Patterning the ascidian nervous system: structure, expression and transgenic analysis of the CiHox3 gene

HOM/HOX genes are involved in the establishment of the animal body plan by specifying the positional identity of different structures along the anteroposterior axis (reviewed in McGinnis and Krumlauf, 1992; Krumlauf, 1994). It is also believed that they contribute to the morphological diversity throughout the animal kingdom. During evolution, Hox genes have conserved both their organization in clusters and the ordered correlation between the position of the genes in the cluster and their expression pattern along the body axis (spatial colinearity) (Lewis, 1978; Graham et al., 1989; Duboule and Dollè, 1989). From an evolutionary perspective, the development of a more complex body organization seems to correlate with the formation and amplification of Hox clusters, probably achieved through successive tandem duplications of an ancestral homeobox-containing gene and subsequent cluster duplication (Field et al., 1988; Schugart et al., 1988; Kappen et al., 1989).

In agreement with this hypothesis, a single cluster with a variable number of Hox genes has been found in most metazoans, such as the Arthropods and Nematodes (reviewed by Holland, 1992; Burglin and Ruvkun, 1993). In chordates, only one cluster is present in the genome of the Cephalochordate Amphioxus (Holland et al., 1994) and it has been suggested that the Urochordate Ciona intestinalis also has a single cluster (Di Gregorio et al., 1995). Four apparently identical Hox clusters are present in all mammals so far analyzed (Boncinelli et al., 1988; McGinnis and Krumlauf, 1992; Duboule, 1994); however, variation in the number and organization of Hox clusters is observed in some lower vertebrates (Aparicio et al., 1997; Amores et al., 1998).

Genetic analysis of vertebrate Hox genes, involving both loss- and gain-of-function mutations, has been carried out primarily in the mouse through the use of gene targeting technology. Together, this large body of work has revealed important functional roles for the overlapping and ordered expression patterns of Hox genes in many different tissues of the embryo, such as the CNS, axial skeleton, limbs and gut (reviewed by McGinnis and Krumlauf, 1992; Maconochie et al., 1996; Duboule, 1993). There are distinct individual roles for most Hox genes, but there is also evidence for synergistic interactions and functional redundancy between members, which are most likely due to overlapping expression patterns and structural conservation between paralogous genes. Hence, Hox genes have multiple roles in patterning diverse axial structures.

Organisms belonging to the Chordate phylum share some characteristics, such as the presence of an axial notochord flanked by muscles, a dorsal hollow nervous system and a ventral endodermal strand. Given the key role of Hox genes in regulation of the body plan, study of their expression patterns and regulatory regions in different Chordates may aid in the identification of both homologous structures and novel structures, thus shedding light on the evolution of morphological diversity. Therefore it is important to understand the regulation of restricted Hox expression in different species, as one suggested means of coupling this gene family to a range of common or unique morphogenic processes could be through the conservation, modification and/or acquisition of short regulatory elements (Gellon and McGinnis, 1998).

The expression and regulation of the Hox genes has been studied in detail in the CNS (reviewed in Keynes and Krumlauf, 1994; Rubenstein and Puelles, 1994; Maconochie et al., 1996; Lumsden and Krumlauf, 1996). In the vertebrate CNS, Hox genes belonging to paralogy groups 1 to 4 are expressed with partially overlapping domains whose anterior boundaries correlate with the subdivision of the hindbrain into segmental units or rhombomeres (Hunt et al., 1991; Keynes and Krumlauf, 1994). Mutational analysis in the mouse has shown that these rhombomere-restricted domains of Hox expression underlie functional roles in multiple steps of segmental patterning, as illustrated extensively for the group 1 genes (Carpenter et al., 1993; Mark et al., 1993; Studer et al., 1996, 1998). The remaining cluster members (paralogy groups 5–13) are expressed along the entire spinal cord in progressively more posterior domains that are also functionally important (Tiret et al., 1998).

The use of a simpler model system than the higher vertebrates might contribute to the study of Hox genes and of their role in the evolution of chordates. Ascidians have a basic body plan and are considered prototypical chordates. Despite their simple organization, recent evidences have pointed out a certain degree of conservation in the genetic pathways involved in the specification of different body structures between ascidians and vertebrates (Di Gregorio and Levine, 1998).

During ascidian embryogenesis, the nervous system develops by the formation of a neural plate, which invaginates and seals dorsally leading to a hollow nerve cord in a process reminiscent of vertebrate neurulation. At the larval stage, the ascidian nervous system comprises an anterior vesicle in the trunk, containing the pigmented sensory organs and thus called sensory vesicle, followed, more posteriorly, by a visceral ganglion and a caudal nerve cord. This latter structure is devoid of neuronal cell bodies and lies above the notochord (Nicol and Meinertzhagen, 1991).

Recent experiments have demonstrated the usefulness of C. intestinalis as a model system for studying gene regulation (Corbo et al., 1997a,b). In the present report, we have focused our attention on a C. intestinalis Hox gene belonging to the anterior group, sharing the highest degree of homology to the class 3 Hox genes of vertebrates and thus referred to as CiHox3. The analysis of CiHox3 spatial and temporal expression patterns during C. intestinalis embryogenesis revealed that its mRNA is confined to the larval CNS and in particular to a well-defined region of the visceral ganglion. In an effort to understand the mechanisms involved in the regulation of CiHox3 expression during development, we have used electroporation (Corbo et al., 1997b) to introduce reporter constructs, containing different fragments from the 5′-flanking region of the CiHox3 gene fused to the lacZ cistron, into fertilized Ciona eggs. With this approach, we identified an enhancer element responsible for the specific expression of the reporter gene in the nervous system.

Furthermore, the CiHox3 promoter was tested in transgenic mouse embryos and mouse Hox3 regulatory elements were electroporated in Ciona in order to address whether regulatory elements of Hox3 genes have been conserved during chordate evolution.

Adult C. intestinalis were collected in the Bay of Naples by the fishing service of the Stazione Zoologica. Gametes were used for in vitro fertilization and embryos were raised in filtered sea water at 16–18°C. Samples at appropriate stages of development were collected and used for whole extract protein or RNA extraction, or fixed for whole-mount in situ hybridization.

A cDNA library made from embryos at larva stage (Gionti et al., 1998) was screened using as a probe a genomic fragment, named CiHbox1, previously characterized by Di Gregorio et al. (1995) and containing the homeobox sequence of CiHox3. A positive clone insert was sequenced on both strands by the dideoxynucleotide termination procedure (Sambrook et al., 1989).

A C. intestinalis cosmid library, constructed by Reference Library Database (RLDB, MPI for Molecular Genetic, Berlin-Dahlem, Germany; Burgtorf et al., 1998), was screened at high stringency using as a probe the [³²P]oligonucleotide 5′-TCTATCGGCAGCCATA-AGAGTC-3′, complementary to the most 5′ coding region of the cDNA sequence. Three positive clones: MPMGc119L0224, MPMGc119B058 and MPMGc119D1338 were amplified and DNA was purified with QIAGEN kit (Qiagen Inc., Chatsworth CA, USA). The genomic inserts were subjected to digestion with the restriction endonucleases EcoRI or XbaI followed by Southern blot hybridization with the oligonucleotide described above. An XbaI- and an EcoRI-positive fragments of about 4 kb and 1 kb in length, respectively, derived from the cosmidic clone MPMGc119B058 and containing the 5′ coding region of CiHox3, have been subcloned in the pBlueScript II KS vector and sequenced.

The 4.2 kb promoter region was obtained by assembling three partially overlapping EcoRI or XbaI genomic fragments obtained from the successive hybridization of Southern blot of the cosmidic clone MPMGc119B058 with oligonucleotides at the 5′ end of each fragment.

Introns position and length have been determined by PCR using the same cosmid clone as template and useful primers complementary to the cDNA sequence.

Total RNA from various stage embryos was prepared according to Chomczynski and Sacchi (1987). Poly(A)⁺ RNA was purified by oligo(dT)-cellulose chromatography (Sambrook et al., 1989). Northern blot hybridization was carried out as described by Gionti et al. (1998) using 15 μg of poly(A)⁺ RNAs per lane and a ³²P-labeled CiHox3 cDNA insert as a probe. In addition, a Ci-CaM cDNA (Di Gregorio et al., 1998) probe was used as a control.

Two Dig-11-UTP-labeled RNA probes were used for whole-mount in situ hybridization experiments: the CiHox3 cDNA insert and a DNA fragment, amplified by PCR, corresponding to the 5′ coding region of CiHox3 (extending from position −70 to +500). Conditions for in vitro transcription were as described by the manufacturers in the Digoxigenin RNA Labeling Kit (Boehringer-Mannheim). Whole-mount in situ hybridizations on Ciona embryos and larvae were carried out as described in Caracciolo et al. (1997). Semithin sections were performed as described by Di Gregorio et al. (1998).

The transcriptional initiation sites were determined by RNase protection assay basically according to the RNase protection kit instructions (Boehringer Mannheim).

A 353 bp fragment (spanning −179 bp to +174 bp), amplified by PCR, was cloned in the EcoRV site of the pBluescript II KS vector. The plasmid was digested with XbaI and used as template for in vitro transcription of [³²P]UTP-labeled antisense riboprobe. 5 μg of poly(A)⁺ RNA from embryos at larva stage were hybridized overnight at 45°C with the probe. After RNase A and T1 digestion, the protected fragments were run on 8% denaturing polyacrylamide gel along with a reference sequence for size determination.

The basic electroporation vector was pBlueScript II KS containing the lacZ and SV40 polyadenylation sequences (pBS+LacZ). Genomic fragments of 0.2, 0.45, 1.0 and 1.5 kb were amplified by PCR from the cosmid clone MPMGc119B058. The amplified fragments were ligated with the pBS+LacZ vector in the 5′ to 3′ orientation. An XbaI-KpnI CiHox3 genomic fragment, obtained from the cosmid clone MPMGc119B058, was subcloned in the 0.45 construct, XbaI and KpnI digested, to obtain the construct 3.0.

The 4.2 construct was obtained by cloning an XbaI (end-filled)-XbaI genomic fragment, from the same cosmid clone, in the KpnI (end-filled)-XbaI-digested 0.45 construct.

To generate the internal deletion mutants Δ4-3 and Δ3-25 the regions from −3 kb to −0.5 and from −2.5 kb to −0.5 kb of the 4.2 and 3.0 constructs, respectively, were removed by KpnI and SnaBI or only SnaBI digestions and the resulting linear DNA fragments were self-ligated. The deletion mutant Δ3-2 was generated by exonuclease III digestion of the region from −2 kb to −0.45 kb of the 3.0 construct according to manufacturer’s instructions (Promega). All the other constructs, P-1 to P-6, were obtained by cloning the corresponding genomic fragments, amplified by PCR, in the end-filled HindIII site of the 0.45 construct.

For transgenic mice, a 2.3 kb KpnI-HindIII fragment (construct 1, Fig. 7A) or a 500 bp PCR product (construct P1, Fig. 7A) from the Ciona Hox3 genomic region were cloned downstream of a lacZ reporter expression construct (pBGZ40) containing the human β-globine promoter (Yee and Rigby, 1993). DNA purified from vector sequences was microinjected in mouse oocytes, and lacZ expression in embryos detected as described by Whiting et al. (1991).

The mouse Hoxb3 and Hoxa3 constructs used for electroporation in Ciona embryos were the constructs N. 5 and 6 by Manzanares et al. (1997) and #1.4, #3.3 (Manzanares et al., 1999).

Electroporation of fertilized and dechorionated C. intestinalis eggs was as described in Corbo et al. (1997b) with a few modifications: capacitance setting was between 700 and 850 μF so that the pulse range was 15–18 mseconds and the final volume in the 0.4 cm cuvettes was 700 μl.

Embryos were allowed to develop at 16°C, in fresh sea water in 0.9% agarose-coated dishes, until the required stage, fixed in 1% glutaraldehyde in sea water for 30 minutes at room temperature, washed twice in PBS 1× and stained at 30°C in a solution containing 3 mM K₃Fe(CN)₆, 3 mM K₄Fe(CN)₆, 1 mM MgCl₂, 0.1% Tween20 and 200 μg/ml X-gal in PBS.

Band-shift assay was carried out essentially as described by Yuh et al. (1994) with 8 μg of protein extracts from embryos at larva stage, 50 ng of poly(dIdC) and 4 fmol of ³²P-labeled AL2 oligonucleotide. The sequence of the random oligonucleotide was: 5′-CTGCTTTGA-TGGATGGAGCTG-3′. Protein extracts were prepared according to Tomlinson et al. (1990).

CiHox3 was isolated from a cDNA library prepared with poly(A)⁺ mRNA from C. intestinalis larvae using a genomic fragment, containing the homeobox sequence (previously named CiHbox1 in Di Gregorio et al., 1995) as a probe.

Sequence analysis of the isolated cDNA clone, 1994 bp in length, showed that the translation start codon was missing. To obtain the remaining 5′ coding sequence, a cosmid DNA library, prepared by the RLDB (Burgtorf et al., 1998), was screened with an oligonucleotide complementary to the anteriormost region of the cDNA. Three positive clones were isolated and analyzed, and an EcoRI fragment from the cosmid clone MPMGc119B058, partially overlapping with the cDNA, was subcloned and sequenced. This fragment was found to contain the remaining 5′ coding region, preceded by a series of stop codons, and part of the promoter sequence of CiHox3. The analysis of the genomic organization of CiHox3 has been performed by PCR amplification on the cosmid clones using oligonucleotides complementary to the CiHox3-coding region. The complete genomic structure of CiHox3 thus obtained is schematically reported in Fig. 1A. It includes two introns of 2700 and 2173 bp respectively, and three exons including the 5′ (90-167 bp) and the 3′ UTR (280 bp) regions, and a coding region of 2199 bp, encoding for a putative protein of 733 amino acids.

It is interesting to note that the first intron, positioned between the sequences encoding the hexapeptide and the homeodomain, is conserved among all Hox3 genes of chordates, while the presence of an intron in the homeobox sequence seems to be a peculiar feature of C. intestinalis homeobox-containing genes (Di Gregorio et al., 1995).

The comparison of the deduced amino acid sequence of CiHox3 with those deposited in the GenBank and EMBL databases revealed the highest degree of homology with the homeodomains of the paralogy group 3 HOX proteins. A comparison of the homeodomain, the hexapeptide and the interposed sequence of CiHox3, AmphiHox3, the HOX3 proteins of mouse and the Drosophila Proboscipedia is shown in Fig. 1B. The highest percentage of identity was found in the homeodomain of AmphiHox3 (81%) and the lowest in the homeodomain of Proboscipedia (72%). Interestingly, the degree of homology (78%) is the same among the three HOX group 3 proteins of mammals (HOXa3, HOXb3 and HOXd3). A lower degree of conservation can be identified in the region interposed between the homeodomain and the hexapeptide where 10–12 identical amino acids are conserved. In this domain, CiHox3 does not include the stretch of glycine residues, characteristic only of mouse and human HOXb3, suggested to have a hinge function (Beachy et al., 1985; Sham et al., 1992). Therefore this may represent characteristic sequence acquired only by the HOXb3 proteins later in evolution.

To establish the site of transcript initiation, RNase protection assay was carried out on poly(A)⁺ RNA from the larval-stage embryos using a [³²P]UTP-labeled riboprobe which extends from position −179 to +174 and includes both part of the 5′ coding region and of the adjacent promoter sequence. Three alternative transcription start sites were found (Fig. 1C): two giving a stronger signal corresponding to position −90 and −101, and a third one giving a weaker signal at position −167 from the translation start codon (Fig. 1A). The same result was obtained using different riboprobes or poly(A)⁺ RNAs from tailbud-stage embryos (data not shown).

The timing of CiHox3 expression during C. intestinalis development was determined by northern blot analysis. Using the cDNA clone as a probe, no signal was detected in stages prior to the early tailbud in which a single positive band of about 2.6 kb appears (Fig. 2A). The signal remained faint in the middle tailbud and gradually increased in the following stages of development, peaking in the swimming larva. The same northern, hybridized with the Ci-CaM cDNA (Di Gregorio et al., 1998), served as a control for RNA loading (Fig. 2B).

In situ hybridization assays were then carried out to determine the localization of CiHox3 message during Ciona embryogenesis. No signal could be detected even after long staining times in early tailbud embryos probably due to the low abundance of the message (not shown). CiHox3 expression became visible at larval stage and was restricted to the nervous system (Fig. 3A,B). In particular, the transcript was localized in the anteriormost region of the visceral ganglion showing well-defined both anterior and posterior limits. Transverse sections, cut at different levels of the hybridized larva, confirmed this conclusion. In particular, staining seemed to be confined in the lateral and most superficial region of the visceral ganglion (Fig. 3C) which, according to the studies by Nicol and Meinertzhagen (1991), contains the cellular bodies of ganglion cells. No signal was detected outside the CNS.

As indicated in Fig. 1A, the CiHox3 promoter region contains several putative TATA and CAAT boxes in the region adjacent to the three alternative transcription start sites. In order to identify the elements required for neural-specific expression of CiHox3, the 5′ genomic sequence of this gene was further examined using an electroporation method (Corbo et al., 1997b). Initially, we assayed the 4.2 kb genomic DNA fragment starting immediately upstream from CiHox3 translation start site and linked to a lacZ reporter gene (Fig. 4A). The electroporated Ciona embryos were allowed to develop until the stage of interest, they were then fixed and assayed for β-galactosidase activity by X-gal staining. As shown in Fig. 4C, this construct drove the expression of the reporter gene in the nervous system of the larvae. In particular, staining was visible in the visceral ganglion showing virtually the same localization as the endogenous transcript. However, this construct also showed ectopic expression in the anteriormost region of the nervous system, at the level of the sensory vesicle and around the otolith. In about 5% of the electroporated embryos, the trunk mesenchyme was also ectopically stained (not shown). The onset of expression was at early tailbud stage (Fig. 4B). In these embryos, staining became visible in the anterior region of the developing nervous system after 10 days of staining. Three internal deletion transgenes and a series of deletion constructs progressively truncated at the 5′ end were then prepared and their effects was analyzed (Fig. 4A). The same pattern of expression found with the 4.2 kb transgene was obtained only with the construct 3.0, while the progressively shorter constructs of 1.5, 1.0, 0.45 and the three deletion mutants Δ4-3, Δ3-25 and Δ3-2 were unable to give expression in the nervous system. These constructs, however, showed strong ectopic labeling in the trunk mesenchyme at the larval stage (Fig. 4E). The embryos electroporated with the 0.45 construct were also analyzed at earlier stages of development and after 2 weeks of staining mesenchymal cells of the neurula stage embryos were weakly labeled (not shown). During the subsequent stages of development, the signal became stronger and was clearly recognizable after one week of staining in the mesenchymal pockets of the early tailbud stage embryo (Fig. 4D). The smallest deletion construct 0.2 (shown in Fig. 4A), containing only the putative TATA and CAAT boxes was unable to drive the expression of the reporter gene (data not shown).

The analysis of the results obtained with the constructs shown in Fig. 4A suggested that the region from −2 kb to −1.5 kb (indicated as red box) of the promoter contained elements responsible for the restricted expression of the lacZ transgene in the larval nervous system. To confirm this hypothesis a construct (P1, shown in Fig. 5A) was prepared by fusing this region upstream from the 0.45 construct of Fig. 4A. The β-galactosidase activity in the embryos electroporated with this construct was assayed at the larval stage. The reporter gene was expressed both in the visceral ganglion and the sensory vesicle of the CNS (Fig. 5C) thus reproducing the result obtained with the 4.2 and 3.0 constructs of Fig. 4C. Also in this case about 5% of the embryos showed ectopic staining in the trunk mesenchyme (not shown). These results indicate that the sequence contains all the element(s) necessary to activate transcription in the CNS.

In an attempt to isolate the minimal sequence responsible for the neural-specific expression pattern, the P1 sequence was subdivided in a series of smaller and partially overlapping fragments obtained by PCR, which were subcloned upstream from the 0.45 construct as shown in Fig. 5A.

Embryos electroporated with the construct P2 presented only ectopic staining in the trunk mesenchyme of the larva. Constructs P3 and P4 were able to drive expression of lacZ in the larval nervous system, both in the visceral ganglion and ectopically in the sensory vesicle (not shown). The P4 sequence was further subdivided into two fragments and embryos electroporated with the construct corresponding to P5 did not show expression in the nervous system while P6 fragment reproduced the result obtained with P4 (Fig. 5D). Therefore, the element(s) necessary for the expression in the CNS are present in the 80 bp of the 5′ flanking region from position −1943 to −1864 (Fig. 6A).

In order to study the ability of the 80 bp sequence to bind nuclear proteins, different overlapping oligonucleotides of 20–24 bp each were used in gel shift assays on whole-cell extracts of Ciona larvae. Only the oligonucleotide AL2, indicated in Fig. 6A, was able to form a specific complex (Fig. 6B). In fact, a 200-fold excess of cold AL2 was able to decrease complex formation while an unrelated oligonucleotide had no effect. Two oligonucleotides, partially overlapping with AL2, named AL1 and AL3 (Fig. 6A), were tested for the ability to compete with AL2 for complex formation. As shown in Fig. 6B, only AL1 was able to decrease the binding, suggesting that the common region between the AL1 and AL2 oligonucleotides was involved in the binding. The analysis of the AL2 sequence with the MATINSPECTOR 2.2 and TRANSFAC databases did not reveal any likely recognition sequence for known transcription factors.

To date, only the CNS-specific regulatory elements of mouse and chicken Hoxb3 or mouse Hoxa3 (Manzanares et al., 1997, 1999) promoters have been studied and several kreisler-like elements have been found to be involved in their activation in rhombomeres (r) 5 and 6. Therefore, we decided to test the transcriptional activity of CiHox3 regulatory elements in mouse embryos. For this purpose, reporter constructs with lacZ for microinjection in mouse oocytes were prepared. We first tested a 2.3 KpnI-HindIII fragment (Fig. 7A), which includes the region involved in the neural-specific expression of CiHox3 in Ciona embryos. Transgenic embryos for this construct consistently gave reporter expression in a pattern reminiscent of mouse Hox regulatory elements. Expression was detected in the ventral portions of rhombomere (r) 4 and also in a domain with an anterior boundary at the r6/r7 boundary that extended posterior into the spinal cord (Fig. 7B,C). We then tested in mouse embryos the CiHox3 P1 promoter fragment that was responsible for specific lacZ expression in the Ciona nervous system (Fig. 7A). In contrast to the first case, this construct did not generate any specific pattern of expression in transgenic mouse embryos, but showed only ectopic expression presumably due to effects of the integration site (data not shown). Therefore, these results suggest that the CNS-specific regulatory elements from CiHox3 gene that direct reporter expression in that species are not the ones recognized by the mouse embryo transcriptional machinery. Hence, other elements must be present in Ciona genomic fragments that are able to mediate a segmental expression in the mouse hindbrain. These elements might correspond to auto-or cross-regulatory regions that are capable of responding to endogenous mouse Hox genes (see Discussion).

Conversely, when the mouse kreisler-dependent enhancer regions, involved in the expression of Hoxa3 and Hoxb3 in the hindbrain (schematized in Fig. 5B), were electroporated in Ciona embryos, no nervous-system-specific expression was observed. Only ectopic expression was detected, such as staining in mesenchymal cell using Hoxa3 and in the muscle of the tail using Hoxb3 (data not shown). Hence the mouse kreisler-like response elements, conserved in higher vertebrates, seem unable to regulate any part of a CiHox3-type expression pattern in Ciona. It is possible that other CNS regulatory regions recently identified in the mouse Hoxa3 locus (Manzanares et al., 1999) might function in Ciona. However, our results reveal that the regulation of Hox3 gene expression in the mouse and Ciona CNS does not appear to be mediated by the same highly conserved elements, as either of the enhancers shown to direct specific expression in one species fails to direct reporter expression in the other. This might indicate that different regulatory elements and components are utilized by these species and/or that similar components are involved but they have slightly diverged and are unable to function across such an evolutionary distance.

We have isolated a homeobox-containing gene called CiHox3 encoding for a protein with 81 and 78% of identity in its homeodomain and hexapeptide sequences with AmphiHox3 and HOX paralogy group 3 of vertebrates, respectively (Fig.1B). The temporal and spatial expression pattern of CiHox3 were analyzed by northern blot and whole-mount in situ hybridization. The earlier stage at which a clear signal could be detected in the northern blot was the early tailbud. CiHox3 mRNA localization was restricted to the larval nervous system and, in particular, to the visceral ganglion.

The larval nervous system in C. intestinalis can be subdivided into three regions, which have been attributed distinct functions (Nicol and Meinertzhagen, 1991): in the trunk, an anterior sensory vesicle, containing the two sensory organs, followed by the visceral ganglion and, in the tail, the spinal cord. The visceral ganglion has been considered the integrating center of the ascidian nervous system and its architecture has been studied in detail and was found to contain, in its outer region, neuronal cell bodies (ganglion cells), while in the inner part are present fibrous neuropil (Barnes, 1971). Transverse sections of hybridized larvae (Fig. 3C) showed that CiHox3 message was localized in the external area of the visceral ganglion suggesting it could be expressed in the neuronal cell bodies. The comparison of Hox3 expression territories between ascidians, cephalochordates and vertebrates might shed light on the evolution of these genes along this phylogenetic line and could permit the establishment of structural homologies among such different taxa. Studies on Amphioxus AmphiHox3 expression revealed that it is restricted to a specific domain of the dorsal nerve cord with a sharp boundary only in the anterior part, in particular, at the level of somite pairs 4 and 5 and it is also expressed in the posterior mesoderm (Holland et al., 1992). In vertebrates, the localization of paralogy group 3 genes transcripts is not confined to the nervous system, where they are expressed in the hindbrain with anterior limit at the level of rhombomere 5, but are also detectable in posterior and lateral mesoderm (Hunt et al., 1991). Therefore, our analysis on CiHox3 expression, which is restricted to the nervous system, demonstrates that the common feature of Chordates Hox3 genes is in the spatial patterning of the nervous system.

Fig. 8 shows a schematic representation of the territories of expression of CiHox3 and Cihox5, the Ciona Hox genes so far analyzed in the larva, compared with the neural expression of mouse Hox counterparts. A previous study on the Ciona homologue of the paralogy group 5 genes, Cihox5 (Gionti at al., 1998), revealed that its expression is limited to the larva neural tube.

This demonstrated that the spatial colinearity rule, i.e. the expression pattern of Hox genes follows their position in the cluster, is fulfilled in Ciona. As shown in the scheme (Fig. 8), CiHox5 is expressed in the most anterior region of the larval neural tube, therefore posteriorly to CiHox3 and without overlapping domains.

Moreover, their territory of expression has well-defined anterior and posterior limits, feature which, together with their restricted expression to the nervous system, seems to be unique to Ciona Hox genes. In cephalochordates and vertebrates, Hox genes are present in different mesodermal regions and their neural expression has only a defined anterior limit, while posteriorly extends and overlaps along the entire spinal cord.

Previous studies suggested that ascidians neural dorsoventral patterning (Corbo et al., 1997a; Glardon et al., 1997; Wada et al., 1997) and neural anteroposterior patterning are regulated by the same mechanisms described in vertebrate CNS. Wada et al. (1998) have proposed a subdivision of the ascidian nervous system based on the comparison of the territories of expression of genes involved in the patterning of anteroposterior structures of vertebrates and their ascidian counterparts.

Our results indicated that CiHox3 is expressed in Ciona visceral ganglion and are in agreement with the suggestion that the mechanisms involved in the anteroposterior regionalization of the nervous system have been conserved between ascidians and vertebrates.

In Halocynthia roretzi, the expression pattern of HrHox1 was analyzed by Katsuyama et al. (1995). In the nervous system of this ascidian, which lacks a defined visceral ganglion, its expression is limited in the larva to the sensory vesicle and to the junction of trunk and tail. It has been suggested that this latter localization corresponds to the region of the visceral ganglion of other ascidian species (Wada et al., 1998). Since HrHox1 territory of expression is equivalent to the visceral ganglion, there would be an overlapping of expression with CiHox3. Therefore, the study of C. intestinalis Hox1 and Hox2 homologues would be necessary to compare the expression territories of Ciona and Halocynthia Hox genes. To better understand the evolutionary relationships of Hox3 genes between ascidians and vertebrates, we carried out the analysis of CiHox3 promoter regulatory elements with the electroporation method. The results indicated that a sequence of 80 bp, from position −1943 to −1864 from the translation start site, contained the positive regulatory element(s) able to activate transcription in both the visceral ganglion and the sensory vesicle.

Analysis of this sequence by band-shift assay indicated that only the AL2 sequence (Fig. 6A) was able to form a complex with protein extracts from Ciona larvae. This suggested the presence of an unknown transcription factor binding site in the AL2 sequence presumably involved in the neural-specific activation of CiHox3.

It has been proposed that the conservation of a cluster organization for Hox genes during evolution was also due to the presence of common regulatory elements for different Hox genes (Lufkin, 1996). Therefore, we decided to see whether the regulatory elements identified from CiHox3 would be recognized by vertebrate transcriptional networks and, on the contrary, whether the mouse HOXa3 and HOXb3 promoters were able to work in Ciona. In fact, although there is ample proof of the common evolutionary origin of Hox complexes throughout all animal phyla, this has been studied basically on the coding regions for the Hox genes. Much less is known about the conservation and/or divergence of regulatory elements. The regulation of vertebrate group 3 Hox genes has recently been analyzed, and it has been shown that the b-ZIP transcription factor encoded by the kreisler gene is responsible for the upregulation of Hoxb3 in r5 and Hoxa3 in r5 and r6 (Manzanares et al., 1997, 1999). Therefore, direct comparisons between regulation of mouse and Ciona group 3 Hox genes is possible.

Mouse Hoxa3 and Hoxb3 rhombomere-specific regulatory elements tested in Ciona embryos were unable to reproduce any neural-specific expression. This suggests that indeed the regulation of Hox3 genes between mouse and Ciona is somewhat different. However, the kreisler-dependent enhancers could represent a recent acquisition by the higher vertebrates, which occurred during the early duplications leading to four Hox complexes. It is possible that other CNS control regions from the mouse group 3 loci, for example from Hoxd3 or those recently isolated in Hoxa3 (Manzanares et al., 1999), might be able to function in Ciona. Since Ciona is believed to have only one complex, its group 3 CNS elements might have been differentially distributed amongst the multiple mouse group 3 members.

The converse experiment, testing two different CiHox3 promoter fragments in transgenic mouse embryos for their ability to drive reporter lacZ expression, showed that they did not function effectively in mice. The smallest fragment that drives nervous system expression in Ciona was negative in mice. This again suggests a divergence of control elements or mechanisms between these species. While this may arise due to totally different mechanisms, it is also possible that small divergence of the binding site sequences, and/or associated changes in the binding capabilities of the same trans-acting factors, would render Ciona sites unrecognizable by the mouse machinery or vice versa.

Nevertheless, it is interesting that when a slightly larger Ciona enhancer fragment was tested, a reproducible segmental pattern of expression in the mouse hindbrain was obtained. This consisted of expression in the ventral portions of r4 and a second domain starting at the r6/r7 boundary, which spreads posteriorly into the spinal cord. This pattern is reminiscent of those seen using the mouse Hoxb1 r4 enhancer (Marshall et al., 1992; Popperl et al., 1995), and the Hoxb4 r6/7 region A enhancer (Whiting et al., 1991). One possible explanation for the similarities in these patterns, comes from the observation that both the Hoxb1 r4 and the Hoxb4 r6/7 domains, are dependant on auto-/cross-regulatory loops involving Hox/Pbx interactions (Popperl et al., 1995; Gould et al., 1997). Therefore, the larger CiHox3 CNS enhancer that also functions in mice, might contain both the minimal CNS element defined in this study and an auto-/cross-regulatory element. The pattern obtained in mice would then reflect a read-out of mouse Hox/Pbx complexes acting through a CiHox3 Hox-responsive element. This may indicate that the Ciona Hox complex also uses auto-/cross-regulatory interactions for maintaining later expression patterns. In conclusion, this comparative regulatory analysis between Ciona and mice demonstrates that, at least for the group 3 Hox genes, there is considerable divergence in sequences implicated in mediating CNS expression. Further analysis in other chordates and intermediate species will be required to elucidate the molecular basis of these differences.

We are indebted to Dr Albertina Fanelli for her invaluable help in the preparation of the manuscript. We thank Professor Roberto Di Lauro and Dr Rosaria De Santis for helpful discussion and Gennaro Iamunno for sectioning C. intestinalis larvae. We thank Amanda Hewett, Peter Mealyer and Rosemary Murphy for help in animal husbandry. M. M. was supported by HFSP and EEC Marie Curie postdoctoral fellowships and by an EEC Biotechnology Network grant (#BIO4 CT-960378). This work was also funded in part by Core MRC Programme support and an EEC Biotechnology Network grant (#BIO4 CT-960378) to R. K.