The general handling of G. marginata is described in Janssen et al. 12. The embryos were allowed to develop at room temperature (22 to 25°C). The developmental stage of the embryos was determined by 4'-6-diamidino-2-phenylindole (DAP) staining. Staging was performed as described previously 1213.

Specimens of L. forficatus were collected from a leaf litter stack in the backyard of the Evolutionary Biology Centre (EBC) in Uppsala/Sweden in spring (May/June). Around 50 centipedes were held at room temperature in a spacious plastic box filled with washed leaf litter (washing away small particles makes the later finding of the eggs easier). The adults were fed with pieces of common earthworms (Lumbricus) every few days. The often detritus-covered eggs were collected by hand and incubated in plastic dishes on damp paper tissues until they reached the desired developmental stage. Staging was performed as described previously 14. Generally, the handling was carried out similarly to the method described for Lithobius atkinsoni 15.

Embryos of the spider Cupiennius salei, the onychophoran E. kanangrensis and the red flour beetle Tribolium castaneum were obtained and treated as described previously (161718, respectively).

Fragments of Ubx and Antp transcripts of G. marginata were obtained via 5' and 3' rapid amplification of cDNA ends (RACE)-PCR (Gene Racer RACE Kit; Invitrogen, Carlsbad, CA, USA). A fragment (383 bp) of Tribolium Ubx corresponding to the C-terminal end of the open reading frame (ORF) (94 bp) and the beginning of the 3' untranslated region (UTR) was isolated with gene-specific primers (Table 1). General Hox primers, as described previously 19, were used to isolate a small fragment of Ubx from Euperipatoides cDNA. An extended fragment was subsequently obtained by 3'-RACE.

A fragment of Lithobius forficatus Ubx was isolated with gene-specific primers based on the published sequence of Lithobius atkinsoni Ubx 15. The isolated L. forficatus fragment is only 221 bp long, but works well in hybridization experiments.

Part of the bicistronic transcripts containing Ultrabithorax and Antennapedia (Ubx/Antp) were isolated from the brine shrimp Artemia (first PCR), the onychophoran Euperipatoides and the millipede Glomeris. The gene-specific primers used were directed against the homeodomains of Ubx (forward primer) and Antp (backward primer). Gene-specific primers to amplify a possible Tribolium Ubx/Antp transcript failed, even though we used the primers (Table 1) in all possible combinations including nested PCRs.

Sequences of the fragments were determined from both strands by sequencing (Big Dye Terminator Cycle Sequencing Kit; Perkin-Elmer Applied Biosystems, Foster City, CA, USA) chemistry on an automatic analyser (ABI3730XL; Perkin-Elmer Applied Biosystems) by a commercial sequencing service (Macrogen, Seoul, Korea). Sequences are available in GenBank under the accession numbers FN687748 (Gm-Ubx), FN687749 (Gm-Antp), FN687750 (Gm-Ubx/Antp_variant II), FN687751 (Ek-Ubx), FN687752 (Ek-Ubx/Antp_variant I), FN687753 (Ek-Ubx/Antp_variant II), FN687754 (Lf-Ubx) and FN687755 (Af-Ubx/Antp_variant II).

Whole-mount in situ hybridization for all species was performed as described previously for Glomeris 20. Double whole-mount in situ hybridization and cell nuclei detection using DAPI was performed as described by Janssen et al. 21. Embryos were analyzed under a dissection microscope (Leica, Heerbrugg, Switzerland) equipped with a digital camera (Axiocam; Zeiss, Jena, Germany) or a DC100 (Leica) digital camera. Brightness, contrast and colour values were corrected in all images using image processing software (Adobe Photoshop CS2., V.0.1 for Apple Macintosh; Adobe Systems Inc. San Jose, CA, USA).

Partial sequences of the transcripts of all ten Hox genes of G. marginata were published previously 19. In all cases except fushi-tarazu, only part of the homeodomain and 3' UTR sequence was obtained. The published Ubx fragment neither ends in a poly-A tail nor has one of the typical polyadenylation sites and is therefore likely to be incomplete. Recent 3'-RACE experiments demonstrated the presence of additional 3' UTR transcript. The extended fragment ends in a poly-A tail, but lacks an obvious polyadenylation site close to this. The 3' UTR region contains nine possible polyadenylation sites more distant from the poly-A tail, allowing for the presence of transcripts with different 3' UTR length. Whether the recovered '3' UTR' sequence is a typical UTR that occurs in the monocistronic transcript of Ubx or if is merely the result of the bicistronic transcript of Ubx and Antp (see following section) is unclear.

We recovered 5'-RACE fragments of Ubx and Antp. The Ubx fragment represents the complete N-terminal region of the ORF and 5' UTR sequence. The 5'-Antp fragment is incomplete and does not include the N-terminal region of the protein coding sequence and the 5' UTR. The fragments encode conserved motifs that are characteristic for Ubx and Antp orthologs in arthropods (Figure 1A). Note that the Glomeris ANTP protein lacks the characteristic SQFE motif between the hexapeptide and the homeodomain. Instead, this short peptide is replaced by a single lysine (K) in Glomeris (Figure 1A). The expression pattern of all newly recovered fragments is identical to those described previously 19 (not shown).

For Glomeris, we identified an Ubx/Antp bicistronic transcript that encodes the Ubx homeodomain C-terminal to the upstream primer position and 38 bp of the Ubx 3' UTR, which is directly adjacent to the complete N-terminal part of the Antp homeodomain up to the downstream primer position (splice variant II; see below) (Figure 1B,B'). Whether the sequence C-terminal to this sequence is part of the fusion transcript is unclear; however, the sequence N-terminal to the described short fusion transcript has been independently recovered by 5' RACE using gene specific primers (GSPs) against the Antp homeodomain that amplified the Ubx/Antp fusion transcript instead of the Antp 5' transcript. This sequence is part of the Ubx transcript as proven by 5'-RACE PCR for Ubx.

We also successfully isolated a splice version (splice variant I) of Ubx/Antp bicistronic transcripts from an onychophoran (Euperipatoides). This splice variant I is also described for a number of several crustaceans including the brine shrimp Artemia 11 (Figure 1B). For Euperipatoides and Artemia, we also isolated the shorter splice variant II of the bicistronic transcript described for Strigamia 9 (Figure 1B,B'). A splice variant I is not described for Strigamia and we could not isolate it from Glomeris either. We failed to detect any Ubx/Antp bicistronic transcripts in the beetle Tribolium (Insecta).

The information on aUbx transcription is based on probes detecting the Ubx antisense strand during in situ hybridization experiments (Figure 1C). It was thus necessary to unravel the true extension of the aUbx transcript by in situ hybridization experiments with minimum size probes (around 300 bp for Glomeris) detecting aUbx complementary to the ends of the available Ubx fragments (Figure 1C). In all cases these sense probes detected the aUbx expression pattern (described below) suggesting their complete transcription. Whether the aUbx transcript extends the Ubx transcript is unclear; however, it does not extend into the transcripts of abdominal-A (abd-A) or Antennapedia (Antp), because in situ hybridization experiments with anti-abd-A and anti-Antp probes did not detect any transcription. The longest possible ORF of the aUbx transcript is 113aa long and encodes a repetitive sequence of the type (LLLLR/cSE) (Figure 1D).

Transcripts of aUbx can already be detected at the blastoderm stage in a broad posterior domain (Figure 2A); at stage 0.2, this expression intensifies (Figure 2B). At the next stage (0.3) the centre of the initial broad domain is cleared from the transcripts (Figure 2C). At stage 0.4, the domain splits into an anterior stripe and a broad posterior domain (Figure 2D). The broad domain lies anterior to the future proctodaeum; the anterior stripe covers the intersegmental indentation between trunk segment two (T2) and T3, and is thus located in the posterior part of T2. At stage 1, the posterior domain has broadened and its anterior and posterior margins show enhanced expression (Figure 2E). At stage 1.2, the complete T2 segment expresses aUbx, although the expression in its anterior part is weak (Figure 3A). The anterior margin of the former broad domain (Figure 2E) has now transformed into an independent stripe in the posterior of T3 (Figure 3A). The posterior-most expression is in the anal valves (AV). Ventrally, the expression of aUbx is weaker than in its corresponding lateral and dorsal tissue (Figure 3A). At the subsequent stage (stage 2) three stripes of aUbx expression are detectable: in the posterior areas of T1, T2 andT3 (Figure 3B). This expression is restricted to the ventral tissue only for the stripes in T1 and the T3, whereas the stripe in T2 extends into the dorsal tissue. All stripes are discontinuous at the ventral midline. At around stage 3, an additional stripe forms in the posterior of T4 (Figure 3C). In subsequent stages (4 to 6), additional discontinuous stripes of aUbx appear in the ventral germ band with the formation of additional segments. Expression in dorsal tissue, legs and anal valves remains unchanged. Expression of the anterior-most aUbx stripe (the posterior stripe in T1 (T1p)), is enhanced at these stages (Figure 3D and data not shown). Note that although the legs posterior to T3 are forming, aUbx is not expressed in their tips (Figure 3D). The posterior-most part of the developing early embryo, which will later give rise to the hindgut and the proctodaeum, remains free from aUbx expression (Figure 2).

The Gm-aUbx transcript is regulated in a similar, but complementary, specific pattern to that of Gm-Ubx (Figure 2, Figure 3, Figure 4). Expression of aUbx starts earlier (stage 0) than expression of Ubx (stage 0.2 or 0.3), but in a comparable posterior area. Double in situ hybridization to detect possible overlap of early Ubx and aUbx expression is not possible because the signal of Ubx is too weak in the early stages (Figure 2G-J). At stage 1, Ubx expression is still restricted to the posterior growth zone and is not present in the nascent segment T3 (Figure 2J), unlike the previously reported expression in T3 in stage 1.2 embryos 19. At this stage, the anterior margin of aUbx is clearly more anterior (T2) than that of Ubx (T3). At stage 2, it becomes obvious that the expression patterns of Ubx and aUbx are indeed broadly complementary (Figure 4A-C). The stripe of aUbx expression extending into the dorsal tissue lies in the posterior of T2, and is thus anteriorly abutting the expression of Ubx (Figure 4A-C). The ventral aUbx stripe in T3 coincides with a lack of Ubx expression in this area (Figure 4A-D). Very faint expression of Ubx extends minimally into T2 ventrally (Figure 4B), and aUbx is weakly expressed anterior to this (Figure 4A-C). Whereas the ventral expression of Ubx at stage 4-6 becomes more complex, the expression of aUbx remains as stripes (Figure 4D), which are complementary to the expression of Ubx. (Figure 4D). The anterior shift of aUbx into the posterior of T1 (Figure 3A,B) coincides with a shift of Ubx expression into the complete ventral part of T2 (Figure 4F) 19. The anterior border of dorsal Ubx is shifted towards the posterior compared with its anterior border in ventral tissue (Figure 4F) 19. In dorsal tissue, the expression of aUbx still abuts the expression of Ubx and is hence also shifted towards posterior (Figure 3C, Figure 4E; also seen in Figure 4F for a stage 4 embryo).

The isolated fragment of L. forficatus (Uppsala/Sweden) Ubx is 93% (206 of 221 bp) identical with the orthologous sequence of L. atkinsoni 15 and 98% (64 of 65 bp) identical with the sequence of L. forficatus (UK) 22. The expression pattern of Lf-Ubx is identical to that described for L. atkinsoni 15). As expected from the data for Glomeris and Strigamia, the antisense transcript (Lf-aUbx) is also transcribed. The expression pattern of Lf-aUbx is complementary to that of Ubx and very similar to that of Strigamia antisense Ubx in embryos with 30 leg-bearing segments (LBS) (Figure 5) 9. A broad central domain in the first walking leg segment (L1) abuts the anterior-most expression of Ubx which extends into the very posterior of L1. Dorsal to that, in the region of the developing legs, aUbx is expressed as a thin stripe at the border of the maxillipedal segment (mxpd) and L1 (Figure 5). We expect that the expression pattern of Lf-aUbx is more complex in older developmental stages 9.

We investigated the possible expression of aUbx in members of other arthropod classes and an onychophoran. Sense probes of the same length as the antisense probes used for the detection of Ubx in Tribolium (Insecta), the two known Ubx paralogs in Cupiennius (Chelicerata) 16, and Ubx in Euperipatoides (Onychophora) failed to detect any transcripts. In all cases, positive controls detecting the Ubx signal were successfully probed with antisense probes in parallel experiments (data not shown).

Sequence and expression data of Ultrabithorax are presently known from four myriapod species: the geophilomorph Strigamia maritima (Chilopoda) 9; the lithobiomorph species L. atkinsoni and L. forficatus (15 and this study); and the pill millipede G. marginata (Progoneata) 19. In all cases, the antisense DNA strand complementary to Ubx is transcribed and the expression pattern of the antisense transcripts (aUbx) is complementary to that of the sense transcript (coding transcript; Ubx) (9 and this study). This finding suggests that complementary expression of sense and antisense transcripts generated from the Ubx locus is conserved between all myriapods.

Because aUbx expression has not yet been detected outside the Myriapoda, but has been detected in Chilopoda and Progoneata, it probably represents a synapomorphy for the Myriapoda, although this conclusion is dependent on the phylogenetic position of symphylans and pauropods 232425. This finding further supports myriapod monophyly, which is to date mainly based on nucleotide sequence data (2627 morphological data are still controversial in this context 10252829.

The fact that Ubx and aUbx are expressed in conserved and complex complementary patterns strongly suggests that one (or its transcription) is involved in the regulation of the other. Striking similarities to the situation in myriapods can be found in Drosophila, in which transcription of bxd non-coding RNAs (ncRNAs) upstream of Ubx prevents transcription of the latter. This repression is probably caused by transcriptional interference as the bxd transcript(s) elongate into the region of Ubx promoters and prevent the binding of the transcription machinery 430. As a result, bxd ncRNAs are expressed in a complementary pattern to that of Ubx, causing a mosaic-type expression pattern of Ubx within its overall expression domain 46

A similar situation is found in myriapods, in which a putative ncRNA, aUbx, is expressed in a complementary pattern to that of Ubx. Like the bxd ncRNAs in Drosophila, aUbx also precedes expression of Ubx, and also as in Drosophila, expression of Ubx in myriapods occurs in the anterior of each segment and expression of bxd and aUbx occur in the posterior of each segment (this study, 931).

The most obvious difference between the expression of bxd ncRNAs in Drosophila and aUbx in myriapods is that aUbx (or its promoter) is located on the complementary DNA strand in myriapods and not oriented in a tandem position to Ubx on the same strand. How can this disparity be explained if we assume that aUbx expression in myriapods is homologous to bxd expression in Drosophila?

The simplest explanation of this pattern would be to postulate an inversion event in the Ubx locus back in the stem lineage leading to the myriapods, placing the aUbx (bxd) promoter on the complementary strand (Figure 6A). Subsequent transcription through the promoter site(s) of Ubx in myriapods would then cause expression of aUbx in a complementary pattern. However, this would require a stage at which Antp and Ubx were on different strands, and as we show in this paper, bicistronic transcripts of Ubx/Antp and their splice versions (variants I and II) are conserved and thus are most probably of strong developmental importance, thus they are unlikely to have been separated in this way. A single inversion event putting Ubx alone on the complementary strand can also be excluded because of the presence of Ubx/Antp bicistronic transcripts that are very unlikely to be a result of a trans-splicing event (discussed below) 932. An alternative to these unlikely possibilities is hat a new bxd/aUbx promoter site evolved on the complementary strand located between Antp and Ubx (Figure 6B). Functional studies or a fully sequenced genome, which could possibly help shed light on the role of aUbx transcription in myriapods and answer the question of whether the mechanisms suggested for Ubx regulation in myriapods are related to those in Drosophila, are currently not available.

A number of theories have been suggested over the past few years to explain how noncoding antisense transcripts or bidirectional transcription may regulate the expression of the coding unit (33 and references therein). A case of possible transcriptional interference displaying much similarity between Drosophila and myriapods has been discussed in the previous section. However, although this possibility appears to be likely, aUbx or its transcription could nevertheless also act differently. We therefore summarize and discuss some of those mechanisms in the light of our data.

First, transcription of the antisense strand can cause epigenetic modifications, methylation of sense-strand promoters, and conversion of the chromosome structure, causing repression of gene transcription on the sense strand 34. Epigenetic modification could explain or cause the complementary pattern of Ubx and aUbx if aUbx represses the transcription of Ubx in tissues or cells that are generally Ubx-competent.

Second, transcriptional interference can also occur via promoter collision, when RNA polymerases meet on opposite strands and cannot pass each other. This can cause the premature termination of one or both transcripts 3035.

Third, sense and antisense transcripts could form double-stranded (ds)RNA, a source for small interfering RNAs that would mediate RNA interference (RNAi) 36. The complementary expression pattern of Ubx and aUbx would be explainable by the rapid degeneration of Ubx due to perfectly matching miRNAs descendent from the possible Ubx-aUbx dsRNA 37.

The fact that aUbx is expressed significantly earlier than Ubx may also have important implications on the regulatory mechanisms discussed. It would guarantee the immediate binding of incorrectly expressed Ubx to pre-existing aUbx in an RNAi-based mechanism, or provide a head start for transcription of aUbx in cases of transcriptional interference. In the case of epigenetic modification, it would prevent the later transcription of Ubx by silencing its promoter(s).

We discovered a repetitive sequence of exactly 21 bp (Figure 1D) in the 3' UTR of Ubx. This sequence most probably represents a minisatellite (or short sequence repeat; SSR) common in bacterial and metazoan genomes 38. It may represent multiple recognition sites for micro (mi)RNAs 39. Alternatively, it could represent an ORF encoding a small 113 amino acid protein, possibly involved in the regulation of Ubx. The finding of an SSR could generally also be of interest for investigating population genetics in Glomeris 40.

The finding that bicistronic transcripts of Ubx and Antp (Ubx/Antp) are present in myriapods and crustaceans suggests that this represents a conserved state of at least the Mandibulata or potentially the Arthropoda. Despite this, we failed to isolate Ubx/Antp fusion transcripts from the beetle T. castaneum. The latter may merely represent a loss in higher insects that finally allowed the Hox complex to split between Ubx and Antp, as is the case in Drosophila melanogaster 41; however, in Tribolium, the Hox cluster is still intact 8. Alternatively, it may represent the early loss of Ubx/Antp in the stem lineage of the insects or hexapods. If the hexapods have evolved from a crustacean ancestor (as in the Pancrustacea theory), a loss of Ubx/Antp may be present in the suggested recent sister-group crustacean orders Remipedia and/or Cephalocarida 42. The presence of Ubx/Antp fusion transcripts in an onychophoran shows that the evolutionary origin of bicistronic transcription of Ubx and Antp dates back to the common ancestor of onychophorans and euarthropods, suggesting that Ubx/Antp is also likely to occur in chelicerates.

Interestingly, only the short splice variant II (Figure 1B,B') has been isolated from myriapods. We therefore believe that variant I may be lacking in myriapods exclusively, again supporting myriapod monophyly. However, we are aware that negative results are less reliable arguments than positive results, and therefore we can only see the lack of splice variant I in myriapods as minor evidence for monophyletic Myriapoda.

The presence of the Ubx/Antp splice variant II in onychophorans, crustaceans and myriapods argues against a mere genomic rearrangement in a population of Ubx as suggested for the centipede Strigamia 9, but rather suggests an important and conserved role in Hox gene regulation across the Arthropoda.

In crustaceans, bicistronic transcripts of Ubx/Antp are not (Daphnia) or only partially (only Ubx in Artemia) translated. Expression of the translated monocistronic transcripts, and therefore the protein, differs significantly from expression of Ubx/Antp 11. It is tempting to speculate that transcription of Ubx/Antp under control of the Ubx promoter interferes with the proper transcription of monocistronic Antp in these crustaceans.

The conserved appearance of Ubx/Antp in arthropods and onychophorans suggests their involvement in the regulation of Ubx, Antp or both Hox genes. In particular, repression of Antp via Ubx/Antp transcription appears likely, not least because the transcript is apparently spliced in such a way that it lacks most of its coding capacity (variant II).

For Glomeris and Euperipatoides, it is unclear whether the detected expression patterns of Ubx and Antp are a result of mono-or bicistronic transcription. However, in both, as in crustaceans 11, the Ubx/Antp transcript is probably under control of the Ubx promoter, as the expression pattern of Ubx/Antp is identical with that of Ubx (not shown). Thus, it is possible Ubx/Antp contributes to or even replaces monocistronic Ubx expression in Glomeris and Euperipatoides as it does in Artemia 11. If part of the detected mRNA expression patterns of Ubx and Antp 19 is a result of Ubx/Antp, it might not correlate with the protein pattern. Specific antibodies to detect UBX and ANTP protein are not available, and the crossreacting antibody FP6.87 43 does not detect UBX in Glomeris (data not shown). Further investigation is thus needed to unravel the role of Ubx/Antp transcription in arthropods.

Ubx expression is likely to be involved in the delayed outgrowth of the walking legs posterior to T3 in Glomeris by repressing Distal-less (Dll) as shown for other arthropods 444546. The finding that aUbx, a possible repressor of Ubx (as discussed above), is strongly expressed in the tips of the legs in T2 and T3 further supports this view, suggesting that the absence of Ubx is indeed crucial for the accelerated development of walking legs in T1 to T3 in Glomeris 19 . The exclusion of Ubx from the distal part of the legs possibly caused or supported by aUbx could represent a developmental novelty in the 'battle' of appendage growth in Ubx-expressing segments. In Strigamia and Lithobius, Ubx seems not to repress Dll, possibly because of a number of phosphorylation sites in the C-terminal end of the protein that interfere with the assumed repressor function of Ubx on Dll 1945. Consequently, there is no need to keep the tips of the legs free from Ubx or, in other words, to express aUbx.