A clonal analysis of neural progenitors during axolotl spinal cord regeneration reveals evidence for both spatially restricted and multipotent progenitors

A unique feature of tail regeneration in urodele amphibians is the restitution of a functional central nervous system, including the complex cohort of neuronal cell types such as sensory, inter- and motor neurons. A fundamental question in this system is how the resident spinal cord cells are activated to produce the regenerating spinal cord and how they come to form the full spectrum of neuronal cell types. Histological characterization of the urodele spinal cord indicates that the animals retain cells with radial glial characteristics throughout life (Holder et al., 1990; O'Hara et al.,1992). Upon amputation and lesioning of the spinal cord, the terminal radial glial cells polarize toward the tail tip and apparently undergo an epithelial to mesenchymal transition to migrate toward the cut surface (O'Hara et al., 1992). These cells then form a single-cell-layered tube of neuroepithelial cells called the ependymal tube that extends posteriorly in concert with overall tail regeneration. After a significant period of growth, cellular differentiation occurs in a rostral to caudal sequence, with the tip of the tail remaining undifferentiated until later stages of regeneration(Arsanto et al., 1992).

Cell tracking experiments confirm that the radial glial cells are the progenitor cells for spinal cord regeneration. Electroporation into luminal cells of a plasmid where enhanced green fluorescent protein (eGFP) expression was driven by the glial fibrillary acidic protein (GFAP) promoter resulted in transfection of radial glial cells. When these cells were monitored in live animals over time, they contributed to the regenerating ependymal tube,underwent some proliferation and generated, among other cell types, neurons(Echeverri and Tanaka, 2002). In a separate set of studies, the contribution of differentiated neurons to the regenerating spinal cord was tested by retrogradely labeling motor neurons with rhodamine dextran prior to tail amputation(Zhang et al., 2003). Labeled neurons were swept along with the growing ependymal tube, and came to reside in the regenerate. There was no evidence of dedifferentiation of these neurons and so this cell source could not account for the growth and formation of all the new neurons in the regenerate.

Although the basic cell tracking experiments indicated that radial glial cells are involved in regeneration, these studies did not address how the progenitor cells produce the various neuronal cell types during regeneration. Several scenarios are possible. First, the spinal cord may harbor highly multipotent stem cells that undergo expansion and populate all regions of the regenerating spinal cord to form all the different cell types. This situation would necessitate the presence of inducing factors that direct the cells to form the different cell types. Second, the spinal cord may contain highly restricted progenitors that are committed to forming specific neuronal cell types. Third, multipotent stem cells may, under normal regeneration conditions, tend to divide within a local domain, and thus could contribute to a limited spectrum of cell types even though they have the potential to form other cell types if induced by strong extracellular cues. Fourth, regeneration may depend on a mixture of multipotent and restricted progenitor cells.

The different neuronal subtypes such as sensory, inter-, commissural and motor neurons derive from distinct domains along the dorsal, lateral and ventral axis of the developing spinal cord. These domains are molecularly defined by a combinatorial code of transcription factors expressed in the progenitor cells well before neuronal differentiation (for reviews, see Briscoe and Ericson, 2001; Shirasaki and Pfaff, 2002; Tanabe and Jessell, 1996). During embryogenesis signals from outside the neural tube direct uncommitted neural progenitor cells to form these domains. Namely, notochord-derived sonic hedgehog (SHH) and ectodermally derived bone morphogenetic proteins (BMPs) are crucial extracellular signaling molecules for these patterning events.

By contrast, grafting experiments in the regenerating axolotl tail indicate that the information for establishing the dorsal/ventral (D/V) domains of the regenerating spinal cord comes from within the mature spinal cord. Holtzer removed a section of spinal cord from the mature tail and then regrafted it after rotating it 180° in the D/V axis(Holtzer, 1956). Upon healing,the tail was cut through the graft to induce regeneration. Under these conditions, the entire tail regenerate including neural, cartilage and muscle structures was inverted 180° along the D/V axis. This experiment indicated, first, that the information for organizing the D/V patterning of the regenerating spinal cord derives from the mature spinal cord. Second, it indicated that factors from the spinal cord determine the patterning of the surrounding tissues. Schnapp et al. defined some of the molecular factors underlying these properties (Schnapp et al., 2005). They showed that the transcription factors MSX1, PAX7 and PAX6, which delineate dorsal, dorso-lateral and lateral progenitor cell domains in the developing neural tube, were expressed in the uninjured axolotl spinal cord. This indicates that the mature spinal cord contains molecularly defined progenitor cell domains along the D/V axis similar to those found in development. These domains were also found in the regenerating ependymal tube. The secreted signaling factor sonic hedgehog was expressed in the ventral-most floorplate spinal cord cells both in the mature and regenerating tissue. The activity of the sonic hedgehog was shown to be required for determining the PAX7 domain size in the spinal cord, the induction of cartilage ventral to the regenerating spinal cord, and for tail blastema cell proliferation. Although the Schnapp et al. work defined the molecular players underlying progenitor cell identity during spinal cord regeneration, there are still many unanswered questions concerning how the progenitor cells come to regenerate the correct complement of cell types in the spinal cord. A prime missing aspect is how the progenitor cells populate the regenerating spinal cord, and whether a single progenitor cell will produce descendents that will populate all D/V regions of the spinal cord, or whether cells remain within their original domain. It is also not known on a single cell level how cells proliferate to allow posterior extension of the regenerating ependymal tube.

Here we examine these issues by tracking cells both at the clonal cell level and in cell groups. We first determine the size of the mature spinal cord region that gives rise to the regenerating spinal cord. Second, we examine the spatial distribution of clone growth along the anterior/posterior(A/P) and D/V axis. We find that although most cells grow along the A/P axis and remain close to their D or V origin, some cell clones expand to produce progeny that populate multiple D/V areas. These cells can change their transcription factor expression. Interestingly, we found that the posterior-most region of the regenerating spinal cord, termed the terminal vesicle, is molecularly and cellularly distinct from the other regions. The cells there are PAX7^- and PAX6^-. Also, in this region,ventrally located cells spread into the dorsal region and exit the spinal cord into the blastema.

For dissection, axolotls were anesthetized in 0.03% Ethyl p-aminobenzoate(E-1501; Sigma) dissolved in water. Approximately 4 mm-long spinal cord portions were removed from 3.5 cm-long white axolotls (d/d alleles; own colony) and a comparable size spinal cord piece was grafted from an eGFP-expressing transgenic animal (Sobkow et al., 2006). After 7 days of healing, tails were amputated and imaged weekly on an Olympus SZX 12 stereomicroscope using a Spot imaging system (Diagnostic Imaging Systems).

Orthotopic transplantations of prospective floor plate or roof plate stripes between eGFP transgenic axolotl donors (d/d) and white (d/d) hosts(both stage 15-16) were performed (Fig. 8) in 1× Steinberg's solution(Steinberg, 1957) supplemented with antibiotics (penicillin/streptavidin/amphotericine). Embryos were grown to 2 cm larvae at room temperature (RT). Tails were then amputated and photographed every 2-3 days using the imaging protocols described below.

Cells were electroporated by cutting the tail of 2 cm-long larval axolotls and inserting a DNA-filled electrode into the spinal cord lumen, as described by Echeverri and Tanaka (Echeverri and Tanaka, 2003). To transfect the dorsal cells in the axolotl tail spinal cord, the animals were orientated with the dorsal side to the ground electrode. To transfect DNA into single cells,optimum electroporation conditions were three pulses (50 V, 200 Hz and a pulse length of 100 mseconds), applied using an SD9 Stimulator (Grass Telefactor,West Warwick, RI).

In these lineage tracing experiments, a total of approximately 2000 axolotls were electroporated with cytoplasmic eGFP, nuclear DsRed (data not shown) or nuclear GFP plasmids. From approximately 800 axolotl larvae electroporated with nuclear GFP, 77 had clearly identifiable single cells and were used for experimental observation. In 55 (60%) of these cases cells disappeared or did not divide during the course of observation. In 22 (40%)animals it was possible to follow dividing cells during regeneration.

Axolotl larvae containing labeled cells were imaged every 2-3 days by anesthetizing the animals in 0.01% Ethyl p-aminobenzoate and placing them on a coverslip. eGFP-expressing cells were imaged using a Zeiss Axiovert 2 microscope controlled by a Metamorph image acquisition system. Cells were imaged using Zeiss Water Ph2 Plan-Neofluar 25×/0.80 Imm Korr and 40×/1.2 W Korr objectives, and the tails were taped to the coverslip to minimize the distance between the cells and the objective.

For PAX7 and PAX6 immunostaining, larval axolotl tail samples were fixed in freshly made 4% paraformaldehyde (PFA) at RT, then overnight (ON) at 4°C. They were then washed 2×30 minutes in 1×PBS and placed in a solution of 30% sucrose/1×PBS ON at 4°C. For sectioning, the samples were embedded in TissueTek (Sakura).

For adult tail and body spinal cord immunostaining, adult axolotls were perfused through the heart with 1×PBS followed by freshly made 4% PFA. Spinal cord pieces were dissected free from surrounding tissue and placed in 10% and 20% sucrose/1×PBS ON at 4°C, and then into 20% sucrose/3.5%gelatin (Bloom 80-120; Merck)/1×PBS ON at 37°C and embedded in 20%sucrose/7.5% gelatin/1×PBS. Cryosections of TissueTek- and gelatin-embedded samples were collected on Histobond Adhesion microslides(Marienfeld). The sections were allowed to air dry for 2-3 hours and were immunostained as described below.

Monoclonal antibodies against axolotl PAX6 were generated by producing a GST-fusion protein against the axolotl PAX6 protein corresponding to amino acid region 70-421 of the mouse protein. The protein was injected into mice(EMBL facility) and resulting hybridoma clones were screened initially by ELISA, and then on cryosections. PAX6 IgG was purified from hybridoma cell supernatant by binding to a HiTrap Protein G column (GE Healthcare) and elution at pH 2. Purified anti-PAX6 IgG was conjugated with N-hydroxysuccinimidyl ester-digoxygenin (NHS-DIG) in 0.1 M carbonate buffer, pH 9, at a labeling stoichiometry of 20:1 (DIG:AB). After labeling reactions, PAX6-DIG was dialyzed overnight at 4°C in 1×PBS using Spectra/Pore membrane MWCO 6000-8000.

For PAX7/PAX6 double immunolabeling, the cryostat sections were incubated ON at 4°C in block buffer (1×TBS, 0.03% Triton X-100, 10% rabbit or goat serum), incubated with primary PAX7 antibody (Developmental Studies Hybridoma Bank), washed with wash buffer (1×TBS, 0.03% Triton X-100),incubated with secondary antibody Cy5 Fab fragments (goat anti-mouse;Dianova), washed well with wash buffer, incubated with PAX6-DIG antibody,washed with wash buffer, incubated with sheep anti-DIG-rhodamine Fab fragments(11 207 750 910; Roche) and stained with 1 μg/ml Hoechst. All primary and secondary antibodies were diluted in block buffer and each antibody reaction was done ON at 4°C.

For PAX7/βIII-tubulin double immunolabeling the cryostat sections were stained with anti-PAX7 antibody as described above using Cy3 Fab secondary antibody fragments. βIII-tubulin antibody was incubated with Cy5 Fab fragments (30 minutes, RT), and to eliminate the superfluous Cy5 Fab fragments the antibody mixture was incubated with mouse anti-BrdU antibody (30 minutes,RT). The antibody mixture was added to the PAX7/Cy3-stained slides (ON at 4°C). In addition, the slides were stained with 1 μg/ml Hoechst.

All animal procedures were performed in adherence to the guidelines regulating work at the MPI-CBG.

We initially wanted to characterize the size of the progenitor cell zone that gives rise to the regenerate after tail amputation. A small founding progenitor cell pool would result in a large expansion of cells during regeneration whereas a large progenitor cell pool would necessitate a smaller expansion. To define this zone chimeric spinal cords were produced where a spinal cord section of approximately 4 mm was removed from white hosts and was replaced by a comparable section of spinal cord from eGFP-expressing transgenic animals (Sobkow et al.,2006) (Fig. 1A,D). After 7 days of healing, the tails were amputated at different distances ranging from 300 to 3500 μm from the rostral junction between the host and donor spinal cord (Fig. 1B,E, Table 1). The length of remaining eGFP-positive spinal cord was measured at 2 days post-amputation to account for tissue loss because of the injury(Fig. 1B,E, Table 1). At day 9 and day 16,the total length of the eGFP-expressing region versus the total length of the regenerating spinal cord was measured (Fig. 1C,F, Table 1). The amputation plane could be determined as the point where the segmented cartilage changes to the newly forming cartilage rod(Fig. 1, broken line).

These experiments provided several insights into the founding progenitor pool and where progenitors are dividing during ependymal tube growth. First, a length of approximately 500 μm eGFP⁺ spinal cord present at day 2 gave rise to the entire portion of the regenerating spinal cord at day 9,and at day 16 (Table 1), while shorter lengths ultimately labeled the distal end of the regenerate(Fig. 1D,E, Table 1). This suggests that the animal recruits progenitor cells from a 500 μm zone for regeneration. In these 3.5 cm-long animals, by day 9 there had been an approximately sevenfold expansion of the 500 μm zone, and by day 16 there had been a greater than 10-fold expansion of this zone. Second, we analyzed how the regenerating spinal cord grows using animals where less than 500 μm of eGFP⁺ tissue remained at day 2, because this resulted in a chimeric regenerate at day 9 and day 16 (Fig. 1D-F). Overall, the proportion of eGFP⁺ spinal cord in relation to the whole spinal cord regenerate increases slightly from day 9 to 16, indicating that the regenerating spinal cord expands (by cell division)along the whole length, and the tip of the spinal cord is growing slightly faster than the rostral portion of the regenerate. This would be consistent with the rostral to caudal wave of differentiation that sets in at approximately 12 days post-amputation. Third, in a sample where only 300 μm remained post-amputation, eGFP⁺ cells were present in the very tip of the 9-day regenerate (Table 1). Surprisingly, this label was lost by day 16, indicating either cell death, silencing of the eGFP expression, or cell migration out of the spinal cord, an issue that will be addressed in the Discussion.

Although this experiment defined the source of regenerating neural progenitors and the uniform proliferation along the spinal cord, it did not give insight on how clones deriving from individual progenitor cells expand during tail regeneration. Considering the posterior growth of the regenerating spinal cord, we expected significant extension of clones along the A/P axis. More importantly, we wanted to understand whether a single progenitor cell may produce descendents that form diverse neural cell types. Because neural cell types form in characteristic positions along the D/V axis of the spinal cord,we endeavored to label a cell in a defined dorsal or ventral position, follow its division and the position of its descendents and then finally to determine whether the daughters expressed different molecular markers reflecting different identities.

Schnapp et al. had demonstrated the existence of a dorsal PAX7⁺,lateral PAX6⁺ and a ventral SHH⁺ domain in both the uncut and the regenerating spinal cord, providing the essential markers required for our study (Schnapp et al.,2005). Prior to undertaking lineage analysis, we extended the expression analysis of PAX7 and PAX6 to examine the variation of the expression along the A/P axis of the regenerating spinal cord.

In the previous study PAX6 was examined via in situ hybridization(Schnapp et al., 2005). In order to ultimately combine our lineage tracing studies described below with PAX6 localization, we generated a monoclonal antibody against the axolotl PAX6 protein that we use for the current localization studies.

To determine the A/P dependence of the PAX7 and PAX6 expression, we examined immunolabeled cross-sections at different A/P levels of the regenerating spinal cord spanning the terminal vesicle to the amputation plane in 2 cm-long larvae. At the very caudal terminus of the regenerating spinal cord, PAX7 was barely detectable (Fig. 2A,C, arrowhead), but expression was found in an increasing number of cells toward the amputation plane (Fig. 2D,F,G,I). Mature spinal cord showed strong PAX7 expression relative to regenerated spinal cord regions(Fig. 2J,L). PAX6 expression was also not detectable close to the terminal vesicle(Fig. 2A,B) but was expressed in the lateral domains of the regenerating tail spinal cord in more rostral regions (Fig. 2D,E). Closer to the amputation plane, PAX6 and PAX7 were co-expressed in some dorso-lateral cells (Fig. 2G,H, yellow arrows). In mature tissue PAX6 was expressed in lateral and all dorsal domain cells, so that in the dorsal region, all PAX7⁺ cells were PAX6⁺ (Fig. 2J,K). These expression patterns were observed at every timepoint of regeneration. The strong PAX7 and PAX6 expression in mature tail spinal cord suggests that progenitor cells with D/V identity reside in the mature tissue.

This histological analysis of PAX7 and PAX6 as well as the previously published analysis (Schnapp et al.,2005) was performed on 2 cm-long larvae, which represents the stage where our experimental lineage analysis was carried out. We were concerned whether the expression in the mature region represented a situation specific to an animal that was still undergoing rapid growth. We therefore examined PAX7 and PAX6 distribution in the spinal cord of fully grown adults. We found PAX7⁺ and PAX6⁺ cells in the spinal cord of the adult body (Fig. 3A) and uninjured tail (Fig. 3D) spinal cord. Interestingly, in the adult body spinal cord the PAX7⁺ cells were found in a layer of cells (Fig. 3A, arrowheads) that was clearly distinct from the ependyma and the neuronal layer as defined by βIII-tubulin immunostaining(Fig. 3B) and Neuronal Nuclei(NeuN) staining (Fig. 3C). We did not observe any clear co-expression of PAX7 and βIII-tubulin (an early neuronal marker), suggesting that the PAX7-positive cells probably represent a progenitor cell in the mature spinal cord. PAX6 expression was restricted to the lateral cells of the mature body and tail spinal cord ependymal layer. These results indicate that spatially restricted PAX7 and PAX6 expression is a feature of the fully adult spinal cord.

To understand how individual cells proliferate and whether descendents from a single cell populate different D/V domains, we electroporated single cells in the spinal cord with a nuclear-eGFP expression plasmid. We biased the electroporation into dorsal or ventral cells by the relative orientation of the positive and negative electrodes. We then examined cells 2 or 3 days post-electroporation and selected samples that contained a single cell (or single cells spaced distinctly apart) expressing nuclear-eGFP in a clearly identifiable dorsal, ventral or lateral domain (approximately 10% of starting electroporated samples, see Materials and methods). The selected samples were then imaged every second or third day to track the behavior of the cells. Here we present the results of 21 successful single-cell lineage tracing samples(Table 2, clones 1-21).

In the majority of cases (13/21, 62%) cell clones distributed along the A/P axis, remaining close to the initial D/V domain of the parent cell. Of these samples, eight consisted of dorsal cells whose descendents remained dorsal. Immunohistochemical analysis of a dorsally restricted clone showed PAX7⁺ expression of all descendents (clone 4). Five lateral cells remained within the lateral domain(Table 2, clones 1-8 and 12-16). In eight of the 21 samples (38%), however, a cell started in a specific dorsal or ventral location but gave rise to daughters that moved into other domains (Figs 4, 6, Table 2 and clones 10, 11, 17,18, 19, 20). In two of these eight cases, a cell initially located in the ventral domain generated daughters that spread dorsally(Fig. 4), whereas in the remaining six, a dorsal or ventral cell clone spread to the lateral domain.

Fig. 4 shows a ventral spinal cord cell that produced a clone spanning ventral, lateral and dorsolateral domains. On day 5 after amputation and electroporation the cell divided (Fig. 4B) and continued to divide (Fig. 4D-F) so that by day 23 (Fig. 4F) the cell had generated a 12-cell clone that occupied ventral, lateral and dorso-lateral regions of the regenerated spinal cord. On day 23 this sample was fixed for cryosectioning and PAX7/PAX6 double immunostaining to analyze the marker profile of the daughter cells (Fig. 5). The clone consisted of both PAX6^- cells(Fig. 5A,B, arrows) and PAX6⁺ cells (Fig. 5D,E, arrows with asterisks), indicating that a single cell had produced daughters with different identities.

Fig. 6 shows an example of a single cell transfected in the dorsal spinal cord(Fig. 6A). This cell first divided on day 8 after electroporation(Fig. 6B). On day 10 and day 14 one daughter cell moved from a dorsal to a lateral position(Fig. 6C,D, white arrow with asterisk) whereas the other cell maintained its dorsal position(Fig. 6C,D, white arrow). On day 16 (not shown) the lateral cell had moved further laterally and showed decreased GFP expression. In order not to lose the eGFP signal in this cell,the sample was fixed without imaging on that day for domain-marker analysis. In Fig. 7 the two nuclear-eGFP⁺ cells are visible in a single section double immunostained for PAX7 and PAX6. The dorsally located cell was PAX7⁺ (Fig. 7C, arrow) and the laterally located cell was PAX6⁺ and PAX7^-(Fig. 7B, arrow with asterisk).

The behavior of the 21 clones is summarized in Table 2 and the images of the clones are included in supplementary material Figs S1-S19. In Table 2, the grey bars span the days on which the clone was followed either before the cells lost expression or the clone was fixed for analysis. The numbers within the bars denote the number of cells in the clones on a given day. From this data we find that the average clone size was 4, with a range of 2 to 12 cells. Because a cell within a clone would sometimes disappear during observation, this number represents the maximum number of cells in the given clone during its observation period. Disappearance of cells could either be because of dilution of plasmid and loss of nucGFP expression or cell death. The average time of observation of the clones was 17 days. The fourfold expansion of the clones is roughly consistent with the overall growth of the spinal cord when we consider that we initially amputated approximately 2-3 mm from the spinal cord and considering our evidence that the regenerated portion of the spinal cord derives from a 500μm length of the spinal cord.

From these single-cell lineage tracing experiments we conclude that most clones remain associated with their original D/V domains but that some progenitors can give rise to descendents that switch domains. These descendents can express distinct D/V markers(Fig. 7), indicating the action of inductive signals in the regenerating axolotl spinal cord that regulate D/V patterning during regeneration. We observed no defect in the animals harboring multipotent clones that might indicate domain switching was induced by experimental circumstances.

Lineage tracing experiments using cytoplasmic-eGFP showed results that were very similar to the nuclear-eGFP experiments(Table 3). However, the cytoplasmic-eGFP experiments were not grouped together with the nuclear-eGFP results because of the technical limitations of unambiguously identifying a single cell in vivo at the beginning of lineage tracing. It was, however,possible to follow the division of a cell (or cells) that began in a specific region and the distribution of their descendents (supplementary material Fig. S19). From 18 animals used for this analysis, seven (≈40%) of the cell groups remained in one region, whereas in 11 (≈60%) the cells changed between the different domains (Table 3).

In the single cell tracing experiments we found that most cell clones spread along the A/P axis, remaining closely associated with their original D/V location, and only a small minority of clones spread across the entire D/V spinal cord axis. We wanted to understand whether cells that displayed such flexibility in cell fate were located in distinct spinal cord regions. In order to gain a broader overview of dorsal and ventral spinal cord cell fate,we devised a protocol to label groups of dorsal or ventral cells by transplantation of eGFP⁺ neural plate tissue at embryonic stages. Small pieces of eGFP⁺ neural plate tissue, either prospective ventral or dorsal regions, deriving from a germline eGFP⁺transgenic animal were transplanted into unlabeled hosts at embryonic stage 15-16 (Fig. 8). The operated embryos were grown to 2 cm-long larvae and animals harboring dorsally or ventrally restricted cell labeling were collected for cell tracking experiments. In these experiments, seven animals harbored specific labeling,two with dorsally and five with ventrally restricted cell groups. It should be pointed out that this labeling protocol naturally selects for progenitors that maintained a specific dorsal or ventral location throughout the course of development and excludes any cells that had populated multiple D/V domains during embryogenesis.

Fig. 9 illustrates a sample with dorsal cell labeling, whereas Fig. 10 depicts a ventrally labeled animal. The tails were amputated close to dorsally or ventrally restricted eGFP⁺ cell populations(Fig. 9A, Fig. 10A). Dorsal labeling showed that dorsal/dorso-lateral cell groups proliferated and expanded within dorsal and dorso-lateral domains in the regenerating spinal cord(Fig. 9A-C).

Tracing of ventral/ventro-lateral cell populations mainly showed spatial restriction of cells to ventral and ventro-lateral domains during regeneration(Fig. 10A-C). Strikingly, in several ventral cell tracking experiments (3/5) the cells lying close to the terminal vesicle eventually distributed to the dorsal domain during the time of observation (Fig. 10C,E,arrowheads). These new dorsal cells seemed to migrate out of the terminal vesicle (Fig. 10D,arrowheads). Thus, through these embryonic transplantation experiments we conclude that dorsal and ventral cells generally remain distinct, except when close to the terminal vesicle where cells can relocate from a ventral to a dorsal domain.

By combining molecular marker analysis with three types of cell fate tracing, we have addressed several questions regarding how neural progenitor cells regenerate the axolotl spinal cord. By grafting spinal cord tissue from eGFP⁺ donors into unlabeled hosts we determined that a 500 μm zone of the spinal cord behind the amputation plane produces the neural progenitors for the regenerating ependymal tube. Given that the diameter of a radial glial cell is approximately 25 μm, and there are approximately 40 progenitor cells in a spinal cord cross-section at the amputation plane, the founding population for spinal cord regeneration is expected to consist of approximately 800 cells. It should be noted that the ependymal zone of the mature spinal cord is a multilayered structure whereas the early, regenerating ependymal tube is a single cell epithelium, and so the migration and rearrangement of cells in addition to rapid proliferation could account for the initial elongation of the regenerating ependymal tube, a process we have not studied here. Preliminary evidence in our laboratory indicates that the 500 μm zone is independent of A/P location of the amputation plane (A. Tazaki and E.M.T., unpublished), which suggests that the size of the founding progenitor cell pool does not determine the amount of spinal cord regenerated. Analysis of the chimeric spinal cords at later timepoints allowed us to conclude that in the first two weeks of regeneration, cells all along the regenerating ependymal tube are proliferating, leading to relatively uniform expansion of the ependymal tube, although the posterior end of the spinal cord grows slightly more than the anterior portion.

In accordance with the extensive posterior growth of the regenerating ependymal tube, tracking of eGFP-marked progenitor cells, either via single cell electroporation or by embryonic transplantation of neural plate tissue,primarily revealed spreading of proliferating progenitors along the A/P axis. We were, however, particularly interested in the distribution of clones along the D/V axis, as this would indicate whether a single progenitor cell was multipotent; producing daughter cells that would contribute to multiple neuronal cell types. In a significant number of cases, we observed clones that spread from a dorsal or ventral position to lateral positions, and in a small minority of cases we observed ventral cells giving rise to lateral and dorsal cells. We could confirm using PAX7 and PAX6 immunohistochemistry that the daughters from a single cell had different molecular expression profiles,indicating that cells had changed fate during clonal expansion. This is consistent with the presence of extracellular inducing molecules such as sonic hedgehog in the floorplate region of the mature and regenerating spinal cord that could alter progenitor cell fate(Schnapp et al., 2005). Our results indicate that at least some of the neural progenitors in the regenerating spinal cord are multipotent, but they do not resolve whether all progenitor cells are multipotent or if the regenerating spinal cord consists of a mixture of multipotent and lineage-restricted progenitor cells. Precise D/V transplantation experiments to test this issue are technically demanding.

It is interesting to consider our results in the light of those from Gabay et al., who have analyzed D/V identity in cultured neural stem cells in vitro(Gabay et al., 2003). They noted that in vivo, oligodendrocyte precursors derive from ventral neural tube regions whereas astrocyte precursors derive from a distinct dorsal domain,implying that individual neural progenitor cells must generate at most two cell lineage types in vivo. By contrast, trilineage multipotent stem cells are observed in in vitro cultures from the same tissue. They investigated this discrepency and showed that dorsally derived progenitors were ventralized by fibroblast growth factor (FGF) and sonic hedgehog when cultured in vitro, thus leading to tripotency. The authors question the relevance of trilineage stem cells that derive from such conversions in vivo. Our cell tracing experiments show that in vivo, during outgrowth of the ependymal tube, cells from a dorsal or ventral domain can switch to an alternate location and acquire a new identity, as assayed by PAX6 and PAX7 in these studies. Although we have not demonstrated whether our progenitor cells generate the three lineages(oligodendrocyte, astrocyte and neuron), our results strongly indicate that the kind of conversion between dorsal and ventral identities observed by Gaby et al. in culture are relevant to in vivo stem cell behaviors.

Our data indicate that the posterior tip of the spinal cord close to the terminal vesicle is a zone where progenitor cell identity becomes destabilized. Cell tracking experiments using embryonic transplantation of eGFP⁺ floorplate tissue revealed that the ventral to dorsal spreading of cells occurs in the terminal region of the ependymal tube. On a molecular level, we found that the terminal region of the ependymal tube is reduced or absent for PAX7 and PAX6 expression. These results suggest that in this region, the neural progenitor cells are either converted to an alternative cell type or to a more generalized progenitor. We are currently investigating other molecular markers that may shed light on the exact identity of the terminal vesicle cells.

A fascinating aspect of terminal vesicle cell identity is the acquisition of migratory properties. Through our embryonic grafting experiments, we clearly observed a trail of eGFP⁺ cells leaving the dorsal surface of the terminal vesicle into the surrounding blastema during tail regeneration. This phenomenon accounts for the disappearance of eGFP⁺ terminal vesicle cells in our spinal cord transplantation experiments. When we produced a chimeric regenerating spinal cord with an eGFP cell label only in the distal-most region of the regenerating spinal cord, the eGFP label had disappeared after 16 days. This is presumably because of the migration of all the labeled cells into the blastema. Indeed, in these spinal cord transplants and in the embryonic tissue-grafted animals, we see abundant eGFP⁺ cells outside the spinal cord, including blood vessels,apparently in Schwann cells and occasionally in muscle and cartilage. However,both of these grafts are not pure enough to allow us to conclude which of these cell types derive from bonafide neural progenitor cells.

The terminal vesicle has in fact long been characterized as a region where an epithelial to mesenchymal transition occurs, and where cells are proposed to delaminate from the ependymal tube to join the surrounding blastema tissue(Benraiss et al., 1997; Egar and Singer, 1972; O'Hara et al., 1992). Our eGFP⁺ spinal cord transplantation and the eGFP⁺embryonic neural tube grafting experiments allowed us to confirm that cells escape the regenerating spinal cord from this region, as we observe trails of cells emanating from the posterior-dorsal wall of the terminal vesicle.

Clearly an important question is what these migrating cells form during regeneration. The epithelial to mesenchymal transition is reminiscent of embryonic neural crest. Benraiss et al., using biolistic transfection of an alkaline phosphatase expression vector into the newt spinal cord, found alkaline phosphatase-expressing cells emanating from the ependymal tube(Benraiss et al., 1997). At later timepoints alkaline phosphatase expression was observed in melanophores and Schwann cells. We are currently working to confirm whether the primary fate of the migrating terminal vesicle cells is neural crest. Because the cell tracking experiments were performed in white mutant animals (d/d alleles),which are deficient in melanophore migration and survival, it was not possible to assay the contribution to melanophores in the current experiments.

A further question is whether the migrating cells contribute to cell types outside of classical neural crest lineages. Echeverri and Tanaka reported the formation of muscle and cartilage after electroporation into the spinal cord of an expression plasmid where eGFP was driven by the GFAP promoter(Echeverri and Tanaka, 2002). Considering the frequency of the events reported in this study, we expected a higher number of muscle fibers to be labeled after our spinal cord transplantation experiments. The basis for the discrepency between the two experimental results is not clear. One possibility is that the electroporation technique itself induced an increase in cellular plasticity or a tendency to fuse with other cell types. The injury caused by insertion of the glass microcapillary could have caused such events. Another possibility is the use of transient transfection in the electroporation experiments versus the cells from the transgenic animals. One issue is whether the integrated transgene can become silenced during the process of cellular plasticity. For example, such silencing phenomena were observed in cells from a transgenic mouse where eGFP was driven by the same ubiquitous CAGGs promoter(Torensma and Figdor, 2004). eGFP was robustly expressed in mature lymphocytes but was barely expressed in immature thymocytes. We are currently investigating these issues.

The authors thank S. Bramke for her kind help with some of the immunohistochemical preparations and Jan Peychl for his support and advice and for providing light microscopy equipment. Special thanks to Heino Andreas for axolotl care.