Zebrafish hox genes: genomic organization and modified colinear expression patterns in the trunk

Clustered homeobox genes were first described for Drosophila melanogaster, where two divided clusters form the homeotic complex, Hom-C (Lewis, 1978; reviewed by McGinnis and Krumlauf, 1992). Similarly clustered homeobox genes (Hox genes) have since been described for a wide range of species. The invertebrates studied to date, for example arthropods (including Drosophila, Tribolium, and Artemia; Akam et al., 1994), an annelid (Dick and Buss, 1994), the nematode C. elegans (Wang et al., 1993), and amphioxus (Garcia-Fernandez and Holland, 1994), possess a single Hox cluster. By contrast, the tetrapod vertebrates, including human (Acampora et al., 1989), mouse (Duboule et al., 1986; Graham et al., 1989), and Xenopus (Harvey et al., 1986), have four separate Hox clusters lying on four different chromosomes. In mouse, 39 Hox genes (McGinnis and Krumlauf, 1992; Zeltser et al., 1996) are organized into four clusters, termed clusters A to D. Within each cluster the genes are assigned, according to sequence homology and location in the genome, to one of 13 possible paralogy or cognate groups, where paralogue group 1 lies most 3′ on the genome and paralogue group 13 lies most 5′ (see Scott, 1992, for a full description of Hox nomenclature). The four tetrapod Hox clusters have probably arisen via duplication and divergence events from a single ancestral cluster (Fig. 1). Supporting evidence for such an ancestral condition is provided by the description of a single Hox cluster for amphioxus, a member of the cephalochordates, sister taxon to the vertebrates, which may represent an extant example of an intermediate stage in Hox gene evolution (Garcia-Fernandez and Holland, 1994). The existence of a single Hox cluster for amphioxus suggests that the duplication of the Hox clusters could have occurred very close to the origins of the vertebrate line.

In addition to conservation of the 3′ to 5′ genomic organization of the Hox clusters, spatial colinearity of Hox gene expression has also been conserved during evolution. In general, Hox gene expression commences in the posterior of the embryo then spreads anterior until a well defined expression limit is reached; this ‘mature’ pattern of expression is then maintained (Deschamps and Wijgerde, 1993). Colinearity describes the observation that the anterior expression limit of a given Hox gene mirrors its position within the cluster; i.e. a more 3′ located gene has a more anterior expression limit within the embryo. Colinearity has been demonstrated by careful expression analyses for the mouse Hoxa, Hoxb, Hoxc and Hoxd clusters (Gaunt et al., 1988; Graham et al., 1989; Peterson et al., 1994; Duboule and Dolle, 1989), and the Xenopus Hoxb cluster (Dekker et al., 1992; Godsave et al., 1994).

Hox genes are implicated in imparting anteroposterior (AP) identity to the embryonic body plan (McGinnis and Krumlauf, 1992), both by analogy to the role of the Drosophila homeotic complex and from the results of functional studies in mouse and Xenopus (reviewed by Krumlauf, 1994; McGinnis and Krumlauf, 1992). Major sites of Hox gene expression include the CNS and the sclerotomal component of the somites. Functional studies have frequently concentrated on the effects of loss or gain of Hox gene function on the sclerotome-derived vertebrae, in part due to the ease of observation of these structures. These experiments have lent support to the notion that the AP body axis is patterned by a ‘Hox code’ in which a given combination of Hox gene products specify a particular axial identity within the mesoderm (Kessel, 1992; Kessel and Gruss, 1991). Furthermore, a careful comparison of the expression domains of multiple Hox cluster genes in mammals and avians, which have distinct and characteristic vertebral organizations, has revealed that Hox gene expression domains are closely correlated with specific axial structures rather than with a specific somite or prevertebra (Burke et al., 1995). For example, Burke and colleagues have shown that in a variety of vertebrates the anterior limit of Hoxc-6 gene expression lies in the mesoderm adjacent to the site of forelimb outgrowth, although this point is at different AP levels in different species.

We were interested to study the hox genes in the teleost zebrafish (Danio rerio) because it represents an example of a non-tetrapod vertebrate, and is additionally distinguished by its unique advantages as a developmental model system. We examined both genomic organization and gene expression patterns in order to further investigate the overall extent of conservation of Hox gene organization and function amongst the vertebrates. Hox clusters have been revealed in a wide variety of phyla, often by PCR-based screens that provide sequences of multiple short cDNAs. Such recent PCR surveys for the teleost fishes Fundulus and zebrafish have provided data consistent with the idea that these non-tetrapod vertebrates may have a similar hox cluster organization to the tetrapod vertebrates (Misof and Wagner, 1996; Misof et al., 1996). In addition, previous studies have provided cDNA or genomic clones for the zebrafish homologues of several hox genes. These include: the most 5′ genes in the hoxd cluster, hoxd10, hoxd11, hoxd12 and hoxd13 (van der Hoeven et al., 1996), several hoxa cluster genes, hoxa1 (Alexandre et al., 1996), hoxa9, hoxa10, hoxa11 and hoxa13 (Sordino et al., 1996) and two hoxb cluster genes, hoxb5 and hoxb6, which were found on a single genomic clone, as were two hoxc cluster genes, hoxc5 and hoxc6 (Njølstad et al., 1990, 1988c). Thus, to date, there has been no evidence to suggest that the organization of hox genes of the teleost zebrafish differs from that of the tetrapod vertebrates. However, a recent analysis of hox genes in another, highly derived, teleost fish, Fugu rubripes has revealed significant differences in genomic organization in comparison to that of the tetrapod vertebrates (Aparicio et al., 1997); for example, the Fugu hox clusters have lost at least 9 genes and include a novel member of paralogue group 2.

Expression analyses have been carried out for several of the posteriorly expressed members of the zebrafish hoxa and hoxd clusters; hoxa9, hoxa10, hoxa11, hoxa13, hoxd10, hoxd11, hoxd12 and hoxd13 (van der Hoeven et al., 1996; Sordino et al., 1996). These genes exhibit colinear expression in the trunk, as seen in the tetrapod vertebrates, however the anterior limits of expression are approximately 10 metameres more anterior than those of their murine counterparts (van der Hoeven et al., 1996; Sordino et al., 1996). The expression domain of zebrafish hoxc-6 also shares several characteristics with that of the tetrapod gene (Molven et al., 1990); this was revealed using an antibody raised against the equivalent Xenopus gene. Those zebrafish hox genes with anterior expression limits in the hindbrain also tend to show strong conservation of expression patterns with other species, although some specific spatial and temporal differences do exist (Prince et al., 1998; Alexandre et al., 1996).

We have taken a comprehensive approach to screen for zebrafish hox genes; we used 3′RACE-PCR with degenerate primers targeted to specific paralogue groups to obtain cDNAs long enough for direct use in expression studies. In addition, we have analyzed the linkage relationships of these genes by making use of a novel resource, a series of zebrafish/mouse somatic cell hybrids (Ekker et al., 1996). In this manner we have revealed that the zebrafish hox gene organization differs from that of both tetrapod vertebrates and Fugu. The zebrafish has at least three additional genes within the four hox clusters, and at least three extra genes beyond the four clusters. We have also analyzed expression of hox genes during early zebrafish development. The genes in paralogue groups 1, 2 and 3, which are primarily expressed in the hindbrain region, have been described elsewhere (Prince et al., 1998); in this study we concentrate on those genes with a predicted role in patterning the trunk region. We have assessed expression limits of zebrafish hox genes within the CNS and the paraxial mesoderm, facilitating comparisons across paralogue groups, across hox clusters, and across species. We find that in general the phenomenon of colinearity is conserved, but that anterior expression limits are compacted over a shorter AP extent of the embryonic axis than in tetrapod species, and that members of paralogue groups 7 and 8 share similar expression limits, perhaps reflecting their common evolutionary origin.

The hoxd4 gene was cloned from an embryonic zebrafish cDNA library in lambda ZAP (kindly provided by D. J. Grunwald), by screening with a murine Hoxd-4 cDNA (kindly provided by M. Featherstone). A clone of about 2.8 kb containing the entire zebrafish hoxd4 coding region was obtained.

Additional hox genes were cloned by 3′RACE-PCR, carried out as previously described (Frohman, 1993). cDNA was reverse transcribed, using the Gibco-BRL Superscript kit according to manufacturer’s instructions, from 24-hour zebrafish embryo RNA prepared as described by Chomczymski and Sacchi (1987) PCR reaction conditions were : 1 cycle at 94°C 2 minutes; 35 cycles at 94°C 1 minute, 45-50°C 2 minutes, 72°C 2.5 minutes; 1 cycle at 72oC 10 minutes. Degenerate primers were designed from published tetrapod Hox sequences to be specific for one or more paralogue groups, in all cases the primer sequences lie in the first half of the homeobox:

Paralogue groups 6 and 7:-5′ KRGRQTYT and 5′ TLELEKEF.

Paralogue group 8:-5′ TLELEKEF and 5′KEFLFNP.

Paralogue groups 9 and 10:-5′ TLELEKEF and 5′KEFLFNM.

Paralogue group 11:-5′EREFFFN and 5′VYINKEK.

Paralogue group 12:-5′EFLVNEF and 5′NEFITRQ.

PCR products were cloned into the Promega pGEM-T cloning vector (or pBluescript, Stratagene) according to the manufacturer’s instructions.

The zebrafish hoxb4 (zf-13;Njølstad et al., 1988a), hoxb5 (zf-21;Njølstad et al., 1988c), hoxb6 (zf-22; Njølstad et al., 1990), hoxc5 (zf-34; Ericson et al., 1993), hoxc6 (zf-61; Njølstad et al., 1990) and hoxa5 (zf-54; Njølstad et al., 1988b) genes have been cloned previously. hoxb4, hoxc5, hoxa5 and hoxc6 were cloned by the 3′RACE-PCR technique as described above but using specific primers to the published homeobox sequences. Subclones of the genomic hoxb5 and hoxb6 (generously provided by Anders Molven) were derived by a PCR based approach. For hoxb5 a 600 bp cDNA encompassing exon 1 was synthesized; for hoxb6 a 600 bp cDNA encompassing exon 2 was synthesized. Some additional hox clones were derived using 3′RACE-PCR with primers designed directly to the published sequences of Misof et al. (1996). cDNAs for the following genes were obtained in this manner (Misof et al. nomenclature indicated in parentheses): hoxc4 (z-96), hoxc8 (z-179), hoxb9 (bz23), hoxc9 (bz39), hoxx9 (z10), hoxd9 (z-28), hoxa10 (z-140), hoxd10 (z-82).

Clones were screened by sequencing of double stranded templates (Sequenase, US Biochemicals Inc.) from forward and reverse primers, the most 5′ 400 bp, of coding sequence and 3′ untranslated region, was then sequenced in both directions using internal oligonucleotide primers. Sequence analyses and comparisons were performed using the Wisconsin genetics GCG software package. All the sequences described have been submitted to the EMBL database and are available under accession numbers Y14526-14548 and Y13944-13950. The sizes of the cDNAs obtained for each gene are as follows: hoxa5 900 bp; hoxa7 800 bp; hoxa8 1050 bp; hoxa11 800 bp; hoxb4 1600bp; hoxb7 700 bp; hoxb8 1100 bp; hoxb9 500 bp; hoxb10 1000 bp; hoxc4 900bp; hoxc5 900 bp; hoxc6 1200 bp; hoxc8 1500 bp; hoxc9 800 bp; hoxc10 800bp; hoxc11 600 bp; hoxd9 700 bp; hoxd10 750 bp; hoxd11 750 bp; hoxd12 800 bp; hoxx4 900 bp; hoxy6 1000 bp; hoxx9 1000 bp.

Embryos were staged as described by Kimmel et al. (1995). In situ hybridizations were performed as previously described (Prince et al., 1998). Some embryos were processed after in situ hybridization for immunostaining with the F59 anti-myosin antibody (Crow and Stockdale, 1986; kindly provided by F. Stockdale), essentially as previously described (Devoto et al., 1996). Briefly, embryos were rinsed in phosphate-buffered saline, 0.1% TWEEN-20 (PBT), preblocked with 2%BSA, 5% goat serum in PBT, then incubated in a 1/10 dilution of F59 antibody tissue culture supernatant overnight at 4°C. Embryos were rinsed in PBT and incubated in a goat anti-mouse secondary antibody conjugated to horseradish peroxidase (Jackson Immunochemicals) at room temperature for 4 hours, then rinsed again before processing with diaminobenzidine.

We have obtained 32 different cDNAs of zebrafish hox genes from paralogue groups 1 through 12. (The members of groups 1, 2 and 3, which are primarily expressed in the developing hindbrain region, have been described elsewhere; Prince et al., 1998). We have assigned the genes to specific hox clusters by PCR screening of zebrafish/mouse somatic cell hybrids (Ekker et al., 1996) with gene-specific primers. Hybrids were first identified that contained specific hox complexes, by making use of published sequences. Using the hoxa1 sequence (Alexandre et al., 1996), hybrids ZFB 54, 57, 65, 206 and 211 were found to contain the hoxa complex. Using hoxb5 and hoxb6 sequences (Njølstad et al., 1988c; 1990), the hybrids ZFB-73, 212, 239 and LFFB-3 were found to contain the hoxb complex. Using hoxc5 and hoxc6 sequences (Ericson et al., 1993; Njølstad et al., 1990), the hybrids ZFB-11, 50, 207 and LFFB-13 were found to contain the hoxc complex, and finally, using hoxd4 (this study), hybrid LFFB-10 was found to contain the hoxd complex. In all but three cases, each of our hox gene cDNAs was identified in each hybrid line that contains one of the four hox clusters, thus we have positively identified 29 members of zebrafish hox clusters A-D. Consistent with these results, hoxb genes were found to be absent from the DNA of a mutant strain of fish lacking the hoxb cluster, and hoxc genes were found to be absent from a mutant strain lacking the hoxc cluster (Fritz et al., 1996; Andreas Fritz personal communication). The three remaining hox genes were not located on any of the four linkage groups corresponding to hox clusters A-D, and therefore must lie beyond the expected four clusters. These genes fall into paralogue groups 4, 6 and 9, and we have termed them hoxx4, hoxy6 and hoxx9, (hoxx4 has previously been described, Prince et al., 1998). Interestingly, hoxx4 and hoxx9 are both present within the same set of hybrids, suggesting that these genes are linked and hinting at the existence of an additional hox cluster in the zebrafish. Diagnostic amino acids both within and beyond the homeobox region have been described for the individual hox paralogue groups of the vertebrates (Sharkey et al., 1997); we were therefore additionally able to use the deduced amino acid sequences for the homeobox and the 3′ coding region to assign our genes to specific paralogue groups (Table 1).

By combining the information on location within specific hox clusters and paralogue groups for each gene, we have been able to deduce the genomic organization of the zebrafish hox genes (Fig. 1). Although this organization shares many properties with that of the tetrapod vertebrates, we have found additional genes both beyond and within the four hox clusters. The hoxx and y genes lie beyond the clusters as described above. Within the clusters, there is a member of paralogue group 8 on the hoxa cluster, the zebrafish hoxa8 gene, and a member of paralogue group 10 on the hoxb cluster, zebrafish hoxb10; no equivalent genes have been described for the tetrapod vertebrates. In addition, the previously identified zebrafish hox gene zf-114 (Molven et al., 1992), has been identified as hoxc3, and thus represents a third gene within the 4 clusters that has no tetrapod equivalent. The existence of apparent ‘gaps’ in the clusters, where we have not isolated a zebrafish gene equivalent to a described tetrapod gene, does not necessarily imply that these genes have been lost in the zebrafish. Rather, such gaps may merely reflect the incomplete nature of our survey; confirmation of any gene losses from the zebrafish hox clusters will require direct analysis of genomic DNA.

The amino acid sequences of the zebrafish hox genes in paralogue groups 4-11 are compared to consensus sequences for the appropriate vertebrate paralogue group (Sharkey et al., 1997) in Table 1. In general, residues within the homeobox that have previously been characterized as invariant for particular paralogue groups are conserved in the zebrafish. We see only two exceptions to this rule: in hoxc10, where homeobox residue #59 is a leucine rather than a methionine, and in hoxd11, where homeobox residue #44 is a serine rather than an asparagine. Beyond the homeobox region, diagnostic amino acids for each paralogue group (Sharkey et al., 1997) are also generally conserved in the zebrafish. The hoxa8 gene, which has no direct homologue in the tetrapod vertebrates, nevertheless maintains all the diagnostic residues expected for a member of paralogue group 8. The other gene with no direct homologue, hoxb10, similarly maintains all the diagnostic amino acids for group 10 within the homeobox, although only 4 out of 10 diagnostic residues beyond the homeobox are conserved; however, it should be noted that the other zebrafish paralogue group 10 genes also do not show 100% conservation of these diagnostic residues (for example hoxc10 conserves 7 out of 10 diagnostic amino acids beyond the homeobox). The hoxx and y genes similarly share the majority of the diagnostic amino acids of the paralogue groups we have allocated them to. The hoxx4 gene maintains all 7 of the diagnostic amino acids beyond the homeobox, and shows just a single unexpected change within the homeobox (homeobox residue #43 is a methionine rather than the predicted leucine). The hoxy6 gene also has a single unexpected amino acid change within the homeobox (homeobox residue #46 is a serine rather than a threonine), but the diagnostic leucine beyond the homeobox is maintained. Finally, hoxx9 shows complete conservation of invariant group 9 residues. Thus, it appears that the unique zebrafish hox genes, with no tetrapod counterparts, are nevertheless bona fide members of the hox gene family, falling into specific paralogue groups.

The linked hoxx4 and hoxx9 genes are most closely related to Hoxa cluster genes of other species, according to phylogenetic tree analysis (performed with the Megalign module of the Laser gene programme; DNASTAR, Inc.). The predicted amino acid sequence of hoxx4 shows 69% identity over 90 amino acids with that of murine Hoxa-4, and that of hoxx9 shows 84% identity over 37 amino acids to murine Hoxa-9. Furthermore, hoxx9 shares 29 out of 30 amino acids with zebrafish hoxa9 (Sordino et al., 1996), although there are multiple differences between these genes at the nucleotide level. No zebrafish hoxa4 gene has yet been identified. The unlinked hoxy6 gene is most closely related to zebrafish hoxc6 (76% identity in amino acid sequence), and also to Hoxc-6 of other species.

We have used our zebrafish hox gene cDNAs to generate antisense digoxigenin-labeled riboprobes for use in whole-mount in situ hybridizations. We analyzed expression patterns of the zebrafish hox genes primarily at the 10 somite (s) and 20s stages of development. We found that the 20s stage was particularly convenient for assessing the anterior limits of expression within the CNS, as at this stage expression was at relatively high levels, yet the anterior limit had reached its ‘mature’ position (i.e. the anterior expression limits were not found to shift further anterior within the CNS after the 20s stage). To assist in accurately assessing the anterior expression limit within the CNS, we made use of two other molecular markers to provide fixed reference points. The krox-20 gene is expressed discretely in rhombomeres (r) 3 and 5 of the developing hindbrain (Oxtoby and Jowett, 1993) and the F59 antibody (Crow and Stockdale, 1986) recognizes myosin and thus reveals the location of the somites.

Expression of hox genes in paralogue groups 4, 5 and 6, at the 20s stage, is shown in Fig. 2. In general, the expression patterns share certain characteristics: (1) there are obvious anterior limits of expression in the CNS; (2) expression tends to be at its highest level at the anterior limit and gradually reduces toward the posterior; (3) several of the genes are expressed in the tailbud and in a segmental manner in the posterior paraxial mesoderm. For each gene the anterior expression limit in the CNS was assessed by observing several embryos, in both lateral and dorsal views, and comparing the expression limits of each hox gene to the krox-20 and myosin expression domains. (It should be noted that myosin expression is at lower levels in the first somite than in the remaining somites, and in some cases somite 1 was only visible at particular focal planes using Nomarski optics). For the more anteriorly expressed genes (paralogue group 4) krox-20 labelling enabled an expression limit within the hindbrain to be determined. For the remaining, more posteriorly expressed genes, anterior expression limits within the spinal cord were determined relative to the adjacent somite or somite boundary.

Within individual clusters, spatial colinearity appears to be maintained in the CNS for the genes of paralogue groups 4-6. Thus, considering the hoxb cluster: hoxb4 has an anterior limit at the rhombomere (r) 6/7 boundary (Fig. 2C,D), hoxb5 at the anterior of somite 1 (Fig. 2K,L), and hoxb6 at the s1/2 boundary (Fig. 2Q,R). Similarly, for the hoxc genes: hoxc4 has an anterior limit within r7 (Fig. 2E,F), hoxc5 within s1 (Fig. 2M,N), and hoxc6 at the s2/3 boundary (Fig. 2S,T). Note that although each cluster shows colinearity over this subset of genes, the absolute anterior limits are not maintained for different members of a given paralogue group. For example, for paralogue group 4, hoxb4 and hoxd4 share an anterior limit at the r6/7 boundary (Fig. 2D,H), whereas hoxc4 has a slightly more posterior limit, within r7 (Fig. 2F). We were also interested to compare the expression limits of the hox genes from beyond clusters A-D, hoxx4, hoxy6 and hoxx9, with those of the clustered hox genes. The hoxx4 gene has a fairly similar expression limit in the CNS to those of the other group 4 genes (Fig. 2A-H), at about the rhombomere 7/8 boundary; in addition the general expression profile appears conserved with the other paralogue group 4 members (Fig. 2A,B). Conversely, the hoxy6 gene has a rather different expression profile to the other paralogue group 6 members (Fig. 2O-T). Expression of hoxy6 is at low levels in the CNS, with an anterior limit slightly posterior to that of hoxb6 and hoxc6, at the s3/4 boundary, but with high expression levels ventrally, in the endoderm (Fig. 2O,P).

Several of the genes (hoxa5, Fig. 2I,J; hoxb5, Fig. 2K,L; hoxb6, Fig. 2Q,R and hoxc6, Fig. 2S,T), also have expression domains lateral to the hindbrain or anterior spinal cord, in very discrete domains which may represent neural crest derivatives or mesenchymal cells. In addition, as mentioned above, many of the genes show expression in the tailbud and in a segmental pattern in the posterior paraxial mesoderm. This expression pattern is shared by all of the hoxb and hoxc cluster genes within paralogue groups 4-6 (Fig. 2C,E,K,M,Q,S), and by hoxd4 (Fig. 2G) and hoxy6 (Fig. 2O). The hoxb6 gene is also expressed very ventrally along the length of the embryo in the region where the pronephric ducts are forming (Kimmel et al., 1995; Fig. 2Q). Several genes exhibit low level expression in the more anterior mesoderm with an anterior limit posterior to that in the CNS; this paraxial mesoderm expression is at higher levels at earlier stages and is considered in more detail below. Ventral expression, probably correlating with the location of the endoderm, was also observed for several of the hox genes, including hoxb5, hoxb6 and hoxc6 (Fig. 2K,Q,S).

Expression of hox genes in paralogue groups 7, 8, 9 and 10 is shown in Fig. 3. Once again we were able to assess the anterior expression limits in the CNS according to the adjacent somite. Considering the hoxb cluster genes: hoxb7 has an anterior expression limit adjacent to the s3/4 boundary (Fig. 3C,D), this limit is shared by hoxb8 (Fig. 3G,H), and by hoxb9 (Fig. 3M,N), the anterior limit for hoxb10 is at the s7/8 boundary (Fig. 3T). Thus, surprisingly, hoxb7, hoxb8 and hoxb9 appear to have very similar anterior expression limits in the CNS (compare Figs 3D,H and N). There are, however, obvious differences in the expression patterns of these three genes. For example, hoxb8 expression in the CNS rapidly reduces towards the posterior, from an initial high level near the anterior limit (Fig. 3G), whereas hoxb7 and hoxb9 expression in the CNS is maintained at similar levels along its AP extent (Fig. 3C,M). Beyond the CNS, hoxb8 is expressed in the pronephric ducts, in anterior paraxial mesoderm (Fig. 3G), and in bilateral domains adjacent to the neural tube in the anterior spinal cord (Fig. 3H). By contrast, hoxb7 and hoxb9 expression beyond the CNS is confined to the posterior paraxial mesoderm and tailbud at the 20s stage (Fig. 3C,M). Considering the hoxa cluster genes: hoxa7 has an anterior expression limit in the CNS adjacent to the s2/3 boundary (Fig. 3A,B), this limit is shared by hoxa8 (Fig. 3E,F), hoxa10 has an anterior limit adjacent to the s10/11 boundary (Fig. 3Q,R; this is in approximate agreement with a previous description of hoxa10 expression which placed the anterior limit at s9; Sordino et al., 1996). Thus, similar to the situation for the hoxb7 and hoxb8 genes, hoxa7 and hoxa8 share identical anterior expression limits in the CNS (compare Fig. 3B and F). Considering the hoxc cluster genes: hoxc8 has an anterior expression limit in the CNS adjacent to somite 4 (Fig. 3I,J) and hoxc9 adjacent to the s6/7 boundary (Fig. 3O,P). The anterior limit for hoxc10 fades out towards the anterior rather than having a sharp limit but is approximately adjacent to the s14/15 boundary (Fig. 3U,V).

Two of the genes in groups 7-10, hoxa8 and hoxb10, do not have equivalents in the tetrapod hox clusters; nevertheless, the expression patterns of these genes follow the same basic pattern seen with the other zebrafish hox genes. In addition, the unclustered gene, hoxx9 has a rather similar expression profile to its paralogues, hoxb9 and hoxc9 (Fig. 3 K-P); the anterior limit of hoxx9 expression in the CNS is at the s4/5 boundary (Fig. 3L), intermediate between the limits for hoxb9 and hoxc9. Similar to the more anteriorly expressed genes, many of the hox genes in groups 9-10 are expressed in the tailbud and in a segmental fashion in the posterior mesoderm: hoxb7, hoxc8 and all of the group 9 and 10 genes analyzed show this pattern (Fig. 3). Analysis at the 15s stage and the 24 hour stage, has revealed that the segmentally arrayed posterior mesodermal expression domains correlate with somitogenesis. Expression is dynamic such that it is restricted to the last few somites formed or forming at any given stage, leading to maintenance of a constant distance between the posterior mesoderm expression and the tailbud expression domains. We have not considered the expression patterns of genes in paralogue groups 11 and 12, as the signals obtained with probes to these genes tended to be rather weak, and thus it was difficult to accurately assess expression limits (data not shown); however the expression patterns of several of the posterior genes have been previously investigated using longer length riboprobes (van der Hoeven et al., 1996; Sordino et al., 1996).

Taken together, the expression data at the 20s stage show that there is spatial colinearity of expression in the CNS for the zebrafish hox genes i.e. the more 3′ genes in the clusters have more anterior expression limits in the developing embryo. However, our detailed expression analysis has revealed that, rather than necessarily stepping posteriorly along the axis, the expression limits for members of an individual cluster may be in equivalent positions in some cases. Thus, hoxa7 and hoxa8 share equivalent anterior limits and hoxb7, hoxb8 and hoxb9 share equivalent anterior limits. As mentioned above, although each cluster shows colinearity, the absolute anterior limits are not maintained for different members of a given paralogue group. This becomes more pronounced for the more 5′, or posterior, genes. Thus for the paralogue group 5 genes the anterior expression limits span a 1 somite range, whereas for the paralogue group 10 genes, the anterior expression limits span a 7 somite range.

We were interested to assess the anterior expression limits of the zebrafish hox genes in the paraxial mesoderm. The expression limits in this tissue have previously been carefully investigated and compared for two tetrapod vertebrates (mouse and chick; Burke et al., 1995), and we wished to extend this analysis to the zebrafish. Furthermore, we wished to compare the expression limits in the paraxial mesoderm with those in the CNS, and in particular, to discover whether the phenomenon of equivalent anterior expression limits for different members of the same cluster would be conserved in the mesoderm. To facilitate assessment of anterior expression limits in the paraxial mesoderm we carried out in situ hybridizations at the 10s stage. This stage was chosen because it precedes tailbud eversion, and thus embryos can easily be mounted on their ventral surface to facilitate a dorsal view. Moreover, at this stage the embryo is still undergoing convergent extension movements and in dorsal view the somites can readily be observed lateral to the CNS; at later stages the embryo will become narrower in the mediolateral axis but deeper in the dorsoventral axis, thus expression domains in the CNS and mesoderm are not so easily differentiated in whole-mounted preparations. In order to confirm that the expression limits had arrived at their most anterior point by the 10s stage, mesodermal expression limits were also observed at the 15s stage; where expression was at high enough levels to facilitate an accurate measurement, the limits were generally found to be maintained from the 10s stage.

Expression of hox genes from paralogue groups 6-10, in embryos at the 10s stage, is shown in Fig. 4; once again we made use of the additional molecular markers, krox-20 and the F59 anti-myosin antibody to help assign precise expression limits. At the 10s stage myosin expression is confined to the medial part of each somite, although expression of the hox genes generally extends throughout the mediolateral extent of each somite as observed by Nomarski optics. Although in most cases the anterior expression limit was clear and robust, for some genes (e.g. hoxa7) the expression was at a relatively low level leading to a poor signal to noise ratio; in these cases many embryos were compared to assist in obtaining the most accurate assessment possible of the anterior expression limit. The members of paralogue group 4 genes are expressed at low to negligible levels in the paraxial mesoderm; for hoxc4 and hoxd4 this expression is very close to the limits of detection, for hoxb4 and hoxx4 the expression is at detectable levels and extends as far anterior as somite 1 (data not shown). Similarly, for hoxa5 and hoxc5, paraxial mesoderm expression is close to being undetectable, hoxb5 expression is at higher levels in the paraxial mesoderm and extends as far anterior as somite 1 (data not shown). For the hoxb cluster genes, there is spatial colinearity of expression in the paraxial mesoderm; hoxb6 has an anterior limit at s4 (Fig. 4B) hoxb7 at s6 (Fig. 4E), hoxb8 at s6 (Fig. 4G), hoxb9 at s7 (Fig. 4J) and hoxb10 at s9 (Fig. 4M). Thus, as in the CNS, the anterior expression limits for hoxb7 and hoxb8 are shared, but unlike the situation in the CNS, hoxb9 does not share this limit. It should be noted that for hoxb7 and hoxb9 (Fig. 4E,J), the most anterior expression in the paraxial mesoderm is confined to the medial aspect of the somite, whereas more posteriorly, expression spreads throughout the mediolateral extent of the somites. For the hoxa cluster genes, hoxa7 is expressed at low levels in the paraxial mesoderm (Fig. 4D), with an anterior limit of expression at approximately s6, hoxa8 is expressed at high levels in the paraxial mesoderm with a very distinct anterior expression limit at s6 (Fig. 4F). Thus, once again as for the CNS, hoxa7 and hoxa8 share an anterior expression limit in the paraxial mesoderm. The hoxa10 gene shows diffuse expression in the tailbud (Fig. 4L), with expression confined to the region posterior to s10. The hoxc cluster genes also show spatial colinearity in the paraxial mesoderm. At the 10s stage hoxc6 has an anterior expression limit at s5 (Fig. 4C), hoxc8 at s7 (Fig. 4H), and hoxc9 at s8 (Fig. 4K); similar to hoxa10, the anterior limit of hoxc10 expression lies posterior to the last somite at the 10s stage (Fig. 4N), and, consistent with this observation, lies at s13 by the 15s stage (data not shown).

Once again we were interested to find out whether the hoxx and y genes would show expression similar to that of their clustered paralogues. The hoxy6 gene has an anterior expression limit at s5 (Fig. 4A), as does the hoxc6 gene (Fig. 4C). Similarly, the hoxx9 gene has an anterior limit at s8 (Fig. 4I), equivalent to the anterior limit of hoxc9 (Fig. 4K) and just one somite posterior to that of hoxb9 (Fig. 4J). Thus, the paraxial mesoderm expression of the hoxx and y genes is again consistent with the idea that they are bona fide members of their specific paralogue groups.

In general the hox gene expression in the paraxial mesoderm appears confined to the most anterior part of each somite, this is especially obvious for hoxy6, hoxb6, hoxc6, hoxb7 and hoxb9 (Fig. 3A-C,E,J). There is also expression beyond the paraxial mesoderm at the 10s stage, for example in the CNS, although this expression is at significantly lower levels than at the 20s stage. CNS expression limits are particularly clear for hoxa7, hoxa8 and hoxb9. For hoxa7 and hoxa8 these CNS limits are similar to those at the 20s stage, namely adjacent to the s2/3 boundary (Fig. 4D,F). However, for hoxb9 the anterior limit in the CNS is slightly more posterior than at the 20s stage, at about the s4/5 boundary (Fig. 4J). This suggests that hoxb9 has not yet reached its anterior expression limit at the 10s stage; consistent with this idea, by the 15s stage the expression limit in the CNS has shifted anteriorly to reach the mature position (data not shown). By the 15s stage, hoxb9 expression in the paraxial mesoderm has already begun to down-regulate, thus we are unable to judge whether the paraxial mesoderm expression limit has also shifted anteriorly. An additional expression domain, in a stripe lateral to the paraxial mesoderm is seen for hoxb6 and hoxb8 (Fig. 4B,G), this domain may correlate with the primordium of the future pronephric ducts, as each of these genes is also expressed in the region of the ducts at the 20s stage.

We have isolated a total of 32 zebrafish hox gene cDNAs from paralogue groups 1-12, primarily by 3′RACE PCR (Prince et al., 1998; this study). We have analyzed the linkage relationships of these genes by allocating them to specific somatic cell hybrids between mouse and zebrafish (Ekker et al., 1996), and have allocated the genes to specific paralogue groups based upon the presence of diagnostic amino acids (Sharkey et al., 1997). By combining these information sets, we have been able to construct the organization of the zebrafish hox genes (shown in Fig. 1), and to reveal the unexpected finding that the zebrafish has additional hox genes both within and beyond the expected 4 clusters, A-D. Thus, from within the clusters, we have isolated a hoxa8 gene and a hoxb10 gene; no equivalent genes have been reported for the tetrapod vertebrates. It should be noted, however, that hoxb13 was only recently isolated from the mouse (Zeltser et al., 1996), and found to lie at a considerable distance from the hoxb9 gene (70 kb); as searches for more 5′ members of the hoxb cluster have previously concentrated on locations relatively close to hoxb9 it remains a possibility that an intermediate hoxb10 gene remains to be found for mouse and other tetrapods.

The modern insect and vertebrate Hox cluster genes are postulated to have derived from a single Hox cluster comprised of five (Schubert et al., 1993) or six (Garcia-Fernandez and Holland, 1994; Fig. 1) genes in a simple, perhaps worm-like, ancestral organism. The Hox gene organization of amphioxus, a cephalochordate, sister group to the vertebrates, supports the idea of lateral duplications within the ancestral cluster. Amphioxus has a single Hox cluster comprising at least 10 genes, corresponding to members of paralogue groups 1-10 (Garcia-Fernandez and Holland, 1994). Two rounds of cluster duplications, followed by secondary losses of genes with no vital or unique role, could have led to the production of the modern day 4 cluster organization of the tetrapod vertebrates. The existence of additional genes within the 4 clusters of the teleost zebrafish is consistent with this scheme, implying that a different set of secondary losses have occurred in the teleost lineage. However, it should be noted that the different set of secondary gene losses may be due to different selective pressures, resulting from the possible existence in zebrafish of clustered hox genes beyond the 4 canonical clusters.

We have isolated 3 genes that are not closely linked to any of the 4 clusters A-D. We have termed these the hoxx and y genes and named them according to their apparent paralogue groups: hoxx4, hoxy6 and hoxx9. The hoxx4 and hoxx9 genes are present in the same set of hybrids; this linkage suggests the existence of at least one additional hox cluster in the zebrafish, although it remains to be ascertained whether further additional hox genes lie on this or other linkage groups. Both hoxx4 and hoxx9 show strong sequence similarity to Hoxa cluster genes, suggesting the possibility that the hoxa cluster has been duplicated during teleost evolution. The possibility of such chromosomal duplications during zebrafish evolution has already been suggested by the isolation of additional members of other gene families. For example, the zebrafish has two additional Dlx genes in comparison to mammals (Stock et al., 1996), possibly as a result of an independent duplication in the teleost lineage. Interestingly, the Dlx genes are themselves linked to the mammalian Hox clusters, and it will therefore be informative to determine the linkage relationships between the zebrafish hoxx or y genes, and Dlx genes. Similarly, there are additional members of the Msx and En gene families in the zebrafish (Akimenko et al., 1995; Ekker et al., 1992), and a large number of metabolic enzymes are encoded by more gene family members in teleosts than in mammals (Morizot, 1990).

To date, hox gene organization has been investigated in depth for only three different teleost species: the zebrafish (this study; van der Hoeven et al., 1996, Sordino et al., 1996; Misof et al., 1996), Fundulus (Misof and Wagner, 1996), and Fugu (Aparicio et al., 1997). Misof and Wagner carried out a PCR survey of the Fundulus hox genes, and did not look directly at linkage between these genes. However, their study revealed an additional gene to the tetrapod complement in paralogue group 1, suggesting that the precise numbers of hox genes in the teleosts may be more variable than in the tetrapod vertebrates. Aparicio and colleagues (1997) analyzed genomic DNA from Fugu rubripes, to reveal distinct changes with respect to the organization of both tetrapod and zebrafish Hox clusters, including several gene losses. The following genes are absent from the Fugu hox clusters, relative to the tetrapod vertebrates: hoxa6, hoxa7, hoxb7, hoxb13, hoxd1, hoxd3, hoxd4, hoxd8 and hoxd12. Thus, unlike the situation in zebrafish, there are no group 7 paralogue genes present in Fugu, and the hoxd cluster is radically diminished. Furthermore, an additional paralogue group 2 gene lies on the Fugu hoxd cluster, and pseudogenes corresponding to hoxc1 and hoxc3 were also found. There is, however, no evidence for additional hox clusters in Fugu. Fugu is a highly derived example of a teleost, belonging to the order Tetradontiformes, which in general lack both ribs and pelvic bones. As mammalian studies have implicated Hox genes in determining skeletal morphology (reviewed by Krumlauf, 1994), many of the Fugu specific changes in hox gene complement may reflect the specialized anatomy of the species. It will be interesting in the future to compare gene organization in a range of different teleosts. Such studies will reveal whether the existence of additional hox genes beyond the expected 4 clusters is specific to zebrafish, and will also facilitate correlation between changes in hox gene complement and differing anatomical features.

We have analyzed the expression of the zebrafish hox genes during early development. In a previous study we analyzed expression of 7 ‘hindbrain’ genes, from paralogue groups 1, 2, 3 and a subset of the group 4 genes (Prince et al., 1998). We have found that the expression patterns of these genes within the developing hindbrain share many similarities with those of the tetrapod vertebrates, although there are a few significant differences in precise timing and location of expression. In this study, we have particularly concentrated on assigning anterior expression limits to the individual ‘trunk’ genes, in both the CNS and the paraxial mesoderm, thus allowing us to investigate whether spatial colinearity of expression is conserved in a teleost vertebrate and to compare the expression limits with those from other species (Burke et al., 1995). Spatial colinearity refers to the phenomenon whereby genes lying 3′ within a hox cluster have more anterior expression limits than those lying 5′. Previous studies have demonstrated colinear expression of the most 5′ members of the hoxa and hoxd clusters (Sordino et al., 1996; van der Hoeven et al., 1996). Our expression analysis of additional zebrafish hox genes has shown that spatial colinearity is indeed conserved, however, there are several instances where expression limits are equivalent for neighbouring genes along an individual cluster, rather than stepping posterior along the body axis. In addition, the anterior expression limits of many of the zebrafish hox genes are generally condensed towards the anterior aspect of the embryo in comparison to those of the tetrapod vertebrates (summarized in Fig. 5).

Within any one paralogue group, colinearity within the CNS can be observed (Figs 2, 3; summarized in Fig. 5A). However, hoxb7, hoxb8 and hoxb9 share very similar CNS expression limits, approximately adjacent to the somite 3/4 boundary (Fig. 3). Similarly, hoxa7 and hoxa8 also share similar anterior expression limits, in this instance adjacent to the somite 2/3 boundary (Fig. 3). The hoxa8 and hoxb10 genes have no equivalents reported from the tetrapod vertebrates. Nevertheless, these genes show overall expression patterns and limits consistent with their paralogy assignments within the hox clusters. The three hox genes which lie beyond the 4 hox clusters, A-D, (hoxx4, hoxy6 and hoxx9) are also expressed as might be predicted for members of their assigned paralogue groups. Thus, it seems likely that these genes are continuing to play the same type of functional roles as their related paralogues within the 4 previously reported tetrapod clusters. The expression limits of zebrafish hox genes within the paraxial mesoderm were ascertained at an earlier stage (10 somites); shortly after this stage expression of hox genes in the paraxial mesoderm is rapidly down-regulated. This early phase of zebrafish hox gene expression may reflect the time at which AP identity of the mesoderm is specified; for example, Kieny and colleagues (1972) used transplantation techniques to show that the skeletal component of unsegmented chick mesoderm is already specified to take on thoracic versus cervical characteristics, supporting the idea that Hox gene expression patterns confer regional identity during an early phase of expression. In the zebrafish trunk region, little regional diversity has been described, however the first few somites do show some distinct features (described in more detail below), which may require specific patterning. Similar to the situation in the CNS, the paraxial mesoderm expression limits show colinearity (Fig. 4; summarized Fig. 5B), although hoxb7 and hoxb8 share equivalent expression limits at somite level 6, as do hoxa7 and hoxa8. For each individual gene, the paraxial mesoderm expression limit is to the posterior of the CNS expression limit, as has been described for other species. Again, similar to the CNS, the genes with no tetrapod equivalents, hoxa8, hoxb10, and the hoxx and y genes behave as would be predicted for members of their paralogue groups. In both the CNS and the paraxial mesoderm, the genes from each individual paralogue group show increasing disparity in precise anterior expression limit as the 5′ ends of the clusters are approached. The anterior expression limits of the hoxb cluster genes cover the shortest AP extent of the trunk, and those of the hoxc cluster the longest extent. However, even for the hoxc cluster, the expression limits are confined well within the ‘trunk’ region as opposed to the developing tail. Expression of the most posterior hoxc cluster member, the hoxc11 gene, similarly reaches anterior to the trunk/tail transition point at somite 17 (data not shown). Indeed, it has been suggested that the posterior extent of the hox system is in some manner ‘fixed’ at the level of the anus (van der Hoeven et al., 1996), which in zebrafish lies relatively anterior, at about somite 17. Such a fixation of the hox system would help to explain our observations that the anterior limits of hox expression are compacted close to one another, in comparison to those of amniote species where the anus lies significantly more posterior (somite 25-30). The zebrafish hox genes in paralogue groups 7 and 8, which share equivalent anterior expression limits, are believed to have derived from a common ancestral gene (Fig. 1). Furthermore, the anterior expression limits of the group 6 genes, which are believed to have derived from the same ancestor, have dispersed only a short distance along the axis. This lack of dispersal of the group 6, 7 and 8 anterior expression limits may reflect an ancestral expression condition. It will be of interest to examine expression of the amphioxus paralogue group 6, 7 and 8 genes as amphioxus has been postulated to represent an extant example of an intermediate stage in Hox cluster evolution (i.e. after lateral duplications to produce an extended cluster, but before whole cluster duplications to produce the 4 clusters of the vertebrates).

The evolution of differing anterior expression limits for individual Hox genes requires changes in the regulatory mechanisms responsible for correct spatial expression. These changes may lie in the regulatory sequences of the Hox genes themselves, or alternatively may lie upstream, for example, in the localization of activators and repressors of Hox gene transcription. In several cases the precise elements responsible for conferring correct spatial expression on individual Hox genes have been unraveled (reviewed by Lufkin, 1996). These studies have been performed in the mouse to take advantage of transgenic technology; however, comparison of regulatory elements across species has already proved to be a useful approach towards dissecting the evolution of Hox gene regulation (Beckers et al., 1996).

The zebrafish hox genes are also expressed in a range of tissues beyond the anterior paraxial mesoderm and CNS. For example, several of the genes are expressed in a segmental manner in the posterior mesoderm. This dynamic expression may play a role in somitogenesis as it is confined to the most recently formed (or just forming) somites, moving posteriorly in concert with somitogenesis. In cases where such expression is observed it is always found in parallel with expression in the developing tailbud. Several of the genes are also expressed in defined regions of the developing endoderm, perhaps reflecting a function in patterning the gut structures. The hoxb6 and hoxb8 genes are also expressed in the location of the forming pronephric ducts, expression is present at the 10s stage when lateral ‘stripes’ can be observed (Figs 2, 3). Finally, several of the genes are expressed in discrete domains immediately lateral to the neural tube, the AP position of these domains extending anterior to the CNS expression limit. We cannot unequivocally identify these structures, but their position is consistent with a correspondence to forming ganglia, for example hoxa5 is expressed immediately posterior to the otic vesicle (Fig. 2I,J), in the position where the primordium of the lateral line ganglia would be predicted to lie (Metcalfe, 1985).

The two major sites of Hox gene expression are the CNS and the paraxial mesoderm, but the patterning function of the Hox genes in the trunk region has primarily been investigated with respect to the paraxial mesoderm-derived vertebrae, due to the ease of recognizing morphological changes in vertebral structure. Axial organization of the tetrapod vertebrae is complex, one of the most obvious manifestations of this being the existence of five basic classes of vertebrae – cervical, thoracic, lumbar, sacral and coccygeal. The precise number of vertebrae within each class differs between species, providing clear structural landmarks along the developing axis. These and other morphological landmarks, such as the limb buds, were used by Burke and colleagues (1995) to test the correlation between Hox gene expression and regional identity. Their comparison of expression patterns in mouse and avians revealed that Hox gene expression domains correlate with specific axial structures, even when these structures are at very different AP levels in different species, consistent with the idea that the Hox code directly imparts regional identity along the axis of the developing embryo.

Unlike the multiple anatomical subdivisions of the tetrapod trunk, only two basic classes of teleost vertebrae are recognized: trunk vertebrae consisting of a centrum articulating with a neural arch dorsally and ribs ventrally, and tail vertebrae, in which the ribs are replaced by ventral hemal arches (Kent, 1992; van Eeden et al., 1996). However, the first few somites of the zebrafish do show some distinct features. For example: (1) the first six somites each form in only 20 minutes, whereas later somites take 30 minutes to form (Kimmel et al., 1995); (2) the first seven somites are the only ones affected by the mutations deadly seven and after eight (van Eeden et al., 1996); (3) the pectoral fin grows out opposite somite 3 (Kimmel et al., 1995); and (4) the pectoral fin is innervated by spinal nerves 3, 4 and 5, which derive from adjacent to somites 3, 4 and 5 (Myers, 1985). Indeed, by analogy to mouse, chick, goose and Xenopus, it has previously been suggested that the hoxc6 anterior limit (which we observe at somite 5, consistent with previous reports; Molven et al., 1990), may correlate with the posterior limit of innervation of the pectoral fin bud (Burke et al., 1995). Furthermore, again consistent with the findings of Burke and colleagues in amniote species, we find that the region between the anterior expression limits of paralogue group 5 and group 6 genes correlates with the region of fin bud outgrowth. Thus, in a similar manner, the hox genes with anterior limits at somite 6 (hoxa7, hoxb7, hoxa8, hoxb8), may also play some role in differentiating between the anterior and posterior subsets of trunk somites, although in mouse and chick the paralogue group 7 and 8 genes have anterior limits that lie within the thoracic segments, not correlating with any obvious transition point.

The anterior expression limits of tetrapod hox genes in paralogue group 9 mark the thoracic/lumbar transition point; no equivalent transition point exists in the teleosts. We find that the zebrafish paralogue group 9 genes have anterior expression limits very close to those of the group 8 genes. This lack of dispersal of anterior expression limits along the AP axis may be a direct reflection of the general lack of diversity of axial structures along the zebrafish AP axis. Thus, in the modern tetrapod vertebrates, Hox genes have not only been duplicated, but in addition, their expression domains have dispersed along the AP axis to pattern the trunk and tail. One would predict that as organisms have become more complex, with increased regional diversity along the AP axis, the overall degree of complexity in the combinatorial Hox code would need to concomitantly increase. In teleost fishes, vertebrates with fewer regional differences along the AP axis, the hox code might be predicted to be simpler. Consistent with this idea, in comparison to the amniotes, we see fewer different anterior expression limits for the zebrafish hox genes, and less dispersal of these limits along the axis. However, contrary to this idea, the existence of additional hox genes in the zebrafish provides the means to encode a broader range of codes. Perhaps, these additional hox genes play important roles in patterning non-mesodermal structures, such as the nervous system or the gut. It remains to be seen whether the presence of additional hox genes is a common feature of the teleosts or a zebrafish-specific phenomenon.

We wish to thank Anders Molven for hoxb5 and hoxb6 genomic clones, Frank Stockdale for anti-F59 antibody, David Grunwald for the cDNA library, Gunter Wagner for sharing sequence information prior to publication and M. Featherstone for mouse Hoxd-4 cDNA. We are very grateful to Annie Burke, Laure Bally-Cuif, Marty Cohn, Mike Coates, Anthony Graham, Andreas Fritz, Tom Vogt and Gunter Wagner for helpful discussions and advice. We would especially like to thank Dr Andreas Fritz for generously sharing unpublished data. We would also like to thank Tracy Roskoph for expert fish care and advice. V. E. P. has been supported by long term fellowships from EMBO and HFSPO. This work was supported by the Canadian Genome Analysis and Technology Program to M. E. and by a donation from the Rathmann Family Foundation to the Molecular Biology Department at Princeton University, by a Basil O’Connor Starter Scholar Research Award from the March of Dimes and by NIH grant RO1 HD34499 to R. K. H. who is a Rita Allen Foundation Scholar.