A β1,4-galactosyltransferase is required for Bmp2-dependent patterning of the dorsoventral axis during zebrafish embryogenesis

Complex carbohydrates are widespread throughout cells and tissues, and are predominant components of the extracellular matrix and the cell surface. The enormous diversity of complex carbohydrate structures results from the concerted action of specific glycosyltransferases and associated enzymes involved in carbohydrate synthesis. Given the ubiquitous expression of complex carbohydrates and their biosynthetic enzymes, it is likely that they play many different roles within the organism(Varki, 1993). However, the complexity of carbohydrate structures and of their biosynthetic enzymes has made it difficult to define their function in vivo, particularly during vertebrate embryogenesis.

Most studies of complex carbohydrate function during vertebrate development have relied upon the characterization of targeted knockouts of specific glycosyltransferases of interest. This approach has yielded important clues about the overall requirement of N-linked glycoside chains during early development, and about the role of specific monosaccharide residues in various physiological events (Domino et al., 2001; Ioffe and Stanley,1994; Lu et al.,1997; Maly et al.,1996). Furthermore, and perhaps more importantly, these studies have shown that glycosyltransferases are a much more polymorphic class of enzymes than previously thought. For example, targeted deletion ofβ1,4-galactosyltransferase uncovered the existence of five additional genes that encode β1,4-galactosyltransferases. In light of the large number of individual glycosyltransferases thought to be active in mammalian tissues (e.g. ∼150-300), it becomes difficult to achieve a more global understanding of carbohydrate function through traditional knockout approaches. To address this limitation, we have taken advantage of the zebrafish system, which is more amenable to a high throughput analysis, to investigate the function of glycosyltransferases during development. Herein,we describe an essential and unexpected role for aβ1,4-galactosyltransferase in patterning of the early embryo by participating in proteoglycan glycosylation that is required for Bmp-dependent specification of the dorsoventral axis.

Zebrafish, Danio rerio, were maintained at 28°C as described previously (Westerfield,1993). All experiments were performed using random matings of *AB animals. Embryos were collected, raised and staged in embryo medium until they reached the desired developmental stage. Developmental stage was determined following previously defined criteria(Kimmel et al., 1995). Allele designations for zebrafish mutants used in this study include: pgy(piggytail), snh (snailhouse) and swr(swirl) (Mullins et al.,1996).

Putative zebrafish β4-galactosyltransferases (β4GalTs) were identified through searches of the zebrafish genomic database (Sanger Institute). Briefly, mammalian β4GalTs (human β4GalT5, NM_004776 and mouse β4GalT5, NM_019835) were used as query sequences. Putative trace sequences were merged into contigs using the Lasergene sequence management software (DNASTAR) to determine the full-length sequences. Identified sequences were BLASTed against the genome assembly to identify putative splice sites. Primary amino acid sequence, as well as genetic structure, was used to define homology. Simple phylogeny was determined based upon parsimony,aligning the known human β4GalTs (NM_001497, NM_003780, NM_003779,NM_212543, NM_004775) with putative zebrafish orthologs using protpars (J. Felsenstein, unpublished). An unrooted phylogenetic tree was inferred from the alignment using megalign (DNASTAR) and ClusalW(Thompson et al., 1994).

β4GalT5 was cloned by RT-PCR from total RNA isolated from 72-hour zebrafish embryos. Primers (Operon) were designed to the central region ofβ4GalT5 and used to amplify a core fragment. Additional primers within the core fragment were used to perform both 5′ and 3′ RACE. The RACE fragments were gel purified and overlapping regions between the 5′RACE, the core fragment and the 3′ RACE were annealed. The full-length transcript was generated by PCR using primers designed to the 5′ and 3′ UTR. The full-length transcript was then cloned into pCRII(Invitrogen). Full-length β4GalT5 was subjected to site-directed mutagenesis to introduce silent mutations into the morpholino recognition site that would eliminate binding (Gene-Tailor, Invitrogen). The MO3 recognition sequence (TAATGCCGACACATCTGAGA) was modified to TTATGCCCACACACCTAAGC.

Whole-mount in situ hybridization was performed as described previously(Thisse and Thisse, 1998). Briefly, staged embryos were fixed in 4% paraformaldehyde and dehydrated in methanol. Following rehydration into phosphate-buffered saline/0.1% Tween-20,embryos were prehybridized for 1 hour at 65°C. Hybridization with gene-specific probes was conducted overnight at 65°C. DIG-labeled probes were detected with α-DIG antibodies (Roche) and visualized with BCIP/NBT(Vector Labs). Plasmids containing chordin were kindly provided by M. Halpern (Carnegie Institution). pax2a, mkp3 and β4galt5RNA antisense probes were synthesized with T7 from BamH1 linearized plasmids. Sense controls were generated from Not1 linearized plasmids and transcribed with Sp6. At least 10 embryos were used in each assay. All in situ hybridizations were repeated at least three times.

Antisense morpholino oligonucleotides were designed to compliment either a sequence in the 5′UTR of β4GalT5 (MO1,5′-CACTGCTGGAAATGTAAATACTCAT-3′; base pairs –245 to–221), an internal splice site of β4GalT5 (MO2,5′-ACGTGAACCCTGTCGCGTCCTGTCA-3′; base pairs +176 to +186 plus 15 intronic base pairs) or the start codon of β4GalT5 (MO3,5′-CGAAATCTCAGATGTGTCGGCATTA-3′; base pairs –2 to +23)(Gene-Tools). Morpholinos were resuspended in 1×Daneau Buffer prior to injection (Nasevicius and Ekker,2000). Various concentrations, as described in the text, were injected into the cytoplasmic stream of two- and four-cell embryos. Morpholinos to other galactosyltransferases or an irrelevant morpholino oligonucleotide (5′-CCTCTTACCTCAGTTACAATTTATA-3′) were injected as controls. Full-length transcripts of β4GalT5 and mutated β4GalT5 were subcloned into the pCRII plasmid (Invitrogen). Mature capped and poly-adenylated mRNA was transcribed from BamH1 linearized plasmids using the T7 or Sp6 mMessage mMachine kit (Ambion). mRNA was diluted in water and various concentrations were injected into the cytoplasmic stream of two-and four-cell embryos. Antisense mRNA at the same concentrations was injected as a control.

Embryos were lyzed in ice-cold lysis buffer (0.5% Triton X-100, 150 mM NaCl, 20 nM Tris-HCl, 10 mM EDTA) and electrophoresed on 7.5% SDS-PAGE forα-Smad blots. Fifteen percent SDS-PAGE gels were used for α-Bmp blotting. Blots were transferred to PVDF (Millipore) membranes and blocked in 5% nonfat dry milk in 20 mM Tris-HCl, 150 mM NaCl (TBS) plus 0.1% Tween-20(TBST) for 1 hour at room temperature. Following block, membranes were incubated in primary antibody (α-Smad5, 1:1000, Cell Signaling;α-phospho-Smad1/5/8, 1:1000, Cell Signaling; α-Bmp2, 2 μg/ml,Sigma; or α-Bmp7, 2 μg/ml, Alpha Diagnostic International) in 5% BSA in TBST for 12 hours at 4°C. Following three washes in TBST, membranes were incubated in secondary antibody [goat α-rabbit-HRP (Santa Cruz) orα-mouse-HRP (Santa Cruz), both at 1:25,000 dilutions] in 5% BSA in TBST for 1 hour at room temperature. Following three washes in TBST and two washes in TBS, antibody reactivity was detected using ECL detection (Amersham).

Proteoglycans were extracted as described previously from deyolked 85%epiboly embryos (Hascall and Kimura,1982). Briefly, embryos were placed in ice-cold 4 M guanidine-HCl,0.2% w/v zwittergent 3-12, 50 mM sodium acetate, 10 mM EDTA with protease inhibitors (Roche), and incubated at 4°C for 1 hour. The lysates were cleared by spinning at 15,000× g for 10 minutes at 4°C. The supernatant was removed and dialyzed against 20 mM Tris-HCl, pH 7.4, for 18 hours at 4°C. The proteoglycans were precipitated using 5%cetylpyridinium chloride at 37°C for 1 hour. The precipitate was washed in 0.5 M sodium acetate/95% ethanol followed by 0.5 M sodium acetate/10% ethanol. The final pellet was resuspended in 20 mM HEPES, pH 7.4. For SDS-PAGE of isolated proteoglycans, samples were run on a 5% polyacrylamide gel. The gel was fixed in 50% methanol/5% acetic acid for 1 hour and washed in 3% acetic acid for 20 minutes with one change of the wash solution. Carbohydrates were oxidized for 30 minutes with 10 mg/ml periodic acid in 3% acetic acid. The gel was again washed for 20 minutes with 3% acetic acid and stained with Pro-Q emerald (Molecular Probes) for 2 hours and visualized at 300 nm. The gel was then stained for total protein by incubating with SYPRO stain (Molecular Probes) and visualized at 300 nm.

For affinity chromatography, 100 μl of purified proteoglycans were coupled to an Affi-Gel 10 (BioRad) column. Briefly, 100 μl Affi-Gel 10 was washed with distilled water. Purified proteoglycans were added to the matrix,and incubated at room temperature for 1 hour. Following ligand binding,unbound sites were blocked with 0.1 M ethanolamine-HCl, pH 8, for 1 hour at room temperature. The bound and blocked matrix was added to a 1 ml syringe filled with approximately 100 μl volume of glass wool. The column was washed with 3 ml HEPES, pH 7.4 and stored at 4°C. All liquids were applied to the column and allowed to flow by gravity. Prior to use, the columns were washed with 1 ml 20 mM Tris-HCl, pH 7.4. Recombinant human BMP2 (Sigma) or recombinant human BMP7 (Sigma) (2 μg) was applied to the columns and allowed to bind. Following binding, the columns were washed with 1 ml 20 mM Tris-HCl, pH 7.4, and the BMP was eluted with increasing concentrations of NaCl in 20 mM Tris-HCl, pH 7.4. Fractionated eluent was separated by SDS-PAGE and probed for BMP2 or BMP7 by western blotting, as described above.

The zebrafish EST and genomic sequence databases were queried for sequences that are similar to mammalian β4GalTs. One transcript identified had 69.9% identity at the amino acid level to the human β4GalT5(Fig. 1A) and was designatedβ4galt5 (Accession number DQ104219). The zebrafish β4galt5 transcript is predicted to contain nine exons(Fig. 1B), and retains the genomic structure predicted for human β4GalT5. mRNA transcribed from the zebrafish β4galt5 gene is 1,143 base pairs in length and encodes a 382 amino acid protein. The zebrafish sequence contains conserved motifs that are thought to be essential for sugar-nucleotide binding, including the FNRA, DVD and WGWG(G)EDDD motifs (Hagen et al., 1999). A simple phylogenetic analysis(Fig. 1C) reveals that zebrafish β4GalT5 is ancestral to a subgroup of the β4GalT family that includes the β4GalT5s and the β4GalT6s.

To define the temporal expression of β4galt5,semi-quantitative RT-PCR was preformed with staged RNA libraries(Fig. 2A). In the oocyte, high levels of control mRNA [brul, a maternally and zygotically expressed RNA-binding protein (Suzuki et al.,2000)] were detected (data not shown); however, there was no detectable expression of β4galt5. Low levels of β4galt5 were first evident at 50% epiboly and reached peak expression by the 12-somite stage. Levels of β4GalT5 remained consistently high in Prim-16 stage embryos. Thus, β4GalT5 is not maternally loaded into the oocyte and peak expression is reached by mid-somitogenesis.

To address the spatial localization of β4GalT5, whole-mount in situ hybridization was performed. Low expression was first detectable in 50%epiboly embryos (Fig. 2B) and was uniform throughout the embryo. Occasionally, slightly higher levels of expression were seen in the hypoblast along the primary embryonic axis (arrow, Fig. 2B), although this may reflect increased tissue mass along the axis. During early somitogenesis (bud stage; Fig. 2C), β4GalT5 was expressed uniformly throughout the embryo, including expression in the polster. Later in somitogenesis (12 somites, Fig. 2D), expression was elevated in the ventral regions of the embryo, as well as at the midline(arrow, Fig. 2D). By the 26-somite stage, it was possible to identify specific structures with noticeably higher levels of β4GalT5, such as in the intersomitic spaces of the trunk (arrows, Fig. 2E),as well as in the otic vesicle (insert, Fig. 2E). Parallel incubations using the sense β4GalT5 probe produced no detectable signal.

The widespread expression of β4GalT5 makes it difficult to predict its developmental role. To address its function in vivo, morpholino oligonucleotides specific to β4GalT5 were injected into two- and four-cell embryos. A total of three independent morpholinos were used: two different translation blocking morpholinos (MO1, MO3) and one splice blocking morpholino (MO2); all injected morpholinos produced nearly identical phenotypes (Table 1), and all subsequent studies were repeated with at least two independent morpholinos. For controls, ∼20 embryos were injected in each experiment with equal amounts of an irrelevant morpholino. The degree of embryonic death was similar(<10% death) in all injections, irrelevant of the morpholino injected. However, developmental defects were only seen in embryos injected withβ4GalT5-specific morpholinos; all surviving embryos injected with control morpholinos appeared normal (Fig. 3B).

The earliest phenotype detectable in morpholino-injected embryos was evident at 80% epiboly as an exaggerated elongation of the embryo. By the 2-somite stage, morpholino-injected embryos were clearly elongated and displayed a defective tail bud (Fig. 3A). Extensive coiling of the tail and other indicators of dorsalization were seen when control embryos had reached the 26-somite stage,at which time three classes of morphant phenotypes of increasing severity were observed.

The β4GalT5 morphant phenotypes were classified according to the criteria defined by Mullins et al.(Mullins et al., 1996) for dorsoventral defects. The three classes of β4GalT5 morphants roughly correspond to the three most severe classes of dorsoventral phenotypes described by Mullins et al. (Mullins et al., 1996). Class 3 embryos are similar to the pgy mutant(Fig. 3C), in that they have a slightly coiled tail indicative of mild dorsalization, and display moderate ear defects, including small otic fields and an absence of otoliths (asterisk, Fig. 3E). Interestingly, both wild-type and β4GalT5MO embryos express pax2a, a marker of the otic vesicle (Pfeffer et al.,1998); however, the pax2a field in β4GalT5MO embryos is smaller and rounder than in wild-type embryos(Fig. 3E).

Class 4 embryos are similar to the snh mutant and display a more significant coiling of the tail, as well as dorsalization within the anterior regions of the embryo, and are considered moderately dorsalized(Fig. 3C). Due to the severity of the phenotype, no otic structures were detected in this class of morphants,although both Class 4 and Class 5 embryos, in our hands, have pax2astaining in a region that is consistent with the otic field (data not shown). Class 5 embryos, similar to swr mutants(Fig. 3C), are the most severely affected and display a completely dorsalized phenotype.

The penetrance of the morpholino phenotype was directly dependent upon the amount of β4GalT5 morpholino injected. As demonstrated forβ4GalT5MO1 (Table 1),injection of 5 ng morpholino resulted in a greater proportion of Class 3 embryos and few Class 5 embryos; whereas 20 ng resulted in high proportions of Class 5 embryos and few Class 3 embryos. Thus, the severity of theβ4GalT5 morpholino phenotype was dose dependent.

Whereas the use of multiple independent morpholino oligonucleotides is taken as evidence that the phenotype results from downregulating the target(i.e. β4GalT5) transcript (Nechiporuk et al., 2005; Yan et al.,2005), we tested the specificity of the β4GalT5 morpholinos by injection of mRNA encoding full-length β4GalT5 in combination withβ4GalT5MO3 (Table 1). Injection of 40 pg full-length mRNA in combination with 10 ng morpholino resulted in a rescue of the β4GalT5 knockdown phenotype, i.e. a wild-type appearance (Fig. 3D). Injection of full-length β4GalT5 mRNA in the absence of morpholino did not produce any noticeable phenotype. These results demonstrate that the phenotype observed following injection of β4GalT5 morpholino oligonucleotides results from a specific reduction of β4GalT5.

Because the β4GalT5MO phenotype grossly phenocopies the swr,snh and sbn mutations, which are characterized by dorsalized embryos resulting from mutations in bmp2b, bmp7 or smad5,respectively (Dick et al.,2000; Hild et al.,1999; Kishimoto et al.,1997), it was important to determine whether downregulatingβ4GalT5 had a synergistic effect with these mutations. Knockdown ofβ4GalT5 in swr (bmp2b) mutants failed to reveal any additional phenotype, which may simply reflect the pre-existing severely dorsalized phenotype in swr mutants. However, knockdown ofβ4GalT5 accentuated the moderate phenotype of snh(bmp7) mutants into a more severe dorsalized appearance, similar to that seen in the swr mutant (22 out of 24 morpholino-injected embryos showed a swr phenotype versus 0 out of 24 control-injected embryos). The failure of β4GalT5 knockdown to influence the swr phenotype and to increase the dorsalization of the more moderate snh mutant is consistent with a β4GalT5 function in Bmp signaling.

Bmp2b and Bmp7 generate a negative-feedback loop with chordin that is required for the proper establishment of the dorsoventral margin. In mutants with defective Bmp signaling, such as swr and snh, chordin expression invades the ventral hemisphere, a result of relieving the Bmp inhibition (Miller-Bertoglio et al.,1997). Therefore, we examined chordin expression in 85%epiboly embryos by in situ hybridization to determine whether Bmp signaling is altered following β4GalT5 knockdown. As reported, chordinexpression was high in the dorsal axis of control-injected embryos(Fig. 4A; asterisk indicates dorsal axis in all panels) and restricted from the ventral hemisphere (arrows, Fig. 4A). β4GalT5MO embryos had no obvious dorsal axis when viewed dorsally(Fig. 4B); however, the axis was evident in the animal view (Fig. 4B). Moreover, chordin expression was expanded both ventrally and anteriorly (Fig. 4B). The expanded chordin expression in the presumptive ventral domain was also apparent when viewed from the vegetal pole(Fig. 4B), where chordin expression had completely enveloped the ventral hemisphere.

Because dorsalization is a common side effect of morpholino injection, we assayed the expression of other signaling pathways that participate during dorsoventral axis patterning. Of these, the Fgf signaling pathway is a crucial component underlying dorsoventral patterning, and, consequently, we examined the expression of mkp3, a downstream mediator of Fgf signaling(Tsang et al., 2004). As shown in Fig. 4, mkp3expression is unaltered in the β4GalT5 morphant background, being localized to the dorsal axis in both control-injected and β4GalT5 morphants (arrowheads). Similarly, we determined that the specification and/or patterning of dorsal structures appears grossly normal in β4GalT5MO embryos, as assayed by in situ hybridization of the dorsal markers ntl and pax2 (data not shown). Collectively, all of the morphological and molecular characterization of the β4GalT5MO phenotype is indicative of disrupted Bmp signaling, and, consequently, Bmp signaling was assayed directly in β4GalT5MO embryos.

Bmps affect gene expression through the activation of the Smad family of transcription factors. Bmp receptor activation leads to phosphorylation of the R-Smad proteins (receptor-regulated Smads1/5/8), which are subsequently released from the receptor and form a heteromeric complex with the common Smad, Smad4. The phosphorylated R-Smad/Smad4 complex is translocated into the nucleus, where it regulates gene transcription(Mehra and Wrana, 2002).

During dorsoventral patterning, Bmp2b signaling results in the phosphorylation of Smad5 (and possibly Smad1 and/or Smad8)(Wrana and Attisano, 2000),and, consequently, the levels of total and activated (i.e. phosphorylated)R-Smads were assayed in 80% epiboly embryos following injection of either control or β4GalT5 morpholinos (Fig. 5). Whereas the total level of R-Smad5 (and other R-Smads as well)was similar in control and β4GalT5MO embryos(Fig. 5A), the level of phosphorylated R-Smads in β4GalT5MO embryos was only 27% of control levels (Fig. 5B), indicating a severe reduction in the activation of the Bmp signaling pathway. Similar results were obtained using β4GalT5MO embryos injected with either splice blocking or translation blocking oligonucleotides. We next examined the basis whereby a defect in β4GalT5-dependent glycosylation alters the activation of Bmp-dependent Smad proteins.

It has become clear during the past few years that the ability of soluble cytokines to maintain stable expression domains is dependent upon their binding to the glycosaminoglycan (GAG) chains of large molecular weight proteoglycans. However, virtually all of our knowledge comes from the study of specific cytokines, such as Egf and Fgf, which bind to defined pentasaccharide structures within the heparan sulfate chains of proteoglycans(Hardingham and Fosang, 1992; Norton et al., 2005). Although other cytokines, including members of the Tgfβ superfamily to which the Bmps belong, have also been shown to bind proteoglycan GAG chains, the overall binding specificity of proteoglycans for these other cytokines has yet to be demonstrated. As we have no evidence to suggest that the expression of either bmp2b or bmp7 is altered, we examined whether the defective Bmp signaling characteristic of β4GalT5MO embryos can be attributed to abnormal proteoglycan biosynthesis.

Initially, proteoglycans were isolated from both wild-type andβ4GalT5MO embryos. On average, each wild-type embryo produced 18.3 ng of proteoglycans; whereas, only 4.8 ng were isolated from a typicalβ4GalT5MO embryo. Similar results were found using three independent proteoglycan isolations. To determine whether this reduction in proteoglycan mass was associated with alterations in protein and/or carbohydrate content,the proteoglycan preparation was resolved by SDS-PAGE under non-reducing conditions and stained for both protein and carbohydrate. Two prominent bands were apparent with relative molecular mass of >400 kDa and ∼300 kDa,both of which contained similar protein levels in control and β4GalT5MO preparations (Fig. 6A). Both polypeptide species were highly glycosylated in the control sample(Fig. 6A); however, the same polypeptide bands showed dramatically less carbohydrate content in theβ4GalT5MO sample (Fig. 6A). Thus, the proteoglycan core proteins appear to be synthesized normally in β4GalT5MO embryos, but are grossly underglycosylated relative to those in control-injected embryos. Surprisingly, the underglycosylated proteoglycans from β4GalT5 morphants resolved at a similar molecular mass to control proteoglycans, suggesting either that the β4GalT5 deficiency does not lead to a global loss of GAG side-chains or that any reduction in molecular weight, given the relative amount of carbohydrate in the native proteoglycan, is too small to be resolved by the 6% SDS-polyacrylamide gel. In any event, the proteoglycans from β4GalT5 morphants show reduced glycosylation, and we therefore determined whether this influenced its ability to bind Bmp2 and/or Bmp7.

Proteoglycans were extracted from 80% epiboly stage embryos and coupled to an affinity support to which recombinant human BMP2 or BMP7 was applied, and any unbound protein was removed by washing. Bound BMP2 or BMP7 was eluted by increasing the salt concentration. BMP2 was eluted from control proteoglycans(Fig. 6B) at 0.8-1.6 M NaCl,similar to what others have shown for the binding of recombinant Noggin and Bmp4 to synthetic heparin sulfate(Paine-Saunders et al., 2002). When BMP2 was applied to proteoglycans isolated from β4GalT5MO embryos(Fig. 6B), BMP2 eluted from the column at 0.2 M NaCl and was completely eluted by 0.8 M NaCl. Similar results were obtained using proteoglycans isolated from β4GalT5MO embryos injected with either splice blocking (MO2) or translational blocking (MO3)oligonucleotides. This demonstrates that proteoglycans isolated fromβ4GalT5MO embryos show a reduced binding affinity for BMP2, relative to control proteoglycans.

We determined whether the decrease in BMP2 affinity to β4GalT5MO proteoglycans was specific to BMP2 or characteristic for other cytokines as well. As Bmp7 is the other major ventralizing agent in the late epiboly stage embryo, the affinity of BMP7 for control and β4GalT5MO proteoglycans was analyzed (Dick et al., 2000). BMP7 showed a distinctly different elution profile from that seen with BMP2,with a broad elution profile between 0.4-1.4 M NaCl(Fig. 6B). Unlike that seen with BMP2, BMP7 showed a similar elution pattern from proteoglycans isolated from both control-injected and β4GalT5MO embryos. This indicates thatβ4GalT5 generates a proteoglycan epitope that has apparent specificity for Bmp2, but not for Bmp7.

The results presented here represent the first functional characterization of a specific glycosyltransferase during early patterning of the zebrafish embryo. β4GalT5 was initially identified by an in silico search, and was subsequently cloned from a 48-hour RNA library. At the amino acid level,β4GalT5 is 69.9% identical to the human β4GalT5. β4GalT5 is not expressed in the oocyte, but is expressed by the early epiboly stage and reaches a steady state level of expression by mid-somitogenesis. Furthermore,β4GalT5 shows widespread expression throughout the embryo during the first 24 hours of development, with enhanced expression within several structures at the 20-somite stage. Insight into the biological function ofβ4GalT5 was addressed by the injection of three specific morpholino oligonucleotides, all of which resulted in a dorsalized embryo. Consistent with the hypothesis that β4GalT5 is essential for patterning of the dorsoventral axis, the expression of chordin was inappropriately expanded into the ventral hemisphere, similar to what is observed with well-studied dorsoventral patterning mutants(Miller-Bertoglio et al.,1997), thus suggesting that Bmp signaling was defective. This was confirmed by a dramatic reduction in the activation of the Bmp-dependent transcription factors Smad1/5/8. Because the trafficking and signaling efficacy of peptide cytokines are thought to be regulated by binding to GAG chains of proteoglycans, it is noteworthy that proteoglycans isolated fromβ4GalT5MO embryos demonstrated defective glycosylation and a greatly reduced affinity for Bmp2. These results suggest that β4GalT5 generates an epitope within the glycoside chains of proteoglycans that is required for proper Bmp signaling.

Previous analysis of glycosyltransferase function during development has relied upon generating knockouts of individual transferases or through random mutagenesis. Although these approaches have yielded insights into the function of a few specific glycosyltransferases, there is still a plethora of glycosyltransferases that remain uncharacterized(Bulik and Robbins, 2002; Furukawa et al., 2001; Lu et al., 1997; Maly et al., 1996; Wandall et al., 2005). Furthermore, because of the large number of glycosyltransferases now known to exist in mammalian tissues, and because of a poor understanding of the substrate specificity for each enzyme, it is difficult to accurately identify individual glycosyltransferases that are likely to be essential during development. For example, there are currently six confirmed β4GalTs in the mammalian genome, and this reflects just one arm of the larger glycosyltransferase `superfamily' (Lowe,1991). An initial search of the human genome suggests upwards of 300 glycosyltransferases are encoded. Generating targeted knockouts in all of these genes to identify those that have functions during development would be a monumental task.

Invertebrate systems have yielded new insight into developmentally essential glycosyltransferases. For example, in C. elegans, sqv-3,sqv-7 and sqv-8 predominantly affect the glycosylation of chondroitin sulfate and heparan sulfate proteoglycans(Bulik and Robbins, 2002). As a result of defective glycosylation, the vulval epithelium that normally invaginates to form a tube, is either collapsed or completely absent. Similarly, the Drosophila mutants fringe, brainiac and egghead all encode glycosyltransferases with important developmental roles. Fringe is a N-acetylgalactosaminyltransferase that is required for regulating Notch signaling by modulating its ligand-binding specificity(Moloney et al., 2000). brainiac and egghead are embryonic lethal mutations that encode glycosyltransferases required for the synthesis of the core carbohydrate in glycosphingolipids(Wandall et al., 2005).

Classically, glycosyltransferases are required for the posttranslational modification of virtually all membrane bound and secreted glycoproteins and glycolipids, so it is not surprising that β4GalT5 showed widespread expression throughout the embryo. Although the widespread expression of glycosyltransferases has made it difficult to determine which ones are likely to play crucial roles during embryogenesis, the power of zebrafish has allowed us to screen the knockdown phenotype of individual transferases to identify those that are essential during development. Through this approach, we have identified a β4GalT transcript that is required for dorsoventral patterning of the early embryo, but which had no precedent for having a role in vertebrate or invertebrate embryogenesis.

Specification of the dorsoventral axis is dependent upon Bmp signaling, as shown by mutations in swirl/bmp2b, snailhouse/bmp7and somitabun/smad5 (Dick et al., 2000; Hild et al.,1999; Nguyen et al.,1998). At the onset of gastrulation, Bmps are present throughout the embryo, and not until the initial stages of dorsoventral patterning do Fgf8, Bozozok, and possibly members of the Wnt family repress Bmp signaling in the presumptive dorsal region of the embryo(Furthauer et al., 2004; Kishimoto et al., 1997; Solnica-Krezel and Driever,2001). Once the dorsal organizer has been established, Chordin,Follistatin and Noggin repress Bmp signaling by physically inhibiting the binding of Bmp to its receptor (Iemura et al., 1998; Piccolo et al.,1996; Zimmerman et al.,1996). A feedback loop is established whereby Bmp2b/Bmp7 inhibit Chordin through the activation of a chordin-specific protease, Tolloid(Blader et al., 1997). Upon Bmp receptor activation, the R-Smad proteins (e.g. Smad5) are phosphorylated and released from the receptor complex to form heteromeric complexes with Smad4;these heteromeric complexes then translocate to the nucleus where they activate the transcription of genes that direct ventralization(Kimelman and Pyati, 2005; Mehra and Wrana, 2002; O'Connor et al., 2006; Padgett et al., 1998).

Defective Bmp signaling leads to the inappropriate expansion of chordin expression into the presumptive ventral domain. In this study, we observed a similar expansion of chordin expression, suggesting a reduction of Bmp signaling in β4GalT5MO embryos. Furthermore, we determined that the activated Smad signaling complex was inefficiently activated in β4GalT5MO embryos, thus confirming a reduction in Bmp signaling. In order to determine how defective galactosylation in β4GalT5MO embryos could affect Bmp-dependent signaling pathways, we compared proteoglycans from control andβ4GalTMO embryos, as proteoglycans are known to regulate cytokine binding and trafficking across the epithelial sheet.

Proteoglycans are a diverse class of extracellular proteins that consist of a protein core anchored to the plasma membrane by either a transmembrane domain (syndecans) or by a glycosylphosphatidylinositol (GPI) anchor(glypicans); some proteoglycans can be secreted as well (perlecans)(Kreis and Vale, 1999). Attached to the core protein are multiple, large molecular weight GAG chains that account for as much as 90% of the proteoglycan mass, and which contain a linker region followed by a repeating disaccharide unit unique to each GAG type (Sugahara and Kitagawa,2000). GAG chains can be further modified by the addition of sulfate groups to specific monosaccharide residues(Chapman et al., 2004), and there is some evidence to suggest that the position of sulfate groups along the GAG chain directs cytokine binding. Knockdown of a specific sulfotransferase in zebrafish, zHS6ST, results in a phenotype similar to knypek, which encodes a glypican involved in non-canonical Wnt signaling, thus suggesting a role for GAG chains in the modulation of Wnt signaling (Bink et al., 2003; Topczewski et al., 2001). However, the Drosophila homolog of zHS6ST, dHS6ST, is not involved in Wnt signaling, although dHS6ST participates in Fgf signaling(Kamimura et al., 2001). Although these studies suggest that sulfation can be crucial to ligand binding, it is still unclear how alterations in sulfation directly regulate ligand specificity. Our work suggests that carbohydrate moieties outside of the traditional GAG chain and independent of sulfation can impact ligand-binding specificity as well.

Of all of the proteoglycans, the most intensely studied are those that contain heparan sulfate GAG chains. dally is a Drosophilaheparan sulfate proteoglycan that has been shown to interact with dpp(a member of the Tgfβ family of cytokines that includes the Bmps) in the Drosophila imaginal disk (Jackson et al., 1997). Dally is also required for wingless (wg) signaling in the wing disc by interacting with the Wg receptor, Frizzled 2 (Cadigan et al., 1998). However, there is no evidence that dallyinteracts with dpp during embryogenesis, where dpp is essential for dorsoventral axis patterning.

We interpret the results presented here to suggest that β4GalT5 participates in the synthesis of oligosaccharide chains of zebrafish proteoglycans that are essential for Bmp2 binding and subsequent presentation to its receptor, thus triggering Smad activation. β4GalT5 has no apparent role in the synthesis or expression of Bmp2, only in its ability to bind and/or activate its receptor. Consequently, injection of recombinant Bmp2, or its mRNA, would not be expected to rescue the morphant phenotype unless one could bypass the requirement for proteoglycans and insure that Bmp2 was presented to its receptor with equal efficacy as wild type. In a similar study, defective synthesis of heparan sulfate proteoglycans that leads to reduced Fgf10 signaling could not be rescued by the injection of Fgf10 protein(Norton et al., 2005).

It is interesting that whereas proteoglycans isolated from β4GalT5MO embryos show reduced affinity for Bmp2, relative to control proteoglycans,binding of the closely related cytokine Bmp7 was relatively unaffected. The structure of the β4GalT5 epitope involved in Bmp2 binding is of obvious interest, but because the substrate specificity of the β4GalT5 identified here remains unknown, identification must await structural analysis of the relevant proteoglycan chains. In any event, this is the first report in which the ligand binding affinity of an endogenous proteoglycan can be modulated by a specific β1,4-galactosylation. Furthermore, these results raise the possibility that the ligand-binding specificity of proteoglycans may be defined by a carbohydrate `code' involving glycoside residues both internal and external to the traditional GAG chains.

The authors thank Drs Win Sale, Karl Saxe and Iain Shepherd for suggestions regarding the manuscript. This work was supported by grant RO1 DE07120 from the NIH to B.D.S. and A.F.