How Notch establishes longitudinal axon connections between successive segments of the Drosophila CNS

The axon scaffold of every bilaterian central nerve cord, whether in insects or mammals, is built along two major axes: lateral connections within a segment and longitudinal connections between segments. Although a great deal is known about lateral patterning in the CNS, the mechanisms that produce longitudinal connections have been elusive and controversial. In Drosophila, for example, longitudinal pioneer axons track along specialized glial cells, the interface glia (also called longitudinal glia) (Jacobs and Goodman, 1989a; Jacobs and Goodman, 1989b; Hidalgo et al., 1995; Hidalgo and Booth, 2000), so it has been speculated that those glia might provide the substratum for the pioneers. Perturbing the glia, however, produces only relatively mild defects in the initial extension of longitudinal pioneers (Hosoya et al., 1995; Jones et al., 1995; Hidalgo and Booth, 2000), arguing against this hypothesis. Genetic experiments implicate Drosophila DCC (Frazzled) and its ligand, Netrin, and also the receptor Notch with its ligand, Delta, in the development of CNS longitudinal axon tracts in the fly: mutation of any of these genes interferes with development of longitudinal tracts between segments (Giniger et al., 1993; Kolodziej et al., 1996; Mitchell et al., 1996). Frazzled is not expressed in the longitudinal pioneer neurons themselves, but rather in cells along the nascent longitudinal pathway. Frazzled marks the track the pioneers are destined to follow by capturing Netrin protein produced elsewhere in the CNS and presenting it for recognition by pioneer growth cones (using another, unidentified Netrin receptor) (Hiramoto et al., 2000). Notch promotes growth of longitudinal axons by regulating the Abl tyrosine kinase pathway (‘Notch/Abl signaling’) (Giniger, 1998; Crowner et al., 2003; Le Gall et al., 2008). The non-canonical Notch/Abl pathway stands in contrast to the usual mechanism by which Notch directly controls nuclear gene regulation: in most contexts, in response to ligand, the intracellular domain of Notch is released proteolytically from the membrane and transits to the nucleus to form a transcription complex that activates genes bearing binding sites for the Notch-associated DNA-binding protein, Su(H) (Lai, 2004). How modulation of Abl by Notch promotes growth cone motility is unknown.

In establishing longitudinal connections in the CNS two problems must be solved. The first is to mark the path for pioneer axons to follow, and the second is driving motility of their growth cones along that path. Whereas much is known about how late-growing axons follow an existing nerve, we do not understand what specifies the exact trajectories assumed by pioneer axons as they first establish a nerve pathway. Within the growth cone, although a host of signaling proteins have been identified in the cascades downstream of guidance receptors (Luo, 2000; Song and Poo, 2001), the logic that maps individual signaling molecules onto particular steps in morphogenesis remains obscure, as manipulating signaling molecules in vivo often yields contradictory results (Wills et al., 1999b; Wills et al., 1999a; Bashaw et al., 2000; Krause et al., 2002; Hsouna et al., 2003; Forsthoefel et al., 2005; Trichet et al., 2008). This has frustrated efforts to understand how any single receptor modulates the cytoskeleton to guide a specific growth cone in vivo.

We now demonstrate how the interaction of four different CNS cell types specifies the pathway for Drosophila longitudinal pioneer axons, and how modulation of signal transduction in those pioneers drives their motility along that pathway. Ventrolateral neurons expressing the Notch ligand Delta project axons medially as they grow towards the commissures. These axons contact Notch-expressing interface glia, activating canonical, Su(H)-dependent Notch signaling in those glia. These glia, in turn, attract a cap of fine filopodial processes from surrounding neurons, and shapes that cap into a continuous track of neuronal membrane that bridges from one segment to the next. The neuronal cap, being enriched for the Drosophila DCC Frazzled, recruits Netrin protein, thus constructing a continuous Netrin domain that is the direct substratum for extending pioneer axons. Meanwhile, the Delta-positive commissural axons that stimulated canonical Notch signaling in the interface glia also act as stepping stones for the growing longitudinal axons. Delta on the commissural axons activates non-canonical Notch/Abl signaling in the longitudinal pioneers, stimulating growth cone motility by de-repressing the actin polymerization factor Enabled and suppressing activity of the Rac GTPases. By increasing filopodial development, and presumably also reducing substratum adhesion, this promotes the ability of advancing pioneer growth cones to cross the segment border and encounter one another, thereby establishing the first connection between successive segments of the fly CNS.

Flies were from the following sources: heartless-GAL4: Alan Michelson, NIH, USA; E(spl)mδ-lacZ: Jas Singh, Mt Sinai School of Medicine, NY, USA; UAS-Notch-lexA: Toby Lieber, Columbia University, NY, USA; GAL4-605: Gerd Technau, University of Mainz, Germany; frazzled⁻: Peter Kolodziej, Vanderbilt University, Nashville, TN, USA; Masaki Hiramoto and Yash Hiromi, NIG, Japan; GAL4-15J2: Andrea Brand, Welcome Trust, Cambridge, UK; UAS-mCD8-GFP, UAS-Rac1[N17], UAS-Rac1[V12] and repo-GAL4: Liqun Luo, Stanford University, Palo Alto, CA, USA; UAS-ena-GFP, UAS-FP4-mito, UAS-cpb: Julie Gates and Mark Peifer, University of North Carolina, Chapel Hill, NC, USA; UAS-trio[GEF1], Df(1)Net[A, B], NetB-myc: Barry Dickson, IMP, Vienna, Austria. Two copies of the modifying transgene were used for experiments using UAS-FP4-mito and UAS-Rac[N17]. All other stocks were from the Bloomington Drosophila Stock Center.

For temperature-shift experiments we used Notch^ts1 (N^ts1) or the heteroallelic combination Dl^6B37/via1. To examine the mature axon pattern (stage 15/16), embryos were collected for 3 hours at 18°C, aged 8 hours at 18°C, shifted to restrictive temperature (32°C) for 6 hours and fixed. To quantify early-growing longitudinal axons (mid-stage 13), embryos were collected for 3 hours at 18°C, aged 7.5 hours at 18°C, shifted to restrictive temperature (32°C) for 4 hours 20 minutes and fixed. To visualize extending pioneer growth cones, embryos were collected for 3 hours at 18°C, aged 7 hours at 18°C, shifted to restrictive temperature (32°C) for 4 hours and fixed. Embryos for quantification of early CNS phenotypes were tightly staged based on embryo morphology: dorsal closure, head involution and gut morphogenesis.

Embryos for most experiments were prepared, fixed and stained by standard methods (Bier et al., 1989). For Netrin-myc and actin-GFP we developed a modified hot triton fixation method (Miller et al., 1989). Embryos were collected, dechorionated and rinsed with 0.7% NaCl, 0.3% Triton X-100 (NaCl/Triton). Embryos were transferred to a 5 ml test tube in a small volume of NaCl/Triton, 3 ml of pre-warmed (90°C) NaCl/Triton was added and the tube placed immediately in a 90°C heat block for 5 seconds. Embryos were then vortexed briefly, poured into 20 ml ice-cold NaCl/Triton and held at 0°C for 10 minutes. Embryos were devitellinized with 1:1 heptane:95% methanol, 25 mM EGTA, by vortexing vigorously for 30 seconds, and rinsed with methanol/EGTA for at least 1 hour. Embryos were rehydrated through a methanol series into PBS, then postfixed with 4% methanol-free formaldehyde in 0.1 M NaPO₄ pH 7.2 for 15 minutes at room temperature. Embryos were washed four times in PBT (0.1 M NaPO₄ pH 7.2, 0.3% Triton X-100), then blocked and stained with antibodies.

Antibodies were from the following sources: rabbit anti-Frazzled: Y. N. Jan, UCSF, San Francisco, CA, USA; Laurie Lee and Peter Kolodziej, Vanderbilt University, Nashville, TN, USA; anti-Heartless: Alan Michelson; anti-Prospero: Chris Doe, University of Oregon, Eugene, OR, USA; Rb anti-Repo: Gerd Technau; rabbit anti-β-gal, mouse anti-β-gal and rabbit anti-HRP: Cappel Research Reagents, Cochranville, PA, USA; chick anti-GFP: Aves Lab, Tigard, OR, USA; mouse anti-GFP: Invitrogen. All other primary antibodies were from the Iowa Developmental Studies Hybridoma Bank. Secondary antibodies were from Jackson ImmunoResearch, except FITC-anti chicken IgY, which was from Aves Labs.

Peroxidase immunohistochemistry was performed using biotinylated secondary antibodies and VectaStain avidin-biotin-HRP (Vector Labs). Immunofluorescence was performed with fluorochrome-conjugated secondary antibodies, except detection of anti-Delta and NetB-myc, which were carried out using TSA-biotin (NEN) and FITC-streptavidin (Jackson ImmunoResearch), and some experiments employing TRITC-conjugated anti-HRP (Jackson ImmunoResearch). Immunofluorescent samples were examined by epifluorescence with a Zeiss AxioImager, using deconvolution or structured illumination (Apotome), or by confocal microscopy.

All embryos were examined in filet preparations. Embryos were mounted in 80% glycerol for 3,3′-diaminobenzidine (DAB) histochemistry, and in Fluorogard (BioRad) for immunofluorescence. In all experiments, embryo genotypes were scored unambiguously using gratuitous markers. For quantification of axonal phenotypes with mAb22C10, intersegment connections were scored at mid-stage 13. Left and right sides of a segment were scored independently. A connection was only scored as ‘absent’ if there were no axons detected connecting a pair of neuromeres; ‘thinning’ of a connection was not considered to be a mutant phenotype. n=200-450 hemisegments per dataset for all experiments. Statistical significance was assessed by ANOVA.

Intersegmental longitudinal connections in the fly CNS are pioneered by four interneurons, pCC, MP1, dMP2 and vMP2 (Fig. 1A) (Jacobs and Goodman, 1989a; Jacobs and Goodman, 1989b; Lin et al., 1995; Hidalgo and Brand, 1997). dMP2 and vMP2 are born in the middle of each segmental ganglion (‘neuromere’), and begin to extend axons during stage 12. The dMP2 axon grows posteriorly and vMP2 grows anteriorly, projecting towards the vMP2 and dMP2 axons pioneering from the next segment. Simultaneously, pCC sends its axon anteriorly in association with vMP2, while the MP1 axon wraps around dMP2 and extends posteriorly in association with it. The encounter of pioneers from successive segments establishes the first intersegment connection early in stage 13. The pioneer functions of these four axons are largely redundant as all must be ablated to produce a severe defect in the mature nerve (Lin et al., 1995; Hidalgo and Brand, 1997). Below, we will focus on dMP2 and vMP2.

Removing Notch midway through embryogenesis, using a temperature-sensitive mutation, prevented formation of mature longitudinal axon tracts (Fig. 1B,C) (Giniger et al., 1993; Giniger, 1998). Examining embryos at early stage 13 revealed that the Notch phenotype is apparent at the earliest stages of the pioneering of these tracts. In temperature-shifted embryos, the dMP2 and vMP2 axons grew in the appropriate direction, but stalled, failing to make contact even by late stage 13 (Fig. 1D-F). Expressivity of this phenotype depends on the timing of the temperature shift (Crowner et al., 2003). We adjusted conditions to give a very modest expressivity (13±1.1% of early hemisegmental connections missing) to minimize unrelated CNS defects. Failure to establish intersegment connections is not due to transformation of the identities of vMP2 and dMP2, as shown by the direction of axon growth and by staining for the molecular marker Odd skipped (Spana and Doe, 1996) (data not shown). Stalling of longitudinal pioneers was also observed in Delta^ts (Fig. 1G).

To ascertain where Notch is activated at this stage in wild type, we examined a transcriptional reporter for Notch activity, E(spl)mδ-lacZ (Cooper and Bray, 1999), and found β-galactosidase expression in a subset of the interface glia. These glia lined the pathway that the advancing longitudinal pioneer axons will follow and were present when those axons were growing (Fig. 1H) (Jacobs and Goodman, 1989a; Hidalgo and Booth, 2000). The β-galactosidase⁺ cells were identified as interface glia based on position and morphology (Fig. 1H), and by co-labeling for Prospero and Repo (Fig. 1I,J). The most prominent β-galactosidase expression is in glia within the neuromeres (Griffiths and Hidalgo, 2004), but at this stage there was also expression in interface glia between neuromeres, along the presumptive axon pathway (Fig. 1H, arrows).

Notch activation in interface glia was surprising because expression of Notch in neurons suppresses the late stage CNS axonal phenotype of Notch^ts (Giniger, 1998; Le Gall et al., 2008). Remarkably, we now found we could efficiently rescue the early axon phenotype of Notch^ts by expressing Notch either just in the glia [repo-GAL4 (all glia): 72% rescue; htl-GAL4 (interface glia): 92% rescue] or just in the dMP2 and vMP2 pioneers (15J2-GAL4: 85% rescue; P<0.01 in all cases; t-test) (Fig. 1K). We verified that in wild type Notch is expressed in interface glia and in both dMP2 and vMP2 (data not shown). We also verified that Notch activity in both cell types is physiologically relevant using loss-of-function experiments (Fig. 1K): RNAi against Notch in interface glia significantly enhanced the Notch^ts phenotype (2.5-fold; P<0.005), and cell-autonomous reduction of functional Notch in dMP2 and vMP2 by cis-inhibition (Sprinzak et al., 2010) using expression of Delta lacking its intracellular domain (Itoh et al., 2003) enhanced the Notch^ts phenotype by 92% (P<0.05). RNAi was not effective on the relevant timescale in dMP2 and vMP2 using 15J2-GAL4.

How can Notch rescue the growth of longitudinal pioneer axons both autonomously, acting in the pioneers, and non-autonomously, acting in the glia? We will first dissect the action of Notch in glia, and then turn to the Notch mechanism in pioneers.

Notch becomes activated in interface glia by contact en passant with the axons of Delta-expressing commissural interneurons (Thomas and van Meyel, 2007). Interface glia were contacted by Delta⁺ axons and displayed activated Notch signaling by early stage 13, in time to modulate growth of longitudinal pioneers (Fig. 2A). Moreover, expression in the glia of a Notch derivative lacking the ligand-binding domain could not rescue longitudinal axons (21% of intersegment connections were absent in Notch^ts; repo-GAL4;UAS-Notch^ΔEGF(10-12) versus 13% for Notch^ts, consistent with previous evidence showing a mild dominant-negative effect of this transgene) (Le Gall et al., 2008).

Notch^ts embryos had the normal number of interface glia, as assayed with anti-Repo (Fig. 2B). Moreover, the position and morphology of those glia appeared largely normal, and they expressed Htl and (where appropriate) Prospero (Fig. 2C,D). Thus, failure of overall glial development does not underlie the Notch axonal phenotype.

We next found that longitudinal growth cones do not directly contact the interface glia, but rather a thin meshwork of neuronal tissue intervenes between. This meshwork stretches in an unbroken band from one segment to the next, potentially providing a pathway for longitudinal axons. Labeling late stage 12/early stage 13 wild-type embryos for longitudinal pioneer axons, interface glia and neuronal membrane revealed a small (~2-3 μm) gap between the glial surface and the advancing growth cones (Fig. 2E,F). This gap was filled by a fibrous-appearing meshwork that was labeled by four well-characterized markers for neuronal membranes: anti-HRP, mAb BP102, anti-Frazzled (DCC) and UAS-mCD8-GFP driven by neuron-specific GAL4 lines (Fig. 2E″-E″″,G). The advancing pioneer growth cones were invariably in intimate contact with this mesh, typically tracking along its edge (Fig. 2E,F, Fig. 3C). The substratum meshwork did not label with glial markers (anti-Htl or UAS-mCD8-GFP driven by glial-specific GAL4; Fig. 2H), but rather was positioned ventral to the glia. At this stage, the pioneer neurons themselves did not express Frazzled (Hiramoto et al., 2000), showing that the layer of neuronal membrane is provided by nearby differentiating neurons, and not by the pioneers. Moreover, previous serial section electron microscopy (Jacobs and Goodman, 1989a) showed that there are no other axons present in this region at the stage when the pioneers are extending, ruling out the possibility that the neuronal labeling in the meshwork derives from other axons. These observations are consistent with previous reports of a bed of unidentified filopodia closely apposed to the surface of the interface glia (Jacobs and Goodman, 1989a; Hidalgo and Booth, 2000). Development of the interface glia precedes morphogenesis of the neuronal meshwork (Fig. 2J), consistent with the glia positioning the meshwork rather than vice versa.

The essential glial function of Notch is to shape the associated neuronal meshwork into a continuous band that bridges between segments to provide the direct longitudinal substratum. In Notch^ts, the nearby differentiating neurons were labeled with the expected molecular markers, including anti-Frazzled (Fig. 3E), anti-Elav, mAb BP102 and anti-HRP (data not shown). In the mutant, however, although the neuronal meshwork formed within each segmental ganglion, it did not spread laterally between ganglia to make an unbroken band stretching from segment to segment. Instead, there were gaps in the mesh between segments (compare Fig. 3B,E), and pioneer axons often (but not always) stalled or misrouted when they encountered those gaps (compare Fig. 3C,F). Whether these gaps reflect a defect in the growth or orientation of the filopodia making up the meshwork has not been determined. Interface glia were still present in segments with stalled axons even though the neuronal meshwork was absent (Fig. 3G). Restoring Notch just to the glia (with repo-GAL4 or htl-GAL4) was sufficient to rescue the continuity of the neuronal meshwork (Fig. 3H), and thus, indirectly, the extension of the longitudinal pioneers. Temperature-shifts of Dl^ts produced gaps in the neuronal mesh, similar to those observed in Notch^ts (data not shown).

Pioneer longitudinal axons track along the edge of a band of Frazzled⁺ tissue, and this Frazzled⁺ domain immobilizes Netrin, presenting it for recognition by an alternate receptor on the pioneer growth cones (Hiramoto et al., 2000). Consistent with this, visualizing Netrin by immunostaining of Notch^ts revealed longitudinal gaps in the pattern of Netrin accumulation that mimicked the gaps in Frazzled (Fig. 3I). The Notch pioneer phenotype, moreover, was strongly enhanced by heterozygosity for a frazzled mutant (P<0.01), and the Notch Netrin double mutant had a significantly higher frequency of longitudinal defects than either mutation by itself (Fig. 3J). Thus, the crucial role of Notch in interface glia is to shape a continuous carpet of Fra- and Netrin-enriched neuronal tissue between segments, and the advancing longitudinal pioneers navigate by tracking along that carpet.

If Notch acts in glia to construct the longitudinal pathway, why does expression of Notch in pioneers rescue their axons? We expressed Notch selectively in the dMP2 and vMP2 pioneer neurons of Notch^ts and stained for early growing longitudinal axons and for Frazzled. Although restoration of Notch to the pioneers rescued growth of their axons, the continuity of the Frazzled⁺ neuronal meshwork was not restored: longitudinal gaps were still present (Fig. 4A-D). Evidently, restoration of Notch to the pioneers endowed them with the ability to extend across gaps in the Frazzled pathway. The ligand for Notch function in pioneers seems to be provided by the same Delta⁺ commissural axons that stimulate canonical Notch signaling in interface glia; after they passed under the glia, these Dl⁺ axons were perfectly positioned to act as stepping stones for advancing longitudinal pioneers (Fig. 4F, Fig. 7D). Consistent with this, expression in the pioneers of a Notch derivative lacking the ligand-binding repeats did not rescue the Notch axon phenotype (24% of intersegment connections absent in Notch^ts; 15J2-GAL4; UAS-Notch^ΔEGF(10-12) versus 14% for Notch^ts; 15J2-GAL4).

In contrast to its action in glia, Notch action in pioneers does not employ canonical Su(H)-dependent signaling, but rather the non-canonical Notch/Abl signaling pathway (Crowner et al., 2003; Le Gall et al., 2008). We expressed two diagnostic reagents in either interface glia or in the dMP2 and vMP2 pioneers of Notch^ts embryos and assayed development of longitudinal axons at mid-stage 13. Su(H)-VP16 is a constitutively active, receptor-independent derivative of the canonical Notch effector Su(H), but has no influence on the Notch/Abl pathway (Crowner et al., 2003). Notch^Δ2155 is a truncated derivative of full-length Notch that has no ability to stimulate canonical signaling but is almost fully active for the Notch/Abl interaction (Le Gall et al., 2008). Expression of Su(H)-VP16 in glia rescued construction of a continuous longitudinal track and thus fully rescued extension of longitudinal pioneers, whereas expression in the pioneers had no effect on the extension of their axons (Fig. 4E). Conversely, expression of Notch^Δ2155 in the glia produced no rescue of either the Frazzled⁺ pathway or the pioneer axons, but in pioneers it was nearly as effective as wild-type Notch for rescuing axon extension (Fig. 4E). Moreover, we did not observe expression of the E(spl)mδ-lacZ canonical Notch reporter in any of the longitudinal pioneers at this stage (Le Gall et al., 2008). Finally, Notch interacted genetically with Abl to modify longitudinal axon development just as we have observed previously for pathfinding of the ISNb motonerve: removing one copy of the Abl gene by mutation suppressed the mature, stage 16 CNS axonal phenotype of Notch (Notch^ts: 29% of intersegmental connections defective; Notch^ts; Abl⁴/+: 13% defective; P<0.01; t-test).

We verified further that the autonomous function of Notch in dMP2 and vMP2 axons is via the Notch/Abl interaction by selectively expressing constitutive or dominant-negative derivatives of various proteins of the Abl pathway in the pioneers and assaying enhancement or suppression of the Notch phenotype. (RNAi was ineffective for this experiment owing to the long half-lives of the proteins.) Previous experiments suggested that Notch suppresses the activity of the Abl signaling pathway (Fig. 5A) (Crowner et al., 2003). Consistent with this, expression of the Abl antagonist Enabled in dMP2 and vMP2 suppressed the early longitudinal axon phenotype of Notch, whereas expression of an Ena antagonist, FP4mito (Bear et al., 2000), enhanced it (Fig. 5B). Enabled, in turn, is an actin ‘anti-capping’ protein (Trichet et al., 2008), and overexpression of actin capping protein β (cpb) enhanced the Notch phenotype (Fig. 5B). The guanine nucleotide exchange factor Trio is a positive element of the Abl pathway (Liebl et al., 2000), and overexpression of the active, Rac-specific GEF1 domain of Trio or of Rac[V12] enhanced the Notch phenotype whereas expression of a dominant-negative Rac[N17] suppressed it, again consistent with Notch acting as a suppressor of Abl signaling (Fig. 5B) (Song and Giniger, 2011). None of these transgenes caused defects in longitudinal axons when expressed in a wild-type background with this driver (<2% of hemisegments defective). As indicated above (Fig. 4E), expression of Su(H)-VP16 in this paradigm had no effect on the early longitudinal axon phenotype. These data argue that Notch promotes motility of the dMP2 and vMP2 pioneer growth cones by suppressing Abl pathway signaling autonomously in these cells.

Visualizing actin in dMP2 and vMP2 growth cones revealed that Notch promotes the presence of filopodial structures on the growth cone. We expressed actin-GFP in the dMP2 and vMP2 growth cones of temperature-shifted Notch^ts mutants with or without wild-type Notch, and prepared embryos using a fixation that preserves actin structure (Miller et al., 1989). Embryos were double-labeled with anti-GFP and with a marker for the microtubule cytoskeleton (Futsch, the Drosophila MAP-1B). Filopodia were defined operationally as thin, whisker-like projections of uniform caliber that were actin-positive, but microtubule-negative, and were counted on dMP2 growth cones that were fixed while in transit between neuromeres (Fig. 6A-F). Filopodia were detected in the Notch mutant in 30% of growth cones, with an average total filopodial length of 1.2±0.4 μm/growth cone (n=33 dMP2 growth cones). By contrast, when Notch was restored to the growth cone, filopodia were detected on 93% of growth cones (n=41; P<10⁻⁷, χ² test), with an average total filopodial length of 4.7±0.5 μm/growth cone (P<10⁻⁶, ANOVA). Clearly, Notch strongly promotes formation and/or stability of filopodia on dMP2 growth cones.

The axons of the longitudinal pioneer interneurons of the Drosophila ventral nerve cord establish the initial connection between successive segments of the animal. The receptor Notch is crucial for making those first connections, performing two parallel, partially redundant but completely separate functions. Canonical Notch signaling in the interface glia constructs an unbroken track for longitudinal pioneer axons to follow by shaping a continuous band of neuronal membrane that bridges from segment to segment. Simultaneously, non-canonical Notch/Abl signaling in the pioneer neurons themselves promotes the motility of their growth cones, suppressing the activity of the Abl tyrosine kinase signaling module to stimulate filopodial development. Either signaling mechanism provides substantial rescue of a Notch mutant, but both are required for full activity in formation of longitudinal connections of the CNS.

The mechanism that guides the very first axon to establish the path of a nascent nerve is one of the most fundamental problems in neural development. For longitudinal pioneers of the Drosophila CNS, we now see that constructing their path requires coordinated contributions from four interacting cell types (Fig. 7). First, the axons of commissural interneurons bearing the Notch ligand Delta contact interface glia. The glia respond by activating canonical Notch signaling, enhancing expression of Notch target genes, including prospero (Griffiths and Hidalgo, 2004; Thomas and van Meyel, 2007) (I.K. and E.G., unpublished). The genetic program stimulated by Notch directly or indirectly allows the glial cells to attract a ‘cap’ of fine filopodial processes from nearby differentiating neurons, and shapes that cap into a continuous longitudinal band that bridges between segments. The neuronal cap atop the glia bears the Netrin receptor Frazzled (DCC), which in turn recruits soluble Netrin (Hiramoto et al., 2000), thus constructing a domain of accumulation of Netrin protein that hugs the surface of the associated glia. Finally, the pioneer growth cones advance along the edge of that domain of immobilized Netrin until they meet and fasciculate with their partners pioneering from the next segment. The consequence of this choreography is a nerve trajectory that follows, indirectly, the shape of the row of interface glia.

This view suggests plausible explanations for many aspects of longitudinal axon development that have, up to now, been confusing. Previous investigators have documented that the pioneer growth cones extend amidst a thicket of filopodia that cap the interface glia (Jacobs and Goodman, 1989a; Hidalgo and Booth, 2000). The provenance and significance of those filopodia were unknown, though it was clear that they did not derive from other axons (Jacobs and Goodman, 1989a). We now see that these filopodia come from surrounding, differentiating neurons, and that their function is to hold Frazzled, and therefore Netrin (Hiramoto et al., 2000), in a pattern dictated by the positions of the overlying glia, creating the Netrin domain along which the pioneers extend. This explains why the positions of the interface glia correlate so closely with the axon trajectory even though the glia are not the direct substratum. Our data, along with other recent results, also suggest why previous experiments investigating the guidance function of interface glia might have given such confusing results. Transformation of the glia into neurons in a gcm mutant would be predicted to place a row of DCC-expressing neurons in precisely the position of the wild-type filopodial carpet. Moreover, genetic experiments ablating or displacing the glia have relied on reagents targeting the progeny of the longitudinal glioblast (Jacobs et al., 1989; Hidalgo and Booth, 2000; von Hilchen et al., 2010), but we now know that only nine of the ten interface glia come from this precursor; the tenth, M-ISNG, is from a different lineage (Beckervordersandforth et al., 2008). M-ISNG is appropriately positioned to anchor the filopodial carpet in the absence of the other interface glia (Beckervordersandforth et al., 2008) and preliminary experiments suggest that it is sufficient for this (I.K. and E.G., unpublished). Moreover, von Hilchen et al. (von Hilchen et al., 2010) have argued that in those rare segments where pioneer axons stall owing solely to manipulations of the interface glia, all ten of them, including M-ISNG, tend to be absent or displaced. Finally, as in previous studies (Hiramoto et al., 2000), we find that the Netrin zygotic mutant has a mild, and genetically enhanceable phenotype, showing that the null for the gene is not null for the genetic pathway. It might be that there is a maternal contribution to Netrin, as there is for frazzled. Alternatively, because the receptor on longitudinal pioneers presumably recognizes a Netrin-Frazzled complex, it might be that this receptor has some affinity for Frazzled even in the absence of Netrin. Identification of the missing Netrin receptor will be necessary to clarify this point. It also seems likely that other neuronal components cooperate with the Netrin-Frazzled complex on the meshwork to provide substratum function, as expression of Fra in interface glia is not sufficient to rescue the Notch axonal phenotype (data not shown).

Once the pathway for an axon has been constructed, there remains the problem of driving the motility of the growth cone along that pathway. Somehow, the information encoded in a pattern of occupancy of cell surface receptors must be transformed into a pattern of cytoskeletal dynamics that drives growth cone motion. At the level of the axon, this is the bedrock problem in axon guidance, and it, too, has resisted analysis. Our data reveal how Notch modulates an elementary property of the actin cytoskeleton to promote motility of longitudinal pioneer growth cones. Through its antagonism of the Abl signaling network, Notch de-represses the Abl antagonist Enabled and suppresses the Rac GEF Trio. Enabled directly promotes filopodial growth (Krause et al., 2002; Trichet et al., 2008); suppressing Rac indirectly promotes filopodia, probably by redirecting various factors away from lamellipodia (P. Bradley and E.G., unpublished observations). Stimulating filopodial development probably promotes longitudinal axon growth in at least two ways. First, converging pioneer growth cones from successive segments need to encounter one another and fasciculate to establish the connection between segments (Fig. 6G). Extension of filopodia increases the area searched by an advancing growth cone, increasing the probability that it will encounter its partners advancing from the adjacent segment. Second, filopodia promote neurite growth by promoting microtubule invasion of the leading edge (Dent et al., 2007).

In parallel with stimulating filopodia, suppression of Trio, and thus of Rac, is expected to reduce substratum adhesion (Kaufmann et al., 1998; Symons, 2000; Crowner et al., 2003). When the pioneers are growing towards the segment border, small gaps in the Frazzled-Netrin pattern are not uncommon, so release of the advancing growth cone from the substratum is likely to aid its forward motion. Moreover, initially there is more Frazzled-bound Netrin within the neuromeres than there is at the segment border (Fig. 3I), requiring advancing pioneers to go down a gradient of Netrin, towards a region with less Netrin. It is possible that both of these properties make it helpful to limit substratum adhesion of the growth cone via reduction of Rac signaling by Notch.

It might seem paradoxical that Notch promotes axon growth by suppressing Abl signaling when Abl has been the archetype of a motility-promoting signaling pathway (Gertler et al., 1993). Indeed, genetic studies of Abl in axon guidance have often appeared to be confusing and contradictory (Bashaw et al., 2000; Hsouna et al., 2003; Forsthoefel et al., 2005). In part, this reflects pleiotropy. Abl appears to act in the glia, and in the cells providing the filopodial carpet, in addition to the pioneers (I.K. and E.G., unpublished). As the phenotype in a whole-animal mutant reflects the sum of unrelated functions in different cells, seemingly similar experiments can produce contradictory results if different cellular processes become limiting. For longitudinal axons, for example, if pathway establishment is limiting (in Abl⁻ or fra⁻ animals), reduction of Notch interferes with axon growth synergistically (Giniger, 1998); if growth cone function is limiting (in Notch⁻ animals), reduction of the Abl pathway restores axon growth (this work). It was therefore essential in the current work to control gene activity, and analyze phenotypes, in single, identified cells.

Beyond pleiotropy, however, complexity arises because the effect of signaling molecules in motility is profoundly context dependent. Ena promotes actin polymerization but often restricts cell motility (Bear et al., 2000; Krause et al., 2002; Trichet et al., 2008); cofilin severs actin filaments but can promote net actin polymerization and cell migration (Loisel et al., 1999; Blanchoin et al., 2000; Ng and Luo, 2004). As axon growth is achieved by throughput through a cycle of actomyosin dynamics, it requires a balance among the steps of that cycle (Sheetz et al., 1998; Suter and Forscher, 2000; Giniger, 2002). Excessive activity or inactivity of any single step in the process inhibits motility by impairing progression through the cycle (Luo et al., 1994; Song and Giniger, 2011). Our data now reveal that, for Drosophila longitudinal pioneers, an essential aspect of growth cone movement is restraint of Abl activity to allow filopodial development, and perhaps also to limit substratum adhesion.

The data reported here reveal that Notch promotes CNS longitudinal axon growth in two very different ways, constructing a pathway using its canonical signaling mechanism and promoting motility via the Notch/Abl interaction. This dual role bears striking parallels to the dual role of Notch in radial migration of neurons in the mammalian cortex. There, as in the fly, canonical signaling by Notch is essential for the development of glial cells that define a migration pathway (Ever and Gaiano, 2005), whereas interaction with the Abl pathway protein Disabled controls neuronal motility and adhesion (Hashimoto-Torii et al., 2008; Song et al., 2010). Further study will be required to assess whether these parallels between Notch function in the fly and vertebrate nervous systems reflect a deeper mechanistic similarity. Similarly, it will be interesting to see whether formation of longitudinal nerve tracts in the spinal cord uses machinery homologous to that which we have described in the fly.

We thank members of our laboratory for helpful advice and assistance. We are also grateful to Roger Jacobs, Don van Meyel, Gerd Technau, Ben Altenhein and Mark Peifer for valuable discussions and for sharing unpublished observations. For comments on the manuscript we are grateful to Ben Altenhein, Ajay Chitnis, Alicia Hidalgo, Maude Le Gall, Chi-Hon Lee, Don van Meyel and Gerd Technau. We thank all of the investigators who generously provided reagents for these studies, and the NHGRI microscopy core for expert assistance. These experiments were supported by the Basic Neuroscience Program of the NINDS Intramural Research Program (Z01 NS003013). Deposited in PMC for release after 12 months.