Spinal Cord Injury Reveals Multilineage Differentiation of Ependymal Cells

1Transplantation of different types of stem cells improves functional recovery after spinal cord injury in rodents and primates. The beneficial effects appear to be mediated by several mechanisms, including replacement of lost cells, secretion of neurotrophic factors, and probably most importantly, the generation of oligodendrocytes that remyelinate spared axons in the vicinity of a lesion.

2Neural stem cells present in the adult spinal cord can be propagated in vitro, and promote functional recovery when transplanted to the injured spinal cord. Endogenous neural stem cells could therefore be attractive candidates to manipulate for the production of desired progeny after spinal cord injury as an alternative to stem cell transplantation. This approach would offer a noninvasive strategy that avoids the need for immune suppression, but has been held back by difficulties in identifying adult spinal cord neural stem cells and developing rational ways to modulate their response to injury. Studies using indirect techniques have suggested that the neural stem cell potential in the adult rodent spinal cord resides in the white matter parenchyma or close to the central canal, either in the ependymal layer or subependymally.

3We have employed genetic fate mapping to characterize a candidate neural stem cell population in the adult spinal cord and show that close to all in vitro neural stem cell potential resides within the population of ependymal cells. Ependymal cells give rise to a substantial proportion of scar-forming astrocytes as well as to some myelinating oligodendrocytes after spinal cord injury. Modulating the fate of ependymal cell progeny after injury could potentially promote the generation of cell types that may facilitate recovery after spinal cord injury.

4In order to fate map candidate neural stem cells close to the central canal, we generated two transgenic mouse lines expressing tamoxifen-dependent Cre recombinase (CreER) under the control of FoxJ1 (HFH4) or Nestin regulatory sequences. FoxJ1 expression is specific to cells possessing motile cilia or flagella. In the adult forebrain, a subset of astrocytes in the subventricular zone contact the ventricle and have an immotile primary cilium, but FoxJ1 expression is restricted to cells with motile cilia. Nestin is expressed in central nervous system stem and progenitor cells during development and in adulthood. In the adult spinal cord, nestin is expressed by cells lining the central canal, endothelial cells, and sparse white matter glial cells. The second intron enhancer in the Nestin gene allows for selective expression of CreER in the neural lineage, eliminating expression in for example endothelial cells.

5CreER expression in the adult spinal cord is limited to cells lining the central canal in both the FoxJ1-CreER and Nestin-CreER mouse lines (Figure 1). Administration of tamoxifen to mice on an R26R or Z/EG Cre reporter background allows inducible, permanent. and heritable genetic labeling by the expression of β-galactosidase (β-gal; R26R) or GFP (Z/EG) in cells expressing CreER (the strategy is schematically depicted in Figure S1). Recombination in the absence of tamoxifen was exceptionally rare (<1 cell/30 coronal 20-μm-thick sections in both transgenic lines) and limited to CreER-expressing cells in the ependymal layer. Administration of tamoxifen (five daily injections) resulted in recombination of the reporter allele (Figure 1A–1D) in 82 ± 4% of transgene-expressing cells in Nestin-CreER mice and 88 ± 4% in FoxJ1-CreER mice (mean ± standard deviation [SD], n = 6 mice for each mouse line).

6The cells at the central canal expressing CreER protein from the Nestin-CreER or FoxJ1-CreER transgene are immunoreactive to Crocc, a marker for ciliated cells (Figure S2). They contain the intermediate filaments nestin and vimentin, associated with immature neural cells, but notably not glial fibrillary acidic protein (GFAP) (Figures S2 and S3), which is present in some neural stem cells in the adult forebrain. The transgene expressing cells display other markers associated with neural stem/progenitor cells such as CD133/prominin-1, Musashi1, PDGFR-α, Sox2, Sox3, and Sox9 but are negative for the oligodendroglial progenitor marker Olig2 (Figures S2 and S3). All above-mentioned proteins appear uniformly expressed by the cells lining the central canal, and we have not found any molecular marker delineating any subpopulations.

7Immunoelectron microscopy established that the Nestin-CreER and FoxJ1-CreER transgenes are expressed in identical cell populations by the central canal; their expression is restricted to lumen-contacting cells with motile cilia (9 + 2 axonemes), and all such cells express both transgenes (Figures 2A–2C, S4, and S5). Ultrastructural analysis in serial sections revealed morphological heterogeneity among the lumen-contacting ciliated cells, with some cells displaying typical cuboidal ependymal cell morphology and others a tanycyte morphology (Figures 2C, 2D, S4, and S5). In addition, there is a less numerous third cell type, which we refer to as a radial ependymal cell. Radial ependymal cells share the morphology of the cytoplasm, and often nucleus, with ependymal cells, but have a long basal process (Figures 2B, 2D, and S5). The radial ependymal cells almost invariably reside in the dorsal or ventral pole of the ependymal layer, with a basal process oriented along the dorsoventral axis (Figures 2A and S5). Although the lumen-contacting ciliated cells can be subdivided into these three groups, cells with intermediary phenotypes are frequent (Figure 2E and Table S1), which together with their homogeneous molecular profile suggests that they are closely related. The naming of ependymal cell types is based solely on morphological criteria and does not imply any function. The central canal-contacting ciliated cells have in common that they reside in the ependymal layer, thus we collectively refer to them as ependymal cells.

8Adult spinal cord neural stem cells can be propagated in vitro, but their precise identity has been difficult to establish unequivocally. We utilized our genetic labeling paradigms to ask whether adult spinal cord ependymal cells have neural stem cell properties in vitro. Adult FoxJ1-CreER and Nestin-CreER mice on Cre recombination reporter background (R26R or Z/EG) received five daily injections of tamoxifen to induce recombination, and primary cultures were initiated after an additional 6 d without tamoxifen (Figure 3A). Tamoxifen and its active metabolite 4-hydroxytamoxifen have a half-life of 6–12 h in the mouse, and accordingly CreER protein was no longer detectable in the nucleus after 6 d without tamoxifen (Figure 1B and 1D). Spinal cords were dissociated and plated at clonal density in standard conditions that allow for neurosphere formation (Figure 3B). We found that 76 ± 5.7% of neurospheres from Nestin-CreER and 85 ± 2.2% from FoxJ1-CreER mice were recombined and thus derived from recombined ependymal cells (mean ± SD, n = 6 mice analyzed separately per line, Figure 3C and 3E). Neurospheres were either homogeneously recombined or not recombined, verifying their clonal origin.

9Since recombination never was fully penetrant in the ependymal cells, the analysis of the proportion of neurospheres that were recombined may underestimate the true contribution of this cell population to neurosphere formation. If there is a stochastic distribution of recombination within the CreER expressing cell population, rather than recombination demarcating a subpopulation that differs with regard to neurosphere-forming potential, one can estimate the contribution of the cell population to neurosphere formation by normalizing it to the observed recombination rate. To estimate the theoretically maximal proportion of neurosphere-initiating cells that are ependymal cells, we analyzed the recombination frequency in CreER-expressing cells in sections from each spinal cord sample that was used for neurosphere cultures (Figure 3B). Normalizing the recombination frequency in neurospheres to the recombination frequency in the CreER-expressing cells in vivo, suggested that close to all neurosphere-initiating potential could reside within the ependymal cell population under these conditions (Figure 3D).

10Progenitor cells with limited self-renewal capacity can give rise to neurospheres, but are incapable of generating new neurospheres when passaged more than twice. We found that 100% of the recombined neurospheres from both Nestin-CreER and FoxJ1-CreER could be serially passaged at least eight times to give rise to new neurospheres (n = 6 neurospheres per 4 transgenic mice). The number of cells increased exponentially during passaging (Figure S6). Analysis of the differentiation potential of ependymal cell-derived neurospheres after three passages revealed that 100% of the neurosphere clones were multipotent and differentiated into neurons, astrocytes, and oligodendrocytes (Figure 3F).

11We also isolated prospectively identified ependymal cells by flow cytometry independently of Cre-mediated recombination by utilizing the green fluorescent protein (GFP) expression under the bicistronic control of the FoxJ1 promoter (Figures 3G and S1). Flow cytometric isolation of adult spinal cord cells substantially reduced neurosphere formation, and 0.18 ± 0.06% (mean ± SD from in average 1,600 GFP-positive (GFP+) cells/mouse, n = 6 mice analyzed separately) of GFP+ ependymal cells formed neurospheres (Figure 3H–3K). In contrast, not a single neurosphere developed from the same number of cells in the GFP− non-ependymal fraction from any animal in the same experiments. Thus, the neural stem cell potential in the adult spinal cord, at least under the conditions employed here, largely resides within the ependymal cell population.

12Cells in the adult spinal cord ependymal layer proliferate, albeit slowly or rarely. Continuous administration for one month in the drinking water of 5-bromo-2-deoxyuridine (BrdU), which is incorporated into DNA in cells in S-phase, resulted in labeling of 19.9 ± 4.2% of ependymal cells (mean ± SD from three mice, Figure 4A and 4B). The BrdU-labeled ependymal cells constituted 4.8 ± 0.9% of all BrdU-labeled cells in a spinal cord segment (mean ± SD from three mice). The majority of BrdU-labeled ependymal cells were found in pairs, indicating that most mitoses resulted in self-duplication rather than the generation of another cell that had left the ependymal layer (Figure 4A and 4B). In line with this, analysis of the distribution of recombined cells up to 8 mo after tamoxifen administration in the FoxJ1-CreER and Nestin-CreER mice did not provide evidence for the generation of cells that leave the ependymal layer under normal conditions (unpublished data).

13Whether a specific cell population is derived from another cell type or it is maintained through self-duplication can be established by analyzing the genetic labeling frequency at different time points after induction of recombination. There was no reduction in the proportion of recombined ependymal cells for up to 10 mo after tamoxifen administration (Figure 4C–4E), indicating that ependymal cells are maintained by self-renewal and are not replenished by another cell population.

14We next assessed the response of ependymal cells to spinal cord injury. We used the same labeling paradigm as before (Figure 3A), with a 6-d period between the last tamoxifen dose and the injury. This ensures that all recombination occurs prior to the insult and that even if other cells than ependymal cells would start to express the FoxJ1-CreER or Nestin-CreER transgene in response to the injury (nestin is indeed expressed by reactive astrocytes), it would not result in recombination. An incision in the dorsal funiculus, which does not compromise the integrity of the ependymal layer, dramatically increased the proliferation of ependymal cells (Figures 5A, 5B, and S7). In contrast to the uninjured spinal cord, where proliferation of ependymal cells appears largely limited to self-renewing divisions, recombined cells started to migrate and were located outside the ependymal layer 4 d after the injury (Figure 5C–5I). Migrating recombined cells lost their ependymal phenotype as judged by the loss of immunoreactivity to Sox2 and Sox3 and lack of CreER expression from the FoxJ1 promoter (Figure 3D and unpublished data). Most emigrating cells expressed Sox9 and some the astrocyte marker GFAP (Figure 5F, 5H, and 5I). Ultrastructural analysis revealed that ependymal cell morphology was largely unaltered by the injury, with the exception of a darker cytoplasm due to a higher content of filaments (Figure 5J and 5K).

15Ependymal progeny migrated towards the injury site in the dorsal funiculus and an increasing number of recombined cells accumulated in the forming glial scar over several weeks and remained there for at least 10 mo after the insult (Figure 6A–6C). The recombined ependyma-derived cells occupied 18.3 ± 6.9% (mean ± SD from three FoxJ1-CreER mice) of the area in the scar tissue 2 wk after the injury, which is likely to be a slight underestimate of the true contribution since recombination never was fully penetrant. The ependyma-derived cells were not evenly distributed throughout the injury site, but the scar consisted of patches of recombined and unrecombined cells (Figure 6H and 6I). The reaction of the ependymal cells was restricted to the injured segment and was absent in adjacent segments (Figures 5A–5H and 6A–6C), which are indirectly affected by the severance of axons and Wallerian degeneration. There were no recombined cells outside the ependymal layer in animals in which only the spinal cord was exposed but no lesion was made (sham lesion, Figure S8), and a lesion did not induce recombination in animals that had not received tamoxifen (unpublished data).

16Since some ependymal cells extend processes along the dorsolateral midline, it was possible that the activation of ependymal cells by a dorsal funiculus incision was triggered by the severance of such processes. To investigate this, we performed incisions in the lateral spinal cord, which do not directly injure the ependymal cell processes in the dorsolateral midline. In these animals, ependymal cell progeny were generated and migrated laterally towards the injury (Figure S8). The ependyma-derived cells migrating to the lesion appeared less numerous after a lateral than after a dorsal incision, suggesting that severance of ependymal cell processes in the midline is not necessary for the activation of ependymal cells, but that it may augment their reaction. The migration of ependyma-derived cells to the site of injury suggests the presence of attractive signals originating in the lesion area. SDF1, through its receptor CXCR4, mediates attraction of progeny from neural stem/progenitor cells after some types of injuries. The majority of ependymal cells as well as their progeny were, however, negative for CXCR4 (Figure S9), making it unlikely that this receptor mediates the attraction of ependymal cell progeny to a spinal cord lesion.

17Analysis of the fate of the ependymal cell progeny by molecular markers and electron microscopy after a dorsal funiculus incision revealed that the majority were immunoreactive to Sox9 and vimentin and had an astrocyte-like morphology (Figures 6E, 6H, 6J, 6M, and S10). A smaller subpopulation of the recombined cells expressed GFAP and nestin, but the vast majority of cells with this phenotype were not recombined (Figure 6D, 6H, 6I, and 6M). Recombined GFAP- and nestin-expressing cells were typically located close to the surface of the spinal cord (Figure 6D, 6H, 6I, and 6L), whereas the Sox9- and vimentin-expressing cells were most abundant in the core of the scar tissue (Figures 6L and S10). We conclude that the glial scar is comprised of two different populations of astrocyte-like cells, where the majority of the Sox9+/vimentin+ population derives from ependymal cells and the GFAP+/nestin+ cells are mainly reactive resident astrocytes.

18We further investigated the contribution of ependymal cells to other lineages. None of the recombined cells in the scar tissue had neuronal morphology or were immunoreactive to the neuron-specific epitope NeuN (unpublished data). A population of recombined cells expressed Olig2 (Figure 6F and 6G). The first month after injury, Olig2-expressing recombined cells were scattered throughout the injury site and had an ultrastructural morphology corresponding to immature oligodendrocytes (Figure 6F, 6K, 6L, and 6M). At later time points, Olig2-expressing ependyma-derived cells were excluded from the scar tissue and were restricted to the uninjured tissue that bordered the scar (Figure 6G, 6L, and 6M). Lesions in the lateral funiculus resulted in the generation of ependymal progeny of the same fates as after a dorsal funiculus incision (Figure S8).

19The scar tissue that forms at spinal cord injuries is thought to inhibit axonal growth. Chondroitin sulphate proteoglycans (CSPG) appear to be the principal axonal growth inhibiting molecules in glial scars. Ependyma-derived cells at the injury formed a complementary nonoverlapping pattern with areas that were CSPG immunoreactive (Figure 7A and 7B), indicating that ependymal cell progeny do not contribute to the production of axonal growth-inhibiting CSPG.

20In parallel with the production of axonal growth-inhibiting factors in the glial scar, there is an increase in some axonal growth-promoting molecules, such as the extracellular matrix molecules laminin and fibronectin. In the injury model employed here, axons send sprouts into the scar tissue, mainly during the first month after an injury, and the axons are preferentially associated with areas in the scar tissue that have high levels of laminin. Both laminin and fibronectin immunoreactivity were widely distributed throughout the scar tissue, overlapping both with CSPG-immunoreactive domains and areas occupied by ependyma-derived cells (Figure 7A and 7B). Neurofilament-immunoreactive axons were present in the center of the scar tissue and were often wiggly and oriented in all directions (Figure 7C–7H). This is in contrast to the rostrocaudal orientation of axons seen in the uninjured dorsal funiculus, suggesting that many of the axons present in the scar were severed and sprouting into the scar tissue. Neurofilament-immunoreactive axons were present in domains dominated by ependyma-derived cells, as well as in other areas of the scar where these cells were less abundant (Figure 7C–7H). Axons were often present in direct proximity to ependyma-derived cells (Figure 7D, 7E, 7G, and 7H). The finding that ependyma-derived progeny is not associated with the main scar-associated axonal growth-inhibiting factor, CSPG, together with their proximity to axonal sprouts, argues against these cells being a major factor in glial scar-associated axonal growth inhibition.

21The finding that some ependymal cell progeny displayed a marker profile and ultrastructural morphology suggesting oligodendroglial differentiation (Figure 6) prompted us to characterize these cells further and to address whether they may contribute to axonal remyelination at later time points. Ten months after spinal cord injury, the majority of ependyma-derived progeny are located in the scar tissue that has formed at the injury site, but a substantial number of cells are sparsely distributed in a large area of the intact grey and white matter bordering the lesion (Figure 8A–8C). Most of these cells are Olig2+ and display mature oligodendrocyte morphology with processes that extend along and enwrap myelin basic protein (MBP)-immunoreactive myelin ensheathing axons (Figure 8B–8D).

22Nuclear regions and processes of two ependyma-derived cells were followed in the electron microscope in serial utrathin sections (Figures 9 and S11, and unpublished data). They both displayed a typical mature oligodendrocyte morphology, such as oval nuclei with clumped chromatin, a cytoplasmic matrix that appeared denser than in surrounding astrocytes, a granular endoplasmic reticulum represented by several short cysternae, and tight junctions with adjacent oligodendrocyte processes (Figure 9). Few processes emerged from the cell body, and unlike those of astrocytes, they did not form many branches and did not contain evident fibrils (Figure 9). The processes of the recombined cells could be traced along axons, surrounding their myelin sheaths (Figure 9D). Thus, in addition to the generation of astrocytes, ependymal cells generate myelinating oligodendrocytes.

23Stem cells are notoriously difficult to identify, and their localization in the adult spinal cord has been controversial. We report that ependymal cells constitute the vast majority of cells displaying in vitro neural stem cell properties in the adult spinal cord. Ependymal cells self-renew in vivo, but do not generate appreciable numbers of other cell types under homeostatic conditions. Their normally limited proliferation increases dramatically after spinal cord injury and they then produce oligodendrocytes, and more abundantly, astrocytes that migrate to the site of injury and make a substantial part of the glial scar.

24The immediate descendants of tissue stem cells, progenitor cells with limited self-renewal capacity and/or lineage potential, can in some situations acquire stem cell properties. For example, spermatogonial progenitor cells can regain stem cell function after injury and during aging and forebrain neurospheres may be derived from committed progenitors. It appears unlikely that this would explain the neural stem cell properties displayed by ependymal cells in vitro, as they are not replenished by any other cell type in the adult, but are self-renewing. However, although ependymal cells at the population level display cardinal stem cell features in vivo, such as self-renewal and generation of diverse progeny, it is difficult to study these properties at the single cell level in the tissue, and we cannot conclude that they act as stem cells in vivo.

25In addition to ependymal cells, neural progenitors (expressing NG2, Olig2, and/or Nkx2.2) reside in the white and gray matter of the adult rodent spinal cord. Different studies have suggested that the parenchymal progenitors represent multipotent neural stem cells or more-restricted glial progenitors. Under the standard neurosphere assay conditions employed here, the vast majority of the neural stem cell potential resides within the ependymal population. However, we cannot exclude that other cells contribute, to a comparatively smaller degree, to neurosphere formation under our conditions or that they may display neural stem cell properties under other conditions. The parenchymal progenitors are likely to serve to replace glial cells in the uninjured spinal cord, which we do not find evidence that ependymal cells do. Parenchymal progenitors are rapidly depleted after spinal cord injury, but are later replaced and may participate in the generation of glial cells after injury. It is possible that some of the ependyma-derived Olig2+ cells observed shortly after injury represent regenerated parenchymal progenitors.

26The limited functional recovery typically associated with central nervous system injuries is in part due to the failure of severed axons to regrow and reinnervate their targets. Axonal regeneration is inhibited by scar formation and growth-inhibitory factors associated with myelin and astrocytes. Modulating the responsiveness to axonal growth-inhibitory factors and glial scar formation are attractive strategies to improve functional recovery after central nervous system injuries. The majority of ependyma-derived cells differentiate to astrocyte-like cells after injury and are found in the core of the scar tissue. However, these cells are found in complementary nonoverlapping domains to areas immunoreactive to CSPG, the most important axonal growth inhibitor associated with glial scars. Moreover, axons in the scar tissue, most likely sprouts from severed axons growing into the scar tissue, were frequently found in direct proximity to ependyma-derived cells. This argues that ependyma-derived cells in the scar tissue do not constitute a major impediment to axonal growth, and may even suggest that they support some local sprouting.

27Spinal cord injury results in the loss of oligodendrocytes and demyelination of axons even at some distance to the lesion. Spinal cord injuries are most commonly incomplete in man, leaving spared tissue connecting the spinal cord above and below the lesion, but the function of remaining axons is often compromised due to demyelination. Without insulating sheaths of myelin, spared axons close to, but not directly affected by the injury, become less efficient in their ability to conduct electrical impulses. Moreover, chronically demyelinated axons are vulnerable to degeneration. Axons are remyelinated with time, and this is thought to occur through the generation of new oligodendrocytes by stem or progenitor cells rather than by self-duplication of mature remaining oligodendrocytes. We report here that ependymal cells contribute to the regeneration of oligodendrocytes and remyelination after spinal cord injury. The differentiation pattern of ependymal cells after injury is reminiscent to that seen for in vitro expanded adult spinal cord neural stem cells transplanted to the injured spinal cord. Transplanted adult spinal cord-derived neurospheres improve functional recovery, and if they are forced to generate more oligodendrocytes, functional recovery is further improved. Since ependymal cells are the main source of neurospheres from the adult spinal cord (Figure 3), promoting oligodendrocyte generation from these cells in vivo could potentially improve recovery after spinal cord injury. The development of pharmacological strategies to modulate endogenous stem cells and their progeny may be an attractive alternative to cell transplantation for the treatment of spinal cord injury.

28For Nestin-CreER, we used the enhancer found in the second intron of the rat nestin gene fused to a minimal hsp68 promoter that controls the expression of CreERT2, as previously described. For FoxJ1-CreER, we used a human FOXJ1 promoter fused to a CreERT2 ires-EGFP construct. Transgenic mice were generated at the Karolinska Center for Transgene Technologies by standard procedures utilizing pronuclear injection of CBA × C57BL/six fertilized eggs. Potential founder animals were screened by Southern blot analysis and PCR analysis using a CreERT2-specific fragment as probe or PCR template. Founder mice were bred to wild-type C57Bl/6 mice. Expression of the transgene was analyzed by confocal microscopy of sections stained with anti-Cre antibodies and cell-specific markers. Recombination was induced by five daily intraperitoneal injections of 2 mg of tamoxifen (Sigma; 20 mg/ml in corn oil).

29Adult mice were perfused transcardially with PBS followed by 4% formaldehyde in PBS, spinal cords were post-fixed overnight at 4 °C and then cryoprotected in 30% sucrose. Coronal (14 or 20 μm) or sagittal (20 μm, from ∼9-mm-long pieces) sections were collected alternating on ten slides (8–10 sections per slide). Sections were incubated with blocking solution (10% donkey serum in PBS, with 0.3% Triton-X100) for 1 h at room temperature, then incubated at 4 °C or room temperature in a humidified chamber for 12–48 h with primary antibodies diluted in blocking solution. For MBP staining, sections were first delipidized. For antibodies raised in mouse, the M.O.M. kit (Vector), ABC kit (Vector), and TSA system (Perkin Elmer) were used following the manufacturers' instructions. The following primary antibodies were used: β-galactosidase (1:5,000, rabbit; ICN Biomedicals, or 1:1,000, goat; Biogenesis), BrdU, (1:200, rat; Accurate), CD133 (1:500, rat, clone 13A4; eBioscience), chondroitin sulfate (1:1,000, mouse; Sigma), Cre (1:2,000, mouse; Nordic BioSite), Crocc (1:5,000, rabbit, Root6; gift from T.Li), CXCR4 (1:500, mouse; BD Pharmingen), fibronectin (1:1,000, rabbit; Sigma), GFAP (1:1,000, mouse, clone G-A-5; Sigma), Ki67 (1:1,000, rabbit; Neomarkers), Olig2 (1:500, goat; R&D Systems), laminin (1:1,000, rabbit; Sigma), MBP (1:500, rabbit; Chemicon), musashi-1 (1:2,000, rat, clone 14H1; gift from H.Okano), nestin (1:5,000, rabbit or 1:500, mouse; BD Pharmingen), neurofilament heavy (1:1,000, chicken; Chemicon), PDGFRα (1:500; BD Pharmingen), RC1 (1:200, mouse; DSHB), Sox2 (1:500, mouse; Chemicon, or 1:1,000, goat; gift from J. Muhr), Sox3 (1:500, rabbit; gift from T.Edlund), Sox9 (1:500, goat; R&D Systems), vimentin (1:1,000, chicken; Chemicon). After washing, antibody staining was revealed using species-specific fluorophore-conjugated (Cy3, Cy5 from Jackson, and Alexa 488 from Molecular Probes) or biotin-conjugated secondary antibodies (Jackson). Biotinylated secondary antibodies were revealed using the ABC kit (Vector Labs) with TSA fluorescent amplification kit (Perkin-Elmer). Sections were counterstained with DAPI (1 μg/ml; Sigma). Control sections were stained with secondary antibody alone. Pictures were taken using a Zeiss Axioplan 2, Zeiss Axiovert 200M or a LSM510 META confocal microscope with Zeiss and Openlab (Improvision) software. Image processing and assembly was performed in ImageJ and Photoshop.

30Anaesthetized mice were perfused transcardially with 4% paraformaldehyde in PBS. The spinal cord was dissected out and post-fixed for 4 h. Sections (60 μm) were immunolabled with Cre or β-gal antibodies in 0.1% Triton X100 and 10% donkey serum in PBS. A secondary antibody conjugated to biotin was used with an ABC kit (Vector Labs). In some sections, a fluorescent secondary antibody was used, and sections were labeled with DAPI to map the location of the nuclei of the surrounding cells (Figure S11). The sections were post-fixed in 3% glutaraldehyde in 0.1 M cacodylate buffer (pH 7.4) and in 1% osmium tetraoxide in 0.1 M cacodylate buffer, dehydrated in a graded series of ethanol, stained with uranyl acetate, and embedded in Durcupan resin (Fluka). Serial sections (200 nm) of FoxJ1-CreER spinal cord were used to define the ultrastructural morphology of cells lining the central canal. The complete series of sections from individual cells were traced in 90 sections to characterize the morphology in three dimensions. Serial semi- and ultrathin sections (2 μm and 70 nm, respectively) were used in correlative light and electron microscopic evaluation (CLEM) of the fate of ependyma-derived cells 4 wk after a spinal cord injury (Figure S6). The semithin sections were used to identify immunopositive cells, and the ultrathin sections to show the morphology of the identified cells. Sections (70–200 nm) were placed on Formvar-coated copper grids, counterstained with 2% uranyl acetate and Reynold's lead citrate. Sections were examined in a Tecnai 12 electron microscope (FEI) equipped with 2kx2k TemCam-F224HD camera (TVIPS).

31Spinal cords were dissected and cells dissociated using papain (Worthington). Neurospheres were cultured as described in DMEM/F12 medium supplemented with B27 and EGF and bFGF (both 10 ng/ml). Approximately 200,000 cells were plated in 10-cm cultures dishes, corresponding to a density of 20 cells/microliter, which allows the generation of clonal neurospheres. For assaying self-renewal and multipotentiality, FoxJ1-CreERxR26R (n = 6) and Nestin-CreERxR26R (n = 6) adult mice were administered with tamoxifen intraperitoneally for 5 d with a washout period of 6 d (see Figure 3). Single spheres were manually collected and split into two wells. One well was used for continuous passaging and subsequent neural stem cell differentiation, whereas the other well was used for X-gal staining. For assaying self-renewal, four clonal recombined neurospheres per animal were manually isolated after 12 d of primary neurosphere formation. All recombined neurospheres were serially passaged eight times. In vitro differentiation by growth factor withdrawal for 10 d was assessed in passage 3 and passage 6 by staining as described above for βIII-tubulin (Tuj1, 1:1,000; Covance), GFAP (1:5,000; DAKO), and O4 (1:200; Chemicon).

32Spinal cords were dissected from FoxJ1-CreER mice and dissociated using papain (Worthington) and DNase in 1× HBSS at 37 °C for 1 h. Ovomucoid inhibitor (Worthington) was added and cells were collected by centrifugation at 300g for 5 min. Cells were resuspended in Leibovitz-15/B27 with 7AAD, which labels dead cells. Single GFP+ (based on the ires-GFP signal), 7AAD− cells were isolated using a FACSAria (BD). Singlets were identified based on forward scatter width (FSC-W) versus forward scatter height (FCS-H). Single cell sorting and GFP fluorescence was confirmed by microscopic examination.

33Mice were anesthetized with 2.5% Avertin, and the dorsal funiculus at mid-thoracic level was cut transversely and was extended rostrally with microsurgical scissors to span one segment. In other animals, the lateral funiculus was cut transversally and the lesion extended rostrally to span one segment.

34BrdU (1 mg/ml and 1% sucrose, exchanged every 3–4 d) was administered in the drinking water to label dividing cells.

35In order to correlate recombination frequency in neurosphere cultures to the in vivo recombination frequency of spinal cord tissue, the ratio between CreER+ and β-gal+ cells (n = 60) was quantified in a small postfixed biopsy from the same spinal cords used for neurosphere cultures.

36The percentage of BrdU+ ependymal cells was obtained from three animals treated for 4 wk with BrdU in the drinking water (3–5 coronal sections/animal analyzed). The total number of cells per section was obtained counting all nuclei stained with DAPI.

37The in vivo recombination frequency was assessed by counting the number of recombined cells over the total number of ependymal cells (Vimentin+) from five coronal sections per animals at 2 d (4 animals: 2 Nestin-CreERxR26R, 2 FoxJ1-CreERxR26R) and 10 mo (3 animals: 3 FoxJ1-CreERxR26R) after tamoxifen treatment (Figure 4E).

38The contribution of recombined cells at the site of injury was established by measuring the relative area occupied by β-gal+ cells within the epicenter of the lesion (using ImageJ software) of 3 FoxJ1-CreERxR26R animals (2 sagittal sections per animal) 2 wk after spinal cord injury.

39The cell fate distribution of ependyma-derived progeny was obtained by scoring recombined cells positive for Olig2, GFAP or Sox9 in either coronal (3–5 sections per animal) or sagittal (1–2 sections per animal) sections encompassing the lesion site from 1 month (3 animals: 1 Nestin-CreERxZ/EG, 2 FoxJ1-CreERxR26R) and 8–10 mo (4 animals: 1 Nestin-CreERxZ/EG and 3 FoxJ1-CreERxR26R) after spinal cord injury (Figure 6M).

40The frequency of proliferation of ependymal cells and their progeny was assessed by counting the number of Ki67+ recombined cells over the total number of recombined cells from three segments (rostral to, caudal to, and at the injury site; average of 15 coronal sections, or 300 recombined cells, per segment analyzed) from 2 FoxJ1-CreERxR26R animals 4 d after spinal cord injury (Figure S7C).