The Notch signaling pathway mediates communication between neighboring cells, controlling cell-fate determination and the formation of fine-grained patterns in a wide range of animal tissues (reviewed in Lewis, 1998; Artavanis-Tsakonas et al., 1999; Irvine and Rauskolb, 2001; Pourquié, 2001; Rida et al., 2004). One of its major roles consists in the specification of border cells at boundaries between juxtaposed cell populations (reviewed in Irvine and Rauskolb, 2001). This process depends on Fringe, a glycosyltransferase (Moloney et al., 2000; Brückner et al., 2000) that can glycosylate Notch. This process is thought (in Drosophila at least) to make Notch more susceptible to activation by its ligand Delta and less susceptible to activation by the alternative ligand Serrate. Thus, when a patch of cells expressing Notch, Serrate, and Fringe confronts a patch of cells expressing Notch and Delta, Notch is strongly activated only at the interface, because it is only there that cells expressing a given ligand make contact with cells susceptible to activation by that ligand (Panin et al., 1997). fringe was first found in Drosophila (Irvine and Wieschaus, 1994); subsequently, several homologues were discovered in vertebrates (Table 1 and references therein). Fringe-modulated deployment of Notch signaling has been shown to be critical for formation of developmental boundaries in Drosophila wing, Drosophila eye, Drosophila leg, chicken limb, vertebrate somites, and chicken zona limitans intrathalamica (zli) and may have a similar role in other tissues such as the hindbrain, although the detailed mechanisms may vary (reviewed in Irvine and Rauskolb, 2001). Fringe is also important for some other Notch-dependent processes; in chick and mouse, in particular, it plays an essential part in the “segmentation clock”—the oscillation of gene expression in the presomitic mesoderm during somitogenesis (Dale et al., 2003).
Table 1. Comparison of Expression Pattern of Fringe Genes*
Xenopus and newt
Mouse and rat
Note that entries in the Table marked NR do not necessarily imply that expression is absent, only that it has not been reported. NR, not reported; PSM, presomitic mesoderm. Superscript letters in table body indicate the following:
In this study, we describe the cloning and expression pattern of three zebrafish fringe homologues: lunatic fringe (lfng), radical fringe (rfng), and manic fringe (mfng). One of these, lfng, has already been described (Prince et al., 2001; Leve et al., 2001). The other two are novel. We note that lfng is expressed in the sensory patches of the inner ear in addition to the sites reported before (Prince et al., 2001; Leve et al., 2001); rfng is expressed in adaxial cells, tectum and rhombomere boundaries and at a low level in somites; and mfng at early stages is expressed only at a low level, such that we detect it only by reverse transcription-polymerase chain reaction (RT-PCR) and not by in situ hybridization, although later it becomes detectable by in situ hybridization in the ear. We find that rfng expression in the hindbrain is lost in mib mutants, in which Notch signaling is defective and rhombomere boundaries are not formed. In this tissue at least, therefore, normal rfng expression depends on positive regulation by Notch signaling, suggesting the existence of a positive feedback tending to reinforce the expression of rfng in boundary cells. Although lfng is expressed in nascent and formed somites, none of the three zebrafish fringe genes is detectably expressed in the posterior presomitic mesoderm, suggesting that, in contrast with chick and mouse, the segmentation clock in the zebrafish does not depend on Fringe activity.
RESULTS AND DISCUSSION
Cloning and Sequence Analysis
Chicken Lunatic fringe (Lfng) was used as a probe to screen a zebrafish cDNA library (see Experimental Procedures section for details) at low stringency. Two fringe genes were identified from sequences present in several independent clones of various lengths, including full-length. These were compared by BLAST with other vertebrate fringe genes in NCBI GenBank (http://www.ncbi.nlm.nih.gov/BLAST/) and were named lunatic fringe (lfng, AF510992) and radical fringe (rfng, AF510993), respectively (Fig. 1A). Zebrafish Lfng is 70%, 70%, and 61% identical in amino acid sequence to Lunatic fringe of Xenopus, chicken, and mouse, respectively, as described elsewhere (Prince et al., 2001; Leve et al., 2001). Zebrafish Rfng is 67%, 65%, 59%, 59%, 57%, and 56% identical in amino acid sequence to Radical fringe of newt, chicken, Xenopus, human, mouse, and rat, respectively, and has not been described previously. A TBLASTN search of the most recent ENSEMBL release (v19.3a.2, 3 March 2004) of the zebrafish whole genome sequence, querying with the amino acid sequences of each of the three mammalian Fringe homologues, revealed the existence of a third zebrafish fringe gene (and no others beyond this). A BLAST search of the expressed sequence tag database (dbEST) identified a zebrafish EST clone, AL923400 (Lo et al., 2003), corresponding to a fragment of a transcript of this gene, and a further BLAST search with this fragment identified a further EST clone (CF348026, NIH-MGC, http://mgc.nci.nih.gov/), which, when fully sequenced, was found to contain full-length coding sequence for a novel 360-amino acid Fringe protein. Sequence comparisons indicate that this is a homologue to mouse, rat and human Manic Fringe, with 50%, 50%, and 49% amino acid sequence identity to them, respectively. It matches human Lfng, mouse Lfng, mouse Rfng and human Rfng slightly less well with 49%, 47%, 44%, and 43% amino acid sequence identity, respectively, and the similarity tree and features of the sequence that are consistent between species tend to confirm that it is a zebrafish ortholog of Manic fringe (see Fig. 1). Thus, we have named the protein zebrafish Mfng. According to the latest ENSEMBL release, the zebrafish lfng, rfng, and mfng genes are located on chromosomes 3, 1, and 22, respectively. We will mainly focus on rfng and mfng here.
The protein sequences of zebrafish Rfng and Mfng, inferred from the cDNA sequences, show almost all the structural features of the corresponding tetrapod Radical and Manic Fringe proteins (see Fig. 1A for details, Wu et al., 1996; Johnston et al., 1997; Cohen et al., 1997; Laufer et al., 1997; Rodriguez-Esteban et al., 1997; Cadinouche et al., 1999; Mikami et al., 2001), including a putative signal peptide, potential N-linked glycosylation sites, a DXD-motif responsible for the glycosyltransferase catalytic activity (Moloney et al., 2000; Brückner et al., 2000) and six conserved cysteines that are thought to form disulfide bonds (Irvine and Wieschaus, 1994). Interestingly, the putative tetrabasic proteolytic sites are only found in Xenopus and chicken Rfng, suggesting that zebrafish, newt, rat, mouse, and human Rfng proteins may not require regulated proteolytic activation.
Expression Patterns of lfng and mfng
Zebrafish lfng is expressed in the central nervous system—notably in rhombomeres 2 and 4—and in somites, lateral plate mesoderm, and prechordal plate mesoderm (Prince et al., 2001; Leve et al., 2001; and data not shown). It is also expressed in the otic vesicles, which has not been documented previously in the zebrafish. In the mouse, Mfng and Lfng are expressed in the developing sensory regions within the otic epithelium (Johnston et al., 1997; Morsli et al., 1998), and the same is true for lfng in Xenopus (Wu et al., 1996) and for Lfng in the chick, where Lfng and Serrate-1 have been shown to have closely similar otic expression domains (Laufer et al., 1997; Cole et al., 2000). In zebrafish, we find likewise that lfng and genes of the serrate and delta families are expressed together in the developing sensory patches at the anterior and posterior ends of the early otocyst (Fig. 2A,B and Haddon et al., 1998a). At later stages, expression of serrateB (and, transiently, of the delta genes) in sensory patches becomes confined to differentiated hair cells (Haddon et al., 1998a) while lfng and serrateA become confined to the supporting cells (Rachael Brooker, personal communication, and data not shown).
No mfng expression can be detected by whole-mount in situ hybridization (WISH) in the early stages. From the 20-somite stage, the expression of mfng in the anteroventral part of otic vesicle is gradually up-regulated and becomes obvious by ∼30 hr postfertilization (hpf; Fig. 2C,D). To test for low-level mfng expression, below the detection threshold of WISH, we used RT-PCR. As shown in Figure 3A, mfng expression is undetectable up to 13 hpf (8 somites) but can be detected from 15 hpf (12 somites) onward. mfng may be expressed in the kidney as well, although not detected by WISH, because there is an EST clone isolated from a kidney-specific cDNA library (CD284742, from NCI_CGAP_ZKid1 library).
While the fringe expression patterns in the ear seem to be well-conserved, there are important species differences in the presomitic mesoderm (PSM). Mouse and chick embryos show oscillating expression of Lfng throughout the PSM, and this gene is required for oscillating expression of other genes in this broad region. By contrast, zebrafish lfng is only expressed in the rostral parts of formed somites and in the most anterior extremity of the PSM where the new somites are about to form (Forsberg et al., 1998; McGrew et al., 1998; Aulehla and Johnson, 1999; Prince et al., 2001; Leve et al., 2001; Dale et al., 2003; data not shown). It is possible that the described lfng is not the ortholog of mouse/chick Lfng and that some other fringe is cyclically expressed in the zebrafish PSM. Our data argue against this, however. Our search of the current assembly of the whole zebrafish genome (admittedly still not quite complete) has revealed only three zebrafish fringe homologues, and none of these—neither lfng nor rfng nor mfng—is expressed at detectable levels in the posterior PSM (with the exception of the adaxial cells, where rfng is expressed in a nonoscillatory pattern—see below). This finding suggests a fundamental difference between zebrafish and tetrapods in the function of Fringe in the segmentation clock. Consistent with this idea, the expression pattern of a core component of the segmentation clock, Delta, in the PSM is different: whereas expression of zebrafish deltaC is oscillatory, that of mammalian Dll1 and Dll3 and of chick C-Delta-1 is not (Bettenhausen et al., 1995; Dunwoodie et al., 1997; Palmeirim et al., 1998; Jiang et al., 2000; and reviewed in Pourquié, 2001 and Rida et al., 2004).
Expression Pattern of rfng
rfng is maternally expressed (Fig. 3B), and its expression is faintly visible by the end of gastrulation in parts of the future brain and mesoderm (Fig. 2E and data not shown). Its mesodermal expression starts in the two parallel rows of cells flanking the midline with cuboidal epithelioid morphology (adaxial cells) in the undifferentiated PSM (Fig. 2E,F). At the three-somite (s) stage, rfng is also expressed in the tail bud at a low level (Fig. 2F). The expression in adaxial cells is prominent in the segmentation period (Fig. 4E–G and data not shown). When the somites mature and the intersomitic boundaries form, expression of rfng continues in the medial part of each somite, with low-level expression in the lateral part, but appears excluded from the regions immediately facing the intersomitic clefts (although this finding may simply reflect the high concentration of cell nuclei in these regions; Fig. 4E). At approximately the 12s stage, rfng expression is more obvious in the midbrain and hindbrain regions (Fig. 2G). The midbrain expression persists and becomes restricted to the tectum (Fig. 2I,J). From approximately the 14s stage, rfng expression becomes gradually up-regulated in rhombomere boundaries, where it persists up to at least 30 hpf (Fig. 2H,K and data not shown); this pattern is similar to that of mariposa (Fig. 4C and Moens et al., 1996).
We have found that several genes of the Notch signaling pathway are expressed segmentally in rhombomeres in a pattern similar to that of notch1a (Bierkamp and Campos-Ortega, 1993), including deltaA, deltaB, and deltaD (Fig. 2K and data not shown). To further confirm that rfng is expressed in the rhombomere boundaries, a two-color in situ hybridization was performed. Whereas deltaA (in red) is expressed in neurons within rhombomeres, rfng (dark purple) is expressed along rhombomere boundaries between deltaA territories (Fig. 2K), in a narrow stripe with a width of approximately two cells (Fig. 2L).
A rhombomere boundary defect has been reported in mibta52b mutant embryos (Jiang et al., 1996), where Delta-Notch signaling fails (Itoh et al., 2003). In these mutants, rfng expression in the hindbrain is dramatically reduced or completely lost, while rfng expression in the tectum is largely unaffected (Fig. 4B and data not shown). This finding suggests that normal rfng expression in the hindbrain depends on positive regulation by Notch activity, suggesting the existence of a positive feedback loop tending to reinforce rfng expression at these sites. Taken together, these findings raise the interesting possibility that Notch signaling is involved in rhombomere boundary formation—a hypothesis that we have explored and confirmed in a separate paper with Cheng et al. (2004). In accord with this hypothesis, we also find that the expression of mariposa is disorderly in mib mutants (Fig. 4D), suggesting a disorganization or loss of rhombomere boundary cells, which could be a consequence of the down-regulation of rfng in mib mutants. The situation in the somites in mutants with defects of Notch signaling is somewhat different: in des/notch1a, aei/deltaD, bea, and mib mutants, somitic expression of rfng is somewhat disorderly in the anterior regions of the body axis where somite boundaries do form and even more disorderly in the middle and posterior regions where no boundaries form; here, although expression is not generally reduced, it is nonsegmental, reflecting the disorder of somite segmentation (Fig. 4F and data not shown). Similarly, in fss mutants, the rfng expression in adaxial cells is maintained but loses its segmental organization (Fig. 4G).
Expression Patterns of fringe Genes in Different Vertebrates Compared
The complicated and dynamic expression patterns of fringe genes in different vertebrates are summarized in Table 1. In some tissues where the expression pattern has been documented in detail, e.g., nervous system (Cohen et al., 1997; Ishii et al., 2000), epidermis (Cohen et al., 1997; Thelu et al., 2002), teeth (Harada et al., 1999; Pouyet and Mitsiadis, 2000; Mustonen et al., 2002), and hair follicles (Favier et al., 2000; Chen and Chuong, 2000), fringe expression appears to mark a contrast between uncommitted cells, which express lunatic fringe, and their committed progeny, which express radical and/or manic fringe, but it is clear from our present data in the zebrafish that this is by no means a universal rule. Moreover, patterns of expression of fringe genes are not always conserved. For example, there are differences of expression between vertebrates in the PSM (McGrew et al., 1998; Forsberg et al., 1998; Aulehla and Johnson, 1999; Leve et al., 2001; Prince et al., 2001; and this study), the hindbrain (Johnston et al., 1997; Leve et al., 2001; Prince et al., 2001; and this study), and the limb buds (Laufer et al., 1997; Christen and Slack, 1998). It is likely that more detailed comparisons would reveal additional differences in expression pattern in other tissues. Such discrepancies suggest differences in mechanisms of developmental processes such as the dorsoventral patterning of limb buds (Christen and Slack, 1998) and the segmentation clock that governs somitogenesis (Lewis, 2003). The contrast with regard to the role of Lfng in this latter mechanism is particularly striking: in the chick and mouse, it is an essential component of the oscillator, whereas in the zebrafish, it is simply not expressed in the relevant region. Expression pattern analysis of the Amphioxus genes AmphiFringe and AmphiNotch reinforces this point, suggesting that while the function of Notch signaling in chordate mesoderm segmentation is ancestral, the involvement of Fringe in this process, as seen in birds and mammals, is derived (Holland et al., 2001; Mazet and Shimeld, 2003).
Fish Stocks and Embryos
Zebrfish embryos were obtained by natural spawnings and maintained normally at 28.5°C in system water. We used the following alleles for somite mutants (van Eeden et al., 1996; Jiang et al., 2000): fssti1 (Nikaido et al., 2002), desth35b (Holley et al., 2002), and mibta52b (the white tail allele, Jiang et al., 1996; Itoh et al., 2003). Embryos were staged as described (Kimmel et al., 1995).
Cloning of Zebrafish lunatic fringe, radical fringe, and manic fringe
Chicken Lfng (a gift from Isabelle Le Roux) was used as a probe to screen a 15–19 hour zebrafish λZAP II cDNA library (a gift from Bruce Appel) at low stringency, yielding two sets of clones corresponding to the zebrafish lfng and rfng genes, respectively. cDNAs were excised from the vector using Rapid Excision Kit (Stratagene). For mfng, EST clones AL923400 (a gift from Jinrong Peng) and CF348026 (IMAGE:7001272, received from RZPD) were used. Complete cDNA sequences were determined in both directions by using an ABI PRISM system and submitted to GenBank under the accession nos. AF510992 (lfng), AF510993 (rfng), and AY608926 (mfng).
Sequence Comparisons and Phylogeny Analysis
Protein alignments were performed by using Clustal W (Thompson et al., 1994; http://clustalw.genome.ad.jp/). A similarity tree was generated with PAUP. The GenBank accession numbers of the compared genes are zebrafish lfng, AF510992; chicken Lfng, U91849; Xenopus lfng,U77640; mouse Lfng, U94351; human Lfng, U94354; zebrafish rfng, AF510993; chicken Rfng, U91850; human Rfng, ENSP00000307971; mouse Rfng, U94350; Xenopus rfng, U77641; newt rfng, AF115388; rat Rfng, AB016486; zebrafish mfng, AY608926; human Mfng, U94352; mouse Mfng, U94349; and rat Mfng, NM_199110. Signal peptide cleavage sites are predicted by SignalP V1.1 (Nielsen et al., 1997; http://www.cbs.dtu.dk/services/SignalP/).
Gene-specific reverse primer was used with Expand Reverse Transcriptase (Roche) to synthesize the first-strand cDNA from total RNA, and this product was then amplified by PCR using a pair of specific primers (predenaturation, 94°C, 2 min; denaturation, 94°C, 30 sec; annealing, 62°C, 30 sec; elongation, 72°C, 45 sec; 35 cycles, from the 11th cycle onward, the elongation time increases 20 sec per cycle). For a negative control, total RNA without RT was used as a template for PCR. cyclin b was RT-PCR–amplified as a positive control. Primers were 5′-TTACTACCATCCCTGCGTGCCAT-3′ (mfng, forward for PCR), 5′-GCAGTGAACTGTCCTGAAC-3′ (mfng, reverse for PCR), 5′-AGCTCAGTGCATCACATGTTG-3′ (mfng, reverse for RT), 5′-TGGGAGACAAGACTAATATTGAGC-3′ (rfng, forward for PCR), 5′-ATCTAAAGGGCTGTGCTGGGC-3′ (rfng, reverse for RT and PCR), 5′-CGCTTCCTTCAGGATCATCCAG-3′ (cyclin b, forward for PCR), 5′-TTGGCAATATGCTGCATCACAG-3′ (cyclin b, reverse for RT and PCR).
In Situ Hybridization and Imaging
Digoxigenin RNA antisense probes were made by using Stratagene's RNA transcription kit. Whole-mount in situ hybridization was essentially as described (Oxtoby and Jowett, 1993) and either 5-bromo-4-chloro-3-indolyl phosphate/nitroblue tetrazolium (BCIP/NBT) or BM purple is used as a chromogenic substrate; two-color in situ hybridization was performed basically as previously described (Hauptmann and Gerster, 1994; Jowett and Lettice, 1994). lfng was linearized with EcoRI and transcribed with T7; mfng was linearized with HindIII and transcribed with T3; rfng was linearized with EcoRI and transcribed with T7; deltaA was linearized with EcoRI and transcribed with T7 (Haddon et al., 1998b); serrateB was linearized with Xba I and transcribed with T7 (Haddon et al., 1998a); and mariposa was linearized with BamHI and transcribed with T7. Fixed and stained embryos were mounted in glycerol under a coverslip supported at its corners by high-vacuum grease, or were dehydrated in serial methanol (from 20% to 100%) and cleared in benzyl benzoate: benzyl alcohol (2:1) solution before mounting. Images were taken with Zeiss Axioskop II or Leica MZ microscopes and edited by using Adobe Photoshop.
We thank Ying Chang for the help in constructing the similarity tree. Y.-J. J. was supported by the Agency of Science, Technology, and Research (A*STAR), Singapore; and J.L. was funded by Cancer Research UK.