Comparative analysis of neurogenesis in the myriapod Glomeris marginata (Diplopoda) suggests more similarities to chelicerates than to insects

Recent molecular and morphological data have challenged the traditional view that insects and myriapods are closely related(Hwang et al., 2001; Friedrich and Tautz, 1995; Boore et al., 1995; Akam et al., 1988; Patel et al., 1989a; Patel et al., 1989b; Abhzanov and Kaufman, 1999; Abhzanov and Kaufman, 2000; Damen et al., 2000; Abhzanov et al., 1999; Damen and Tautz,1998; Damen et al.,1998; Telford and Thomas,1998; Scholtz,1990; Scholtz,1992; Dohle and Scholtz,1988; Harzsch and Dawirs,1996; Whitington et al.,1991). This view was based on supposedly shared characters such as loss of second antennae, formation of Malpighian tubules, postantennal organs and trachea. However, a re-evaluation of these characters shows that they are prone to convergence (Friedrich and Tautz,1995; Dohle,2001). Instead, it is possible to find other characters that are strikingly similar between insects and higher crustaceans, but cannot be found in equivalent form in myriapods (Dohle,2001). The characters in common in insects and crustaceans are the presence of neuroblasts, patterns of axonogenesis in early differentiating neurons, the fine structure of ommatidia and the expression patterns of segmentation genes (Dohle,1997; Dohle,2001). Some molecular data sets even suggest that the chelicerates and the myriapods are sister groups(Friedrich and Tautz, 1995; Hwang et al., 2001). However,morphological data supporting this hypothesis is still missing.

It has been shown recently that neurogenesis in the spider Cupiennius salei (chelicerate) shares several features with Drosophila, but that there are also several important differences. Similar to the generation of neuroblasts in Drosophila, invagination sites arise sequentially and at stereotyped positions in regions that are prefigured by proneural genes(Stollewerk et al., 2001),while neurogenic genes restrict the proportion of cells that adopt the neural fate at each wave of neural precursor formation(Stollewerk, 2002). However,in contrast to Drosophila, groups of cells, rather than single cells,adopt the neural fate at a given time. In addition, neural stem cells,comparable to Drosophila neuroblasts, could not be detected in the ventral neuroectoderm of the spider. Furthermore, there is no decision between epidermal and neural fate in the ventral neuroectoderm of the spider as in Drosophila; instead all cells of the neurogenic region enter the neural pathway.

In all four myriapod groups (Diplopoda, Chilopoda, Symphyla and Pauropoda),the general development of the ventral neuroectoderm follows the same pattern. Ventral to the limb buds, thickenings form as a result of cell proliferation. When the embryo begins to bend about a transverse fold in the middle of the trunk, these thickenings flatten (Anderson,1973). After completion of ventral flexure, the middle part of the hemisegment sinks into the embryo forming a groove(Dohle, 1964). Cell proliferation takes place within this groove pushing newly formed cells towards the basal side and leading to the formation of stacks of cells that project out as rays from the edges of the groove. This structure is called the`ventral organ'. During the course of neurogenesis the ventral organs are gradually incorporated into the embryo while epidermal cells overgrow the ventral nerve cord (Dohle,1964). Neurogenesis has been analysed in a variety of representatives of all myriapod groups, but failed to reveal stem cell-like neural precursors with morphological characteristics of insect or crustacean neuroblasts (Heymons, 1901; Tiegs, 1940; Tiegs, 1947; Dohle, 1964; Whitington et al., 1991). Furthermore, Whitington and co-workers(Whitington et al., 1991)showed that in the centipede Ethmostigmus rubripes the earliest central axon pathways do not arise from segmentally repeated neurons as in insects but by the posterior growth of axons originating from neurons located in the brain. In addition, the axonal projections and the cell body positions of the segmental neurons clearly diverge from the pattern described in insects and crustaceans (Whitington et al.,1991).

Here, we have analysed the pattern of early neurogenesis in the myriapod Glomeris marginata and compared it to the recently obtained data for the spider Cupiennius salei(Stollewerk et al., 2001; Stollewerk, 2002).

Adult Glomeris marginata (Myriapoda, Diplopoda) were collected in the city forest of Cologne, Germany, between April and August 2002. Glomeris were kept in 20×10 cm plastic containers at room temperature with high humidity provided by wet cloths and earth. The females made egg chambers out of earth to cover the eggs. The eggs were collected daily and kept separately until they were between 6- and 12-days old.

Adult Archispirostreptus were obtained from the Aquazoo in Düsseldorf, Germany. Ten adults were kept at 28°C in large terraria filled to a depth of at least 20 cm with earth. Females laid a clutch of eggs into the earth, approximately every 3 weeks, which were collected, staged,cleaned with bleach (under 5%) and frozen at –80°C for RNA extraction.

Glomeris eggs were removed from their egg chambers by submerging them in water. They were transferred to a 2 ml Eppendorf tube and washed several times in water. They were then dechorionated by leaving them in bleach(under 5%) for 2 minutes and rinsing several times with water. Embryos were frozen at –80°C for RNA extraction or fixed in 1 ml heptane, 50μl formaldehyde (37%) for later use. Embryos for antibody stainings were fixed for 20 minutes on a wheel and then washed several times with 100%ethanol and stored at –20°C. For in situ hybridizations, embryos were fixed for 4 hours on a wheel and washed with 100% methanol before storage at –20°C. After storage, at least overnight, the embryos could be devitelinized with tweezers for further staining.

GmASH, GmNotch, GmDelta, AsAsh, and AsDelta were initially found by RT-PCR on cDNA synthesized from RNA extracted from 7- to 12-day Glomeris, or 2- to 3-week Archispirostreptus embryos,respectively. Degenerate primers for the respective genes were used as described by Stollewerk et al. (Stollewerk et al., 2001; Stollewerk,2002). The obtained PCR fragments were cloned into the pZero vector (Stratagene) and sequenced. Larger fragments were obtained by rapid amplification of cDNA ends (Marathon cDNA Amplification Kit, Clonetech;GeneRacer Kit, Invitrogene). Sequence reactions were performed on plasmid preparations with BIG DYE and run on an Abi Prism automatic sequencer. The sequences obtained were deposited with GenBank (Accession Numbers: GmDelta, AJ36341; GmNotch, AJ36342; AsDelta,AJ36343; AsNotch, AJ36344; AsAsh, AJ36345; GmAsh,AJ36346; TcASH, AJ36347).

Sequences were analysed and aligned with related amino acid sequences taken from the NIH Blast database in Bioedit. Trees were constructed using the PAUP NJ minimum evolution algorithm with 1000 bootstrap replicates. Positions where an amino acid insertion was present in only one sequence were removed, as was the variable part of the loop for the Ash alignment. Since the portion of the Ash sequences that could be aligned is very short (the BHLH domain), the presence or absence of a loop was used as an extra character (32/57 informative characters). For Delta, the DSL domain and EGF repeats 1 and 2 were aligned (70/109 informative characters), for Notch the 5′ sequences up to EGF repeat 12 were aligned (266/311 informative characters).

Whole-mount in situ hybridizations were performed as described for Danio rerio, with the modification that 20× SSC pH 5.5 was used instead of 20× SSC pH 7.4, to reduce the background(Bierkamp and Campos-Ortega,1993).

Phalloidin-rhodamine staining of Glomeris embryos was performed as has been described for flies (Stollewerk,2000). Immunocytochemistry was performed as described previously(Stollewerk et al., 2001). The anti-phospho-histone 3 (PH3) antibody was provided by F. Sprenger (Institute for Genetics, Cologne). Histology was performed as described previously(Stollewerk et al., 1996).

The embryonic segments of Glomeris marginata arise sequentially. Shortly after formation of the germband [stage 1; stages after Dohle(Dohle, 1964)], the first five anterior segments that contribute to the head are visible: the antennal segment, the premandibular segment, the mandibular segment, the maxillar segment and the postmaxillar segment. Three leg segments can also be distinguished at this stage. The next leg segment forms from the posterior growth zone at stage 2, while owing to the formation of intersegmental furrows the remaining segments are more clearly visible(Dohle, 1964). At stage 3 limb buds arise on the antennal, the mandibular and the maxillar segments, as well as on the three leg segments. It is important to note that these anlagen are formed simultaneously. At the end of stage 3 limb buds are also visible on the fourth leg segment and a fifth segment has been generated by the posterior growth zone. A thickening of the cephalic lobe and the ventral neuroectoderm can be observed at stage 4. Limb buds are now also visible on the fifth and sixth segment. At stage 5 a ventral-dorsal furrow forms at the level of the postmaxillar segment, so that the embryo curves inward and the head eventually approaches the anal pads at stage 6.

To analyse the morphology of the ventral neuorectoderm in Glomeriswe stained embryos at stage 4 with phalloidin-rhodamine, a dye that stains the actin filaments, and investigated the cell shapes in the confocal laser-scanning microscope (LSM). At this stage a thickening of the neuroectoderm is already visible (see above) and the extension of the ventral neuroectoderm is clearly demarcated medially by the ventral midline and laterally by the limb buds. We made flat preparations of embryos stained with phalloidin-rhodamine and scanned them from apex to the base using the LSM. Similar to the situation in the spider, we detected dots of high phalloidin-rhodamine staining in apical optical sections of the neuroectoderm of the head segments and the first five leg segments(Fig. 1, see also Fig. 3). More basal optical sections of the same regions (at a depth of 11-21 μm from the apical surface of the embryo) revealed that groups of basally enlarged cells are located underneath the strongly stained dots, indicating that these dots mark the sites of invagination of neuroectodermal cells(Fig. 1).

As in the spider, 30-32 invaginating cell groups are arranged in a regular pattern of seven rows consisting of four to five invaginaton sites each. However, analysis of serial transverse sections revealed that up to 11 cells contribute to an individual invagination site(Fig. 2), while in the spider only five to nine cells were counted. Furthermore, in contrast to the spider,the ventral neuroectoderm has a multi-layered structure: the apical region covered by a single invagination site seems to be larger and the spacing between the individual invagination sites is narrower than in the spider(Fig. 1A,B). The reason for these morphological features is that the invaginating cell groups are located closer together and because of limited space come to lie over and above each other (Fig. 2B). The invaginating cells of a group do not all occupy a basal position as in the spider, but they also form stacks of cells(Fig. 2D). Since more cells contribute to an invagination site and the cell processes of the invaginating cells are not as constricted as in the spider, the apical region occupied by an individual invagination site is larger than in the spider(Fig. 2A).

A detailed analysis of different embryonic stages revealed that the invagination sites form sequentially in G. marginata. The same numbers of invaginating cell groups arise simultaneously in the five head segments and the first two leg segments, while the invagination sites are formed in an anterior to posterior gradient in the remaining segments. A tight comparison of the relevant embryonic stages showed that the invagination sites are formed in four waves generating five to 13 invaginating cell groups each,as in the spider (Fig. 3, see also Fig. 7A-E). When the limb buds form, the first invagination sites arise in the medial region of each hemisegment in the five head segments and the first two leg segments at stage 2 (Fig. 3B,G). During the second wave of invagination site formation at stage 3, new invaginating cell groups arise anteriorly, posteriorly and in-between the existing invagination sites (Fig. 3C,H). The next invagination sites to arise form a semicircle around the central region where invagination sites have already formed(Fig. 3D,I). During the last wave of generation of invagination sites, invaginating cell groups arise in an anterior medial region and in between the existing invagination sites (compare also Fig. 7E). Although the embryo then curves inward and the ventral neuroectoderm stretches along the mediolateral axis, the arrangement in seven rows is maintained(Fig. 3E,J).

In summary, the data show that, as in the spider, groups of invaginating cells are generated in four waves that show a regular pattern strikingly similar to the arrangement of invagination sites in the spider.

The thickening of the ventral neuroectoderm is a result of cell proliferation (Anderson, 1973). To see whether there is a connection between cell proliferation and formation of invagination sites, we double stained embryos with the mitotic marker anti-phospho-histone 3, to analyse the pattern of cell divisions, and phalloidin-rhodamine, to visualize the invagination sites.

In contrast to the spider, in which cell proliferation does not coincide with the formation of invagination sites(Stollewerk et al., 2001), in Glomeris mitotic cells are associated with invagination sites and seem to prefigure the regions where invagination sites arise in the ventral neuroectoderm (Fig. 4). During formation of the first invagination sites at least one mitotic cell abuts the invaginating cell group (Fig. 4D-F), while groups of cells and individual cells could be detected in the regions where invagination sites form hours later(Fig. 4A-C). Most cell divisions occur in the apical cell layer (data not shown), as in the spider. It was not possible from these experiments to determine whether the cells of a particular invagination group are clonally related.

The sequential formation and the regular pattern of the invagination sites in the ventral neuroectoderm of Glomeris suggest that, as in Drosophila and the spider Cupiennius salei, proneural and neurogenic genes regulate the recruitment of neural precursor cells from the neurogenic region. It has been shown recently in the spider that the proneural gene CsASH1 is expressed before formation of the invagination sites in the appropriate regions of the ventral neuroectoderm(Stollewerk et al., 2001). Functional analysis revealed that the gene is responsible for the establishment of the neural precursors. Furthermore, one Notch and two Delta homologues of the Drosophila neurogenic genes restrict the proportion of cells that are recruited for the neural fate at each wave of neural precursor formation. To see whether a similar genetic network is involved in early neurogenesis in Glomeris we have cloned achaete-scute, Notch and Delta homologues. We also cloned homologues from Archispirostreptus sp., a distantly related millipede, to ensure that the data collected from Glomeris is representative of Diplopod millipedes. Conserved features found in Notch,Delta and acheate-scute homologues allowed us to amplify small fragments of these genes from Glomeris and Archispirostreptus sp. cDNA. Only one fragment per species was found for each gene.

Rapid amplification of the 5′and 3′ ends (RACE) of GmASH resulted in a 1000 bp fragment with an 804 bp open reading frame (ORF), Archispirostreptus 5′ and 3′ RACE led to a 1200 bp fragment with an ORF of 864 bp. Both sequences have a single start codon with a short conserved motif also found in the CsASH genes, as well as upstream and downstream stop codons. The deduced amino acids of full-length GmASH and AsASH sequences showed a similarity of 61%, with 86% identical amino acids in the bHLH domains. The deduced amino acid sequence of GmASH is 83% identitical to Homo sapiensAchaete-Scute Complex homolog-like 1, while the Archispirostreptussequence is 81% identical to the Gallus gallus transcriptional regulator CASH over the region of the bHLH domain. Outside of this domain, it is only possible to align a short conserved domain at the end of the protein. The alignment of the bHLH domains with other ASH proteins showed that, in contrast to insects, the millipede sequences, like their spider and vertebrate homologues, have a highly reduced loop(Fig. 5A). A tree was constructed from an alignment of the bHLH domains of nine insect, five vertebrate, two Cupiennius salei and the myriapod sequences(Fig. 6A). The node joining the myriapod, spider and vertebrate sequences has very high bootstrap support(94), while that joining the insects has low support.

Larger fragments of the GmDelta and GmNotch genes were amplified by 5′ RACE. This resulted in a 850 bp GmDeltasequence with an open reading frame of 280 amino acids covering a part of the N-terminal signal sequence, the DSL domain and EGF repeats 1 and 2, and a 1110 bp GmNotch sequence with a 377 deduced amino acid sequence comprising the N terminus and the first 12 EGF repeats. For both sequences, it was not possible to unambiguously identify the start codon.

For Delta, an Archispirostreptus sequence covering the DSL domain and EGF repeats 1 and 2 with 81% similarity to the CsDelta1 protein was isolated. The complete GmDelta sequence shares 57% identical amino acids with CsDelta1, while the fragment similar to AsDelta has a similarity of only 62% with CsDelta1. An alignment of the DSL domain shows that these sequences are highly conserved(Fig. 5B). The DSL domain and EGF repeats 1 and 2 from two insect species, five vertebrates, the two myriapods, the spider Cupiennius salei and one ascidian sequence(Ciona savigny) were aligned to create the tree shown in Fig. 6B. Here, the insects and the vertebrates form two clear groups, while the myriapods and the spider form another. All three of these groups have relatively high bootstrap support:insects, 86%; vertebrates, 100%; spider and myriapods, 74%.

The obtained Glomeris Notch sequence, which shares 68% of its amino acids with the Boophilus microplus (chelicerate) Notch homologue, was aligned with seven Notch homologues to create the tree shown in Fig. 6C. The high sequence similarities between the Glomeris and the Boophilus proteins are reflected by the tree, where the node joining the Chelicerates with the vertebrates and Glomeris has a bootstrap support of 100%(Fig. 6C). The insect sequences, in contrast, are joined by a node with less than 50% support.

GmASH transcripts were first detected before formation of the limb buds at stage 1. At this time no invagination sites are visible in the ventral neuroectoderm (Fig. 7A). The gene is expressed in neuroectodermal cells in the middle of each hemisegment in the head and the first two leg segments at heterogeneous levels(Fig. 7F). Groups of cells express high levels of the gene, while there is a weak uniform expression in the remaining regions (Fig. 7F). At stage 2 invagination sites arise in the expression domains of GmASH (Fig. 7B). At this time transcripts can be detected anterior, posterior and in between the first invagination sites (Fig. 7G, Fig. 8A). Again the next invagination sites to arise are generated in the regions of GmASH expression (Fig. 7C). Although the gene is simultaneously expressed in the head segments and the first two leg segments, the expression domains in the antennal, premandibular, mandibular and maxillar segments seem to be smaller than in the remaining segments (Fig. 8A-D). At stage 3 the expression domains of GmASH form a semicircle around the area where invagination sites have already formed(Fig. 7H, Fig. 8B). This expression pattern again prefigures the regions where invagination sites will be formed hours later (Fig. 7D). Before the last wave of formation, GmASH is expressed in the corresponding regions in between and anterior-medial to the existing invagination sites. In addition, the gene is transiently expressed in the invaginating cell groups and in the neural precursors of the peripheral nervous system(Fig. 7J, Fig. 8C,D).

In summary, the data show that the Glomeris achaete-scutehomologue is expressed prior to invagination of the neural precursors in the appropriate regions, similar to the spider gene.

GmDelta is first expressed during the first wave of neural precursor formation at stage 2. Transcripts can be detected at low levels in all ventral neuroectodermal cells, but accumulate at higher levels in the invaginating cell groups, similar to the expression pattern of the spider CsDelta2 gene. GmDelta is also expressed in all invagination sites generated subsequently (Fig. 9D-F). The expression seems to be rapidly down regulated, since transcripts cannot be detected in all invagination sites generated during different waves (Fig. 9A,B,D,F).

GmNotch is expressed in segmentally repeated stripes at stage 1,but shows a stronger expression in the ventral neuroectoderm. During formation of the first invagination sites the expression in stripes becomes restricted to the dorsal part of the embryo, lateral to the limb buds(Fig. 10D). GmNotchis expressed at weak levels in almost all cells of the ventral neuroectoderm up to leg segment 3 (Fig. 10D). During the next wave of neural precursor formation at stage 3, there is a clear heterogeneity in the expression levels of GmNotch(Fig. 10A). This expression pattern is maintained during subsequent waves of neural precursor formation. GmNotch expression extends to more posterior segments during the course of neurogenesis. As in the anterior segments, the expression is uniform during the first wave of neural precursor formation and shows heterogeneous expression during formation of the remaining invagination sites.

In summary, the data show that homologues of the Drosophila Notchand Delta genes are expressed in the ventral neuroectoderm during neurogenesis in a spatiotemporal pattern, suggesting that these genes are involved in the recruitment of neural precursors.

The cells in the ventral neuroectoderm of Drosophila have a choice between an epidermal and a neural fate. It has been shown recently, that this decision does not take place in the neurogenic region of the spider. Rather,all cells of the ventral neuroectoderm enter the neural pathway(Stollewerk, 2002).

Analyses of transverse and horizontal sections of the ventral neuroectoderm of Glomeris embryos revealed that the invaginating cell groups detach from the apical surface at stage 6. At this stage a medial thickening forms in each hemineuromere (Fig. 11C). Subsequently, the neuroectoderm thickens at the lateral border adjacent to the limb buds (Fig. 11D) and the whole ventral neuromere sinks into the embryo while the epidermis overgrows the ventral nerve cord (Fig. 11A,D). At this time, a ladder-like axonal scaffold is already visible on the basal side (Fig. 11B), suggesting that there is no decision between epidermal and neural fate during the formation of neural precursors in the ventral neuroectoderm of Glomeris, as is the situation in the spider.

Our results show that, as in the spider, groups of cells invaginate from the ventral neuroectoderm of Glomeris. Number and arrangement of the invagination sites are strikingly similar to the spider pattern. In both the spider and Glomeris, there are about 30 invaginating cell groups arranged in a regular pattern of seven rows consisting of four to five invagination sites each. In addition, in both species the invaginating cell groups are generated in four waves. In contrast, in insects approx. 25 neuroblasts are generated per hemisegment that delaminate as individual cells from the ventral neuroectoderm in five waves. The first two populations of neuroblasts are arranged in three longitudinal columns and four rows per hemisegment. This regular pattern is lost after delamination of the next population of neuroblasts, because earlier born neuroblasts are shifted into a more basal position (Goodman and Doe, 1993). In summary these data show that the pattern of neural precursors and their mode of generation in Glomeris are more similar to those in the spider than in the insects.

However, there are some special features in the millipede that are different from spider neurogenesis. After formation of the first invagination sites, the ventral neuroectoderm forms a multi-layered structure of small cells, while in the spider there is only one single cell layer. The reason for this morphological difference is that because of limited space in the ventral neuroectoderm the invagination sites are located closer together and come to lie over and above each other. In addition, the invagination sites in Glomeris consist of up to 11 cells as compared to a maximum of eight in the spider and they do not all occupy a basal position. The invaginating cells do not all have the typical bottle-like shape as in the spider, so that their cell processes cover larger apical areas, and the dots of high phalloidin staining appear bigger in the millipede.

Furthermore, although the pattern of the invagination sites is very similar in the spider and Glomeris, the relative timing of generation of individual invagination sites is different. While in the spider the first invagination sites arise in the most anterior lateral region of the hemisegments, the first invaginating cell groups of the millipede are visible in the middle of the hemisegment. The next wave of invagination sites in the spider generates invaginating cell groups in coherent medial and posterior regions abutting the former generation sites. In contrast, in Glomeris newly formed invagination sites are distributed all over the hemisegement. Furthermore, the two most lateral anterior invagination sites that occupy strikingly similar positions in the spider and the millipede (see Fig. 1A,B) are generated during the first wave of invaginations in the spider, while they are not visible until the third wave of neural precursor formation in Glomeris.

However, the generation of neural precursors at stereotyped positions seems to be an ancient feature that has been maintained throughout the evolution of arthropods. Future analysis will show if the cells of an individual invagination site give rise to an invariant pattern of neurons in spiders and millipedes similar to the progeny of an identified neuroblast in insects and crustaceans.

Studies of neurogenesis in different representatives of all myriapod groups have failed to reveal stem cell-like cells with the characteristics of insect or crustacean neuroblasts (Heymons,1901; Tiegs, 1940; Tiegs, 1947; Dohle, 1964; Whitington et al., 1991). It is assumed that neurons are produced by a generalized proliferation of the ventral neuroectoderm. However, Knoll(Knoll, 1974) proposed that neuroblasts are present in the apical layer of the centipede Scutigera coleoptrata generating vertical columns of neurons; a mode of neural precursor formation that would be very similar to the crustacean pattern. Analysis of neurogenesis in another centipede, Ethmostigmus rubripes,led Whitington and co-workers (Whitington et al., 1991) to the assumption that neural precursors with the characteristics of insect neuroblasts are absent in this species. They could not detect sites of concentrated mitotic activity or dividing cells that are significantly larger than the surrounding cells. Similar results have been obtained for the spider: scattered mitotic cells are distributed over the neuroectoderm that do not prefigure regions where neural precursors form(Stollewerk et al., 2001). However, our analysis of the mitotic pattern in the ventral neuroectoderm of Glomeris revealed that dividing cells are associated with invaginating neural precursors. Furthermore, groups of dividing cells seem to prefigure the regions where invagination sites arise. In contrast to the results from Ethmostigmus the dividing cells are significantly larger in size than the surrounding cells in the millipede. Therefore, we assume that stem cell-like cells are present in the apical layer of the ventral neuroectoderm in Glomeris, although this has to be confirmed by single cell labelling and BrdU injections. Since the dividing cell groups are located in the apical cell layer and are present before formation of the invagination sites, they are different from insect and crustacean neuroblasts. In Drosophila the neuroblasts do not divide until they delaminate from the outer layer. In contrast, the crustacean neuroblasts do not delaminate but remain in the apical layer, dividing parallel to the surface,so that the progenies are pushed to the basal side(Dohle and Scholtz, 1988; Scholtz, 1990; Scholtz, 1992).

The millipede pattern of neural precursor formation is intermediate when compared between chelicerates and insects. While in the spider neuroectodermal cells seem to divide randomly and are recruited for the neural fate because of their positions in the hemisegment, the presence of neural stem cells in the millipede links cell proliferation and generation of neural precursors in the apical cell layer. The necessity to single out epidermal and neural precursor cells from the ventral neuroectoderm in insects and crustaceans has led to a separation of the generation sites: epidermal cells are produced in the apical layer, while neural cells are generated on the basal side. Furthermore, the fact that neuroblasts are missing in almost all lower crustaceans analysed and that their mode of neurogenesis seems to be similar to that of myriapods, i.e. a separation of the ganglia into the interior(Anderson, 1973) indicates that an entirely neurogenic ventral neuroectoderm may be the ancestral state.

It is known that after completion of ventral flexure, the middle part of each hemisegment sinks into the embryo forming a groove(Dohle, 1964). During the course of neurogenesis this region is gradually incorporated into the embryo while epidermal cells overgrow the ventral nerve cord. We show here that this process does not take place until the final differentiation of the invaginated neural precursors, since a neuropil is already visible on the basal side, when the neuromeres sink into the embryo. This means that as in the spider, and in contrast to the situation in the insects, there is no decision between epidermal and neural fate in the central neurogenic region of the millipede.

As in the spider and also in the insects, the Glomeris achaete-scute homologue is expressed before formation of neural precursors in the ventral neuroectoderm. Like the spider CsASH1 gene, GmASH is expressed in patches of cells in the neuroectoderm and becomes restricted to the invaginating cell groups, while in insects proneural gene expression is reduced to one cell of the proneural cluster. After each wave of neural precursor formation GmASH is re-expressed in the regions where invagination sites form, indicating that the gene is necessary for the formation of all neural precursors. However, this has to be confirmed by functional analysis.

During formation of neural precursors in Drosophila, the neurogenic genes Notch and Delta appear to be uniformly expressed in the neuroectoderm (Baker,2000). Although it is assumed that within a proneural cluster the cell expressing the highest amounts of Delta is selected for the neural fate, no variation in Delta expression has yet been observed in the ventral neuroectoderm of fly embryos. In contrast, the expression patterns of the spider Delta genes can be correlated to the formation of neural precursors. While CsDelta1 is exclusively expressed in neural precursors, CsDelta2 transcripts are distributed uniformly throughout the neuroectoderm and accumulate in nascent neurons. The only Delta gene we have found in Glomeris is expressed similarly to the spider CsDelta2 gene: at low levels in almost all neuroectodermal cells and at higher levels in the invaginating cell groups. Furthermore, although Glomeris Notch is initially expressed uniformly in the neuroectoderm, it resolves into a heterogeneous expression during the first wave of neural precursor formation similar to the spider CsNotch transcripts.

To summarize, our data support the hypothesis that myriapods are closer to chelicerates than to insects. The spider and the millipede share several features that cannot be found in equivalent form in the insects: in both the spider and the millipede, about 30 groups of neural precursors invaginate from the neuroectoderm in a strikingly similar pattern. Furthermore, in contrast to the insects, there is no decision between epidermal and neural fate in the ventral neuroectoderm of both species analysed. The sequence data also suggest a closer relationship of the millipede to the spider than to the insects. However, to confirm a sister group relationship of these arthropod groups,more morphological data from different representatives of myriapods,chelicerates and outgroups will be necessary.

We thank Diethard Tautz for continued support, critical discussions and helpful comments. We are grateful to José A. Campos-Ortega for providing access to the histological equipment and Dr Löser for supplying an Archispirostreptus stock. We thank Daniela Krackl and Sylvia Niciporuk for help with collecting myriapods and taking care of them. This work was supported by a grant from the Deutsche Forschungsgemeinschaft (Sto 361/1-3).