A Mesenchymal-Like ZEB1+ Niche Harbors Dorsal Radial Glial Fibrillary Acidic Protein-Positive Stem Cells in the Spinal Cord§


  • Jean-Charles Sabourin,

    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
  • Karin B. Ackema,

    1. Growth & Development, Biozentrum, University of Basel, Basel, Switzerland
    Search for more papers by this author
  • David Ohayon,

    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
  • Pierre-Olivier Guichet,

    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
  • Florence E. Perrin,

    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
  • Alain Garces,

    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
  • Chantal Ripoll,

    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
  • Jeroen Charité,

    1. Department of Cell Biology, Erasmus Medical Center, Rotterdam, The Netherlands
    Search for more papers by this author
  • Lionel Simonneau,

    1. Laboratoire interdisciplinaire de recherche en didactique et formation, University Montpellier, Montpellier, France
    Search for more papers by this author
  • H. Kettenmann,

    1. Max-Delbrück-Centrum für Molekulare Medizin (MDC), Berlin, Germany
    Search for more papers by this author
  • Azel Zine,

    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
  • Alain Privat,

    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
    • A.P. and J.P.H. contributed equally to this work.

  • Jean Valmier,

    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
  • Alexandre Pattyn,

    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
  • Jean-Philippe Hugnot

    Corresponding author
    1. Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    • Institut National de la Santé et de la Recherche Médicale (INSERM) U583, Physiopathologie et Thérapie des déficits sensoriels et moteurs, Institut des Neurosciences de Montpellier, Hôpital St. ELOI, Montpellier, France
    Search for more papers by this author
    • A.P. and J.P.H. contributed equally to this work.

    • Tel: 00 33 4 99 63 60 08; Fax: 00 33 4 99 63 60 20

  • Author contributions: J.C.S., K.B.A. P.O.G., F.E.P., and C.R.: conception and design; D.O., J.C., L.S., H.K., A.Z., and A.P.: provision of study material; A.G.: collection and/or assembly of data; A.P. and J.V.: financial support; J.P. H.: conception and design, manuscript writing, final approval of manuscript.

  • Disclosure of potential conflicts of interest is found at the end of this article.

  • §

    First published online in STEM CELLS EXPRESS September 25, 2009.


In humans and rodents the adult spinal cord harbors neural stem cells located around the central canal. Their identity, precise location, and specific signaling are still ill-defined and controversial. We report here on a detailed analysis of this niche. Using microdissection and glial fibrillary acidic protein (GFAP)-green fluorescent protein (GFP) transgenic mice, we demonstrate that neural stem cells are mostly dorsally located GFAP+ cells lying ependymally and subependymally that extend radial processes toward the pial surface. The niche also harbors doublecortin protein (Dcx)+ Nkx6.1+ neurons sending processes into the lumen. Cervical and lumbar spinal cord neural stem cells maintain expression of specific rostro-caudal Hox gene combinations and the niche shows high levels of signaling proteins (CD15, Jagged1, Hes1, differential screening-selected gene aberrative in neuroblastoma [DAN]). More surprisingly, the niche displays mesenchymal traits such as expression of epithelial-mesenchymal-transition zinc finger E-box-binding protein 1 (ZEB1) transcription factor and smooth muscle actin. We found ZEB1 to be essential for neural stem cell survival in vitro. Proliferation within the niche progressively ceases around 13 weeks when the spinal cord reaches its final size, suggesting an active role in postnatal development. In addition to hippocampus and subventricular zone niches, adult spinal cord constitutes a third central nervous system stem cell niche with specific signaling, cellular, and structural characteristics that could possibly be manipulated to alleviate spinal cord traumatic and degenerative diseases. STEM CELLS 2009;27:2722–2733


Compared with brain neural stem cell niches (subventricular zone [SVZ]) and hippocampus), very little is known about stem cells in the adult spinal cord [1–3]. Bona fide neural stem cells able to form long-term passageable neurospheres appear to be located only around the central canal [4]. After spinal cord trauma and also in various degenerative diseases, these migrate to the lesion site and differentiate into both astrocytes and oligodendrocytes [5]. This may contribute to the exacerbation of the glial scar or inversely to axonal regeneration. Formation of neuronal-like cells and oligodendrocytes from the adult spinal cord both in physiological and pathological situations has also been reported [6–10]. The precise identity and diversity of stem/progenitor cells around the central canal is not yet known. Single cell purification and studies with transgenic animals expressing green fluorescent protein (GFP) in cells around the canal led to the conclusion that neural stem cells are glial fibrillary acidic protein (GFAP)-negative ependymal cells [11, 12]. Stem cells in the adult spinal cord can also be sorted based on their high aldehyde dehydrogenase activity and these can then be induced to differentiate in vitro to form motoneurons thus opening new approaches to treat motoneuron disease [13]. We recently reported the existence of a Sox2+ Nestin+ stem/progenitor cell niche around the human central canal cord [14].

The two brain neural stem cell niches have been extensively described but on the contrary the cellular composition, organization, and molecular signalings in the spinal cord niche remain ill-defined. Here we report a detailed analysis of this third central nervous system (CNS) niche and show an unexpected cellular and molecular complexity. Neural stem cells were accurately identified as mostly dorsal radial GFAP+ cells lying in a multicellular niche displaying mesenchymal features. These cells maintain correct rostro-caudal Hox gene pattern and require the epithelial-mesenchymal transition transcription factor zinc finger E-box-binding protein 1 (ZEB1) for their expansion in vitro. A better knowledge of this niche may ultimately allow its manipulation, permitting endogenous regeneration of the spinal cord.


Spinal Cord Dissection and Neural Stem Cell Culture

Mice were handled in compliance with French regulations. Spinal cords from 12- to 15-week-old OF1 wild-type (Janvier, Le Genest-Saint-Isle, France) or transgenic (hGFAP-GFP [15]) mice euthanized with CO2 were dissected and rapidly processed for neurosphere assay as described in Johansson et al. [12] without the bovine serum albumin step. For flow cytometry, erythrocytes were lysed between the dissociation and sucrose centrifugation steps with a solution containing 0.15 M NH4Cl, 1 mM KHCO3 and 0.1 mM Na2EDTA. Neural stem cells were cultured as described in Deleyrolle et al. [16] and Dromard et al. [17]. Neurosphere assays were performed with a cell density of 1 cell per microliter in T25 or T75 flasks. Flask manipulation was minimized to decrease neurosphere aggregation. Using mixtures of fluorescent (actin-GFP) and nonfluorescent neurospheres, we checked that under these conditions the vast majority of spheres were clonal. The total number of neurospheres formed was assessed 10 days later by carefully microscopic scanning of the entire flask. For the neurosphere assay using spinal cord sections, lumbar spinal cord was sliced with a tissue chopper (McIlwain; Mickle Laboratory Engineering Co., Guildford, U.K., http://www.micklelab.co.uk) set at 750 μm and slices were collected in Hanks' balanced salt solution (HBSS). Under a binocular microscope, each section was microdissected in two or three parts (Fig. 1) using a diamond blade (Sapphire knife 0.22 mm, 15°; World Precision Instruments, Sarasota, FL, http://www.wpiinc.com) or the central canal region was removed by punching with a fire-reduced Pasteur pipette (Fig. 1). Spinal cord fragments were then incubated in 250 μl HBSS, 30 μl trypsin 13 mg/ml, 30 μl hyaluronidase 7 mg/ml, 7 μl DNaseI 10 mg/ml, and 13 μl kinureic acid 4 mg/ml for 30 minutes at 37°C and dissociated by pipetting with yellow tips. The cell suspension was filtered on a 40-μm sieve (BD, Franklin Lakes, NJ, http://www.bdbiosciences.com) and plated in neurosphere medium at clonal density (1 cell per microliter).

Figure 1.

Proliferation in the niche and localization of spinal cord neural stem cells.

(A): Example of a Ki67+ cell (red) in the niche. (B): Average number of Ki67+ cells per section (13-μm thick, 32 sections per animal) in the niche from 0–13 weeks after birth. Values are means ± standard error of mean (SEM) (n = 96 sections from three animals). (C): Spinal cord length from 0–13 weeks after birth measured with a ruler. Values are means ± SEM (n = 3 spinal cords analyzed). (D): Photographs of spinal cord 1 or 9 weeks after birth (scale bar = 5 mm). (E, G, I): Schematic representations of lumbar spinal cord microdissections. (E): An approximately 550-μm diameter central fragment (red circle) was dissected from the parenchyma and both parts were processed for neurosphere formation. (G): Spinal cords were cut in half. (I): A narrow dorsal medial strip (3) was dissected. (F, H, J): Number of neurospheres formed per 1,000 cells seeded from dissections shown in (E, G, I). Values are means ± SEM (n = 7 flasks) of one experiment representative of three completely independent experiments. For experiment in (G), the absolute numbers of neurospheres (deduced from the total number of cells obtained in the dorsal and ventral parts × frequency of neurosphere formation) were 84,000 in the dorsal versus 36,000 in the ventral part. For experiment in (I), the absolute numbers of neurospheres were 1,500, 5,500, and 22,690 for the three indicated parts. Long dorsal GFAP+ cells are depicted in green in (E, G, I) to indicate that they may be damaged by the excision procedure in (E) but preserved in (G) and (I). (K): Diameter of neurospheres derived from the different spinal cord regions shown in (I). Dorsal medial-derived neurospheres are larger. Number of neurospheres measured from regions 1, 2, and 3 are 21, 9, and 30, respectively. (L): Multipotency of dorsal medial neurospheres. After differentiation for 4 days, most individual neurospheres show the presence of three cell types: astrocytes (GFAP; green, yellow arrow), oligodendrocytes (O4; blue, white arrow), and young neurons (Dcx; red, arrowhead). Abbreviations: C, central canal; D, dorsal; Dcx, doublecortin protein; GFAP, glial fibrillary acidic protein; P, parenchyma; V, ventral.

Statistical Analysis

Tests were performed using one-way analysis of variance. Simple, double and triple asterisks indicate, respectively, 0.05%, 0.01% and 0.001% statistically significant differences.


Proliferation in the Central Region Ceases Around 13 Weeks and Correlates with Postnatal Spinal Cord Extension

As the postnatal development of the central canal region has been barely explored, we started by exploring the rate of proliferation from 0–13 weeks after birth using immunodetection of Ki67+ cells (Fig. 1A). We found a progressive decrease of proliferating cells around the central canal until 9 weeks, after which proliferation was no longer detected (Fig. 1B). This was confirmed by injecting bromodeoxyuridine (BrdU) for 4 days every 4–8 hours and counting the number of BrdU+ cells. No BrdU+ cells were found around the canal in 42 sections corresponding to 500 μm in thickness, whereas positive cells were constantly found in the parenchyma. To test the possibility that postnatal central canal proliferation is linked to postnatal spinal cord extension, spinal cord length was measured during the 13 postnatal weeks. As shown in Figure 1D, spinal cord length increases 2-fold after birth and a clear opposite correlation exists between cellular proliferation and extension (Fig. 1C). This substantiates the involvement of the central canal proliferation in the increase of spinal cord length after birth and shows that this region is not or hardly proliferative in adult.

Spinal Cord Neural Stem Cells Are Principally Located Dorsally

Although proliferation in the central region is completed around 13 weeks, neural stem cells are still present after this age as evidenced by the formation of neurospheres by approximately 0.2% of clonally seeded cells (not shown). We specified the location of these neurosphere-forming cells using microdissection. As the lumbar spinal cord has more neural stem cells than at other levels [1], our study was concentrated on this region. Using a fire-reduced Pasteur pipette (diameter 540 μm) as a puncher on adult spinal cord sections, we first confirmed that neurosphere-forming cells are essentially located around the central canal (Fig. 1E, 1F). Microdissection was then used to divide spinal cord sections into dorsal and ventral regions (Fig. 1G, 1H), which were then processed for neural stem cell assay. Surprisingly, the majority of neurosphere-forming cells were found to be located dorsally. Dissecting further down to a dorsal strip including the central canal showed the highest enrichment for neurosphere-forming cells (Fig. 1I, 1J). After 2 weeks, the neurospheres reached a size larger than 500 μm, suggesting that they originate from stem rather than progenitor cells [18]. Upon differentiation (Fig. 1L; supporting information Fig. 3F), the neurospheres generate astrocytes, O4+ oligodendrocytes, and young Dcx+ neurons. The latter were GABAergic as indicated by GABA immunodetection (data not shown). Neurosphere formation and multipotentiality were maintained even after nine passages (supporting information Fig. 3F). Neurospheres arising from the other regions might be due to contamination or to the presence of few non dorsal-medial located neurosphere-forming-cells. Such neurospheres are smaller (Fig. 1K), suggesting that they may be derived from progenitors rather than stem cells [18]. Of note, fourfold more neurospheres were obtained from 1,000 cells of the dorsal medial part (Fig. 1J) than from the central part (Fig. 1F), indicating that punching the central region for excision may damage part of the neurosphere-forming cells, as would be expected if these possessed long radial processes (see below).

The Central Canal Region Is Composed of Several Cell Types

The predominant dorsal location of neural stem cells prompted us to investigate the identity of these cells using a series of antibodies known to label different immature cells of the neural lineage. Simple histological examination of the adult mouse central canal shows that this region is not composed of a single layer of cells but like in humans contains several cell types differing in their morphology and position (ependymal and subependymal) (Fig. 2A). Here, the term “ependymal” indicates that the cell nucleus is within the first row of cells. Most cells around the canal express the immature intermediate filament proteins vimentin (Fig. 2B) and nestin (not shown), the Sox2 and Sox9 transcription factors (Fig. 2E, 2F), the polysialated form of neural cell adhesion molecule (PSA-NCAM, Fig. 2D), and Cadherin 13 (also known as t-cadherin, Fig. 2H), an adhesion molecule expressed by various neural precursor cells in vitro [19]. Cadherin 13 levels are however higher in young animals than in adults (supporting information Fig. 1A). Sox4, a marker of the oligodendrocytic lineage, was not detected in adults; however we found positive cells in younger animals (Fig. 2G). Anti-CD133, a well-known marker of neuroepithelial and neural stem cells, stains the apical side of central canal cells (Fig. 2C). The endothelin receptor B (EdnrB), which is strongly expressed in spinal cord neurospheres [16], is expressed by most cells around the canal but, significantly, dorsal cells extending a process toward the pial surface are strongly stained (Fig. 2I). Anti-brain lipid binding protein (BLBP; also known as fatty acid binding protein 7), a radial glial marker, labels a small population of cells with radial morphology most often situated in the dorsal part (Fig. 2J). Staining for GFAP, an astrocytic/neural stem cell marker, labels several cell types (Fig. 2K, 2L; supporting information Fig. 1B, 1C, 1H). Of note, these GFAP+ cells are detected with a polyclonal antibody and only a fraction can be detected with a monoclonal antibody (not shown). A group of cells extending very long GFAP+ processes toward the pial layer were always observed in the dorsal part (Fig. 2L; supporting information Fig. 1B, 1C). These cells were either in the ependymal layer adjacent to the canal (supporting information Fig. 1B, 1C) or further above in a subependymal location (supporting information Fig. 1B, 1C, 1H) when they sent a process toward the canal (Fig. 2K, 2L; supporting information Fig. 1H). Interestingly a fraction of these cells express the radial glia marker BLBP (Fig. 2K). Ependymal and subependymal GFAP+ are also occasionally observed in the lateral and ventral part of the canal (Fig. 2L; supporting information Fig. 1C). CD15 (also known as Lewis X antigen or stage-specific embryonic antigen-1), a marker for various stem cells including neural stem cells [20], stained the central spinal cord region dissymmetrically with the dorsal region around the canal displaying a higher diffuse staining than the ventral part (Fig. 2P). Central canal cells were not labeled but remarkably a fraction of dorsal GFAP+ cells and their luminal extension were labeled (Fig. 2Q). Double-labeled BLBP and CD15 could also be observed (not shown).

Figure 2.

Diversity of cells in the spinal cord niche.

(A): A lumbar spinal cord section stained with toluidine blue. Arrows indicate subependymal cells and arrowhead points to a cell with neuronal features. (B–Q): Immunofluorescence detection of indicated marker on adult spinal cord sections. (C): CD133 labeling is confined mainly to the apical pole of cells around the canal (arrowhead). (G): Sox4 labeling (arrow) is often observed in twin cells, suggesting recent mitosis. (I): The EdnRB stains central canal cells and long dorsal radial processes (arrow). (J): BLBP (red) stains a few dorsal cells with a radial process. (K): High magnification of the BLBP+ cell in (J). This cell coexpresses GFAP (middle picture). (L): GFAP staining shows processes contacting the lumen (arrows) and a dorsal group of GFAP+ cells (arrowhead). (M, N): β-III-tubulin+ and Dcx+ neuronal-like cells extending a process into the lumen (arrow). A bulge (arrowheads) is often observed at the tip of the processes. (O): Nkx6.1 cells (red nuclei) are located mostly subependymally. These cells coexpress Dcx (inset in [N]). (P): CD15 staining is present mostly dorsally. (Q): High magnification of box in (P). CD15 stains a fraction of GFAP+ dorsal cells (arrowhead) and some luminal extension (arrow). Ventral central canal is at the bottom on all images. Scale bar = 10 μm. Abbreviations: BLBP, brain lipid binding protein; Dcx, doublecortin protein; EdnrB, endothelin receptor B; GFAP, glial fibrillary acidic protein; PSA-NCAM, polysialated form of neural cell adhesion molecule; Vim, vimentin.

Staining for β-III-tubulin (Fig. 2M) revealed the presence of neuronal cells lying close to the ependymal region often extending a process toward the lumen. These neurons express Dcx, a marker for young neurons, indicating that they may retain immaturity characteristics (Fig. 2N). Interestingly, these cells show a large bulge at the tip of their process in the canal lumen (Fig. 2M, 2N). The presence of these neurons was confirmed by the observation of fluorescent cells using a transgenic mouse expressing GFP under a Dcx promoter (supporting information Fig. 1D). To gain further insight on this neuronal subtype, we colabeled Dcx with the homeoprotein Nkx6.1, as during spinal cord development, the central canal region is derived from a ventral domain expressing this gene [21]. Nkx6.1 cells are mostly subependymal (Fig. 2O) and coexpressed Dcx, suggesting a ventral developmental origin of these central canal neurons (Fig. 2N, inset). Nkx6.1+ rarely stained GFAP+ dorsal cells (data not shown). Since we detected postnatal proliferation within the niche, we evaluated the possibility that these neuronal cells were born postnatally. Three-week-old mice were injected with BrdU for 5 days, then the presence of Dcx or β-III-tubulin neurons was assessed after 3 or 7 days. No BrdU+ neuronal cells were detected, suggesting that these cells are generated during the embryonic or the early postnatal period. As in the lumbar central canal, a diversity of cells (CD133+, GFAP+, and Nkx6.1+) is also observed at the cervical level (supporting information Fig. 1E–1G).

We further examined the cellular diversity around the canal with electron microscopy and immunogold techniques (supporting information Fig. 2). There is a clear ventral/dorsal dissymmetry of the central canal organization with the most dorsal part showing a higher cell density (supporting information Fig. 2A) often organized in two or three rows (supporting information Fig. 2C). In agreement with immunofluorescence analysis, cells with long radial extensions containing intermediate filament recognized by GFAP antibodies were observed in this region (supporting information Fig. 2B). Subependymal cells were common all around the central canal (supporting information Fig. 2D). These cells had either similar ultrastructural features to ependymocytes or were clearly distinct (supporting information Fig. 2D). This analysis provides evidence for a dorsal-ventral dissymmetry of the central canal and the presence of several ependymal and subependymal cell types.

Spinal Cord Neural Stem Cells Are Radial GFAP+ Cells

Since in both humans and mice, many data indicate that neural stem cells in the adult brain are GFAP+ cells closely associated with the ependymal cell layer, we investigated whether the GFAP+ cells around the canal behaved like neural stem cells in vitro using a human GFAP promoter-GFP transgenic mouse [15]. Around the canal of the lumbar spinal cord, most bright GFP+ cells (>80%) were found in the dorsal part (Fig. 3A, 3B, supporting information Fig. 1H), located in the ependymal layer in direct contact with the lumen (Fig. 3A), or located subependymally, sending a process toward the lumen (Fig. 3B). GFP further revealed the presence of very long radial processes originating from the GFAP+ dorsal cells (Fig. 3A). All GFP+ cells around the central canal were found to be GFAP+ (52 of 52 cells) (Fig. 3B; supporting information Fig. 1H), but inversely a fraction of GFAP+ cells are GFP. To examine whether GFP+ cells could generate neurospheres, after spinal cord dissection GFP+ and GFP cells were separated by flow cytometry and seeded at clonal density (supporting information Fig. 1I). We found that 90% of the neurospheres are formed in the GFP+ fraction (Fig. 3C). These neurospheres could be cultured for at least 10 passages (not shown). In all neurospheres (179 of 179) cultured for only a few days at clonal density at least one or several strongly expressing GFAP+ cells were found (Fig. 3D, D′), some of which expressed CD133 (supporting information Fig. 3A). We confirmed the presence of GFAP+ cells by deriving neurospheres from the hGFAP-GFP mouse. We found that 104 of 158 neurospheres contain one or several GFP+ cells coexpressing GFAP (Fig. 3E). Spheres that do not contain GFP+ cells may be derived from the subpopulation of GFAP+ GFP cells mentioned above. The observation of both GFAP+ and GFAP cells within a clonally expanded neurosphere suggests cellular diversity, probably resulting from asymmetric division as would be expected for neural stem cells. The vast majority of GFAP cells were found to express Olig2, Map2ab, CD15, and Dcx markers, presumably representing bipotent neuronal-glial progenitors (supporting information Fig. 3B–3E). Of note, all neurospheres were found to strongly express the markers we observed in the niche, namely Cadherin 13, Sox4, Nkx6.1, DAN (see below), BLBP, PSA-NCAM, and EdnRB, (Fig. 3F–3L). All together, these results demonstrate that the adult spinal cord neural stem cells are mostly dorsal medial GFAP+ cells directly or indirectly in contact with the central lumen.

Figure 3.

Identification of spinal cord neural stem cells.

(A, B): GFP+ cells around the canal of hGFAP-GFP mouse. (A): Example of three dorsal GFP+ cells abutting the central canal. These cells are located ependymally (inset). Note the presence of long tortuous GFP dorsal radial processes (arrow) and, in the ventral part, a group of weakly GFP+ cells (arrowhead). (B): Example of one dorsal GFP+ cell located subependymally (box) sending a process into the lumen (arrowhead). This cell coexpresses GFAP (red, left-hand image). (C): Neurospheres are derived mostly from GFP+ cells. After dissection of hGFAP-GFP spinal cord GFP+ and GFP cells were separated by flow cytometry (supporting information Fig. 1I) and seeded at clonal density. Values are means ± SEM (n = 7 flasks) of one experiment representative of four completely independent experiments. (D, D′): All primary derived neurospheres contain one or several strongly GFAP+ cells (green). Two examples are shown. (E): Most of primary neurospheres from the hGFAP-GFP mice contain one or several GFP+ cells that coexpressed GFAP (arrows). Inversely, some GFAP+ cells did not express GFP (arrowhead). (F–L): Immunofluorescence detection of indicated antigen in undifferentiated neurospheres. Note that after centrifugation and attachment on coverslips for immunofluorescence analysis, undifferentiated neurospheres can partially disaggregate giving an aspect of dissociated cells (F, H). Scale bars = 10 μm. Abbreviations: BLBP, brain lipid binding protein; DAN, differential screening-selected gene aberrative in neuroblastoma; EdnrB, endothelin receptor B; GFAP, glial fibrillary acidic protein; GFP, green fluorescent protein; PSA-NCAM, polysialated form of neural cell adhesion molecule.

Elevated Signalings and Mesenchymal Features of the Niche

Various adult stem cells are commonly maintained in their niche through activation of key signalings. In the central canal, we evaluated the presence of such local stem cell signalings by detection of the HES1 protein, a key effector of Sonic hedgehog and Notch pathways. HES1 is expressed in the nuclei of most cells around the canal (Fig. 4A), but also frequently in the cytoplasm, notably in long dorsal processes (Fig. 4B). This nuclear/cytoplasmic localization may reflect regulatory shuttling of the protein. The Jagged1 protein, one of the ligands of the Notch pathway, is also readily observed on the luminal side of the cells around the canal (Fig. 4C). Several stem cell niches are characterized by the presence of bone morphogenetic proteins (BMPs) and anti-BMP signaling molecules, such as BMP4 and noggin in the SVZ. In the adult spinal cord, we found that DAN, a BMP inhibitor highly expressed in spinal cord neurospheres [16], is expressed by, depending on the section, most or a restricted number of central canal cells (Fig. 4D). Double labeling indicated that dorsal GFAP+ cells express DAN (Fig. 4D). In addition to canonical stem cell signalings, recent studies have shown the implication of transcription factors of the epithelial-mesenchymal transition (EMT) in the maintenance of stem cells both in vertebrates and invertebrates [22]. This prompted us to examine the expression of ZEB1/2, SNAI1/2, and TWIST1/2 in neurospheres and in the niche. As shown in Figure 4E, reverse-transcription polymerase chain reaction (PCR) analysis shows the expression of SNAI2, ZEB1, and ZEB2 genes in spinal cord neurospheres. Expression of ZEB1 and ZEB2 was also demonstrated at the protein level by immunofluorescence (Fig. 4F, 4G). The mesenchymal feature of cultured neural stem cells was further confirmed by the expression of vimentin, fibronectin, CD44, and smooth muscle α-actin (SMA also known as actin α 2), four proteins commonly found in mesenchymal cells (Fig. 4H–4K and PCR on Fig. 4E). Upon neurosphere differentiation, the majority of GFAP+ astrocytic cells are SMA+ (Fig. 4L). In the spinal cord, the majority of cells around the canal were found to express CD44, SMA, and ZEB1 (Fig. 4M–4O), whereas ZEB2 was not detected. Dorsal-medially or ventral-medially located cells tended to be stained more strongly for ZEB1 and dorsal GFAP+ cells were intensely labeled (Fig. 4O, 4P). ZEB1 expression was also examined in the adult brain SVZ (supporting information Fig. 4A, 4B). The majority of GFAP+ cells, considered to be stem cells [23], were positive, whereas only a fraction of PSA-NCAM+ young neuronal cells expressed ZEB1 at a detectable level.

Figure 4.

Elevated signaling and mesenchymal features of the spinal cord niche.

(A, B): Hes1 staining. Hes1 protein is mostly nuclear (A, red) or in the cytoplasm (B, arrowhead), notably in long dorsal processes (B, arrow). (B): A magnification of white box is presented on right-hand image. (C): Jagged1 staining (red) is confined to the apical part of central canal cells. (D): Cells in the niche express DAN (red), notably the GFAP+ cells (green) on right-hand image. (E): Reverse-transcription polymerase chain reaction expression analysis of indicated mRNA in neurospheres with (+) or without (−) reverse transcriptase during cDNA synthesis. Sizes in base pair. (F): ZEB1 staining was heterogeneous in neurosphere cells: absent (yellow arrow), nuclear (arrow), or cytoplasmic (arrowhead), whereas in (G), ZEB2 was mainly cytoplasmic. (H–K): Neurospheres express high level of vimentin, fibronectin, CD44, and SMA. (L): Immunocytochemistry on differentiated neurospheres showing cells coexpressing the astrocytic marker GFAP (green) and SMA (red). Most of cells around the canal stained for CD44 (M, red staining on luminal side), SMA (N, red), ZEB1 (O, red). ZEB1 is often more strongly expressed in the dorsal part (arrow in [O]). GFAP+ dorsal cells display intense ZEB1 staining (P). SMA staining was also observed with a polyclonal and another monoclonal antibody (clone O-N-5) (data not shown). Scale bars = 10 μm. (Q): Velocity of indicated cells measured by videomicroscopy during 6 hours. NSC and fibroblast velocity did not differ significantly. Abbreviations: bp, base pair; DAN, differential screening-selected gene aberrative in neuroblastoma; Fibro, primary mouse skin fibroblasts; GFAP, glial fibrillary acidic protein; NSC, mouse adult spinal cord neurosphere cells; Sma, smooth muscle α-actin; Snai2, Snail homolog 2; Vim, vimentin; Zeb, zinc finger E-box-binding protein.

Collectively, these data provide evidence for the maintenance of several typical developmental signalings and more unexpectedly of EMT genes in the spinal cord niche and in cultured neural stem cells. Compared with epithelial cells, mesenchymal cells are characterized by a high ability to migrate in vivo and in vitro. We thus assessed this mesenchymal feature on cultured adult spinal cord neural stem cells by plating neurosphere cells on laminin and measuring the cellular mobility by videomicroscopy. Primary mouse skin fibroblasts were used as a positive control for the mesenchymal phenotype, whereas HeLa and HEK293 cells were used as two control lines for epithelial phenotype. As shown on Figure 4Q and supporting information Videos V1–V4, we found that neurosphere cells were as motile as fibroblasts, whereas as expected, epithelial cells were nearly immobile within this studied time frame.

Critical Role of ZEB Proteins in Spinal Cord Neural Stem Cells In Vitro

The ZEB transcription factors constitute a family of transcriptional factors acting as essential regulators of the epithelial to mesenchymal transition [22]. They are also involved in the regulation of apoptosis, proliferation, senescence, and cellular polarity. As their expression in spinal cord neural stem cells has not been reported, we explored their role in these cells by generating a transgenic mouse conditionally expressing DB-ZEB, a dominant negative form of ZEB proteins [24], and GFP (Fig. 5A). DB-ZEB contains only the DNA-binding domain of ZEB1 and blocks the activity of both ZEB1 and ZEB2. Translation of DB-ZEB and GFP is blocked by an upstream loxed-stop cassette that can be removed by cre-mediated recombination. Neurospheres were derived from the adult spinal cord of DB-ZEB animals and no overt proliferation deficit was observed, as expected since no expression of DB-ZEB was detected in these cultures (data not shown). Recombination and DB-ZEB expression were performed in this culture by infecting with a canine adenovirus (CAV) expressing cre recombinase (CAV-cre). More than 75% of the cells were infected and recombined as measured by the presence of GFP+ cells (Fig. 5B). The capacity of this culture to form new clonal neurospheres was compared with culture infected with control CAV-GFP virus. Figure 5C and 5D show that the expression of DB-ZEB causes a massive reduction of the neurosphere number and a concomitant reduction of the total number of cells. As a control, neurosphere cells derived from nontransgenic animals and infected with CAV-cre and CAV-GFP show no significant differences for the formation of new neurospheres (Fig. 5E), demonstrating the absence of cre recombinase-induced toxicity. To examine how DB-ZEB affects cells, during the 3 days following infection, the number of apoptotic cells was measured by terminal deoxynucleotidyl transferase-mediated dUTP-biotin nick-end labeling. Figure 5F shows that CAV-cre infection triggers a rapid and drastic increase of apoptotic cells, suggesting an essential role of ZEB transcription factors in the maintenance of spinal cord neural stem cells.

Figure 5.

Critical role of ZEB proteins for survival of cultured neural stem cells. (A): Schematic representation of cre-inducible ZEB dominant negative (DB-ZEB) construction injected to obtain the DB-ZEB mouse. Upon recombination, the loxed Stop cassette is removed allowing translation of an IRES-mediated bicistronic DB-ZEB and GFP messenger. (B): Immunodetection of GFP after infection with CAV-cre adenovirus in DB-ZEB transgenic neurosphere cells. For this experiment, neurospheres were plated on adherent substrate. IRES-mediated GFP expression was too low to be detected by direct fluorescence. (C, D): Histograms show the number of neurospheres and the total number of cells obtained 7 days after seeding 1,000 noninfected or infected cells in T25 flasks. In this experiment, neurospheres were derived from adult spinal cord of DB-ZEB mice, passaged four times, then infected with CAV virus as described in Methods. This experiment is representative of three independent experiments. Value given are means ± standard error of means (n = 7 flasks). (E): Control experiment of (C) performed with neurospheres derived from nontransgenic mice showing that CAV-cre virus is not toxic to the cells. Value given are means ± standard error of means (n = 7 flasks). (F): Percentage of TUNEL+ cells after infection with CAV-GFP and CAV-cre. The latter induces massive cell death of neurosphere cells. Value given are means ± standard error of means (n = 3 coverslips). Abbreviations: CaV, canine adenovirus; CaV-cre, CaV expressing cre recombinase; CaV-GFP, CaV expressing green fluorescent protein; GFP, green fluorescent protein; h, hours; TUNEL, terminal deoxynucleotidyl transferase-mediated dUTP-biotin nick-end labeling.

Maintenance of a Hox Gene Pattern in Adult Spinal Cord Neural Stem Cells Along the Rostro-Caudal Axis

Recent data have demonstrated that the SVZ harbors a heterogeneous population of neural stem cells derived from several embryonic regions that generate specific neuron subtypes upon differentiation ([23] for review). This diversity of stem cells relies on the adult maintenance of different embryonic patterning genes such as homeobox genes. During spinal cord development, the regionalization of the neural tube results from expression of different rostro-caudal combinations of genes of the large Hox family [25]. It is not known whether these genes are still expressed in adult spinal cord neural stem cells and whether stem cells from different rostro-caudal levels would maintain regional fetal combinations of Hox genes. This was explored by growing neurospheres derived from the cervical, thoracic, and lumbar levels for four passages and assessing the expression of the 39 Hox genes by semiquantitative PCR. For comparison, Hox gene expression was also measured in the three adult spinal cord levels. We found that Hox genes are differentially expressed in the adult spinal cord, with a clear propensity for genes located toward the 5′ end of a given cluster to be uniquely or more highly expressed in the adult lumbar region (see for instance Hox a13, b13, c13, d13) (Fig. 6). In neurospheres, Hox genes were expressed and their rostro-caudal pattern was, overall, very well conserved. Whereas some genes expressed in the lumbar spinal cord were not detected in lumbar-derived neurospheres (a13, d11, d13), all other genes expressed only in lumbar spinal cord were expressed only in lumbar-derived neurospheres (b13, c11, c12, d10). Hoxa11, expressed mainly in lumbar spinal cord but also weakly expressed in thoracic spinal cord (maybe as a result of a small dissection contamination), was detected exclusively in lumbar-derived neurospheres. The only exception was Hoxc13, which is moderately expressed in lumbar spinal cord and neurospheres but was also detectable, albeit at a lower level, in cervical neurospheres. These data provide evidence for the preservation in the adult spinal cord of a range of regionally specified neural stem cells expressing specific rostro-caudal Hox gene combinations.

Figure 6.

Hox gene expression in spinal cord neural stem cells.

(A): Examples of RT-PCR showing expression of 5′ Hox genes restricted to the lumbar part of the adult spinal cord and in corresponding neurospheres. Hypoxanthine phosphoribosyl transferase (HPRT) amplification was used as a PCR quality control. (B): Schematic representation of expression of the 39 Hox genes in adult spinal cord and in neurospheres. A black square (nd) indicates that no expression was detected using 2 μl of cDNA and 40 cycles. A colored square indicates that a product of the predicted size was obtained with 2 (dark red), 0.2 (red), and 0.02 (orange) μl of cDNA corresponding to low, moderate, and high expression, respectively. Abbreviations: C, cervical; L, lumbar; NS, neurospheres; SP, spinal cord; T, thoracic.


In contrast with recently published data [11], we provide evidence here that the spinal cord ependymal niche is composed of at least three cell types (Dcx+ Nkx6.1+ neuronal cells, GFAP+ radial cells, and BLBP+ cells) in addition to typical ependymocytes. A schematic view of the central canal niche is presented in Figure 7. Proliferation within this niche corresponds to the period of spinal cord extension. The neural stem cells present in this niche are represented mainly by dorsal GFAP+ cells in contact with the central canal. In vitro, spinal cord neural stem cells derived from different spinal cords accurately maintain the rostra-caudal Hox gene pattern. The niches expresses high levels of Notch signaling and anti-BMP protein and, more surprisingly, display typical mesenchymal traits. Finally, we demonstrated the requirement of the ZEB transcription factors for neural stem cell expansion in vitro.

Figure 7.

Schematic representation of adult spinal cord niche.

The presence of a single short cilia on GFAP+ cells was not investigated in this study but was extrapolated from the study in the SVZ performed by [26, 27]. Red shapes indicate CD15 staining.

Several cell types in the ependymal region of the adult spinal cord (tanycytes, cerebral-fluid contacting neurons) have been described in several species [28, 29] since the first observations of Ramon y Cajal [30]. Less is known about molecular markers characterizing cellular subpopulations and differences between the ventral and dorsal parts of the canal. Whereas most of cells around the canal express Sox2, Sox9, Cadherin 13, and EdnrB, the dorsal part is distinguished by the presence of higher CD15 staining and by a group of GFAP+ radial cells, at least some of which have stem cell properties. This is also consistent with our observation that in the hGFAP-GFP mouse, the dorsal part consistently contains GFP+ cells, whereas in the ventral and lateral of the canal this was only occasionally observed. The central canal dorsal-ventral dissymmetry can be further demonstrated by checking the GENSAT expression database (http://www.gensat.org), which reports that some markers are restricted to the dorsal or ventral ependymal regions. In human spinal cord, CD15 and PSA-NCAM labeling of central canal cells also revealed a manifest ventral-dorsal dissymmetry [14].

Recently Meletis et al. [11] did not find any molecular markers delineating subpopulations of central canal cells and concluded that spinal cord neural stem cells are GFAP unlike in the SVZ. Here, in contrast, we found abundant GFAP+ cells around the canal, the discrepancy most probably resulting from the use of different antibodies. There are several isoforms of GFAP, some of which are expressed mostly around the ventricle [31]. The use of a monoclonal rather than polyclonal antibody (as in this study) may not reveal all GFAP+ cells. The presence of such cells was confirmed in our study using polyclonal and monoclonal GFAP antibodies, a hGFAP-GFP transgenic mouse, and immunogold analysis. These cells either are located directly in contact with the cerebrospinal fluid or lie subependymally sending a process between the cells, a situation quite different from that in the SVZ where GFAP+ cells are mostly subependymal.

As in the SVZ, we found that most spinal cord neural stem cells express GFAP. However considering that in the SVZ, GFAP transit-amplifying cells cell can be converted into stem cells with epidermal growth factor [32] and that neurospheres are composed mainly of GFAP cells, the possibility still exists that, similarly, in the spinal cord a limited fraction of neurospheres could be derived from GFAP cells. Dorsal GFAP+ cells have a radial morphology and some expressed BLBP, a typical radial glial marker. In the spinal cord, these cells have been classified as tanycytes by other investigators [33–35]. This cell population is however heterogeneous [29] and we could distinguish a subpopulation of dorsal radial GFAP+ expressing BLBP or CD15. This subpopulation may constitute a significant fraction of adult spinal cord neural stem cells as (a) during development BLBP+ radial glial cells are considered as the main population of multipotent neural cells [36], (b) neurospheres are composed mainly of BLBP+ and CD15+ cells (this study and [17]), and (c) CD15 can be used to purify neural stem and progenitor cells in the adult SVZ and radial glial stem cell in the embryo [20]. The interaction of tanycytes with the vascular system is reminiscent of the situation observed with type B1 astrocyte in the SVZ [26, 37, 38] and, significantly, we observed that these cells strongly express the B receptor for endothelins, a family of cytokines released by endothelial cells that influences neural stem cell proliferation in vitro [16].

Using FoxJ1-GFP transgenic mice, a gene involved in the formation of motile cilia and expressed only by cells around the canal in the spinal cord, Meletis et al. [11] found that neurospheres are derived from FoxJ1+ cells. This supports the conclusion that neural stem cells are GFAP in the spinal cord since, in the SVZ, type B astrocytes have a single primary cilium [27] considered as immotile and are thus supposed to be FoxJ1, assumptions remaining to be formally demonstrated. In contrast, dorsal radial cells are likely to express FoxJ1 as these were clearly GFP+ upon cre activation in FoxJ1-cre Z/EG transgenic mice [11].

In addition to GFAP+ and BLBP+ cells, cerebrospinal-fluid contacting neurons (CSF-Ns) were commonly observed around the canal. They have already been noted in various brain regions and in the spinal cord from cyclostomes to mammals [28], although their function is still unknown. They could act as mechanoreceptors sensitive to pressure or flow of the cerebrospinal fluid (CSF) or to spinal cord flexion, or they might release neurotransmitters into the CSF as they express GABA and vasoactive intestinal peptide. The Dcx expression reported here may reflect an immature trait of these cells as Dcx is expressed transiently during development. It is however unlikely that these CSF-Ns are produced from a continuous adult spinal cord neurogenesis as we did not detect BrdU+ cells after 13 weeks. In addition, we provided evidence that these neurons are not generated postnatally and a recent study performed in the rat showed that they are in fact produced during embryogenesis [39]. The expression of the Nkx6.1 homeogene we found in these CSF-Ns is consistent with the fact that the adult central region is derived from the embryonic ventral Nkx6.1+ pMN/p2 domains [21].

The persistence of immature cells around the central canal coincides with the presence of a high level of both Notch (HES1, Jagged1) and transforming growth factor β (TGFβ; DAN) signaling proteins. As in CNS development [40], Notch signaling in the adult spinal cord may help to maintain the BLBP radial cell. HES1 and Jagged1 expression is however not restricted to the dorsal stem cells but extends to other cells around the canal. The latter appears to be able to generate neuroblasts, astrocytes, and oligodendrocytes after lesions [10, 11, 41] and can thus be considered as progenitor cells. This region is thus a new illustration of an adult CNS niche where developmental signals maintain and regulate various cells with different proliferation and differentiation potentials (stem cells, progenitors, young neurons) [42]. One remarkable protein we observed in the niche is DAN, also known as NBL1, a member of a cystine knot protein family that includes Noggin, Cerberus, Gremlin, and protein related to DAN and cerberus [43]. These proteins act as secreted inhibitors of BMP and growth differentiation factor proteins and are thus key regulators of TGFβ signaling. In the brain SVZ, Noggin is expressed by ependymocytes [44]. Here we found that DAN is expressed by cells of the spinal cord niche, notably by the dorsal group, and in vitro by most cells of the neurospheres. DAN causes growth suppression of a variety of cells lines and thus its expression in the spinal cord niche and in neural stem cells may serve a similar function.

In addition to the presence in the niche of typical stem cell signaling pathways, one important finding is the expression of SMA and ZEB1, two genes typically expressed during EMT. This suggests the existence of close relationships between EMT and stemness, as has recently also been found in breast cancer stem cells [45, 46]. We confirmed this hypothesis by showing that (a) ZEB protein inactivation in cultured neural stem cells led to a massive cell death, (b) neurospheres expressed several mesenchymal markers, and (c) neurosphere cells are as motile in vitro as typical mesenchymal cells. ZEB transcription factors are important regulators of the EMT required for development and cancer [22] and their expression is controlled by a number of ligands including TGF β1, progesterone, estrogen, insulin-like growth factor 1, and several micro-RNAs. They operate as transcriptional activators or repressors (notably for E-Cadherin) for controlling several cellular processes such as migration, senescence, and apoptosis. Significantly, ZEB1 transactivates the SMA promoter [47] and thus may be directly implicated in the expression of smooth muscle actin in central canal niche cells and in the neurospheres. As both ZEB1 and ZEB2 are expressed in neurospheres and DB-ZEB inactivates both, it is difficult to decide how much each protein contributes to the survival of neurosphere cells. The detection of only ZEB1 in the spinal cord ependymal region indicates it plays a predominant role in the maintenance of several cell types in the niche. In Drosophila, the orthologous ZEB protein Zfh-1 was recently shown to have a critical role in maintenance of the somatic stem cell compartment in the testis stem cell niche [48]. ZEB proteins may thus have a general role in the maintenance of stem/progenitor cells in the SVZ and spinal cord adult niches. Finally, EMT is associated with cellular delamination and migration, two phenomenon observed in CNS lesions such as in spinal cord injury where neural precursor cells readily leave their niche to migrate to the lesion site [10, 11]. The expression of EMT genes in the niche may be involved in keeping the cells competent to, or to prime them to, perform this process. In a normal situation, the EMT genetic program could be counterbalanced by local signalings to keep the cells in their normal position.

Whereas CNS adult neural stem cell niches display comparable gene expression and organization, it is now becoming clear that these niches harbor different types of neural stem cells capable of generating distinct types of neurons and glial cells [49]. This implies that expression of particular genes and epigenetic modifications operate to retain neural stem cell diversity in adult CNS. Here, we found that cultured spinal cord stem cells maintained distinct Hox gene expression profiles typical of the embryonic spinal cord pattern [25], suggesting that different Hox gene combinations still operate in adult spinal cord stem cells probably to specify their rostro-caudal features, as recently observed by Kulbatski and Tator [50]. The maintenance within the SVZ and along the rostro-caudal spinal cord axis of different patterning genes in neural stem cells suggests that not all neural stem cells may be equivalent and useful for stem cell-based therapy [51].


Compared with the brain SVZ and hippocampal niches, the role of this adult spinal cord niche is unclear, as the production of new cells is absent or low. In contrast, we found that after birth, this niche is mitotically active until 9 weeks in association with a higher expression of two markers (Sox4, Cadherin 13) so its role may be limited to postnatal spinal cord development. Since the spinal cord extends 2-fold in length after birth, this niche may be involved in the production of new cells for the central canal wall to extend in parallel with the spinal cord. As the spinal cord niche cells are able to produce neurons, oligodendrocytes, and astrocytes in vitro, this niche may produce such cells in the postnatal period. In the adult brain, the activity of the neural stem niches is modulated by various environmental parameters and, likewise, the adult spinal cord niche may be reactivated in particular physiological situations.


This work was supported by the European Union FP6 “Rescue” STREP, the “Association Française contre les Myopathies” (Evry, France), the Princess Grace of Monaco Fundation (Monaco). We are very grateful to Dr N. Heintz (New York), Dr P. Jay (Montpellier, France), Dr Darling (Louisville, KY), Dr Sudo (Toray, Japan), Dr Wegner (Erlangen, Germany), Dr A. Postigo (St. Louis), and Dr T. Walker (Brisbane, Australia) who generously provided us with BLBP, Sox9, ZEB1, Hes1, Sox10 antibodies, DB-ZEB construct, and Dcx-GFP mice, respectively. J.C.S. is a recipient of an AFM Ph.D. fellowship. We thank Dr C. Cazevielle (electron microscopy platform), Dr C. Duperray (cytometry platform), Dr H. Boukhaddaoui and C. Sar (cell imaging platform), I. Sassetti (histology platform), and Y. Gerber/G. Bonniface for high-quality technical assistance and experiment designs. We are grateful to Dr K. Langley for critical reading of the manuscript. A.P. and J.P.H. contributed equally to this work.


The authors indicate no potential conflicts of interest.