Precise connections of neurons with their targets during neural development are responsible for the normal physiological and behavioral patterns of animals (Dorskind and Kolodkin, 2021; Engle, 2010). Neurons generated during embryonic and postnatal development extend axons that navigate along distinct paths to find their appropriate synaptic targets (Chédotal and Richards, 2010; Kolodkin and Tessier-Lavigne, 2011). Axon pathfinding is controlled by the coordinated action of attractive and repulsive cues (Tessier-Lavigne and Goodman, 1996). These opposing guidance cues act at either short-range or long-range (Kolodkin and Tessier-Lavigne, 2011; Tessier-Lavigne and Goodman, 1996). In general, long-range guidance cue molecules travel a long distance to bind to their cognate receptors expressed on growth cones, and then mediate either attractive or repulsive axon guidance (Kolodkin and Tessier-Lavigne, 2011; Tessier-Lavigne and Goodman, 1996). In contrast, short-range guidance cue molecules largely regulate axon–axon, axon–cell, and axon–extracellular matrix interactions (Kolodkin and Tessier-Lavigne, 2011; Tessier-Lavigne and Goodman, 1996). The question then arises of how these different types of axon guidance cues are integrated to steer growth cones toward their synaptic targets.

A wide range of axon guidance molecules, including four major classes of guidance cues-semaphorins, slits, netrins, and ephrins, have been discovered, and their guidance functions appear to be evolutionarily conserved across the animal kingdom (Bashaw and Klein, 2010; Dickson, 2002). Slits found in Drosophila, Caenorhabditis elegans, and vertebrates, are secreted guidance molecules that can serve as short-range repulsive cues (Brose et al., 1999; Hao et al., 2001; Kidd et al., 1999; Li et al., 1999). Netrins can mediate attractive axon guidance through their cognate receptors, such as Frazzled in Drosophila, UNC-40 in C. elegans, and DCC (Deleted in Colorectal Carcinomas) in vertebrates (Chan et al., 1996; Keino-Masu et al., 1996; Kolodziej et al., 1996). Semaphorins are a large family of transmembrane and secreted proteins that contain a conserved semaphorin domain (Pasterkamp, 2012). Most semaphorins can act as repulsive cues during neural circuit development (Kolodkin and Tessier-Lavigne, 2011). Vertebrate ephrins have been shown to mediate either repulsion or attraction in axon guidance, whereas Drosophila ephrin was reported to play an important role in axonal repulsion during neural development (Bossing and Brand, 2002; Kania and Klein, 2016; Liu et al., 2017). However, signaling mechanisms that function downstream of the guidance molecules and their crosstalk should diverge to meet the huge number of demands for wiring complex and diverse neural networks among animal taxa (Bashaw and Klein, 2010; Pasterkamp, 2012). To this end, understanding the molecular signaling mechanisms underlying axon pathfinding is one of the key goals of neural development. The Drosophila motor axon guidance is an excellent model system to study axon pathfinding mechanisms at the level of guidance receptors and their crosstalk, due to its simplicity and powerful genetic tools. This review summarizes the current understanding of molecular mechanisms underlying motor axon guidance in Drosophila, and discusses how the complex wiring observed in the nervous system of many higher organisms can be achieved by a limited number of guidance genes.

Embryonic muscle development begins at stage 12, and somatic muscle patterning and specification are established before stage 16 (approximately 13 h after egg laying) (Bate, 1990). A repeating and identical pattern of 30 somatic muscle fibers is observed in each abdominal hemisegment A2-A7 (Fig. 1; Bate, 1990). In each hemisegment, these muscle fibers are innervated in a cell-type specific manner by 36 motor neurons that are generated and reside in the ventral nerve cord (VNC) (Fig. 1; Landgraf and Thor, 2006; Landgraf et al., 1997; Van Vactor et al., 1993). How can an individual axon, which extends from the cell bodies of motor neurons, follow the correct path to reach specific target muscle(s)? One of the most promising strategies is the control of motor axon pathfinding by the combined action of selective fasciculation at the initial step, and selective defasciculation at specific choice points (Fig. 2; Raper and Mason, 2010; Tessier-Lavigne and Goodman, 1996; Wang and Marquardt, 2013). In fact, two major nerve fascicles, namely the intersegmental nerve (ISN) and the segmental nerve (SN), and the transverse nerve (TN) as a minor nerve fascicle are observed within the VNC when late stage 16 embryos are immunostained with anti-Fas II antibody (Fig. 1; Grenningloh et al., 1991; Jeong, 2017; Van Vactor et al., 1993). This indicates that selective fasciculation of motor axons occurs before exiting the VNC. In the periphery, sequential defasciculation of the ISN at specific choice points creates three nerve branches called the ISN, ISNb, and ISNd, while selective defasciculation of the SN produces the SNa and SNc (Fig. 1; Landgraf and Thor, 2006; Van Vactor et al., 1993). Additional and sequential defasciculation of motor axons along each nerve pathway is also required to innervate particular target muscles (Fig. 2B; Jeong, 2017; Landgraf and Thor, 2006; Van Vactor et al., 1993).

Selective fasciculation of motor axons during neural development is thought to require attractive interaction and specific adhesion among axons. Therefore, cell adhesion molecules (CAMs) that primarily mediate cell–cell adhesion function may contribute to the formation of motor nerve fascicles. One of the best candidate adhesion molecules for selective fasciculation is Fasciclin II (FasII), an immunoglobulin superfamily (IgSF) protein, since panneuronal overexpression of FasII resulted in a range of motor axon defasciculation defects due to hyperfasciculation (Figs. 3A and 3B; Lin and Goodman, 1994). In addition, loss-of-function studies showed that FasII is required to recognize a specific axon pathway through axon–axon attraction in the VNC, even though no motor axon pathfinding defects in the periphery are detected in FasII mutants (Grenningloh et al., 1991; Lin et al., 1994). However, the selective formation of five major nerve fascicles seen in the Drosophila peripheral nervous system (PNS) could not be explained merely by FasII-mediated cell adhesion function, because during neural development, FasII is expressed on all motor axons (Van Vactor et al., 1993). Another candidate adhesion molecule is Connectin (Con), a member of the leucine-rich repeat family, which is expressed on eight lateral and ventral muscle fibers, and also the motor axons that innervate these eight muscles (Nose et al., 1992). When Con is ectopically expressed on a subset of ventral muscles innervated by ISNb motor axons under the control of Toll enhancer element, the motor axon guidance phenotypes observed in the ISNb pathway are like those seen in FasII-overexpressing embryos (Fig. 3B; Lin and Goodman, 1994; Nose et al., 1994). As in the case of FasII, no significant axon pathfinding defects were found in con loss-of-function mutants (Nose et al., 1994), indicating functional redundancy in axonal fasciculation. Furthermore, some of the Drosophila CAMs, such as Neuroglian, Fasciclin III, and Capricious, which are expressed on either all or a subset of motor neurons, were also shown to be required for motor axon guidance (Abrell and Jäckle, 2001; Chiba et al., 1995; Hall and Bieber, 1997). These observations suggest that axon–axon adhesion is regulated by redundant and cooperative roles of multiple CAMs, and also demonstrate that CAMs are necessary, but do not seem to be sufficient for the selective fasciculation of motor axons during neural development.

One intriguing question remains as to what classes of guidance molecules collaborate with CAMs to control selective axon fasciculation. In vivo genetic analysis in Drosophila showed that transmembrane semaphorin-1a (Sema-1a), which is a member of the semaphorin protein family of axon guidance cues, has been implicated in the generation of three smaller nerve branches, including the ISN and ISNb via selective defasciculation (Yu et al., 1998). In Sema-1a mutant embryos, a small but significant fraction of ISNb motor axons fail to defasciculate from the ISN nerve, and thus bypass their normal target muscles, which are referred to as a “fusion bypass phenotype” (Fig. 3B; Yu et al., 1998). Similar fusion bypass phenotypes are also observed in plexin A (PlexA) mutant embryos (Fig. 3B; Winberg et al., 1998a). PlexA functions as a neuronal receptor for Sema-1a in motor axon pathfinding (Winberg et al., 1998a). In addition, it is well established that Sema-1a/PlexA-mediated signaling at specific choice points induces selective axon–axon repulsion (Jeong et al., 2012; 2017; Winberg et al., 1998a; Yu et al., 1998). Therefore, these results strongly suggest that Sema-1a/PlexA repulsive signaling contributes to the formation of the ISNb branch. This idea is further supported by strong genetic interactions between Sema-1a and FasII, and conn (Yu et al., 2000). The fusion bypass phenotype observed in Sema-1a mutants is enhanced in a synergistic manner by increased expression of FasII, but suppressed by loss-of-function mutations in conn and/or FasII (Yu et al., 2000). These findings support the idea that the selective formation of five nerve branches in the PNS depends on the combined action of Sema-1a-mediated repulsion and CAMs-induced adhesions.

Interestingly, highly penetrant fusion bypass phenotypes of the ISNb were also found in Dptp69D Dptp99A double mutant embryos (Fig. 3B; Desai et al., 1996). Both Dptp69D and Dptp99A encode receptor-linked protein tyrosine phosphatases (RPTPs) which during embryonic neural development are expressed on motor axons that include ISNb, ISN, and SNa axons (Desai et al., 1996; Hamilton et al., 1995). The motor axon guidance phenotypes observed in double mutant embryos suggest that these RPTP proteins play an important role in defasciculation at specific choice points (Desai et al., 1996). However, it is still unknown if Dptp69D and Dptp99A directly regulate axon–axon interactions, because to date, no ligands for these RPTPs have been discovered (Arzan Zarin and Labrador, 2019). The most valuable insights into the formation of five nerve branches that include ISN, ISNb, and SNa emerge from the genetic ablation of embryonic muscles (Landgraf et al., 1999; Prokop et al., 1996). In the absence of muscles, peripheral nerve branches, including the ISNb and SNa, fail to defasciculate from the main nerve bundles, resulting in a fusion bypass phenotype (Fig. 3B; Landgraf et al., 1999; Prokop et al., 1996). These genetic ablation studies further demonstrate that founder myoblasts produce specific defasciculation cues required for the formation of peripheral nerve branches (Landgraf et al., 1999). Taken together with the axon guidance functions of CAMs, Sema-1a/PlexA, and the two RPTPs mentioned above, these results suggest that the selective defasciculation of motor axons in the periphery is controlled by the cooperation between axon–axon adhesion and selective axon–axon repulsion in response to muscle-derived guidance cues at specific choice points.

After five nerve branches that include ISN, ISNb, and SNa are formed in the periphery, each projects to its target muscle region, and then its bundled axons sequentially defasciculate at specific choice points to innervate target muscle fibers (Fig. 1). With respect to the selective defasciculation of motor axons for targeted muscle innervation, this review will for several reasons primarily focus on the ISNb motor axon guidance and its regulation. First, when stained with anti-FasII antibody, the visibility of the axonal projection pattern of the ISNb is greater than that of other nerve branches (Jeong, 2017; Van Vactor et al., 1993). In addition, the embryonic pattern of ventrolateral muscles 7, 6, 13, and 12, which are innervated by ISNb motor axons (Fig. 2A), is also well recognized under differential interference contrast (DIC) microscopy (Van Vactor et al., 1993). These patterns appear to be very helpful for identifying more subtle axon pathfinding defects. Second, no nonmuscle mesodermal cells, which could function as guidepost cells, have yet to be recovered in the peripheral ISNb pathway (Van Vactor et al., 1993). Therefore, axon–axon and axon–muscle interactions seem to play an important role in precise navigation of the ISNb motor axons. Third, it is likely that no pioneer axon is required for axon pathfinding and growth of the ISNb (Krueger et al., 1996; Van Vactor et al., 1993). This may suggest that each growth cone of the ISNb is able to differentially respond to guidance cues emanating from the surrounding, and also able to recognize its own target muscle(s) (Krueger et al., 1996). Fourth, loss-of-function studies have identified a relatively larger number of guidance molecules required for correct axon pathfinding in the ISNb pathway, compared to other nerve pathways (Arzan Zarin and Labrador, 2019). This could in part be ascribed to the absence of guidepost cell and pioneer axon in the ISNb (Krueger et al., 1996; Van Vactor et al., 1993). Finally, compared to other peripheral nerve branches, more severe and diverse guidance phenotypes seem to be observed in the ISNb (Desai et al., 1996; Fambrough and Goodman, 1996; Krueger et al., 1996; Sink et al., 2001; Van Vactor et al., 1993; Winberg et al., 1998a; Yu et al., 1998). Therefore, when one single gene or one copy of a gene is absent, subtle axon guidance errors could easily be detected. This is probably due to relatively low levels of genetic redundancy, which indicate that the individual guidance molecule retains distinct non-overlapping guidance functions.

The question then arises as to how motor axons can recognize defasciculation choice points. The recognition of specific choice points should be a prerequisite for the selective defasciculation of motor axons. Therefore, the fusion bypass phenotype observed in genetically muscle-ablated embryos indicates that muscle founder cells and muscle-derived cues seem to play an essential role in recognizing defasciculation choice points (Landgraf et al., 1999; Prokop et al., 1996). Moreover, the presence of a single founder cell in muscle-ablated embryos is sufficient to induce defasciculation of motor axons, further supporting this idea (Landgraf et al., 1999). The absence of both Dptp69D and Dptp99A results in highly penetrant fusion bypass phenotypes of the ISNb, which are like those of genetic muscle ablation (Fig. 3B; Desai et al., 1996; Landgraf et al., 1999). These findings may suggest that RPTP proteins expressed on motor axons both serve as signaling molecules required for the recognition of specific choice points, and seemingly the selective axon–axon defasciculation. In addition, Sema-1a/PlexA signaling may also be involved in the recognition of choice points due to the fusion bypass phenotypes, even though these phenotypes are occasionally observed in either Sema-1a or PlexA mutants (Fig. 3B; Winberg et al., 1998a; Yu et al., 1998). However, the most frequently observed guidance defect in Sema-1a mutants is a stall phenotype, in which the ISNb largely stalls between muscles 6 and 13 along the ISNb pathway (Fig. 3D; Jeong, 2017; Yu et al., 1998). This stall phenotype of Sema-1a mutants may reflect that the mutant ISNb axons still possess the ability to recognize specific choice points, and to respond differentially to multiple guidance cues emanating from muscles. Therefore, the differential responsiveness of ISNb axons to surrounding guidance cues, combined with the failure in axon–axon repulsion at specific choice points, could lead to stall phenotypes in Sema-1a mutants. Similar stall phenotypes were also detected in the absence of PlexA, suggesting that Sema-1a/PlexA signaling is required for the selective defasciculation of ISNb axons at specific choice points (Fig. 3D; Jeong et al., 2012; Winberg et al., 1998a; Yang et al., 2016). The secreted semaphorins Sema-2a and Sema-2b, which both function as ligands for PlexB, are required for proper defasciculation of the ISNb axons at the last choice point (Ayoob et al., 2006; Roh et al., 2016; Wu et al., 2011). Moreover, Sema-2a genetically interacts with Netrin A, Netrin B, and FasII in the control of target selection by the ISNb axons (Winberg et al., 1998b). These results demonstrate that Semaphorins/Plexins work together with other guidance molecules to regulate motor axon pathfinding and/or target recognition. On the other hand, numerous axon guidance molecules are also involved in the synaptic partner choice (Sanes and Yamagata, 2009; Sanes and Zipursky, 2020). In many cases, it is difficult to distinguish between their axon pathfinding and synaptic target recognition functions. Therefore, synaptic recognition within target muscle region will not be addressed here.

In addition to fusion bypass phenotypes, different types of pathfinding defects that include U-turn, split/detour, and split/stall phenotypes (Figs. 3E and 3F), were observed in Dptp69D Dptp99A double mutants, strongly suggesting that while Dptp69D and Dptp99A are also responsible for making the correct pathfinding decisions, they barely contribute to the recognition of muscle targets (Desai et al., 1996). In contrast, Drosophila Lar (Dlar), which is a member of type IIa RPTPs, plays an important role in recognizing muscle targets or fields (Johnson and Van Vactor, 2003; Krueger et al., 1996). In Dlar mutants, ISNb axons normally defasciculate from the ISN at the initial choice point, but frequently fail to enter their target muscle fields, and instead extend along the ISN pathway as a separate bundle, referred to as the “parallel bypass phenotype” (Fig. 3C; Krueger et al., 1996). Interestingly, this parallel bypass phenotype is robustly suppressed by the loss of Dptp99A (Desai et al., 1996). Given that no ISNb guidance defects were observed in Dptp99A null mutants (Hamilton et al., 1995), the negative genetic interactions between Dlar and Dptp99A, along with the positive genetic interactions between Dptp69D and Dptp99A, demonstrate that Dptp99A collaborates in a redundant manner with Dptp69D and Dlar to recognize defasciculation choice points and target muscles, and to make the correct pathway decisions. Other types of RPTPs, such as Dptp10D, Dptp52F, and Dptp4E, were shown to be involved in motor axon guidance (Arzan Zarin and Labrador, 2019). Dptp10D interacts genetically with Dptp69D, Dptp99A, and Dlar in potentiating ISNb defaciculation (Sun et al., 2001). Dptp52F also acts together with Dptp10D and Dptp69D to regulate defasciculation of ISNb axons at specific choice points (Schindelholz et al., 2001). Dptp4E Dptp10D Dptp69D triple mutants show genetic enhancement of clump phenotypes observed in the ISNb pathway (Jeon et al., 2008). These complex genetic interactions among RPTPs reflect their functional redundancy, and further demonstrate that a combinatorial code of RPTPs controls the recognition of defasciculation choice points and the pathfinding decision-making.

Highly penetrant bypass phenotypes of the ISNb were also observed in either beaten path-Ia (beat-Ia) or sidestep (side) mutant embryos (Fambrough and Goodman, 1996; Sink et al., 2001). Both beat-Ia and side encode membrane-tethered immunoglobulin superfamily (IgSF) proteins that form a complex through direct protein–protein interaction (Li et al., 2017; Özkan et al., 2013; Siebert et al., 2009). Beat-Ia protein is mainly detected on motoneuron axons and growth cones, while Side is expressed on all muscles and sensory neurons as guidance substrates in the periphery (Fambrough and Goodman, 1996; Sink et al., 2001). Genetic and immunohistochemical analyses propose that Beat-Ia functions as a receptor to recognize Side-expressing muscles, and then to mediate growth cone turning of the ISN pioneer neurons (Siebert et al., 2009). However, it remains to be determined whether the same mechanism underlies the ISNb motor axon guidance, since beat-Ia and side mutations seem to result in a different bypass phenotype (Fambrough and Goodman, 1996; Sink et al., 2001). Parallel bypass phenotype often appears to be observed in beat-Ia mutant embryos, whereas fusion bypass phenotype is frequently found in side mutants (Figs. 3B and 3C; Fambrough and Goodman, 1996; Sink et al., 2001). In addition, FasII interacts genetically with beat-1a, but not with side (Fambrough and Goodman, 1996; Sink et al., 2001). Lastly, a negative genetic interaction between beat-Ia and side was observed in the larval innervation pattern of the ISNb neurons in muscles 12 and 13, but not in muscles 6 and 7 (Siebert et al., 2009). These observations might suggest that Beat-Ia and Side not only mediate a common signaling pathway, but also have distinct functions in the ISNb pathfinding. The fly genome contains a total 14 beat-like genes and 8 side-like genes, showing a Beat/Side protein interaction network (Li et al., 2017; Özkan et al., 2013; Pipes et al., 2001). Several other members of the Beat subfamily, including Beat-Ic, also appear to be involved in ISNb motor axon guidance (Pipes et al., 2001). In contrast, the functional roles of other Side subfamily members have not yet been characterized.

One of the most prominent mechanisms underlying motor axon guidance is the selective fasciculation and defasciculation that occurs at specific choice points. At least two criteria for selective fasciculation should be fulfilled. First, the individual growth cones of motor neurons born at different locations must be guided to meet each other or pioneer axon(s). Second, these axons associate to form a fascicle via axon–axon attraction. Selective defasciculation of motor axons preferentially requires the recognition of defasciculation choice points in response to muscle-derived guidance cues. Multiple muscle-derived guidance cues should not only induce selective axon–axon repulsion, but also mediate attractive axon guidance in a differential manner. Once motor axons selectively defasciculate at specific choice points, additional guidance receptor molecules mainly expressed on the growth cones are involved in making a correct pathfinding decision for targeted muscle innervation in response to muscle-derived guidance cues.

One unsolved question in neural development is how the complex and stereotyped patterns of neural circuits observed in many higher organisms can be shaped by a limited number of guidance molecules (Dickson, 2002). This could be explained by several mechanisms that include combinatorial codes of guidance molecules, differential regulation of guidance receptors and their ligands, and the reiterative use of a limited set of guidance molecules (Bonanomi and Pfaff, 2010; Dickson, 2002; Pasterkamp, 2012). In addition, I here propose that selective fasciculation and defasciculation contribute much to the formation of complex and precise neural circuits with a relatively small number of guidance molecules, based on the following reasons. First, selective fasciculation of growing follower axons ensures their correct pathfinding between fasciculation and defasciculation points, even in the absence of the signaling mechanisms necessary for axon pathfinding. In the Drosophila ISNb nerve in which no pioneer axon is found, interestingly, the individual axon appears to respond to guidance cues present along the pathway. Therefore, fasciculated ISNb axons are enabled to respond better to surrounding guidance cues, so that they can arrive at defasciculation choice points in a more coherent manner. Second, the recognition of each defasciculation point, which seems to be mediated by combinatorial codes of guidance molecules, is tightly associated with selective defasciculation. These combinatorial mechanisms of guidance molecules might produce a much larger number of divergent cellular outputs for axonal segregation. Third, once axons selectively defasciculate at specific choice points, segregated axons enter their own target field to choose correct synaptic partners, largely through the short-range interactions between growth cones and potential targets (Sanes and Yamagata, 2009). These short-range interactions within a restricted target region are likely to require a lesser number of guidance molecules. Lastly, the myriad number of nerves and nerve tracts observed in the mammalian PNS and central nervous system (CNS) highlights the importance of selective fasciculation and defasciculation in their complex wiring (Chédotal and Richards, 2010; Luxey et al., 2020; Takahashi et al., 2010; Thiebaut de Schotten et al., 2015), and also suggests that the cellular strategy using selective fasciculation and defasciculation should increase the fidelity of axon pathfinding, even with the relatively small number of guidance molecules in mammals. Given that major classes of axon guidance molecules serve evolutionarily conserved guidance functions across the animal kingdom (Bashaw and Klein, 2010; Dickson, 2002; Kolodkin and Tessier-Lavigne, 2011), in summary, the molecular and cellular mechanisms underlying selective fasciculation and defasciculation could be highly conserved from flies to mammals.

This work was supported by a National Research Foundation of Korea (NRF) grant funded by the Korea government (MIST) (2018R1A2B6008037), and research funds of Jeonbuk National University in 2019.

S.J. contributed to the manuscript preparation.

The author has no potential conflicts of interest to disclose.