hand2 and Dlx genes specify dorsal, intermediate and ventral domains within zebrafish pharyngeal arches

Specification of pharyngeal arch-derived facial skeleton by transcription factor-encoding genes is a topic of considerable recent interest. Pharyngeal arches are comprised of neural crest-derived mesenchymal cells, with mesoderm-derived cores, surrounded medially by endoderm and laterally by ectoderm. Edn1 is a secreted protein important for dorsoventral jaw patterning: in mouse, mutations in Edn1 and its receptor Ednra cause homeotic transformations of lower jaw skeleton into upper jaw skeleton (Ozeki et al., 2004; Ruest et al., 2004). Studies of Edn1 signaling in zebrafish (Danio rerio) indicate that early pharyngeal arch patterning results in discrete dorsal, intermediate and ventral domains (D-I-V) in pharyngeal arch mesenchyme and pharyngeal-arch-derived skeleton (Kimmel et al., 1998; Miller and Kimmel, 2001; Miller et al., 2000; Miller et al., 2003; Walker et al., 2006). Edn1 is known to activate expression of many genes proposed to mediate D-I-V patterning, including hand2, gsc, nkx3.2 (formerly bapx1) and the Dlx genes (Miller et al., 2000; Miller et al., 2007; Walker et al., 2006). However, the boundaries of early D-I-V patterning genes have not yet been examined at later timepoints when the D-I-V skeletal regions are visible. In this study, we propose a unified definition of D-I-V domains, and examine interactions between genes that pattern these domains. We place a particular focus on the patterning of intermediate-domain joints and jointed skeleton. In this study, `joint' refers specifically to mesenchyme connecting early larval skeletal elements, whereas `joint region' includes both this joint mesenchyme and connected skeleton. We refer to the joint between Meckel's and palatoquadrate cartilages as the `jaw joint' of larval zebrafish.

Dlx genes are homeodomain-containing transcription factors, homologs of the single Distal-less gene in Drosophila (for a review, see Panganiban and Rubenstein, 2002). Mammalian Dlx genes are found in three bi-gene clusters (Qiu et al., 1997). Zebrafish also have three Dlx bi-gene clusters, containing dlx1a and dlx2a, dlx3b and dlx4b, dlx5a and dlx6a, as well as two additional Dlx genes, dlx2b and dlx4a (Stock et al., 1996). The two genes in each Dlx bi-gene cluster are approximately co-expressed (Ellies et al., 1997; Qiu et al., 1997), probably owing to shared enhancers (Ghanem et al., 2003; Park et al., 2004; Sumiyama et al., 2003). In mouse and zebrafish, functional redundancy is present both within and between these bi-gene pairs (Depew et al., 2005; Jeong et al., 2008; Qiu et al., 1997; Sperber et al., 2008; Walker et al., 2006). Within mouse pharyngeal arches, Dlx1 and Dlx2 (collectively referred to as Dlx1/2) expression extends further dorsally than Dlx5/6, which themselves show expression further dorsal than Dlx3/4 (Depew et al., 2002; Qiu et al., 1997). Dlx1^–;Dlx2^– mice primarily show dorsal skeletal defects (Qiu et al., 1997), and loss of dorsal specific molecular markers (Jeong et al., 2008). Conversely, Dlx5^–;Dlx6^– mice show homeotic transformations of lower jaw into upper jaw, corresponding to the exclusion of Dlx5/6 expression from dorsal arch regions (Beverdam et al., 2002; Depew et al., 2002). The skeletal homeosis of Dlx5/6 loss is mirrored by a ventral expansion of dorsal molecular markers, whereas several ventral markers (including Hand2) are lost.

Hand2 encodes a basic helix-loop-helix protein crucial for ventral facial pattern. Mice carrying a deletion in the pharyngeal arch-specific promoter of Hand2 have dramatically shortened lower jaws, but relatively normal patterning in joint regions and the upper jaw (Yanagisawa et al., 2003). When Hand2 is ectopically expressed throughout pharyngeal arches the upper jaw was partially transformed into an ectopic lower jaw (Sato et al., 2008). Thus, in mouse, the Edn1 targets hand2 and Dlx are directly implicated as homeotic selector genes along the pharyngeal arch dorsoventral axis.

In zebrafish, dlx3b and dlx5a are redundantly required for patterning specifically within intermediate domain-derived skeleton (Walker et al., 2006). By contrast, zebrafish hand2 nulls exhibit loss of lower jaws, but not upper jaws (Miller et al., 2003). hand2 is expressed ventral to nkx3.2, a marker of the jaw joint region (Miller et al., 2003). In zebrafish, hand2 mutants, nkx3.2 expands ventrally, indicating that hand2 patterns lower jaw identity in part by repressing jaw joint identity (Miller et al., 2003). However, it was unclear whether hand2 represses intermediate domain identity, because hand2 mutants consistently lose jointed-jaw skeleton (Miller et al., 2003).

Fate-mapping experiments have indicated approximately where skeletal patterning domains arise within early pharyngeal arches (Crump et al., 2006; Crump et al., 2004; Eberhart et al., 2006). However, these fate maps lacked the precision to directly connect early gene expression patterns to later skeletal shapes. Here, we present expression patterns that allow us to precisely define the dorsal, intermediate and ventral domains within zebrafish pharyngeal arches. We propose that the ventral domain comprises the hand2-expressing pharyngeal arch region, and the skeletal elements that are formed in this region. The ventral domain contains most of Meckel's and ceratohyal cartilages, and the dentary bone. The intermediate domain is the region of pharyngeal arches that expresses all Dlx genes, besides dlx2b (which is not expressed in anterior arches). Expression of the most restricted Dlx gene, dlx4a, reveals the borders of the intermediate domain. The intermediate domain includes the jaw joint region, and the second arch joint region, as well as the opercle and branchiostegal bones. Arch mesenchymal expression of dlx3b and dlx4b is also restricted to the intermediate domain. The dorsal domain is the region of the pharyngeal arch dorsal to dlx4a expression. Because dlx2a is expressed throughout the arch dorsoventral axis, co-labeling of dlx2a and dlx4a reveals the dorsal domain. The dorsal domain contains most of the palatoquadrate cartilage, including the distinctive pterygoid process, the hyomandibular cartilage and the maxillary bone. dlx5a and dlx6a expression does not correspond to a single domain.

In addition to defining D-I-V domains, this report examines the functional requirements for D-I-V patterning. We show that along with dlx3b and dlx5a, dlx4b is also redundantly required for intermediate domain skeleton. We report a transgenic revealing the expression pattern of trps1, a general marker of skeletal joint identity. We show that nkx3.2 and trps1 require dlx3b/4b/5a function for normal expression. We examine regulation between domains, noting that hand2 inhibits ventral expression of dlx3b, dlx4a, dlx4b and dlx5a. In hand2 mutants, nkx3.2 and trps1 expand to fill ventral space beneath expanded intermediate domain skeleton. However, even in hand2 mutants, expression of trps1 and nkx3.2 still requires dlx3b/4b/5a function. Despite differences in patterning ventral versus intermediate domains, we provide evidence that hand2 and dlx3b/4b/5a act in concert to repress dorsal domain identity.

Fish were raised and maintained under standard conditions and staged as described previously (Kimmel et al., 1995; Westerfield, 1995). Mutant lines were maintained on the AB background, and morpholinos were injected into AB fish. Df(Chr1)hand2^S6 (a null allele, hereafter: hand2^S6) and Is(Chr1)hand2^C99 (a hypomorphic allele, hereafter: hand2^C99) homozygotes were identified using previously described fully penetrant phenotypes, including dramatic heart defects (Miller et al., 2003; Yelon et al., 2000). edn1 mutants were identified as previously described (Miller et al., 2000).

trps1^j1271aGt (at most a hypomorphic allele) and dlx5a^j1073Et (a likely hypomorph, based on comparison with morpholinos) were generated using the Tol2 transposon T2KSAG, which contains enhancerless eGFP (Kawakami et al., 2004), during a screen for vital markers with specific expression patterns. trps1^j1271aGt and dlx5a^j1073aEt stocks have been submitted to ZIRC. After identification, carriers were outcrossed to AB background fish for several generations. Tail-PCR (Parinov et al., 1999) was used to identify genomic flanking regions, revealing that the J1271a insertion is at chr19: 43671269, inside the first intron of trps1, and the J1073a insertion site is chr19: 40245837, inside the first exon of dlx5a. A PCR primer in the transposon sequence (GCAAGGGAAAATAGAATGAAGTG) and primers in the flanking genomic DNA sequence (trps1^J1271aGt: TGTATTTTGACTCCTCAGTTCTGC, TACGCTCGAGTGAAGTGTGG or for dlx5a^J1073Et: ATTCCTGAG ACGGATGATGC, CGTAACAGCGCAATTTAGGA) were then designed and tested on 24 embryos segregating the expression pattern to show that we had correctly identified the insertion generating the expression pattern.

Alcian Blue and Alizarin Red staining was as described (Walker and Kimmel, 2007). For vital bone staining, fish were treated overnight with 0.000033% Alizarin Red in embryo medium, followed by de-staining in embryo medium. Fluorescent RNA in situ hybridization was carried out with a protocol modified from those described previously (Jowett and Yan, 1996; Welten et al., 2006). DNP-labeled probes were revealed with tyr-Cy5, dig-labeled probes were revealed using tyr-Cy3, fluorescein-labeled probes were revealed with tyr-fluorescein (available from Perkin-Elmer). Our full RNA in situ protocol is available online (http://wiki.zfin.org/display/prot/Triple+Fluorescent+In+Situ). Probes used are dlx2a (Akimenko et al., 1994), dlx3b (Akimenko et al., 1994), dlx5a (Walker et al., 2006), dlx6a (Walker et al., 2006), gsc (Schulte-Merker et al., 1994), dlx4a (Ellies et al., 1997), dlx4b (Ellies et al., 1997), hand2 (Angelo et al., 2000), nkx3.2 (Miller et al., 2003), sox9a (Yan et al., 2002) and eng2 (Ekker et al., 1992).

Antibody labeling was essentially as described (Nusslein-Volhard, 2002). For RNA in situ experiments and antibody staining experiments, embryos were raised in 0.0015% PTU (1-phenyl 2-thiourea) to inhibit melanogenesis (Westerfield, 1995). Confocal imaging was performed on a Zeiss LSM5 Pascal microscope, followed by image processing with Volocity software. Colors are digitally enhanced to increase visibility.

Morpholinos are injected at 2-3 nl into one- to two-cell stage embryos. Translation blocking morpholinos to dlx1a (Sperber et al., 2008), dlx2a (Sperber et al., 2008), dlx3b (Liu et al., 2003) and dlx5a (Walker et al., 2007), as well as a splice blocking morpholino to dlx4b (Kaji and Artinger, 2004) were purchased from Gene Tools using previously described sequences. dlx1a-MO and dlx2a-MO have previously been shown to be specific and effective through RNA rescue, and knockdown of transgenic dlx1a-GFP and dlx2a-GFP expression (Sperber et al., 2008). We confirm that dlx3b-MO strongly reduces Dlx3b immunolabeling (data not shown) (Liu et al., 2003). We also confirm that dlx4b-MO strongly disrupts dlx4b transcripts (Kaji and Artinger, 2004), without affecting any other Dlx gene (data not shown). Furthermore, in support of previous work (Liu et al., 2003), co-injection of dlx3b-MO with dlx4b-MO phenocopies otolith losses seen in a deletion that contains dlx3b and dlx4b (data not shown). In addition to the dlx5a translation blocking morpholino, we tested a splice blocking morpholino to dlx5a (dlx5aE2I2-MO: 5′-TATTCCAGGAAATTGTGCGAACCTG-3′). This morpholino had only nominal effects on splicing, and produced a different phenotypic suite from either the dlx5a translation blocking morpholino or the dlx5a mutant. As a result, dlx5aE2I2-MO was not used in any further analysis.

Co-injection of dlx3b-MO and dlx5a-MO causes intermediate-domain-specific defects without affecting dorsal or ventral structures (supporting Walker et al., 2006). Because dlx4b is in the same bi-gene cluster as dlx3b (Ellies et al., 1997), we hypothesized that dlx4b also functions in intermediate domain patterning. Injection of dlx4b-MO and co-injection of dlx3b-MO with dlx4b-MO fails to cause striking phenotypes (Fig. 1C,G). However, co-injection of dlx4b-MO with dlx5a-MO causes low penetrance intermediate defects (Fig. 1G). Furthermore, fish co-injected with dlx3b-MO;dlx4b-MO;dlx5a-MO (henceforth called dlx3b;4b;5a-MO) show defects throughout the intermediate domain at high penetrance (Fig. 1E,G). This synergism indicates that dlx3b, dlx4b and dlx5a function partially redundantly in facial patterning. dlx5a^J1271aEt homozygotes co-injected with dlx3b-MO and dlx4b-MO fish showed defects specifically within the intermediate domain (data not shown), similar to dlx3b;4b;5a-MO fish. By contrast, the most frequent defect in uninjected dlx5a^J1271aEt homozygotes is a low penetrant shortened symplectic phenotype, similar to dlx5a-MO treatment (Fig. 1D and data not shown). Hence, with both a morpholino and a mutant, we confirm that dlx5a acts largely redundantly with dlx3b and dlx4b to pattern the intermediate domain.

In mouse, Dlx1/2 have patterning requirements dorsal to Dlx5/6 (Depew et al., 2002; Qiu et al., 1997). To test whether zebrafish dlx1a/2a has patterning requirements dorsal to dlx3b/4b/5a, we injected dlx1a-MO and dlx2a-MO together and separately. When injected alone, dlx1a-MO and dlx2a-MO cause little skeletal deformity (data not shown). In support of previous work (Sperber et al., 2008), dlx1a-MO;dlx2a-MO co-injection results in low penetrance intermediate domain defects (Fig. 1F). In addition, dlx1a-MO;dlx2a-MO-treated fish often showed defects within the dorsal domain cartilages (palatoquadrate and hyomandibular cartilage) (Fig. 1F), indicating that the dorsal requirements of dlx1a and dlx2a are conserved.

Several models have been proposed in which Dlx genes function combinatorially to impart dorsoventral skeletal identities (e.g. Depew et al., 2005; Walker et al., 2006). To understand Dlx combinatorial patterning properly, we must understand how the expression domains of the Dlx gene fit together, which we can directly assay using multi-color fluorescent RNA in situ hybridization.

dlx2a is expressed throughout the dorsoventral axis of pharyngeal arches, excluding mesodermal cores (Kimmel et al., 2001). At 36 hours post fertilization (hpf), dlx4a expression is intermediate along the dorsoventral axis of zebrafish pharyngeal arches, and expression is not seen in dorsal or ventral arch regions (Fig. 2F,K), as revealed by labeling dlx4a expression alongside dlx2a. edn1 and hand2 expression is ventral to dlx4a at 36 hpf (Fig. 2I,N). Thus, we can delineate the 36 hpf ventral domain by hand2 expression, intermediate domain by dlx4a expression and dorsal domain by the expression of dlx2a dorsal to dlx4a (Fig. 2F). Similarly, at 36 hpf, dlx3b and dlx4b show intermediate specific expression, coincident with dlx4a boundaries within arch mesenchyme (Fig. 2C,H,M), although dlx3b also shows prominent epithelial expression (arrowheads in Fig. 2M).

Other Dlx genes show broader expression than dlx3b, dlx4b and dlx4a at 36 hpf (Fig. 2). The dorsal limit of dlx5a expression lies between the dorsal limits of dlx4a and dlx2a expression (Fig. 2G). In arch 1, dlx5a expression extends ventral to dlx4a expression (Fig. 2G) and is co-expressed with hand2 (data not shown), indicating that dlx5a is expressed in the first arch ventral domain. However, in the second arch, dlx5a expression shares a ventral boundary with dlx4a and is restricted from the ventral hand2-expressing region (Fig. 2G and data not shown). Matching the in situ analysis, Dlx3b protein is nested both dorsally and ventrally within the dlx5a^j1073aEt-expressing domain (see Fig. S1 in the supplementary material). dlx5a and dlx6a are largely co-expressed (Fig. 2B,G), although dlx6a has weaker expression intensity. Similar to dlx2a, the expression of dlx1a is seen broadly within pharyngeal arch mesenchyme, though with faint intensity (data not shown). dlx2b expression is not detected in the first two arches (Stock et al., 2006) (data not shown). Collectively, these results reveal a complex pattern of expression by 36 hpf, with the expression of hand2 ventral to dlx3b/4a/4b, which is nested within dlx5a/6a, which themselves are nested within dlx1a/2a boundaries (Fig. 2K-O).

Although the expression of dlx4a is intermediate-specific at 36 hpf (Fig. 2I), the earliest dlx4a expression is found in both ventral and intermediate arches (Fig. 2J). This observation of ventral dlx4a loss between 26 hpf and 36 hpf in wild-type fish, combined with the previous observation that hand2 represses dlx3b (Miller et al., 2003), suggested that hand2 ventrally represses Dlx expression. Indeed, the expression of dlx3b, dlx4b and dlx5a expands ventrally in hand2^S6 mutants at 36 hpf (Fig. 3). In wild-type fish, first arch expression of dlx5a extends more ventrally than dlx3b and dlx4b, whereas in hand2^S6 mutant fish, the three genes share a ventral expression border (Fig. 3C,D). Antibody staining for Dlx3b also expands ventrally in hand2^S6 (data not shown). Furthermore, in hand2^S6 fish, the expression of dlx4a expands, and fills the mesenchyme around edn1-expressing ventral mesodermal cores and ectoderm at 36 hpf (Fig. 3E,F). These results indicate that in wild type, hand2 inhibits the transcription of intermediate-domain-Dlx genes from the ventral domain.

The ventral inhibition of several Dlx genes by hand2 suggests that Dlx and hand2 may have some opposing roles in arch development. We examined the effect of dlx3b/4b/5a knockdown on two known hand2 targets: gsc and nkx3.2 (Miller et al., 2003), and the pre-skeletal marker sox9a [see Fig. S2 in the supplementary material, building upon Yan et al. (Yan et al., 2005)]. In the wild-type first arch, we see co-expression of the jaw-joint-region marker nkx3.2 with dlx4a and sox9a, but not with hand2 at 48 hpf (Fig. 4; see Fig. S3 in the supplementary material). nkx3.2 expression is reduced in dlx3b;4b;5a-MO (Fig. 4J). Conversely, we see strong expansion of nkx3.2 in hand2^S6 mutants (Fig. 4K). The expanded nkx3.2-expressing cells in hand2^S6 also express sox9a (Fig. 4C). When we inject dlx3b;4b;5a-MO into hand2^S6 (hand2^S6;dlx3b;4b;5a-MO), nkx3.2 expression is dramatically reduced (Fig. 4L), suggesting that hand2 represses nkx3.2 expression via its repression of dlx3b/4b/5a.

gsc is expressed in ventral and dorsal bands within the first two pharyngeal arches, avoiding the first arch intermediate domain (Fig. 4; see Fig. S3 in the supplementary material). In agreement with previous reports (Miller et al., 2003), ventral first arch gsc expression is lost in hand2^S6 (Fig. 4O). Conversely, in dlx3b;4b;5a-MO there are low penetrance fusions of the dorsal and ventral gsc expression bands (Fig. 4N). In hand2^S6;dlx3b;4b;5a-MO, there is an overall reduction in gsc expression (Fig. 4P). However, in hand2^S6;dlx3b;4b;5a-MO there are sometimes small protrusions of gsc expression attached to the dorsal gsc domain (Fig. 4P). This ectopic gsc expression may represent expansions of the dorsal gsc domain. Hence, the wild-type function of hand2 activates gsc and represses nkx3.2 (in agreement with Miller et al., 2003), whereas dlx3b/4b/5a acts to repress gsc and activate nkx3.2.

The expansion of dorsal identity in Dlx5^–/–;Dlx6^–/– mice (Depew et al., 2002) raises the issue of whether dorsal identity also expands in zebrafish injected with dlx3b;4b;5a-MO. To assay dorsal identity we used the dorsal muscle marker eng2 (Hatta et al., 1990), which specifically labels a region of the first arch mesodermal core, dorsal to dlx4a expression (Fig. 5A). Injection of dlx3b;4b;5a-MO into wild-type fish causes an increase in eng2 expression volume (Fig. 5N,U). However, these expanded eng2 expression domains are still located dorsal to dlx4a expression (Fig. 5F). In hand2^S6 mutants, eng2 expression is found ventral to its location in wild type (Fig. 5O), supporting Miller et al. (Miller et al., 2003). In hand2^S6, ectopic ventral nodules of eng2 expression sometimes appear within mesoderm ensconced by dlx4a expression (Fig. 5G). Although hand2^S6 mutants show changes in eng2 expression shape, the average volume of eng2 expression in hand2^S6 mutants does not differ from wild type (Fig. 5U). When dlx3b;4b;5a-MO is injected into hand2^S6, eng2 expression expands in volume (Fig. 5U) and is ventrally elongated (Fig. 5H), indicating that dlx3b/4b/5a and hand2 separately repress eng2. The overall expression of dlx4a is reduced in hand2^S6;dlx3b;4b;5a-MO (Fig. 5U), indicating a further loss of intermediate identity in these fish. Despite the shifting patterning domains seen with dlx3b/4b/5a and hand2 loss, we see no change in overall arch size, as assayed by dlx2a expression (Fig. 5U). Collectively, these results indicate that hand2 and dlx3b/4b/5a act in concert to inhibit dorsal identity in ventral/intermediate pharyngeal arches at 36 hpf.

To clarify the connection between hand2/Dlx expression and skeletal domains, we co-labeled fish for hand2 and Dlx gene expression alongside the pre-skeletal marker sox9a. Early in arch development, pharyngeal sox9a-expressing cells express dlx2a (see Figs S2, S3 in the supplementary material). However, by 60 hpf, most of the Dlx expression that we observe is lateral to sox9a expression (see Movie 1 in the supplementary material). dlx2a expression is maintained in cartilages near the Meckel's-palatoquadrate joint and the hyosymplectic-ceratohyal joint (Fig. 6A-C; see Fig. S2 in the supplementary material) and in mesenchyme lateral to these cartilages (see Movie 1 in the supplementary material). All arch expression of dlx2a is ventral to the neurocranium (see Fig. S2 in the supplementary material), consistent with previous findings (Verreijdt et al., 2006). dlx5a is expressed within cartilages in the Meckel's-palatoquadrate and the hyosymplectic-ceratohyal joint regions at 60 hpf. dlx5a is also expressed in mesenchyme lateral to much of the skeleton, except for dorsal aspects of the palatoquadrate cartilage, hyomandibular cartilage and most of the ceratohyal cartilage (Fig. 6E-L). dlx5a^J1073aEt expression is very similar to dlx5a in situ, but probably owing to the longevity of GFP proteins, dlx5a^J1073aEt is detectable in cartilages longer than dlx5a RNA (see Fig. S1 in the supplementary material). dlx6a expression is very similar to dlx5a, although the dorsal dlx6a expression border may not extend as far dorsally as dlx5a (Fig. 6I-L). Dlx3b and dlx5a^J1073aEt expression is found within precursor cells for both the opercle and branchiostegal bones (see Fig. S1 in the supplementary material). At 60 hpf, dlx4a expression is found in the Meckel's-palatoquadrate joint, and in the hyosymplectic-ceratohyal joint, as well as in mesenchymal cells lateral to these cartilages (Fig. 6M-P; see Movie 2 in the supplementary material). At 60 hpf, dlx3b, dlx4b and dlx4a show similar expression; however, as at 36 hpf, dlx3b is also strongly expressed in ectoderm (Fig. 6M-T and data not shown). By contrast, at 60 hpf, hand2 is expressed within much of the Meckel's and ceratohyal cartilages, as well as the surrounding mesenchyme, ventral to dlx3b and dlx4a expression (Fig. 6M-T). Hence, the relative dorsoventral expression borders of hand2 and the various Dlx genes are maintained from 36 hpf to 60 hpf, although outside of joint regions there is a progressive loss of Dlx gene expression in chondral elements. The D-I-V boundaries revealed by hand2 and Dlx at 60 hpf reveal which skeletal elements are formed from each expression domain.

Our expression data suggest that ventral arch cells lose their ventral identities and acquire intermediate identities in hand2-null mutants. When we also lower dlx3b/4b/5a functions, we see a gene expression shift, suggesting that dorsal identity expands. By these interpretations, we might also expect to see dorsalized homeotic phenotypes in arch-derived skeletons of such mutant and morpholino-injected fish. We constructed a phenotypic series of skeletal preparations to learn if the predicted homeosis is present (Fig. 7). Although there is extensive phenotypic variation (see Fig. S4 in the supplementary material), we found that the first arch skeletal phenotypes show the predicted changes most clearly.

In wild-type fish, there is a clear distinction between Meckel's and palatoquadrate shapes (Fig. 1A; Fig. 7A). In dlx3b;4b;5a-MO-injected fish, the jaw joint region is fused, but Meckel's cartilage is still immediately recognizable (Fig. 1D; see Fig. S4C in the supplementary material). Conversely, with just a partial loss of hand2, Meckel's cartilage is shortened, and the dentary bone is misshapen, but the joint-cleft between Meckel's and palatoquadrate cartilage is still clearly present (Fig. 7B; homozygous mutants for the hand2 hypomorphic allele c99). However, with stronger loss of hand2 function, the distinction between Meckel's and palatoquadrate is blurred (Fig. 7D, homozygotes for the hand2 deficiency s6; and Fig. 7C, transheterozygotes of S6 and C99). Instead, we interpret the midline cartilages in hand2^S6 as being transformed into ectopic palatoquadrate cartilage. Consistent with this interpretation, structures shaped like ectopic pterygoid cartilages variably seen in the hand2^S6 midline (arrows in Fig. 7C,D; Fig. S4E,F in the supplementary material). The ectopic expression of dlx3b, dlx4b and dlx5a seen in hand2^S6 raises the possibility that the ectopic cartilages seen in hand2^S6 require dlx3b/4b/5a function. Consistent with this hypothesis, the ectopic midline cartilages seen in hand2^S6 homozygotes are reduced when dlx3b;4b;5a-MO is injected (Fig. 7E; Fig. 8D; Fig. S4D in the supplementary material). Instead, the cartilages protruding from the reduced palatoquadrate are shaped like ectopic pterygoid processes (arrows in Fig. 7E). Injection of dlx5a-MO, or co-injection of dlx3b-MO with dlx4b-MO into hand2^S6 homozygotes produced subtler shifts in skeletal shape than injection of dlx3b;4b;5a-MO (see Fig. S4L in the supplementary material). When the hypomorphic hand2^C99 homozygotes are injected with dlx3b;4b;5a-MO, these pterygoid shapes are also seen, and there is a remarkable symmetry along the dorsoventral axis, consistent with the predicted homeosis (arrows in Fig. 7F).

Joints are key structures in the intermediate domain, and thus are predicted to expand in hand2 mutants. We used a transgenic line, trps1^j1271aGt (see Fig. 8P for details of the construct), in combination with cartilage labeling, to examine the joint and skeletal phenotypes more closely. trps1^J1271aGt is strongly expressed in joint regions of wild-type fish (Fig. 8A; matching our in situ results, not shown), consistent with findings in mouse (Kunath et al., 2002). Although reduced in intensity, we surprisingly see distinctive expression of trps1^J1271aGt in the joint region of dlx3b;4b;5a-MO-injected fish, even though a joint-cleft is lost (Fig. 8B). Instead, the trps1^J127aAGt-expressing cells lie just next to fused cartilages (Fig. 8B). In the corresponding region of hand2^S6 fish, trps1^j1271aGt labeling was dramatically expanded (Fig. 8C), revealing expansion of joint-cell fate that is completely unrecognized by skeletal staining alone. In marked contrast, trps1^j1271aGt expression is highly reduced in the first arch of hand2^S6;dlx3b;4b;5a-MO compared with uninjected mutants, similar to edn1 loss (Fig. 8D,E). Hence, we infer that joint cell identity is established by Edn1 signaling, is repressed by hand2, and requires dlx3b/4b/5a function.

The homeotic shape changes and molecular marker shifts we observe (Fig. 9B) indicate that hand2 and Dlx genes impart distinct identities to D-I-V domains in the first two arches (Fig. 9C). In previous modeling, all Dlx genes were thought to be co-expressed with hand2 in ventral aspects of arches (Depew and Simpson, 2006; Walker et al., 2006). Indeed, we show that there is initial co-expression of ventral dlx4a and hand2. However, dlx3b, dlx4b and dlx4a expression soon becomes restricted both dorsally and ventrally in the first two arches, indicating that by 36 hpf, zebrafish Dlx genes are more fully nested than was previously thought (Fig. 9). Intermediate-restricted Dlx nesting is also present in lamprey, which, together with our finding, suggests that dorsal/ventral Dlx restriction is basal within vertebrates (Daniel Medeiros, personal communication). We provide new evidence that dlx3b, dlx4b and dlx5a have overlapping functions in intermediate domain patterning, coincident with their overlapping expression within the intermediate domain. By 36 hpf, hand2 is expressed ventral to dlx4a, correlating with its specific requirements in ventral domain patterning (Miller et al., 2003). The stacked expression of dlx4a and hand2 persists until after major cartilage domains have been formed. We recognize that owing to the dynamic nature of gene expression, only precise fate maps can definitively connect expression patterns between different time-points. Nonetheless, the differential expression and requirements of dlx1a/2a, hand2 and dlx3b/4b/5a provides a mechanism to generate discrete D-I-V domains within pharyngeal arches and skeleton (Fig. 9).

Similar to the findings in mouse (Qiu et al., 1997), zebrafish dlx1a and dlx2a function redundantly to pattern dorsal identity. However, more ventrally restricted Dlx genes dlx3b/4b/5a lack dorsal requirements, supporting a correlation between Dlx expression and function. We have also noted additional Dlx nesting: dlx3b/4b nest within the dlx5a expression domain. It will be important for future work to test the functional relevance of this deeper Dlx nesting, which may reveal patterning sub-domains.

Skeletal shape changes in Edn1 signaling pathway mutants are correlated with changes in hand2 and Dlx expression. mef2ca and furina mutants, which only partially reduce Edn1 signaling, result in the loss of dlx4b and dlx5a expression, but no persistent losses in hand2 expression (Miller et al., 2007; Walker et al., 2006). The skeletal defects in mef2ca and furina mutants include joint loss, ectopic cartilages and second arch bone fusion, but no ventral defects (Miller et al., 2007; Walker et al., 2006), similar to dlx3b;4b;5a-MO. By contrast, edn1 mutants and plcb3 mutants, in which Edn1 signaling is strongly reduced, have strong loss of hand2, dlx3b and dlx5a expression (Miller et al., 2000; Walker et al., 2006; Walker et al., 2007). The skeletal defects seen in edn1 and plcb3 mutants include severe defects in both intermediate and ventral skeleton (Miller et al., 2000; Walker et al., 2006; Walker et al., 2007), similar to hand2^S6;dlx3b;4b;5a-MO. Furthermore, prominent expansions of the dorsal marker eng2 are seen in both edn1 mutants (Miller et al., 2003) and hand2^S6;dlx3b;4b;5a-MO. Hence, the overall arch patterning domains identified in this study of hand2/Dlx expression and function closely mirror the domains identified previously from studies of Edn1 signaling.

We examined skeletal phenotypes in fish treated with morpholinos to various combinations of dlx1a, dlx2a, dlx3b, dlx4b and dlx5a, revealing redundant patterning roles for these genes. However, the conclusions we draw are limited because we lack known null alleles in any Dlx gene. Furthermore, all dlx4a and dlx6a morpholinos tested to date have failed to disrupt splicing convincingly, or produce any skeletal phenotype (data not shown). It will be very important for future studies to examine null alleles of Dlx genes. For example, zebrafish dlx5a^–;dlx6a^– nulls could conclusively test whether loss of these two genes in fish results in the homeotic transformations observed in Dlx5^–;Dlx6^– mutant mice. Although we have demonstrated that dlx3b/4a/4b expression does not extend as far dorsally as dlx5a/6a, we have not observed a functional consequence of this expression difference. The expression difference between dlx3b/4a/4b and dlx5a/6a may be present because the major D-I-V domains are further subdivided into smaller patterning domains by Dlx expression. With genetic nulls, we could conclusively assay the functional relevance of expression differences between dlx3b/4a/4b and dlx5a/6a.

In wild-type zebrafish, trps1 expression faithfully labels joint regions. However, in our mutants, we found several examples of trps1-expressing joint cells that do not connect skeletal elements. For example, although dlx3b;4b;5a-MO injection causes a fusion between Meckel's and palatoquadrate cartilages, some trps1 expression is found in cells surrounding the location where the joint would have been. Similarly. some expression of nkx3.2 remains, indicating that even when normally jointed cartilages are fused together, remnants of joint pattern can remain. In dlx3b;4b;5a-MO, trps1 expression spans the fused opercle-branchiostegal bone, including a region of the bone that does not connect to skeleton. As a more extreme example, in hand2^S6, Meckel's cartilage is lost, and instead there is an enormous mass of ectopic trps1-expressing cells. In hand2^S6, the most anterior trps1-expressing cells sometimes extend well beyond any apparent bone or cartilage, indicating that joint cells can arise separately from skeleton. The disassociation of joint cells from jointed skeletons in our mutants leads us to ask how wild-type fish obtain a perfect correlation of jointed skeleton with jointing cells. It will be intriguing to discover the developmental relationship between joint cells and jointed skeletal elements.

Losses in ventral or intermediate domain identity result in compensatory expansion of identity from other domains (Fig. 9). When hand2 is lost, dlx3b, dlx4a, dlx4b and dlx5a expression expands ventrally at 36 hpf. New research indicates that hand2 also inhibits Dlx gene expression in mouse (David Clouthier, personal communication). Coincident with the expansion of these Dlx genes, we observe expansion of intermediate domain cartilages, trps1 expression and nkx3.2 expression in hand2^S6. Injecting dlx3b;4b;5a-MO into hand2^S6 mutants results in a loss of joint identity, indicating that hand2 represses joint identity via its repression dlx3b/4b/5a. When both hand2 and dlx3b/4b/5a functions are reduced, the arch volume (indicated by 36 hpf dlx2a expression) remains fairly constant, and dorsal identity expands. The expansion of dorsal identity in hand2^S6;dlx3b;4b;5a-MO is similar to expansions of dorsal identity observed in Edn1 pathway mutants. For example, dorsalizing homeoses are seen in both zebrafish and mouse Edn1 mutants (Kimmel et al., 2003; Ozeki et al., 2004), and Ednra mutants/morpholinos (Nair et al., 2007; Ruest et al., 2004), as well as mouse Dlx5^–;Dlx6^– mutants (Beverdam et al., 2002; Depew et al., 2002). In our study, we examined markers broadly and specifically required for ventral and intermediate domain identity, but a broadly expressed dorsal domain specific marker has remained elusive. Candidates for dorsal specification genes have been recently proposed (Jeong et al., 2008; Zuniga et al., 2010).

We thank Macie Walker, Craig Miller, Ryan Macdonald, Gage Crump and April DeLaurier for comments during manuscript preparation. We thank Joseph Martial's lab for sharing a fluorescent in situ protocol prior to publication. We thank Jose Morillo and other members of the Johnson lab for identifying and maintaining enhancer or gene traps, and Ryan McAdow for identifying transposon insertion sites. We thank John Dowd and the UO fish facility for the care and maintenance of our fish. This work was supported by NIH grants DE13834, HD22486 (to C.B.K.), DK069466 (to S.L.J.) and DTG GM007257 (to J.C.T.). Deposited in PMC for release after 12 months.