The control of rostrocaudal pattern in the developing spinal cord: specification of motor neuron subtype identity is initiated by signals from paraxial mesoderm

Vertebrate motor behavior is coordinated by the activity of motor neurons (MNs) located in the spinal cord and brain stem. The refinement of motor function depends, in part, on the existence of subclasses of MNs which participate in distinct circuits in the central nervous system and innervate different target cells in the periphery. Some insight into the molecular events involved in establishing the generic identity of MNs has been obtained (Tanabe and Jessell, 1996). In particular, the differentiation of MNs has been shown to depend on a Sonic Hedgehog-mediated inductive signal transmitted from axial mesodermal cells of the notochord to overlying neural plate cells (Roelink et al., 1995; Tanabe et al., 1995; Marti et al., 1995; Chiang et al., 1996; Ericson et al., 1996). How developing MNs acquire their distinct subtype identities remains poorly understood.

The diversification of MN subtypes first becomes apparent through differences in the settling pattern of MNs in the spinal cord and by the distinct pathways selected by motor axons in the periphery (Landmesser, 1980). One major subdivision of spinal MNs is based upon their segregation into discrete columns. In the chick embryo, four major columnar subtypes of somatic MNs are evident (Landmesser, 1978; 1980; Hollyday, 1980; Gutman et al., 1993). MNs in the medial subdivision of the median motor column (MMC) project axons to axial muscles; MNs in the lateral MMC project axons to body wall muscles, and MNs in the medial and lateral subdivisions of the lateral motor column (LMC) project axons respectively to ventrally and dorsally derived limb muscles. In addition, a class of visceral MNs located in the Column of Terni (CT) innervates sympathetic neurons (Levi-Montalcini, 1950; Prasad and Hollyday, 1991). With the exception of the medial MMC, each MN columnar subtype is generated in a discontinuous manner along the rostrocaudal axis of the spinal cord (Fig. 1A). Somatic MNs within individual columns can be further subdivided into pools on the basis of their specific muscle targets (Landmesser, 1978; Hollyday, 1980).

The analysis of motor axon projection patterns after manipulation of the chick embryo has suggested that the columnar and pool identities of somatic MNs are acquired prior to target innervation and enable motor axons to respond appropriately to guidance cues encountered in the periphery (Tosney, 1991; Landmesser, 1992). Molecular evidence for the early acquisition of MN subtype identity has been provided by the demonstration that the expression of LIM homeodomain (HD) proteins defines each of the columnar subclasses of chick MNs, prior to target innervation (Tsuchida et al., 1994). Moreover, genetic analyses have shown that LIM-HD proteins control neuronal differentiation and axonal pathfinding in both vertebrates and invertebrates (Lundgren et al., 1995; Pfaff, et al., 1996; Thor and Thomas, 1997; Hobert et al., 1997) suggesting that they have a role in specifying the columnar subtype identity of spinal MNs.

The source and nature of inductive signals that control the generation of spinal MN subtypes have not been defined. Early rostrocaudal inversion of chick lumbosacral neural tube results in a respecification of the pool identity of MNs, as assessed by the emergent pattern of muscle innervation (Matise and Lance-Jones, 1996). Furthermore, the intrasegmental displacement of primary MNs along the rostrocaudal axis of zebrafish embryos changes the initial pattern of motor axon projections (Eisen, 1991) and LIM homeobox gene expression (Appel et al., 1995). Thus, the differentiation of certain MN subtypes appears to be dependent on the rostrocaudal position of neural tube cells. Since notochord inductive activity and Shh signaling appear to be constant along the rostrocaudal axis (Yamada et al., 1991; Fukushima et al., 1996; Ericson et al., 1997), axial mesodermal signals are unlikely to control the formation of distinct MN columnar subtypes at different segmental levels of the spinal cord.

In the embryonic brain there is evidence that signals provided by the paraxial mesoderm can impose rostrocaudal identity on neural cells. An activity expressed transiently by paraxial mesoderm can confer an early caudal character on cells derived from prosencephalic levels of the chick neural plate (Muhr et al., 1997) and similar activities have been detected in zebrafish (Woo and Fraser, 1997) and Xenopus (Bang et al., 1997). A later-acting signal from the somitic mesoderm can posteriorize the pattern of Hox gene expression within the developing chick hindbrain (Itasaki et al., 1996; Grapin-Botton et al., 1997). At spinal cord levels, signals from the somites have been shown to restrict the rostrocaudal spread of clonally related cells (Stern et al., 1991). Moreover, different rostrocaudal domains of the embryonic spinal cord, as in the hindbrain, can be defined by the overlapping expression of Hox genes (Kessel and Gruss, 1991; Krumlauf, 1994). Signals from the paraxial mesoderm may, therefore, also control the rostrocaudal identity of cells at more caudal levels of the neural tube and initiate the generation of distinct MN subtypes at different segmental levels of the spinal cord.

We have examined this possibility through an analysis of the expression of Hox and LIM-HD proteins that define the rostrocaudal identity of neural tube cells and the columnar subtype of spinal MNs. Neural tube and mesodermal grafting studies in vivo show that the rostrocaudal identity of neural cells remains plastic at the time of neural tube closure. Positionally restricted signals from the paraxial mesoderm appear to have a critical role in establishing the rostrocaudal identity of cells in the caudal neural tube and later in determining MN subtype identity. Our results suggest that the generation of MN subtypes in the spinal cord involves the integration of rostrocaudal and dorsoventral patterning signals that derive from distinct paraxial and axial mesodermal cells groups.

This study is based on >200 quail to chick embryonic grafts, and an analysis of the 74 embryos which survived and contained quail tissue in appropriate locations.

To determine the location of prospective brachial (B) and thoracic (T) levels of the neural tube for use in grafting experiments we constructed a fate map of the quail neural tube at Hamburger-Hamilton (HH) (1951) stages 10-13. DiI was injected focally at rostrocaudal levels of the neural tube estimated to correspond to B and T levels. Embryos were incubated until HH stage 28-29, the position of labeled cells determined by fluorescence microscopy and mapped onto the segmental level of that defined by vertebral segmental identity. Based on this analysis, we found that the B level of the neural tube in 10-11 somite stage (s) embryos was located 1-2 somites more rostral than the position predicted by extrapolation of the rostrocaudal length occupied by segmented somites. In 14s embryos, this fate map showed that the T neural tube was located adjacent to the caudal half of the unsegmented mesoderm.

In grafting experiments, T level neural tissue at the 22-25 somite level was isolated from quail donors at the 13-18s and grafted to the 16-21 somite level of 11-15s chick hosts. B level neural tissue was isolated from 11-13s stage quail embryos and grafted to the T level of 11-15s chick hosts. Neural tube tissue for grafting was separated from surrounding mesodermal tissue by digestion with Dispase at 20°C for 7 minutes. Enzymatic treatment was stopped by incubation of neural tissue in L-15 medium containing 2% heat-inactivated normal goat serum (HINGS) at 4°C and tissue was kept for less than 40 minutes prior to transplantation into chick hosts. Grafting and DiI injection methods were generally as outlined by Stern and Holland (1993).

Prospective B and T regions of quail and chick mesoderm were located on the basis of the fate maps of Chaube (1959) and Hornbruch and Wolpert (1991). Quail embryos were placed in L-15 medium and mesodermal tissue isolated, incubated in Dispase at 20°C for approx. 7 minutes and transferred to L-15 medium containing 4% HINGS before grafting. Grafting methods were essentially as described by Stern and Holland (1993).

Embryos were fixed in 4% paraformaldehyde and serial 10-15 μm cryostat sections through donor quail and host chick tissue were processed for immunohistochemistry. Rabbit (Rb) (K5) and mAb (4D5) antibodies were used to detect Isl1 and Isl2. Rb antibodies (A7, A8) and mAb 1D5 were used to detect chick Isl1 (Tsuchida et al., 1994; Ericson et al., 1996). Rb antiserum (T4) or mAb 4F2 (Tsuchida et al., 1994) were used to detect Lim1/Lim2. A mAb was used to detect Hoxc8 (Belting et al., 1998). Fluorophore- and peroxidase-conjugated antibodies (Jackson Inc.) were used at 1:100 to 1:500. Quail tissue was detected with mAb QCPN (Developmental Studies Hybridoma Bank). Confocal images were collected on Bio-Rad MRC 600 or Leica TCS4D confocal microscopes.

To distinguish the columnar subtypes of MNs generated at different rostrocaudal positions in the spinal cord we analyzed the combinatorial expression of LIM-HD proteins (Tsuchida et al., 1994), focusing on MN subtypes generated at brachial (B) and thoracic (T) levels of the spinal cord (Fig. 1A). At B levels, MNs are organized into three columns: neurons of the medial MMC (MMC_M) which co-express Isl1, Isl2 and Lim3; neurons of the medial LMC (LMC_M) which coexpress Isl1 and Isl2; and neurons of the lateral LMC (LMC_L) which co-express Isl2 and Lim1 (Fig. 1B-E; Tsuchida et al., 1994). At T levels, MNs are also organized into three columns: MMC_M neurons; lateral MMC neurons which coexpress Isl1 and Isl2 but not Lim3 and dorsomedially positioned CT neurons which express Isl1 alone (Fig. 1F-I; Tsuchida et al., 1994). This profile of LIM-HD protein expression is identical in chick and quail embryos, becomes apparent by HH stages 25-29, prior to the onset of MN cell death, and accompanies the segregation of MNs into discrete columns (Tsuchida et al., 1994, data not shown).

Since neural cells remain sensitive to dorsoventral patterning signals at the time of neural tube closure (Tanabe and Jessell, 1996) we examined whether the columnar identity of MNs can be respecified at a similar developmental stage. Segments of quail B or T neural tube were grafted into chick embryos, exchanging prospective T and B levels of the neuraxis. Grafts were placed in a rostrocaudally inverted orientation to control for a possible influence of the normal rostral-to-caudal gradient of spinal cord maturation on changes in neural pattern. Embryos were permitted to develop until HH stages 28-29 and spinal cord histology, the settling pattern of MNs and the profile of LIM-HD protein expression were examined in both donor and host spinal cord. Graft tissue was identified by expression of a quail-specific antigen, QCPN.

Grafts of 13-15s quail T neural tube were placed rostrally at the B level of 12-15s chick hosts (Fig. 2A; for summary of grafts in this and subsequent experiments see Table 1). Spinal cord segments that developed from grafted T neural tube exhibited a change in histology and in the settling pattern of MNs, compared to normal T segments. The ventral horn at T levels is normally compact (Fig. 1G) but in grafted tissue was greatly enlarged and resembled the ventral horn observed normally at B levels (Fig. 1B,C). Dorsomedial Isl1⁺/Isl2⁻ neurons characteristic of the CT were not detected (Fig. 2C). Ventrally, MNs were organized into an ovoid cluster (Fig. 2C) characteristic of the LMC at B levels (Fig. 1C) rather than the crescent-shaped MMC MN cluster normally present at T levels (Fig. 1G). Consistent with this, MNs in a ventrolateral position coexpressed Isl2 and Lim1 but not Isl1, a LIM-HD code characteristic of the LMC_L (Fig. 2D-F). These findings suggest that grafted neural cells have been diverted from their normal T fates and that their neuronal progeny acquire the molecular properties of B MNs.

Grafts of 11-13s quail B neural tube were placed caudally into the T level of 13-15s chick hosts (Fig. 2G). In a host T environment, the grafted B neural tube developed into spinal cord tissue that exhibited histological features of T spinal cord (Fig. 2H). The grafted tissue contained dorsomedial Isl1⁺/Isl2⁻ MNs characteristic of the CT (Fig. 2I,L). Moreover, ventral Isl1⁺/Isl2⁺ MNs were organized into a crescent-shaped cluster characteristic of the MMC (Fig. 2I,L) and no Isl2⁺/ Lim1⁺ LMC_L-like MNs were detected within the graft (Fig. 2J-L). Thus, cells in such grafts appear to have been diverted from their normal B fates and give rise to MN subtypes characteristic of the T spinal cord. Segments of 12-15s quail T or B level neural tube were also grafted into chick hosts at their appropriate axial level but still in an inverted rostrocaudal orientation. These grafts developed into spinal cord segments with the MN columnar settling pattern characteristic of their position of origin (Table 1A; data not shown), although as discussed below, under this condition the pool identity of MNs in the B LMC is altered.

We next determined the period over which cells at T and B levels of the neural tube remain competent to respond to signals from the host environment with a change in MN columnar subtype. To assess this, quail T neural tube was isolated from later stage donors and grafted into the B level of chick hosts (Table 1A). Spinal cord tissue derived from 15-16s T grafts contained only a few Isl2⁺/Lim1⁺ LMC_L-like MNs and dorsal Isl1⁺/Isl2⁻ CT-like MNs were still detected (Table 1 A; data not shown). Spinal cord tissue derived from 18s quail T neural tube contained no Isl2⁺/Lim1⁺ LMC_L-like MNs and all MNs were characteristic of the T spinal cord, (Table 1A, data not shown). Conversely, 14-16s quail B neural tube grafted into the T level of 14-15s chick hosts contained both dorsal CT-like MNs and LMC_L-like MNs (Table 1A, data not shown). 17s quail B neural tube grafts to T levels generated LMC but no CT MNs (Table 1A). These results suggest that by these stages cells from B and T levels of the neural tube have been exposed to signals that specify their rostrocaudal identity and thus are no longer able to respond to host signals with a change in the columnar identity of MNs.

Early rostrocaudal inversion of lumbosacral level neural tube has been shown to change the pool identity of MNs, as assessed by the pattern of muscle innervation (Matise and Lance-Jones, 1996). We examined whether the rostrocaudal inversion of B neural tube results in a change in the molecular properties of MNs within individual motor pools. To address this issue we made use of an observation that certain motor pools within the rostral two segments of the B LMC can be distinguished at HH stage 29 by the combination and/or level of LIM-HD protein expression. At rostral B levels, a dorsomedial group of MNs within the LMC_L expresses high levels of Isl2 and Lim1 (Fig. 3C) and appears to correspond to the Deltoid (DL) motor pool (Fig. 3B; Hollyday and Jacobson, 1990). A ventromedial group of LMC_L MNs expresses low levels of Isl2 and Lim1 and appears to correspond to the Extensor Metacarpi Radialis/Ulnaris (EMR) motor pools (Fig. 3B,C; Hollyday and Jacobson, 1990). A third MMC_M-like group of rhomboideus (RB) MNs is embedded in the lateral region of the LMC_L (Hollyday and Jacobson, 1990; Straznicky and Tay, 1983) and expresses Isl1, Isl2 and Lim3 but not Lim1 (Fig. 3B,C; Tsuchida et al., 1994). This pool-dependent variation in LIM-HD protein expression was not observed within the two caudal segments of the B spinal cord where all LMC_L MNs co-expressed high levels of Isl2 and Lim1 (Fig. 1D; data not shown).

We first analyzed the pattern of LIM-HD protein expression in segments of a 12s quail B neural tube grafted in a rostrocaudally inverted orientation into the B level of a 10s host chick embryo (Fig. 3A). The profile of LIM-HD expression by MNs within the LMC_L was altered in the rostral two segments of the graft, formerly the caudal level of the donor B neural tube. A dorsomedial high level Isl2⁺/Lim1⁺ (DL-like) pool, a ventromedial low level Isl2⁺/Lim1⁺ (EMR-like) pool and a lateral Isl1⁺/Isl2⁺/Lim1⁻(RB-like) pool were detected (Fig. 3D). The LIM-HD protein profile of these MN pools resembles closely that normally generated in the LMC_L in rostral B spinal cord (Fig. 3C). In the caudal (formerly rostral) two segments of the inverted B spinal cord, all LMC_L MNs co-expressed high levels of Isl2 and Lim1 (data not shown), similar to the normal caudal B spinal cord. 3A). In the graft, dorsal LMC_L neurons expressed higher levels of Isl2/Lim1 than ventromedial LMC_L neurons and laterally located MNs coexpressed Isl1 and Isl2 but not Lim1 (Fig. 3E,F). Thus a similar, albeit less ordered, change in the organization of MN pools was detected. These results provide evidence that the emergence of a pool-specific pattern of LIM-HD protein expression by MNs within the LMC is dependent on the rostrocaudal position of neural tube cells.

The alteration in histological features of the spinal cord after rostrocaudal displacement of the neural tube suggested that the positional identity of cells other than MNs might be changed. To assess this, we analyzed the pattern of expression of Hoxc8, a Hox protein with a domain of expression known to encompass B and T levels of the spinal cord (Le Mouellic et al., 1988; Awgulewitsch and Jacobs, 1990; Belting et al., 1998).

We first analyzed in detail the normal pattern of expression of Hoxc8 in chick and quail embryos at HH stage 28/29. Hoxc8 is expressed over distinct rostrocaudal domains of the spinal cord and paraxial mesoderm. Neural expression of Hoxc8 extends throughout both T and B levels (Fig. 4A), but the pattern varied markedly at different levels. In rostral B segments (Fig. 4A), expression is largely restricted to cells in the dorsal spinal cord whereas at more caudal B levels, expression is detected in ventral interneurons and at the level of the most caudal B segments, also in ventral progenitor cells and MNs (Fig. 4A). In the paraxial mesoderm, the rostral boundary of Hoxc8 expression is located at the junction of T and B levels and expression extends caudally throughout the T level, decaying at rostral lumbar levels (Fig. 4A; Burke et al. 1995; Belting et al., 1998). The pattern of Hoxc8 expression, therefore, provides a sensitive indicator of the rostrocaudal positional identity of cells at B and T levels of the spinal cord and in addition, distinguishes T and B paraxial mesoderm.

We therefore examined whether the pattern of neural Hoxc8 expression is changed after rostrocaudal displacement of T and B segments of the neural tube. In 15s quail T neural tube grafted rostrally into the B level of a 14s chick embryo (Fig. 4B), Hoxc8 expression was restricted to dorsal cells at rostral levels of the graft (Fig. 4C). However, expression expanded ventrally at more caudal levels (Fig. 4D,E), mimicking closely the normal B pattern (Fig. 4A). In 11-12s quail B neural tube grafted in a rostrocaudally inverted orientation into the B level of an 11s chick host (Fig. 4F), Hoxc8 expression was dorsally restricted at rostral, formerly caudal, levels of the graft, (Fig. 4G). Again expression expanded ventrally at more caudal levels (Fig. 4H), generating a pattern similar to that normally detected at B levels. Finally, in 11-13s rostral quail B neural tube grafted caudally into the T level of 13-15s chick hosts (Fig. 4F), a uniform pattern of Hoxc8 expression was detected (Fig. 4I), similar to that found normally at T levels (Fig. 4A). This analysis of Hoxc8 expression patterns provides evidence that the change in cell identity after neural tube displacement is not restricted to MNs, occurs in a graded manner along the rostrocaudal axis of the spinal cord and is associated with both a rostral and a caudal respecification in cell fate. In contrast, neural tube grafts between B and T levels did not change the pattern of Hoxc8 expression in the flanking paraxial mesoderm (Fig. 4C-E,G-I).

The change in identity of MNs detected after displacement of the neural tube led us to examine the source of the patterning signals. Studies of the neuraxis, at hindbrain levels, have raised the possibility that signals transmitted along the neural tube contribute to the early establishment of rhombomere identity (Grapin-Botton et al., 1997). We therefore examined whether there is any change in MN settling pattern or in the profile of LIM-HD protein expression in host chick spinal cord tissue adjacent to the grafted quail neural tube. We did not detect any change in the cellular or molecular features of MN differentiation in adjacent host chick T or B spinal cord (data not shown). We conclude, therefore, that signals transmitted rostrocaudally within neural tube are unlikely to underlie the marked change in positional identity and pattern of cells detected within grafted neural tissue.

We next examined whether signals derived from non-axial mesoderm cell groups might control the subtype identity of MNs. The neural tube at prospective spinal cord levels is flanked by paraxial, intermediate and lateral plate mesoderm (collectively termed non-axial mesoderm) and signals transmitted between these tissues appear to control many aspects of mesodermal differentiation (Stephens et al., 1991; Pourquie et al., 1996; Michaud, et al., 1997). In view of this, our initial analysis of mesodermal influences on rostrocaudal neural tube patterning attempted to preserve any ongoing intra-mesodermal signaling by transplanting all non-axial mesodermal tissues coordinately.

11-12s quail B non-axial mesoderm was grafted ipsilaterally to the T level of 12-15s chick hosts (Fig. 5A) and embryos were analyzed at stage HH 28-29. QCPN expression showed that quail cells contributed to mesodermal tissues only on the side of the graft (Fig. 5B). These embryos developed an additional limb bud at the T level (data not shown), indicating that the signaling activities of non-axial mesoderm required for limb generation (Stephens et al., 1991; Cohn et al., 1995; Crossley et al., 1996; Fernandez-Teran et al., 1997) are preserved. Moreover, there was no expression of Hoxc8 in donor quail mesodermal cells grafted to T levels, whereas Hoxc8 was expressed by contralateral host paraxial mesoderm (Fig. 5F). Thus, the positional identity of the grafted mesodermal tissue appears unchanged by its rostrocaudal displacement.

A marked change in MN settling pattern and LIM-HD protein expression was detected within the host spinal cord on the side ipsilateral to the B mesoderm graft. Ipsilateral T spinal cord did not contain dorsally positioned Isl1⁺/Isl2⁻ CT-like MNs (Fig. 5C,D). Ventrally, MNs settled in a pattern characteristic of the LMC (Fig. 5C-E) and lateral LMC MNs co-expressed Isl2 and Lim1 (Fig. 5D,E,H, data not shown). In contrast, the MN settling pattern and profile of LIM-HD protein expression on the contralateral side remained characteristic of T levels (Fig. 5C-E,G). The widespread neural expression of Hoxc8 on the side of the graft (Fig. 5F) suggested that the T neural tube possessed a caudal B character and consistent with this, there was not a clear distinction in the level of Isl2/Lim1 co-expression by LMC_L-like MNs (Fig. 5H; data not shown).

17-20s quail T non-axial mesoderm was grafted rostrally in place of B mesoderm (Fig. 5I). QCPN⁺ mesodermal cells remained on the side of the graft (Fig. 5J). The positional identity of the grafted mesoderm appeared unchanged, as assessed by the maintenance of Hoxc8 expression in quail mesodermal cells (Fig. 5N) and by the generation of ribs and the absence of a limb bud ipsilaterally (data not shown). These non-axial T mesoderm grafts induced a marked reorganization of the ipsilateral side of the host spinal cord. Isl1⁺/Isl2⁺ MNs were characteristic of the MMC_M at T levels, not of B level LMC (Fig. 5K-M,O) and no MNs coexpressed Isl2 and Lim1 (Fig. 5K-M,P). Although there was a change in somatic MN identity, we detected few if any dorsally positioned Isl1⁺/Isl2⁻ CT-like MNs on the side of the spinal cord ipsilateral to the graft (Fig. 5K,L). In these experiments, successful mesodermal grafts were always positioned at caudal B levels and consistent with this, widespread neural expression of Hoxc8 was detected (Fig. 5N). However, because of this the analysis of these grafts did not reveal whether changes in MN columnar identity are accompanied by a rostralization in the identity of other cell types in the spinal cord.

This series of grafts therefore provides evidence that signals from non-axial mesoderm can change the columnar subtype identity of somatic MNs and can suppress but apparently not induce the generation of visceral MNs.

To determine more precisely the origin of non-axial mesodermal signals, quail paraxial mesoderm was separated from intermediate and lateral plate mesoderm and grafted alone in place of chick paraxial mesoderm.

13-15s quail B level paraxial mesoderm was placed unilaterally at the T level of 12-16s chick host embryos (Fig. 6A). QCPN⁺ mesodermal cells remained on the side of the graft (Fig. 6B) and retained their B identity, as assessed by the absence of Hoxc8 expression (Fig. 6E). The host T spinal cord on the side adjacent to grafted B paraxial mesoderm assumed a B identity: it did not contain dorsally positioned Isl1⁺/Isl2⁻ CT-like MNs (Fig. 6C) and ventral MNs settled in a pattern characteristic of LMC neurons with the most lateral of these co-expressing Isl2 and Lim1 (Fig. 6C,D,H).

In contrast, grafts of later (18-20s) segmented B paraxial mesoderm into the T level of chick hosts did not change the pattern of MN differentiation in the adjacent T level spinal cord (Table 1B; data not shown), suggesting that the patterning activity of the paraxial mesoderm is expressed transiently. Moreover, grafts of 11-15s quail B lateral plate mesoderm adjacent to the T neural tube of chick hosts produced no detectable change in MN settling pattern or in LIM-HD protein expression (Table 1B; data not shown). Thus, the patterning activities of the non-axial mesoderm appear to be mediated by signals from the paraxial mesoderm.

15-18s T paraxial mesoderm was grafted rostrally in place of B level paraxial mesoderm (Fig. 6I). QCPN⁺ donor cells remained on the side of the graft (Fig. 6J) and retained their positional identity as assessed by the maintenance of Hoxc8 expression (Fig. 6M; data not shown). In these experiments embryos were analyzed at HH stage 25 since we encountered problems with survival at later stages. This precluded an analysis of the generation of CT neurons which have not yet begun to migrate dorsally (Prasad and Hollyday, 1991). However, the ipsilateral side of the host level B spinal cord did not contain Isl2⁺/Lim1⁺ LMC_L-like MNs (Fig. 6K,L,O,P). Furthermore, there was an increase in the number of ventral Hoxc8⁺ cells on the side of the graft (Fig. 6M), suggesting that the B neural tube had acquired a more caudal identity.

We also examined whether B neural tube could be induced to acquire a more rostral identity in response to signals from paraxial mesoderm. To test this, we grafted 5s cervical level paraxial mesoderm caudally, in place of B paraxial mesoderm (Fig. 7A). QCPN⁺ quail mesodermal cells remained on the side of the graft (Fig. 7B) and did not express Hoxc8 (Fig. 7F), as expected for cervical or B mesoderm (Fig. 4A). The ipsilateral B spinal cord contained MNs with a settling pattern characteristic of cervical MNs (Fig. 7C; Ericson et al., 1997; data not shown) and few if any MNs exhibited the B property of co-expression of Isl2 and Lim1 (Fig. 7G,H). The LIM-HD protein code does not clearly distinguish somatic MN subtypes generated at the cervical and T levels of the spinal cord (Tsuchida et al., 1994; Ericson et al., 1997). This precluded a molecular assessment of whether the depletion of LMC_L MNs resulted from a rostral (to cervical) or a caudal (to T) shift in MN identity. However, Hoxc8 expression was almost completely extinguished on the side of the spinal cord ipsilateral to the graft (Fig. 7F). Since Hoxc8 is not expressed by cells at cervical levels of the spinal cord (Fig. 4A), this result suggests that the loss of LMC_L MNs elicited by cervical paraxial mesoderm grafts is associated with a rostral, rather than a caudal transformation in the identity of B neural tube cells. This result also provides additional evidence that the pattern of Hoxc8 expression by neural cells can be altered by signals from the paraxial mesoderm.

These studies provide evidence that the paraxial mesoderm has a critical role in imposing a rostrocaudal identity on cells at caudal levels of the neural tube that give rise to the spinal cord. The acquisition of such positional identity appears to underlie the generation of distinct columnar subclasses of MNs at different rostrocaudal positions. Our results add to the emerging evidence (Itasaki et al., 1996; Grapin-Botton et al., 1997; Muhr et al., 1997) that signals from the paraxial mesoderm influence rostrocaudal pattern within the avian neural tube. These prior studies on the patterning of the hindbrain and forebrain have indicated that the paraxial mesoderm exerts a caudalizing influence on neural cell fates.

In contrast, the present studies suggest that at spinal cord levels the paraxial mesoderm is able to respecify cell fates in both a rostral and a caudal direction. We discuss these findings in the context of the progression of progenitor cell differentiation into MNs, the rostrocaudal patterning of the neural tube and the signaling properties of paraxial mesoderm.

The pattern of cell differentiation observed after displacement of the neural tube or paraxial mesoderm indicates that the rostrocaudal identity of neural cells remains plastic at the time of neural tube closure but that there is only a brief period (approx. 12 hours) within which respecification is possible. These observations probably explain the failure of later-stage grafts to alter the pattern of MN differentiation (Wenger, 1951; Narayanan and Hamburger, 1971; O’Brien et al., 1990). An apparent transformation of MN fate, assessed histologically, has been described after grafting cervical neural tube to T levels (Shieh, 1951). However, the age at which most of these grafts were performed is beyond the critical period defined here and more recently, similarly late grafts have not resulted in a change in cervical MN fate (Yaginuma et al., 1996).

Although the critical period for specification of the rostrocaudal identity of neural cells appears to end soon after stage 11, post-mitotic MNs are not generated at B and T levels until stage 15, some 18-24 hours later (Hollyday and Hamburger, 1977; Prasad and Hollyday, 1991; Ericson et al., 1992). The loss of competence for transformation could reflect the time at which progenitor cells in the neural tube commit to MN columnar fates. However, the acquisition of a generic MN fate remains sensitive to local Shh signaling until late in the final cell cycle of MN progenitors (Ericson et al., 1996) which would require that MNs commit to their columnar subtype identity before their generic identity. An alternative possibility is that the loss of competence reflects the time at which progenitor cells within the neural tube acquire secondary inductive signaling activities that are independent of the paraxial mesoderm. In this case the columnar subtype identity of MNs may be determined only at later stages. In support of this idea, studies of MN differentiation in zebrafish have provided evidence that primary MN subtype identity is determined in post-mitotic cells after the specification of generic MN fate (Eisen et al., 1991; Appel et al., 1995).

Our analysis of LIM-HD protein expression suggests that both the columnar and pool subtype identities of MNs are influenced by signals from the paraxial mesoderm. The molecular evidence for a change in MN columnar identity is strongest in the case of the LMC_L, a subtype for which an unambiguous LIM-HD protein code exists. Analysis of the pattern of LIM-HD expression by MNs, however, suggests that signals from the paraxial mesoderm coordinately regulate the generation of all LMC neurons. The sufficiency of paraxial mesodermal signals for the differentiation of CT neurons is less clear. Displacement of B neural tube adjacent to T level paraxial mesoderm leads to the generation of CT neurons but placement of T paraxial (or non-axial) mesoderm adjacent to B level neural tube does not. This result could reflect the requirement for a higher level of paraxial mesodermal signaling in CT neuron generation. Since all mesodermal grafts were performed unilaterally, a competing influence of signals from the contralateral host B mesoderm might therefore be sufficient to prevent the generation of CT but not somatic MNs. We cannot exclude, however, that an additional T-level restricted signal provided by cells other than the paraxial mesoderm, normally contributes to the generation and/or dorsal migration of CT neurons.

An alteration in motor pool identity, determined on the basis of the emergent pattern of muscle innervation, has been described after inversion of lumbosacral neural tube segments at early stages (Matise and Lance-Jones, 1996). Our results show that at B levels, distinct LMC motor pools express different levels and/or combinations of LIM-HD proteins and it is likely that this pool-specific pattern is also established in response to signals from the paraxial mesoderm. Since the organization of motor pools varies between individual segmental levels of the B spinal cord, signals from the paraxial mesoderm may therefore influence MN subtype at a spatial resolution finer than that revealed by the columnar organization of MNs. Consistent with this idea, the mesoderm induces a graded reorganization of the pattern of Hoxc8 expression within the B spinal cord. Thus, a graded control of the positional identity of neural tube cells may result later in the establishment of the distinct segmental (pool) and plurisegmental (column) organizational features of MN subtypes.

Although our studies show that the paraxial mesoderm changes the LIM-HD coding of MN columnar identity they have not addressed whether this code predicts the pattern of motor axon projections. An anatomical analysis of axonal projection patterns after early interchange of B and T levels of the neural tube, however, is unlikely to be informative in assessing the degree to which the molecular and anatomical changes in MN columnar identity are coordinated. Large displacements of the neural tube at stages beyond the critical period for respecification of MN identity lead to axonal projection patterns that are inappropriate for both the normal columnar (O’Brien and Oppenheim, 1990; O’Brien et al., 1990) and pool (Lance-Jones and Landmesser, 1980) identities of MNs. The intrasegmental displacement of primary MNs in zebrafish, has, however, been shown to result in a coordinate change in the expression of LIM homeobox genes and motor axon projections (Eisen et al., 1991; Appel et al., 1995). It is likely, therefore, that LIM-HD protein expression by avian MNs is a pertinent indicator of their columnar subtype identity.

The present studies provide evidence that rostrocaudal differences in the signaling properties of the paraxial mesoderm specify MN subtype identity. Studies of primary MN differentiation in spadetail mutant zebrafish embryos have revealed changes in MN identity and in the profile of LIM homeobox gene expression at levels in which mesodermal development is perturbed (Eisen and Pike 1991; Tokumoto et al., 1995). These changes, however, may reflect a general permissive requirement for mesodermal signaling rather than a more specific positional influence.

What is the basis of the positional signaling properties of the paraxial mesoderm? After its segmentation, the paraxial mesoderm gives rise to sclerotomal and myogenic cells. Sclerotomal cells possess a positional identity which is retained when cells are transplanted to different rostrocaudal levels of the embryonic axis (Kieny, 1971; Kieny et al., 1972; Chevallier et al., 1977; Chevallier, 1979). Myogenic progenitor cells may also express stable and heritable markers of their rostrocaudal position of origin (Donoghue et al., 1992; Grieshammer et al., 1992). The mechanisms that establish intrinsic rostrocaudal distinctions in sclerotomal and/or myogenic cells might also confer upon the paraxial mesoderm its positional signaling properties, reflected later in the pattern of cell types generated in the neural tube. The positional identity of the scelerotomal derivatives of the paraxial mesoderm is dependent on Hox gene activity (Krumlauf, 1994). In turn, there is evidence that the mesodermal expression of Hox genes is regulated by the Cdx class of homeobox genes and by FGF signaling (Subramanian et al., 1995; Shashikant et al., 1995; Pownall et al., 1996; Chawengsaksophak et al., 1997). Defining how the positional identity of the paraxial mesoderm is established may help to determine whether its neural patterning activity results from the spatially restricted expression of different signals or from variations in the level or timing of expression of a single inductive signal.

It remains unclear when signaling from the paraxial mesoderm is initiated and whether cells of the caudal neural plate exhibit relevant anteroposterior positional differences before the action of paraxial mesoderm signals. Analysis of the pattern of Hox gene expression has revealed early and dynamic domains of Hox gene expression at caudal levels of the neural plate that give rise to the spinal cord (Wilkinson et al., 1989; Sundin and Eichele, 1992; Deschamps and Wijgerde, 1993; Gaunt and Strachan, 1994; Krumlauf, 1994). Consistent with this, Hoxc8 is initially expressed over a wider rostrocaudal domain of the neural tube than it is at HH stage 28 (Le Mouellic et al., 1988; Awgulewitsch and Jacobs, 1990; Belting et al., 1998). Paraxial mesoderm-derived signals that operate at the time of neural tube closure may therefore refine pre-existing but coarse positional values that are established at neural plate stages. Such early positional differences in Hox gene expression appear, however, to be subordinate to the patterning information provided later by paraxial mesoderm, at least in the context of MN differentiation.

Signals from paraxial and/or intermediate mesoderm have also been implicated in the induction of limb differentiation in cells of the lateral plate mesoderm (Stephens et al., 1991; Crossley et al., 1996; Ohuchi et al., 1997; Fernandez-Teran, 1997). The induction of LMC neurons by limb level paraxial mesoderm thus raises the possibility that the paraxial mesoderm controls, in a coordinated manner, the differentiation of limb mesenchyme and the generation of MN subtypes whose axonal projections to and within the limb depend on limb mesenchymal cells (Landmesser, 1980; Lance-Jones and Dias, 1991).

At hindbrain levels of the neuraxis the rostrocaudal identity of neural cells appears to be established prior to the specification of dorsoventral cell fates (Simon et al., 1995). Our results suggest that at spinal cord levels progenitor cells in the ventral neural tube are still sensitive to both rostrocaudal and dorsoventral patterning signals. Graded Sonic Hedgehog signaling along the dorsoventral axis appears to establish several distinct progenitor cell populations, notably those defined by expression of Pax6 and Nkx2.2 (Ericson et al., 1997). In the caudal hindbrain, the specification of these two progenitor cell populations by Sonic Hedgehog appears to direct the generation of somatic and visceral MNs (Ericson et al., 1997). In the spinal cord, however, Pax6 and Nkx2.2 are expressed uniformly along the rostrocaudal axis and virtually all MNs appear to derive from Pax6 progenitors (Ericson et al., 1997; unpublished observations).

Thus, in the spinal cord the interpretation of rostrocaudal patterning signals from the paraxial mesoderm is likely to involve the activation of a transcriptional program distinct from that controlling dorsoventral pattern. The identity of relevant factors remains unknown but since Hox genes are expressed in restricted domains along the rostrocaudal axis of the spinal cord (Kessel and Gruss, 1991; Krumlauf, 1994), they could be involved in the interpretation of signals provided by the paraxial mesoderm. In support of this, the ectopic expression or targeted inactivation of Hox genes at hindbrain levels has been shown to result in changes in cranial MN identity (Zhang et al., 1994; Goddard, et al., 1996; Studer et al., 1996).

In conclusion, the present studies indicate a role for the paraxial mesoderm in initiating the diversification of MN subtypes, but the source of signals involved in many subsequent steps in the acquisition of MN subtype identity remain to be resolved. For example, within a single limb-level segment of the spinal cord it is still unclear how the distinctions between MNs of the MMC and LMC or of the medial and lateral subdivisions of the LMC are defined.

We thank C. Stern for guidance in grafting methods, M. Cohn for discussions on limb development and S. Morton for antibodies to LIM-HD proteins. Part of this study was performed by M. E. at University College, London and we are grateful to C. Tickle and members of her lab for hospitality, advice and discussion. We also thank F. H. Ruddle for encouragement and permission to use the Hoxc8 antibody. S. Arber, J. Briscoe, J. Dodd, T. Edlund, J. Ericson, N. Itasaki, R. Krumlauf, J. P. Liu, C. Stern and Y. Tanabe provided helpful criticism of the paper and K. MacArthur and I. Schieren much help in its preparation. This work was supported in part by grants to T. M. J. from the NIH and the ALS Association. M. E. was supported by the Telethon Foundation; T. N. T. by NIH grant 5T32GM07367 and H.-G. B. by NIH grant GM09966 to F. H. Ruddle. T. M. J. is an Investigator of the Howard Hughes Medical Institute.