Drosophila multiplexin (Dmp) modulates motor axon pathfinding accuracy

Authors


  • Author contributions: Experiments were designed by FM and BM and carried out by FM. Manuscript was prepared by FM and BM.

*Author to whom all correspondence should be addressed.
Email: bernard.moussian@tuebingen.mpg.de

Abstract

Multiplexins are multidomain collagens typically composed of an N-terminal thrombospondin-related domain, an interrupted triple helix and a C-terminal endostatin domain. They feature a clear regulatory function in the development of different tissues, which is chiefly conveyed by the endostatin domain. This domain can be found in proteolytically released monomeric and trimeric versions, and their diverse and opposed effects on the migratory behavior of epithelial and endothelial cell types have been demonstrated in cell culture experiments. The only Drosophila multiplexin displays specific features of both vertebrate multiplexins, collagens XV and XVIII. We characterized the Drosophila multiplexin (dmp) gene and found that three main isoforms are expressed from it, one of which is the monomeric endostatin version. Generation of dmp deletion alleles revealed that Dmp plays a role in motor axon pathfinding, as the mutants exhibit ventral bypass defects of the intersegmental nerve b (ISNb) similar to other motor axon guidance mutants. Transgenic overexpression of monomeric endostatin as well as of full-length Dmp, but not trimeric endostatin, were able to rescue these defects. In contrast, trimeric endostatin increased axon pathfinding accuracy in wild type background. We conclude that Dmp plays a modulating role in motor axon pathfinding and may be part of a buffering system that functions to avoid innervation errors.

Introduction

The extracellular environment provides numerous cues that direct the behavior of many migratory cell types, including pathfinding decisions of neuronal axons and dendrites. While being exposed to multiple guidance signals at the same time, a growth cone has to integrate these signals, and a guidance decision thus results from the interplay of multiple attractive and repulsive signals of different strength. During the last two decades, a large number of extracellular proteins that act in the process of motor axon pathfinding have been identified and functionally characterized in several vertebrate and invertebrate model systems. In the limelight of genetic and biochemical studies were almost exclusively transmembrane molecules of adhesive or signaling function and their diffusible or membrane-bound ligands, like Robo and Slit, semaphorins and their plexin-containing receptor complexes, netrins and deleted in colorectal cancer (DCC)/frazzled receptors, ephrins and their receptors, fasciclinII/neuronal cell adhesion molecule (NCAM) and DsCam (Araujo and Tear 2003; Chilton 2006). Other components of the extracellular matrix (ECM) have received much less attention.

Abundant ECM constituents are glycosaminoglycans (GAGs); especially heparan sulfates (HS) and chondroitin sulfates (CS) that are covalently linked to a core protein. In the nervous system, both HS and CS proteoglycans (HSPGs and CSPGs) have been implicated in various aspects of development, including axonal pathfinding (Bandtlow and Zimmermann 2000; Carulli et al. 2005). Additional sulfation, epimerization and deacetylation of HS and CS side chains can create microdomains of high specificity for certain binding partners, thus adding a high degree of functional specificity to their core protein (Bulow and Hobert 2004). To date, it has been shown that a loss of enzymes involved in HS biosynthesis and modification causes axon guidance defects (Bulow and Hobert 2004; Lee et al. 2004) and that the guidance function of the Robo/Slit pathway requires HS (Bulow and Hobert 2004).

In Drosophila, four classes of HSPG core proteins are present. The membrane-bound ECM components syndecan (Sdc) and, to a lesser extent, the glypican Dally-like (Dlp) have been shown to function in axonal guidance at the ventral midline and in the visual system via the action of the Slit/Robo signaling system, but seem to play a minor role in the motor axon guidance system (Johnson et al. 2004; Steigemann et al. 2004; Fox and Zinn 2005; Rawson et al. 2005). Of the two basement membrane HSPG core proteins encoded in the Drosophila genome, only perlecan has been studied so far, and found to be involved in neuromuscular junction (NMJ) development (Arikawa-Hirasawa et al. 2002; Voigt et al. 2002), but it has not been implicated in axonal guidance. The only potential Drosophila HSPG core protein entirely uncharacterized to date is a homologue of vertebrate collagens XV and XVIII.

Collagens XV and XVIII are the only two members of a vertebrate collagen family termed multiplexins (Oh et al. 1994). While collagen XVIII is an HSPG, collagen XV is modified with CS side chains (Li et al. 2000). Both molecules are characterized by a collagen triple helix region that is interrupted by several non-collagenous stretches, as well as a C-terminal non-triple helical NC1 domain (Oh et al. 1994; Rehn and Pihlajaniemi 1995). The NC1 domain is composed of three functionally distinct regions: an association region in its most N-terminal part that trimerizes NC1 monomers, a protease-sensitive hinge region, and the C-terminal globular endostatin (ES) domain (Sasaki et al. 1998).

Biochemical work on vertebrate collagen XVIII has identified a number of proteases capable of releasing monomeric ES (Wen et al. 1999; Felbor et al. 2000; Ferreras et al. 2000; Lin et al. 2001; Heljasvaara et al. 2005) and proteolytic release has also been shown to be necessary to activate biological ES function in vitro (Heljasvaara et al. 2005). Free monomeric ES domains occur in vivo and have been extensively studied as inhibitors of endothelial cell migration and thus as angiogenesis inhibitors (O’Reilly et al. 1997; Sasaki et al. 1999; Yamaguchi et al. 1999; Rehn et al. 2001; Abdollahi et al. 2004; Hurskainen et al. 2005; Marneros et al. 2007). It has also become clear in several experimental systems that the association of two or three NC1 domains, forming dimeric or trimeric ES, changes the effects of ES: In tissue cultures of different endothelial and other cell types, ES dimers or trimers activate migratory cell behavior, whereas free ES monomers produce an inhibitory effect even in the presence of multimeric ES (Kuo et al. 2001; Clamp et al. 2006). Similarly, in Caenorhabditis elegans lacking the NC1-encoding region of the multiplexin gene cle-1, migration of several cell types is defective, including that of mechanosensory neurons. This phenotype can be restored to wild type by overexpressing NC1, but not by ES, and ES overexpression in wild type impairs mechanosensory neuron migration (Ackley et al. 2001).

The Drosophila neuromuscular system is a very powerful and extensively used genetic model for studying axonal pathfinding. Neuromuscular connectivity gets established during embryogenesis, starting with the outgrowth of pioneer axons from the central nervous system towards the regions where the respective target muscles develop. Ensuing events are guidance of the growth cone towards its target region, target recognition and synapse formation (Tessier-Lavigne and Goodman 1996). We believe that the results summarized above suggest a role for the collagen XV/XVIII homologue in this process. To the end of characterizing this role of the protein and its subdomains, we generated both mutant and overexpression fly strains, and assayed them for motor nerve misrouting phenotypes. Our observations indicate that Dmp is involved in axonal navigation decisions, and possibly by more than one mechanism. We also analyzed the gene, its expression and products on the sequence level.

Materials and methods

Molecular biology

RNA was isolated from pools of stage 1–17 w1118 embryos using the RNeasy tissue kit (Qiagen). After oligo-dT-primed reverse transcription, dmp transcripts were polymerase chain reaction (PCR)-amplified, cloned into TOPO pCRII and sequenced using the BigDye Terminator kit (PE Applied Biosystems). Transcript 14 indicated in Figure 1(B) is BDGP clone GH14382. Sequences of Transcripts 1–13 were submitted to GenBank (Accession numbers EU523228–EU523240). Primer sequences are available in Supplement 1.

Figure 1.

Dmp gene, transcript and protein structures. Colour codes refer to protein domains as depicted in (C) and their respective coding sequences. ?(A) Genomic structure of the dmp gene. Exons are numbered in colours according to (B) and (C). Open reading frames of CG8647 and CG33171 are indicated as currently annotated in FlyBase. Triangles refer to PiggyBac insertion sites; they were used to generate dmp deletion alleles as indicated.? (B) Summary of splicing patterns observed from 14 different transcripts that were fully or partly sequenced. Within these 14 transcripts, eight different splice variants occur, some of which are varied only by the in- or exclusion of very short exons. Exons are numbered as in (A). Sequences of transcripts 1–13 have been submitted to GenBank (Accession nos. EU523228-EU523240). Transcript 14 represents the RH14382 cDNA clone from BDGP. For a detailed description of the alternative splicing events, refer to Supplement 1.? (C) Domain structure of the three major protein isoforms produced from the dmp locus. Left: Primary structures drawn to scale. Right: Schematic quartenary structures. The longest isoform DmpTSPN3hNC1 contains a domain with similarity to the N-terminal domain of thrombospondin (TSPN), which is absent from the middle isoform Dmp3hNC1. The shortest isoform DmpES is a single monomeric ES domain.

In situ hybridization

Digoxygenin (DIG)-labeled RNA probes were generated using the following templates: an AflIII-digest of BDGP-clone GH14382 (ES antisense), a cDNA clone containing dmp exons 1 and 2 generated by ourselves (TSP antisense), and a PCR product of exon 3 using GH14382 as template and a reverse primer containing the sequence of the T7 promoter (exon 3 antisense). In situ hybridizations were carried out following the standard protocol (Tautz and Pfeifle 1989) and sense probes used for control purposes.

Genetics and fly stocks

Strains carrying deletions in the dmp locus were generated by recombination between piggyBac transposable elements inserted within the dmp locus, as described by Parks et al. (2004). This method includes crossing together two different FLP recombination target (FRT)-bearing P-element or piggyBac insertion lines and a heat shock inducible FLP recombinase. Larvae were heat shocked at 37°C for 1 h daily on five consecutive days. As a result, it was possible to recover chromosomes where a recombination has taken place between the two FRT-bearing insertions, thus containing a deletion as well as a residual P-element or piggyBac insertion. For the dmpΔN4–11 strain, PBac{WH}CG33171f07253 and PBac{WH}CG33171f03008 were recombined, while PBac{WH}CG33171f03008 and PBac{PB}c06608a were recombined for the dmpΔC12–25 strain. PiggyBac insertion lines used were generated by Thibault et al. (2004) and obtained from Bloomington Drosophila Stock Center (PBac{WH}CG33171f07253, http://flystocks.bio.indiana.edu/) and from the Exelixis collection at Harvard (PBac{WH}CG33171f03008 and PBac{PB}c06608a, http://drosophila.med.harvard.edu/). Successful deletions were identified by PCR screening for the correct intersections between piggyBac element and genomic DNA, as well as by carrying out long-range PCRs spanning the whole residing piggyBac element. Note that we successfully used piggyBac elements of the PB type, which were not used in the work published by Parks et al. (2004).

For overexpression constructs, respective dmp fragments were PCR-amplified from GH14382 and cloned into pUAST (Brand and Perrimon 1993). In this vector, coding sequences were placed under the control of the so-called upstream activating sequence (UAS), so that in vivo their expression could be controlled by the Gal4 transcriptional activator binding to its UAS target site. The construct of the middle isoform Dmp3hNC1 contained the endogenous extracellular localization signal, while for the NC1 and ES constructs the signal sequence of wingless (Zecca et al. 1996) was PCR-amplified with primers introducing EcoRI restriction sites. These were then used to clone the signal sequence to the 5′ end of NC1 and ES coding sequences, to which an EcoRI site had also been introduced by using respective PCR primers. Transgenic lines were generated by construct injection into w1118 fertilized eggs using standard methods.

Wild-type flies used in all experiments were w1118, which were also the background used for generating the rescue lines. The Gal4 driver line used for overexpressing the different Dmp fragments was 24B-Gal4 (Brand and Perrimon 1993; Fyrberg et al. 1997). Rescue crosses were carried out by crossing stocks of 24B-Gal4/24BGal4; dmpΔC12–25/dmpΔC12–25 genotypes to UAS-x/UAS-x; dmpΔC12–25/dmpΔC12–25 flies.

Immunohistochemistry and embryo staging

The nervous system of late embryos was stained as described (Meyer and Aberle 2006) using a murine fasciclinII (FasII) antibody (1D4, DSHB). Embryos were dissected on microscope slides and imaged with an LSM510 confocal microscope (Zeiss) using the Cy3- and DIC channels. Signals from the Cy3 channel were converted to black to enhance their contrast in overlays with DIC images.

Embryos for dissection (stage 16/17) were selected based on the FasII signal under a UV light dissection scope. Since the ventral nerve cord (VNC) contracts during development, the maturity of the nervous system can be inferred from VNC length. As a second step, each resulting embryo filet was only included into statistical analysis if in some of its segments the ISNb had already invaded the ventral muscle field, making sure that the developmental time point for the ISNb steering decision was already past.

Quantification of axon guidance phenotypes and statistical analysis

Innervation phenotypes were quantified in abdominal segments A2–A7 of dissected FasII-stained embryos. Upon phenotypic evaluation, genotypic identity of all embryos was masked to avoid bias.

Statistical significance of error rate differences between different lines was evaluated using a χ2 test for independence or Fisher's exact test. Fisher's exact test is designed to deal with rare events. Comparisons between genotypes were carried out pairwise, one partner of the pair being the wild type or dmpΔC12–25, depending on the phenotype of the pair's second partner: A mutant phenotype was compared to the wild type, while a rescue phenotype was compared to dmpΔC12–25. Theoretically expected frequencies of correct and defective hemisegments were calculated using a contingency table. Values used were the absolute numbers of correct and defective hemisegments observed for each genotype and nerve. If any expectancy value in the table was below 10, Fisher's exact test was applied, which complies with the more rigorous of statistic standards (Becker and Genschel 2005; Bortz 2005). This was the case for most data presented here, exceptions being Figure 4(D) and all genotypes of Figure 5(A) except for 24B-Gal4xUAS-ES; dmpΔC12–25. Here, all expectancy values were above 10 and therefore the χ2 test for independence was used. p-values were calculated by JUMBO (Java-using Münster Biometry Online-System, http://imib.uni-muenster.de/fileadmin/template/conf/imib/lehre/skripte/biomathe/jumbo.html) using Java-Applet ‘4.8 Vierfeldertafel’.

Figure 4.

Motor axon guidance phenotypes caused by deleting the parts of the anterior (dmpΔN4–11) or the posterior (dmpΔC12–25) region of the dmp gene locus.? (A) Schematic of the larval hemisegmental set of muscles and the five motor nerves that innervate them. The target muscle fields of the five different nerves are depicted in different colours and the names of the respective nerves are given on the right. Phenotypes studied in this work concern the ISNb, that innervates one of the three ventral muscle fields (yellow) and the ISN, innervating the dorsalmost muscle field (green).? (B–D) ISNb wild type and deletion phenotypes. (B) Ventral region of two wild type hemisegments, displaying the characteristic nervous structures that innervate muscles 12, 13, 6 and 7, formed by the ISNb (arrow in left hemisegment). (C) Ventral region of three dmpΔC12–25 hemisegments. In the left hemisegment, the ISNb behaves similar to wild type. In the middle segment, it braches off the ISN at the ventral choice point, but instead of invading the ventral muscle field, it steers back to the ISN and fuses with it for part of its length (arrow), but detaches again later (small arrowhead). In the right hemisegment, the ISNb also detaches from the other nerves at the ventral choice point, but then steers to the anterior, even crossing the segment boundary and invading the anterior segment (large arrowhead). This may be an attempt to substitute for the missing ventral innervation in the anterior segment. ?(E–G) ISN wild type and deletion phenotypes. (E) Dorsal region of three wild type hemisegments and (F) dorsal region of three dmpΔC12–25 hemisegments. In contrast to wild type, mutant ISN nerves bifurcate and cross segment boundaries. ?(D, G) Statistical evaluation of the incidence of ISNb (D) and ISN (G) guidance mistakes in wild type, dmpΔC12–25, and dmpΔN4–11. Deletion of the 3’ portion of the dmp gene, encoding the NC1 region, leads to a significant increase in ISNb error frequency, while deletion of the 5’ region, starting with exon 4, leads to a less penetrant phenotype. These tendencies are identical for the ISN, although statistical significance is on a lower level, due to the lower absolute error number.? Absolute numbers of hemisegments evaluated are given at the base of each bar. Figures above the bars are p-values calculated by the χ2 test for independence for (D) and by Fisher's exact test for (G).

Figure 5.

Overexpression effects of different Dmp fragments in dmpΔC12–25 and wild type background.? (A, B) Effect on ISNb guidance accuracy in mutant (A) and wild type (B) background. A rescue effect of the mutant phenotype is evident for the longest construct UAS-3hNC1 and for the ES domain alone. With UAS-NC1, there is no rescue activity, since the error rate is still significantly different from that in wild type. In a wild type background, UAS-NC1 significantly reduces the wild type error rate. UAS-ES has a disruptive effect on the general morphology of some embryos which get not fully reflected by motor nerve error statistics (see text and supplementary data).? (C, D) Effect on ISN guidance accuracy in mutant (C) and wild type (D) background. In contrast to the ISNb, UAS-3hNC1 does not exert a rescue effect on the ISN deletion phenotype. In contrast, the error rate reduction observed in the 24BxUAS-ES; dmpΔC12–25 genotype statistically significant (p = 0.045). In wild type background, there is a clear disruptive effect of UAS-ES on the ISN (see above and text). At the base of each column, absolute numbers of hemisegments evaluated are given; figures above columns indicate p-values. The χ2 test for independence was applied to all genotypes in (A) except for 24B-Gal4xUAS-ES;dmpΔC12–25. Here, and to all data represented in (B), (C), and (D), Fisher's exact test was applied.

Results

CG8647 and CG33171 are one gene from which various isoforms of Dmp are expressed

CG33171 is one of only three conserved collagen genes annotated in the Drosophila genome and the only one that encodes an endostatin domain. The ES domain is positioned at the protein's C-terminus and qualifies it as a multiplexin. But in contrast to vertebrate and C. elegans multiplexin genes (Kivirikko et al. 1994; Muragaki et al. 1994; Muragaki et al. 1995; Rehn and Pihlajaniemi 1995; Ackley et al. 2001), CG33171 lacks the coding sequence for any N-terminal globular domain. This annotation is supported by 14 BDGP cDNA clones, one of which is full-length, and three more of which confirm this N-terminus. CG8647 is predicted as the next gene upstream of CG33171, and its two exons encode a moiety related to the N-terminal domain of thrombospondin (TSPN), as found at the N-terminus of vertebrate and C. elegans multiplexins. This suggests that the two annotated genes CG8647 and CG33171 really are only one gene from which several different transcripts are produced, including a fusion transcript of the two predicted loci. Indeed, we were able to amplify respective cDNAs from the pooled stage 1–17 wild-type embryos and found that a variety of alternative transcripts are produced from CG8647 and CG33171, which we now refer to as the Drosophila multiplexin (dmp) gene.

We found transcripts having the two exons of CG8647, now referred to as exons 1 and 2 of dmp, as their 5′ ends, which then skip the first two exons of CG33171 (now exons 3 and 4 of dmp), as well as transcripts beginning with exons 3 and 4 (see Figure 1A,B). Hence, one of the alternative transcripts will produce a protein carrying a TSPN domain at its N-terminus, the other will result in a multiplexin molecule starting with the collagen triple helix domain; we call the two DmpTSPN3hNC1 and Dmp3hNC1, respectively (see Fig. 1C).

We also detected one alternative transcript where exon 4 is directly joined to the first exon encoding the C-terminal endostatin domain. So the protein product of this transcript will be a single globular ES domain (Fig. 1C), which is a multiplexin isoform that has not been described previously. We refer to it as DmpES.

We additionally found a number of splicing variations that affect only short sequence stretches (see Fig. 1B). In summary, the alternative splicing events we observed fall into three classes: There are two alternative 5′ ends, each of which is made of two exons, there are seven exons (exons 5, 6, 7, 8, 12, 20, and 21) that individually participate in cassette-type alternative splicing, and there are three alternative 5′- or 3′ splice sites, within exons 2, 9, and 10.

Drosophila Dmp shares characteristic features with both vertebrate collagen XV and XVIII

Sequence alignment illustrates the relatively distant relatedness between TSPN domains of the different multiplexins and the N-terminal domain of human TSP-1 (TSPN-1) on the primary structure level (Fig. 2A). Sequence identity is around 25% between multiplexin TSPN sequence stretches, depending on the alignment program used. Identity between multiplexin TSPN domains and TSPN-1 is well below 20%, but secondary structure prediction using the homology detection tool HHpred (Soding et al. 2005) suggests Dmp TSPN adopts the same structure as shown for TSPN-1 (Tan et al. 2006), as the prediction shows the 13 beta strands that form the beta sandwich structure of human TSPN-1. The two cysteine residues that form an intradomain disulfide bond are also conserved, while the arginine and lysine residues involved in heparin binding in TSPN-1 (Tan et al. 2006) are only partially conserved (Fig. 2A).

Figure 2.

Protein sequence alignment of Dmp domains and their C. elegans and human homologues.? (A) TSPN domains of Dmp, human collagen XVIII and XV, C. elegans Cle-1, and human thrombospondin-1. Sequence identity on the amino acid level is low (around 20–25%). However, according to HHpred structure prediction, Dmp-TSPN has the same topology as reported for TSPN-1 (Tan et al. 2006). Here, Dmp-TSPN and TSPN-1 β-sheets (dark blue) and α-helices (red) as described by these two sources are indicated above and below the two sequences.? Yellow: Two cysteine residues that stabilize the typical TSPN fold by forming a disulphide bond are conserved. Light blue: Basic residues that contribute to the major heparin binding site in TSPN-1 (Tan et al. 2006) are not conserved.? (B) C-terminal NC1 domain of Dmp, human collagen XVIII and XV, and C. elegans Cle-1. Within the ES domain, identity between fly and human sequences is around 50%. *: 4-residue-loop that distinguishes collagen XV and XVIII. Light blue: Arginine residues that make important contributions to heparin binding in collagen XVIII (Sasaki et al. 1999). Green: zinc coordination residues (Sasaki et al. 2000). Purple: Putative integrin-binding RGD motif. Red and yellow: residues encoded by alternative exons 20 and 21 (see Fig. 1). They are omitted in the majority of transcripts we recovered, and their sequence provides no cue about their function.

There are further alternative N-terminal domains found in nematode and vertebrate collagen XVIII. In Drosophila, the next predicted gene upstream of dmp encodes the transcription factor Biniou, and the intergenic sequence stretch of 17.8kb has been subjected to extensive use of gene prediction programs and BLAST searching. No additional putative exons could be identified. We therefore expect that the TSPN domain is the only N-terminal globular domain occurring in Dmp. This is a feature that the Drosophila multiplexin shares with vertebrate collagen XV (Kivirikko et al. 1994; Muragaki et al. 1994).

For the C-terminal NC1 domains of type XV and XVIII collagens of different organisms, homology is much higher already on sequence level (Fig. 2B). The Drosophila ES domain is 52% and 49% identical to that of human collagen XV and XVIII in a conserved core region of 164 residues, respectively. Of the two vertebrate multiplexins, only collagen XVIII carries the N-terminal Zn2+ coordination motif, and collagen XVIII also possesses a longer hinge region compared to collagen XV, as well as a four-residue loop that is thought to be important for heparin binding (Sasaki et al. 1999). Sequence alignment with the Drosophila and C. elegans homologues shows that for the first two features, both of the fly and worm multiplexins are more similar to vertebrate collagen XV. However, both the fly and worm multiplexins also possess the four-residue loop within the ES domain. Hence, the single multiplexin gene present in the fly genome exhibits characteristic features of both of the two vertebrate multiplexin types, and we therefore refrain from categorizing it as any of the two, but continue to refer to it as Drosophila multiplexin (Dmp).

Multiplexins are qualified as collagens by a region of triple helix repeats that is interrupted several times by a non-triple helix sequence. In Dmp this region spans about 290 amino acids and is thus considerably shorter than in human collagen XV and XVIII, where it extends over 570 and 690 amino acids, respectively. In contrast, the two type IV structural collagens of Drosophila possess triple helix regions of 1900 and 1560 amino acids. So the structural domain of Dmp is small both in relation to the rest of the molecule and to the same domain in other molecules, and this suggests that Dmp does not play an important role as a structural molecule.

Collagen XVIII has been shown to be a heparan sulfate proteoglycan (Halfter et al. 1998), while collagen XV is modified with chondroitin sulfate side chains (Li et al. 2000). Both types of modification occur via a Ser-Gly consensus sequence (Whitelock and Iozzo 2005) with acidic residues nearby, and a couple of such motifs are present in the Dmp core protein. The first of them is situated in the last part of the TSPN domain, encoded by one of the variably spliced exons. Additional putative attachment sites are present in the remainder of the molecule, with one of them being situated within the ES domain, giving ample potential for posttranslational glycosylation.

Transcription of dmp begins in late stages of embryonic development

We carried out in situ hybridization experiments using antisense probes for different characteristic regions of dmp transcripts in order to shed light on the different gene products’ possible spatial and temporal requirement during development.

Using a probe against the region encoding the TSPN domain of the DmpTSPN3hNC1 isoform produces weak signals at embryonic stage 16 from head skeleton, filzkörper, dorsal vessel, and central nervous system (CNS) (Fig. 3A). All earlier embryonic stages do not show any TSPN expression. However, the TSPN probe yields a very weak signal, possibly reflecting a low expression level of the TSPN domain. In a ventral view of a stage 16 embryo, a symmetric, dotted and segmentally repeated pattern within the ventral region of the VNC is discernible (Fig. 3B), where each dot presumably corresponds to an individual cell. The identity of these cells is unknown.

Figure 3.

 mRNA expression patterns from different regions of dmp. ?All embryos are stage 16, except for (F) and (K) that are early (F) and late (K) 17. (A, B) TSPN-like region antisense probe, (C-F, L) exon 3 antisense probe, (G-K, M) endostatin antisense probe. With all probes, we detected expression in head skeleton, filzkörper, dorsal vessel, and CNS beginning at stage 16 (A, C, G). All tested regions of the gene were expressed in single cells of unknown identity within the VNC (B, D, H). Dorsal vessel expression for the TSPN-like region was very weak, whereas for exon 3 and the ES region, strong signals of a segmentally interrupted fashion were detected for the heart and the aorta (E, I, L, M). CNS expression persisted into stage 17, but the cell-wise pattern observed at stage 16 was lost (F, K). Note that ISNb invasion of the ventral muscle field as well as ISN dorsal development occurs throughout embryonic stage 16, i. e. before the onset of strong expression inside the CNS. In the mutant, we did not observe any strong heart or CNS defects in spite of the gene's strong expression in these two tissues.

The expression of the 5′-UTR of the non-globular N-terminus of Dmp3hNC1 and DmpES, encoded by exon 3, has a tissue distribution similar to that seen for TSPN, (Fig. 3C), but the signal is considerably stronger. Expression onset is during stage 16 as well. In contrast to the TSPN probe, there is a detectable expression in the CNS of late stage 17 embryos and early larvae (Fig. 3F). In these stages, the cell-wise expression of stage 16 VNCs (Fig. 3D) is lost, and the remaining signal has its main focus in the dorsal VNC, where the neuropil is located. Expression in the dorsal vessel is strongest within its posterior part, also referred to as the heart. In strongly stained embryos, there is also an expression in the aorta, reaching as far anterior as segment A2 (Fig. 3E,L). Dorsal vessel expression occurs in an interrupted, segmented fashion, so it seems that only a subset of cardioblasts produces Dmp.

Using an ES antisense probe yields an equally strong signal as the exon 3 probe, and from the same tissues as the two 5′ probes: the dorsal vessel, the central nervous system, the head skeleton, and the filzkörper starting at stage 16 (Fig. 3G–I) and lasting until the end of embryogenesis. During stages 16 and 17, expression occurs in the whole central nervous system, and with similar expression peaks in single cells of the VNC as the TSPN and exon 3 region (Fig. 3G,H). Like with the exon 3 probe, ES expression in the CNS persists into early larval stages, but loses its single-cell pattern. Instead, CNS expression is strongest within the neuropil region (Fig. 3K). Also the segmented staining pattern from the dorsal vessel is identical to that observed with the exon 3 probe (Fig. 3I,M). These results indicate that there is no tissue-specific usage of the different Dmp isoforms.

Dmp partial deletion mutants are homozygous viable and the dmpΔN4–11 deletion does not abolish NC1 transcription and processing

Wanting to investigate the contribution of Dmp to Drosophila development in general and specifically to motor axon pathfinding, we generated two strains that carry partial deletions of the dmp gene by making use of FLP recombinase-induced recombination between FRT sites present in piggyBac transposons, as described by Parks et al. (2004). In the first strain, dmpΔN4–11, 21.26 kbp of genomic DNA were removed, that encode the second alternative start codon and 759 bp of the triple helix repeat region (exons 4 to 11, see Fig. 1A). The second strain, dmpΔC12–25, lacks a genomic stretch of 19.89 kbp, including the whole coding sequence for NC1 (exons 12 to 25, see Fig. 1A).

We found that both deletions are homozygous viable, causing no obvious defects apart from a general developmental retardation of homozygous mutant animals compared with their heterozygous siblings, since homozygous adults start eclosing several days later than their balancer-bearing siblings, but then appear at the expected ratio. So in spite of the strong expression of Dmp by cardioblasts, it is not vital for a functional heart. Nonetheless, it is possible that a lack of Dmp affects its morphogenesis or has other functional consequences for the dorsal vessel. Likewise, in spite of the dmp expression inside the CNS, we did not observe any malformation of the CNS in dmp mutants, neither in conventional light microscopy nor when using the FasII marker (see below) that labels peripheral motoneurons as well as six characteristic longitudinal fascicles inside the CNS.

We detected some residual transcription of the ES domain from the dmpΔN4–11 allele (see Supplement 2) and consequently, only dmpΔC12–25, but not dmpΔN4–11, can be considered as a functional null mutant with respect to the protein's C-terminus.

Deletion of different regions of the dmp gene causes axon guidance errors in ISNb and ISN nerve tracts

We then turned to a close examination of motor axon guidance accuracy in dmp mutants. Each of the abdominal hemisegments A2 to A7 of the Drosophila larva contains 30 different, individually identifiable muscles, which are innervated by five different nerves (Fig. 4A) that are FasII-positive. Branching sites of these nerves and the resulting innervation patterns are stereotypically reproduced in each wild type abdominal hemisegment. A ventral bypass phenotype can affect the ISNb as well as the other two nerves that innervate the ventral muscle field, the SNc and the ISNd. In these cases the affected nerves fail to detach from the main nerve tract and to invade the ventral muscle field. Instead, they continue to travel dorsally together with the main nerve tract, with the distance travelled varying between segments. Several molecules have already been recognized for the regulation of this process, and many of the molecules influencing ISNb guidance also affect pathfinding of other motor nerves, reflecting the complex interplay between different guidance cues and their differential effects on different cells. These include fasciclinII (FasII), Integrins, the Leukocyte antigen related (LAR) family of receptor protein tyrosine phosphatases (RPTPs), and the extracellular protease Tlr1 and matrix metalloproteases (Lin and Goodman 1994; Lin et al. 1994; Desai et al. 1997; Hoang and Chiba 1998; Sun et al. 2001; Meyer and Aberle 2006; Miller et al. 2008).

We analyzed stage 16 and 17 embryos of both dmp deletion strains, that were FasII-stained and fileted. Our quantitative mutant analysis was focused on one of the nerves innervating part of the ventral muscle region, the ISNb, and the nerve leading to the dorsalmost muscles, the ISN. All quantifications were carried out blindfolded with respect to the genotypic identity of each sample, and by a single person. The other nerves, the SNa, SNc and ISNd did not show any phenotype that could be reliably distinguished from wild type. To the eye it appeared that in dmpΔC12–25 mutants the third, most anterior SNa branch sometimes gave a stronger FasII signal than in wild type, where it was hardly visible in the embryo. This trait, however, could not be evaluated in quantitative terms.

Both the ISNb and the ISN showed an increased rate of guidance errors in homozygous mutants of both deletion strains. In the C-terminal deletion dmpΔC12–25, ISNb pathfinding errors, leading to a lack of innervation of the ventral muscle field, were found in 21.4% of all examined hemisegments (n = 210, derived from 41 embryo filets). In a wild type genetic background, similar ISNb errors were observed at a rate of around 7.1% (nsegment= 181, nembryo = 30, Fig. 4D), which is in accordance with rates described in other studies (Meyer and Aberle 2006; Yu et al. 2000; Miller et al. 2008). A χ2 test for independence qualifies this difference as clearly statistically significant (p = 0.00008). Hence, Dmp does play a role in ensuring motor axon guidance quality. For dmpΔN4–11, we observed an error rate of 14.6% (nsegment = 123, nembryo = 22, p = 0.035), so the removal of one of the alternative N-termini from the gene and the at least reduced dosage of the protein's C-terminal NC1 domain impair axon guidance accuracy, but less strongly than the complete loss of the endostatin region. In both deletion strains, ISNb pathfinding errors are of the same quality. They include full ISNb bypass relative to the ventral muscle field that it is due to innervate, so that, instead of steering off into the ventral muscle field, an additional nerve tract can be seen travelling along the ISN (Fig. 4C, middle segment), or the disposition of the ISNb axon bundle remains entirely obscure. A partial bypass behavior has also been observed, where very faint FasII-positive structures have left the main nerve tract and adhered to muscle surfaces within the ventral muscle field immediately adjacent to the ISN. In this case, it seems that the ISNb fails to branch off the main nerve tract, but individual axon tips still follow a guiding mechanism towards the muscle field. Another type of guidance error is an overgrowth of the ISNb or parts of it towards the transversal nerve (TN) of either the adjacent anterior or posterior hemisegment, sometimes including intrusion into an adjacent hemisegment (Fig. 4C, right segment) and/or fusion of the ISNb to the TN.

For the ISN, similar tendencies were observed as for the ISNb. In both deletion strains, the ISN error rate is about threefold higher than in the wild type: 5.8% in dmpΔC12–25 (with nsegment = 191, nembryo = 40) and 5.3% in dmpΔN4–11 (with nsegment = 114, nembryo = 22), compared with 1.7% (nsegment = 181, nembryo = 30, see Fig. 4G). For rare events, for example, those found in all experiments relating to the ISN, Fisher's exact test was used for assessing probabilities (see Materials and methods). This results in p = 0.054 for dmpΔC12–25. This formally means that the observed differences of ISN phenotypes between wild type and dmpΔC12–25 are not significant, but note that statistical significance is hard to achieve in this context due to the low absolute number of ISN errors, and a p-value of 0.054 is sufficiently small to let true differences between the genotypes seem possible. One of the ISN pathfinding mistakes we observed is its intrusion into an adjacent segment, leading to a fusion of ISN neighbors (Fig. 4F). This can result either from a premature deviation of the ISN from its route, or occur together with an ISN overgrowth, where the respective axonal growth cone(s) do not stop at the dorsalmost site of synapse formation, but continue growing perpendicular to the previous direction. We also observed bifurcation events of single ISNs, with either both new branches remaining in their original segment or the branches invading the neighboring segments. Taken together, the results from the two deletion strains strongly suggest a functional role for Dmp in ensuring motor axon guidance accuracy, and point at a specific relevance of the molecule's C-terminal NC1 domain in the process.

Different rescue activities on the dmpΔC12–25 guidance phenotype are exerted by different Dmp fragments

In vertebrate tissue culture it has been shown that monomeric and oligomeric (as produced by natural trimerization of NC1 domains) ES has converse effects on migratory behavior of endothelial and other cells, with oligomeric ES activating migration and monomeric ES having no impact (Kuo et al. 2001). Similarly, only NC1 constructs were able to restore defective migratory behavior of mechanosensory neurons in C. elegans (Ackley et al. 2001). We therefore wanted to test the effect of different Dmp fragments on motoaxon pathfinding in wild type and in our dmpΔC12–25 mutant.

The dmp gene is highly expressed in the mesodermal heart and in some yet unidentified cells within the VNC at the end of mid-embryogenesis. The driver line 24B-Gal4 (Brand and Perrimon 1993; Fyrberg et al. 1997) produces an expression pattern that is partially similar to that of dmp, as it drives expression in somatic mesoderm, heart and some cells within the VNC (see Supplement 3).

The basement membrane is a continuous ECM covering all organs of the embryo and has been shown to be formed by non-cell-autonomous distribution of ECM components. For example, the classic basement membrane component Collagen IV is expressed by hemocytes and fat body in Drosophila embryos (Mirre et al. 1992; Yasothornsrikul et al. 1997), but in spite of this is present in the basement membranes of all tissues, including muscles and nerves (Fessler and Fessler 1989). Mirre et al. (1992) show that Collagen IV protein is present inside the VNC. A good example of non-cell-autonomous action in motoaxon guidance is Tlr1, an extracellular protease, mutants of which exhibit a motoaxon pathfinding phenotype similar to that of Dmp. In tlr1 mutants, pathfinding defects can be rescued using Gal4 driver lines of various different tissue specificities, clearly indicating non-cell-autonomous action (Meyer and Aberle 2006). Therefore, we believed that a perfect match between endogenous dmp expression and Gal4 used to express transgenic Dmp is unlikely to be necessary for rescue, and hence did not hesitate to use 24B-Gal4 for the rescue crosses.

We produced UAS lines for overexpressing the endostatin domain (UAS-ES), NC1 domain (UAS-NC1) and Dmp3hNC1 (UAS-3hNC1) and assessing their potential rescue effect on the dmpΔC12–25 motor axon guidance phenotype. Upon systematic quantitative evaluation, we found a reduction of ISNb guidance inaccuracies to wild type level when overexpressing UAS-3hNC1 (nsegment = 79, nembryo = 15, p(dmpΔC12–25) = 0.0003) or UAS-ES (nsegment = 291, nembryo = 43, p(dmpΔC12–25) < 0.00001). By contrast, overexpression of UAS-NC1, that is expected to create ES trimers, had no effect on the dmpΔC12–25 phenotype (nsegment = 67, nembryo = 13, p = 0.002, Fig. 5A). So in this context, the monomeric version of ES is the active compound in conferring correct guidance instructions to axonal growth cones, whereas the NC1 version is inactive. This opens up the possibility that this active compound can be proteolytically generated from the Dmp3hNC1 molecule, but not from the UAS-NC1 product, that lacks the triple helix region.

In contrast to the ISNb, UAS-3hNC1 did not alter ISN error frequency (Fig. 5C), whereas both nerves experienced an error rate reduction to the wild type level by UAS-ES (for ISN: p = 0.045). This suggests that different fragments or isoforms of Dmp act through different mechanisms in ventral and dorsal regions of the embryo.

Different effects on wild type embryos are exerted by different Dmp fragments

We used the same UAS lines and the 24B-Gal4 driver line to overexpress the different Dmp fragments in a wild type background. Overexpression of UAS-3hNC1 did not have a significant impact on motoaxonal pathfinding, neither for the ISNb nor for the ISN nerve (Fig. 5B,D). Conversely, overexpression of the NC1 domain in wild type background reduced the number of ISNb pathfinding errors from 7.2% in wild type (nsegment = 181, nembryo = 30), that is typically observed in wild type (Yu et al. 2000; Meyer and Aberle 2006; Miller et al. 2008) to an extremely low error rate of 0.75% (nsegment = 134, nembryo = 19) that is significantly different from that in wild type (p = 0.005, Fig. 5B). For the ISN, a low error rate was found as well, but since ISN guidance mistakes are rare in wild type as well, there is no significance in this figure. For overexpression of the ES domain alone we found an impairment of ISNb and ISN pathfinding, and the resulting error rates we recovered are 14.6% (nsegment = 88, nembryo = 12) and 11.5% (nsegment = 78, nembryo = 11), respectively (p = 0.12 and 0.0012, Fig. 5B,D). These figures, however, do not reflect the full impact of ES overexpression on embryonic development, as a high proportion of individuals of this genotype showed a generally defective body morphology. It was the healthier fraction that was used for dissection and assessment of axon pathfinding, which is thus not representative of the genotype as a whole. Defective embryos possessed strands of FasII-positive cells that did not at all resemble a functional nervous system. Other embryos showed a recognizable, but disordered nervous system and a somewhat deformed general morphology (see Supplement 4). But on any account, it has become clear that a monomeric ES domain has a disadvantageous impact on axonal pathfinding, in contrast to trimeric ES, which is actually enhancing guidance accuracy.

Discussion

Axonal pathfinding is a complex process depending on the balance of various attractive and repulsive guidance cues that act on an individual growth cone. The C-terminal domains of vertebrate multiplexins have received ample attention as activators and inhibitors of endothelial and epithelial cell migration, and some evidence also exists for a role of multiplexins as modulators of neuronal migratory behavior (Ackley et al. 2001; Kliemann et al. 2003; Schneider and Granato 2006). Multiplexins are multidomain proteins, containing an N-terminal domain with homology to the N-terminal domain of thrombospondin (TSPN), an interrupted collagen triple helix domain and a C-terminal non-collagenous domain (NC1), that subdivides into a trimerization region, a protease-sensitive hinge region and a globular endostatin (ES) domain. One of the two vertebrate multiplexins, collagen XVIII, and C. elegans collagen XVIII occur in three different isoforms that can contain further N-terminal domains (Muragaki et al. 1995; Rehn and Pihlajaniemi 1995; Ackley et al. 2001). We found that the Drosophila melanogaster multiplexin (Dmp) is encoded by a complex locus, comprising CG8647 and CG33171, that encodes a TSPN domain (Tan et al. 2006), but no further N-terminal moieties.

We have generated two deletions within the dmp gene, and both deletion strains show pathfinding deficiencies of the same quality. ISNb growth cones fail to detach from the main nerve tract and steer into the ventral muscle field but continue to grow on dorsally. Additionally, ISNb as well as ISN nerves sometimes disrespect segment boundaries and fuse to nerves of adjacent segments. The two deletion strains differ by the incidence of pathfinding errors, as dmpΔC12–25, lacking the larger part of the gene, including part of the triple helix repeat region and the NC1-encoding region, shows greater penetrance than dmpΔN4–11, where the second alternative N-terminus and the following triple helical repeat regions are missing. We have shown that dmpΔN4–11 is not likely to be a null mutation since there is wild type-like transcript present from the NC1-encoding region of the gene. This situation suggests that a low dosage of a truncated Dmp version leads to an improved phenotype compared with dmpΔC12–25. However, an effect caused by the absence of the protein's N-terminal portion cannot be completely ruled out. For example, one can speculate about a function of a putative integrin-interacting RGD motif present therein.

The role of multiplexin is clearly distinct from that of the classic structural collagen IV

The spatially and temporally very restricted expression of dmp contrasts with the landmarks of the two collagen IV molecules, that are the only other conserved collagens in Drosophila and general important ECM constituents. In Drosophila, their expression begins in embryonic stage 12 (Mirre et al. 1988) or even earlier (Natzle et al. 1982) and hemocytes produce and disperse them all over the embryo while circulating in the hemolymph throughout the rest of embryonic development. During late embryogenesis, specific hemocytes also migrate along the ventral midline between the CNS and the epidermis, but these midline hemocytes are dispersed very asymmetrically (Wood et al. 2006), and we do not believe that any of the dmp expression we observed originates from these cells. Hence, Dmp production is largely different from collagen IV expression, since it is restricted to limited domains of immobile cells and occurs markedly later in development.

Both dmp deletion strains we generated are homozygous viable. This is in correlation with collagen XV and XVIII single knockouts in mice (Eklund et al. 2001; Fukai et al. 2002) and even with the double knockout (Ylikarppa et al. 2003), as well as the C. elegans cle-1 mutation (Ackley et al. 2001), none of which are lethal. However, the dmp gene and two type IV collagen genes (Natzle et al. 1982; Yasothornsrikul et al. 1997) are the only collagen genes in the Drosophila genome, whereas vertebrates and nematodes possess a far greater number of collagen types and genes (Myllyharju and Kivirikko 2004). This renders the dispensability of Dmp in flies even more surprising than in other organisms, and underlines that its role is very different from that of collagen IV.

Endostatin monomers harbor a specific activity in axon pathfinding, and their dosage is critical

It was possible to rescue the dmpΔC12–25 ISNb phenotype by overexpressing the Dmp3hNC1 isoform in dmpΔC12–25 mutant background, while its overexpression in wild type had no effect on axon pathfinding. This indicates that the protein product from the transgene functions in a fashion similar to the endogenous protein. Monomeric ES had a rescue effect on the dmpΔC12–25 ISNb phenotype that is indistinguishable from the effect of UAS-3hNC1, whereas UAS-NC1, which is expected to lead to the formation of trimeric ES, did not improve the phenotype. We conclude from this that a beneficial activity resides specifically within the monomeric ES version, which is necessary and sufficient for ensuring ISNb pathfinding accuracy. Biochemical work on vertebrate collagen XVIII has identified a number of proteases capable of releasing monomeric ES (Wen et al. 1999; Felbor et al. 2000; Ferreras et al. 2000; Lin et al. 2001; Heljasvaara et al. 2005) and proteolytic release has also been shown to be necessary to activate biological function (Heljasvaara et al. 2005). In our dmp mutants, a proteolytic release of monomeric ES might account for the rescue effect of UAS-3hNC1. This implies that ES release is not possible from UAS-NC1, since this construct does not improve the mutant phenotype. A possible explanation for this is that the triple helix region of the molecule is needed for specific enzyme-substrate recognition.

The overexpression effect of ES in wild type background is qualitatively different from its impact on the dmp mutant phenotype. While ES overexpression was largely beneficial in the mutant, it caused a variety of phenotypes in wild type background, ranging from a defective body morphology giving the impression that morphogenesis was hampered on a general level, to wild type looking individuals and such that appeared to exhibit a specific motor axon guidance phenotype. We suggest that the phenotypic differences between the two genotypes are due to an ES dosage effect. ES overexpression in dmp mutant background produces an ES quantity that can be fully absorbed by the standard interaction partners, whereas the ES excess in the wild type overexpression situation leads to an ES overcharge of standard or non-standard interaction partners that inhibits vital interactions with other molecules. In an alternative scenario, the detrimental effect of ES in wild type background would arise from an interaction between ectopic ES and the trunk of the Dmp molecule.

When overexpressing Dmp3hNC1, no detrimental effect is observed, so the full-length protein dosage increase over wild type level apparently does not pose a problem. This can be due to the endogenous mechanisms that balance proteolytic generation of monomeric ES, such as the characteristics of the proteases involved, which can include their spatial and temporal distribution, quantity, substrate affinity and the resulting turnover numbers.

NC1 does not function without endogenous Dmp and might differ in its action mechanism from monomeric ES

Transgenic expression of trimeric ES, as formed by the NC1 transgene product, and monomeric ES had exactly converse effects on the ISNb phenotypes of wild type and dmpΔC12–25 mutants. Free ES was able to rescue the pathfinding defects observed in dmpΔC12–25 and had a detrimental effect in wild type background. In contrast, the NC1 transgene, while having no impact in the mutant background, significantly reduced the wild type ISNb error rate of 7.2% to a virtually flawless ISNb guidance with 0.75% errors. These observations indicate that the function of NC1 depends on the presence of endogenous Dmp. In one scenario, endogenous Dmp plays a role in creating a microenvironment or architecture in the basement membrane that is permissive for NC1 activity. Conversely, a basement membrane without Dmp deteriorates in quality resulting in an inactive form of NC1. Alternatively, a direct interaction between Dmp or parts of it and NC1 may be required for NC1 function. Each feature of Dmp may potentially convey the relevant interactions, including the collagen triple helix region, the TSPN-1 domain, the NC1 portion, or several of these domains cooperate. Since both the Dmp triple helix region and the ES domain contain putative GAG attachment sites (Dong et al. 2003), options for functional moieties include heparan sulfate side chains.

Another question concerns the mode of action of NC1. One possibility is that it acts, if supported by the presence of functional Dmp, by releasing ES. In this scenario, NC1 simply is an inhibited version of ES. This option, however, is contradicted by the observation that neither overexpression of ES nor of 3hNC1 improved pathfinding in wild type background. An alternative idea is that ES monomers and NC1 trimers play distinct roles. In this setting, ES cannot be released from NC1, but NC1 exerts its own action as a trimer when Dmp is present. One option for NC1 activity is that its trivalence is used to crosslink different ECM components. This might stabilize structures that contain guidance information and eventually enhance the stringency of guidance interactions.

The modulating capacity of Dmp derivatives may buffer guidance processes against perturbations

Our results show that the wild type ISNb error rate of 7.2% can be reduced by 10-fold by the expression of additional trimeric endostatin from the UAS-NC1 transgene. This illustrates that Dmp possesses a capacity to influence pathfinding accuracy that is not fully taken advantage of in the normal wild type situation. This could be explained in that a low rate of statistically no more than one error per individual does not decrease the affected individual's fitness and hence does not exert any selective pressure.

Another possible reason is that increasing the fidelity of guidance interactions would interfere with the system's error prevention and/or correction mechanisms. As already described, correct pathfinding depends on the appropriate balance of attractive and repulsive forces, and this system is susceptible to perturbations. The observation that nerves are occasionally misrouted in wild type is a reflection of this fact. An ability to correct errors would therefore convey a certain degree of robustness in dealing with such perturbations. Indeed, error correction has been described in the Drosophila motor axon guidance system for SNa misrouting phenotypes that occur due to axonal overexpression of FasII. In third instar larvae, incidence of this phenotype is much lower than in embryos, suggesting that errors can get corrected during larval stages (Lin and Goodman 1994). For zebrafish retinal axons, it has also been stated that errors occur in wild type, but get corrected (Hutson and Chien 2002). In the example of ISNb pathfinding, excess trimeric endostatin might impair the system's plasticity by overly stabilizing misleading guidance interactions, and thus impair correction mechanisms, and by consequence reduces plasticity of the system.

Our observation that dmp mutations lead to pathfinding defects shows that certain features of Dmp do normally act to improve guidance accuracy. Hence, Dmp seems to play a dual modulating role, and as detailed above, these roles are likely exerted by different mechanisms. On the one hand, Dmp helps leading nerves where they belong, but on the other hand, does normally not make full use of its capacity to do so. This opens up the possibility that Dmp is an agent that has multiple means to modulate axon guidance accuracy by buffering axonal navigation against perturbations. It will be interesting to see if excess NC1 is able to improve guidance phenotypes of mutants other than Dmp.

Opposed effects of mono- and trimeric endostatin are a widespread phenomenon and differ in a context-dependent manner

Different effects of monomeric versus oligomeric endostatin have also been observed in C. elegans (Ackley et al. 2001), but contrarily to our results, the C. elegans cle-1 neuronal migratory phenotype gets rescued by the putatively trimeric NC1 domain, whereas the ES domain does not provide rescue activity.

Mono- and oligomeric ES have also been tested in vertebrate tissue culture systems for their effects on tubule formation and angiogenesis. Oligomeric ES is an inhibitor of tube morphogenesis in HUVEC (human umbilical vein endothelial cell) cultures, since an addition of oligomeric ES to the culture before tube formation inhibited the process, which is motility-dependent, whereas after completion of tube formation, oligomeric ES had a motogenic activity leading to dispersal of the tubular structures. Monomeric ES alone did not have any effect on cell behavior, but pre-incubation with ES inhibited the motogenic effect of oligomeric ES (Kuo et al. 2001). In a CAM (chorioallantoic membrane) angiogenesis assay system, mono- and oligomeric ES of collagen XV and XVIII are converse in their inhibitory activity, that also depends on the cytokine used for stimulating angiogenesis (Sasaki et al. 2000). In another context, using IBE cells (intrahepatic biliary epithelial), monomeric ES even enhanced tubule formation (Dixelius et al. 2000). Hence, the action of mono- and trimeric ES seems to greatly depend on the biological context, on the cell types involved and additional signals they receive from their environment. The only common theme seems to be an antagonistic action of monomeric versus oligomeric ES. In this light, it is not surprising that the rescue activity we observed on the Drosophila motor axon guidance phenotype is exactly opposite to the results for mechanosensory neuron (MSN) migration in C. elegans (Ackley et al. 2001). Notably, the biological process of axonal growth cone steering depends on directional cues and their correct interpretation, which is a fundamental difference to all other experimental systems that compared the action of mono- and oligomeric ES so far, as these systems all focused on the presence or absence of migratory activity per se. This includes the observations made for C. elegans MSN migration, as the phenotypes described by Ackley et al. (2001) mainly consist of errors in migration distance of whole neurons and not direction. So the observations of ES effects we made are placed in a novel biological context, and differences compared with other contexts are not surprising.

For murine multiplexins biochemical interactions with several other ECM components have been observed (Sasaki et al. 2000). Together with the variety of biological effects described for multiplexin fragments by us and others, it seems likely that its exact mechanisms of action depends very much on the biological context. We have here provided the starting point for functionally integrating Drosophila multiplexin into the biological process of motor axon pathfinding.

Acknowledgments

Bernard Moussian was supported by Deutsche Forschungsgemeinschaft (DFG). We would like to thank Christiane Nüsslein-Volhard for her extensive support. We also thank Andrew Renault, Jana Krauss, Uwe Irion and Hermann Aberle for many helpful discussions.

Ancillary