Neural crest patterning: autoregulatory and crest-specific elements co-operate for Krox20 transcriptional control

The patterning mechanisms controlling craniofacial morphogenesis constitute and important issue and an interesting paradigm for the study of vertebrate development. Numerous studies have demonstrated the crucial roles of the hindbrain and its associated neural crest in this process(Lumsden and Krumlauf, 1996;Le Douarin and Kalcheim,1999). During early development, the hindbrain undergoes a transient segmentation which is visible at the morphological level by the appearance of 7 bulges known as rhombomeres (r)(Lumsden and Keynes, 1989;Lumsden, 1990). Each rhombomere consists of a polyclonal, cell lineage restricted, anteroposterior(AP) compartment that expresses a specific set of genes and behaves as a developmental unit (Fraser et al.,1990; Birgbauer and Fraser,1994; Lumsden and Krumlauf,1996). The cranial neural crest (NC) arises at the border between the neural plate and surface ectoderm. It comprises a pluripotent cell population that delaminates just prior to and following neural tube closure in the mouse and chick, respectively, and, after migration, differentiates into various tissues including ganglia, cartilage, bone and connective tissue(Le Douarin, 1982; Noden,1988). In register with the rhombomeric organisation, the cranial NC cells migrate in three segmented streams adjacent to r2, r4 and r6(Anderson and Meier, 1981;Lumsden et al., 1991;Serbedzija et al., 1992;Liem et al., 1995;Selleck and Bronner-Fraser,1995) contributing to the V, VII/VIII and IX ganglia and colonising the 1st, 2nd and 3rd branchial arches, respectively(Le Lièvre and Le Douarin,1975; Noden, 1983;Lumsden et al., 1991;Köntges and Lumsden,1996). r3 and r5 produce fewer NC cells than even-numbered rhombomeres and these migrate rostrally and caudally to join the adjacent streams (Sechrist et al.,1993; Lumsden et al.,1991; Trainor et al.,2002a).

Previous studies, involving transplantations of neural tube or neural fold,led to the conclusion that the NC exerts an instructive role in the specification of the regional identity of the branchial arches(Le Douarin, 1982;Noden, 1983; Noden, 1988). Furthermore, the implication of Hox genes in the AP patterning of both rhombomeres and NC (Gendron-Maguire et al., 1993; Rijli et al.,1993; Zhang et al.,1994; Barrow and Capecchi,1996; Goddard et al.,1996; Studer et al.,1996; Gavalas et al.,1997; Manley and Capecchi,1997), together with correlated Hox codes between NC cells and their rhombomeric region of origin (Hunt et al., 1991a; Hunt and Krumlauf, 1991), led to the elaboration of a pre-patterning model for craniofacial development: NC cells acquire positional information (Hox gene pattern), according to their AP level of origin in the hindbrain, which is passively carried to peripheral tissues where it is used to instruct ectodermic, mesodermic and endodermic tissues in co-ordinated craniofacial development (Hunt et al.,1991b; Hunt et al.,1991c). This view has been recently tempered by several observations, including: (i) NC plasticity in Hox gene expression(Schilling et al., 2001;Trainor and Krumlauf, 2001;Couly et al., 2002;Trainor et al., 2002b); (ii)independent Hoxa2 gene regulation in the hindbrain and in the NC(Prince and Lumsden, 1994;Maconochie et al., 1999);(iii) patterning roles of the isthmus, cranial mesoderm and foregut endoderm(Trainor and Krumlauf, 2000;Couly et al., 2002;Trainor et al., 2002b); (iv)acquisition of AP specification by the branchial arches in the absence of NC input (Veitch et al., 1999;Gavalas et al., 2001). Together these data suggest that the NC cells are able to respond to signals from the environment in which they migrate and that the regional specification of the branchial arches involves a complex integration of patterning information from different tissues.

In spite of this new evidence, the contribution of the NC to craniofacial development is likely to depend in part on prepatterning events that define NC migration pathways and AP identity. We are investigating this question in the case of the Krox20 gene. Krox20 encodes a zinc finger transcription factor that is expressed in r3 and r5, and has been shown to play a key role in hindbrain segmentation: it controls and co-ordinates the formation of the r3 and r5 territories and their acquisition of odd-number characteristics (Schneider-Maunoury et al., 1993; Schneider-Maunoury et al., 1997; Giudicelli et al., 2001; Voiculescu et al.,2001). These roles are exerted through the transcriptional control of many genes, including Hox genes and members of the Eph receptor gene family(Sham et al., 1993;Vesque et al., 1993;Nonchev et al., 1996;Theil et al., 1998;Giudicelli et al., 2001;Manzanares et al., 2002). Together with their ephrin ligands, the Eph receptors prevent cell mingling between adjacent rhombomeres and are involved in controlling NC migration in separate streams (Bergemann et al.,1995; Gale et al.,1996; Smith et al.,1997). In the NC, Krox20 expression is mainly restricted to cells migrating lateral to r6 towards the third branchial arch and presumably originating from r5 in the mouse(Schneider-Maunoury et al.,1993). Krox20 is rarely expressed in the NC cells migrating rostral from r5 and has not been detected in those produced in r3(Schneider-Maunoury et al.,1993). Indeed, fate-tracing analyses have not revealed any significant contribution of Krox20-expressing cells to the r3 NC(Voiculescu et al., 2001). In the r5-derived NC, Krox20 controls the transcription of several key regulatory genes, including Hoxa2, Hoxb2, Hoxb3 and EphA4(Sham et al., 1993;Nonchev et al., 1996;Manzanares et al., 2002;Theil et al., 1998), and may therefore play an important patterning role.

In order to understand the regulatory network controlling Krox20in the NC, we have searched for cis-acting regulatory elements responsible for this aspect of its expression. In this paper we present the identification and characterisation of a neural crest-specific element (NCE) located 26 kb upstream of the mouse Krox20 gene that drives specific expression in the r5 stream of the NC in transgenic mice. The NCE was shown to require both Krox20 and putative high mobility group (HMG) box binding sites for proper activity. The HMG box factor Sox10, expressed specifically in the NC, was found to co-operate with Krox20 to activate the NCE and to be required for the maintenance of Krox20 expression in the r5-derived NC. This organisation suggests a mode of patterning of the r5-derived NC according to its rhombomeric origin: following the initial Krox20 expression in the premigratory NC under the control of cis-regulatory elements active in r5,Krox20 and crest-specific Sox proteins combine to activate the NCE,maintaining its expression in the delaminating NC.

The Krox20^Cre(Voiculescu et al., 2000),Krox20^lacZ(Schneider-Maunoury et al.,1993) and Sox10^Dom(Herbarth et al., 1998;Southard-Smith et al., 1998)alleles and the -31/+7 Krox20/lacZ transgenic line(Ghislain et al., 2002) have been described previously. These lines were maintained in a mixed C57B16/DBA2 background. Genotyping was performed by PCR on extracted DNA using primers described previously (Schneider-Maunoury et al., 1993; Voiculescu et al., 2000; Ghislain et al.,2002) except for the Sox10^Dom allele, which was genotyped as follows. PCR primers:5′-ctccggatgcagcacaaaaaggacc-3′ (forward; T4 polynucleotide kinase labelled using [γ-³²P]ATP) and 5′-cctgaatagcagcagcccctccttg-3′ (reverse) were used to amplify sequences containing the extra nucleotide at position 929 of theSox10^Dom allele(Southard-Smith et al., 1998). The product of the wild-type and mutant alleles, 137 and 138 bp, respectively,were resolved by denaturing gel electrophoresis.

The pEMBL2 cosmid clone 8/4 (Chavrier et al., 1989) was used as a source of murine Krox20extragenic sequences. Subfragments were cloned upstream of a minimalβ-globin promoter/lacZ reporter in plasmid pBGZ40(Yee and Rigby, 1993), as described previously (Ghislain et al.,2002). Site-directed mutagenesis of Krox20 binding sites was performed on fragment #11 cloned in pBS using the Transformer™Site-Directed Mutagenesis kit (Clontech). The oligonucleotides used were:5′-ggaggaaagcgtcggttgcaggcagg-3′ (binding site 1);5′-gcctcgaagaagtcggcgggagcctc-3′ (binding site 2);5′-ggaattcgatatctctagaatcgataccgtc-3′ (selection). Deletion of the putative HMG box binding sites was performed using the Exsite™ PCR-based site-directed mutagenesis kit (Stratagene) with the following oligonucleotides: 5′-ggctggggacgggca-3′ (forward);5′-tgttctcctttccctgtctcagcag-3′ (reverse). All mutated fragments were verified by sequencing and cloned into pBGZ40. Multimerisation of sequences containing the putative HMG box binding sites (fragment #24) was performed on double-stranded oligonucleotides containing terminalSpeI sites by ligation into pBS. A 7× multimer was selected and cloned into pBGZ40.

Transgenesis and identification of transgenic embryos by PCR was performed as described previously (Sham et al.,1993; Ghislain et al.,2002). In ovo electroporation in the chick neural tube was performed as described previously(Giudicelli et al., 2001). Krox20 transactivation studies were performed by coelectroporating thelacZ reporter constructs with a mouse Krox20 expression plasmid, pAdRSV Krox20(Giudicelli et al., 2001).

Whole-mount in situ hybridisation was performed as described previously(Wilkinson et al., 1992). Alkaline phosphatase activity was revealed using the NBT/BCIP substrate (Roche). Antisense RNA probes were prepared from aKrox20 cDNA (Wilkinson et al.,1989) and from Sox9 and Sox10 est clones obtained from the MRC geneservice (IMAGE ID 4165469 and 3675437,respectively). Mouse and chick embryos were stained for β-galactosidase activity in toto following fixation in 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) for 15 minutes at 4°C. Staining was performed in 1 mg/ml X-gal, 5 mM K₃Fe(CN)₆, 5 mM K₄Fe(CN)₆, 2 mM MgCl₂ and 0.1% Tween-20 in PBS at 30°C for 15 hours. For staining prior to in situ hybridization embryos were incubated for a maximum of 4 hours in the staining solution containing 0.02% PFA. All tissues were postfixed in 4% PFA in PBS. Embyos for sectioning were dehydrated and embedded in paraffin wax and 10 μm sections were prepared. Sections were stained with Nuclear Fast Red at 1 mg/ml for 10 seconds, dehydrated and mounted in Eukitt (Merck).

A pBS-HindIII sub-library of a chicken genomic BAC clone carrying the Krox20 gene and flanking sequences(Giudicelli et al., 2001) was transferred to Hybond-N+ membranes and hybridised at low stringency with the mouse NCE fragment #17, labelled with [α-³²P]dATP by random priming, according to the manufacturers instructions (Amersham Pharmacia Biotech). A positive clone, containing a 6.4 kb insert, was digested with a series of restriction enzymes. Following gel electrophoresis, the DNA was transferred under alkaline conditions to Highland-N+ membrane and hybridised at low stringency with fragment #17. This identified a 1.7 kbApaLI/StuI subfragment that was cloned into pBS for sequencing and pBGZ40 for electroporation in the chick neural tube.

The mouse Krox20 protein was expressed in bacteria using the pET3a system(Novagen). Extracts were prepared from Krox20-expressing and control bacteria as described previously (Nardelli et al.,1992). For footprinting and band shift experiments, fragment #17,cloned in the SpeI site of pBS, was digested with eitherBamHI or XbaI, dephosphorylated, T4 polynucleotide kinase labelled using [γ-³²P]ATP and digested with NotI orSmaI, respectively. Labelled fragments were purified by polyacrylamide gel electrophoresis. DNase I footprinting experiments were performed as described previously (Galas and Schmitz, 1978) and reaction products resolved on a 6%denaturing polyacrylamide gel. Band shift experiments have been described previously (Nardelli et al.,1992).

HeLa cells were cultured in DMEM supplemented with 10% FCS. Cells(150,000/35 mm well) were transfected with 1 μg of DNA in duplicate using the Fugene 6 transfection reagent (Roche). Expression plasmids encoding the mouse Krox20 protein, pAdRSV Krox20(Giudicelli et al., 2001) and human Sox10 protein, pECESOX10(Bondurand et al., 2000) were used. Reporter constructs were the same as those used in in vivo experiments. The total quantity of DNA was kept constant using empty pBS vector. Following transfection, cells were incubated for 48 hours and extracts were assayed forβ-galactosidase activity using the chemiluminescent β-gel reporter gene assay (Roche). Transfections were normalised by cotransfecting a chloramphenicol acetyltransferase (CAT) expression plasmid (pSV2CAT)and quantitating protein using the CAT ELISA assay (Roche).

The GenBank accession numbers for human and mouse genomic contigs containing the Krox20 gene and flanking sequence are AL133417 and AC068424, respectively. The 1.7 kb chick sequences containing homology to the mouse NCE have been submitted to GenBank (accession number AY117679). Sequence alignments were performed using Dialign(Morgenstern et al., 1996) and identification of putative Krox20 binding sites using the rVista program (G. G. Loots, I. Ovcharenko, L. Pachter, I. Dubchak and E. Rubin,unpublished).

Recent studies have suggested that autoregulatory mechanisms contribute toKrox20 expression in the developing hindbrain(Giudicelli et al., 2001;Voiculescu et al., 2001). However, previous analysis of Krox20 mutants homozygous for aKrox20^lacZ allele did not reveal any obvious difference in the level of β-galactosidase activity in the NC when compared to heterozygotes (Schneider-Maunoury et al.,1993). We reasoned that an autoregulatory role of Krox20in the NC may have been masked in these experiments by the presence of two copies of the lacZ gene in the homozygotes, leading to a higher initial expression relative to heterozygotes. To circumvent this problem we made use of another Krox20 null allele,Krox20^Cre, and analysed compound mutant embryos,Krox20^Cre/lacZ, which carry only one copy of thelacZ gene. The effect of the Krox20 mutation in r3 and r5 has been described previously(Schneider-Maunoury et al.,1993; Schneider-Maunoury et al., 1997; Voiculescu et al.,2001). Consistent with these studies r3 and r5 were dramatically affected in the compound mutants at both 9.0 and 9.5 dpc (compareFig. 1A with B and C with D). In the NC, the β-galactosidase activity in heterozygotes(Krox20^+/lacZ) was similar to the compound mutants at 9.0 dpc (compare Fig. 1A with B),however, β-galactosidase-positive NC cells were seen to migrate both rostrally and caudally around the otic vesicle in the compound mutants, a profile described previously for the Krox20 mutation(Schneider-Maunoury et al.,1993). At 9.5 dpc a significant down-regulation was observed in the compound mutants (compare Fig. 1C to D). Since the production and survival of the crest population which should express Krox20 is not noticeably affected inKrox20 null mutants(Schneider-Maunoury et al.,1993) (O. Voiculescu, personal communication), these data suggest a decrease in Krox20/lacZ expression and support an autoregulatory mechanism contributing to Krox20 expression in the NC.

In order to identify cis-acting regulatory elements controllingKrox20 expression, we generated transgenic lines with several constructs containing large genomic regions surrounding a Krox20allele with a reporter lacZ gene inserted in-frame in exon 2(Krox20/lacZ) (Ghislain et al.,2002). This insertion is identical to the knock-in allele(Krox20^lacZ) generated previously in our lab, which was shown to reproduce the expression profile of the endogenous gene(Schneider-Maunoury et al.,1993). One of these constructs, carrying sequences between -31 kb and +7 kb relative to the transcription start site of Krox20 (-31/+7Krox20/lacZ), was active at the dorsal edge of the neural tube at the level of r5 and in a stream of NC cells migrating caudally from r5, as early as 8.5 dpc (8-10 somites) in 3/3 transgenic lines(Fig. 2A,Fig. 3, construct #1). In the neuroepithelium, the β-galactosidase-expressing cells were largely restricted to the dorsal edge consistent with the premigratory NC(Fig. 2B). The presence of more ventrally located β-galactosidase-positive cells(Fig. 2B), which are probably of neuroepithelial identity, may be explained by a lack of commitment of some dorsal cells to a NC fate. The expression in migratory NC cells was maintained at later stages, the majority migrating caudal to the otic vesicle(Fig. 2C,D). By 9.5 dpc expression was reduced, particularly in those cells distal to their site of origin in the third branchial arch (Fig. 2D). A small number of lacZ-positive crest cells were also found to migrate rostal to the otic vesicle(Fig. 2C,D). This pattern of expression of the transgene in the r5-derived NC is identical to those described for knock-in Krox20 alleles as well as for the endogenous gene (Fig. 1C)(Wilkinson et al., 1989;Schneider-Maunoury et al.,1993; Nieto et al.,1995; Voiculescu et al.,2001).

Given that autoregulation contributes to Krox20 crest expression(Fig. 1), we asked whether such a mechanism may be involved in the control of the activity of the -31/+7Krox20/lacZ transgene in the NC. When this transgene was analysed in a Krox20 null background (Krox20^Cre/Cre), noβ-galactosidase activity was detected in embryos at either 9.0 or 9.5 dpc(Fig. 2E,F). As the absence ofβ-galactosidase activity in the NC is not a consequence of the loss of these cells (Schneider-Maunoury et al.,1993), we conclude that Krox20 is required for the expression of this transgene. Taken together, these results identify a neural crest-specific element (NCE) that recapitulates the wild-type Krox20expression in the NC and whose activity is dependent on an autoregulatory mechanism.

Although the genetic studies described above clearly indicate the involvement of autoregulation in the control of the activity of the NCE, the identification of the cis- and associated trans-acting factors is required to determine whether the role of Krox20 is direct or indirect. As a first step,the NCE was delimited from the original -31/+7 Krox20/lacZ transgene. Subfragments of the -31 kb to +7 kb region containing either theKrox20/lacZ chimera (Fig. 3, construct #2) or linked to a β-globin minimal promoter/lacZ reporter (Fig. 3, constructs #3-6) were tested by `transient' transgenesis at 9.5 dpc. Of the five constructs spanning this region, only one was active in the NC, the -31 kb to -23.5 kb fragment, which reproduced the profile in 5/9 transgenic embryos (Fig. 3,Fig. 4A, construct #6S). Interestingly, one of the embryos, which was positive in the NC, also showed widespread expression in r5 (Fig. 3 and data not shown). Fragment #6 was also tested by transgenesis in antisense orientation relative to the promoter/lacZ reporter(Fig. 3, construct #6AS). This construct was active in 2/6 transgenic embryos in the NC, suggesting that the NCE acts as a canonical transcriptional enhancer (data not shown).

In ovo electroporation in the chick neural tube offers a quick and easy procedure to test the activity of cisregulatory elements in the neuroepithelium (Itasaki et al.,1999). To determine whether this approach could facilitate the delimitation of the NCE, constructs #2-6 were electroporated into chick embryo hindbrains at Hamburger and Hamilton (HH) stage 9 and analysed forβ-galactosidase activity at HH stage 13-14. Similar to the results for these constructs in mouse transgenesis, constructs #2-5 were negative whereas construct #6 was active in the NC (Fig. 3, Fig. 4B). Interestingly, in chick electroporation this construct was also active in r3 and r5 (Fig. 4B; see discussion). These results support the applicability of in ovo electroporation to delimit the NCE.

Subsequently, a series of deletions of fragment #6 were analysed for activity in the chick neural tube linked to the β-globin/lacZpromoter/reporter (Fig. 3,constructs #7-11). These studies led to the identification of an approximately 1 kb fragment extending from -26.5 kb to -25.5 kb which, much like construct#6, was active in the NC and r3 and r5(Fig. 3,Fig. 4C, construct #11). When tested by transgenesis in the mouse, construct #11 produced a high frequency of expression in the NC and a low frequency in r5, similar to construct #6(Fig. 3,Fig. 4D and data not shown). The NCE was further delimited by testing external 5′ and 3′deletions of the 1 kb fragment (Fig. 3, constructs #12-16). Using chick electroporation, we identified a 247 bp sequence with activity similar to the original fragment(Fig. 3,Fig. 4E, construct #17). When this fragment was tested in mouse transgenesis, it gave results similar to those of fragment #11, although the level of expression in the NC was reduced and the frequency of expression in r5 was increased(Fig. 3,Fig. 4F, construct #17). This raises the possibility that some cis-regulatory elements that contribute to the activity in the NC are located outside of fragment #17. However, as these sequences control only the level of activity in the NC and not the specificity, the characterisation of the 247 bp fragment was pursued.

As Krox20 was shown to act upstream of the NCE in the mouse(Fig. 2C-F), we tested the effect of Krox20 ectopic expression on the activity of this element by co-electroporation of construct #17 with a Krox20 expression vector (Giudicelli et al.,2001) in the chick hindbrain. In contrast to construct #17 alone,whose activity was limited to the endogenous Krox20 expression domain, co-electroporation led to activation of the reporter throughout the electroporated region (Fig. 3,construct #17+; compare Fig. 4E with G). This unrestricted response to Krox20 provides additional evidence in favour of an autoregulatory mechanism controlling the activity of the NCE.

In summary, deletion studies identified a fragment (#11) located between-26.5 kb and -25.5 kb, which reproduced the profile of the -31/+7Krox20/lacZ transgene in the NC. Further deletions of this element resulted in the isolation of a minimal, 247 bp, Krox20-responsive NCE, which contains all of the sequences necessary for specific activity in the NC.

As the mouse NCE performed comparably in the NC of both mouse and chick, we searched for orthologous regulatory elements in the chicken genome to aid the identification of cis-and trans-acting factors regulating Krox20 in the NC. The putative chick NCE was isolated from a bacterial artificial chromosome (BAC) clone containing the chicken Krox20 genomic sequence and flanking regions (Giudicelli et al.,2001) by hybridisation at low stringency, using fragment #17 as a probe. A 1.7 kb, cross-hybridising subfragment was tested in chick electroporation after linking to the β-globin/lacZpromoter/reporter. This construct was active in the r5 NC stream and in r3 and r5 similar to the mouse NCE, although occasional ectopic expression in the rostral neuroepithelium was also observed(Fig. 4H). The nucleotide sequences of mouse fragment #11 and of the 1.7 kb chick fragment were then established and compared together with the human sequences that were available in GenBank. These analyses revealed the existence of blocks of homology corresponding approximately to mouse fragment #17 (247 bp) and consisting of 211 bp in the chick and 243 bp in man. They showed 65% and 83% sequence identity to mouse fragment #17, respectively(Fig. 5). Interestingly,despite the indication that sequences flanking fragment #17 participate in NC activity, no significant homology was detected in this region in mouse/chick comparisons, supporting the conclusion that the essential cis-regulatory sequences are located within fragment #17 (data not shown).

The availability of NCE sequences in different species was then used to investigate whether the activity of fragment #17 was mediated by direct binding of Krox20 to sequences within this fragment. A matrix for the Krox20 binding site consensus sequence (Swirnoff and Milbrandt, 1995; Wingender et al., 2000) was used to search for putative Krox20 sites within fragment #17 and the human and chick orthologues. This led to the identification of three conserved candidate binding sites (data not shown). The capacity of these sites to bind Krox20 was assayed by footprinting with bacterially expressed Krox20 protein (Fig. 6). These analyses revealed the existence of two protected regions of about 20 bp, each centred over two of the three conserved sequences(Fig. 5,Fig. 6A). No other binding site was detected. In order to analyse the role of these two Krox20 binding sites,they were mutagenised by introducing a single G to C substitution, a mutation that has previously been shown to eliminate Krox20 binding activity both in vitro and in vivo (Nardelli et al.,1992; Sham et al.,1993). Indeed, bandshift experiments performed with wild-type fragment #17 and bacterially expressed Krox20 revealed specific complexes that were essentially eliminated when both Krox20 sites were mutated(Fig. 6B). These mutations were then introduced into fragment #11 and their effect was tested in both transgenesis in the mouse and electroporation in the chick(Fig. 3, construct #18). Inactivation of both Krox20 binding sites led to a dramatic decrease in the activity of the element in the NC in both the chick and the mouse (compareFig. 4C,D withFig. 7A,B). In addition, in the chick, where the NCE is normally active in r3 and r5, fragment #18 was completely inactive in these rhombomeres(Fig. 7A). These results indicate that the activity of the NCE is largely dependent on its Krox20 binding sites and strongly suggest the implication of a direct autoregulatory mechanism.

In conclusion, together with the genetic data obtained with theKrox20 null allele, these results demonstrate the existence of a direct Krox20 autoregulation in the NC that is required for the majority of the NCE activity. The limited crest activity that remains following the elimination of the two Krox20 binding sites may be due to either residual in vivo Krox20 binding activity and/or the action of other factors.

In subsequent experiments, we searched for cis-acting sequences involved in the weak crest activity observed with mutant fragment #18. We introduced a series of internal deletions into fragment #18 and tested their effects in the chick electroporation assay. We focused our analysis on the region of interspecies sequence similarity since it is likely to contain the essential regulatory elements (Fig. 5). Two deletions, together covering the majority of this region, were first generated (Fig. 3, constructs#19 and #20). Whereas construct #19 behaved in a manner similar to the original construct #18, construct #20 was completely inactive (data not shown). The region deleted in construct #20 was further subdivided into two deletions (Fig. 3, constructs#21 and #22). Interestingly, both deletions were inactive in the NC (data not shown), raising the possibility that important regulatory sequences may lie at the junction between the two regions.

The sequences deleted in construct #20 were analysed for the presence of putative transcription factor binding sites that were conserved between the mouse, human and chick sequences. We identified two adjacent, conserved sequences in a head-to-head orientation, similar to HMG box protein binding sites (A/TA/TCAAAG, Fig. 5)(Travis et al., 1991;van de Wetering et al., 1991;Harley et al., 1994). These sites are located precisely at the junction between the regions deleted in fragments #21 and #22. A mutant fragment consisting of a 15 nt deletion was generated to eliminate the two putative HMG box binding sites(Fig. 5). When tested in chick electroporation this construct was inactive, suggesting that these sites contribute to the activity of the NCE (Fig. 3, Fig. 7C,construct #23). A 41 bp sequence encompassing these sites(Fig. 5) was then multimerised and tested for enhancer activity in both chick electroporation and mouse transgenesis (Fig. 3, construct#24). In the chick, this construct was active throughout the NC along the electroporated region (Fig. 7D). A similar profile was observed in mouse transgenesis:although weak, NC cells migrating from r2, r4 and r5/6 were clearly labelled as well as some trunk crest cells (Fig. 7E).

In conclusion, the search for additional cis-acting sequences in the NCE identified an essential subregion containing conserved sequences similar to HMG box binding sites which, when multimerised, appears to exhibit enhancer activity throughout the NC, implicating the action of crest-specific HMG box factors.

Several HMG box, SRY-related Sox genes are expressed in the NC and may therefore participate in Krox20 expression. These include two group E Sox genes, Sox9 and Sox10(Ng et al., 1997;Zhao et al., 1997;Kuhlbrodt et al., 1998;Southard-Smith et al., 1998). Recently, Sox9 was shown to be important in the specification of NC cells inXenopus (Spokony et al.,2002). Sox10, although not required for crest formation and migration, participates in subsequent differentiation steps(Herbarth et al., 1998;Southard-Smith et al., 1998;Kelsh and Eisen, 2000;Britsch et al., 2001;Dutton et al., 2001). The possibility that Sox10 contributes to Krox20 expression in the NC was investigated. Firstly, we analysed its expression profile by in situ hybridisation. Sox10 expression was detected as early as 8.5 dpc in the pre/post migratory NC, with the highest levels in r2 and r4(Fig. 8A). However, significant levels were also detected in dorsal r5(Fig. 8A). Double labelling ofKrox20, by detection of β-galactosidase activity inKrox20^+/lacZ embryos, and Sox10 revealed a co-localisation in migrating NC, a profile that continued throughout the stages of crest migration (Fig. 8B and data not shown).

In a second series of experiments we investigated the possibility that Krox20 and Sox10 co-operate in the activation of the NCE. This was performed in a transactivation assay in cultured HeLa cells, which provides a quantitative evaluation of the contribution of each factor(Fig. 9). Using the wild-type NCE construct #11 as reporter, transfection of either a Krox20 or Sox10 expression construct led to a moderate level of induction. In contrast, the combination led to a synergistic induction. In order to investigate whether the synergism between these factors requires the elements important in NCE activity, a series of NCE mutants were tested. Construct #18, mutated in the two Krox20 binding sites, was weakly responsive to Krox20, confirming that the majority of the Krox20 binding activity had been eliminated. Although the Sox10 response was not affected, the combination of Krox20 and Sox10 resulted in a considerably reduced synergistic induction when compared to the wild-type construct. This suggests that the two Krox20 binding sites are required but that additional, weaker sites, are active. These results are also consistent with our in vivo data showing that construct #18 is weakly active in the NC(Fig. 7A,B). A new mutant derivative of construct #11, carrying uniquely the 15 nt deletion of the putative HMG box binding sites (Fig. 5), was tested. While the response to Krox20 was similar to the wild-type construct, this construct responded poorly to Sox10, consistent with an interaction of this factor with these sites. The combination of Krox20 and Sox10 together resulted in a considerably weaker synergistic response than with the wild-type construct. Although the reduced activity implicates additional sites, these results suggest an important contribution of the putative HMG box binding sites. Indeed, a construct containing a 7X multimer of these sites (#24) was dramatically induced by Sox10. Finally, a construct deficient in both the Krox20 and HMG box binding sites (#23) showed only an additive response to the combination of Krox20 and Sox10, definitively establishing the importance of these sites in the synergistic response. This latter result is consistent with our in vivo data indicating that construct#23 is defective in NC activity in chick electroporations(Fig. 7C).

To establish the involvement of Sox10 in Krox20regulation in vivo we analysed its expression in the Sox10 mutant line, Dom (Herbarth et al.,1998; Southard-Smith et al.,1998). Comparison of wild-type (data not shown), heterozygous(Fig. 8C) and homozygous mutant embryos (Fig. 8D) at 9.0 dpc revealed a strong decrease in Krox20 levels in the migrating neural crest of the homozygous mutant, whereas the heterozygote was not affected. This phenotype was observed throughout the stages of NC migration (data not shown). As Sox10 mutation does not affect neural crest production and migration (Southard-Smith et al.,1998; Britsch et al.,2001), this effect is likely due to a decrease in gene transcription.

In conclusion our experiments indicate that Krox20 and Sox10 are co-expressed in the migrating NC, they can co-operate in the autoactivation of the NCE in vitro and Sox10 is involved in the control of Krox20expression in vivo. This suggests that Sox10 is an essential coactivator ofKrox20 autoregulation in the NC.

This paper presents the identification and characterisation of a transcriptional enhancer element, the NCE, responsible for direct, positive autoregulation of the Krox20 patterning gene in the r5-derived NC. In addition to the Krox20 binding sites, the NCE contains essential recognition sequences for a putative, crest-specific HMG box factor. A candidate for such a factor, Sox10, was shown to be involved in Krox20 crest expression and to synergise with Krox20 to promote NCE activity. These data provide insight into the molecular mechanisms controlling the identity of postmigratory NC cells and how this identity is maintained.

We have identified the NCE as an enhancer located upstream ofKrox20, which can drive specific expression in the r5-derived NC stream in transgenic mice, reflecting the normal Krox20 expression pattern in the NC. The activity of this element is largely dependent on the integrity of two Krox20 binding sites and is abolished in a Krox20null background. These data establish that the NCE can mediate direct,positive autoregulation of Krox20 in the NC, raising the question as to what extent this element contributes to Krox20 expression. Sequence similarity searches identified elements in both man and chick that are the orthologues of the mouse NCE (Fig. 5). The chick element exhibited an activity in the NC similar to that of the mouse element when tested by electroporation in the chick neural tube. These observations suggest that the NCE probably has an important function in these cells as it has been conserved throughout the evolution of birds and mammals. More directly, analyses performed on compound mutant embryos, Krox20^Cre/lacZ, revealed a role ofKrox20 in maintaining the level of β-galactosidase activity and therefore its own level in the NC (Fig. 1). Furthermore the contribution of autoregulation may be partially masked in these studies by the stability of the β-galactosidase protein which is likely to be higher than that of the Krox20 protein itself. We can conclude that autoregulation allows for prolonged Krox20expression in the NC following the migration of these cells from the neural tube. This may be critical for maintaining NC positional identity since several Hox genes (Hoxa2, Hoxb2 and Hoxb3) are underKrox20 regulation (Sham et al.,1993; Nonchev et al.,1996; Manzanares et al.,2002). In addition to Hox genes, Krox20 controls the expression of the receptor tyrosine kinase gene, EphA4, in the NC(Seitanidou et al., 1997;Theil et al., 1998). By repulsion from cells expressing their Ephrin ligands, the Eph/Ephrin signalling system may guide NC cells along their pathway (reviewed byHolder and Klein, 1999). Indeed, blocking EphA4 function in Xenopus embryos causes r5-derived NC cells, normally destined for the third arch, to migrate into the second and fourth arches (Smith et al.,1997). This may partially explain why Krox20-expressing NC, which normally migrate caudally to the otic vesicle, migrate both rostrally and caudally in Krox20 mutant mice(Schneider-Maunoury et al.,1993).

The involvement of autoregulation in the control of Krox20expression is not specific to the NC. We have also observed this mode of regulation in r3 and r5 (Giudicelli et al., 2001; Voiculescu et al.,2001) and in the bone-forming cells (M. F., unpublished). Autoregulation secures Krox20 expression when the factors required for its initiation are no longer present, reflecting the importance and the duration of the role of this factor in developmental processes. Interestingly,distinct cis-acting elements mediate autoregulation in these tissues (this study; M. F., unpublished; D. Chomette, unpublished). The existence of multiple autoregulatory elements may derive from the necessity of tight coupling of autoregulation with the presence of other tissue-specific factors that have binding sites within each of these elements.

Analysis of the NCE carrying mutations in the two Krox20 binding sites led to the identification of an essential subregion containing two adjacent,head-to-head, putative HMG box binding sites(Fig. 7C). In addition, when a multimerised oligonucleotide containing these sites was tested in vivo,enhancer activity was detected throughout the NC(Fig. 7D,E). Together these studies suggested the action of crest-specific HMG box proteins inKrox20 autoregulation. We then provided evidence specifically implicating members the group E Sox subfamily of HMG box proteins: (i) two members of this group, Sox9 and Sox10, are expressed in the pre/post migratory NC (Ng et al.,1997; Zhao et al.,1997; Kuhlbrodt et al.,1998; Southard-Smith et al.,1998) and Sox10 is co-expressed with Krox20 in r5-derived NC (Fig. 8); (ii)Sox10 can synergise with Krox20 in NCE transactivation studies in cell culture(Fig. 9); (iii) Krox20levels in migrating neural crest are not maintained in Sox10 mutant embryos (Fig. 8). Together these data indicate that Sox10 is a crest-specific factor essential forKrox20 autoregulation in the NC. This conclusion is consistent with the co-operation recently observed between Sox10 and Krox20 in the activation of the connexin 32 gene in the Schwann cell lineage(Bondurand et al., 2001),suggesting that such co-operation may be a more general phenomenon.

Although our studies highlight an important role of Sox10, it may not be the only HMG box factor co-operating with Krox20 in the r5-derived NC.Sox9, which is expressed in the newly formed r5-derived neural crest(data not shown) may also be involved in an earlier step of Krox20autoregulation. Similarly, although the mutation of either the Krox20 or HMG sites resulted in a significant decrease in the synergistic response, the response was not completely eliminated, suggesting that additional Krox20 and HMG binding sites may be involved. This is supported by sequence analyses of the NCE indicating the presence of additional putative binding sites for both factors (data not shown).

In addition to the demonstration of synergism between Krox20 and crest-specific Sox proteins, our data suggest that these factors may also have distinct roles in the activation of the NCE. In the mouse, the NCE is generally active in r5-derived NC and less frequently in r5. In the electroporated chick neural tube, the NCE is active in all territories expressing Krox20: r3, r5 and r5-derived neural crest. Furthermore,in the chick, ectopic expression of Krox20 leads to generalised activation of the NCE in both the neuroepithelium and the NC. A hypothesis to explain these data is that activation of the NCE by Krox20 may be very dependent on the chromatin state of the NCE: in normal chromosomal chromatin configuration, Krox20 absolutely requires co-operation with a crest-specific factor to activate the NCE; in contrast, in `relaxed chromatin'configurations, reached at a low frequency in mouse transgenesis, where it may depend on the site of insertion of the transgene, and systematically after electroporation in the chick neural tube, where the reporter plasmid is likely to remain episomal, Krox20 can activate the NCE without the crest factor. According to this hypothesis, the crest-specific factor might modify the local conformation of the chromatin in order to allow activation by Krox20. The properties of Sox proteins are consistent with such a possibility(Wegner, 1999).

The present data allow us to propose a model on how Krox20expression is set up in r5-derived NC according to its level of origin in the hindbrain. This study has established an absolute requirement for Krox20 in NCE activity (Fig. 2). Consequently, the Krox20 protein must accumulate in the premigratory NC prior to the activation of the NCE by another mechanism. This may involve the action of regulatory elements that establish Krox20 expression in the neuroepithelium, without discrimination between neural and NC precursors. We propose that establishment of Krox20 expression in r5-derived NC involves the following steps (Fig. 10). (i) Initiation of Krox20 expression in r5 by hindbrain cis-acting initiator element(s). We have recently identified an r3/r5-specific regulatory element, located upstream of the NCE, that constitutes a good candidate to perform this function (D. Chomette,unpublished). (ii) Activation of the NCE in premigratory NC cells in r5 by co-operation between Krox20 and crest-specific Sox proteins. (iii) Maintenance of Krox20 expression under control of the NCE in postmigratory r5-derived NC once the initiators are downregulated. The model explains the establishment of a registry between Krox20-expressing NC and r5 on a cell lineage basis: sufficient amounts of Krox20 have to be inherited in delaminating NC cells for them to initiate or maintain the autoregulatory loop under the control of crest-specific Sox proteins. Such a mechanism ensures the acquisition of positional information in the NC according to its rhombomeric origin and may be involved in the control of the expression of other patterning genes. More generally, autoregulatory mechanisms, such as those controlling the expression of Hox genes in the hindbrain neural crest(Pöpperl et al., 1995;Maconochie et al., 1997), are likely to be important in maintaining the neural crest prepattern.

We thank Sylvie Schneider-Maunoury, Christine Vesque-Kishtoo and Marion Wassef for critical reading of the manuscript. We are greatful to Michel Goossens for providing us with the Sox10^Dom mouse line. J. G. was supported by postdoctoral fellowships from the European Union (TMR),the AFM and the Institut de France. This work was supported by grants from INSERM, MENRT, EC, ARC and AFM.