Paraxial mesoderm specifies zebrafish primary motoneuron subtype identity

The vertebrate nervous system consists of many specialized cell types that form at distinct characteristic positions. We investigated how neurons acquire appropriate, position-specific fates by studying how primary motoneurons(PMNs), the first motoneurons (MNs) to form in the zebrafish spinal cord, are specified and patterned.

In zebrafish, as in other vertebrates, Hedgehog signals from the embryonic midline induce MNs in the ventral neural tube on both sides of the floor plate(Eisen, 1999; Lewis and Eisen, 2001). In all vertebrates examined so far, additional signals from paraxial mesoderm then specify distinct subpopulations of MNs that occupy specific motor columns at particular anteroposterior (AP) axial levels(Eisen, 1999; Ensini et al., 1998; Liu et al., 2001). For example, lateral motor column MNs are generated only at limb levels and visceral MNs are generated at thoracic levels.

Embryonic zebrafish have individually identifiable MNs, facilitating analysis of mechanisms that pattern neurons at the level of single cells. In addition to distinct motor columns at particular AP axial levels(Eisen, 1994), zebrafish also have a more fine-grained, segmentally reiterated pattern of different PMN subtypes along the spinal cord AP axis. Such a fine-grained, reiterated pattern of distinct MN subtypes has not yet been described in other vertebrates, probably because there are many more MNs, and individual cells cannot be recognized. Zebrafish have three different PMN subtypes: rostral primary (RoP), middle primary (MiP) and caudal primary (CaP). One PMN of each subtype forms per spinal hemisegment, with the exception that about half the hemisegments initially have two CaP-like PMNs, one of which is called variably present (VaP) and usually dies (Lewis and Eisen, 2003). Each PMN subtype is uniquely identifiable by soma position relative to overlying somites and by axon trajectory. For example,MiP and RoP somata are adjacent to overlying somite boundaries and CaP somata are adjacent to overlying somite middles (e.g. Fig. 1E,J). CaP axons project into ventral myotome, MiP axons project into dorsal myotome and RoP axons project into medial myotome (e.g. Fig. 1T,Z). Although many aspects of PMN development have been characterized (Lewis and Eisen,2003), it is still unclear how these different subtypes are specified.

PMNs can first be identified molecularly by expression of islet1. Prospective PMNs express islet1 soon after birth; at mid-somitogenesis stages CaPs initiate expression of islet2 and then within 1 hour downregulate expression of islet1. By contrast, MiPs and RoPs never express islet2(Appel et al., 1995; Inoue et al., 1994; Tokumoto et al., 1995), but can be distinguished by their temporal expression of islet1: RoPs express islet1 later than MiPs or CaPs(Appel et al., 1995). Therefore, at mid-somitogenesis stages CaPs can be identified by islet2 expression and MiPs by islet1 expression (see Fig. 1); RoPs do not yet express islet1.

The tight spatial correlation between the reiterated pattern of PMNs and the overlying somites suggests that signals from paraxial mesoderm might specify different PMN subtypes. Consistent with this idea, transplantation experiments have shown that environmental signals can specify zebrafish PMN subtypes (Appel et al., 1995; Eisen, 1991). For example,when MiP is transplanted 2-3 hours before axogenesis to the position where CaP normally develops, the transplanted cell forms a CaP-like axon and initiates expression of islet2. However, when MiP is transplanted in the same way just 1 hour before axogenesis, it remains committed to its original fate,extends a MiP-like axon and does not express islet2(Appel et al., 1995; Eisen, 1991). Furthermore, PMN specification and development is disturbed in spadetail(spt) mutants, which have a dramatic reduction of trunk paraxial mesoderm (Bisgrove et al.,1997; Eisen and Pike,1991; Inoue et al.,1994; Tokumoto et al.,1995); when somite segmentation is disturbed by heat shock, the position and axonal morphology of PMNs are also disturbed(Kimmel et al., 1988; Roy et al., 1999).

Together these observations suggest that signals from paraxial mesoderm may specify PMN subtypes. To test this hypothesis, we investigated PMN subtype specification in several zebrafish mutants that affect paraxial mesoderm development in different ways. We concentrated on MiP and CaP specification because these PMNs can be identified both molecularly and by axon trajectory. Our findings demonstrate that signals from paraxial mesoderm are required to specify MiPs and CaPs, and they suggest that additional signals from the somites are also required to fine-tune or maintain correct spatial organization of PMN subtypes.

Zebrafish (Danio rerio) embryos were obtained from natural spawnings of wild types (AB) or crosses of identified carriers heterozygous for specific mutations. Fish were maintained in the University of Oregon Zebrafish Facility on a 14 hour light/10 hour dark cycle at 28.5°C and embryos staged according to Kimmel (Kimmel, 1995) by number of somites or hours post fertilization at 28.5°C (hpf). Production of parental fish heterozygous for mutations at two different loci was carried out as in Lewis and Eisen (Lewis and Eisen,2001). Mutant embryos were identified by morphology, except for fss;yot mutants younger than 24 hpf. At these stages the fssphenotype masks the yot phenotype. However, yot mutants lack lateral floor plate (Odenthal et al.,2000), so young fss;yot mutants were identified by fss morphology and lack of expression of a lateral floor plate marker nkx2.2b (kindly provided by M. Schäfer and C. Winkler prior to publication).

Df(LG12)dlx3b^b380 (hereafter referred to as Df^b380) is a deficiency on linkage group 12 that removes 21-24 cM that includes dlx3b, dlx4b, sox9a, irbp, rbp4 and dkk1 (Fritz et al.,1996; Liu et al.,2003). Df(LG05)her1^b567(hereafter referred to as Df^b567) is a deficiency on linkage group 5 that removes up to 22 cM that includes her1, her7,ndr3 and a number of ESTs (Henry et al., 2002). Other lines used were: after eight^tr233 (aei); fused-somites^te314 (fss); you-too^ty17 (yot)(van Eeden et al., 1996); no tail^b195 (ntl)(Halpern et al., 1993); spadetail^b104 (spt)(Ho and Kane, 1990); trilobite^m209 (tri)(Hammerschmidt et al., 1996); knypek^b639 (kny)(Topczewski et al., 2001); cyclops^b16 (cyc)(Hatta et al., 1991); and floating headⁿ¹ (flh)(Talbot et al., 1995).

All analyses were carried out on PMNs in the trunk except that PMNs in the tail were analysed where noted in the text. Whenever possible, we assayed CaP and MiP identity using both axon trajectory (znp1 antibody staining) at 26-30 hpf and gene expression (islet1 and islet2) at 17-19 hpf,when MiPs, which are located directly under somite boundaries express islet1 (Fig. 1E) and CaPs, which are located under somite middles express islet2(Fig. 1J). In all cases we examined at least seven embryos in detail using a compound microscope, and in several cases we analysed both whole mounts and serial cross-sections. islet1 + islet2 double in situ RNA hybridization did not provide strong enough signals to assess gene expression unequivocally. Therefore, to determine the distribution of islet1-expressing and islet2-expressing PMNs, we conducted in situ RNA hybridization for each gene alone as well as Islet antibody staining followed by islet2in situ RNA hybridization (Fig. 1). The Islet antibody recognizes both Islet1 and Islet2. Therefore, PMNs recognized by both Islet antibody and islet2riboprobe express islet2 and hence are CaPs, whereas PMNs recognized by Islet antibody alone only express Islet1 and are MiPs(Fig. 1O). In some mutants most or all PMNs expressed islet2; comparison with islet1 in situ RNA hybridization suggested that many of them also expressed islet1. Therefore, these PMNs had a hybrid identity. In these cases we examined how many PMNs expressed islet1 by Islet antibody staining followed by islet1 in situ RNA hybridization; PMNs labeled by antibody alone express only Islet2, whereas PMNs labeled by antibody and islet1riboprobe express islet1 (Fig. 1Y). We were able to identify cells that expressed only islet1, only islet2, or both islet1 and islet2 unequivocally using this combination of markers.

In situ RNA hybridization was performed as previously described(Concordet et al., 1996). her1 probe was synthesized as described by Müller (Müller,1996), cs131 probe was synthesized as described by Durbin et al.(Durbin et al., 2000), and islet1 and islet2 probes were synthesized as described by Appel et al. (Appel et al.,1995). Islet antibodies originally isolated by the Jessell Laboratory were obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by the University of Iowa, Department of Biological Sciences, Iowa City, IA 52242. Antibody 39.4D5 was used alone at a final concentration of 1/200 or a 1:1 mixture of 39.4D5 and 40.2D6 was used; in the latter case, both antibodies were used at a final concentration of 1/300. In cases in which Islet antibody staining and islet1 or islet2 in situ RNA hybridization were performed on the same embryos, antibody staining was carried out first. Embryos were fixed for 5-6 hours at 4°C in 4% PFA in PBS, permeabilized by washing with PBS +0.5% Triton for several hours and distilled H₂O for 30 minutes and blocked in a serum-free block solution [2% BSA; 1×PBS; 5-10% DMSO; 0.2%Triton plus RNAse inhibitor (20-40 units/ml)] for 1 hour. Embryos were then incubated with Islet antibody in fresh serum-free block overnight at 4°C. Antibody staining was developed using the Sternberger Clonal PAP system and detected using a Vector Laboratories DAB kit. The same serum-free block was used for secondary and tertiary antibodies. Embryos were then fixed for 15-20 minutes in 4% PFA in PBS and processed for in situ RNA hybridization.

Axon trajectories were assessed as described by Eisen et al.(Eisen et al., 1989) using znp1 monoclonal antibody (Trevarrow et al., 1990), the Sternberger Clonal PAP system and Vector Laboratories DAB kit. In all cases, CaP axons were clearly visible both in whole mount and in cross-section. In some mutants, MiP axons were easier to identify in cross-section.

Specimens were analysed using a Zeiss Axioplan microscope and photographed with Kodak Ektachrome 64T or 164T film. Images were scanned on a Nikon LS-1000 35mm film scanner and processed using Adobe Photoshop software.

Morpholino antisense oligonucleotides (MOs) were obtained from Gene Tools. About 5 nl of a MO mix (ntl MO 1 mg/ml; spt MO #1 0.75 mg/ml; spt MO #2 0.075 mg/ml) was injected into one- to two-cell wild-type embryos. This concentration reliably phenocopied ntl;sptmutants. The MO sequences were: ntl MO, GACTTGAGGCAGGCATATTTCCGAT(see also Nasevicius and Ekker,2000); spt MO #1, AGCCTGCATTATTTAGCCTTCTCTA; and spt MO #2, GATGTCCTCTAAAAGAAAATGTCAG.

For all blastula stage transplants and some somite transplants we used ntl;spt MO-injected embryos as hosts. We confirmed that MO-injected embryos phenocopy ntl;spt mutants by examining their morphology, islet1 and islet2 expression patterns and tropomyosin expression (to confirm absence of somitic mesoderm).

Wild-type donor embryos were injected with a mixture of 2.5% 3 kDa fluorescein dextran and 2.5% 3 kDa rhodamine dextran in 0.2M KCl at the one-to two-cell stage. Host embryos were injected with a mixture of ntland spt MOs at the one- to two-cell stage. Fluorescently labeled,wild-type cells were transplanted into the margin of ntl;sptMO-injected embryos between blastula and 30% epiboly stages(Fig. 6A); embryos were analysed and fixed at 18-22 somites.

We transplanted fluorescently labeled whole somites from wild-type donors at the 7- to 10-somite stage into similarly staged ntl;spt mutant or ntl;spt MO-injected hosts (Fig. 6E); results were identical in both cases. These experiments need to be done early enough that PMN subtype identities are still labile(Appel et al., 1995; Eisen, 1991); this was possible because ntl;spt mutants completely lack somitic mesoderm(Amacher et al., 2002) and therefore we did not need to remove somites from host embryos. Fluorescently labeled wild-type donor somites were prepared as described previously(Beattie and Eisen, 1997). Donors were dissociated and somites, recognized by their characteristic morphology, were removed to a separate culture dish. ntl;spt mutants or ntl;spt MO-injected hosts were mounted in agar with their spinal cords facing up. Several wild-type somites were inserted into the trunk using a wide-bore micropipette (Fig. 6E). Host embryos were cultured at 28.5°C for about 4 hours in L15 medium with 50 units/ml of penicillin and 0.05 mg/ml streptomyocin before being fixed for analysis. PMN identity was assayed using Islet antibody and islet2 riboprobe. In some cases, we were able to see PMN labeling through the transplanted somites (see Fig. 6E). However, in many cases we had to remove the somites to assay PMN identity because the Fast Red stain we used to recognize donor somites obscured PMN labeling.

Individual PMNs were labeled in live embryos(Eisen et al., 1989).

If PMN subtypes are specified by signals from paraxial mesoderm, altering the relationship between somites and PMNs should change PMN subtype identities. trilobite;knypek (tri;kny) mutants have defects in convergence and extension that result in very narrow somites that have normal AP patterning but are only about two cells wide along the AP axis(Table 1); wild-type somites are about five cells wide (Henry et al.,2000). If PMNs form in these mutants, all of them will be next to both a somite boundary and a somite middle. Therefore, we reasoned that if our hypothesis that localized signals from overlying somites specify PMN subtypes is correct, all PMNs in these mutants should be exposed to the same signals and therefore have the same subtype identity.

MiPs and CaPs form normally in tri and kny single mutants that have somites that are three or four cells wide, which is intermediate between wild types and tri;kny double mutants. However, PMNs are often slightly closer together in these single mutants, corresponding to the slightly narrower somites (Fig. 1C,D,H,I,M,N,R,S; Table 2). By contrast, tri;kny mutants have continuous stretches and clumps of PMNs, possibly because of their more severe defects in convergence and extension (Fig. 1A,F,K,U). Most PMNs in double mutants express both islet1 and islet2 and hence have a hybrid identity, at least in terms of gene expression (Fig. 1A,F,K,U; Table 2). None of the PMNs in tri;kny mutants expresses only islet2(n=8; Fig. 1U), but in some double mutants there are a few PMNs that apparently express islet1 alone (<10% of the PMNs in any one embryo; 4/16 embryos had no islet1 only cells). However, these could be later-forming secondary motoneurons (SMNs) that are just initiating islet1expression (Appel et al.,1995).

tri;kny mutants have many ventrally projecting CaP-like axons,although these axons have some aberrant branches. By contrast, dorsally projecting MiP-like axons are very rare and can be best seen in cross-section(Fig. 1P,X). In seven double mutants analysed in cross-section, we saw 122 CaP axons, four fairly normal MiP axons (e.g. Fig. 1X) and four shorter dorsal axons that were less convincingly MiPs. This suggests that nearly all PMN axons in tri;kny mutants are CaP like. We confirmed this result by dye-labeling PMNs in 14 live embryos. In contrast to wild-type embryos in which PMNs are readily identified by soma position(Eisen et al., 1989), the unusual morphology of tri;kny double mutant embryos makes it difficult to identify PMNs unambiguously using these criteria. Therefore, we labeled essentially every cell of the right size in the ventral spinal cord. We saw 36 CaPs, 10 PMNs with short CaP-like axons and two MiPs (both in the same embryo).

Because CaPs normally have turned off islet1 expression by the time they extend axons, our results suggest that in tri;kny mutants most PMNs have a hybrid identity with respect to gene expression but a CaP-like identity based on axon trajectory. This is consistent with our hypothesis that signals from somites specify MiPs and CaPs. The simplest interpretation of our results is that in wild types there are two spatially separated signals: one that specifies MiPs and one that specifies CaPs,whereas in tri;kny mutants the narrow somites cause these signals to overlap so all PMNs are exposed to both signals. However, it is also possible that these mutations affect somites and PMNs independently. Therefore, to test further the hypothesis that signals from somites specify MiPs and CaPs, we examined PMN subtype specification in mutants that lack proper somite segmentation.

Several zebrafish mutations disturb somite segmentation and block formation of at least some somite boundaries (Fritz et al., 1996; Henry et al.,2002; Liu et al.,2003; van Eeden et al., 1996b). In all of these mutants examined so far, genes that are normally AP restricted within individual somites are either missing or ubiquitously expressed within the somitic mesoderm(Durbin et al., 2000; Henry et al., 2002; Holley et al., 2000; Jiang et al., 2000; van Eeden et al., 1996). We reasoned that if localized signals emanating from somites specify PMN subtypes, these signals might be missing or mislocalized in these mutants. Therefore, we examined MiP and CaP specification in several mutants that affect somite segmentation.

fused somites (fss) and Df^b380mutants lack all somite boundaries (Liu et al., 2003; van Eeden et al.,1996) (Table 1). after eight (aei) mutants resemble fss mutants in the posterior trunk, but the first eight or so somites form normally and aei is thought to be involved in a different step in somite formation than fss (Durbin et al.,2000; Holley et al.,2000; Jiang et al.,2000) (Table1). Df^b567 mutants form somites with irregular widths and boundary defects (Henry et al.,2002) (Table 1). Despite their defects in somite boundary formation, fss,Df^b380, aei and Df^b567 mutants all form myotome boundaries at later stages, although these are irregular(Henry et al., 2002; van Eeden et al., 1998)(K.E.L. and J.S.E., unpublished). We therefore also analysed fused somites;you-too (fss;yot) mutants because they lack even this later morphological segmentation of somitic mesoderm(van Eeden et al., 1998)(Table 1).

During our study the fss locus was cloned (see Table 1) and mapped to the same linkage group as Df^b380(Nikaido et al., 2002). We performed complementation analysis and found that Df^b380and fss mutations do not complement, suggesting the Df^b380 deletion uncovers at least part of the fssgene or its regulatory sequences and that at least part of the somite phenotype of Df^b380 is due to loss of fssfunction. However, as we describe below, presomitic mesoderm expression of at least one gene differs between Df^b380 and fssmutants, suggesting that the somite defect in Df^b380mutants is more severe than in fss mutants.

We analysed PMN subtype specification in these different mutants and found that in Df^b567, Df^b380 and fss;yot mutants islet2-expressing and islet1-expressing PMNs still form in normal numbers. In addition,both CaP and MiP axons are present, although they often have some aberrant branches, and as in yot single mutants, fss;yot mutants have fewer PMN axons than wild types (Fig. 2) (van Eeden et al.,1996). However, the precise spacing and alternation of MiPs and CaPs is disturbed in all of these mutants, although the severity of this phenotype varies among embryos and even sometimes between the two sides of the same embryo. In most cases the alternation of MiPs and CaPs is less regular than in wild types, the spacing between PMNs varies and PMNs on the two sides of an embryo are out of register (Fig. 2).

In contrast to the mutants described above, aei mutants have a slight `neurogenic' phenotype in which excess PMNs are produced, resulting in small clusters of islet2-expressing and islet1-expressing PMNs instead of individual cells (see also Holley et al., 2000). Nevertheless, at 18-20 hpf there is a regular alternation and spacing of islet1-expressing and islet2-expressing PMNs in aeimutants with islet2-expressing PMNs forming adjacent to somite middles and islet1-expressing PMNs forming adjacent to somite boundaries. This is consistent with observations in other Delta/Notch pathway mutants and embryos expressing a dominant negative Delta construct(Appel and Eisen, 1998; Appel et al., 2001; Gray et al., 2001). The somite defect in aei mutants is specific to posterior somites, so we also examined tails of 24 hpf aei mutants. Even at this stage, the alternation and spacing of PMNs resembles wild types of a similar stage(compare Fig. 2D with 2E). We also see both CaP and MiP axons in aei mutants, although there is aberrant branching posterior to somite 8(Fig. 2K,L)(van Eeden et al., 1996).

One interpretation of these results is that segmentation of somitic mesoderm is required for fine-tuning or maintaining the spacing and precise alternation of MiPs and CaPs, but is unnecessary for specifying PMN subtypes. However, it is also possible that all of these mutants have some remaining cryptic or early segmentation that is sufficient to specify MiPs and CaPs. Consistent with the latter possibility cs131, which encodes a cell cycle arrest protein, is segmentally expressed in presomitic mesoderm of both aei and fss mutants(Durbin et al., 2000), and her1 is segmentally expressed in presomitic mesoderm of fssmutants, although it is not segmentally expressed in aei mutants(Durbin et al., 2000; Holley et al., 2000; van Eeden et al., 1998). We therefore examined expression of cs131 in Df^b567, Df^b380 and fss;yot mutants, and expression of her1 in Df^b380 and fss;yot mutants. In all of these mutants at least one gene is segmentally expressed in presomitic mesoderm. Df^b380 and fss;yot mutants have segmental expression of her1 in presomitic mesoderm(Fig. 3F,G), and Df^b567 and fss;yot mutants have segmental expression of cs131 in presomitic mesoderm(Fig. 3B,D). However, compared with wild types in which cs131 is also expressed in the posterior of each somite, cs131 expression is weak and unlocalized in somitic mesoderm of these mutants (Fig. 3B-D). By contrast, Df^b380 mutants lack the presomitic mesoderm stripe of cs131, although they still have weak,unlocalized expression of cs131 in somitic mesoderm(Fig. 3C). This difference in cs131 expression in presomitic mesoderm of Df^b380and fss;yot mutants suggests that the somite phenotype of Df^b380 mutants is more severe than that of fss or fss;yot mutants. However, even in Df^b380 mutants, there is still some early molecular segmentation of paraxial mesoderm as evidenced by segmental expression of her1.

These results suggest that in addition to signals that specify MiP and CaP subtype identities, there are also signals that fine-tune or maintain correct spatial organization of PMN subtypes and that these signals are missing from somite segmentation mutants. Surprisingly, even though these mutants lack proper segmentation of the somitic mesoderm, CaPs and MiPs are still specified, suggesting that neither somite boundaries, nor AP-restricted gene expression within individual somites are required for PMN subtype specification. However, at least some segmental gene expression remains in presomitic mesoderm in all somite segmentation mutants we examined. This early paraxial mesoderm segmentation may be sufficient to specify MiPs and CaPs, but insufficient for correct spatial organization of these PMN subtypes. Therefore, as no mutations have been described that lack all aspects of paraxial mesoderm segmentation, we examined PMN subtype specification in mutants that lack all paraxial mesoderm.

If signals from either presomitic or somitic mesoderm normally specify PMN subtypes, PMNs should be misspecified in mutants lacking these signals. spadetail (spt) mutants have a severe reduction of trunk paraxial mesoderm (both somitic and presomitic) and no tail;spadetail(ntl;spt) mutants completely lack paraxial mesoderm(Amacher et al., 2002; Ho and Kane, 1990)(Table 1). Thus, any signals from presomitic or somitic mesoderm should be reduced in spt mutants and completely lacking in ntl;spt mutants.

In ntl;spt mutants, the vast majority of PMNs express both islet1 and islet2 and thus have a hybrid identity with respect to gene expression (Fig. 4A-D; Table 2). An occasional cell expresses just islet1(Table 2). These cells may be SMNs that are just initiating islet1 RNA expression or interneurons that were mistakenly counted as motoneurons because of the misshapen axis in these embryos. By contrast, no PMNs express only islet2 (0% of PMNs, n=251; Fig. 4D). spt mutants have a similar but less severe phenotype(Fig. 4E-H; Table 2). Most PMNs express both islet1 and islet2 but there are more islet1-only expressing cells than in ntl;spt mutants(Table 2). In addition, very occasionally in spt mutants (1.4% of PMNs, n=142) a PMN expresses only islet2. We could not analyse the axon trajectories of PMNs in ntl;spt mutants, because in the absence of somitic tissue PMN axons do not exit the spinal cord (see also Eisen and Pike, 1991).

In addition to their lack of paraxial mesoderm, ntl;spt mutants also lack both notochord and floorplate. Therefore, to check whether these midline structures are required for correct specification of MiPs and CaPs we examined PMN subtype specification in cyclops;floating head(cyc;flh) mutants that lack both notochord and floorplate(Halpern et al., 1997), and ntl single mutants that lack notochord(Halpern et al., 1993); all of these mutants have somitic and presomitic mesoderm. We previously showed that cyc;flh mutants have fewer islet1-expressing and islet2-expressing PMNs due to reduced levels of Hedgehog signalling(Lewis and Eisen, 2001) but we did not assess whether these PMNs have a hybrid identity. In our current study, we found that with respect to both gene expression and axon trajectory,specification of MiPs and CaPs is normal in cyc;flh and ntlmutants although the spatial organization of these PMNs is sometimes slightly perturbed (Fig. 5).

All of these results are consistent with the hypothesis that signals from paraxial mesoderm specify MiP and CaP subtype identities. They also suggest that in the absence of these signals, PMNs assume a hybrid identity, at least with respect to islet gene expression.

All of the mutants we examined that have a severe somite phenotype provide strong support for our hypothesis that signals from paraxial mesoderm pattern PMN subtype identities. However, it is still formally possible that mutations in these genes could affect PMNs and somites independently. We tested this possibility by using transplantation methods to create genetically mosaic embryos to ask whether normal PMN subtype identity requires ntl and spt function in the CNS or in somites. First, we addressed whether wild-type CNS restored normal subtype identities to ntl;spt mutant PMNs. We transplanted large numbers of fluorescently labeled, wild-type donor cells into ntl;spt MO-injected host embryos at blastula stage(Fig. 6A) and analysed PMN gene expression in eight embryos that lacked somites but in which most trunk spinal cord was derived from wild-type donor cells. In every case, all of the PMNs expressed islet2; thus they remained misspecified(Fig. 6B-D). We further analysed four of these embryos in cross-section and confirmed that all PMNs that developed from wild-type donor cells (n>40) expressed islet2 (Fig. 6C,D). These results suggest that wild-type spinal cord cells are insufficient for correct PMN subtype specification in the absence of paraxial mesoderm,consistent with our hypothesis that MiP and CaP subtypes are specified by signals extrinsic to the spinal cord.

Next we addressed whether wild-type somites could restore normal PMN subtype specification in ntl;spt mutants. We transplanted fluorescently labeled whole somites from wild-type donor embryos at the 7-10 somite stage into similarly staged ntl;spt mutant or ntl;sptMO-injected hosts (Fig. 6E). We carried out this experiment on three separate occasions and each time we restored PMN subtype specification in the region of the somite transplant in at least one embryo. We divided our results into two categories: restoration over two segments (two groups of islet1-expressing cells separated by islet2-expressing cells) and restoration over more than two segments(restoration of more than two groups of islet1-expressing cells separated by islet2-expressing cells; Fig. 6F; Table 3). Some transplanted embryos had no restoration of PMN subtype specification. There are a number of technical reasons why this may have been the case: we may have damaged the somites during transplantation, inserted the somites too far from the neural tube, or these host embryos may have been older and PMN fates less labile(Eisen, 1991; Appel et al., 1995). We also processed 28 control embryos: two with transplants in the head, two in which the transplanted somites fell off and 24 ntl;spt mutant or MO-injected embryos without transplants processed in parallel to the transplanted embryos.

In a significant number of our transplants, but not in our controls, we restored an alternating pattern of islet1-expressing and islet2-expressing PMNs in the spinal cord region adjacent to the transplanted wild-type somites (Fig. 6F; Table 3). In cases in which we could see somite boundaries, islet1-expressing PMNs were located adjacent to these boundaries, whereas islet2-expressing PMNs were located between them (Fig. 6E). These results show that lack of paraxial mesoderm in ntl;spt mutants causes mis-specification of PMN subtypes, therefore specification of MiPs and CaPs requires signals from paraxial mesoderm.

We provide the first analysis of how a segmentally reiterated pattern of neurons is specified along the AP axis of the vertebrate spinal cord by analysing how CaP and MiP motoneuron subtypes are specified and spatially organized in embryonic zebrafish. These different PMN subtypes occupy different locations within the ventral spinal cord relative to overlying somites, express different genes and innervate different muscle territories(Lewis and Eisen, 2003). We show that signals from paraxial mesoderm specify this reiterated, segmental pattern of zebrafish PMN subtypes. In the absence of these signals, PMNs express both islet1 and islet2, and hence have a hybrid identity with respect to gene expression; under these conditions, the CaP axon trajectory appears dominant. Our results also suggest that there is a distinct mechanism for ensuring correct spatial organization of PMN subtypes that requires at least some aspects of normal somite segmentation.

Our evidence that signals from paraxial mesoderm are required for correct specification of MiPs and CaPs is threefold. First, our analysis of tri;kny mutants shows a correlation between very narrow somites and misspecified PMNs. Second, our analysis of ntl;spt and sptmutants shows a correlation between loss of paraxial mesoderm (presomitic mesoderm and somites) and mis-specification of PMNs. Third, our transplantation experiments demonstrate that mis-specification of PMN subtypes in ntl;spt mutants is caused by lack of paraxial mesoderm.

Our analysis of PMNs in spt mutants is consistent with previously published data. Inoue et al. (Inoue et al., 1994) reported an increase in islet1-expressing MNs in spt mutants at 24 hpf and Bisgrove et al.(Bisgrove et al., 1997)reported an increase in islet2-expressing PMNs at 18 hpf. Both of these observations are consistent with our findings that most PMNs in spt mutants express both islet1 and islet2 at 18-20 hpf. However, Tokumoto et al. (Tokumoto et al., 1995) reported a reduction in the number of islet2-expressing MNs in spt mutants slightly later, at 24 hpf. This is surprising, given our results and those of Bisgrove et al.(Bisgrove et al., 1997). However, it is possible that SMNs, some of which also express islet2by 24 hpf, are reduced or delayed, or that some PMNs are dying, resulting in a reduction in islet2-expressing MNs at this later stage.

tri;kny mutants form PMNs with a hybrid identity as assayed by islet gene expression. The simplest interpretation of this result is that there are two signals from the paraxial mesoderm, one that induces or maintains islet1 expression in MiPs and one that induces islet2 expression in CaPs. In wild-type embryos, these signals are spatially distinct so that each PMN experiences only one of them. By contrast,in tri;kny mutants these signals are so close together that they overlap and all PMNs experience both signals and respond by expressing both islet genes. The idea of inducing signals is supported by experiments showing that individual MiPs transplanted to the CaP position turn on expression of islet2. These experiments suggest that localized signals normally induce islet2 expression in CaPs(Appel et al., 1995; Eisen, 1991), and also demonstrate that PMNs that do not normally experience this islet2-inducing signal still have the ability to respond to it.

In the light of these results, it was surprising to find that in ntl;spt mutants that lack all paraxial mesoderm-derived signals, PMNs also express both islet1 and islet2. The simplest interpretation of this result is that there are two signals from paraxial mesoderm: one that normally represses islet2 expression in MiPs and one that normally represses islet1 expression in CaPs. This simple model, with two repressive signals, is inconsistent with the simple model suggested by the tri;kny double mutant and PMN transplantation results, which postulated two inducing signals. This suggests to us that the signalling that specifies CaP and MiP subtypes may involve both repressive and inductive signals, and hence be more complicated than either of these simple models. Thus, although our results demonstrate that signals from paraxial mesoderm are required for PMN subtype specification, it is still unclear exactly how these signals act. Further studies will be needed to identify these signals and the mechanisms by which they specify distinct PMN subtypes.

In tri;kny mutants most PMNs express both islet1 and islet2 but have a CaP axon trajectory, suggesting that CaP identity is dominant over MiP identity. tri;kny mutants do have rare PMNs with a MiP-like axon trajectory that probably correspond to the occasional PMNs that express only islet1. However, an intriguing possibility that remains to be tested is that, as suggested from studies in mouse,over-occupation of a particular axon pathway may cause an occasional axon to be `shunted' to an alternative target(Sharma et al., 2000).

We offer three possible interpretations of why PMNs in tri;knymutants have CaP-like axons despite their hybrid subtype identity as indicated by gene expression. First, somite-derived signals necessary for MiP axon pathfinding may be lacking in tri;kny mutants. This seems unlikely as the occasional MiP-like axon still forms, and all of the genes examined so far are expressed normally in the somites of these mutants(Henry et al., 2000) (K.E.L. and J.S.E., unpublished). Second, CaP-specifying signals may be dominant over MiP-specifying signals and even though PMNs in tri;kny mutants express both islet1 and islet2, they may otherwise molecularly resemble CaPs more than MiPs. This possibility can be assessed when additional molecular markers for PMN subtypes are identified. A third related possibility is that expression of islet2 may specify a CaP-like axon trajectory in PMNs, irrespective of other islet genes the cell expresses. We favor this possibility, because it is consistent with studies showing that reducing Islet2 function changes CaPs into VeLD spinal interneurons (Segawa et al.,2001).

To our initial surprise, both MiPs and CaPs were specified in normal numbers in mutants with disturbed somite boundaries and AP somite patterning. This suggests that neither of these aspects of paraxial mesoderm segmentation are required to specify different PMN subtypes, although we cannot rule out the possibility that there are as yet unidentified genes expressed segmentally in the somites of these mutants. Interestingly, although MiPs and CaPs formed in all of these mutants, in most cases their spatial organization was disturbed. The precise alternation and spacing of different PMN subtypes was lost and PMNs on the two sides of an embryo were out of register. This shows that PMN subtype specification and PMN spatial organization are separable aspects of PMN patterning and it suggests that these somite segmentation mutants are missing signals that normally fine-tune or maintain the precise spatial organization of different PMN subtypes. One mechanism by which these signals may act is by controlling the cell adhesion properties of particular PMNs and/or neighboring cells. In this model the normal, precise, alternating pattern of MiPs and CaPs would be fine-tuned or maintained by cell adhesion and this mechanism would be disturbed or lacking in the somite segmentation mutants. This would explain why transplanted PMNs sometimes migrate back to their original spinal cord positions relative to overlying somites(Eisen, 1991). In addition, if these cell adhesion properties develop after CaPs are first specified, it would explain why the initial spatial organization of CaPs is not as regular as at later stages (Appel et al.,1995).

When we started our analyses, one large class of genes that were obvious candidates for specifying PMN subtype identities were genes that are normally expressed in an AP-restricted pattern in somites. However, all of these genes that have been examined so far are either not expressed, or are mislocalized in somite segmentation mutants (Durbin et al., 2000; Henry et al.,2002; Holley et al.,2000; Jiang et al.,2000; van Eeden et al.,1996). Yet both MiPs and CaPs form in normal numbers in these mutants. This strongly suggests that none of these genes are required for specifying PMN subtypes.

The signals that normally specify MiPs and CaPs might emanate from presomitic mesoderm. Consistent with this possibility, although the somite segmentation mutants we examined affect different aspects of paraxial mesoderm segmentation, in every case at least one gene is still segmentally expressed in presomitic mesoderm. If the signals that specify PMN subtypes come from presomitic mesoderm, MiPs and CaPs would be specified before we can currently identify them molecularly. However, if PMN subtypes are specified this early,additional, somite-derived signals would be required to explain how PMNs transplanted at mid-somitogenesis stages can adopt a new, position-specific PMN subtype identity (Appel et al.,1995; Eisen, 1991)and transplanted somites can restore specification of PMN subtype identities in ntl;spt mutants. These later signals might normally only fine-tune the spatial organization of MiPs and CaPs, in which case they are presumably missing in the somite segmentation mutants.

An alternative possibility is that signals that specify MiPs and CaPs emanate from the somites. In this case, there must be some cryptic aspect of segmentation that persists in the somitic mesoderm in somite segmentation mutants that is sufficient to specify MiPs and CaPs. Consistent with this possibility, vertebrae form with almost normal periodicity in these mutants,and most of the mutants form irregular myotome boundaries at later developmental stages (van Eeden et al.,1996; van Eeden et al.,1998). Distinguishing between these two alternatives will require the identification of additional genes involved in segmentation and in PMN subtype specification.

FGFs are another class of signals that might be candidates for specifying PMN subtype identities. FGF signalling is crucial for somite segmentation(Dubrulle and Pourquie, 2002)and correct timing of neural differentiation(Diez del Corral et al., 2002),and has also been implicated in AP patterning of MNs in chick(Liu et al., 2001). In zebrafish, fgf8 is expressed in the posterior presomitic mesoderm and in the anterior region of newly formed somites(Reifers et al., 1998). fgf17 is also expressed at the anterior margin of somites after about the eight-somite stage (Reifers et al.,2000). However, our preliminary data show that, although somite boundaries are variably disturbed in embryos with reduced FGF8 signalling,MiPs and CaPs are specified normally in ace (fgf8) mutants,embryos injected with fgf8 MOs and ace mutants injected with an fgf17 MO. However, in some of these embryos, PMN spacing is irregular and PMNs on the two sides of an embryo are out of register in a manner reminiscent of somite segmentation mutants (K.E.L. and J.S.E.,unpublished). Our treatments have probably not entirely abolished FGF signalling. Thus, we have not ruled out the possibility that FGFs participate in PMN subtype specification, although this may be difficult to assess because embryos become highly necrotic when fgf8 levels are severely reduced(Draper et al., 2001) (K.E.L. and J.S.E., unpublished).

In conclusion, our data provide strong evidence that signals from paraxial mesoderm are required to specify and spatially organise distinct PMN subtypes in a segmentally reiterated pattern along the AP axis. It will be exciting in the future to learn the nature of the signals involved in these processes and when and how they act during PMN subtype specification.

We thank Michael Brand, Sharon Amacher, Bruce Draper and Jon Muyskens for sharing MOs prior to publication; Roger Albertson for help with some initial analysis of Df^b567; Sharon Amacher for sharing this line with us prior to publication; Christoph Winkler for sharing the nkx2.2b probe prior to publication; the staff of the UO Histology Facility for help with sectioning; and Alan Arnett, Martha Jones, Ellie Melançon, Chapell Miller, Mary Swartz and the staff of the UO Zebrafish Facility for fish husbandry. We also thank Bruce Appel, Estelle Hirsinger,Julie Kuhlman, Lisa Maves and Monte Westerfield for comments on previous versions of this manuscript. Supported by NIH grants NS23915 and HD22486 and Wellcome International Prize Travelling Research Fellowship 054975 to K.E.L.