bif1, a new BMP signaling inhibitor, regulates embryonic hematopoiesis in the zebrafish

Red blood cells (RBCs) constitute the majority of the circulating blood cells and play a major role in the transport of respiratory gases. They are constantly replenished through a process called erythropoiesis, and are derived from hematopoietic stem and progenitor cells (HSPCs) in adults. The molecular basis of this process has been largely elucidated through the use of different in vivo models, such as Xenopus, zebrafish and mice, and in vitro models, such as human umbilical cord blood differentiation (Nandakumar et al., 2016) or the study of embryonic stem cell differentiation. In all vertebrates, the production of RBCs occurs in two major waves during development: primitive and definitive erythropoiesis (Palis, 2014). Both processes lead to the formation of RBCs with molecular differences such as the expression of different hemoglobin genes (Sankaran and Orkin, 2013).

In the zebrafish embryo, multiple waves of hematopoiesis give rise to distinct waves of erythropoiesis. During primitive erythropoiesis, gata1, a master regulator of erythropoiesis, is expressed in the posterior lateral plate mesoderm (PLPM) as early as 12 h post fertilization (hpf) (Detrich et al., 1995; Long et al., 1997). At 18 hpf, after PLPM stripes have converged to the midline, erythroid progenitors differentiate and proliferate in the intermediate cell mass (ICM) (Al-Adhami Ma, 1977). This primitive erythropoiesis is similar to what is observed in mammals, where primitive RBCs arise in the yolk sac directly from mesodermal progenitors (Palis et al., 1999; Keller et al., 1999). As definitive hematopoiesis arises, a transient wave of erythro-myeloid progenitors (EMPs) appears in the posterior blood island of the zebrafish embryo. Although the contribution of these progenitors is hard to quantify in vivo, we have previously shown that they could differentiate into erythroid cells in vitro (Bertrand et al., 2007). Mouse EMPs develop in the yolk sac and give rise to a clearly identified RBC population that expresses specific hemoglobins (McGrath et al., 2011). Finally, HSPCs emerge in the ventral floor of the dorsal aorta in zebrafish (Bertrand et al., 2010), similar to the aorta-gonad-mesonephros in mammals (Medvinsky and Dzierzak, 1996). HSPCs migrate to the caudal hematopoietic tissue (CHT), where they expand and differentiate into all major hematopoietic lineages (Murayama et al., 2006). Although the cell and non-cell autonomous signals that govern RBC production from definitive hematopoietic progenitors are well characterized, the signals that specify mesoderm into primitive RBCs are still largely unknown. gata1, as well as lmo2 and scl/tal1 can be detected as early as 12 hpf in the PLPM during zebrafish development (Thompson et al., 1998; Liao et al., 1998; Detrich et al., 1995). However, the signals that control the expression of these genes, and hence the commitment of mesoderm progenitors towards the erythroid lineage, are still poorly understood.

Bone morphogenetic proteins (BMPs) are members of the TGF-β family and are central players in the BMP signaling pathway, which is involved in several processes, such as cancer, bone formation and mesoderm specification (Zhang et al., 2016; Wu et al., 2016; Chen et al., 2004). During gastrulation, high BMP activity specifies the ventral part of the embryo, while BMP inhibitors are expressed dorsally (Pomreinke et al., 2017). At this stage, high BMP activity specifies mesodermal cells into blood lineage, as shown by severe hematopoietic defects in bmp4- and smad5-deficient mouse embryos (Winnier et al., 1995; Liu et al., 2003). For example, bmp4 mutant mouse embryos die early during embryogenesis, but the few embryos that survive show a strong reduction in primitive erythropoiesis (Winnier et al., 1995). Upon the binding of BMP ligands to their receptors, R-smads (smad1, smad5 and smad9 in zebrafish) are phosphorylated. Activated R-smads bind the co-smad smad4 and translocate to the nucleus to activate the transcription of target genes by binding to BMP-responsive elements (BRE) (Katagiri et al., 2002). To date, several genetic screens have identified many zebrafish mutants with mutations in components of the BMP signaling pathway, such as bmp2b, bmp7 or smad5 (Kishimoto et al., 1997; Hild et al., 1999; Schmid et al., 2000). These mutants share a common dorsalized phenotype due to a problem in dorso-ventral patterning during gastrulation (Kondo, 2007). This signaling pathway is tightly regulated during development at the cell- and non-cell-autonomous level. At the non-cell-autonomous level, chordin and noggin are the major secreted inhibitors against BMP signaling, and are expressed in the organizer during gastrulation (Fürthauer et al., 1999; Smith and Harland, 1992; Sasai et al., 1994). Mutants for chordin (chordino) show a strong ventralization phenotype owing to the change in dorso-ventral BMP gradient (Hammerschmidt et al., 1996). At the cell-autonomous level, the inhibitory Smads (smad6 and smad7) inhibit the binding of R-smads to the BMP receptors, to the co-smad and to DNA, leading to an inhibitory effect on BMP activity. Smad7 was found to increase the self-renewal of HSPCs and to promote myeloid differentiation in human and murine systems, respectively (Chadwick et al., 2005; Blank et al., 2006). SMAD6, by interfering with SMAD5 and SMAD4, inhibits erythropoiesis from human umbilical cord blood progenitors (Kang et al., 2012).

Here, we have elucidated the role of an uncharacterized gene, si:ch73-299h12.2, that was primarily described as highly enriched in endothelial and hematopoietic cells sorted from 26-28 hpf Tg(fli1a:egfp) zebrafish embryos (Cannon et al., 2013). This gene of unknown function encodes a C2H2 zinc-finger protein, the expression of which progressively becomes restricted to the primitive erythroid lineage during early development, and is maintained in definitive hematopoietic progenitors. We named this gene bif1, as our results show that this new gene is important for the development of erythropoiesis (blood-inducing-factor 1) and that it does so by inhibiting the BMP pathway (BMP-inhibitory-factor 1).

bif1 is a gene of unknown function in a cluster of the telomeric region of chromosome 21. It has a simple structure consisting of two exons and one intron (Fig. 1A). bif1 encodes a protein containing ten C2H2 zinc-finger motifs (Fig. 1A). These domains are known to bind DNA and act as transcription factors, but these motifs can also bind RNA or proteins (Brown, 2005; Brayer and Segal, 2008; Wolfe et al., 2000). The prediction site also indicates a classical bipartite nuclear localization site (NLS) in Bif1. To test the ability of Bif1 to translocate to the nucleus, we cloned a mRNA coding for a HA-tagged Bif1. This bif1-HA mRNA was injected into wild-type embryos, and we performed anti-HA immunofluorescence at 6 hpf. In injected embryos, we observed staining in the nucleus (Fig. 1B). Therefore, Bif1 is able to translocate to the nucleus, and could potentially bind DNA to act as a transcription factor.

During embryogenesis, the maternal-to-zygotic transition occurs around 2.5 hpf (Tadros and Lipshitz, 2009). As bif1 starts to be expressed at 3 hpf, it is likely a zygotic mRNA (Fig. 1C). At 6 hpf, bif1 is expressed in the margin and in the animal pole of the embryo following a dorso-ventral gradient, but no expression was seen in the dorsal organizer (Fig. 1C). From 12 hpf to 6 days post fertilization (dpf), bif1 expression is highly similar to that of gata1: it is first expressed in the PLPM at 12 hpf, then its expression becomes specific to the ICM at 18 hpf, before it is expressed in circulating erythroblasts at 24 hpf (Fig. 1C). It is then expressed in the CHT between 2 dpf and 5 dpf (Fig. 1C and Fig. S1). At this stage, bif1 also starts to be detected in the glomerulus area, the ultimate site of hematopoiesis (Fig. S1). To confirm expression in the PLPM, we investigated the expression of bif1 in cloche mutants, as well as its relationship to pax2a-expressing cells (intermediate mesoderm). We found that bif1 was absent in cloche mutants, and that bif1-positive cells were juxtaposed to pax2a-positive cells (Fig. 1D,E). To confirm the erythroid expression, we generated double transgenic animals Tg(bif1:GFP-LT;gata1:DsRed). Between 20 and 24 hpf, GFP-LT expression recapitulated the same pattern as observed by whole-mount in situ hybridization, and showed fluorescence in the ICM, overlapping with that of Tg(gata1:DsRed) in double transgenic embryos (Fig. 1F). Altogether, these data show that bif1 becomes progressively restricted to the erythroid lineage as embryogenesis occurs.

During definitive hematopoiesis, HSPCs are produced from the ventral floor of the dorsal aorta (VDA) (Bertrand et al., 2010; Kissa and Herbomel, 2010). In the double transgenic animals Tg(bif1:GFP-LT;flk1:mCherry), we found double-positive cells in the VDA showing that bif1 is expressed in HSPCs at 36 and 48 hpf (Fig. 2A,B). Later, we identified GFP-LT-positive cells in the CHT, indicating that bif1 is still expressed in HSPCs when they expand in the CHT (Fig. 2C,D). Finally, in the thymus, where definitive lymphopoiesis occurs, we observed double-positive cells in double transgenic Tg(bif1:GFP-LT;rag2:DsRed) animals (Fig. 2E,F). Although bif1 was never observed by whole-mount in situ hybridization in the thymus, the presence of these rare double-positive cells could be due to the long lifespan of GFP-LT expressed by HSPCs. Altogether, these results show that bif1 is expressed during definitive hematopoiesis.

We took advantage of the Tg(gata1:DsRed) transgenic line and sorted, after dissecting the whole kidney marrow (WKM), different hematopoietic cell populations based on scatter populations and levels of DsRed expression: the lymphoid/progenitor fraction was subdivided into lymphocytes (L/PNE – lymphoid/precursor gate non erythroid; gata1:DsRed⁻), erythroid progenitors (gata1:DsRed^low) and mature RBCs (gata1:DsRed^high). Myeloid cells were used as control (Fig. 2G). By qPCR we found an enrichment of bif1 in the two erythroid compartments (Fig. 2H). Therefore, bif1 also marks the definitive erythroid lineage in the adult; however, we never observed this erythroid expression with our bif1:GFP-LT animals (data not shown).

In order to identify the function of this new gene during erythropoiesis, we overexpressed bif1 and analyzed gata1 expression at 12 hpf. About 50% of the embryos showed an ectopic expression of gata1 in the posterior part of the PLPM (Fig. 3A). However, flk1 or acta2 were not altered, showing that bif1 gain-of-function did not interfere with angioblasts or paraxial mesoderm (Fig. 3B). To assess the specific role of bif1 in this phenotype, we also injected a mutated bif1 mRNA (bif1MUT) resulting in a non-functional NLS. This mutated mRNA failed to induce ectopic gata1 at 12 hpf (Fig. S2A,B). The erythroid and pronephric lineages are both derived from the same mesoderm territory. It has previously been shown that BMP signaling affects the balance between these two lineages (Gupta et al., 2006) and we wondered whether bif1 also played a role here. By performing a double whole-mount in situ hybridization against pax2a and gata1, we observed an increase of gata1 expression compared with pax2a in the embryos injected with bif1 mRNA (Fig. 3C,D). This increase in gata1 occurred at the expense of the pax2a territory (Fig. S3), suggesting that bif1 affects the balance between the erythroid and pronephric lineages. As the expression of bif1, by in situ and in transgenic animals, always seems to correlate with early stages of erythroid differentiation, we speculated on a potential role for bif1 in regulating early stages of erythropoiesis. After bif1 mRNA injection, we collected blood samples from 4 dpf embryos and performed a May–Grünwald–Giemsa staining. Interestingly, in bif1-injected embryos, erythroid cells appeared bigger than in control embryos, therefore less differentiated, although the nucleus was condensed (Fig. 3E,F). Their morphology was still very different from adult erythrocytes (Fig. 3F). Altogether, we conclude that bif1 alters/controls the terminal differentiation of erythroid progenitors.

As bif1 affects primitive erythropoiesis, we wanted to know whether it could also affect definitive erythropoiesis. According to previous reports, definitive erythropoiesis is initiated in the CHT (Murayama et al., 2006) at around 4 dpf, when primitive erythrocytes are still present in the embryo (Willett et al., 1999). In order to specifically investigate the role of bif1 in definitive erythropoiesis, we first ablated primitive RBCs, by treating embryos with phenylhydrazine (PHZ) (Shafizadeh et al., 2004). Control embryos treated with PHZ recover from the induced anemia 4 days after the drug is removed. To analyze the potential role of bif1 during recovery, we generated a transgenic line in which bif1 is expressed under the control of the heat shock-70 promoter (hsp70l). This construct allowed us to overexpress bif1 every day, by heat-shocking the embryos (Fig. 3G). In order to measure the difference in recovery, we performed O-dianisidine staining at 4 dpf. In transgenic embryos, there was a better recovery of erythroid progenitors in the CHT at 4 dpf, based on hemoglobin content as measured by O-dianisidine (Fig. 3H,I). In normal conditions, without chemically induced anemia, we observed a slight increase of gata1 expression but no differences in HSPC or neutrophil content in the CHT at 4 dpf (Fig. S4). Therefore, overexpression of bif1 aids the recovery from a chemically induced anemia by increasing definitive erythroid progenitors in the CHT.

To complement our gain-of-function experiments, we knocked down bif1 using a morpholino. We observed a strong decrease of circulating RBCs at 3 dpf (scored by O-dianisidine staining), which was rescued by the injection of bif1 mRNA (Fig. 4A,B and Fig. S11A,B,C,E). This phenotype was quite specific, as we did not observe any change in primitive myelopoiesis or in the development of vasculature or somites when bif1 was knocked down by a morpholino (Fig. S5A-C), thus confirming that bif1 is necessary for erythropoiesis only during primitive hematopoiesis. In order to determine whether bif1 was also required for erythroid progenitors in the CHT, we performed whole-mount in situ hybridization against gata1 at 4 dpf. We observed a loss of gata1 expression in the CHT (Fig. 4C), linked to an increase in cell death, as measured by Acridine Orange staining in bif1 morphants (Fig. S6). However, this was not specific to the erythroid lineage, as rag1 was lost at 5 dpf in the thymus (Fig. 4D) and cmyb was lost at 4 dpf in the CHT (Fig. 4E), showing a defect in definitive hematopoiesis. As we found that bif1 is expressed in HSPCs during their specification in the VDA (Fig. 2B), we wondered whether the emergence of HSPCs was also altered in bif1 morphants. Both runx1 and cmyb were affected in bif1 morphants, whereas gata2b was not (Fig. 4F). Altogether, these results point out the importance of bif1 in definitive hematopoiesis, as well as in primitive erythropoiesis.

As stated above, the balance between intermediate and lateral plate mesoderm is partially controlled by BMP signaling. Moreover, the ectopic expression of gata1 at 12 hpf in the PLPM has already been described in BMP-deficient embryos, such as bmp4 mutants (Stickney et al., 2007). Additionally, BMP-deficient embryos exhibit dorsalized morphologies, as well as defective cloaca development and the formation of ectopic tail structures (Pyati et al., 2006; Yang and Thorpe, 2011; Mullins et al., 1996). We could observe all these phenotypes in embryos injected with bif1 mRNA (Fig. 5A,B and Fig. S7A,B). Again, this dorsalizing phenotype was specific as it was strongly reduced when the NLS in bif1 was mutated (Fig. S2C,D). Similarly, we overexpressed si:ch73-299h12.3 mRNA, a close paralog of bif1 sharing more than 58% similarity at the protein level (Fig. S8A) with an expression pattern that is highly similar to bif1 (Fig. S8B). This highly similar paralog did not induce dorsalization, but rather cyclopia (Fig. S8C), and did not affect gata1 in the way bif1 mRNA did (Fig. S8D). This suggests that the phenotypes observed in bif1 gain of function were bif1 specific. This strong dorsalizing phenotype did not seem to impair primitive myelopoiesis (Fig. S5D), and vasculogenesis was difficult to assess due to the overall morphology of the embryos (Fig. S5E). From these results, we therefore hypothesized that bif1 inhibits the BMP signaling pathway. In order to confirm our hypothesis, we analyzed the level of BMP signaling by measuring the levels of phosphorylated Smad1, Smad5 and Smad9, as these proteins are specifically phosphorylated in response to BMP ligands. bif1 gain of function caused a decrease in the ratio of p-Smad/Smad, therefore arguing in favor of an inhibition of the BMP signaling pathway (Fig. 5C,D). Moreover, in the Tg(BRE:eGFP) reporter line, where eGFP is expressed under the control of BMP responsive elements, bif1 decreased the number of eGFP⁺ cells (Fig. 5E), as well as their mean fluorescence intensity (data not shown), showing a strong decrease in BMP activity in vivo. The dorsalization induced by bif1 was rescued through co-injection of bmp2b mRNA, one of the BMP ligands (Fig. 5B), which by itself led to ventralization (Fig. S9). In bif1 morphants, we found a systematic reduction in eye size (Fig. 5F,G) which was not due to developmental delays, according to the size of otoliths (Fig. S10A,B). This phenotype is known to be linked with a higher level of BMP activity (Song et al., 2013), as scored by higher levels of p-Smad1, Smad5 and Smad9 (Fig. 5H). This phenotype was rescued by the injection of bif1 mRNA (Fig. S11D). Altogether, these results point to bif1 as a new endogenous inhibitor of the BMP signaling pathway. In order to precisely elucidate the molecular function of bif1, we analyzed, using whole-mount in situ hybridization, the expression of genes known to play a role in BMP signaling. At 6 hpf, bmp4 and bmp2b were decreased after bif1 overexpression, whereas the other members tested were not affected (Fig. S12A,B). Further analyses will be required to fully understand the precise molecular function of bif1. Finally, we looked at a potential feedback loop from BMP signaling over bif1 expression. We found an increase and a decrease of bif1 expression when we injected smad6a and bmp2b, respectively (Fig. 5I,J). These results show that the BMP pathway negatively regulates bif1 expression. During gastrulation, BMP signaling and bif1 inhibit each other, which probably contributes to fine-tuning the dorsoventral gradient of BMP signaling. Finally, we sought to investigate whether bif1 could mediate its effect on erythropoiesis by inhibiting the BMP signaling pathway. Therefore, we analyzed the morphology of RBCs after smad6a mRNA overexpression to inhibit the BMP pathway. After examining RBC morphology, we obtained a similar result to that obtained after bif1 overexpression (Fig. S13). Therefore, we conclude that bif1 regulates erythropoiesis through its ability to inhibit the BMP pathway.

We investigated the presence of bif1 orthologs in other species. In order to find a potential ortholog, we blasted the protein sequence of bif1 against known proteomes (blast.ncbi.nlm.nih.gov/) and identified more than 30 zebrafish proteins due to the highly conserved C2H2 sequences (data not shown). Moreover, we also found a potential ortholog in Sinocyclocheilus rhinocerous (S. rhino) – LOC107737464 – that shares 57.8% of similarity at the protein level (Fig. S14). Although zebrafish and S. rhino do not belong to the same subfamily order, they are both cyprinids. Both species share a similar genome organization near these two genes, as observed by synteny analysis (Fig. 6A). We injected the full-length mRNA for S. rhino LOC107737464 into zebrafish embryos. At 24 hpf, these embryos were dorsalized, although to a lesser extent than the dorsalization produced by bif1 overexpression (Fig. 6B), demonstrating that S. rhino bif1 can also inhibit BMP signaling in the zebrafish embryo. However, no significant alteration in gata1 expression was observed in embryos overexpressing this potential ortholog of bif1 (Fig. 6C). We wondered whether bif1 was also present in mammals. We blasted each Bif1 protein sequence from these two fish species against the Mus musculus protein database and identified one potential ortholog. Zfp944 shows 39.1% similarity to Danio rerio Bif1 at the protein level (Fig. S15A). Moreover, this gene seems to be expressed at low levels in erythroid cells during primitive, fetal and adult erythropoiesis, according to the Erythron database (www.cbil.upenn.edu/ErythronDB/). This gene is present on chromosome 17, in a cluster of genes encoding zinc-finger proteins, a situation that is reminiscent of the cluster where bif1 lays in the zebrafish genome. When we injected the full-length Zfp944 mRNA into one-cell stage zebrafish embryos, we observed a dose-dependent dorsalization (Fig. 6D,E), correlated to a decrease of BMP signaling activity, as observed in Tg(BRE:eGFP) (Fig. S15B). However, no effect was found on gata1 expression at 12 hpf, similar to S. rhino bif1 (Fig. 6F), but at 48 hpf, the overexpression of the three transcripts induced a similar decrease of gata1, as measured by qPCR on whole embryos (Fig. 6G), which might explain the abnormal erythroid differentiation observed at 4 dpf (Fig. 3E,F). Altogether, we found that Zfp944 in M. musculus is also able to inhibit BMP signaling in zebrafish embryo. However, additional experiments will be required to address its potential role during mouse erythropoiesis.

In this study, we have identified a new endogenous inhibitor of BMP signaling. Previously uncharacterized, bif1 encodes a C2H2 zinc-finger protein that can translocate to the nucleus. C2H2 ZF represents a family of protein present in all species, from archaea bacteria to humans. Since the discovery of these domains in TFIIIA (Miller et al., 1985), many studies focused their research on this group of proteins, which represent 3% of human proteins (Klug, 2010). This family of proteins is still poorly characterized and no perfect prediction software can determine their molecular function based on their amino acid sequences. It was previously thought that C2H2 domains were only binding DNA until the discovery, in many proteins, of their ability to also bind proteins and RNA (Brayer et al., 2008; Brown, 2005). Moreover, in the same protein, different C2H2 can bind either DNA or proteins, making their studies even more complex. In the case of bif1, owing to its capacity to translocate to the nucleus, we hypothesized a predominant role in the nucleus as a transcription factor, but we cannot exclude other functions, such as stabilizing RNA. From this hypothesis, we speculate that bif1 can inhibit, by binding to DNA, the transcription of specific genes involved in BMP signaling, such as bmp2b and bmp4.

During gastrulation, BMP ligands and inhibitors are expressed according to opposite gradients, such as bmp2b and bmp7, and chordin and noggin. These opposite expressions are responsible for establishing a gradient of BMP activity (Ramel and Hill, 2012). Here, we have identified bif1 as a new intracellular cell-autonomous inhibitor of BMP signaling that regulates the dorso-ventral patterning during gastrulation. Owing to the expression of BMP members in erythroid progenitors, the role of BMP signaling during erythropoiesis has been deeply investigated. The deletions of Smad5 or Bmp4 are embryonic lethal in mice. However, the few mice that survive show strong defects in primitive erythropoiesis (Hegde et al., 2007; Coleman et al., 1969; Winnier et al., 1995), revealing the importance of BMP signaling for the establishment of primitive erythropoiesis. In zebrafish, smad5 mutants also have defects in primitive erythropoiesis (McReynolds et al., 2007). In these different models, BMP signaling was shown to be required for primitive erythropoiesis. Our experiments show that bif1 increases the number of gata1-positive cells during primitive erythropoiesis, whereas its absence decreases the number of erythrocytes in circulation. This indicates a new role for BMP signaling during primitive erythropoiesis.

Under stress conditions, such as PHZ treatment (induced anemia), the recovery of the mice was delayed in the absence of Smad5 (Coleman et al., 1969; Lenox et al., 2005; Hegde et al., 2007). Moreover, Bmp4 is induced in the spleen of mice following anemia (Wu and Paulson, 2010). Through the activation of Smad5, Bmp4 then activates the expansion of erythroid immature BFU-E cells (Perry et al., 2007; Lenox et al., 2005). Moreover, other BMP ligands have been shown to be important during zebrafish stress-induced erythropoiesis, as shown by treatment with ginger, which increases bmp2b and bmp7 expression in the CHT, leading to an increase of erythroid progenitors (Ferri-Lagneau et al., 2012). These data show that, during stress-induced erythropoiesis, BMP signaling is activated in order to increase the expansion of erythroid progenitors. Surprisingly, in our chemically induced anemia experiment, we discovered an expansion of erythroid progenitors in the CHT of embryos injected with bif1, where BMP signaling was therefore attenuated. We hypothesize that during stress erythropoiesis, two steps are important for the recovery of erythroid cells. One step needs an increase of BMP signaling to promote erythroid progenitor expansion, whereas a later steps needs to shut down BMP signaling, probably to proceed with normal differentiation. Further studies will be needed to understand this dual role of BMP signaling during erythropoiesis. In mammals, BMP ligands were also shown to be required for the differentiation into erythroid cells from ES cells (Adelman et al., 2002). In fact, BMP4 is needed to activate the transcription of erythroid genes, such as Eklf (Klf1) and Gata1, whereas SMAD6 blocks this induction (Adelman et al., 2002; Schmerer and Evans, 2003). In human cells, BMP2 also increases the number of early BFU-E cells, whereas activin A increases both BFU-E and CFU-E. This result shows that BMP2 is acting in more immature cells (Maguer-Satta et al., 2003). The maturation of zebrafish primitive erythroid cells was delayed or blocked when bif1 or smad6a were overexpressed. Our results are consistent with the previously described role of BMP signaling during erythroid maturation. By blocking BMP signaling, bif1 blocks erythroid differentiation into mature RBCs, probably by decreasing gata1. By contrast, the absence of bif1 induced a complete absence of primitive erythropoiesis.

We have also found that bif1 is important for the normal development of HSPCs along the VDA, as bif1-morphants exhibit reduced levels of runx1 and cmyb in the hemogenic endothelium. Although BMP signaling positively regulates HSPC emergence (Wilkinson et al., 2009), it was also recently shown that BMP signaling needs to be turned off after HSPC specification for normal HSPC maturation (McGarvey et al., 2017). Therefore, we propose that bif1 plays such a role during HSPC emergence in the zebrafish embryo.

In summary, we have discovered a new intracellular BMP inhibitor in zebrafish, bif1, that is specifically expressed in primitive and definitive hematopoietic lineages during embryogenesis, as well as in adults. We generated a new bif1:GFP-LT transgenic reporter that faithfully recapitulates endogenous expression in the embryo only, as we could not observe GFP-LT expression in adult erythroid progenitors.

Through the inhibition of BMP signaling, bif1 affects dorso-ventral patterning during gastrulation and affects primitive and definitive erythropoiesis, as well as the normal development of HSPCs. This gene is not restricted to Danio rerio but is also present in another fish (S. rhino), where it probably plays a similar role. Finally, we also found a potential ortholog in the mouse genome, Zfp944, that is expressed in the mouse erythroid lineage and induces dorsalization of zebrafish embryos. Further work will be required to understand the precise inhibitory role of this new zinc-finger protein on BMP signaling.

Zebrafish were maintained and bred according to the University of Geneva procedures and authorizations from local veterinary authorities. All experiments were performed using AB* and the following transgenic animals: Tg(gata1:Dsred) (Yaqoob et al., 2009), Tg(BRE:eGFP) (Collery and Link, 2011) and cloche^m34 mutants (Stainier et al., 1995).

Whole-mount in situ hybridization was performed on 4% paraformaldehyde-fixed embryos. Digoxigenin-labeled and FITC-labeled probes were synthesized using a RNA Labeling kit (SP6/T7; Roche). RNA probes were generated by linearization of TOPO-TA vectors (Invitrogen) containing the PCR-amplified cDNA sequence. Whole-mount in situ hybridization was performed as previously described (Thisse et al., 1993). Double whole-mount in situ hybridization was performed by using IBT/BCIP (Roche). Embryos were imaged in 100% glycerol using an Olympus MVX10 microscope. The oligonucleotide primers used to clone cDNA for probes are listed in Table S1.

Full-length cDNAs were cloned into the pCS2 vector from which mRNA was synthesized using the mMessage mMachine Kit (Life Technologies). In all experiments, 300 pg of bif1 mRNA, 100 pg of bmp2b or smad6a mRNA and 1300 pg of bif1 S. rhino were injected into one-cell stage embryos. Except for the rescue experiment with morpholino, bif1 was injected at 100 pg. Site-directed mutation of bif1 was performed with the QuickChangeII Site-directed mutagenesis Kit (Agilent Technologies). Morpholino oligonucleotides were purchased from Gene Tools and are listed in Table S2. MO efficiency was tested by reverse transcription polymerase chain reaction (RT-PCR) from total RNA extracted from 10 embryos at 24 hpf using primers listed in Table S1. bif1-MO was injected at 17 ng for all experiments except for the rescue experiment, where it was injected at 16 ng.

Embryos were lysed in Pierce lysis buffer (Thermo) with a protease and phosphatase inhibitor cocktail (Roche) for 15 min on ice after mechanical disruption. Debris was removed by centrifugation. Samples were then subjected to SDS-polyacrylamide gel electrophoresis and immunoblotting according to standard protocols (Sun et al., 2011), using anti-Smad1/5/9 (1:1000; Creative Diagnostics, DPABH-17143167), anti-phospho-ser463/465-smad1/pser463/465-smad5/pser465/467-smad9 (1:1000; Cell Signaling, 13820) and anti-β-actin (1:25,000; Sigma-Aldrich, A2066) antibodies.

Zebrafish embryos (6 hpf) were fixed in 4% PFA before washes and blocking (PBS+0.1% triton X-100+4% BSA+0.02% NaN₃). The primary anti-HA antibody (1:1000; Abcam, ab18181) was incubated with the samples overnight at 4°C followed by washing and incubation with the secondary anti-mouse IgG Alexa-fluor488 antibody (1:1000; Invitrogen, A32723). Nuclei were stained using DAPI (Sigma).

Tg(BRE:eGFP) embryos (12 hpf) were dissociated as previously described (Bertrand et al., 2007). Adult WKM were mechanically dissociated. SYTOX Red (Molecular Probes) was used to detect dead cells. Cells were analyzed using a FACSAccuriC6 (BD Biosciences) or a FACS Fortessa (BD Biosciences) and sorted with a FACSAriaII cell sorter (BD Biosciences).

RNA was extracted using RNeasy minikit (Qiagen) and reverse transcribed into cDNA using a SuperscriptIII kit (Invitrogen). Quantitative real-time PCR (qPCR) was performed using SensiFAST SYBR Lo-ROX kit (Bioline) and run on a QuantStudio3 real-time system (Thermo Fisher). In each sample, levels of expression were normalized to ef1a. Statistical analysis was carried out using an unpaired Student's t-test in Prism. qPCR experiments were performed in biological triplicates. Primers are listed in Table S1.

For Tg(bif1:GFP-LT) fish generation, we cloned a sequence of 1 kb upstream of the start codon in a Tol2 vector containing GFP-LT. For Tg(Hsp70:bif1-p2a-tfp) fish generation, we cloned a bif1-coding sequence without the stop codon into Tol2 vector containing the HSP70l promoter and a p2a-TFP fusion cassette (a kind gift from Benjamin Martin, Stony Brooks University, NY, USA). Zebrafish embryos were injected with 40 pg of the final Tol2 vector, along with 40 pg Tol2 transposase mRNA. Injected F0 adults were mated to AB* or Tg(gata1:Dsred), and the F1 offspring were screened to assess germline integration of the Tol2 construct. Primers are listed in Table S1.

Phenylhydrazine (0.5 µg/ml final) was added into the incubation medium of 33 hpf zebrafish embryos, followed by extensive washes at 48 hpf. Detection of hemoglobin by o-dianisidine was performed as described previously (Ransom et al., 1996).

After tail clipping using a scalpel, blood cells were collected by pipetting, in 0.9× PBS and 1% BSA (Sigma-Aldrich). Then they were cytospun onto slides by centrifugation at 23 g for 5 min using a shandon cytospin 2. The slides were then air-dried and subjected to May–Grünwald–Giemsa staining according to a standard protocol (Bertrand et al., 2007).

We thank Dr Claire Pouget for the acta2 probe, Dr Richard Fish for the ntla and bmp4 probes, and Dr Benjamin Martin for the evx1 probe and the tol2-hsp70l-p2A-TFP vector. We also acknowledge C. Pasche and C. Compebine for excellent technical support. We thank M. Cavanaugh and Dr Stefania Nicoli for allowing us to use their lab facilities.