordin ORFs were amplified from mid-gastrulation stage cDNA and cloned into pGEM-T and pCS2+, respectively. The following mRNAs were injected: CA-alk8 (7.5 ng/μl) (Bauer et al., 2001), bmp2b (50 ng/μl) (Wu et al., 2011), chordin (320 ng/μl), membrane mRFP (50 ng/μl) (Kieserman and Wallingford, 2009) and cherry H2B (80 ng/μl) (Woolner and Papalopulu, 2012). The following MOs (Genetools, LLC) were used: dharma (Shih et al., 2010), chordin (Nasevicius and Ekker, 2000), gdf6a (French et al., 2009), bmp2b (Imai and Talbot, 2001) and bmp4 (Chocron et al., 2007). All MOs were injected at 6 ng/nl except the gdf6A and bmp2b MOs which were at 3 ng/nl. A volume of approximately 2.5 nl of DNA, mRNA and/or MO was injected per embryo.

In situ hybridisation, immunofluorescence, and movies

The following antisense probes were used: mRFP (Wu et al., 2011), dharma (Koos and Ho, 1998), noggin1, chordin (Miller-Bertoglio et al., 1997), goosecoid (Stachel et al., 1993), bmp2b (Kishimoto et al., 1997), bmp4 (Ramel, 2005), and bmp7a (Schmid et al., 2000). The standard ISH protocol was essentially as described (Jowett, 2001) with the addition of 5% dextran sulphate to the hybridisation buffer (Lauter et al., 2011a). The double fluorescent ISH (DFISH) using DIG- and DNP-labelled probes and revealed with fluorescein tyramide and Fast Blue substrates was modified from Lauter et al. (2011b) (detailed protocol available upon request). For p-Smad1/5 (Cell Signalling 9511, 1:100) immunofluorescence (IF), the protocol was as described (Tucker et al., 2008). The BMP2B (Tebu-Bio 55707, 1:500) and pMAPK (Sigma M8159, 1:500) IF protocols required an antigen retrieval step (Inoue and Wittbrodt, 2011). For ISH combined with IF, ISH was performed first. After ISH substrate development (Cy5 tyramide), embryos were washed in 1X PBS to stop the colourimetric reaction, then immersed in IF blocking solution before proceeding with IF. Embryos that had undergone colourimetric ISH were equilibrated in 80% glycerol and photographed using a Leica DFC420C camera mounted on a Leica MZ-FLIII dissecting microscope. DFISH, immunostained and/or BRE-mRFP samples were counterstained with DAPI (Roche 10 236 276 001), Hoechst 33342 (Invitrogen H1399), or DRAQ5 (Biostatus DR50200), mounted in 0.8% low melt agarose (Sigma-Aldrich A9045) in glass-bottom microwell dishes (MatTek P35G-1.5–14-C) and imaged using a Zeiss LSM710 inverted confocal microscope. Movies were acquired using a Zeiss LSM710 inverted or a Zeiss LSM700 upright confocal microscope and live embryos were mounted in low-melt agarose in fish water supplemented with anaesthetic. Supplementary movie 4 was obtained using a Nikon ECLIPSE TE2000-E inverted wide field microscope and an Andor iXonEM+ DU-888 back-illuminated EMCCD scientific camera. Images and movies were acquired and/or processed with the following software packages: Leica Application Suite, Zeiss Zen, Zeiss LSM Image Browser, Metamorph, Image J and Adobe Photoshop.

Protein preparation and Western blotting

Embryonic protein extracts were prepared as described (Wu et al., 2011). The following antibodies were used: rabbit anti-mRFP (Invitrogen R10367) (Fig. S3), mouse anti-mRFP (SignalChem R46-61M-100) (Fig. S2), rabbit anti-pSmad2 (a mixture of Millipore AB3849 and 04953), mouse anti-pMAPK (Sigma M8159), mouse anti-Actin (Sigma 3853) and goat anti-MCM6 (Santa Cruz sc-9843).

Chemical inhibitor treatments

LDN-193189 (gift from Paul Yu), SU5402 (Calbiochem 572631), and SB-505124 (Tocris Bioscience 3263) stocks were made in DMSO. These inhibitors were diluted in fish water to obtain the desired final concentration. 8- to 16-cell stage embryos were immersed in fish water supplemented with either DMSO (control) or the chemical inhibitors.

Results

Characterisation of mRFP expression in BRE-mRFP embryos

To study BMP transcriptional activity and gradient formation during zebrafish development, we used our previously generated BRE-mRFP reporter transgenic line (Wu et al., 2011), employing both ISH for mRFP and direct observation of mRFP fluorescence in this line to assess BMP/GDF activity.

To observe early BMP signalling, we visualised mRFP mRNA, which gives a very sensitive and dynamic readout. mRFP RNA is expressed ubiquitously in 8-cell stage embryos ( Fig. 1A), suggesting that some mRFP mRNA is maternally deposited. mRFP mRNA is also ubiquitous during blastula stages before and after the onset of zygotic transcription, at the 512-cell and sphere stages, respectively (Fig. 1B). During gastrulation, mRFP transcripts are strongly reduced dorsally and are enriched ventrally in both the mesoderm and ectoderm, and thus transgenic embryos display a clear ventral to dorsal BMP activity gradient (Fig. 1C). During somitogenesis, mRFP expression is highly dynamic (Fig. 1D, S1). Most notably, mRFP mRNA is detected in the newly formed somites, the developing eyes, tailbud and presomitic mesoderm, consistent with studies implicating BMP activity in these tissues (Esterberg et al., 2008; French et al., 2009; Patterson et al., 2010; Pyati et al., 2006). Finally, at 28 hpf, mRFP mRNA is found in additional tissues such as the pineal gland, the developing heart, and the pectoral fin buds (Fig. 1E).

Fig. 1

mRNA expression in BRE-zebrafish embryos ((A)–(E)) ISH for mRFP mRNA in BRE-mRFP embryos. In the rightmost panel of (C), double ISH with mRFP and the dorsal marker goosecoid (gsc). V, ventral; D, dorsal; vs, ventral somites; ds, dorsal somites; tb, tailbud; psm, presomitic mesoderm; pg, pineal gland; dr, dorsal retina; ht, heart tube; pf, pectoral fin bud; s, somites.

In BRE-mRFP embryos, mRFP fluorescence is only first visible around the 6-somite stage, when a strong signal is seen in the tail bud and presomitic mesoderm (Fig. S2A). This delay in detection is due to insufficient mRFP protein being synthesised during gastrulation for direct mRFP fluorescence visualisation, as full-length mRFP protein is virtually undetectable before the 6-somite stage using Western blotting (Fig. S2B). During somitogenesis, the strongest mRFP protein expression domains are the tail bud and the presomitic mesoderm (Fig. S2A; Videos S1,S2), thereby confirming the in situ results. Most notably, at 24 hpf, strong mRFP fluorescence is detected in the dorsal retina, the epiphysis/pineal gland, the epidermis, the cloaca, and the somites (Fig S2A and Video S3) (Alexander et al., 2011; Collery and Link, 2011; French et al., 2009; Laux et al., 2011; Patterson et al., 2010; Pyati et al., 2006; Row and Kimelman, 2009). In addition, mRFP fluorescence is found in the cells of the midbrain–hindbrain border and the trigeminal ganglia (Fig. S2A) (data not shown). Finally, mRFP fluorescence is detected in the atrioventricular canal of 50 hpf transgenic embryos ( Video S4), consistent with reported BMP activity during mammalian early heart valve formation in the BRE-GFP transgenic mouse (Monteiro et al., 2008).

Video S3

Supplementary material related to this article can be found online at doi:10.1016/j.ydbio.2013.03.003.

Video S4

Supplementary material related to this article can be found online at doi:10.1016/j.ydbio.2013.03.003.

The BRE-mRFP transgenic zebrafish is a bona fide in vivo reporter for BMP/GDF activity.

To establish unequivocally that mRFP expression reflects BMP/GDF signalling, we manipulated BMP signalling at different times during embryogenesis and assessed its effect on mRFP mRNA or protein levels.

Stimulation of BMP signalling using overexpression of BMP2B (a zygotic BMP ligand) (Nguyen et al., 1998) and the constitutively-active BMP receptor CA-ALK8 (Bauer et al., 2001) or knockdown of the BMP inhibitor Chordin (Schulte-Merker et al., 1997) resulted in increased mRFP expression and expansion of the mRFP domain to the dorsal side ( Fig. 2A). Conversely, overexpression of Chordin and the BMP antagonist Noggin (Zimmerman et al., 1996) completely abolished mRFP expression at 70–80% epiboly (Fig. 2A) (Wu et al., 2011). Thus, changes in BMP activity are reflected in mRFP expression along the D/V axis of the gastrulating embryo.

Fig. 2

mRFP expression in BRE-embryos is areadout for BMP and GDF activities (A) mRFP ISH in BRE-mRFP embryos injected with chordin, bmp2b or CA-alk8 mRNAs or chordin MO. (B) BRE-mRFP expression in swrTA72A/swrTA72A homozygous embryos compared with +/+ or swrTA72A/+ siblings. The embryos shown resulted from a cross between swrTA72A/+ heterozygotes that contained the BRE-mRFP transgene. (C) mRFP ISH in BRE-mRFP embryos injected with a control or bmp4 MO. Arrows point to the cloaca. (D) BRE-mRFP embryos were injected with a control or gdf6a MO and mRFP expression in the dorsal retina was analysed (fluorescence and ISH). pg, pineal gland; s, somites. In (A) and (B), lateral views are shown; V, ventral; D, dorsal; A, anterior; P, posterior. In all cases, stages are indicated and the number of embryos out of the total analysed that showed the presented staining pattern is given. For (B), this number represents the number of homozygous mutant embryos out of the total number of embryos analysed.

We also used a genetic approach to inhibit zygotic BMP activity, investigating mRFP expression in a bmp2b-deficient background (Fig. 2B). At 70% epiboly, mRFP expression is strongly reduced in homozygous swr TA72A (bmp2b −) embryos. From bud stage/early somitogenesis, swr TA72A/swr TA72A embryos display a striking elongated morphology (Mullins et al., 1996) unlike their swr TA72A/+ and wild-type siblings. In 4-somite stage swr TA72A homozygous embryos, mRFP expression is also strongly reduced. No difference in mRFP expression between wild-type and swr TA72A heterozygous siblings was detected (data not shown).

BMP activity is important at the end of gastrulation to specify ventro-posterior fates, including formation of the cloaca, the common opening of the urogenital and digestive tracts in zebrafish (Pyati et al., 2006, 2005). BMP4 is thought to be the critical zygotic ligand for the specification of cloacal fates as severe bmp4 ST72 homozygous zebrafish mutants do not form a proper cloaca (Stickney et al., 2007). Injection of a splice-blocking MO that targets bmp4 mRNA (Chocron et al., 2007) results in severely reduced mRFP expression in the cloaca of 28 hpf embryos, but strikingly, has no effect on mRFP expression in the somites where other BMP ligands must contribute to mRFP transcription (Fig. 2C).

The dorsal retina is another region of high mRFP expression (Fig. 1E, S1, S2A). The GDF6A ligand is required for p-Smad1/5 activation in the dorsal retina (French et al., 2009), suggesting that it might be the ligand responsible for mRFP expression in this tissue. Injection of a gdf6a splice-blocking MO results in reduced dorsal retina mRFP mRNA expression as well as reduced mRFP fluorescence in 26 hpf embryos (Fig. 2D), confirming that the BRE-mRFP transgenic line is also sensitive to changes in signalling by ligands from the GDF subfamily. We were unable to detect any effect of GDF6A knockdown on mRFP expression in other tissues such as the somites, despite a recent study implicating GDF6A as a factor required for somitic BMP activity during myogenesis (Nguyen-Chi et al., 2012).

Finally, we decreased BMP/GDF signalling globally in fish embryos using the chemical inhibitors Dorsomorphin and LDN-193189, which target the type I BMP/GDF receptors, and lead to a dorsalized phenotype (Cannon et al., 2010; Yu et al., 2008) (Fig. S3). The chemical inhibitors are autofluorescent, resulting in considerable background when mRFP fluorescence is visualised. However, inhibition of mRFP expression is readily detected by mRFP ISH and Western blotting for mRFP protein (Fig. S3).

Taking all the results in this section together, we conclude that mRFP expression in the BRE-mRFP transgenic embryos is an accurate readout of dynamic BMP/GDF activity at all stages of development examined.

The initial ventral to dorsal BMP transcriptional activity gradient is established between 30% and 40% epiboly

The ability to visualise the ventral to dorsal BMP activity gradient in the BRE-mRFP embryos by monitoring levels of mRFP mRNA gave us a unique opportunity to investigate in detail how this gradient forms.

From the 8-cell to the 30% epiboly stage, mRFP expression is uniform ( Fig. 3). The mRFP transcripts detected at the 8-, 256-, and 512-cell stage must be deposited maternally. In contrast, specific nuclear p-Smad1/5 staining is only apparent from the 512-cell stage onwards (Fig. 3) and must occur as a result of maternal BMP signalling initially (Goutel et al., 2000; Sidi et al., 2003), then as a result of zygotic BMP signalling. At 40% epiboly, mRFP mRNA is clearly enriched ventrally and expressed in a graded fashion in BRE-mRFP embryos (Fig. 3). Comparison of the p-Smad1/5 staining with the mRFP expression pattern revealed that decrease of p-Smad1/5 from the dorsal side at 30% epiboly fractionally preceded loss of mRFP expression (Fig. 3). Hence, the BMP transcriptional activity gradient (mRFP) is established between 30% and 40% epiboly and occurs minutes after ventral enrichment of p-Smad1/5. Notably, clearance of mRFP transcripts dorsally occurs in a very short time frame (~20 min), confirming that the mRFP mRNA half-life is very short. Thus, observing mRFP transcripts at these early stages provides a very dynamic readout of the BMP activity gradient. Moreover, this comparative experiment shows that in the case of early zebrafish embryogenesis, it is also possible to use p-Smad1/5 immunostaining as a readout for BMP pathway activation.

Fig. 3

Establishment of the BMP activity gradient in zebrafish embryos. p-Smad1/5 IF and mRFP ISH in early stages BRE-mRFP embryos. Stages and views are indicated. V, ventral; D, dorsal. Arrows indicate the extent of mRFP staining. Note that the DAPI staining at the 8-cell stage did not resolve to single nuclei as the cells are dividing very quickly.

The BMP transcriptional activity gradient is established in a domain of graded ligand transcription

Having determined exactly when the BMP activity gradient is initially formed, we explored the mechanism underlying it. The expression of zygotic bmp ligands is thought to be set up in two key steps. First, after the mid-blastula transition (MBT; ~3.25 hpf), a combination of the maternal BMP ligand GDF6A (possibly acting through maternal ALK8 and Smad5) and the transcription factor POU2 induce the ubiquitous expression of zygotic bmp2b and bmp7a, which is shortly followed by bmp4 expression (Langdon and Mullins, 2011). Second, the sequential action of dorsally-expressed negative regulators contributes to the increase in ventral expression of the BMP ligands. As early as sphere stage (~4 hpf), Dharma, a transcriptional repressor that is expressed in the dorsal-most blastomeres, directly represses bmp2b transcription (Koos and Ho, 1999; Leung et al., 2003). Subsequently, a combination of the dorsally-secreted BMP antagonists Noggin1 and Chordin acts to inhibit BMP signalling, with Chordin being essential for BMP signalling repression (Dal-Pra et al., 2006). It is not clear though when Chordin begins to confer its effect on BMP ligand expression. In addition, there is evidence that the FGF signalling pathway acts on the dorsal side to repress bmp transcription (Furthauer et al., 2004). As a result of all these interactions bmp ligand expression is clearly graded by the end of the blastula stage (Furthauer et al., 2004; Sidi et al., 2003). However, it is not known how the expression domain of the ligands relates to the BMP activity gradient or to transcriptional output during these critical stages.

To establish the sequence of events between dome stage and 40% epiboly, we compared the timing and pattern of mRFP mRNA expression with that of the three zygotic BMP ligands: BMP2B and BMP7A, which act as obligate heterodimers (Little and Mullins, 2009), and BMP4. We also investigated expression at the level of mRNA of the BMP antagonists expressed dorsally at blastula stages, Chordin and Noggin1, the transcriptional repressor Dharma, as well as dorsal FGF activity. At dome stage, dharma, noggin1 and chordin are all expressed dorsally, and in the case of noggin1 and chordin, expression increases at 30% and 40% epiboly ( Fig. 4A). We monitored FGF activity by immunostaining for di-phosphorylated MAPK (pMAPK) (Jurynec and Grunwald, 2010), and demonstrated that FGF is active at all stages examined, with stronger activity detected on the dorsal side. This is especially clear for pMAPK staining at dome stage (Fig. 4A). bmp2b and bmp7a transcripts are expressed uniformly throughout most of the embryo at dome stage, except on the most dorsal side, where a clear gap in staining is visible (Fig. 4A). At 30% and 40% epiboly, bmp2b and bmp7a show a clear asymmetric pattern, being stronger on the ventral side and weaker on the dorsal side (Fig. 4A), in agreement with reports that bmp2b and bmp7a are expressed uniformly upon activation of zygotic transcription, then become enriched ventrally (Hammerschmidt and Mullins, 2002; Leung et al., 2003). bmp4 expression, in contrast, is detected ubiquitously at dome stage (Fig. 4A) and only becomes ventrally enriched later at 30% and 40% epiboly (Fig. 4A) (Langdon and Mullins, 2011). Importantly, despite bmp ligands being expressed in a graded manner from 30% epiboly, mRFP transcripts remain uniform in the majority of embryos examined. In contrast, at 40% epiboly, the gradient of mRFP, reflecting BMP transcriptional activity, was evident, and strikingly, it appears very similar to the expression patterns of the ligands (Fig. 4A).

Fig. 4

The establishment of the BMP activity gradient occurs in a pre-established gradient of BMP ligand expression. (A) BRE-mRFP or wild-type embryos stained for mRFP, dharma, noggin1, chordin, bmp2b, bmp7a or bmp4 mRNA by ISH or immunostained for pMAPK. Arrows indicate the extent of the domain of staining for a particular mRNA. The number of embryos out of the total analysed that showed the presented staining pattern is given. (B) DFISH for mRFP and bmp2b in blastula stages BRE-mRFP embryos. The fluorescent substrates used were fluorescein tyramide (mRFP) and Fast Blue (bmp2b, pseudocoloured in red). The images are the projection of a 10–12 picture stack taken at the margin of the blastoderm to avoid fluorescence from the yolk. V, ventral; D, dorsal.

To confirm that the gradient of BMP ligand expression was the same as the gradient of BMP activity, we used DFISH for bmp2b and mRFP (Fig. 4B). As DFISH caused substantial autofluorescence from the yolk we visualised mRFP and bmp2b mRNAs at the blastoderm margin only. As expected, bmp2b is expressed in a ventral to dorsal gradient at both 30% and 40% epiboly and consistent with the single ISH results (Figs 3 and 4A), mRFP is only expressed in a gradient from 40% epiboly. DFISH shows a very good overlap of bmp2b and mRFP transcripts at 40% epiboly, confirming that at this stage the BMP transcriptional activity gradient occurs in a pre-established domain of graded ligand transcription.

Graded bmp transcription results in a BMP protein gradient with little or no transcriptional activity at a distance from BMP protein-expressing cells

Our observations that the transcripts of the three major BMP ligands are expressed in a gradient as early as 30% epiboly does not necessarily mean that the BMP proteins are also expressed in a gradient at that time. Indeed, post-transcriptional processing, diffusion, transport and other regulatory mechanisms could result in BMP protein expression differing from that of the bmp transcripts. As no data are available for endogenous BMP protein expression during early stages of zebrafish development, we used a new commercial zebrafish specific BMP2B antibody to visualise BMP2B protein expression.

Analysis of BMP2B protein expression revealed a graded staining pattern at all stages examined, from 30% to 75% epiboly ( Fig. 5A and data not shown). At 30% epiboly, strong BMP2B protein expression is detected on one side of the embryo, while BMP2B protein levels are much lower on the opposite side. At 60–70% epiboly, a clear gradient of BMP2B protein expression is detected in both the mesoderm and ectoderm. To confirm that the enrichment of BMP2B protein is indeed ventral, we compared BMP2B with pMAPK expression and found that strong BMP2B and pMAPK staining was obtained on opposite sides of the embryo at 30% epiboly (Fig. S4A). Using combined ISH for the dorsal marker chordin and IF for BMP2B, it was also evident that BMP2B proteins are enriched ventrally and expressed in a gradient during gastrulation (Fig. S4B). Close analysis of wild-type embryos injected with membrane mRFP mRNA revealed that the BMP2B protein is expressed in punctate cytoplasmic structures (Fig. S4C). Since the antibody used was raised against the prodomain region of zebrafish BMP2B (Tebu-Bio/Anaspec, personal communication), it is expected to detect pre-processed BMP2B. As a further check for specificity we performed IF in embryos overexpressing BMP2B and the nuclear marker Cherry H2B in a mosaic fashion and found that increased BMP2B immunoreactivity correlates with cells overexpressing BMP2B/Cherry H2B (Fig. S4D). We also injected embryos with a translation blocking bmp2b MO (Imai and Talbot, 2001) and examined BMP2B expression in the blastoderm at the 30% epiboly stage. BMP2B expression was reduced in the majority of the bmp2b morphants compared with controls (Fig. 5B).

Fig. 5

Graded expression of endogenous BMP2B protein results in graded transcriptional activity. (A) IF for BMP2B in wild-type embryos. (B) IF for BMP2B in control and bmp2b morphants. (C) Combined IF for BMP2B and ISH for mRFP in BRE-mRFP embryos. Note that there is some background staining for BMP2B that is due to autofluorescence from the yolk. Stages and views are indicated. The fluorescent substrate used for mRFP in (C) was Cy5 tyramide (pseudocoloured in red). In (A) and (C, bottom row), the pictures are the projection of a 25–30 stack of images taken from the edge of the embryo until about half way. In (B) and (C, top row), only the blastoderm margin was imaged. The white dotted line in (C, bottom row) shows the outline of the embryo. V, ventral; D, dorsal.

We have shown at 40% epiboly that the gradient of bmp2b mRNA expression matches the BMP activity gradient read out by mRFP mRNA (Fig. 4B). In contrast to what has been reported for Dpp in the Drosophila wing (Entchev et al., 2000; Gibson et al., 2002; Teleman and Cohen, 2000), this result strongly suggests that active BMP ligands do not diffuse significantly beyond the cells that synthesise them in the zebrafish embryo. Our ability to detect both an endogenous BMP protein and BMP transcriptional activity in the same embryo allowed us to determine whether transcriptional activity can be detected at a distance from BMP protein expressing cells. Using combined ISH for mRFP with IF for BMP2B in 40% and 70% epiboly BRE-mRFP embryos, we found that the expression of mRFP mRNA and BMP2B protein strongly overlap (Fig. 5C). Our antibody detects pre-processed BMP2B, and in principle, processed mature BMP2B, which we cannot detect, may diffuse further than the BMP2B-expressing cells. However, the fact that we do not detect any BMP transcriptional activity (mRFP) beyond cells expressing BMP2B suggests that processed BMP2B is not significantly diffusible.

For the first time therefore, we have shown the protein expression pattern of an endogenous BMP ligand in zebrafish. BMP2B protein is expressed in a ventral to dorsal gradient in the early zebrafish embryo, which recapitulates both the bmp2b mRNA expression, and the activity gradient read out by the mRFP reporter. Thus, taken together, our results provide no evidence that extensive diffusion of BMP2B protein occurs to generate the BMP transcriptional activity gradient detected with mRFP.

Graded bmp ligand transcription, established by dharma and FGF signalling and maintained through auto-regulation, determines the BMP activity gradient

Since we have shown that graded bmp transcription precedes the formation of the BMP activity gradient, we next investigated how graded bmp transcription is established. Transcription of the BMP ligands is known to be under autoregulation from about 60% epiboly (Kishimoto et al., 1997; Nguyen et al., 1998; Schmid et al., 2000). However, this does not appear to occur during blastula stages (Furthauer et al., 2004). To inhibit BMP signalling during blastula stages, we used LDN-193189 treatment from the 8–16 cell stage onwards (Fig. S3) (Cannon et al., 2010) and found that most embryos still retained graded bmp2b transcript expression at 30% epiboly ( Fig. 6A), despite a clear effect on overall transcript levels, probably due to reduced maternal BMP signalling. This result demonstrates that graded bmp2b transcription at 30% epiboly is not dependent on zygotic BMP activity. Consistent with this view, Chordin knockdown, which should lead to increased BMP signalling, did not result in expanded bmp2b expression at 30% epiboly (Fig. 6B) (Furthauer et al., 2004).

Fig. 6

Regulation of graded bmp ligand transcription by Dharma, FGF and BMP signalling and consequences for the transcriptional activity of the BMP pathway. (A) bmp2b ISH in 30% epiboly wild-type embryos treated with DMSO, LDN-193189 (LDN) or SU5402. (B) bmp2b ISH in 30% epiboly wild-type embryos injected with either control, dharma, or chordin MOs. Ectopic bmp2b expression is detected at the margin upon Dharma knockdown but not in the dorsal ectoderm (white dotted lines). (C) mRFP ISH in BRE-mRFP embryos injected with either control, dharma, or chordin MOs. The fluorescent substrate used in (C) was Fast Blue (pseudocoloured in red). Note that the images are projection of a 10–12 picture stack taken at the margin of the blastoderm to avoid fluorescence from the yolk. (D) Graphical representation of the levels of mRFP, chordin, and bmp2b mRNAs, as well as BMP and p-Smad1/5 proteins across the D/V axis at 30% and 40% epiboly. Between these two stages, Chordin acts to antagonise residual BMP activity dorsally, resulting in a gradient of mRFP at 40% epiboly that reflects the gradient of BMP ligand transcription. For all panels in ((A)–(C)), stages and/or views are indicated. Arrows indicate the limit of the expression domain of the transcripts analysed. V, ventral; D, dorsal.

Having ruled out positive feedback by autocrine BMP signalling as important for establishing graded bmp2b expression at 30% epiboly, we tested the involvement of Dharma and FGF signalling (Furthauer et al., 2004; Koos and Ho, 1999; Leung et al., 2003). The lack of bmp2b and bmp7a transcripts dorsally at dome stage is a strong indication of localised transcriptional repression by Dharma (Fig. 4A), and this was confirmed by the observation that injection of a dharma MO resulted in expanded bmp2b transcription at 30% epiboly in the majority of embryos examined (Fig. 6B). Notably, we found that ectopic bmp2b is only detected at the margin, where Dharma is expressed (Fig. 4A), but not in the dorsal ectoderm (see white dotted lines in Fig. 6B). Consistent with dorsal bmp2b being detected upon Dharma knockdown, we found that dharma morphants display increased mRFP expression dorsally at the margin in 40% epiboly embryos (Fig. 6C). Expression of pMAPK at dome and 30% epiboly stages indicates active FGF signalling in a dorsal to ventral gradient, with stronger activity occurring at dome stage (Fig. 4A). Using the FGF receptor inhibitor SU5402 (Mohammadi et al., 1997), we demonstrated that FGF signalling inhibition results in expansion of the domain of bmp2b expression to the dorsal side at 30% epiboly (Fig. 6A) (Furthauer et al., 2004). Interestingly, SU5402 treatment also results in reduced chordin expression at this stage (Fig. S5). Thus, Dharma and FGF signalling are the critical factors for establishing the graded bmp transcription that is present at 30% epiboly, rather than auto-regulation via BMP signalling.

Chronologically, it is generally assumed that activity gradients are established as a direct consequence of morphogen diffusion/transport and that therefore these gradients are formed de novo. However, our mRFP ISH results show that BMP activity in blastula stage embryos is initially ubiquitous, then graded (Fig. 3). Hence there must be a molecular mechanism that results in clearance of mRFP transcripts dorsally (Fig. 6D). Indeed, our data indicate that at 40% epiboly, when the gradient of BMP activity is established, it faithfully reflects the gradient of BMP ligand transcription (Fig. 4B). However, our mRFP ISHs and p-Smad1/5 staining indicates that BMP activity is initially ubiquitous and is cleared from the dorsal side between 30% and 40% epiboly (Fig. 3) (Tucker et al., 2008). This occurs after the bmp transcripts disappear dorsally, suggesting that this might occur at the level of BMP protein inhibition (residual dorsal BMP protein), if BMP receptors act to constantly monitor ligand levels. This inhibition of signalling is most likely mediated by Chordin. To test this idea we knocked down Chordin with a MO and investigated the effect on mRFP mRNA levels at 40% epiboly when they are normally graded. We could clearly demonstrate that in chordin morphants the BMP activity gradient was abolished and mRFP expression was uniform (Fig. 6C). This occurs despite the fact that auto-regulation of bmp2b transcription only occurs after that stage (Fig. S6). We therefore conclude that between 30% and 40% epiboly, expression of Chordin on the dorsal side of the embryo must contribute to a loss of dorsal BMP activity, and that after 40–50% epiboly the BMP activity gradient becomes maintained by the gradient of ligand transcription.

Discussion

Using ISH for mRFP in our BRE-mRFP transgenic line, which is a highly sensitive reporter for BMP/GDF signalling, we can visualise the dynamics of BMP signalling in early zebrafish embryos. We have shown that the BMP transcriptional activity gradient is established by 40% epiboly and is preceded by graded expression of BMP ligands (mRNA and protein). For the first time, we are able to detect the expression of an endogenous BMP protein, BMP2B, and found that it is expressed in a gradient during blastula and gastrula stages and that no extensive diffusion of BMP2B occurs to generate transcriptional activity at a distance. Our analysis strongly supports the notion that the BMP activity gradient in zebrafish occurs as a result of graded BMP expression and does not fit the classical models of gradient formation, where ligands synthesised from a local source would diffuse away in order to produce a signalling gradient.

A powerful tool to visualise BMP/GDF activity

We confirmed using various experimental approaches that mRFP expression in our BRE-mRFP transgenic fish is a faithful readout of BMP/GDF activity. We found that mRFP fluorescence is only detected from early somitogenesis stages, which is most likely due to insufficient amounts of mRFP protein being produced. A similar lack of fluorescence at early stages has been reported for other transgenic reporter lines such as the Wnt reporter line TOPdGFP (Dorsky et al., 2002), the Hedgehog pathway reporter line Gli-d:mCherry (Schwend et al., 2010) and the other zebrafish BMP reporter lines (Alexander et al., 2011; Collery and Link, 2011; Laux et al., 2011). However, mRFP ISH in our strong BRE-mRFP lines visualised with either colourimetric or fluorescent substrates provides an excellent method to detect highly dynamic BMP transcriptional activity.

Timing and mechanisms of BMP activity gradient formation during blastula stages

We found that the ventral to dorsal BMP activity gradient, revealed with mRFP ISH, is set up by 40% epiboly during the blastula period. The appearance of this gradient is preceded by the establishment of graded expression of BMP ligands which occurs between dome stage and 30% epiboly, and graded pathway activation read out by p-Smad1/5 levels occurs fractionally before the appearance of the mRFP gradient.

Using our results, combined with those from other studies (Furthauer et al., 2004; Langdon and Mullins, 2011; Maegawa et al., 2006), we propose a model for BMP activity gradient formation that features the temporal and spatial requirement for specific signalling events ( Fig. 7). First, immediately after MBT, bmps are induced ubiquitously in the embryo. In contrast, the various BMP antagonists are locally induced by maternal β-catenin2 on the future dorsal side of the embryo. At dome stage, Dharma acts dorsally to locally repress bmp2b transcription. This symmetry-breaking event is followed by the action of FGF signalling at around 30% epiboly, which appears to act in a dorsal to ventral graded fashion to inhibit bmp ligand transcription in a dose-dependent manner. It is not yet clear how FGF signalling achieves this. As a result, by 30% epiboly, all three bmp transcripts are expressed in a graded manner. However, at this stage, our experiments suggest that there must be some residual BMP protein present on the dorsal side and that the BMP protein gradient lags behind the bmp transcript gradient (Fig. 6D). One clear indication of residual dorsal BMP protein is that p-Smad1/5 is still detected dorsally at 30% epiboly, albeit at slightly lower levels, and that mRFP transcripts are also strongly detected there (Fig. 3). Between 30% and 40% epiboly, inhibition of residual BMP protein function dorsally, via Chordin, rapidly leads to reduction in p-Smad1/5 levels and the disappearance of mRFP transcripts. Coupled with fast mRFP mRNA turnover, this contributes to the detection of the BMP transcriptional activity gradient at 40% epiboly. Interestingly, the rapid clearance of mRFP transcripts dorsally and the fact that Chordin knockdown inhibits this, suggest that BMP receptors must be able to monitor ligand activity at the cell surface and thus respond to rapid changes in signalling. We hypothesise that from 40% to 50% epiboly onwards, the BMP activity gradient becomes proportional to the amount of BMP transcripts and proteins, the expression of which is maintained through auto-regulation, combined with pathway inhibition by dorsally-expressed Chordin (Fig. 7; Fig. S6). One important feature of this model is that there is a complex level of interaction between all the molecular players. One example is that Chordin expression in the blastula is partially under the control of FGF and Nodal signaling (Fig. S5) and that optimal repression of bmp2b expression by FGF signalling also appears to require Chordin function (Maegawa et al., 2006).

Fig. 7

Sequence of signalling events leading to graded BMP ligand transcription and hence a gradient of BMP activity. While the dorsal determinants are actively induced (simplified representation; cross-regulation is not shown), the transcription of zygotic BMPs is initially ubiquitous. The symmetry-breaking event for bmp expression is the transcriptional repression of bmp2b (and possibly bmp7a) by Dharma. FGF signalling is critical for establishing graded bmp expression. Both graded bmp ligand expression and the BMP activity gradient are subsequently mostly maintained through auto-regulation of bmp expression. For details, see text.

The zebrafish ventral to dorsal BMP activity gradient: An alternative model for the formation of BMP gradients

BMPs have long been assumed to be morphogens in the classical sense; that is that they are expressed from a local source and diffuse in the extracellular space to generate an activity gradient that is proportional to the amount of extracellular proteins. Not only is the issue of BMP extracellular diffusion still questionable, but the ways in which the activity gradients are formed in various developmental systems also show that the BMP signalling activity does not necessarily correlate with the amount of signalling BMP proteins (Wolpert, 2011).

Our understanding of how BMP gradients first form comes predominantly from studies in Drosophila, where two different mechanisms have been uncovered (reviewed by Ramel and Hill, 2012). To generate the Dpp activity gradient that determines A/P patterning of the Drosophila wing, the Dpp ligand is secreted from a central stripe of cells in the wing imaginal disc and acts at a distance to form a concentration gradient both anteriorly and posteriorly (Kicheva and Gonzalez-Gaitan, 2008; Raftery and Umulis, 2012). In contrast, the D/V Dpp activity gradient that forms in the embryonic Drosophila blastoderm results from redistribution of ligand within a uniform dpp expression domain. This is achieved by binding and transport of Dpp and a related ligand Screw with the ligand antagonists Twisted Gastrulation and Short Gastrulation (Drosophila Chordin) (O'Connor et al., 2006). An analogous ligand shuttling mechanism was proposed to be responsible for the D/V BMP activity gradient in Xenopus, and was preferred over an alternative inhibition-based mechanism where an inhibition gradient of diffusible inhibitors like Chordin is created over a uniform field of activators (Ben-Zvi et al., 2008). However, there is little evidence for this mechanism in the Xenopus embryo and this issue is very controversial (Francois et al., 2009).

We have shown that the early D/V BMP activity gradient is formed in a very different manner in zebrafish. Our study demonstrates that the BMP activity gradient is actually created in a tissue where activity is initially ubiquitous. Therefore this particular gradient is not created de novo. We also found that the control of BMP ligand transcription is the crucial determinant for generating a BMP activity gradient that is robust and sustainable - rather than extensive diffusion or redistribution of BMP proteins. Indeed, consistent with published data on the diffusion of BMP4 in Xenopus (Jones et al., 1996) and cell transplantation experiments in zebrafish (Nikaido et al., 1999), we find no evidence that zebrafish BMP proteins diffuse to generate signalling activity at a distance. Our data highlights the fact that each BMP gradient is unique in the way it is established and maintained.