Impaired generation of mature neurons by neural stem cells from hypomorphic Sox2 mutants

Sox genes (Gubbay et al.,1990) encode transcription factors that regulate crucial developmental decisions (Kamachi et al.,2000; Wilson and Koopman,2002; Wegner and Stolt,2005). In mouse, Sox2 is expressed in, and is essential for,multipotent stem cells of the blastocyst inner cell mass, and its ablation causes early embryonic lethality (Avilion et al., 2003).

In the nervous system, Sox2 is expressed, and is functionally important, at the earliest developmental stages, in both chick and Xenopus(Kamachi et al., 2000; Pevny and Placzek, 2005; Wegner and Stolt, 2005). In humans, Sox2 neural expression is conserved, and heterozygous SOX2mutations cause hippocampal defects, forebrain abnormalities and anophtalmia(Fantes et al., 2003; Sisodiya et al., 2006; Kelberman et al., 2006). In the mouse nervous system, Sox2 is expressed in stem cells and early precursors, and in few mature neurons(Zappone et al., 2000; Ferri et al., 2004). Adult Sox2-deficient mice, in which Sox2 expression is decreased by about 70%,exhibit neural stem/precursor cell proliferative defects in the hippocampus and periventricular zone (Ferri et al.,2004). Moreover, neurons containing neurofilament/ubiquitin-positive aggregates are observed, together with dead neurons, in thalamic and striatal parenchyma, which are already substantially reduced in size at early developmental stages. These observations point to a possible role for Sox2 in the maturation and/or survival of embryonic and adult neurons. In these mutant mice, abnormalities of ependyma and choroid plexi (the source of growth and trophic factors/signaling molecules)(Lim et al., 2000) were also observed (Ferri et al., 2004). This raises the issue of whether neuronal defects observed in vivo represent an intrinsic defect, or a response to abnormalities in the environment.

We performed in vitro differentiation studies on neurosphere-derived neural cells. Neural stem cells from Sox2-deficient mice produce reduced numbers of mature neurons, but generate normal glia. Normal Sox2 levels are required at early differentiation stages. In vivo, subsets of GABAergic neurons are affected.

Neurosphere cultures were derived from adult or E14.5 mouse forebrain(Zappone et al., 2000; Ferri et al., 2004). For differentiation, neurospheres were dissociated to single cells, and plated onto MATRIGEL (Becton-Dickinson)-coated chambered slides (LabTec, Nunc) at 1-5×10⁴ cells/cm²(Zappone et al., 2000; Gritti et al., 1996; Gritti et al., 2001), with bFGF only as mitogen. After 3 days, the medium was changed to neural stem cell medium without bFGF, supplemented with 1% fetal calf serum (FCS). After further six days (differentiation day 9), cells were analyzed by immunocytochemistry.

Immunocytochemistry was as described by Zappone et al.(Zappone et al., 2000). For single-cell Sox2 immunofluorescence quantitation, see Fig. S2 in the supplementary material. Apoptosis was assayed by the DeadEnd Fluorimetric TUNEL system (Promega). Immunohistochemistry and BrdU labeling were as in Ferri et al. (Ferri et al.,2004); in the latter, sacrifice was 3 days after the last injection. Five olfactory bulb sections (20 μm; one every 16) were counted per animal.

Primary antibodies were: mouse anti-β-tubulin III (Covance 1:500),rabbit anti-β-tubulin III (Covance 1:2000), rabbit anti-calretinin(Chemicon 1:1000; 1:500 for immunohistochemistry), rabbit anti-connexin 43(Sigma 1:2000), rabbit anti-GABA (Sigma 1:2000), mouse anti-GALC (Chemicon 1:200), mouse anti-GFAP (Sigma 1:400), rabbit anti-GFAP (Zymed 1:100), mouse anti-GFP (Molecular Probes 1:100), rabbit anti-GFP (Molecular Probes 1:300),mouse anti-MAP2 (Biomeda 1:100), mouse anti-MAP2 (Immunological Sciences 1:200), rabbit anti-MAP2 (Chemicon 1:1000), mouse anti-nestin (Chemicon 1:200), mouse anti-NeuN (Zymed 1:100 or Chemicon 1: 400, for immunohistochemistry), mouse anti PSA-NCAM (AbCys 1:800), rabbit anti-Sox2(Chemicon 1:200 or 1:500 for immunohistochemistry), mouse anti-Sox2 (R&D 1:10 or 1:50 for immunohistochemistry), rabbit anti-S100 (DakoCytomation,1:400) and mouse anti-RC2 [Developmental Hybridoma Bank (ascites fluid)1:250]. Secondary antibodies were: anti rabbit or anti mouse Alexa 488 (green)or Alexa 594 (red) (Molecular Probes 1:1000-1:2000), anti rabbit or anti mouse FITC or TRITC (Jackson 1:200).

For immunofluorescence, 4% paraformaldehyde-fixed cells were pre-incubated with 10% FCS, 0.2% Triton X-100 in PBS for 30-60 minutes at room temperature,than the primary antibody was added (in 10% FCS in PBS) and left overnight at 4°C (or 1 hour at 37°C); cells were washed in PBS, the secondary antibody was added (in 10% FCS in PBS) for 1 hour at room temperature,followed by wash in PBS, DAPI nuclear counterstaining (4-8 minutes), and mounting in Fluorsave. Cells immunopositive for the various markers were counted under a fluorescence microscope; a minimum of 3000 total cells distributed on five fields was evaluated. Negative controls (equal cell samples treated the same way but omitting the primary antibody) were always performed in parallel for each reported experiment, and gave no signal.

DNAse-treated RNA was reverse transcribed and assayed by PCR for Sox2 as described by Zappone et al.(Zappone et al., 2000). Results were normalized using 18S RNA primers:5′TTTCGGAACTGAGGCCATGATTAAG3′ and 5′AGTTTCAGCTTTGCAACCATACTCC3′.

For ChIP, see Weinmann and Farnham(Weinmann and Farnham, 2002). Antibodies were anti-Sox2 (R&D) and rabbit anti-SV40 large-T (Santa Cruz). Primers for GFAP upstream region were 5′AAAGAATTCCCTGTGTTAGTCAGGGTTCTCTAG3′ and 5′AAACTCGAGTACAGTGAATGGGTAATAAAAATA3′. For SRR2 and nestin primers, see Miyagi et al. (Miyagi et al.,2006). For EMSA, see Catena et al.(Catena et al., 2004). Oligonucleotides are shown in Fig. 9.

For P19 transfection, the 0.6 Gfap region(Fig. 9; amplified with above ChIP primers) was cloned upstream to the TK promoter in the TK-luciferase vector (Miyagi et al., 2006). P19 cells (5×10⁵), plated the previous day in 3 cm dishes,were transfected with 0.5 μg luciferase reporter and 0.5 μg Sox2 expression vector (the CMV-Sox2-GFP lentiviral genome described below, or the same empty vector) using Lipofectamine 2000 (Invitrogen). Lysates were assayed for luciferase (Promega-E1980 kit) after 24 hours.

The Sox2 cDNA (XhoI-Bsu36I 1.3kb fragment) was cloned into the pRRLsin.PPT.CMV.NTRiresGFPpre lentiviral vector(Brunelli et al., 2007),between the CMV promoter and the IRES-GFP. The same vector, empty or carrying a Cre gene, was used as negative control (with comparable results). Lentiviruses were prepared as described by Brunelli et al.(Brunelli et al., 2007). Cells were transduced at MOI 100 at day 1 or 4(Fig. 1A) overnight. The following day the medium was changed to proliferation (day 1 transductions) or differentiation medium (day 4 transductions), and differentiation continued to day 9.

P0 cortical neurons (Wagenaar et al.,2005; Li et al.,2005) were plated on polyethyleneimine-laminin-coated slides at 10⁶ cells/ml. After 3 hours, the plating medium was replaced with Neurobasal medium with B27, 1 mM glutamine, 5 ng/ml bFGF. The culture was maintained for 4-10 hours, prior to fixation with 4% paraformaldehyde.

Neurosphere cultures were derived from the subventricular zone (SVZ) of adult normal and Sox2-hypomorphic mice, carrying a null allele(Sox2^β-geo) together with a `knockdown' allele(Sox2<sup>ΔENH</sup>) (Ferri et al.,2004). The null allele is a `knock-in', where the β-geo gene replaces Sox2. In the `knockdown' allele an upstream Sox2 enhancer is deleted. The level of Sox2 mRNA in Sox2^β-geo/ΔENH neurosphere cultures is 25-30% of the wild type (Ferri et al., 2004).

In vitro, the growth (Zappone et al.,2000) of undifferentiated cultures (measured as numbers of total cells, or neurospheres) from mutant mice was not significantly different from that of normal controls (not shown).

Differentiation was carried out according to Gritti et al.(Gritti et al., 1996; Gritti et al., 2001)(Fig. 1A). Undifferentiated neurospheres, dissociated to single cells, were made to adhere to slides, in the presence of bFGF. After 3 days, bFGF was removed, and 1% FCS was added,leading to differentiation within 9 days from initial plating. We studied differentiation of neurons and glia, as well as Sox2 expression, during this time window. For Sox2 evaluation, we used mouse monoclonal (R&D) and rabbit polyclonal (Chemicon) antibodies, of which we carefully confirmed the specificity (Fig. 1B; see Fig. S1 in the supplementary material) by testing wild-type cells versus Sox2 conditionally deleted (null) cells.

In undifferentiated neurospheres, Sox2 is expressed, together with nestin(a marker of undifferentiated precursors) in virtually all cells (not shown). In differentiating cells, Sox2 is expressed at variable levels (dim to bright)in most cells until day 9, although the bright population was much reduced after differentiation day 1 (Fig. 1C; see Fig. S2 in the supplementary material); nestin colocalized with Sox2 at day 1 (Fig. 1C)but disappeared in most cells by day 3 (see Fig. S4 in the supplementary material). This result is mirrored by a 80% reduction of Sox2 mRNA in differentiated cells (Fig. 1D). In mutant cells, at the beginning of differentiation, Sox2 mRNA(Ferri et al., 2004) and protein (Fig. 1E) are lower than in normal cells, as expected. By single-cell immunofluorescence, at day 1, the Sox2-bright population is much decreased in mutant cells; between days 5 and 9, the difference between normal and mutant cells is progressively reduced (see Fig. S2 in the supplementary material).

β-Tubulin-positive cells (neuronal lineage) appear towards day 5, and persist until day 9; MAP2, a more differentiated marker, is well visible at day 9. Neuronal lineage cells express relatively high levels of Sox2(Fig. 2A,B); however, not all Sox2-bright cells expressed these markers. Similarly, the few GALC-expressing cells (oligodendrocytes) clearly retained Sox2 expression(Fig. 2C). However, the predominant population of (GFAP-positive) astroglia exhibited little Sox2-fluorescence (however, glial nuclei are more expanded than other nuclei,and thus may tend to be less Sox2 bright)(Fig. 2D). As in wild-type cultures, most mutant MAP2-positive (Fig. 2B) and β-tubulin- and GALC-positive cells (see Fig. S2 in the supplementary material and data not shown) retained significant, though slightly decreased (see Fig. S2C in the supplementary material), Sox2 expression.

In cultures from normal adults, most neuronal cells show mature morphology,with extensive arborization, at differentiation day 9(Fig. 3A,B, left). However, in mutant cultures, β-tubulin-positive cells with developed arborization were very rare (Fig. 3A,B,right) and most (undeveloped) β-tubulin-positive cells showed much weaker staining (Fig. 3A). Thus,although the total number of β-tubulin-positive cells is similar between normal and mutant cultures, the absolute number of morphologically `mature'mutant neurons is strikingly decreased (see Table S1 in the supplementary material; Fig. 3).

The immature morphology of mutant β-tubulin-positive cells correlates with impaired expression of mature neuronal markers(Fig. 4). In normal cells, mostβ-tubulin-positive cells were positive for NeuN (80%) or MAP2 (60%)(Fig. 4, see Table S1 in the supplementary material), whereas in the mutant, cells positive forβ-tubulin/NeuN, β-tubulin/MAP2 and PSA-NCAM were strikingly decreased (Fig. 4). We obtained similar results using cultures from E14.5 forebrains (not shown).

Differentiated neuronal cells express the GABA neurotransmitter(Fig. 5)(Gritti et al., 1996; Gritti et al., 2001), and Ca²⁺-binding proteins (calretinin and calbindin), which define inhibitory neurons and their different subpopulations(Wonders and Anderson, 2006; Levitt et al., 2004; Makram et al., 2004). We evaluated, at day 9, the number of cells expressing GABA or calretinin as a proportion of β-tubulin or MAP2-positive cells(Fig. 5; see Table S1 in the supplementary material). Only cells giving strong signals, covering cell body and processes, were scored positive. In both embryonic and adult cultures from normal mice, most of the strong β-tubulin- or MAP2-positive cells were also GABA positive (Fig. 5; see Table S1 in the supplementary material); a few GABA-positive cells (10-15% of the GABA-positive population) were MAP2 negative. In the mutant, most of the(rare, see Table S1 in the supplementary material) MAP2- and (well-developed)β-tubulin-positive cells were also GABA positive, as in the normal cells,but absolute numbers were reduced by more than ten times(Fig. 5); in addition, many GABA-positive cells were MAP2 negative(Fig. 5). Similarly, calretinin expression in the normal cells was frequent in MAP2-positive cells (30-40%),whereas in the mutant it was very rare(Fig. 5; see Table S1 in the supplementary material).

We further studied differentiation into GFAP-positive astroglia, and GALC-positive oligodendroglia. Contrary to results with neuronal differentiation, GFAP-positive cells with mature astroglia morphology were detected in similar proportions in cultures from normal and mutant cells (not shown and see Table S1 in the supplementary material)

Unexpectedly, in mutant cultures, some (∼30%) of theβ-tubulin-positive cells also showed clear, although quite low, GFAP expression (Fig. 6). These cells often showed some neuron-like arborization(Fig. 6, rows 2, 3), but it was not as developed as in wild type β-tubulin-positive cells; however, these cells were obviously distinguished from normal astrocytes, which were highly GFAP-positive (but β-tubulin-negative) and morphologically well developed(Fig. 6, row 4). In normal cultures, we never observed such cells, although a very low proportion ofβ-tubulin-positive cells (∼3%) showed double staining(Fig. 6, top, arrowhead); these cells, however, were very poorly developed, and might represent an early maturation stage. Interestingly, β-tubulin/GFAP double-positive cells were observed in differentiated cultures of glioblastoma multiforme neural stem cells (Galli et al.,2004; Lee et al.,2006a). Notably, these cells aberrantly express Sox2(Hemmati et al., 2003; Lee et al., 2006a; Nicolis, 2007; Pomeroy et al., 2002). Finally, oligodendrocytes were slightly reduced (not shown; see Table S1 in the supplementary material).

The observed results are neither caused by differentiation delay nor by increased apoptosis of mutant cells, as indicated by normal kinetics of nestin and β-tubulin expression and by TUNEL assays (see Fig. S4 in the supplementary material). In conclusion, Sox2 is important mainly in neuronal,but not in astroglial differentiation.

As shown above, Sox2-mutant cells show significantly lower levels of Sox2 than normal cells at the onset of differentiation(Fig. 1E, see Fig. S2 in the supplementary material); but not at later stages (see Fig. S2A-C in the supplementary material).

To evaluate if restoration of Sox2 levels might rescue the differentiation defect of mutant cells, we used a Sox2-IRES-GFP lentiviral construct. We transduced mutant cells at the end of day 1 after plating(Fig. 1A); after 16 hours, we washed the well to remove the virus, adding fresh medium to allow differentiation to proceed until day 9. Control cells were treated similarly,without virus or with control virus expressing only GFP. In an alternative experiment, cells were transduced at day 4, after the switch from mitogen-containing medium to mitogen-free, serum-containing medium. A high proportion (75-80%) of the cells were transduced, expressing GFP and Sox2(Fig. 7A). Transduction at day 1 did not change the overall number of β-tubulin-positive cells, but resulted in a dramatic increase in the proportion of well-arborizedβ-tubulin-positive cells (Fig. 7B,C,D), and of cells expressing the more mature MAP2 marker(Fig. 7C,D).

Importantly, well-arborized morphology in β-tubulin or MAP2-positive cells was observed almost exclusively in efficiently transduced (i.e. GFP-positive) cells (Fig. 7C;arrowheads). Most of the untransduced (GFP-negative) β-tubulin-positive cells showed poor arborization (Fig. 7C; arrow). This latter result represents an `internal' control,indicating that the rescue of the normal phenotype is due to viral-dependent expression, but not to any `environmental' change (caused by the transduction procedure) affecting the efficiency of differentiation. Moreover, control virus expressing GFP but not Sox2 had no effect(Fig. 7B,D). In contrast to the results obtained when the virus was transduced at day 1, no significant effect of Sox2 transduction was observed at day 4(Fig. 7B,D). Thus, appropriate Sox2 levels are required at a crucial early stage of differentiation.

We further examined the astroglia population from cultures transduced with the Sox2-GFP-expressing lentivirus. Unexpectedly, cells expressing high levels of GFP (thus presumably of Sox2) showed reduced or no GFAP expression, while retaining astroglia morphology (Fig. 8A, left) and expression of astrocyte markers S100 and connexin 43(Fig. 8B; see Fig. S3 in the supplementary material); by contrast, cells that had not been transduced showed the expected astroglia morphology with high GFAP expression(Fig. 8A, left). The loss of GFAP expression is not due to toxicity from high levels of GFP, as cells transduced with a GFP-lentivirus without the Sox2 gene were not affected(Fig. 8A, right). Furthermore,the inhibitory effect of excess Sox2 levels on GFAP expression was observed both when the virus was added at day 1 and at day 4(Fig. 8A).

This surprising result prompted an investigation of the possibility that Sox2 might directly affect GFAP expression. Upstream to the GFAP promoter(Morita et al., 1997; Kuzmanovic et al., 2003) lies a region containing three potential consensus Sox2-binding sites (conserved between mouse and man) (Fig. 8C). We cloned this region upstream to the thymidine kinase (TK)minimal promoter, linked to a luciferase reporter, and transfected this construct into P19 embryonic carcinoma cells, together with a Sox2 expression vector or, as control, the same vector without Sox2. The upstream promoter region stimulated luciferase activity by twofold in the absence of Sox2;however, the stimulation was abolished by Sox2 overexpression(Fig. 8D). This suggests that Sox2, expressed at high levels, is a repressor at this regulatory element.

In gel shift analysis (Fig. 8E), recombinant Sox2 (expressed in COS cells) or endogenous Sox2 from P19 cells (Fig. 8E left panels, lanes 1, 4) forms a retarded complex with a GFAP probe containing the two upstream putative Sox2 sites. This complex has mobility similar to that formed on a bona fide Sox2-binding site from an Oct4 gene enhancer(Chew et al., 2005)(Fig. 8E, left panels, Oct4 probe, lanes 2, 5). The complex was abolished by mutation of the Sox2 sites of the probe (MutGfap, lanes 3, 6) and by competition with excess unlabelled Oct4(not shown) and wild-type, but not mutant, GFAP oligonucleotide(Fig. 8E, right). Furthermore,in in vivo chromatin immunoprecipitation (ChIP) experiments, an anti-Sox2 antibody specifically precipitates the upstream GFAP regulatory region in chromatin from both P19 (which express Sox2) and embryonic (E12.5) neural tube cells (Fig. 8F). Control experiments with other Sox2-binding sequences (SRR2 and nestin) indicate that the anti-Sox2 antibody correctly precipitates these chromatin regions in P19 and spinal cord cells, respectively, although SRR2 is not precipitated in spinal cord cells, as expected (Miyagi et al., 2006). These experiments, which demonstrate binding of Sox2 to the GFAP upstream region in vivo and in vitro, and Sox2-dependent transcriptional inhibition (Fig. 8C-F), demonstrate that the repression of GFAP by Sox2 shown in differentiating neural cells (Fig. 8A) may be mediated, at least in part, by direct Sox2 regulation of transcription.

In vitro studies provided three main observations: (1) mutant cells show impaired neuronal maturation, with cells exhibiting abnormal morphologies; (2)GABAergic markers are significantly reduced; and (3) Sox2 levels are higher in early than in more differentiated neural cells, but significant Sox2 protein is retained in many neurons.

To analyze in vivo neuronal differentiation, we examined cortical neurons of newborn mice and newly generated rostral migratory stream (RMS) neurons. P0 cortical neurons derive from embryonic radial glia, and had only a few days to mature since their terminal cell division. Neurons, made to adhere to slides,were stained for neuronal markers. Most cells were positive for β-tubulin and MAP2 at variable intensities and had comparable levels of staining between normal and mutant brains (see Fig. S5 in the supplementary material). However,GABA-positive and calretinin-positive cells were decreased by 50-60% in mutant cortical cells (Fig. 9A-C),confirming a defect, in mutant brain in vivo, of at least one class of mature neurons: the GABAergic neurons.

Cortical GABAergic neurons originate from precursors in the ganglionic eminences, which migrate after terminal division by tangential routes (Makram et al., 2004; Wonders and Anderson,2006). In normal E17.5 embryos, we found several calretinin-positive (i.e. GABAergic) cells within the cortical plate(Fig. 10A-D), whereas in mutant embryos calretinin-positive cells were detected along subcortical fiber bundles but were very scarce or absent in the cortical plate(Fig. 10E-H). This migration abnormality might be part of the suggested differentiation defect. GABA staining at the same stage reveals a disorganized labeling pattern of GABAergic neurons in the mutant (Fig. 10I-N). GABAergic cells which reach their final destination in the cortex progressively develop postnatally into several more mature interneurons subtypes, which include calretinin-positive ones(Markram et al., 2004; Wonders and Anderson, 2006). In adult mutant cortex, calretinin-positive cells showed significant abnormalities, such as reduced dendritic and axonal arborizations(Fig. 11). In conclusion, a subpopulation of embryonically generated neurons (GABAergic neurons) is not only decreased in numbers in postnatal cortex, but also shows significant morphological abnormalities in embryo and adult.

In adult mouse, stem cells within the SVZ generate neurons (many of them GABAergic) that migrate to the olfactory bulb, where they complete differentiation with the expression of mature markers (NeuN in all neurons,calretinin and calbindin in GABAergic neurons subclasses)(Doetsch, 2003; Lledo et al., 2006). We administered BrdU to adult mice, and measured the proportion of NeuN-positive cells within the BrdU-positive population in the olfactory bulb. The newly generated neurons (BrdU/NeuN-positive cells) are substantially (∼40%)decreased in granule (GL) and in periglomerular (PGL) layers of mutant mice(Fig. 12A), indicating a significant maturation defect.

Does this maturation defect result in reduced steady-state levels of GABAergic neurons? Calretinin-positive cells are strongly decreased (40%)within the most external (periglomerular) layer, where mature calretinin-positive cells reside (Fig. 12B). This suggests that mutant cells destined to develop as calretinin-positive cells in the periglomerular layer may fail to reach it and/or complete their maturation. Additionally, calretinin-positive cells in the external layers of the olfactory bulb showed an important decrease in their degree of arborization (Fig. 12C).

In mouse, Sox2 deficiency causes defects in adult hippocampal and subventricular zone stem/progenitor cells, decreased neurogenesis and neuronal defects (Ferri et al., 2004). Here, we show that normal Sox2 levels are essential for proper neuronal differentiation in vitro and, in vivo, for at least one class of neuron, the GABAergic neuron.

In vitro, Sox2 expression is high in undifferentiated cells, significantly declines during differentiation, but is not completely extinguished in many cells (Figs 1, 2). The observed Sox2 expression is not due to antibody crossreactions, as shown by control experiments, using Sox2-null neural cells.(Fig. 1B; see Fig. S1 in the supplementary material), and by RT-PCR(Fig. 1D). This agrees with Bani-Yaghoub et al. (Bani-Yaghoub et al.,2006), who showed significant Sox2 expression in P3 cortex (glia and neurons), relative to high levels in embryonic cortex (mostly neural precursors).

Both in vitro and in vivo, Sox2 expression is decreased in the mutant,although much more in early than in more mature cells(Fig. 1E; see Figs S2, S5 in the supplementary material). It is possible that the enhancer that is deleted in the knockdown allele may be less relevant in mature cells, allowing some compensation. Notably, in vivo (Ferri et al., 2004) (see Fig. S5 in the supplementary material) Sox2 expression is maintained in subsets of differentiated neurons, within P0 cortical neurons, in adult SVZ-generated precursors/neurons in the olfactory bulb and in other cells. In the mutant, Sox2 is already decreased within early precursors, but much less significantly in neurons (see Fig. S5 in the supplementary material), in agreement with the in vitro observations.

In vitro, Sox2-deficient cells exhibit a striking differentiation defect,characterized by abnormal morphology and decreased expression of mature differentiation markers. As the defect is apparent at differentiation day 5(Fig. 3C), Sox2 is already required at early stages. This is confirmed by the in vitro rescue experiment with a Sox2-expressing lentivirus (Fig. 7). Sox2 overexpression in mutant cells at the onset of differentiation is necessary to rescue the well-arborizedβ-tubulin-positive, MAP2-positive phenotype observed in normal, but not mutant cells. However, late expression does not rescue the phenotype(Fig. 7). Preliminary data (in preparation) indicate that neurons originate only from cells that are still dividing at early differentiation stages (day 2, but not day 4); moreover,progenitors at early, but not late stages, express transcription factors known to be involved in neuronal differentiation. Correct expression of Sox2 at early stages may be required to establish a downstream transcriptional program for differentiation, perhaps by generating a `poised' chromatin structure at loci crucial for subsequent neuronal development (as exemplified for Sox2 itself in ES cells) (Boyer et al.,2005; Boyer et al.,2006a; Boyer et al.,2006b; Szutoriz and Dillon, 2005; Azuara et al., 2006; Bernstein et al., 2006; Lee et al., 2006b). When such a program is compromised by insufficient Sox2 levels, as in the mutant, all successive maturation steps (from β-tubulin to MAP2/NeuN expression)would be altered. Indeed, clearly decreased levels of Sox2 are found, in the mutant, at early, but not at late, stages of neurogenesis.(Fig. 1E; see Figs S2, S5 in the supplementary material).

The rescue experiment, while highlighting an essential role of Sox2 in early cells, does not rule out additional, but not yet demonstrated, roles of Sox2 at later stages, as suggested by the presence of Sox2 in well-developed MAP2-positive cells in vitro (Fig. 2) and a few neurons in vivo (ssee Fig. S5 in the supplementary material) (Ferri et al.,2004).

In the mutant, some cells with poorly developed neuronal morphology co-express a neuronal (β-tubulin) with a glial (GFAP) marker(Fig. 6). In neuronal committed cells, Sox2 might act to repress part of a gliogenic transcription program. Indeed, Sox2 binds to the GFAP promoter in vitro and in vivo(Fig. 8E,F); moreover, when overexpressed, it silences the endogenous GFAP activity in differentiating neural cells (Fig. 8A), and inhibits a co-transfected GFAP promoter-driven reporter transgene(Fig. 8D). Thus, at least part of the Sox2-dependent inhibition of GFAP is explained by a direct repressor activity of Sox2.

We hypothesize that Sox2 has a dual role in neural cell differentiation; in early precursors committing themselves to neurogenesis, it `programs' later neuronal differentiation events, while repressing some alternative(glial-specific) transcription programs. In cells undergoing gliogenesis, its decline would allow proper glial-specific gene expression. Similar models have been proposed for other differentiation systems(Enver and Greaves, 1998; Hu et al., 1997; Laslo et al., 2006; Mikkola et al., 2002; Nutt et al., 1999). In mutant neural precursors, Sox2 levels would be too low to upregulate the neuronal differentiation program efficiently and/or to switch-off the glial program.

An important role of Sox2 in neural stem/precursor cells proliferation/maintenance was identified previously(Graham et al., 2003; Bylund et al., 2003; Ferri et al., 2004). This is consistent with the high level of Sox2 detected in such cells(Fig. 1B-E; see Fig. S5 in the supplementary material). Our present results point to an additional role of Sox2 in differentiated cells. Sox2 might participate in different networks of transcription factors in stem versus differentiating cells. A precedent exists for Oct4, a factor co-expressed with Sox2 in ES cells, the levels of which affect both pluripotency and differentiation(Niwa et al., 2000).

Graham et al. (Graham et al.,2003) and Bylund et al. (Bylund et al., 2003) showed that increasing Sox2 levels in normal chick embryo neural tube prevents their initial (day 1) differentiation intoβ-tubulin-positive cells and maintains their self-renewal. Bani-Yaghoub et al. (Bani-Yaghoub et al.,2006) obtained similar results in embryonic neural precursors in vitro. These results are apparently at variance with our observation that Sox2 overexpression in Sox2-mutant cells increases their differentiation(Fig. 7).

Several important differences in species, cellular models, stages and differentiation techniques may explain these discrepancies. In particular, we transduced Sox2 in cells that had previously been induced to initiate differentiation by adherence to matrigel, whereas the above-mentioned authors overexpressed Sox2 in proliferating early precursors prior to their entry into differentiation. Furthermore, most importantly, we overexpressed Sox2 in mutant cells that already have an abnormally low Sox2 level, whereas the above authors overexpressed Sox2 in wild-type cells expressing the physiological level of Sox2. Thus, the rescue we observe may simply reflect the reestablishment of Sox2 levels appropriate for differentiation in cells that already entered the differentiation pathway; the fact that the majority, but not all, of the transduced cells were rescued may indicate the need for a critical Sox2 level, that is neither too low (as in some transduced cells, Fig. 7A) nor too high. By contrast, their results may be due to Sox2 levels too high to allow entry of stem and early precursor cells into the differentiation pathway.

Sox2 overexpression in mutant cells did not change the balance between neuronal (as measured by β-tubulin expression) and glial cells. Rather,it modulated their differentiated characteristics (increased neuronal maturation, decreased glial GFAP expression). Thus, Sox2 does not control the choice between neuronal and glial differentiation.

In agreement with in vitro neural defects, we detect, in vivo, significant abnormalities of a subset of neurons, GABAergic neurons. These are decreased by 40-60% in P0 cortical cells and in the olfactory bulb, indicating that both embryonic and adult genesis of this neuronal type is compromised (Figs 9, 12). Additionally, we detect morphological abnormalities in embryonic GABAergic neurons, during their migration to the cortex from the ganglionic eminences, and in early postnatal cortex (Figs 10, 11), as well as, to a lower extent, in newly generated calretinin-positive cells in the adult olfactory bulb (Fig. 12C). These results confirm the in vitro results (Figs 3, 4 and 5) and extend preliminary in vivo evidence of loss of neural parenchyma and reduced maturation of postnatal neurons (Ferri et al.,2004).

From a quantitative point of view, the overall population in the P0 cortex and postnatal olfactory bulb is not as deeply affected as in the in vitro experiments. We suggest several, not mutually exclusive, explanations for this discrepancy.

First, only selected neuronal populations may be vulnerable to low Sox2 dosage; these might be more represented in vitro than in vivo. Indeed, in vivo, among the neuron types tested, only the GABAergic subset is detectably compromised; significantly, in our in vitro system, the majority of differentiated neurons are of this type(Fig. 5) (see Gritti et al., 2001; Conti et al., 2005).

Second, in vitro stem cells may differ to some extent from in vivo stem cells. Indeed, most bona fide in vivo stem cells are in a low cycling state,and are a radial glia cell type (Doetsch,2003), whereas in vitro stem cells are highly proliferating. Moreover, many in vitro stem cells actually arise from more differentiated in vivo precursors (transit-amplifying progenitors, astroglia and oligodendrocytes), which have been reprogrammed in vitro to a stem cell status by growth factor stimulation (Doetsch et al., 2002). Interestingly, reprogramming of oligodendrocyte precursors to stem cells requires Sox2 reactivation(Kondo and Raff, 2004); thus,Sox2 mutant neural stem cells might have been `reprogrammed' less efficiently than wild-type cells.

Third, in vitro culture conditions, while allowing efficient differentiation of normal neural stem cells, might be subtly deficient relative to the in vivo environment. This might exaggerate the proportion of mutant Sox2 cells that fail to undergo appropriate differentiation in vitro. Indeed, in vitro not all differentiated markers are developed, and very few cells express appropriate electrophysiological properties, in contrast to ex vivo neurons (Gritti et al.,1996; Gritti et al.,2001).

Finally, cell selection effects normally operate in vivo, and only a minority of post-migratory cells survive(Ferrer et al., 1992; Muotri and Gage, 2006; Oppenheim, 1991). Abnormal neurons, that fail to properly develop and establish connections, will probably be selected against in vivo. The neuronal loss observed in vivo in specific brain areas (striatum, thalamus), and the reduced cortical extension(Ferri et al., 2004), might reflect these phenomena.

The in vitro culture system, by demonstrating a role for Sox2 in neuronal differentiation, will allow the identification of early Sox2 targets important for neuronal differentiation, by functional rescue experiments. Rare cases of Sox2 deficiency in man are characterized by hippocampal abnormalities,epilepsy, eye and pituitary defects(Fantes et al., 2003; Ragge et al., 2005; Sisodiya et al., 2006; Kelberman et al., 2006), also reported in mutant mice (Ferri et al.,2004; Taranova et al.,2006). Loss of GABAergic inhibitory neurons leads to epilepsy in mouse and man (Noebels, 2003; Cobos et al., 2005). Our observation of GABAergic neuron deficiency in mouse points to a plausible cellular basis for epilepsy in humans with SOX2 mutations. Other neuronal subsets remain to be tested for their Sox2 requirement.

We thank Akihiko Okuda for the TK-luciferase vector, Anna Ferri and Valentina Tosetti for help with experiments, and Daniela Santoni for animal care. This work was supported by grants from Telethon (GGP05122), Fondazione Cariplo (2004-1503 and NOBEL) and MIUR (Cofin 2005; FAR 2004-6) to S.K.N.