Conserved POU binding DNA sites in the Sox2 upstream enhancer regulate gene expression in embryonic and neural stem cells.

The Sox2 transcription factor is expressed early in the stem cells of the blastocyst inner cell mass and, later, in neural stem cells. We previously identified a Sox2 5'-regulatory region directing transgene expression to the inner cell mass and, later, to neural stem cells and precursors of the forebrain. Here, we identify a core enhancer element able to specify transgene expression in forebrain neural precursors of mouse embryos, and we show that the same core element efficiently activates transcription in inner cell mass-derived embryonic stem (ES) cells. Mutation of POU factor binding sites, able to recognize the neural factors Brn1 and Brn2, shows that these sites contribute to transgene activity in neural cells. The same sites are also essential for activity in ES cells, where they bind different members of the POU family, including Oct4, as shown by gel shift assays and chromatin immunoprecipitation with anti-Oct4 antibodies. Our findings indicate a role for the same POU binding motifs in Sox2 transgene regulation in both ES and neural precursor cells. Oct4 might play a role in the regulation of Sox2 in ES (inner cell mass) cells and, possibly, at the transition between inner cell mass and neural cells, before recruitment of neural POU factors such as Brn1 and Brn2.

Neural stem cells are found throughout the developing embryo, where they are involved in the progressive generation of the mature central nervous system, and also in the adult brain, where neurogenesis persists in selected locations (1)(2)(3). In mammalian development, neural stem cells originate, soon after gastrulation, from the early, highly multipotent stem cells of the blastocyst inner cell mass, through sequential development of these cells into primitive ectoderm (epiblast) and neural plate (3,4). This process, still poorly understood in molecular terms, has many features of a "default" developmental pathway, in embryogenesis (5) as well as in in vitro neural development of embryonic stem (ES) 1 cells (6 -8). Together with "general" neural properties (e.g. ability to generate neurons and glia), neural stem cells acquire (following the action of signaling centers) region-specific transcriptional programs (3,9,10) that will translate into region-specific characteristics of their differentiated neuronal and glial progeny.
The Sox2 transcription factor is expressed in the ES cells of the inner cell mass and its immediate descendant, the epiblast (11); with the beginning of neurogenesis, Sox2 expression becomes restricted to the neural primordium (neural plate) (10 -13). Later, Sox2 is expressed in neural stem cells residing in the ventricular zone of the developing neural tube and in their immediate progeny through embryogenesis (10,14) and into adulthood (15).
Recent experiments indicate that Sox2 plays critical functional roles within all of these stem/multipotent cell types. Stem cells constituting the early epiblast, which represents the multipotent precursors to all embryonic cell types, are lost in mouse embryos homozygous for a Sox2-null mutation, leading to early embryonic lethality (11); epiblast cells cease to proliferate and self-renew, and some of them instead turn on differentiation markers characteristic of trophoblast cells (11). Indeed, Sox2 was also shown to regulate genes, such as fibroblast growth factor 4 (16), essential for epiblast stem cell maintenance (17). The subsequent role of Sox2 in early neuroectoderm has not yet been addressed in mouse. However, in Xenopus, experiments with dominant-negative forms of Sox2 (interfering with the activity of Sox2 and the related Sox1 and 3 genes) have indicated a requirement for Sox family members in the generation of early neuroectoderm (18). Further, electroporation of Sox2 dominant-negative constructs in the chick embryo neural tube impairs the maintenance of proliferating, multipotent neural precursors and elicits their premature differentiation, whereas conversely, ectopic Sox2 (or Sox1) gene expression maintains the proliferative state of neural precursors and prevents their differentiation (14,19). Finally, targeted mutations in mice decreasing, but not abolishing, Sox2 expression in neural cells cause an impairment of adult neural precursor cell proliferation and of neurogenesis, together with a depletion of early stem/precursor cells in the hippocampal adult neurogenic region, indicating a requirement of Sox2 for adult neural stem/ precursor cells function (15).
Although Sox2 is expressed panneurally, its transcriptional regulation in the central nervous system is guided from its early stages by region-specific regulatory elements (10,20), suggesting that Sox2 might be differentially regulated by regional signals, and might control general neural properties (e.g. maintenance of a proliferative, self-renewal state) according to the specific developmental needs of the different regions of the nervous system.
In previous work, we identified a 5Ј-regulatory region of mouse Sox2 which drives expression of a ␤-geo transgene to neural stem cells and precursors of the developing (and adult) dorsal telencephalon (10). In transgenic mice, this region drives expression of the ␤-geo reporter already in the blastocyst inner cell mass, followed by restriction of expression to the neural plate and, progressively, to the telencephalon (10).
Here, we have investigated the sequence requirements of this regulation by transgenic analysis and ES cell transfection and, in parallel, the nuclear proteins binding to functionally relevant sequences by gel shift, footprinting, and chromatin immunoprecipitation studies. We find that conserved POU factor binding sites in the upstream Sox2 enhancer are critical for transgenic Sox2 expression in both embryonic and neural stem/ precursor cells.

DNA Constructs
Modified Fragments of the HSI (Chromatin DNase I-hypersensitive Site 1) Region-Progressively smaller fragments encompassing the HSI region ( Fig. 1A) were amplified by PCR from genomic subclones, with primers complementary to the extremities of the different fragments (0.4a, 0.4b, 0.1; Fig. 1, A and B) and carrying a terminal (5Ј or 3Ј) HindIII restriction site; the upstream primer also carried an SalI site internal to the HindIII site, for later transgene excision. PCR-amplified fragments were verified by sequencing.
HSI-Sox Promoter Constructs-Fragments amplified as above were cloned (in the correct 5Ј 3 3Ј genomic orientation) into the unique HindIII site (Fig. 1A) upstream from the Sox2 "minimal" 3.3-kb promoter, in the 5.7-kb Sox2 ␤-geo vector (10), in which the Sox2 5Јflanking region lies upstream from the ␤-geo reporter, which encodes a ␤-galactosidase/neomycin-G418 resistance fusion gene. SalI digestion (cutting upstream from the amplified HSI fragment, see above, and downstream from ␤-geo, see Ref. 10) was used to release the appropriate transgene-containing fragment (HSI fragment-Sox2 promoter-␤geo), which was then separated from all residual sequences and purified by agarose gel electrophoresis for subsequent pronuclear injection or cell transfection.
HSI-TK Promoter Constructs-The 3-TK pr. (see Fig. 2A) construct was obtained from the full-size 5.7-kb Sox2 ␤-geo vector (Ref. 10, and above) by replacement of the Sox2 promoter region (3Ј to the HSI element) with a herpes simplex TK minimal promoter fragment. The Sox2 promoter was removed by complete HindIII digestion (cutting just upstream from the promoter; Ref. 10 and Fig. 1A) and partial NotI digestion (cutting in the Sox2 5Ј-untranslated region, at the boundary with ␤-geo). The TK promoter, carrying appropriate upstream HindIII and downstream NotI sites for cloning, was obtained by PCR from the pMC1neo poly(A) plasmid (Stratagene) with the primers 5Ј-ACG TAA  GCT TCT CGA GCA GTG TGG TTT TG-3Ј and 5Ј-TTT TCC TTA AGC  GGC CGC TTG GCT GCA GGG TCG CTC-3Ј. Underlined sequences represent an HindIII (in the upstream primer) and an NotI site (in the downstream primer); sequences in italics are those complementary to TK.
All other HSI-TK pr. constructs were derived from the 3-TK pr. construct by cloning the appropriate short modified HSI-encompassing fragments (flanked by a 5Ј-HindIII-SalI linker and a 3Ј-HindIII linker) (see above) into the unique HindIII site at the 5Ј-extremity of the TK promoter. The presence of an SalI site at the 5Ј of all HSI fragments allowed subsequent excision and purification of the transgenes by SalI digestion, as above (see under "HSI Sox Promoter Constructs").
PCR Mutagenesis-To generate point mutations in both POU sites 1 and 2 ( Fig. 1B; 0.4a MUT and 0.1 MUT-Sox pr. constructs), the 0.4a fragment (Fig. 1, A and B) (from the 0.4a Sox pr. construct, see above) was cloned (as an HindIII fragment, see above) into the HindIII site of pBluescriptSK(ϩ).
To mutate POU site 2, two partially overlapping primers containing the desired mutation (underlined) were used to amplify separately, from the wild-type sequence as template (in conjunction with primers from Bluescript sequences flanking the HindIII cloning site), the 5Ј-and 3Ј-portions of the 0.4a HSI fragment. The amplified fragments were gel purified, mixed, and reamplified further with the same external Bluescript primers to yield the full-length mutated 0.4a fragment. This was subsequently cloned into AT vector (Promega). The same strategy was used to mutate POU site 1 separately.
To join the two mutations, a PstI site was used, situated between POU sites 1 and 2 (Fig. 1B); a PstI-HindIII fragment containing POU site 1 mutation was cloned in replacement of the corresponding wildtype sequence in the AT vector plasmid containing the POU site 2 mutation. This generated a 0.4a MUT subclone mutated in both POU sites; the mutated 0.4a fragment was then cloned (as an HindIII fragment) upstream from the Sox2 promoter and ␤-geo (see above, "HSI-Sox Promoter Constructs").
The 0.1 MUT fragment was obtained by PCR from the 0.4a MUT subclone, with primers carrying external HindIII sites (see above) and subcloned upstream from the Sox2 promoter as done for the 0.4a MUT fragment. Transgenes were isolated as SalI fragments.
0.8⌬0.1-Sox Promoter Construct-Portions of the 0.8-kb fragment flanking (5Ј and 3Ј) the 110-bp region to be deleted were amplified separately by PCR. The upstream extremity of the 5Ј-portion was positioned to include an NheI site (Fig. 1A); the downstream extremity of the 3Ј-portion included an NdeI site (Fig. 1A). Both primers flanking the region to be deleted (corresponding to the 0.1 segment in Fig. 1B) carried a terminal EcoRI recognition sequence. After amplification of the upstream (NheI-EcoRI) and downstream (EcoRI-NdeI) fragments, these were inserted (by successive ligations) as an NheI-EcoRI-NdeI fragment, into Bluescript (appropriately modified with an NheI-EcoRI-NdeI linker). This (110-bp deleted) NheI-NdeI fragment was then used to replace the corresponding wild-type sequence in the 0.8-kb HSI fragment.

Transgenic Embryos and X-gal Staining
Transgenic embryos were produced and analyzed by X-gal staining (at E12.5) as described previously (10).

Nuclear Extracts and Electrophoretic Mobility Shift Assay (EMSA)
Nuclear extracts were prepared from ES cells and from dissected embryonic cortex according to the method described in Ref. 21. For ES cell nuclear extracts, one or two subconfluent 10-cm plates were used. For embryonic brain nuclear extracts, 2-3 liters (20 -30 embryos) were used (at embryonic day (E)12.5, or as indicated). The brain was dissected in PBS with protease inhibitors (21) and separated into cortex and into posterior (excluding the spinal cord) plus ventral regions. These were dissociated to single cells by gentle Dounce pestling in PBS with protease inhibitors. The following steps of cell lysis, nuclear pellet preparation, and nuclear extraction were carried out exactly as described, with the exception that a protease inhibitor mixture (1 mg/ml leupeptin, 1 mg/ml pepstatin, 1 mg/ml chymostatin, 1 mg/ml antipain; all from Sigma) was added to buffers A (1:500) and C (1:300).
EMSA was performed according to Refs. 22 and 23. Briefly, protein binding reactions were performed by combining 32 P-labeled doublestranded oligonucleotide probe (15,000 -20,000 disintegrations/min) and 2-4 g of nuclear extract with 1.5 g of poly(dI-dC) and 10 g of bovine serum albumin in binding buffer (10% glycerol, 12 mM HEPES, 4 mM Tris-HCl, 60 mM KCl, 1 mM EDTA, 1 mM dithiothreitol, pH 7.9) and incubating for 20 min at 22°C. For "supershift" reactions, 1 l of the appropriate antibody was added to the complete binding reaction just prior to the addition of the labeled probe. Unlabeled competitor oligonucleotides were added in a 50 -100-fold excess. Electrophoresis was at 150 V on a 5% 29:1 acrylamide:bisacrylamide gel in 50 mM Tris borate, 1 mM EDTA, pH 8.2 (0.5 ϫ TBE), buffer.
The following antibodies were used undiluted for supershift assays: FIG. 1. Sox2 5-regulatory region. A, mouse Sox2 5Ј-flanking region. The thick line (from the 5Ј-end to H) is the DNA region required for telencephalic transgene expression, which comprises the chromatin DNase I-hypersensitive site HSI (vertical arrow) and the PstI-sensitive site P* (10). The thin line (from the H site to the Sox2 gene) is the Sox2 minimal promoter, which is inactive, by itself, in the central nervous system (10). goat anti-Brn1, goat anti-Brn2, goat anti-Brn3, rabbit anti-Oct1 (all Santa Cruz Biotechnology), and anti-Oct4 (24).
The amplified fragments were run on an 8% 29:1 acrylamide:bisacrylamide gel and eluted. DNase I footprinting was essentially according to Ref. 25, with some modifications. Each probe (about 15,000 disintegrations/min) was mixed with 0.6 or 1.8 g of nuclear extract, 10 g of bovine serum albumin, 1.5 g of poly(dI-dC) in footprinting buffer (10% glycerol, 12 mM HEPES, 4 mM Tris-HCl, 60 mM KCl, 1 mM EDTA, 1 mM dithiothreitol, pH 7.9) in a total of 20 l for 40 min on ice. In control reactions, the nuclear extract was omitted. At the end of the reaction, 50 l of footprinting buffer containing 2.5 mM MgCl 2 and 2.5 mM CaCl 2 was added just before the addition of 6 l of DNase I (0.4 -0.8 g, RQ1 DNase; Promega). Control reactions without nuclear extracts received 0.2-0.4 g of DNase I. Digestion proceeded for exactly 2 min at 22°C before being stopped with 70 l of stop buffer (0.6 M NaCl, 20 mM Tris-HCl, 20 mM EDTA, 1% SDS, pH 7.5). After phenol-chloroform extraction and ethanol precipitation, the DNA was resuspended in formamide loading dyes, heated 5 min at 90°C, and analyzed on a 6% acrylamide:bisacrylamide (19:1), 7 M urea sequencing gel.

ES Cell Transfection (10)
CCE ES cells were electroporated at 10 7 cells/0.55 ml of PBSCM (PBS ϩ 1 mM CaCl 2 , 1 mM MgCl 2 ) with a Bio-Rad gene pulser at 220 V, 960 microfarads, with 4 g of purified transgene DNA. After 5 min on ice, cells were plated on two 10-cm dishes onto STO feeder layers. G418 selection at 300 g/ml was applied after 24 h. Resistant clones were counted at day 10 of selection.

PCR on Immunoprecipitated Cross-linked Chromatin
PCR amplifcation was performed in a final volume of 50 l, using 1 ng of genomic ES (E14) DNA or 1 ng of immunoprecipitated cross-linked chromatin as template, 10 pmol of each primer, 2.5 mM dNTPs (Amersham Biosciences), 2.5 units of TaqDNA polymerase (Invitrogen), and 1ϫ PCR buffer (Invitrogen). The PCR consisted of 3 min at 94°C for one cycle and then 33 cycles at 94°C for 30 s, 54°C for 45 s, and 72°C for 45 s, and 1 cycle of 72°C for 10 min. Upstream and downstream primer pairs were as follows: Sox2, 5Ј-ATGAAATTAAAACTTTTGGAACC-3Ј and 5Ј-CTTAGTGGCAAATAGAGGTT-3Ј; ␤-actin, 5Ј-GGTCAGAAG-GACTCCTATGT-3Ј and 5Ј-ATGAGGTAGTCTGTCAGGTC-3Ј. (10) we showed that a 5.7-kb region 5Ј to mouse Sox2 directs transgene expression to telencephalic neural stem cells and precursors; we further demonstrated that this expression requires the most 5Ј 2.5 kb of this region, which contains a cell type-specific chromatin DNase I-hypersensitive site, termed Sox2 HSI, which we identified in neural and ES cell nuclei (10) (Fig. 1A). Two more DNase I-hypersensitive sites (HSII and HSIII) lie in the Sox2 minimal promoter, inactive by itself in the central nervous system (10). The comparison of genomic sequences 5Ј to Sox2 in mouse, human, and chick identifies three conserved regions, of about 400, 110, and 90 bp, respectively ( Fig. 1; see also Ref. 20), the last corresponding to the transcriptional start site region. The position of the extensive most 5Ј-region closely matches the DNase I and PstI chromatin-hypersensitive site, HSI (10) (Fig.  1). This homology thus lies within the region required for transgene expression in the brain.

A Conserved 400-Base Pair Element 5Ј to Sox2 Is Required and Sufficient to Direct Transgene Expression to Brain Neural Precursors-In previous experiments
We asked whether smaller fragments comprising HSI are sufficient to direct telencephalic expression of the ␤-geo reporter gene in transgenic assays (Fig. 2). A 400-bp fragment, centered on HSI (and comprising part of the 5Ј-homology region), was linked to the minimal Sox2 promoter, inactive in the central nervous system (10; Fig. 1A). This construct (0.4-Sox pr., Fig. 2A and 2B, image 2) is sufficient to direct telencephalic transgene expression. To assess whether telencephalic activity requires cooperation with the Sox2 3.3-kb promoter region or can be specified autonomously by the HSI region, we also fused the latter region (as progressively smaller fragments of 3 kb, 800 bp, or 447 bp, the last including the complete homology region) to the heterologous TK minimal promoter, to generate the 3-, 0.8-, 0.4b-TK pr. constructs ( Fig. 2A and 2B, images [3][4][5][6]. The 3-kb and 800-bp fragments, linked to the TK promoter, drive high level telencephalic activity (Fig. 2B, images 3  and 4). Similarly, the 447-bp fragment drives expression in the telencephalon, although at a lower level (Fig. 2B, images 5 and  6), together with some ectopic activity (in one embryo, Fig. 2B,  image 5). These experiments indicate that the HSI element is sufficient to specify telencephalic transgene expression, in the absence of any additional neural regulatory element.
110 Base Pairs within the HSI Element Direct Brain-specific Expression-To identify functionally important sequences, we tested smaller fragments within the HSI element. A 110-bp fragment, centered on a site that is accessible to PstI digestion in telencephalic neural cell chromatin (10) (Fig. 1), was linked to the Sox2 minimal promoter (0.1-Sox pr. construct, Fig. 2). This construct is still able to drive high level telencephalic Sox2 Regulation in Stem Cells expression ( Fig. 2A and 2B, image 7). Notably, however, the expression domain is slightly expanded posteriorly, to include the dorsal mesencephalon (Fig. 2B, image 7).
We subsequently tested the same 110-bp sequence linked only to the TK minimal promoter (0.1-TK pr. construct, Fig. 2A  and 2B, image 8). Again, expression was detected, although at very low levels, in the mesencephalon and telencephalon, particularly along the medial region (Fig. 2B, image 8). We conclude that the 110-bp element is able to specify brain-specific transgene expression, acting more efficiently when linked to the Sox2 promoter region.
Octamer Sites within the 110-Base Pair Core Region Are Bound by CNS-specific POU Transcription Factors-The 110-bp Sox2 enhancer defined in vivo was mapped in vitro for binding sites recognized by nuclear proteins of the brain. Fig. 3 shows a footprinting assay on DNA probes containing the 110-bp region and spanning the entire 400-bp enhancer. Two regions are protected from DNase I digestion by telencephalic nuclear extracts from E12.5 dorsal telencephalon (Fig. 3A). These protections center on two octamer consensus sequences (Figs. 1 and 3A); these sequences are recognized by members of the POU transcription factor family, some of which are expressed in brain (25). By contrast, no protection is detected using extracts from non-neural cells (CH27 lymphoid cells, Fig.  3A, and data not shown) in this assay.
Two probes encompassing the protected regions were subsequently used in EMSAs (Fig. 4). Both probes form two major complexes with embryonic brain nuclear extracts at E12.5 (Fig.  4, A and B), E11.5 and E10.5 (not shown). The upper complex is more abundant in extracts from dorsal telencephalon, and the lower one in extracts from more posterior central nervous system regions (Fig. 4, A and B; see also Fig. 6A). Non-neural cell extracts (from lymphoid CH27 cells) generate a much slower complex and a band with mobility similar to the upper brain band (Fig. 4, A and B); however, this band is generated by a protein different from the neural factor (see below).
Upon incubation with antibodies specific for the neural POU transcription factors Brn1 and Brn2, a specific decrease of the intensity of the higher complex (with anti-Brn1), or of the lower complex (with anti-Brn2), with concomitant appearance of more slowly migrating bands (Fig. 4, asterisks) was observed (Fig. 4, A and B). This supershift is caused by the formation of a complex between the factor bound to DNA and the antibody; no such supershift was observed with an antibody against Brn3. The mobility of the CH27 band comigrating with the upper telencephalic band was not modified by any of the Brn antibodies, indicating that this band is the result of a different protein.
The intensity of both complexes observed with brain extracts is strongly diminished by competition of the binding of the labeled oligonucleotide probe with nonlabeled oligonucleotides carrying canonical octamer-POU binding sites as well as by a previously characterized POU factor binding site in the neural precursor-specific enhancer of the nestin gene, which is essential for the activity of this enhancer in vivo (25) (Fig. 4C).
Mutations by base substitutions within the octamer binding sequences of POU site 1 or POU site 2 probes (shown in Fig. 1B) abolish binding of both complexes (Fig. 4D).
Octamer Mutations Reduce Neural Transgene Expression in Vivo -To assess the functional importance of the POU binding sites in brain Sox2 transgene expression, point mutations that abolish transcription factor binding (Fig. 4D) to both POU sites 1 and 2 were introduced in the 400-bp, as well as in the 110-bp elements, which were linked to the Sox2 minimal promoter upstream from ␤-geo and assayed in transgenic embryos (0.4aMUT-and 0.1MUT-Sox pr. constructs; Fig. 2A and 2B, images 9 and 10). As shown in Fig. 2, a reduction in activity is observed in both cases; this is seen both as quantitative expression levels (compare Fig. 2B, images 2 and 9 and images 7 and 10) and number of integrations showing detectable specific expression ( Fig. 2A). However, brain-specific expression is not abolished, as can be seen with both mutated constructs (Fig.  2B, images 9 and 10).
HSI Sequences Outside the 110-Base Pair Core Region Are Also Able to Specify Telencephalic Transgene Expression-Because the 110-bp core region had proven able to specify high level brain-specific transgene expression from the minimal Sox2 promoter (0.1-Sox pr. construct; Fig. 2B, image 7), we next asked about the consequences of deleting this region from the full size HS1 element. In the 0.8⌬0.1-TK pr. construct ( Fig. 2A), the 110 bp were deleted from the 0.8-TK pr. construct, which had been shown to drive high level, brain-specific transgene expression ( Fig. 2A and 2B, image 4; see above). Surprisingly, the 800-bp HSI enhancer deleted of the 110-bp central region is still able to specify telencephalic transgene expression (Fig. 2B,  image 11). Thus, HS1 sequences both within the core 110 nucleotides (see above) and in the surrounding region, are able to specify telencephalic transgene expression independently.

The Sox2 HSI 400-Base Pair Element Is Active in ES Cells, and This Activity Depends on the Integrity of the Octamer
Binding Sites-We found previously that the full 5.7-kb Sox2 5Ј-flanking region drives transgene expression already in the blastocyst inner cell mass (10). Further, DNase I-hypersensitive sites are detected in ES cell chromatin, including the HSI region (10). We thus sought to determine whether the HSI element active in neural cells after gastrulation (see above) is also active in ES cells. To this end, the 0.4a-Sox pr. construct, previously found to be active in neural cells in transgenic embryos after gastrulation (Fig. 2), was stably transfected into ES cells. Exploiting the bifunctional nature of the ␤-geo reporter gene (encoding lacZ and G418 resistance), the number of G418-resistant colonies was taken as a measure of the activity of the linked regulatory sequences. As shown in Fig. 5, the 400-bp element efficiently stimulates activity in ES cells, compared with the promoterless ␤-geo control and with the minimal 3.3-kb Sox2 promoter cloned upstream from ␤-geo. However, when the octamer-mutated version of the enhancer was

Sox2 Regulation in Stem Cells
assayed (0.4aMUT-Sox pr. construct, see Fig. 2), this activity was reduced to only 17% of the nonmutated control, marginally above the level observed with the 3.3-kb promoter only, and with the promoterless control (Fig. 5). In cells that do not express Sox2, e.g. NIH3T3, both the wild-type and the octamermutated constructs were essentially inactive, with the same very low expression levels regardless of octamer integrity (not shown).

Different POU Transcription Factors Bind to the Sox2 HSI Octamers in ES Cell
Nuclear Extracts-EMSAs were done to investigate proteins binding to the HSI octamer sequences in ES cell nuclear extracts (Fig. 6), using the same probes previously used with neural cells (see above, Fig. 4). Two major complexes are seen with both probes: a higher mobility one, observed specifically in ES cells, and a lower mobility one (Fig.  6A). The latter comigrates with the most abundant complex formed by these probes in cells (such as STO fibroblasts, Fig.  6A, or CH27, see Fig. 4A), which do not express Sox2, and where HSI is not accessible. These ES cell complexes differ from those seen with neural cells (Fig. 6A). Both complexes are competed by an unlabeled "canonical" octamer binding site, known to be bound by the POU factors Oct4, Oct1, and Oct6 in ES cells (26), as efficiently as by the Sox2 gene octamer sequence itself (Fig. 6B).
Specific antibodies were used to investigate these complexes further (Fig. 6C). Incubation with an antibody recognizing the ubiquitous Oct1 POU transcription factor results in absence of the slower complex (Fig. 6C). The faster complex is abolished by an antibody specifically recognizing Oct4 (Fig. 6C), a POU transcription factor specifically found in ES cells (and their in vivo counterpart, the inner cell mass and epiblast) and in germ cells.
Chromatin Immunoprecipitation from ES Cells Demonstrates in Vivo Binding of Oct4 to the HSI Element-To assess whether Oct4 does bind to HSI in vivo, we immunoprecipitated ES cell chromatin with an anti-Oct4 antibody (24), and we amplified DNA from the precipitated chromatin with primers for the Sox2 HSI region, flanking the Oct4 binding sites, and  Fig. 1B); asterisks mark nucleotides within these sequences whose mutated version has been functionally assayed (see text). B, diagram showing probes ␣ and ␤ and the footprinted regions (filled bars) seen with neural extracts in A, with respect to the 0.4-kb enhancer element. Round dots mark the labeled end of the probes used in A. In probe ␣, the forward strand was labeled; in probe ␤, the reverse strand. Sequence numbering is as in Fig. 1B. Filled bars represent the strongest footprints, dotted lines connecting them show a region that is also protected, although more weakly (see A: probe ␣ also shows some protection upstream from position 387 (left panel), and the footprint seen with probe ␤ appears to continue beyond the PstI site (position 368, right panel)).

FIG. 4. The two POU sites in the Sox2 enhancer are recognized by neural POU transcription factors.
A, EMSA of a probe encompassing POU site 1. With E12.5 dorsal telencephalic nuclear extracts (dors. tel.) or E12.5 extracts of brain devoid of dorsal telencephalon (res. brain), two major DNA-protein complexes are seen (black and white arrows at the left of the figure), of which the upper one is stronger with dorsal telencephalic extracts, and the lower one with residual brain extracts. EMSA with CH27 lymphoid cells shows two different complexes (black and white arrowheads at the right of the figure). The addition of antibody against Brn1 disrupts the upper complex in neural cells (lanes 2 and 7) and produces a supershift, stronger in lane 2 (asterisks); no change is observed in CH27 (lane 12). Brn2 antibody disrupts the lower band in neural cells (lanes 3 and 8), generating a supershift (asterisk; stronger in lane 8) and has no effect in CH27 (lane 13). Anti-Brn3 antibody does not affect any of the gel shifts (lanes 4, 9, and 14). B, EMSA of a probe encompassing POU site 2 with E12.5 neural extracts (lanes 1-4) or with non-neural CH27 (lane 6). Two complexes are seen with neural extracts (black and white arrows on the left of the panel) of the same mobility as those seen with POU site 1 probe (compare lanes 3, 4, and 5). The addition of Brn1 or Brn2 antibodies supershifts (asterisks) the upper complex (lane 1) or the lower complex (lane 2), respectively. Non-neural CH27 cells (lane 6) shows two different complexes of the same mobility as those seen in A with POU site 1 (black and white arrowheads). C, binding of neural proteins to POU sites 1 and 2 is competed efficiently by an octamer site in the nestin gene neural precursor-specific enhancer (25) (lanes 5-7 and 16 -18). The competition efficiency is similar to that of the self competitors (lanes 2-4 and  13-15). A nonrelated oligonucleotide (GATA-1-binding oligonucleotide), by contrast, does not compete in the same concentration range (lanes 9 -11  and 20 -22). D, point mutations in POU sites 1 and 2 abolish nuclear protein binding. Oligonucleotides carrying nucleotide substitutions in the octamer sequences (see Fig. 1B) no longer form the protein-DNA complexes (compare lane 4 with lane 1, and lane 10 with lane 7) and fail to compete for binding to the nonmutated probes (lanes 2 and 3 and lanes 8 and 9). wt, wild-type.
with primers for a control gene (actin), not regulated by Oct4 (Fig. 7). The HSI region was amplified efficiently from the immunoprecipitated chromatin, whereas the actin region was not amplified (Fig. 7B). As a further control, preimmune serum was unable to immunoprecipitate both the HSI and actin regions (Fig. 7C). These results indicate that Oct4 is bound to the HSI element in vivo in ES cells.

DISCUSSION
In previous work, we identified a regulatory region of the Sox2 gene which drives reporter gene expression in transgenic mice to neural stem and precursor cells within a specific subset of the panneural Sox2 expression domain, the telencephalon (10). In this paper, we have restricted the critical region for Sox2 telencephalic expression and identified a sequence encompassing a cluster of POU factor binding sites. Mutation of these sites considerably decreases transgene expression. Interestingly, the same mutation affects expression of the transgene in ES cells. These results highlight the role of POU factors in Sox2 regulation, both in totipotent cells and in early neural precursors.
Sox2 Regulatory Sequences Involved in Telencephalic Expression-In vivo work in mouse, Xenopus, and chicken (10,11,14,15,18,19) has suggested important functional roles for Sox2 in stem/precursor cells, ranging from the highly multipotent early epiblast cells to neural stem cells of the adult brain. In the nervous system, Sox2 is expressed during embryogenesis along the entire neural tube. We showed previously (10) that a 5.7-kb fragment, encompassing the Sox2 promoter, drives reporter gene expression in transgenic embryos exclusively in the telencephalon; this suggested that Sox2 regulatory sequences may respond in the telencephalon to local signals. In this work, we sought to identify the shortest Sox2 regulatory regions required for the specification of telencephalic expression.
By linking progressively shorter fragments of DNA surrounding the HSI site (0.8-, 0.4-, 0.1-Sox pr. constructs, Figs. 1 and 2A) to the Sox2 minimal promoter (10), we have shown that 110 bp (centered on the PstI site shown to be hypersensitive in chromatin; Ref. 10) drive, in conjunction with the Sox2 promoter, strong reporter expression to the telencephalic region (Fig. 2B, images 1, 2, and 7). Because the Sox2 3.3-kb promoter alone is not able to direct telencephalic expression (10), these results suggest that the HSI sequences are sufficient to specify telencephalic expression. However, we cannot rule out that sequences within the minimal Sox2 promoter may be required to cooperate with HSI and potentiate its activity. To test this point, we linked HSI fragments to a TK minimal promoter (inactive in the telencephalon, 25). The HSI fragments (3, 0.8, and 0.4 kb) still drive clear (Fig. 2B, images 3-6) reporter gene expression to the telencephalon. This proves that the HSI fragment contains sequences that are sufficient by themselves to specify regional expression in neural cells. The 110-bp fragment is also active in the telencephalon, when linked to the TK promoter, although to a much lesser extent than when linked to the Sox2 promoter (Fig. 2B, image 8). These results further indicate that, although the 110-bp sequence is sufficient to specify telencephalic expression, cooperation with promoter sequences (Fig.  2B, image 7) or with other sequences within the HSI region (Fig.  2B, images 4 and 5) increases the overall efficiency.
An interesting and unexpected observation is that restriction to telencephalic expression requires more than the 110-bp fragment, as indicated by the relatively strong mesencephalic activity detected with the constructs including the 110-bp element (Fig. 2B, images 7, 8, and 10; see also below) and the absence of such expression with all other constructs. These results indicate that the HSI region contains sequences negatively regulating the activity of the core HSI 110-bp element.
The 110-and 400-bp HSI elements are, to our present knowledge, the only short enhancer sequences able to drive specific expression to the telencephalon.
POU Binding Sequences in the HSI 110-Base Pair Region Contribute to Transgene Expression-Within the core 400-bp HSI region, in vitro footprinting (Fig. 3) and subsequent in vitro EMSAs (Fig. 4) reveal prominent binding of neural specific POU factors, such as Brn1 and Brn2, to conserved binding sites, here named POU1 and POU2 sites. Similar sites, bound by the same factors in neural cells extracts, are known to contribute to the level of expression of the nestin and the BFAP genes in neural regions of transgenic mice (25). Mutation of POU sites reduces both the number of expressing embryos and the level of reporter gene activity, within the 0.4-and 0.1-Sox pr. constructs ( Fig. 2A and 2B, compare images 2 and 9, and images 7 and 10). Thus, POU factors binding to the 110 bp are likely responsible for at least part of the HSI activity. It should be noted that other potential POU-binding elements, although not detected as in vitro footprints (not shown), lie both 5Ј and 3Ј to the core region, within the 0.8-kb fragment, and might contribute to the overall activity of the HSI region.
A Modular Organization of the Sox2 Regulatory Region?-Using relatively large Sox2 constructs, consisting of 5.7-12 kb of promoter-enhancer, including HSI, the majority of integrations within transgenic mice was specifically active in the telencephalon, with little or no ectopic expression (10). By dissecting the region to fragments as small as 0.1 kb, telencephalic expression is maintained, even with the smallest fragment (0.1 kb) (see above), although the proportion of active integrations, and/or the level of expression, may be decreased, and some ectopic expression is observed. These effects may be the result of the smaller size of the constructs, allowing stronger position of integration effects. An additional explanation might be that the Sox2 promoter-enhancer region may consist of several, partially redundant modules, able to specify telencephalic expression independently (see above) and to interact to promote optimal activity. Indeed, the 110-bp sequence that is sufficient for telencephalic expression (Fig. 2B, images 7 and 8) is clearly not strictly required because its deletion from the 0.8-kb region is still compatible with specific telencephalic expression (Fig.  2B, images [11][12][13]. A number of data in the literature suggest a modular organization of the Sox2 regulatory sequences, in which multiple partially redundant elements are spread along the gene and control expression level and region-specific activity. Although Sox2 is expressed from the endogenous gene along the entire neural tube, the HSI enhancer is active specifically in the telencephalon in transgenic mice (Ref. 10 and present paper). Other elements may participate in regulating telencephalic expression because a targeted deletion of the HSI enhancer from the endogenous mouse Sox2 gene does not completely abolish Sox2 telencephalic expression, but rather reduces it by about 50% in neural stem/precursor cells from adult and embryonic mouse brains (15). In keeping with this, Uchikawa et al. (20) reported several distinct regulatory elements along the chicken Sox2 gene, active at distant and/or overlapping neural tube regions during development, and Tomioka et al. (27) and Miyagi et al. (28) demonstrated, by transfection and by retroviral transduction assays, a Sox2 enhancer 3Ј to the mouse Sox2 gene itself, active in neural cells. Our present dissection of the HSI region points to further functional heterogeneity even within a short DNA sequence active within a restricted brain region.
HSI POU Factor Binding Sites Are Essential for Sox2 Transgene Expression in ES Cells-ES cells represent an "in vitro" counterpart of early epiblast stem cells, the highly multipotent precursors to all embryonic cell types (including neural cells). Sox2 is essential, in mouse, for growth and maintenance of these cells (11), and the phenotype of Sox2 Ϫ/Ϫ mutants has some resemblance to that of Oct4 Ϫ/Ϫ embryos (29). The regulation of the expression of Sox2 is still poorly understood; however, we reported previously that the same transgenic con-  1B) (lanes 2 and 4) no longer show the complexes with Oct4 (white arrow) and Oct1 (black arrow) seen with the corresponding wild-type probes (lanes 1 and 3). struct that is highly expressed in neural stem cells is also active, prior to that stage, in the blastocyst inner cell mass and epiblast (10; see also 27).
Starting from our observation that POU-binding elements are important in neural expression of Sox2, we asked whether the same sites are recognized by Oct4, a POU factor that is expressed exclusively in inner cell mass and epiblast, germ cells, and very early embryonic tissues (29).
EMSA experiments (Fig. 6, A and B) with ES cell nuclear extracts, in combination with the use of specific antibodies (Fig.  6C), show that at least two different proteins, Oct4 and Oct1, bind to the POU site 1 and site 2 probes. Oct1 is ubiquitous, and interestingly, there is little, if any, Oct1 binding to these sites with neural nuclear extracts (Figs. 4 and Fig. 6A, lanes 9  and 10). Oct4 is specifically expressed in the blastocyst inner cell mass (and in in vitro derived ES cells) and in germ cells (30 -32). Sox2 is highly expressed in the inner cell mass and in ES cells (11). The same POU site mutations that affect neural expression of the Sox2 constructs both abolish Oct4 and Oct1 in vitro binding and drastically decrease (by more than 80%) the activity of the 0.4 Sox pr. construct in stable transfections in ES cells (Fig. 5).
These data suggest that Oct4 (and/or Oct1) are important for Sox2 expression in ES cells.
Transfection experiments in ES cells previously showed a dependence on the cooperation between Sox2 itself and Oct4 for the autoregulation of a different Sox2 enhancer, located downstream from the gene (27). To evaluate the in vivo relevance of Oct4 to Sox2 expression, we performed chromatin immunoprecipitation with anti-Oct4 antibodies; indeed, Oct4 was found to be present on the HSI region of the endogenous Sox2 gene in ES cell nuclei, thus implying a functional in vivo role (Fig. 7). The importance of Oct4 for optimal Sox2 expression is further suggested by the observation that Oct4-deficient ES cells have a lower level of Sox2 mRNA than wild-type cells (33).
The Same Element May Be Sequentially Recruited to Direct Sox2 Expression in Totipotent ES Cells and Multipotent Neural Cells-Both Oct4 and Sox2 are highly expressed in the inner cell mass and are essential for proliferation/survival, maintenance of the totipotent state and commitment to differentiation to appropriate lineages of these cells (11,29). At later stages, Oct4 is repressed in the embryo except in primordial germ cells (30 -33). In contrast, there is a continuum of Sox2 expression between epiblast cells, definitive ectoderm, neural plate, germinative layers of the neural tube, and finally neural embryonic and adult stem cells (10,11,13,15).
Our experiments (Figs. 5 and 7) indicate that Oct4 may contribute to Sox2 expression in totipotent cells (ES cells and presumably inner cell mass cells), acting at HSI, possibly in cooperation with the 3Ј-enhancer defined by Tomioka et al. by transfection (27). In neural cells, the same POU sites control the level of Sox2 expression by binding a different set of factors, including Brn1 and Brn2. When does the replacement of Oct4 with neural factors occur? Recent in vitro experiments with ES cells induced to neural differentiation showed that the level of Oct4 is important for in vitro neurogenesis; suppression of Oct4 abolished neuronal ES differentiation, whereas up-regulation of Oct4 enhanced neurogenesis (6). Thus, Oct4 remains important also for the earliest stages of neurogenesis. Indeed, Oct4 is down-regulated after E7.5 (neural plate stage) and becomes restricted exclusively to primordial germ cells by E8.5-8.75 (30,32,34). A role for the POU2 zebrafish homolog of Oct4 in controlling the level of expression of the neural Pax2.1 gene in zebrafish embryos at the midbrain-hindbrain boundary has been suggested by analyses of the spiel-ohne-grenzen mutant (in which POU2 is inactivated) (35,36). Similarly, Oct4 binds to a POU motif essential for full activity of an early enhancer of the mouse Pax2 gene (37). Taken together, these results suggest a role for Oct4 in early neural cells. Brn2 first appears at E8.5 in the neural plate, close to the time of Oct4 disappearance (38), whereas Brn1 is first detected at E10 (39). We would like to hypothesize that these factors take on, in early neural cells, the role played by Oct4 in the preceding stages of Sox2 expression and further regulate it by interaction with additional neural factors.
Other examples that a single regulatory element within a gene may play a role in gene expression in different cell types have been reported previously (40,41). For example, Katsuoka et al. (40) showed that the same GATA motif within the mafK enhancer is required for transcription in both hematopoietic and cardiac cells, presumably recruiting different members of the GATA family of transcription factors. In these cases, however, gene regulation by the same element was demonstrated in cells belonging to different, but not sequentially related, embryonic lineages.