A subset of oligodendrocytes generated from radial glia in the dorsal spinal cord

Most spinal cord oligodendrocytes are generated from proliferative,migratory progenitor cells that originate in the ventral neuroepithelium– specifically in the pMN domain, which also gives rise to motoneurons(MNs) (Warf et al., 1991; Pringle and Richardson, 1993; Yu et al., 1994; Timsit et al., 1995; Ono et al., 1995; Sun et al., 1998; Lu et al., 2002; Zhou and Anderson, 2002; Takebayashi et al., 2002; Agius et al., 2004) (reviewed by Richardson et al., 2000; Rowitch, 2004). Induction of oligodendrocyte progenitors (OLPs) is dependent on sonic hedgehog (Shh)secreted from the notochord and floor plate at the ventral midline(Orentas et al., 1999; Poncet et al., 1996; Pringle et al., 1996). Shh induces expression of the basic helix-loop-helix transcription factor Olig2,which is required for the generation of both MNs and OLPs(Lu et al., 2002; Zhou and Anderson, 2002; Takebayashi et al., 2002). Soon after OLPs are generated in the pMN domain, they migrate away in all directions, populating the spinal cord and maturing mainly in the white matter.

Evidence for a restricted ventral source of OLPs seems to conflict with other studies that favour more widespread generation of OLPs. For example,immunohistochemical studies have suggested that radial glial cells can transform into oligodendrocytes at the end of neuronogenesis; because radial glia are present in all regions of the embryonic spinal cord VZ, this implies that oligodendrocyte generation might also be widespread(Choi et al., 1983; Choi and Kim, 1985; Hirano and Goldman, 1988). Furthermore, studies based on transgenic mice that express lacZ under transcriptional control of the myelin proteolipid protein (PLP/DM20)gene suggest that there might be additional dorsal sources of OLPs, at least in the cervical spinal cord and brainstem(Spassky et al., 1998). There are also contradictory data from chick-quail chimeras that raise the possibility of dorsal as well as ventral sources of OLPs in birds(Cameron-Curry and Le-Douarin,1995; Pringle et al.,1998; Richardson et al.,1997).

In the present study, we used Cre-mediated recombination in transgenic mice to follow the neuronal and glial fates of neural precursors in a restricted part of the spinal cord neuroepithelium that expresses the transcription factor Dbx1. The initial expression of Dbx1 (and Dbx2) spans four adjoining domains of the neuroepithelium, two dorsal (dP5 and dP6) and two ventral (p0 and p1) (Pierani et al.,2001). We demonstrate that, in addition to the expected interneuron populations, both oligodendrocytes and astrocytes are generated from the Dbx1/2 domain. The appearance of Dbx-derived cells with a hybrid oligodendrocyte/radial glial antigenic phenotype and a radial morphology suggests direct transformation of radial glia into oligodendrocytes, as was suggested previously (Choi et al.,1983; Choi and Kim,1985; Hirano and Goldman,1988). The Dbx-derived oligodendrocytes comprise less than 5% of the total oligodendrocyte population in the cord and have a more restricted distribution, mainly settling in the lateral white matter radially opposite their point of origin. In addition to oligodendrocytes, the Dbx1/2 neuroepithelium generates at least two classes of astrocytes – fibrous astrocytes in the white matter and protoplasmic astrocytes in the grey.

To identify clones containing the Dbx1 locus, we screened a mouse genomic phage artificial chromosome (PAC) library (RCPI 21 from the UK HGMP Resource Centre) with a probe spanning the 3′ UTR sequence of Dbx1 (derived from IMAGE clone 426959 by BamHI and NotI digest). Six PAC clones were isolated and characterized by restriction enzyme mapping and pulsed-field gel electrophoresis (PFGE). The extremities of the PAC inserts were further confirmed by inverse PCR cloning and DNA sequencing. PAC 631-M19, which contained a genomic fragment 235 kb in length (Fig. 1A), was used to generate transgenic mice.

The targeting vector used to modify PAC 631-M19 is shown in Fig. 1C. The 5′ and 3′ homology regions were amplified by PCR from the PAC clone using Expand High Fidelity Taq Polymerase (Roche). Codon-improved Cre recombinase(iCre) (Shimshek et al., 2002)with a nuclear localization signal attached was fused to the initiation codon of Dbx1 using a PCR-based approach. A chloramphenicol resistance cassette flanked by loxP sites was inserted between iCre and the 3′ homology sequence to allow for selection of the correctly recombined clones. PAC modification was carried out in a bacterial system as previously described (Lee et al., 2001),thereby removing the entire coding region of Dbx1 and replacing it with iCre.

The modified PAC DNA was linearized by AscI digestion. PFGE was used to purify the PAC insert away from small vector fragments. This linear PAC DNA was concentrated into a 4% low-melting point agarose gel. Followingβ-agarase (New England Biolabs) digestion and dialysis with microinjection buffer, the DNA was injected into pronuclei of fertilized mouse eggs. Transgenic mouse pups were screened for the presence of the Dbx1-iCre transgene by PCR and Southern blot (primer sequences available on request).

Embryos were fixed in 4% (w/v) paraformaldehyde (PFA) in phosphate-buffered saline (PBS). Generally, material was fixed overnight for in situ hybridization, whereas fixation times for immunohistochemistry were varied according to the age of the embryo and the epitope in question. Fixed embryos were cryoprotected overnight in 20% (w/v) sucrose in PBS. Postnatal animals were perfused with 4% (w/v) PFA through the right ventricle of the heart under terminal anaesthesia. All tissues were embedded in Tissue-Tek OCT compound (R. A. Lamb) and frozen on dry ice. Sections (5 to 15 μm) were cut on a cryostat and collected on Superfrost Plus glass slides (BDH).

All solutions used for in situ hybridization were pre-treated with diethyl pyrocarbonate (0.1% v/v). The following plasmids were used to generate digoxigenin-labelled probes: mouse Dbx1, a 2 kb fragment spanning the 3′ end of the coding sequence and the entire 3′ UTR, from IMAGE clone 426959 (UK HGMP Resource Centre); mouse Dbx2, a 500 bp PCR product that includes parts of the final exon and the 3′ UTR; mouse Olig2, a 1.7 kb 3′ BamHI-NcoI fragment downstream of the open reading frame; mouse Nkx6.1, a 2.2 kb full-length cDNA (gift from T. Jessell); and Mash1, a 700 bp fragment that includes most of the coding sequence and part of the 3′ UTR, from IMAGE clone 481779 (UK HGMP Resource Centre). Details of the in situ hybridization techniques have been described previously(Pringle et al., 1996; Fruttiger et al., 1999)(http://www.ucl.ac.uk/~ucbzwdr/MandM.htm)

For immunohistochemistry, the antibodies used were: rabbit anti-GFP, at 1:8000 dilution (AbCam); rabbit anti-Olig2 (a gift from D. Rowitch); mouse monoclonal anti-RC2 IgM, at 1:5 dilution (Developmental Studies Hybridoma Bank, DSHB); rat anti-PDGFRα, at 1:500 (BD Biosciences); mouse monoclonal anti-GFAP, at 1:400 (Sigma); guinea-pig anti-Sox10, at 1:2000 (a gift from M. Wegner); monoclonal anti-NeuN, at 1:400 (Chemicon); monoclonal anti-Pax7, at 1:10 (DSHB); monoclonal pan-Islet 4D5, at 1:20 (DSHB);guinea-pig anti-Evx1, at 1:8000 (a gift from T. Jessell); monoclonal CC1, at 1:400 (Merck Biosciences); guinea pig anti-Lbx1, at 1:8000 (gift from T. Jessell); monoclonal anti-Lim2, at 1:5 (DSHB); monoclonal anti-Lim3, at 1:20(DSHB); and monoclonal anti-S100β, at 1:400 (Sigma). Primary antibodies were applied overnight at 4°C. Secondary antibodies were rhodamine-, AMCA-or fluorescein-conjugated anti-rabbit, -goat, -rat, -guinea pig or -mouse IgM,or anti-mouse IgG, at 1:200 (Pierce), and were applied for 60 minutes at room temperature. To allow Olig2 and GFP double labelling, a goat anti-rabbit IgG Fab fragment (Jackson) was used to effectively convert rabbit Olig2 into a goat antibody, prior to detection with an anti-goat IgG secondary antibody. All antibodies were diluted in PBS containing 10% serum and 0.1% (v/v) Triton X-100. Following antibody treatment, sections were stained with Hoechst(Sigma) and post-fixed for 5 minutes in 4% PFA. Sections were mounted under coverslips in Citifluor anti-fade reagent (City University, UK).

Spinal cord cultures were prepared as previously described(Kessaris et al., 2004). Briefly, spinal cords from E12.5 mouse embryos were dissected away from surrounding tissue in Hepes-buffered minimal essential medium (MEM-H)(Invitrogen), and dissociated by incubation in 0.0125% (w/v) trypsin in Earle's balanced salt solution (EBSS, Invitrogen) for 30 minutes at 37°C in 5% CO₂. The cells were mechanically dissociated in the presence of DNaseI and seeded onto 13-mm poly-D-lysine-coated coverslips in a 50 μl droplet of Dulbecco's modified Eagle's medium (DMEM, Invitrogen) containing 0.5% FCS at a density of 2.5×10⁵ cells/coverslip. The cells were allowed to attach for 30 minutes. Defined medium (350 μl)(Bottenstein and Sato, 1979)was added and incubation continued at 37°C in 5% CO₂. Cyclopamine was used at a final concentration of 1 μM, and the FGFR inhibitor PD173074 at 100 nM.

To generate mice expressing Cre recombinase under Dbx1transcriptional control we used PAC transgenics. A 230-kb PAC (RP21-631-M19),containing 147 kb upstream and 85 kb downstream of the Dbx1 gene was modified to replace the coding region of Dbx1 with that of iCre (Lee et al.,2001) (Fig. 1A-Cand see Materials and methods). The modified PAC insert was purified and microinjected into fertilized mouse eggs. Three independent transgenic lines were generated, all of which showed similar expression of the transgene in a pattern resembling that of the endogenous Dbx1(Fig. 1D). From E9.5 to E15.5,expression was restricted to a small region of the neuroepithelium. From E16.5 onwards, all transcript expression was abolished (data not shown). We selected one of these lines for further study.

To establish the dorsal and ventral limits of Cre expression in the neuroepithelium, we crossed the Dbx1-Cre mice to the Rosa26-GFP reporter line (Mao et al., 2001) and analyzed the offspring at embryonic day 10.5(E10.5) by in situ hybridization. GFP expression was activated in a broad neuroepithelial domain, wider than the domain of Dbx1 expression at the same age and corresponding more closely to that of Dbx2expression (Fig. 2A-C). This is consistent with previous reports that Dbx1 is initially expressed(prior to E10.5) in an identical pattern to Dbx2 and only narrows down later (Pierani et al.,2001). The region of GFP activation therefore probably corresponds to the early Dbx1/Dbx2 co-expression domain. The ventral limit of GFP expression abuts the dorsal limit of Nkx6.1 expression, many cell diameters away from the Olig2-expressing domain in pMN(Fig. 2D-F). This is consistent with previous data showing mutually exclusive expression of Nkx6.1and Dbx2 in the spinal cord(Briscoe et al., 2000). The dorsal limit of GFP activation overlaps with Mash1 and Pax7, which are expressed within dP5 or dP6, respectively, at their ventral limit(Fig. 2G-I)(Caspary and Anderson, 2003; Helms and Johnson, 2003). Collectively, the data demonstrate that expression of iCre is restricted to Dbx1/Dbx2-positive neuroepithelial precursors, and activation of the GFP reporter gene is therefore restricted to four domains of the neuroepithelium: dP5, dP6, P0 and P1 (Fig. 1E).

We examined the distribution of GFP in Dbx1-iCre×Rosa26-GFP embryos around the time of neurogenesis in the spinal cord (E12.5). At E12.5, GFP-labelled cells were found exclusively in the ventral and lateral spinal cord(Fig. 3A). GFP-labelled axons were observed in the developing ventral white matter, including in commissural axon tracts (Fig. 3A). Subsets of GFP-labelled neurons in the lateral cord co-expressed Lbx1, which marks early postmitotic neurons derived from dP4, dP5 and dP6(Muller et al., 2002)(Fig. 3D). Other GFP-labelled neurons in the ventral spinal cord expressed Evx1, which marks neurons derived from the p0 domain (Moran-Rivard et al.,2001) (Fig. 3E),and En1, which marks V1 interneurons in the lateral cord (not shown)(Saueressig et al., 1999). The majority of the GFP-labelled cells observed in the spinal cord at this stage co-expressed the homeodomain transcription factor Lim2(Fig. 3F), which is expressed in a variety of postmitotic neurons(Tsuchida et al., 1994).

By contrast, there was no co-expression of GFP in Isl1/Isl2-positive neurons derived from pMN or dP3 (Fig. 3B) (Tsuchida et al.,1994), or in Lim3-positive interneurons derived from p2 or motoneurons from pMN (Fig. 3C)(Tsuchida et al., 1994). In conclusion, and as predicted from previous studies, the Dbx-expressing neuroepithelium gives rise to at least four recognizable neuronal subtypes that have different distributions, different antigenic phenotypes and presumably distinct functions within the spinal cord.

Radial glia were one of the first cell types to express GFP in Dbx1-Cre×Rosa26-GFP offspring(Fig. 4A-F). They had a typical radial glial morphology with a single thin process extending from the ventricular zone all the way to the pial surface (reviewed by Bentivoglio and Mazzarello,1999). Immunolabelling with monoclonal antibody RC2 showed co-localization along the entire length of the process(Fig. 4A-F). Co-expression of RC2 and GFP persisted throughout embryogenesis. However, the morphology of these cells changed with time. From E15.5, an increasing number of (RC2, GFP)double-labelled cells appeared to translocate their cell bodies close to the pial surface (shown in Fig. 4G-L at E16.5) and to change their antigenic profile (see below). We thought it possible that these might correspond to radial glial cells in the process of transforming into astrocytes and/or oligodendrocytes at the end of neuronogenesis (Schmechel and Rakic,1979; Choi et al.,1983; Choi and Kim,1985; Hirano and Goldman,1988; Voigt,1989).

We immunolabelled spinal cord sections from Dbx1-Cre×Rosa26-GFP offspring for GFP, together with either Olig2 or Sox10, markers of the oligodendrocyte lineage. Starting around E15.5, cells with a GFP-labelled radial process reminiscent of a radial glial cell and either an Olig2- or a Sox10-positive nucleus were found near the pial surface (Fig. 5A,B). At this stage, Olig2 expression was observed in approximately 25% of the radially-orientated GFP-positive cells. Triple immunolabelling with RC2,anti-GFP and either anti-Sox10 or anti-Olig2 showed co-localization of all three markers in a small subset of cells suggesting that these represent oligodendrocyte lineage cells arising directly from radial glia(Fig. 5E-P). These triple-labelled cells were infrequent (18% of the Olig2+/GFP+ radial cells),consistent with them being in a transitional state. The radially orientated appearance of GFP-labelled oligodendrocyte lineage cells was maintained until late embryogenic stages (Fig. 5C,D).

To further confirm the oligodendroglial identity of these cells and to determine their distribution at later developmental stages, we looked in postnatal spinal cords 10 days after birth (P10). GFP-expressing oligodendrocytes and OLPs, identified by immunolabelling with anti-Sox10,anti-Pdgfra or monoclonal antibody CC1, were concentrated in white matter,mainly in the lateral funiculus, although rare cells also were present in the dorsal and ventral funiculi (Fig. 6 and data not shown). At this stage they had a more rounded cell body with multiple processes, typical of maturing oligodendrocytes. They comprised around 3% of the total Sox10-expressing oligodendroglial cells in the spinal cord at P10.

The Rosa26 promoter remains active in adult Rosa26-GFPmice but its activity is reduced and GFP is difficult to detect by immunohistochemistry. Therefore, to determine the contribution of Dbx-derived oligodendrocytes to the adult spinal cord, we switched to using the Rosa26-lacZ reporter strain (Soriano,1999). β-Gal-positive cells were observed throughout the spinal cord, and especially in the white matter, in a distribution similar to that seen at earlier stages, indicating that Dbx-derived oligodendrocytes make a limited contribution to the oligodendrocyte population of the adult spinal cord (data not shown).

The Dbx-expressing neuroepithelium is exposed to low levels of Shh in vivo and neuronal specification within the Dbx domain depends on signals other than Shh (Pierani et al., 1999). The emergence of oligodendrocytes from this region was therefore surprising given the known requirement for Shh in specifying OLPs in the ventral pMN domain. FGF has recently been shown to be capable of promoting oligodendrocyte specification in cultures of neural precursors from embryonic cerebral cortex or spinal cord (Chandran et al.,2003; Kessaris et al.,2004). To determine the signalling requirements for the generation of Dbx-derived oligodendrocytes in vitro, we dissociated embryonic spinal cords from Dbx1-iCre×Rosa26-GFP mice at E12.5,approximately three days before the first appearance of Dbx-derived oligodendrocyte lineage cells in vivo, and cultured the cells in the presence of the hedghog inhibitor cyclopamine(Cooper et al., 1998; Incardona et al., 1998) or the Fgfr inhibitor PD173074 (Skaper et al.,2000). In the absence of any inhibitors, GFP-labelled oligodendrocytes – identified by co-labelling with anti-Sox10 antibody– developed in vitro within 5 days(Fig. 7A,E). Their development was not inhibited by cyclopamine (Fig. 7B,E) but was completely abolished by PD173074(Fig. 7C-E), indicating a strict requirement for FGF signalling.

In addition to oligodendrocytes and interneurons, two types of astrocytes were generated from the Dbx-expressing neuroepithelium(Fig. 8): GFAP-positive`fibrous' astrocytes in the lateral funiculus(Fig. 8A,B) and `protoplasmic'astrocytes in the grey matter (Fig. 8A-F). The position and arrangement of Dbx-derived fibrous astrocytes was reminiscent of the distal ends of the radial glial processes,suggesting that they, like oligodendrocytes, might be generated directly from radial glia by trans-differentiation(Voigt, 1989). Protoplasmic astrocytes had a very complex morphology with a small cell body and an extensive network of fine branched processes labelled with GFP(Fig. 8C-F). Some protoplasmic astrocytes co-labelled weakly with anti-GFAP(Fig. 8D) and all expressed S100β (Fig. 8C). They were interspersed among dorsal neurons (not shown) and motoneurons(Fig. 8E). They also formed bilateral, longitudinal columns of cells positioned adjacent to the central canal (arrows, Fig. 8F). They did not express NeuN (Fig. 8E,F) or Sox10 (data not shown). It was noteworthy that there was no contribution from Dbx-derived cells to the ependymal layer of the postnatal spinal cord (Fig. 8D).

Using transgenic mice that express Cre recombinase under Dbx1control, we fate-mapped a discrete region of the spinal cord neuroepithelium near the dorsoventral midline and found that it generates a subpopulation of oligodendrocytes. Dbx-derived OLPs appeared about three days later than the majority of OLPs (which are generated from the pMN domain starting on E12.5)and had a more limited distribution. Their initial morphology and antigenic profile resembled radial glia, suggesting that they are formed by trans-differentiation of radial glia. Unlike their ventrally derived counterparts, specification of Dbx-derived OLPs in vitro was independent of Shh but required FGF signalling. However, like pMN-derived OLPs, they must require Olig gene function because oligodendrocyte lineage cells are completely absent from the spinal cords of Olig2 or Olig2/Olig1 compound knockout mice(Lu et al., 2002; Takebayashi et al., 2002; Zhou and Anderson, 2002). In keeping with this, we found that Dbx-derived oligodendrocytes, like those from pMN, also expressed Olig2.

Our experiments show unequivocally that some oligodendrocytes are generated from the Dbx neuroepithelial domain. We estimate that the Dbx-derived population comprises around 3% of all Sox10-positive oligodendrocyte lineage cells. This might be an underestimate, for not all of the neuroepithelial cells in the Dbx domain were GFP-labelled in all sections, presumably because of the transient nature of Dbx1-iCre expression. However, a majority of Dbx-expressing neuroepithelial cells was always labelled, so we think it unlikely that their real contribution to oligodendrocyte production is more than 5%. Most of the remainder presumably come from pMN, although further sub-populations of oligodendrocytes might turn out to be derived from outside either pMN or the Dbx domain (for a summary, see Fig. 9). It would be interesting to determine whether the Dbx-derived oligodendrocytes are functionally different from the rest. They are found predominantly in the lateral white matter suggesting that their progenitors might migrate less widely than the majority do.

There have been previous suggestions of a dorsal source(s) of oligodendrocytes in the rodent cervical spinal cord and hindbrain, based on early expression of the myelin proteolipid protein (PLP/DM20) or on expression of a PLP/DM-20-lacZ transgene(Spassky et al., 1998; Spassky et al., 2000). No estimate could be made of the relative contributions of dorsal versus ventral sources in those studies. A previous fate-mapping study from our laboratory using chick-quail chimeras suggested that oligodendrocytes develop only from ventral grafts (Pringle et al.,1998), although Cameron-Curry and Le Douarin concluded from similar experiments that oligodendrocytes are generated from all parts of the neuroepithelium (Cameron-Curry and Le Douarin, 1995). It might be that Pringle et al. overlooked a small number of dorsally derived oligodendrocytes(Pringle et al., 1998). Alternatively, the dorsal source of oligodendrocytes might be present in rodents but not birds.

The dorsally derived oligodendrocyte lineage cells expressed the same markers as their ventrally derived counterparts – at least those we tested (Olig2, Sox10, Pdgfra and CC1). Thus, we have found no evidence so far that the dorsally derived subpopulation is antigenically distinct.

Shh released from the floor plate and notochord is essential for the development of OLPs in the ventral spinal cord. Chandran et al. recently showed that neural stem cells derived from dorsal spinal cord can be induced by FGF2 to generate OLPs in vitro(Chandran et al., 2003). The same was shown for embryonic cortical precursors by Kessaris et al.(Kessaris et al., 2004). In this study, we made use of Dbx1-Cre×Rosa26-GFP mice to identify neural precursors derived specifically from the Dbx domain in embryonic spinal cord cultures. Our experiments showed that OLPs can develop from Dbx-derived precursors independently of Shh (i.e. development is not inhibited by cyclopamine), consistent with a previous report of Shh-independent specification of interneurons in this region(Pierani et al., 1999). We also showed that OLP development from Dbx-positive spinal cord precursors requires FGF signalling in vitro. To discover whether FGF signalling is actually implicated in vivo would require specific inhibition of FGF signalling in the dorsal spinal cord; this is difficult to contemplate at present because we do not know which of the large number of FGF family members is involved or indeed which FGF receptor.

Some of the first cells to be labelled with GFP in our reporter mice were radial glia. The nuclei of these cells are initially located in or close to the VZ, and their radial processes extend to the pial surface. By analogy with radial glia in the brain, it seems likely that they correspond to the neuroepithelial precursors themselves(Malatesta et al., 2003; Anthony et al., 2004). From E15.5, when the radial glial cell bodies moved towards the pial surface, we observed co-expression of RC2 and the oligodendrocyte lineage markers Olig2 or Sox10 in some radially oriented cells, suggesting that the radial glia transform directly into oligodendrocytes. This supports previous evidence that spinal cord radial glia can give rise to oligodendrocytes as well as astrocytes after neuron generation has ceased(Choi et al., 1983; Choi and Kim, 1985; Hirano and Goldman, 1988).

It is not clear whether the mode of OLP generation from Dbx-expressing precursors is fundamentally different from their mode of generation in the ventral cord. pMN-derived OLPs are presumably derived from radial glia too, as the latter are now thought to correspond to stem cells that generate all the neurons and glia of the CNS. However, the pMN-derived OLPs arise close to the ventricular surface, whereas we first detected Dbx-derived OLPs towards the pial surface. This presumably relates to the position of the radial glial cell bodies at the time they start to express oligodendrocyte lineage markers such as Pdgfra, Sox10 or Olig2. Another difference seems to lie in how widely the OLPs disseminate after they are formed; those formed early from pMN proliferate rapidly at first and migrate away from pMN in all directions,whereas the later-forming Dbx-derived OLPs proliferate less strongly, if at all, and migrate less widely. It remains to be determined whether these different behaviours reflect different inducing signals (Shh versus FGF, say),different environments in the cord at early versus late embryonic stages, or some intrinsic difference between the ventral and dorsal subsets.

In addition to interneurons and oligodendrocytes, at least two morphological subtypes of astrocytes were generated from the Dbx domain. Protoplasmic astrocytes had a typical spherical outline with an extensive and intricate network of sheet-like branching processes and were distributed widely within the grey matter of the spinal cord. They were frequently associated with the cell bodies of neurons in both the dorsal and ventral cord, including motoneurons. Fibrous astrocytes, however, had a restricted distribution within the lateral white matter, in those regions previously occupied by the distal ends of the radial glia processes. The latter observation is consistent with the idea that some radial glia transform directly into fibrous astrocytes at the end of neuronogenesis(Schmechel and Rakic, 1979; Voigt, 1989) (for a review,see Goldman, 2001).

Our Dbx1-iCre mice label the progeny of four neuroepithelial domains – p1, p0, dP6 and dP5 – so we cannot tell whether each one of these domains generates all of the glial cell types described here or whether different subsets of glia are generated from each individual domain,as has been shown for neurons. Resolution of this question would require targeting of the Cre transgene to individual neuroepithelial domains within the broader Dbx domain; however, this is not possible with currently available Cre deleter strains.

We observed that Dbx1/Dbx2 precursors do not contribute to the ependymal layer that surrounds the reduced lumen of the postnatal and adult spinal cord, adding to previous evidence that ependymal cells are derived entirely from ventral (Nkx6.1-expressing) neuroepithelium(Richardson et al., 1997; Fu et al., 2003). Nevertheless, a subset of the Dbx-derived protoplasmic astrocytes settles close to the postnatal ependymal layer where they form narrow bilateral columns of cells along the length of the spinal cord. It is interesting to speculate that these might correspond to a subset of `subependymal astrocytes'analogous to those that have been described as neural stem cells in the postnatal telencephalon (Doetsch et al.,1999).

While this article was being reviewed, two other articles were published that demonstrate production of a subset of oligodendrocyte precursors in the dorsal VZ (dP3, dP4 and dP5) (Cai et al.,2005; Vallstedt et al.,2005) (reviewed by Miller,2005).

We thank Matthew Grist for technical assistance, Ulla Dennehy and Palma Iannarelli for transgenic mouse production, and our colleagues for useful discussions. We also thank Tom Jessell, David Rowitch, Michael Wegner, Rolf Sprengel and Neal Copeland for antibodies and plasmids, Jaime Carvajal for advice about PAC recombination, and Nigel Pringle for help with immunohistochemistry. We thank Stephen Skaper for PD173074 and William Gaffield for cyclopamine. Work in the authors' laboratory is supported by the Wellcome Trust Functional Genomics Initiative, the UK Medical Research Council and a Wellcome Trust Studentship to M.F.