Dlx genes pattern mammalian jaw primordium by regulating both lower jaw-specific and upper jaw-specific genetic programs

Craniofacial development begins when the cranial neural crest cells(CNCCs), migratory multipotent precursors that contribute to most of the face,delaminate from the dorsal brain and migrate ventrolaterally to form the ectomesenchyme of facial primordia known as the frontonasal prominence (FNP)and branchial arches (BAs) (see Fig. S1A in the supplementary material)(Noden, 1978; Couly et al., 1993; Osumi-Yamashita et al., 1994; Köntges and Lumsden,1996; Chai et al.,2000). Subsequently, the FNP becomes the mid- and upper face,while the first branchial arch (BA1) develops into most of the jaw, the lateral skull, palate and the middle ear(Köntges and Lumsden,1996). BA1 is further divided into maxillary arch (mxBA1,prospective upper jaw) on the proximal half, and mandibular arch (mdBA1,prospective lower jaw) on the distal half (see Fig. S1A,B in the supplementary material). The second branchial arch (BA2) mainly contributes to the ear and neck skeleton.

How CNCCs recognize their positional information and develop accordingly is beginning to be understood (reviewed by Depew et al., 2002a; Santagati and Rijli, 2003; Chai and Maxon, 2006). Interactions of CNCCs with the neighboring tissues result in the expression of a diverse set of transcription factors in CNCCs, and the specific combination of transcription factors provides a positional identity to the cells.

The vertebrate Dlx genes are homologs of Drosophila Distal-less;they encode homeodomain transcription factors(Panganiban and Rubenstein,2002). Mice have six Dlx genes, which are organized as three linked pairs in the genome (Dlx1/2, Dlx3/4 and Dlx5/6)(Porteus et al., 1991; Price et al., 1991; Robinson and Mahon, 1994; Simeone et al., 1994; McGuinness et al., 1996; Nakamura et al., 1996; Liu et al., 1997).

During craniofacial development, mouse Dlx genes are regionally expressed within BAs as well as in olfactory and otic placodes (see Fig. S1A,C in the supplementary material) (Dolle et al.,1992; Bulfone et al.,1993; Robinson and Mahon,1994; Simeone et al.,1994; Qiu et al.,1997; Depew et al.,2002b). In the ectomesenchyme of BA1 in mid-gestation stage embryos, Dlx1/2 are expressed in both mxBA1 and mdBA1, whereas Dlx5/6 are expressed in mdBA1 only. Dlx3/4 expression is further restricted to a narrow domain within mdBA1. The same proximodistal arrangement is also found in BA2. Since Dlx3/4 expression is dependent on Dlx5/6 (Depew et al., 2002b) (this study), essentially two different combinations of Dlx partition much of BA1: Dlx1/2 for mxBA1 and Dlx1/2+5/6 for mdBA1. The functional importance of this `Dlx code' in BA patterning has been investigated using mouse loss-of-function mutants. Owing to the tight linkage in the genome, the double mutations of Dlx1/2 and Dlx5/6 pairs were achieved by deleting both genes in one allele (Qiu et al.,1997; Merlo et al.,2002; Depew et al.,2002b). Inactivation of Dlx1 and/or Dlx2(Dlx1^-/-, Dlx2^-/- and Dlx1/2^-/-) caused abnormalities in upper jaw skeleton with little effect on the lower jaw (Qiu et al., 1995; Qiu et al.,1997; Depew et al.,2005). By contrast, Dlx5^-/- exhibited defects in lower jaw development (Depew et al.,1999). Most strikingly, the simultaneous inactivation of Dlx5 and Dlx6 (Dlx5/6^-/-) resulted in homeotic transformation of the lower jaw into upper jaw(Beverdam et al., 2002; Depew et al., 2002b). Therefore, the differential expression of Dlx genes along the proximodistal axis is important for the regional specification of BA1; Dlx1/2 are necessary for the proper development of mxBA1, whereas Dlx5/6 confer mdBA1 identity.

Our current work addresses three important issues on how the Dlx genes regulate BA patterning. First, the mechanism through which Dlx5/6specify lower jaw fate needs to be understood. To this end, we performed a genome-wide transcriptional profiling and obtained a comprehensive list of genes with altered expression in Dlx5/6^-/- mdBA1. Second,we provide the first evidence of the upper jaw-specific genetic program and show that it is partly regulated by Dlx genes. Prior to this study, the abundance of mdBA1-specific markers but the lack of any known mxBA1-specific markers has been compatible with the idea that the upper jaw is the default state upon which the lower jaw fate is imposed. Finally, we investigated the functional relationship of different Dlx genes expressed in BA1 by comparing the morphological and molecular phenotypes of various combinations of Dlx single and double mutants. We found that Dlx1, 2, 5 and 6act both partially redundantly and antagonistically, depending on the context,to achieve differential expression of their downstream genes in mxBA1 and mdBA1.

All experiments using mice were performed following UCSF institutional regulations on the care and use of laboratory animals.

Dlx5/6^-/- and Dlx5/6^+/+ littermates were collected from Dlx5/6^+/- intercrosses at E10.5. The mdBA1s were dissected, flash-frozen and stored in liquid nitrogen until the day of RNA extraction. The tissue was homogenized in Trizol reagent(Invitrogen) using Pellet Pestle (Kontes). RNA was extracted using chloroform and then concentrated by isopropanol precipitation. After a rinse with 80%ethanol, the RNA pellet was dissolved in water and purified using the RNeasy Mini Kit (Qiagen). From 13 Dlx5/6^-/- and 13 Dlx5/6^+/+ E10.5 embryos, we recovered 14.7 μg and 13μg of mdBA1 total RNA, respectively.

All subsequent steps of the microarray experiment were performed by the Translational Genomics Research Institute (TGen, Phoenix, AZ), through the NIH Neuroscience Microarray Consortium. The RNA sample from each genotype was hybridized onto GeneChip Mouse Genome 430 2.0 arrays (Affymetrix) in triplicate. Data acquisition and analysis employed GeneChip Operating Software(GCOS, Affymetrix) version 1.2.

Whole-mount and section in situ hybridizations were performed using digoxigenin-labeled RNA probes as described(Jeong et al., 2004; Jeong and McMahon, 2005),except that 20 μm sections were used for the section in situ hybridization. The control and mutant embryos were stage-matched using a combination of several morphological criteria, including the size and shape of the limb buds,morphogenesis of the eye, and somite numbers. Skeletons of E18.5 or P0 animals were stained with Alcian Blue and Alizarin Red as described(Jeong et al., 2004). For the in situ hybridization and skeletal preparations shown in Figs 1, 2, 3 and 4, embryos of +/+ and +/- genotypes were used indiscriminately and are referred to as `wild type'. For Figs 5 and 6, `wild type' refers to +/+for all the genes in question, or Dlx1^+/-.

DNA templates for in situ hybridization probes were obtained by PCR from a wild-type mouse E10.5 BA1 cDNA library or from adult tail genomic DNA,purchased from companies, or kindly provided by other investigators. Further information on the probes is available upon request.

The 1 kb putative Gbx2 enhancer (see Fig. 3A) was amplified by PCR from mouse tail genomic DNA using primers 5′-ACACCTCGAGAGAGGATGACAGCGAGCTTCG-3′ and 5′-GTGTAAGCTTGAGCAAACATTCCAGTTTTAATGC-3′, and cloned into XhoI-HindIII sites of pGL4.23 (Promega). pGL4.23 contains a minimal promoter and the firefly luciferase coding sequence. pCAGGS-Dlx5(Stuhmer et al., 2002) was used to express Dlx5 protein. pGL4.73 (Promega), a plasmid that constitutively expresses Renilla luciferase, was used as a control for variations in transfection efficiency. 3T3 cells were transfected with FuGene 6 (Roche), and 40 hours later the cells were lysed and analyzed for luciferase activity using the Dual Luciferase Reporter Assay System (Promega). The experiments were performed in triplicate and the results combined for statistical analysis.

To generate the Dlx6-lacZ (Dlx6-) allele, a 4.3 kb SpeI-SnaBI genomic fragment spanning Dlx6 exon 3(which encodes the homeodomain) was subcloned into the SpeI-SmaI sites of pBS KS+. Site-directed mutagenesis was performed to engineer a unique NruI site immediately following the 219th amino acid from the N-terminus of the Dlx6 protein (F of the sequence VKIWFQNKRS). Flanking genomic sequences were added between the unique 5′ XhoI site (5.8 kb 5′ homology arm) and the 3′ NotI site (3.5 kb 3′ homology arm), and the reporter cassette IRES-lacZ-PGKneo (Robledo et al.,2002) was cloned into the unique NruI site to generate the final targeting construct. The Dlx6-lacZ allele therefore interrupts the Dlx6 protein immediately following the amino acid F as described above. ES cell culture, screening, chimera generation and testing were as previously described (Robledo et al., 2002).

To understand the molecular changes underlying Dlx5/6^-/- jaw phenotypes, we compared transcriptional profiles of the wild-type and mutant mdBA1 at E10.5 using Affymetrix GeneChip Mouse Genome 430 2.0 array. We chose E10.5 for our analysis because E9.0-10.5 is when the Dlx genes exhibit proximodistally nested expression in BA1(Qiu et al., 1997; Acampora et al., 1999) (see Fig. S1 in the supplementary material), and the wild-type and mutant BA1 still appear grossly comparable in size and morphology at E10.5 (see also Fig. S2 in the supplementary material). The Dlx5/6^-/- mutant exhibited downregulation of 39 genes (see Table S1 in the supplementary material) and upregulation of 24 genes (see Table S2 in the supplementary material) with greater than a 2-fold change. Our results included most, but not all, of the genes that were previously shown to be dysregulated in the Dlx5/6^-/- mutant(Depew et al., 1999; Depew et al., 2002b). Thus,although it is a robust procedure, technical limitations exist (see Tables S1 and S2 in the supplementary material).

The complete set of raw data from our transcriptional profiling experiment is available at NIH Neuroscience Microarray Consortium data repository(http://arrayconsortium.tgen.org)under accession ruben-affy-mouse-187820, and at GEO (accession number GSE 4774).

The Dlx3, Hand2 and Alx4 transcription factors were previously identified as being downstream of Dlx5/6(Depew et al., 2002b). Our screen found that their close relatives, Dlx4, Hand1 and Alx3, are also downregulated in Dlx5/6^-/- mdBA1 and BA2(Fig. 1A,B,M-P). In the otic vesicle, Dlx5/6 function is required for the expression of Gbx2, a homeodomain transcription factor (Robledo and Lufkin, 2006). We confirmed the same regulatory relationship in mdBA1 (Fig. 1C,D). The transcription factors Cited1 (Msg1)(Shioda et al., 1996) and Zac1(Lot1, Plagl1) (Abdollahi et al.,1997; Spengler et al.,1997) are expressed at high levels in the medial mdBA1 and BA2;their expression was downregulated in Dlx5/6^-/-(Fig. 1Q-T).

Mouse Dlx1 locus encodes several putative non-coding RNA (ncRNA)transcripts (A/S Dlx1; see Fig. S3A in the supplementary material)(McGuiness et al., 1996; Liu et al.,1997). In BA1 and BA2, the expression of A/S Dlx1 was restricted to the distal region (Fig. 1E), and was completely abolished in Dlx5/6^-/-(Fig. 1F). Evf1/2, the two splicing variants of a ncRNA gene (Dlx6os1 - Mouse Genome Informatics), map adjacent to Dlx5/6; their brain expression depends on Dlx1/2 function (see Fig. S3B in the supplementary material)(Faedo et al., 2004; Kohtz and Fishell, 2004). Evf1/2 expression was greatly reduced in both brain and BAs of Dlx5/6^-/- mutants (Fig. 1G,H). It should be noted that the promoter and the first exon of Evf2 fall within the deletion in the Dlx5/6 null allele(Robledo et al., 2002), and thus the decreased Evf1/2 expression in Dlx5/6^-/-could simply be due to the loss of the Evf2 promoter. However, we found reduced expression of Evf1/2 also in Dlx5^-/- and Dlx6^-/- mutants (see below), in which both Evf1 and Evf2 sequences are intact. This result, and the fact that Dlx1/2 activity is necessary for the expression of Evf1/2 in the brain, argue that Dlx5/6regulate Evf1/2 transcription, directly or indirectly.

A secreted signaling molecule, hepatocyte growth factor (Hgf, also known as scatter factor, SF) (Stoker et al.,1987; Birchmeier and Gherardi,1998), is expressed in BA1 and BA2; this expression was dependent on Dlx5/6 function (Fig. 1I,J). Unc5c encodes one of the receptors for the axon guidance molecule netrin 1 (Ackerman et al., 1997). Unc5c was expressed in the medial domain of mdBA1 and BA2, and was severely downregulated in Dlx5/6^-/-(Fig. 1U,V). BMP-binding endothelial regulator (Bmper, also known as crossveinless-2, Cv2) is a secreted molecule that binds to and enhances the signaling of bone morphogenetic proteins (BMPs) (Coffinier et al., 2002; Moser et al.,2003; Coles et al.,2004; Ikeya et al.,2006). Bmper expression in distal BA1 and BA2 was dependent on Dlx5/6 (Fig. 1K,L). The regulator of G-protein signaling 5 (Rgs5) gene is expressed only at the rostromedial tip of mdBA1; this expression was lost in Dlx5/6^-/- (Fig. 1W,X). Rgs proteins are GTPase-activating proteins that attenuate G-protein-coupled receptor signaling (Chen et al., 1997; Xie and Palmer,2007).

Since Dlx3/4 expression is lost in Dlx5/6^-/-mdBA1 (Depew et al., 2002b)(Fig. 1A,B), it is possible that at least some of the gene expression changes in Dlx5/6^-/- mutants are due to the loss of Dlx3/4 activity.

Pou3f3 (Brn1), a POU-domain transcription factor(Hara et al., 1992), was highly expressed in the entire mxBA1 and maxillary-mandibular junction, but was absent from most of mdBA1 (Fig. 2A). It was also confined to the proximal region in BA2. In Dlx5/6^-/-, Pou3f3 expression expanded into mdBA1 and distal BA2 with the same intensity as in its normal expression domains(Fig. 2B).

We identified two Riken cDNA clones (26100017I09Rik,2900092D14Rik), both suspected to be non-coding, that are closely linked to and have very similar expression patterns as Pou3f3, in both wild-type and Dlx5/6^-/- embryos(Fig. 2D,E,G,H; see Fig. S3C in the supplementary material). Because the proximal BA expression of Pou3f3,2610017I09Rik and 2900092D14Rik overlaps with that of Dlx1/2 (see Fig. S1C in the supplementary material), we tested whether the expression of these genes is dependent on the activity of Dlx1/2 in the mxBA1. We found downregulation of all three genes in BA1 and BA2 of Dlx1/2^-/-, demonstrating that Dlx1/2 are necessary for their normal expression(Fig. 2C,F,I).

Foxl2 encodes a winged helix/forkhead transcription factor(Crisponi et al., 2001). It was expressed strongly in the dorsal mxBA1 just below the eye, in addition to a small domain at the maxillary-mandibular junction(Fig. 2J). In Dlx5/6^-/-, the mxBA1 pattern of Foxl2 expression was duplicated in mdBA1 (Fig. 2K). Riken cDNA E330015D05Rik is a poorly characterized gene closely linked to Foxl2 (see Fig. S3D in the supplementary material). Its expression pattern, both in wild type and Dlx5/6^-/-, was identical to that of Foxl2(Fig. 2M,N).

Cyp26a1, a retinoic acid-metabolizing enzyme cytochrome P450(White et al., 1996; Fujii et al., 1997; Ray et al., 1997), is normally expressed along the border of BA1 and BA2 with higher expression in mxBA1 than in mdBA1 (Fig. 2P). In BA2, Cyp26a1 expression was also strong proximally and weak distally. Removal of Dlx5/6 activity lead to upregulation of Cyp26a1in distal BA1 and BA2 to the same intensity as in the proximal domains(Fig. 2Q). Irx5, which encodes an Iroquios-related homeodomain transcription factor(Bosse et al., 2000; Cohen et al., 2000), was expressed strongly in the dorsal mxBA1 and weakly in the ventral mdBA1(Fig. 2S). The latter expression was moderately increased in Dlx5/6^-/-(Fig. 2T). Unlike Pou3f3,2610017I09Rik and 2900092D14Rik, the mxBA1 expression of Foxl2, E330015D05Rik, Cyp26a1 and Irx5 was not dependent on Dlx1/2 (Fig. 2L,O,R,U).

The genes described above in this section are normally expressed in mxBA1 and upregulated in the Dlx5/6^-/- mdBA1, making the mutant mdBA1 molecularly similar to mxBA1. By contrast, transmembrane protein 30b(Tmem30b), a homolog of yeast endosomal protein Cdc50(Katoh and Katoh, 2004), was barely detectable in either mxBA1 or mdBA1 in wild-type embryos(Fig. 2V), but was strongly upregulated in Dlx5/6^-/- mdBA1, making the mutant mdBA1 molecularly different from mxBA1 (Fig. 2W).

An important question about the Dlx5/6-downstream genes identified by our microarray analysis is whether Dlx5/6 directly regulate them. There is already evidence that three of the genes listed in Table S1 (see Table S1 in the supplementary material) are direct targets of Dlx5/6.Charite et al. (Charite et al.,2001) showed that Dlx6 binds to, and thus most likely directly regulates, a Hand2 BA enhancer. Sumiyama and Ruddle(Sumiyama and Ruddle, 2003)showed that the BA enhancer for Dlx3/4 contains a consensus Dlx-binding motif, and thus it is highly likely that Dlx5/6 directly regulate Dlx3/4 expression in the BA. Thus, the fact that Hand2, Dlx3and Dlx4 were identified by our microarray analysis supports the validity of our approach, and suggests that additional genes listed in Table S1 (see Table S1 in the supplementary material) are directly regulated by Dlx5 and Dlx6 proteins.

To identify direct targets of Dlx5/6, we performed in silico analyses of the genomic sequences surrounding Dlx5/6-downstream genes. Previously, a systematic in vitro binding assay determined that(A/C/G)TAATT(G/A)(C/G) is a consensus binding motif for Dlx proteins(Feledy et al., 1999). In addition, various researchers have analyzed cis-regulatory elements from several Dlx target genes, discovering 13 sequences to which Dlx proteins bind directly, including some that do not conform to (A/C/G)TAATT(G/A)(C/G) (see Table S3 in the supplementary material). Using rVISTA(Loots and Ovcharenko, 2004),we searched genomic sequences flanking several of the genes listed in Table S1(see Table S1 in the supplementary material) for all the known Dlx-binding motifs that are conserved between mouse and human. This analysis identified∼0.6 kb of a highly conserved region (conserved down to chick) with a cluster of three putative Dlx-binding sites located ∼10 kb upstream of Gbx2 (Fig. 3A,B). We tested a 1 kb fragment containing this region (red bar in Fig. 3A, including the adjacent 0.25 kb sequence conserved down to opossum) for transcriptional enhancer activity in 3T3 cells. Co-transfection with a Dlx5 expression vector resulted in a >70-fold activation (Fig. 3C), demonstrating that the fragment contains a transcriptional enhancer regulated by Dlx proteins.

To our knowledge, Pou3f3, Foxl2, 26100017I09Rik, 2900092D14Rik and E330015D05 are the first examples of genes that are expressed specifically in the maxillary domain of BA1 at any stage of mouse development. Therefore, they provide evidence of an upper jaw-specific genetic program (see Discussion).

Among the five genes, Pou3f3 has the strongest and broadest expression in mxBA1 (Fig. 2),and thus we performed further analysis on its expression and function during upper jaw development (Fig. 4). Section in situ hybridization at E10.5 revealed that Pou3f3 is expressed only in the mesenchyme (Fig. 4A,B). At E12.5 (data not shown) and E13.5(Fig. 4C-E), Pou3f3expression was found in both upper and lower jaw; however, its expression in the condensed dental mesenchyme was restricted to upper molars(Fig. 4D). In addition, Pou3f3 was expressed in the caudal, but not rostral, palatal shelves(PS) (Fig. 4C,E).

Pou3f3^-/- (McEvilly et al., 2002) mutant skull revealed the essential role of this gene in the development of a part of the upper jaw; mxBA1-derived squamosal bone (SQ) normally articulates with mdBA1-derived dentary to make the functioning jaw joint in mammals. In Pou3f3^-/-, the squamosal bone is largely missing (except the retrotympanic process, rt), and thus the jaw joint does not exist (Fig. 4F-I,L,M). Jugal bone (JG), which forms the zygomatic arch on the lateral skull together with the maxilla and squamosal bone, was also lost in the mutant (Fig. 4H,I,L,M). In the mutant middle ear, the malleus (mdBA1-derived) appeared normal, whereas the incus (mxBA1-derived) had a slightly elongated short crus(Fig. 4J,K). In addition, the BA2-derived stapes was fused to the styloid process(Fig. 4J,K). Other bones in the skull, and the teeth and palate, appeared unaffected in the Pou3f3^-/- mutant (data not shown).

Simultaneous inactivation of Dlx5/6 results in the homeotic transformation of the lower jaw into upper jaw, whereas Dlx5 single mutants have relatively mild defects in lower jaw morphogenesis(Depew et al., 1999; Acampora et al., 1999; Beverdam et al., 2002; Depew et al., 2002b). Based on these results alone, it was not clear whether Dlx5 and Dlx6are functionally redundant, or whether Dlx6 has a more prominent role. To address this, we generated an allele that inactivated Dlx6alone, by inserting an IRES-lacZ-neomycin resistance cassette within the Dlx6 homeobox coding sequence(Fig. 5A). This insertion prevents translation of one-third of the homeodomain and the entire C-terminal domain, including the nuclear localization signal [amino acids 220-228(Cokol et al., 2000)]; thus,this allele is likely to be null.

Dlx6^-/- mice were born alive but died within a day with aerophagia, as reported for Dlx2^-/-, Dlx1/2^-/-and Dlx5^-/- (Qiu et al., 1995; Qiu et al.,1997; Acampora et al.,1999; Depew et al.,1999). The head skeleton of Dlx6^-/- neonates had several abnormalities that are also found in Dlx5^-/-animals (Acampora et al., 1999; Depew et al., 1999). The mutant had a slightly reduced mandible(Fig. 5B,D), in which the dentary lacked the coronoid process and had a hypoplastic condylar process(Fig. 5E-G, arrows and arrowheads). The ectotympanic of the mutant ear was shortened(Fig. 5H-J, arrows), and the gonial bone was attached to an ectopic piece of bone [named os paradoxicum(Depew et al., 1999)]extending toward the ala temporalis on the skull base(Figs 5H-J; see S4A-C,arrowheads, in the supplementary material). The skeletal elements mentioned thus far are thought to be derivatives of BA1. Therefore, allowing for some individual variations in morphological details, Dlx6^-/-mice have BA1-associated defects that are very similar to those of Dlx5^-/-, but are much less severe than the homeotic transformation observed in Dlx5/6^-/-. This result establishes that Dlx5 and Dlx6 are in large part functionally redundant in lower jaw development.

In addition to the BAs, Dlx5/6 were expressed in otic and olfactory placodes (see Fig. S1A,C in the supplementary material) and, as a result, Dlx5^-/- animals have dorsally deficient otic capsule (Fig. 5H,I, open arrowheads) and hypoplastic nasal cartilage (see Fig. S4D,E, arrowheads, in the supplementary material) (Acampora et al., 1999; Depew et al.,1999). However, both structures appeared normal in Dlx6^-/- (Fig. 5J; see Fig. S4F in the supplementary material), which suggests that Dlx6 is less important than Dlx5 in otic and olfactory placode development.

Next, we examined the expression of the molecular markers that are affected in Dlx5/6^-/- mdBA1 (Figs 1 and 2) in Dlx5^-/- and Dlx6^-/- single mutants. Most of the genes showed no, or moderate, changes in either mutant, indicating that they are directly or indirectly regulated by both Dlx5 and Dlx6 to similar degrees (see Fig. S5A-d in the supplementary material). Surprisingly, however, several genes were differentially changed in Dlx5^-/- and Dlx6^-/-; Gbx2, Bmper,A/S Dlx1, Dlx4 and Evf1/2 were all severely downregulated in Dlx5/6^-/- BAs (Fig. 1), but Gbx2, Dlx4 and Evf1/2 were more downregulated in Dlx5^-/- than in Dlx6^-/-, whereas Bmper and A/S Dlx1 were more affected in Dlx6^-/- than in Dlx5^-/- (Fig. 5K-P; see Fig. S4G-I and Fig. S5e-j in the supplementary material). Similarly, for the genes upregulated in Dlx5/6^-/-,Pou3f3, Foxl2, 2610017I09Rik and E330015D05Rik showed greater changes in Dlx5^-/- than in Dlx6^-/-,whereas Tmem30b was upregulated only in Dlx6^-/-(Fig. 5Q-V; see Fig. S4J-L and Fig. S5k-p in the supplementary material). These results suggest that there are some differences between the transcriptional activities of Dlx5and Dlx6, even though inactivation of each gene results in similar morphological consequences.

Although Dlx1/2^-/- mice have no significant defects in the lower jaw (Qiu et al.,1995; Qiu et al.,1997; Depew et al.,2005), Dlx1/2 can contribute to mdBA1 development, as the lower jaw phenotypes of Dlx5^-/- are greatly exacerbated in Dlx1^-/-;5^-/- and Dlx2^-/-;5^-/- mutants(Depew et al., 2005).

We tested whether Dlx6 is also at least in part functionally redundant with Dlx1/2 by generating Dlx1^-/-;6^-/- and Dlx2^-/-;6^-/- double mutants. The compound mutants exhibited far greater craniofacial defects than those found in any of the single mutants. The dentaries of Dlx1^-/-;6^-/- and Dlx2^-/-;6^-/- were shortened, fragmented, and bifurcated to become bones resembling the maxilla, jugal and pterygoid of the upper jaw (Fig. 6A-L; see S6A-F in the supplementary material). The ventrolateral side of the skull of Dlx1^-/-;6^-/- and Dlx2^-/-;6^-/- had what appears to be a duplicate lamina obturans juxtaposed to the endogenous one(Fig. 6S-X; LO^* in Fig. 6W,X). In the ear, the ectotympanic and middle ear ossicles were progressively reduced or lost in Dlx1^-/-;6^-/- and Dlx2^-/-;6^-/-(Fig. 6Y-d). The basihyoid and lesser horn of the hyoid were also progressively more affected in Dlx1^-/-;6^-/- and Dlx2^-/-;6^-/- (see Fig. S6G-L in the supplementary material).

The phenotypes of Dlx1^-/-;6^-/- and Dlx2^-/-;6^-/- are reminiscent of those reported for Dlx1^-/-;5^-/- and Dlx2^-/-;5^-/-(Depew et al., 2005). In all four cases, the mdBA1-derivatives are reduced in size and/or transformed into structures resembling upper jaw elements, although the extent of transformation is less than in Dlx5/6^-/-.

To elucidate the mechanisms underlying the morphological changes in Dlx1^-/-;6^-/-,Dlx2^-/-;6^-/- and Dlx2^-/-;5^-/-, we examined the gene expression patterns in their mdBA1.

The expression of Gbx2 and Bmper in mdBA1 was minimally or moderately downregulated in Dlx5^-/- and Dlx6^-/- (Fig. 5). However, their expression is severely reduced or completely abolished in Dlx1^-/-;6^-/-,Dlx2^-/-;6^-/- and Dlx2^-/-;5^-/-(Fig. 7A-H). Pou3f3was repressed by Dlx5/6 in mdBA1, and was moderately upregulated in Dlx5^-/- and Dlx6^-/-(Fig. 5). Further removing Dlx1 or Dlx2 activity expanded and intensified Pou3f3 overexpression in mdBA1 (compare Fig. 7J,K with Fig. 5S, and Fig. 7L with Fig. 5R). Therefore, even though Dlx1/2 activity is required for the normal expression of Pou3f3 in mxBA1 (Fig. 2A,C), Dlx1/2 apparently share the repressive effects of Dlx5/6 on Pou3f3 in mdBA1. Hand2 was severely downregulated in Dlx5/6^-/-, but unaffected in Dlx5^-/- and Dlx6^-/-(Depew et al., 2002b) (see Fig. S5A-D in the supplementary material). Hand2 expression appeared normal in Dlx1^-/-;6^-/-, but was gradually reduced and became restricted to caudomedial mdBA1 in Dlx2^-/-;6^-/- and Dlx2^-/-;5^-/-(Fig. 7M-P).

The expression changes of Gbx2, Bmper and Pou3f3 in Dlx1^-/-;6^-/-,Dlx2^-/-;6^-/- and Dlx2^-/-;5^-/- are either similar or identical to the changes in Dlx5/6^-/- (Figs 1 and 2), in line with the overall similar morphological defects of these four mutants. By contrast, Hand2 expression was more severely reduced in Dlx5/6^-/- than in any of Dlx1^-/-;6^-/-,Dlx2^-/-;6^-/- or Dlx2^-/-;5^-/- (see Fig. S5D in the supplementary material). This could explain why the morphological transformations of lower jaw into upper jaw in the latter three mutants are incomplete compared with Dlx5/6^-/-.

Our genome-scale expression profiling experiment greatly expanded the number of potential targets of Dlx5/6 in mdBA1 (see Tables S1 and S2 in the supplementary material), and we confirmed 20 novel Dlx5/6-downstream genes by in situ hybridization (Figs 1 and 2). They include genes encoding transcription factors, signaling molecules, ncRNAs or unclassifiable products. Some of the Dlx5/6-downstream genes have been implicated in craniofacial development based on their mutant analysis in mice (summarized in Table 1); our study added Pou3f3 to this list (Fig. 4). However, the reported mutant phenotypes of several genes identified from our screen do not provide an obvious connection to craniofacial development (Table 1). The possibilities are: (1) BA1 expression of these genes has no biological function; (2) the mutants of these genes do have craniofacial phenotypes but the previous studies did not examine/detect them; or (3) these genes contribute to BA1 development redundantly with others, so that the individual mutation does not result in defects. Finally, the rest of the genes from our screen await mutant generation and analysis to reveal their role in facial development.

A recent study showed that Evf2 ncRNA and Dlx2 protein form a complex in embryonic tissues and that Evf2 functions as a transcriptional coactivator of Dlx2 in tissue culture cells(Feng et al., 2006). Therefore, it is possible that Evf2 similarly controls the activities of Dlx proteins in mdBA1 and thus the downregulation of Evf1/2 expression in Dlx5/6^-/- (Fig. 1) contributes to the gene expression changes observed in Dlx5/6^-/- mutants.

To date, none of the Dlx5/6-downstream genes has been shown to recapitulate the phenotypes of Dlx5/6^-/- when mutated in mice. Given that Dlx5/6 directly or indirectly regulate the expression of dozens of genes, it is likely that Dlx5/6 achieve their function through the combined efforts of many genes.

Among the Dlx5/6-downstream genes listed in Tables S1 and S2 (see Tables S1 and S2 in the supplementary material), BA enhancers have been characterized for only three (Hand2 and Dlx3/4), all of which showed some evidence of direct regulation by Dlx proteins(Charite et al., 2001; Sumiyama and Ruddle, 2003). Our identification of a Dlx5-regulated enhancer near Gbx2 suggests that Gbx2 might also be a direct target of Dlx5/6. Since Gbx2 is downstream of Dlx5/6 in BA1 and the otic vesicle,this enhancer is likely to function in one or both of these tissues.

Our screen discovered several genes, including Pou3f3 and Foxl2, the expression of which is largely restricted to mxBA1. Furthermore, we demonstrated that Pou3f3 is essential for the normal development of a part of the upper jaw(Fig. 4). These results provide the first evidence for the genetic program that specifically regulates upper jaw development, and argue against the idea that upper jaw fate is the default state of BA1, whereas lower jaw fate requires specification.

We found that the normal level of Pou3f3 expression in mxBA1 requires Dlx1/2 activity, whereas Foxl2 expression does not(Fig. 2). Therefore, the upper jaw-specific program in mxBA1 has both Dlx-dependent and Dlx-independent components. In addition, the mxBA1-specific genes that we identified show altered expression in Dlx5/6^-/- mdBA1. It is possible that there are genes, the expression of which is restricted to mxBA1 but which is not upregulated in Dlx5/6^-/- mdBA1; our screen was not designed to identify these.

Morphological analysis of the Dlx5 and Dlx6 single mutants indicates that they have very similar roles in mdBA1 development,despite the intriguing differences in their transcriptional activities(Fig 5; see Fig. S4 in the supplementary material). In addition, the analysis of Dlx1^-/-;5^-/-,Dlx2^-/-;5^-/-(Depew et al., 2005), Dlx1^-/-;6^-/- and Dlx2^-/-;6^-/- (Figs 6 and 7) establishes that the activity of Dlx5/6 to specify lower jaw fate is shared by Dlx1/2. However, Dlx1/2 are clearly less important than Dlx5/6 in this process because lower jaw development is essentially normal in Dlx1/2^-/- mutants(Qiu et al., 1997; Depew et al., 2005). Also,among Dlx1 and Dlx2, Dlx2 appears to have a greater influence on lower jaw development because the defects seen in Dlx2^-/-;5^-/- and Dlx2^-/-;6^-/- are more severe than those in Dlx1^-/-;5^-/- and Dlx1^-/-;6^-/-, respectively(Depew et al., 2005) (this study). These differences could be due to the molecular properties of each Dlx protein (as determined by amino acid sequence), but they could also be owing to differences in expression level, pattern or timing.

Another important conclusion from our analysis of Dlx1^-/-;6^-/- and Dlx2^-/-;6^-/- mutants concerns the functional relevance of the previous classification of Dlx genes based on their sequence homology (Stock et al., 1996)into type A (Dlx2, 3, 5) and type B (Dlx1, 4, 6). If two Dlx genes of the same type were functionally closer than those of different types,then removing one gene of each type would result in milder phenotypes than removing two genes of the same type, owing to compensation. Whereas Dlx1^-/-;5^-/- has milder phenotypes than Dlx2^-/-;5^-/-(Depew et al., 2005), Dlx2^-/-;6^-/- has more severe phenotypes than Dlx1^-/-;6^-/- (Figs 6 and 7), contrary to the prediction. Therefore, it appears that the greater sequence divergence between Dlx genes of different types does not result in more-dissimilar functions.

The Dlx code model proposed in previous studies(Qiu et al., 1995; Qiu et al., 1997; Depew et al., 2002b) is compatible with two different hypotheses on the nature of the codes: the qualitative hypothesis invokes the unique activities of Dlx1/2 versus Dlx5/6 proteins in specifying each domain of BA1, whereas the quantitative hypothesis proposes that the higher level of total Dlx protein in mdBA1 differentiates it from mxBA1. Although we do not have positive proof for the quantitative theory, a growing body of data has accumulated that cannot be explained by the qualitative theory, at least on its own. First of all, a recent study(Depew et al., 2005) and the present work (Fig. 6) showed that both Dlx1/2 and Dlx5/6 can regulate lower jaw development, and that Dlx1/2 and Dlx5/6 co-regulate several genes in mdBA1, directly or indirectly(Fig. 7). More importantly, Dlx1/2 perform opposing roles in mxBA1 versus in mdBA1: in mxBA1, Dlx1/2 positively regulate Pou3f3 and other mxBA1-specific genes to guide upper jaw development (Fig. 2); whereas in mdBA1, Dlx1/2 act partially redundantly with Dlx5/6 to repress Pou3f3 (and mxBA1 fate), and upregulate mdBA1-specific genes to promote lower jaw fate(Fig. 7). These results suggest that the patterning activity of a particular Dlx protein is context-dependent. For example, the activities of Dlx1/2 proteins might be modulated by some factors that are unevenly distributed between mxBA1 and mdBA1. Alternatively, Pou3f3 might be induced by the moderate level of Dlx proteins found in mxBA1, but repressed by the high level of Dlx proteins in mdBA1.

Fig. 8 summarizes our hypothesis on the regionalization of BA1 by Dlx genes. During early stages of craniofacial development, the BA1 ectomesenchyme forms without its regional identity. Subsequently, through a poorly understood mechanism that involves endothelin signaling and Mef2c(Ozeki et al., 2004; Ruest et al., 2004; Miller et al., 2007; Verzi et al., 2007), Dlx genes are expressed such that Dlx5/6 are restricted to mdBA1, while Dlx1/2 are in both mdBA1 and mxBA1. Dlx5/6 induce and/or maintain expression of the genes that promote the development of the lower jaw(Group A; see Fig. 8 legend). At the same time, Dlx5/6 repress other sets of genes (Group B and Group C) so that their expression is mostly confined to mxBA1. Here, Group B and Group C genes promote upper jaw development. Dlx1/2 participate in BA1 patterning by inducing and/or maintaining Group B genes in mxBA1. By contrast, in mdBA1, Dlx1/2 positively regulate Group A genes and repress Group B genes, directly or indirectly, to specify lower jaw fate.

We thank Winnie Liang for microarray analysis; Ayako Kuroda for ES cell blastocyst injection; Ugo Borello for help with manuscript preparation; Ugo Borello and Pooja Agarwal for help with the luciferase reporter experiment;and Gail Martin, Maria Barna, Ross Metzger and members of the Rubenstein and Martin laboratories for helpful discussions. We are grateful to Ken Weiss,Walter Birchmeier, Cam Patterson, Maxime Bouchard and Stephen Gold for providing plasmids for in situ hybridization probes. This work was supported by grants to J.L.R.R. from Nina Ireland, Hillblom Foundation, NIH/NIDCD R01 DC005667 and March of Dimes.