Introduction

Pathfinding axons are directed to their appropriate synaptic targets by a variety of guidance cues. While approaching or traversing the target field, the growth cones of migrating axons encounter secreted or cell surface-attached ligands and respond by steering toward or away from their sources (1–4). Fine-grained control of these factors is critical for the establishment of proper neural circuitry.

The zebrafish’s lateral line provides a tractable in vivo system for exploring the mechanisms that guide the assembly of neural circuits in a peripheral nervous system. In particular, it is possible to detect individual hair cells and the sensory axons that innervate them throughout the first week of development (5). The primary posterior lateral line on each side of the zebrafish’s tail consists of about seven neuromasts, which contain mechanoreceptive hair cells of opposing polarities—half sensitive to headward (rostrad) water motion and the complementary half to tailward (caudad) water motion—that cluster at their centers (6). The sensory axons of the lateral-line nerve branch, arborize, and consolidate around the basolateral surfaces of the hair cells (Figure 1A). To aid in an animal’s swimming behavior, these hair cells sense water currents and relay signals to the brain through the sensory axons (7). The well-defined structure of the ramifications and synaptic contacts of the sensory axons highlight the need to determine the molecular signals behind the assembly of such precise microcircuits (8). We adopted a candidate-gene strategy to seek factors that direct the growth of afferent growth cones in the zebrafish’s lateral line. Analysis of single-cell RNA sequencing data identified the semaphorin7a (sema7a) gene to be highly expressed in hair cells of that organ (9). Semaphorins are important regulators of axonal growth and target finding during the patterning of diverse neural circuits (10). The semaphorin family includes proteins that are secreted, transmembrane, and cell surface-attached (11). Among these, Semaphorin7A (Sema7A) is the only molecule that is anchored to the outer leaflet of the lipid bilayer of the cell membrane by a glycosylphosphatidylinositol (GPI) linker (11). Unlike many other semaphorins, which act as repulsive cues (12,13), Sema7A is involved in promoting axon growth (4) and imparting directional signals by interacting with the integrin and plexin families of receptors residing on the pathfinding axons (14–16). Sema7A likely occurs in vivo in both GPI-anchored and soluble forms (17–19), and studies of neuronal explants confirm that both forms can induce directed axonal outgrowth (4). Furthermore, it has been proposed that the GPI anchor is cleaved by membrane-resident GPI-specific phospholipases (GPI-PLs) or matrix metalloproteases to release the Sema7A into the extracellular environment and thus regulate dynamic cellular processes (20,21). Because these observations indicate that Sema7A might influence neuronal development both as a juxtracrine and as a diffusive signal, we investigated the role of Sema7A in sculpting a vertebrate peripheral sensory organ.

Figure 1.

Results

Sema7A expression and localization in hair cells

The zebrafish genome contains a single sema7a gene that produces two transcripts. One transcript encodes Sema7A-GPI, a full-length, GPI-linked, cell surface-attached protein. The second transcript yields Sema7A^sec, a protein with a truncated C-terminus, which we conjecture is secreted into the local environment. Each transcript encodes an N-terminal signal sequence and a single copy of the conserved sema domain (11). Using primers that flank the regions encoding the sema domain and the distinct C-termini of the two transcripts, we performed reverse transcription and polymerase chain reactions (RT-PCR) to identify both the membrane-anchored and the secreted transcript in developing larvae (Figure 1B; Figure 1—figure supplement 1A). Single-cell RNA sequencing data also have shown that the sema7a transcript is particularly enriched in the immature and mature hair cells of the neuromast during early larval development (9), a period when sensory axons of the posterior lateral line contact hair cells and form synapses with them (5,22). Indeed, we have observed Sema7A to be specifically localized in both mature and immature hair cells (Figure 1C; Figure 1—figure supplement 1B-E).

As neuromasts mature, the contacts between sensory axons and hair cells increase in number, become stabilized, and establish synapses (5,22). Because Sema7A can modulate axon guidance (4,14) and synapse formation (15), we wondered whether the level of Sema7A also changes during neuromast development. Upon quantifying the average intensity of Sema7A, we found similar expression in rostrally and caudally polarized hair cells of neuromasts at 1.5, 2, 3, and 4 days post fertilization (dpf). The average Sema7A intensity increased significantly over this period in both rostrally and caudally polarized hair cells. At 1.5 dpf, the neuromasts harbor primarily immature hair cells that make minimal contacts with sensory axons and do not form stable association with axonal terminals (22). At this stage the average Sema7A intensity in hair cells remained low. By 2-4 dpf the hair cells mature to form stable contacts and well-defined synapses with the sensory axons (5,23). The average Sema7A intensity in hair cells at each of these stages rose to levels that significantly exceeded those of 1.5 dpf neuromasts (Figure 1D). This rise in the amount of Sema7A during neuromast maturation supports a role for Sema7A in the guidance of sensory axons and their interaction with hair cells.

We additionally demonstrated an anisotropic distribution of Sema7A along the apicobasal axis of each hair cell. As neuromasts develop through 2, 3, and 4 dpf, Sema7A remained highly enriched in the subapical region of the hair cell, the site of the Golgi network. This localization suggests that Sema7A—like other proteins of the semaphorin family (24)—is directed to the plasmalemma through Golgi-mediated vesicular trafficking. While Sema7A maintained a high level in the subapical region, the protein increasingly accumulated at the base of the hair cell: the Sema7A level at the hair cell base was low at 2 dpf, but increased sharply by 3 dpf and 4 dpf (Figure 1E; Figure 1—figure supplement 1F, G). During this period the hair cells increase in number and mature, whereas the associated sensory axons arborize to contact multiple hair cells and form robust synaptic boutons at their bases (5,23). The progressive enrichment of Sema7A at the hair cell base and sensory-axon interface (Figure 1E, F; Video 1) suggests that Sema7A acts as a mediator of contacts between sensory axons and hair cells.

Sema7A restricts sensory axons around clustered hair cells

If Sema7A guides and restricts the sensory arbors at and around the hair cells clustered in a neuromast, then inactivating Sema7A function should disrupt this process. To test this hypothesis, we obtained a sema7a mutant allele (sema7a^sa24691) with a point mutation that introduces a premature stop codon in the conserved sema domain (Figure 2—figure supplement 1A). The homozygous mutant, hereafter designated sema7a^−/−, should lack both the secreted and the membrane-attached forms of Sema7A. The average intensity of Sema7A immunolabeling in hair cells diminished by 61% in sema7a^−/−larvae (0.34 ± 0.01 arbitrary units) in comparison to controls (0.87 ± 0.01 arbitrary units). In a few cases we observed minute accumulations of Sema7A at the subapical region, but little or no protein at the base of the mutant hair cells (Figure 2A; Figure 2—figure supplement 1B).

Figure 2.

To characterize the impact of the sema7a^−/− mutation on arbor patterning, we utilized doubly labeled transgenic fish Tg(myo6b:actb1-EGFP;neurod1:tdTomato) that marked the hair cells and the sensory axons of the posterior lateral line with distinct fluorophores (25). In control neuromasts, the sensory axons approached, arborized, and consolidated around the bases of clustered hair cells (Figure 2B). Although the sensory axons approached the hair cells of sema7a^-/- neuromasts, they displayed aberrant arborization patterns with wayward projections that extended transversely to the organ (Figure 2C). To quantitatively analyze the arborization morphology, we traced the sensory arbors in three dimensions to generate skeletonized network traces that depict–as pseudocolored trajectories–the increase in arbor length from the point of arborization (Figure 2D, E).

To visualize the hair cell clusters and associated arborization networks from multiple neuromasts, we aligned images of hair cell clusters, registered them at their hair bundles, from neuromasts at 2 dpf, 3 dpf, and 4 dpf. Throughout development, the arborization networks of control neuromasts largely remained within the contours of the hair cell clusters; the few that reached farther nonetheless lingered nearby (Figure 2F). This result suggests that an attractive cue retained the axons near the sensory organ. In sema7a^-/- neuromasts, however the neuronal arbors failed to consolidate within the boundaries of hair cell clusters and extended far beyond them (Figure 2G; Figure 2—figure supplement 1C-H; Video 2).

We quantified the distribution of the sensory arbors around the center of the combined hair cell clusters. For both the control and the sema7a^-/- neuromasts, the arbor densities peaked proximal to the boundaries of the hair cell clusters. Beyond the cluster boundaries, in control neuromasts at 2 dpf, 3 dpf, and 4 dpf the arbor densities fell sharply at respectively 31.7 µm, 36.6 µm, and 32.7 µm from the center. In sema7a^-/- neuromasts of the same ages, the sensory arbors extended respectively 38.5 µm, 51.4 µm, and 56.6 µm from the center (Figure 2H; Figure 2—figure supplement 1I, J; Figure 2—figure supplement 2A). Furthermore, we quantified the degree of contact of the sensory arbors to their hair cell clusters in individual neuromasts from both control and sema7a^-/-mutants. In control neuromasts, the degree of contact was 83 ± 1 % (mean ± SEM) at 2 dpf and 84 ± 1 % at 3 dpf, but significantly increased to 90 ± 1 % at 4 dpf (Figure 2I; Figure 2—figure supplement 2B). This observation indicates that the sensory arbors reinforce their association to the hair cell clusters as neuromasts develop. However, in sema7a^-/- neuromasts such contact was significantly reduced. Only 69 ± 4 % of sensory arbors at 2 dpf were closely associated with the corresponding hair cell clusters, a value that remained at 67 ± 2 % at both 3 dpf and 4 dpf (Figure 2I). These findings signify that, with simultaneous disruption of both signaling modalities of Sema7A, the organ fails to restrict the localization of sensory arbors and has significantly diminished contact with them.

Sema7A patterns the arborization network’s topology

Accurate spatial arrangement of the neuronal projections around hair cells is essential for proper retrieval and transmission of sensory information. Mutations that perturb proper assembly of sensory neuronal microcircuits lead to hearing impairment (26). Quantitative insight into the wiring pattern is therefore essential to identify systematic aberrations in the microcircuit arising from genetic deficits. The three-dimensional skeletonized network traces allowed us to accurately delineate the microcircuit topology across individuals, developmental stages, and genotypes from living animals (Figure 2D, E). To quantitatively analyze the arbor network, we defined five topological attributes: (1) nodes, denoted as junctions where arbors branch, converge, or cross; (2) loops, comprising minimal arbor cycles forming topological holes; (3) bare terminals, defined as arbors with free ends; (4) the node degree, calculated as the number of first-neighbor nodes incident to a node; and (5) the arbor curvature, measured as the total curvature of a bare terminal (Figure 3A). We quantified these attributes to accurately determine how the circuit topology is established during maturation of the sensory organ in control larvae and how it is impacted by sema7a deficit.

Figure 3.

To determine how internode connectivity is distributed across networks, we quantified the distribution of node degrees along the relative height of the hair cell cluster across individuals, developmental stages, and genotypes. As for most other real-world networks (27), we observed a highly skewed node-degree distribution: the majority of nodes had low node degrees (three and four) and a small number of nodes—often acting as hubs, like the arborization initiation point—had high node degrees (five to seven). The node degree varied from three to seven in control larvae and three to six in sema7a^−/− larvae. Across developmental stages and genotypes, the high node degrees primarily occurred at the base (region from 0 % to 10 % of the cluster height) of the hair cell cluster and acted as hubs for the majority of branching events. Low node degrees primarily occurred along the basolateral region (region from 10 % to 60 % of the cluster height) of the hair cell cluster. The apical region (region from 60 % to 100 % of the cluster height) was sparsely populated with low node degrees. The network’s hierarchical organization along the organ’s axis suggests that arbors receive sensory inputs at multiple levels, allowing the network to carry out robust neural computations. Even though nodes of low degree were most prevalent between genotypes across developmental stages, their distribution along the relative organ axis became significantly different at 4 dpf, with nodes in control larvae localized higher along the axis than in sema7a^−/− larvae (Figure 3A; Figure 3—figure supplement 1A-C; Table 1). These observations indicate that Sema7A regulates both the spatial organization and node-degree complexity of the sensory arbor network.

Table 1.

An average of 9.3 sensory axonal branches approach the hair cell cluster of a neuromast (8). Pathfinding of sensory axons around the hair cell cluster results in the formation of axonal fascicles that form numerous loops and bare terminals, which determine the microcircuit topology (Figure 2B, D) (8). To optimally surround and contact the hair cell cluster, do the sensory arbors prefer to adopt one topological feature over the other? To address this question, we measured the correlation between the number of bare terminals and the number of loops across individuals, developmental stages, and genotypes. In both 2 dpf and 3 dpf larvae, the distributions remained highly scattered across individuals and genotypes. However, by 4 dpf in the control larvae the arbor network contained either a high number of loops with a low number of bare terminals or a high number of bare terminals with a low number of loops. This result suggests that the stereotypical circuit topology observed in the mature organ emerges from distinct events in which individual arbors transition from forming bare terminals to forming closed loops encircling topological holes. That we did not observe a similar transition in sema7a^-/- neuromasts of the same age suggests that Sema7A regulates the switch between these two distinct topological features (Figure 3B; Figure 3—figure supplement 1D, E; Table 1).

Measurement of the contour length of both bare terminals and loops additionally revealed that across developmental stages in control larvae the sensory arbors followed a three-step process to choose the appropriate topological feature: arbors shorter than 10 µm always formed bare terminals; arbors between 10 µm and 30 µm began to switch between features, with arbors longer than 20 µm strongly preferring to form loops; and arbors longer than 30 µm always formed loops. These results imply that the switch between bare terminals and loops is partly dependent on the length of the arbors. However, this characteristic topological preference was significantly perturbed in the sema7a^−/− larvae across developmental stages, as many of the sensory arbors longer than 30 µm remained as bare terminals (Figure 3C; Figure 3—figure supplement 1F, G; Table 1). Furthermore, in 4 dpf control neuromasts the majority of the loops contained four nodes and utilized 30-40 µm of arbors to form closed cycles. These conserved topological attributes were significantly perturbed in the sema7a^−/−larvae of same age (Figure 3D, E; Figure 3—figure supplement 2A-D; Table 1). These findings indicate that Sema7A tightly regulates the sensory arbor loop formation, including the preferred loop length and number of nodes per loop, to properly surround and contact the hair cell cluster.

As observed previously, bare terminals in control neuromasts project transverse to the organ and linger nearby, often curving back to form new loops and contact the hair cell cluster (Figure 2D, F). To analyze the propensity of the bare terminals to curl backward, we measured their total curvatures, defined as the sum of the curvature at each point along the length of an arbor. Henceforth, we address the total curvature as the curvature of a bare terminal. We classified the bare terminals in two categories: high-curvature arbors with curled morphology and curvature greater than π/6 (30 degrees); and low-curvature arbors with linear morphology and curvature less than π /6. In control neuromasts 44.8 % of arbors had low curvature compared to 56.4 % in sema7a^-/- neuromasts at 4 dpf. Concomitantly, 55.2 % of arbors had high curvature in control neuromasts compared to 43.6 % in sema7a^-/- neuromasts at 4 dpf (Figure 3F; Figure 3—figure supplement 2E, F; Table 1). These observations further confirm our hypothesis that a Sema7A-mediated attractive cue is involved in orienting bare neural arbors toward the hair cell cluster to potentially form loops.

In summary, we propose that Sema7A regulates four key aspects of the topology of the arborization network: (1) the node degree complexity along the organ’s axis; (2) the transition of arbors from bare terminals to loops; (3) the preference for forming loops containing four nodes with 30-40 µm of arbors; and (4) the preference of bare terminals to adopt higher curvature.

Sema7A^sec is a sufficient chemoattractive cue for sensory axon guidance

Because genetic inactivation of the sema7a gene disrupts both signaling modalities of the cognate protein, we could not assess the specific activity of the Sema7A^sec diffusive cue in guiding lateral-line sensory axons. To independently verify the potential role of that isoform, we therefore expressed the secreted variant ectopically and investigated the resultant arborization patterns.

Analysis of the microcircuit connectivity of the neuromasts has demonstrated bare sensory-axonal terminals in the perisynaptic compartments, where they do not contact the hair cell membrane (8). We speculate that these bare terminals are attracted by a Sema7A^sec diffusive cue. To test this hypothesis, we ectopically expressed Sema7A^sec protein tagged with the fluorophore mKate2, then analyzed its effect on the morphology of sensory arbors. Fertilized one-cell embryos were injected with the hsp70:sema7a^sec-mKate2 plasmid and raised to 3 dpf. Larvae expressing the transgenesis marker—the lens-specific red fluorescent protein mCherry—were heat-shocked, incubated to allow expression, and subsequently mounted for live imaging. In these larvae, a random mosaic of cells—often embryonic muscle progenitors in the dermomyotome or mature myofibers—expressed the exogenous Sema7A^sec-mKate2 protein. We selected only those neuromasts in which an ectopic integration had occurred near the network of sensory arbors (Figure 4A,B). In all 22 such cases, we observed robust axonal projections from the sensory arbor network toward the ectopically expressing Sema7A^sec cells. When the exogenous sema7a^sec integration occurred in embryonic muscle progenitor cells, which reside in a superficial layer external to the muscle fibers and adjacent to the larval skin (28), the projections were able to form direct contacts with them (Figure 4C; Videos 3, 4). But more often, when mature myofibers expressed the exogenous Sema7A^sec deep in the myotome, the axonal projections approached those targets but failed to contact them (Figure 4D, Videos 5, 6). This behavior likely arose because other components in the myotomal niche—including the fibrous epimysium, other myofibers, myoblasts, and immature myotubes whose surroundings are permeable to diverse diffusive cues (29)—physically obstructed direct interaction of the extended neurites with the ectopically expressing myofibers. All 18 of the injected but not heat-shocked control larvae did not express ectopic Sema7A^sec, and we did not observe aberrant projections from the sensory arbor network (Figure 4E).

Figure 4.

To measure the accuracy of the extended axonal projections in finding the ectopic Sema7A^sec source, we defined three parameters: (1) the source path, denoted as the linear distance from the point of arborization to the boundary of an ectopically expressing Sema7A^sec cell; (2) the projection path, representing the linear distance from the point of arborization to the terminus of the extended projection; and (3) the projection proximity, calculated as the linear distance from the terminus of the extended axonal projection to the nearest boundary of the ectopically expressing Sema7A^sec cell (Figure 4B). We quantified these parameters from 18 of the 22 mosaic integration events; in the remainder the extended axonal projections– although following the ectopic source–reentered the posterior lateral-line nerve so that identification of the axon terminals was not possible (Figure 4—figure supplement 1A-A”; Video 7). When we plotted the projection path length against the source-path length, we observed a nearly perfect correlation (Figure 4F). This result signifies that irrespective of the location of the ectopic source–whether proximal or distal to the sensory arbor network–the Sema7A^sec cue sufficed to attract axonal terminals from the sensory arbor. Furthermore, the low average projection-proximity length of 3.5 ± 0.7 µm (mean ± SEM) indicated that the extended projection terminals remained either in contact (8 of 18) or in the vicinity (10 of 18) of the ectopic Sema7A^sec sources (Figure 4G). These observations together demonstrate that the Sema7A^sec diffusive cue is sufficient to provide neural guidance in vivo.

As the posterior lateral-line ganglion matures, newly formed neurons extend their axonal growth cones along the posterior lateral-line nerve to reach the neuromasts (30). We wondered whether the axonal projections that had yet to be associated with a neuromast could also respond to an ectopic source of Sema7A^sec. Indeed, on five occasions in which sema7a^sec integration had occurred between two neuromasts, we detected neurite extensions from the posterior lateral-line nerve to the ectopic Sema7A^sec source (Figure 4—figure supplement 1B-B”; Videos 8-10). The responsiveness of the posterior lateral-line axons to Sema7A^sec might therefore be an intrinsic property of the neurons that does not require the neuromast microenvironment.

Sema7A deficiency impacts synaptic assembly

Contact stabilization between an axon and its target cell is essential for synapse formation (31). Neural GPI-anchored proteins can act as regulators of various synaptic-adhesion pathways (32). In the mouse olfactory system, GPI-anchored Sema7A is enriched in olfactory sensory axons and mediates sustained interaction with the dendrites of mitral and tufted cells in the olfactory bulb to establish synapses (15). In the developing mouse brain–particularly in Purkinje cells–GPI-anchored Sema7A instead regulates the elimination of synapses onto climbing fibers (33). The role of anchored Sema7A in regulating synaptic architecture can thus vary according to the cell type and developmental stage. Our demonstration that Sema7A is necessary to consolidate the contact between sensory axons and hair cells during neuromast development suggests a possible role in regulating synaptic structure.

Proper synaptogenesis requires the correct spatial organization of the presynaptic and postsynaptic apparatus at the apposition of two cells. At the interface between hair cells and sensory axons, the synaptic network involves two scaffolding proteins in particular: ribeye, a major constituent of the presynaptic ribbons (34), and membrane-associated guanylate kinase (MAGUK), a conserved group of proteins that organize postsynaptic densities (35). Utilizing these scaffolding proteins as markers, we investigated the impact of the sema7a^-/- mutation on synapse formation. To analyze the distribution of presynaptic densities–immunolabeled with an antibody against the ribeye-associated presynaptic constituent C-terminal binding protein (CTBP) (34)–we counted the number and measured the area of CTBP punctae from multiple neuromasts at several stages of neuromast development (Figure 5A, B). The CTBP punctae in each hair cell ranged from zero to seven across developmental stages. In the zebrafish’s posterior lateral line, the formation of presynaptic ribbons is an intrinsic property of the hair cells that does not require contact with lateral-line sensory axons. However, sustained contact with the sensory axon influences the maintenance and stability of ribbons (36). As neuromasts matured from 2 dpf to 4 dpf, we identified similar distribution patterns with a characteristic increase in the number of CTBP punctae in both control and sema7a^-/-larvae (Figure 5C, D). This result implies that the formation of new presynaptic ribbons is not perturbed in the mutants, even though there are on average fewer CTBP punctae (Figure 5—figure supplement 1A-C). Measurement of the areas of CTBP punctae also showed similar distributions in both genotypes (Figure 5E, F) but the average area of presynaptic densities was reduced in the mutants (Figure 5—figure supplement 1D-F).

Figure 5.

Because GPI-anchored Sema7A lacks a cytosolic domain, it is unlikely that Sema7A signaling directly induces formation of presynaptic ribbons. Instead, GPI-anchored Sema7A might induce postsynaptic assembly on the apposed neuronal membrane (15). To characterize the distribution of the postsynaptic aggregates–immunostained with a pan-MAGUK antibody– we performed analyses similar to those above (Figure 5I, J). At 4 dpf, 68.3 % of the hair cells in the control neuromasts were associated with at least one MAGUK punctum. In the sema7a^-/-mutant this value fell to 37.5 % (Figure 5K, L). The average number of MAGUK punctae also declined in the mutants (Figure 5—figure supplement 1G-I). As control neuromasts matured from 2 dpf to 4 dpf, the MAGUK punctae showed a shift toward smaller sizes, suggesting that the postsynaptic structure fragmented into smaller entities as observed in other excitatory synapses (37). At 4 dpf, 77.5 % of the hair cells in the control neuromasts had postsynaptic densities with areas between 0.10 µm² and 0.40 µm². In the sema7a^-/- mutant the corresponding value was only 42.5 % (Fig. 4M,N). The average area of MAGUK punctae was also reduced in the mutants compared to the controls (Figure 5—figure supplement 1J-L). We speculate that the abnormalities in postsynaptic structure arose from reduced contact between the hair cells and the sensory axons in the sema7a^-/-mutants or from the lack of Sema7A-mediated juxtracrine signaling onto the apposed neurite terminals (15).

Discussion

The specification of neural circuitry requires precise control of the guidance cues that direct and restrict neural arbors at their target fields and facilitate synapse formation. In the lateral-line neuromast of the zebrafish, we have identified Sema7A as an essential cue that regulates the interaction between hair cells and sensory axons. We have discovered dual modes of Sema7A function in vivo: the chemoattractive diffusible form is sufficient to guide the sensory arbors toward their target, whereas the membrane-attached form likely participates in sculping accurate neural circuitry to facilitate contact-mediated formation and maintenance of synapses. Our results suggest a potential mechanism for hair cell innervation in which a local Sema7A^sec diffusive cue consolidates the sensory arbors at the hair cell cluster and the membrane-anchored Sema7A-GPI molecule regulates microcircuit topology and synapse assembly.

The role of Sema7A in regulating diverse topological features of the arborization network suggests that microcircuits of control and sema7a^-/-neuromasts display important functional differences. Loss of Sema7A disrupts the hierarchical organization of the node-degree distribution along the organ’s axis, as well as the preference towards forming loops containing four nodes with 30-40 µm of arbors. What benefit does the network gain from surrounding the hair cell cluster with closed loops instead of bare terminals? The network may actively form loops to contribute to the neuromast’s overall structure and function, or else emerge as a consequence of the arbors navigating the intervening spaces between the hair cells and support cells. Through our ectopic expression of Sema7A^sec, we have observed that free spaces and physical obstructions can shape arbor morphologies, such as in the myotomal niche. However, we speculate that the precise patterning of the network—with characteristic looping behaviors—is not merely a byproduct of neuromast architecture, but rather contributes to the network’s ability to perform robust neural computations and faithfully transmit sensory information to higher processing centers in the posterior lateral-line ganglion and medial octavolateralis nucleus in the hindbrain (7). Further studies are necessary to dissect the potential role of these topological features in performing robust neural computations in the lateral-line microcircuit.

Recent studies have begun to dissect the molecular mechanisms of semaphorin signaling in axonal navigation. In explants derived from the murine olfactory bulb, Sema7A interacts directly with integrinβ1 to trigger downstream signaling effectors involving focal adhesion kinase and mitogen-activated protein kinases, which are critical regulators of the cytoskeletal network during axonal pathfinding (4,38). Moreover, the interaction between Sema7A and its receptor plexinC1 activates the rac1-cdc42-PAK pathway, which alters the actin network to promote axonal guidance and synapse formation (15,39). We speculate that lateral-line sensory axons utilize similar mechanisms to sense and respond to Sema7A in the establishment of microcircuits with hair cells.

During early development of the posterior lateral line, the growth cones of sensory axons intimately associate with the prosensory primordium that deposits neuromasts along the lateral surface of the larva (40,41). The growth of the sensory fibers with the primordium is regulated by diverse neurotrophic factors expressed in the primordium (42,43). Impairing brain-derived or glial cell line-derived neurotrophic factors and their receptors perturbs directed axonal motility and innervation of hair cells (42,43). Lateral-line hair cells are particularly enriched in brain-derived neurotrophic factor (44) and other chemotropic cues (Figure 6). Because the sensory axons of sema7a^-/- mutants reliably branch from the lateral-line nerve to reach hair cell clusters, these chemoattractive factors might compensate in part for the absence of Sema7A function. Although sensory axons are strict selectors of hair cell polarity and form synapses accordingly (5), diverse chemoattractant cues including Sema7A occur uniformly between hair cell polarities (44). It seems likely that unbiased attractive signals bring sensory arbors toward hair cells, after which polarity-specific molecules on hair cell surfaces dictate the final association between axonal terminal and hair cells. Notch-delta signaling and its downstream effectors regulate hair cell polarity and the innervation pattern (6,8,25). It would be interesting to determine whether Notch-mediated polarity signatures on the hair cell’s surface can refine the neuromast neural circuitry.

Figure 6.

Materials and Methods

Zebrafish strains and husbandry

Experiments were performed on 1.5-4 dpf zebrafish larvae in accordance with the standards of Rockefeller University’s Institutional Animal Care and Use Committee. Naturally spawned and fertilized eggs were collected, cleaned, staged, and maintained at 28.5 °C in system-water with 1 μg/mL methylene blue. Wild-type AB and heterozygous mutant semaphorin7a^sa24691 zebrafish were obtained from the Zebrafish International Resource Center. In addition, the following transgenic zebrafish lines were used: Tg(myo6b:actb1-EGFP) (45), Tg(neurod1:tdTomato) (25), and TgBAC(neurod1:EGFP) (46).

Genotyping of mutant fish

The semaphorin7a^sa24691 allele harbors an A-to-T point mutation in the seventh exon of the semaphorin7a gene, which creates a premature stop codon (Fig. S2A). Genomic DNA was isolated from the tail fins of adult semaphorin7a^sa24691(sema7a^sa91) heterozygous mutant zebrafish. The mutant locus in the genome was PCR amplified using the following primers: sema7a^sa91F: 5′-AAAGCTGGAAAGCGAATCAA-3′ and sema7a^sa91R: 5′-ATATCCAAGGATCCGCCTCT-3′. The 734-base pair amplicons were sequenced, and heterozygous adults were propagated.

Reverse transcription-polymerase chain reactions (RT PCR)

Total RNA was isolated from whole zebrafish larvae by TRIzol extraction followed by DNAse treatment (TURBO DNA-free Ambion). We used thirty 4 dpf zebrafish larvae to isolate total RNA. We generated cDNA libraries using iScript cDNA Synthesis Kit (Bio-Rad). Each transcript was PCR amplified using the following primers: sema7aF: 5′-GGTTTTTCTGAGGCCATTCC-3′, sema7aR1: 5′-GGCACTCGTGACAAATGCTA-3′, sema7aR2: 5′-TGTGGAGAAAGTCACAAAGCA-3′ (Fig. S1A).

Transient transgenesis with the hsp70l:sema7a^sec-mKate2 construct

The pDNR-LIB plasmid containing the sema7a^sec coding sequence was purchased from horizon (7140389, Perkin-Elmer). The coding sequence was PCR amplified using the following primers: sema7a^secF: 5′-GGGGACAAGTTTGTACAAAAAAGCAGGCTTGATGATTCGACATTATTCT-3′ and sema7a^secR: 5′-GGGGACCACTTTGTACAAGAAAGCTGGGTGCTTTGTGGAGAAAGTCACAAAGCA-3′.

The italicized nucleotides denote the mutated stop codon. The amplicon was then cloned into the pDONR221 vector by BP recombination to generate the middle entry pME-sema7a^sec vector. To generate the hsp70:sema7a^sec-mKate2 construct, gateway cloning was performed by combining the plasmids p5E-hsp70 (47), pME-sema7a^sec, p3E-mKate2-myc no-pA (Addgene, 80812), and pDESTtol2pACrymCherry (Addgene, 64023) with LR Clonase II Plus (Invitrogen). Verified constructs (25 ng/μl plasmid DNA) were injected with Tol2 Transposase mRNA (approximately 25 ng/μl) into one-cell embryos. The transiently transgenic larvae— identified by the expression of red fluorescent protein (mCherry) in their lenses—were raised till 3 dpf, heat-shocked in a water bath at 37 °C for 1 hr, incubated at 28 °C for 1 hr to allow expression, and subsequently mounted for live imaging.

Microscopy and volumetric rendering of living neuronal arbors and cells

Living larvae were dechorionated, anaesthetized in 600 μM 3-aminobenzoic acid ethyl ester methanesulfonate in system-water with 1 μg/mL methylene blue, and mounted in 1 % low-melting-point agarose on a glass-bottom MetTek dish at specific stages. The larvae were maintained at 28 °C during imaging in a temperature-controlled stage top chamber (OKO Lab UNO-T). Images at successive focal depths were captured at 200 nm intervals and deconvolved with Microvolution software in ImageJ (NIH). The three-dimensional datasets were processed and volume-rendered with the surface evolver and filament-tracer tools in Imaris (Bitplane, Belfast, UK). The Imaris workstation was provided by the Rockefeller University Bio-imaging Resource Center.

Immunofluorescence microscopy

For the immunofluorescence labeling of wholemounted wild-type control and sema7a^-/- larvae, 1.5-4 dpf larvae were fixed overnight at 4 °C in 4 % formaldehyde in phosphate-buffered saline solution (PBS) with 1 % Tween-20 (1 % PBST). Larvae were washed four times with 1 % PBST for 15 min each, then placed for 2 hr in blocking solution containing PBS, 2 % normal donkey serum (NDS), 0.5 % Tween-20, and 1 % BSA. Primary antibodies were diluted in fresh blocking solution and incubated with the specimens overnight at 4 °C. The primary antibodies were goat anti-Sema7A (1:200; AF1835, R&D Systems) (14), rabbit anti-myosin VI (1:200; Proteus 25-6791), murine anti-GM130 (1:200; 610822, BD Transduction Laboratories), murine anti-CTBP (1:200; B-3, SC-55502, Santa Cruz Biotech), and rabbit anti-mCherry (1:200; GTX128508, GeneTex). Larvae were washed four times with 0.1% PBST for 15 min each. Alexa Fluor-conjugated secondary antibodies (Invitrogen, Molecular Probes) applied overnight at 1:200 dilutions in 0.2 % PBST included anti-rabbit (488 nm), anti-goat (555 nm), and anti-mouse (647 nm). Larvae were then washed four times in 0.2 % PBST for 15 min each and stored at 4 °C in VectaShield (Vector Laboratories). For the immunofluorescence labeling of postsynaptic density in wholemounted wild-type control and sema7a^-/- larvae, 2-4 dpf larvae were fixed with 4 % formaldehyde in PBS for 4.5-6 hr at 4 °C. Larvae were washed five times with 0.01 % PBST for 5 min each and then rinsed in distilled water for 5 min. The larvae were permeabilized with ice-cold acetone at -20 °C for 5 min. They were rinsed in distilled water for 5 min, washed five times with 0.01 % PBST for 5 min each, and blocked overnight with PBS buffer containing 2 % goat serum and 1 % BSA. Murine anti-pan-MAGUK antibody (1:500; IgG1 NeuroMab, K28.86, 75-029) was diluted in PBS containing 1 % BSA, added to the larvae, and incubated for 3 hr at room temperature. Larvae were washed five times for 5 min each with 0.01 % PBST. Alexa Fluor 555-conjugated anti-mouse secondary antibody (1:1000) diluted in PBS containing 1 % BSA was added to the larvae and incubated for 2 hr at room temperature. Larvae were washed five times with 0.01 % PBST for 5 min each, rinsed in distilled water for 5 min, and stored at 4 °C in VectaShield (Vector Laboratories).

Posterior segments of fixed larvae were mounted on a glass slide and imaged with a microlens-based, super-resolution confocal microscope (VT-iSIM, VisiTech international) under 60X and 100X, silicone-oil objective lenses of numerical apertures 1.30 and 1.35, respectively. Images at successive focal depths were captured at 200 nm intervals and deconvolved with the Microvolution software in ImageJ.

Measurement of Sema7A intensity

The average Sema7A fluorescence intensities of the hair cells from 1.5 dpf, 2 dpf, 3 dpf, and 4 dpf neuromasts were measured by determining the mean gray level within each hair cell labeled by actin-GFP expression. The intensity distribution of Sema7A protein along the apicobasal axis of the hair cell was measured using the line profile tool in ImageJ (Figure 1—figure supplement 1F). Because hair cell lengths differed among samples, lines were drawn with 2 μm widths and varying lengths of 6-8 μm. For each hair cell, apicobasal length was scaled from 0 to 100 arbitrary units and the intensity values from 0 to 1 arbitrary units.

Generation of skeletonized networks

The labeled sensory axons were traced in three dimensions with ImageJ’s semi-automated framework, SNT (48). Each trace depicts the skeletonized form of the posterior lateral-line nerve, the posterior lateral-line branch, and the point of arborization from which the sensory arbors radiate to contact the hair cells (Fig. 2D,E). The three-dimensional pixels or voxels obtained from the skeletonized sensory arbor traces were processed using custom code written in Python for visual representation and quantitative analysis.

Generation of combined hair cell clusters and the corresponding combined skeletonized networks

The hair cell clusters of neuromasts from each developmental stage were aligned by the centers of their apices, which were identified by the hair bundles, and overlayed to generate combined hair cell clusters using custom Python code. The associated skeletonized sensory-arbor traces at each developmental stage were aligned similarly using custom Python code. The details of the procedures are available upon request and the corresponding code is available on GitHub.

Quantification of sensory arbor distributions around hair cell clusters

The region from the center of the combined hair cell cluster to 60 μm in the post-cluster region was divided into 3600 concentric sections of equal area (Figure 2—figure supplement 2A). Each section, as depicted with a distinct color shade, had an area of π μm². Sensory arbor density was defined as Log₁₀[(Area occupied by the arbors in each section)/ (Area of each section)]. The arbor density was then plotted as a histogram against distance from the center of the combined hair cell cluster. Each bin of the histogram represents an area of π µm². The histograms were finally represented as density trace graphs with Kernel Density Estimation (KDE) in Python.

Quantification of contact between sensory arbors and hair cell clusters

For each hair cell cluster, the voxels of the sensory arbor traces that were inside or within 0.5 µm—the average neurite radius—of the cluster boundary were denoted as arbors in contact with their corresponding hair cell cluster (Figure 2—figure supplement 2B). The percentage of arbors in contact with each hair cell cluster was defined as: (Number of voxels that remained inside the cluster boundary/Total number of voxels in the arbor) X 100. The data were analyzed through an automated pipeline with custom code.

Quantification of the sensory arborization network topology

We developed a custom Python script (snt_topology.py; available on GitHub) to convert the skeletonized sensory arborization networks to network graphs (NetworkX, v3.1). We first used a custom ImageJ macro to convert SNT skeletons to csv files, then generated the network graph from the individual arbor paths. As noted previously, we defined five topological attributes: (1) nodes— junctions where arbors branch, converge, or cross; (2) loops—minimal arbor cycles forming topological holes; (3) bare terminals—arbors with free ends; (4) node degree—the number of first-neighbor nodes incident to a node; and (5) arbor curvature—the total curvature of a bare terminal (Figure 3A). To enumerate loops and quantify their lengths and node numbers, we computed the minimum cycle basis for each network graph. The relative organ height of each node in a graph was defined in 3D using Equation (1):

where u#⃗ is the vector from the arborization initiation point to a node and #v⃗ is the vector from the arborization initiation point to the hair cell cluster apex. The relative occurrence of nodes (Figure 3—figure supplement 1A-C) was normalized by both the number of hair cells per neuromast and the total number of neuromasts, for comparison across ages and genotypes. The relative number of bare terminals and loops (Figure 3B; Figure 3—figure supplement 1D, E) was normalized by the number of hair cells per neuromast and scaled between 0 and 1. The density was then normalized by the total number of neuromasts imaged for each developmental stage and genotype. For Figure 3C and Figure 3—figure supplement 1F and G, the normalized occurrence was computed as the number of arbors of a given length that formed either a bare terminal or a loop, divided by the total number of arbors of either feature of that length. Lastly, we defined bare terminals as having either high or low curvature (Figure 3F; Figure 3—figure supplement 2C, D) if their absolute total signed curvature [Equation (2)] in three dimensions was either higher or lower than π/6, respectively:

where 𝑠 is the arc-length parameter of the bare terminal, 𝑇(𝑠) is the unit tangent vector at each point along the arbor, 𝑁(𝑠) is the unit normal vector at each point, and the integral is evaluated from the terminal’s branching node to the arbor’s tip. For a closed curve forming a full loop, this value will equal 2π . In practice, we first fit B-splines (scipy.interpolate.splprep; SciPy, v1.10.1) in 3D to the skeletons for each bare terminal before evaluating their derivatives (scipy.interpolate.splev) and integrating (scipy.integrate.simpson) over their lengths.

Quantification of aberrant sensory arborization in ectopic Sema7A^sec expression

The sensory arbor networks were traced in three dimensions with SNT. The surface of each of the ectopically expressing Sema7A^sec-mKate2 cells was represented by a single point that was closest to the terminal point of the guided neurite. Distances between two points were calculated using standard mathematical operations.

Statistical analysis

Data visualization and statistical analysis were conducted with GraphPad Prism (Version 9) and Python (v3.10). The Mann-Whitney test was used for hypothesis testing unless otherwise noted, and the statistical details are given in the respective figure legends.

Data and code availability

Owing to size restrictions, the experimental image data have not been uploaded to a repository but are available from the corresponding author on request. The code generated during this study is available on GitHub (https://github.com/agnikdasgupta/Sema7A_regulates_neural_circuitry).

Acknowledgements

The authors thank Nicolas Velez and Emily Atlas for valuable discussions. Samantha Campbell and Anna Kaczynska provided expert fish husbandry. Katie Kindt kindly provided various strains of zebrafish. Image processing benefitted from facilities at the Bio-Imaging Resource Center. A.D. was supported by a Kavli Neural Systems Institute Postdoctoral Fellowship from The Rockefeller University. C.C.R. is supported by National Science Foundation Graduate Research Fellowship Grant No. 1946429. A.J., S.P.P., and L.M.S. were supported by Howard Hughes Medical Institute, of which A.J.H. is an Investigator.

Author Contributions

A.D. initiated the project, conducted experiments, analyzed data, and quantified results. C.C.R developed the Python codes for the network topology analysis and quantified those results, as well as analyzed published transcriptomics data. S.P.P. developed the Python codes for three-dimensional visual representation of sensory arbors and quantified arbor morphologies. L.M.S. conducted preliminary experiments. A.J. plotted Sema7A intensity measurements. A.D. wrote the paper with contributions from C.C.R and A.J.H.

Declaration of Interests

The authors declare no competing interests.

Figures and Figure Legends

Video 1. Sema7A is enriched at the hair cell base and sensory-axon interface. A through-focus scan of a 4 dpf control neuromast depicts the hair cells (green) that are innervated by the sensory axons (magenta) at their basolateral surfaces where the Sema7A protein (orange) is highly enriched. Scale bar, 5 µm.

Video 2. Three-dimensional arborization patterns in control and sema7a^-/- neuromasts. Lateral views depict three-dimensionally rendered combined skeletonized network traces from 27 control and 27 sema7a^-/-mutant neuromasts.

Video 3. Sensory axonal projections contact an embryonic muscle progenitor cell expressing Sema7A^sec. A through-focus scan of a 3 dpf larva depicts a muscle progenitor cell in the dermomyotome that expresses the ectopic Sema7A^sec (orange) and the sensory axonal projection (magenta) that makes intimate contact with it. Scale bar, 20 µm.

Video 4. Volumetric surface reconstruction of a sensory arbor and an embryonic muscle progenitor expressing Sema7A^sec. Reconstruction of the muscle progenitor cell (orange) at 3 dpf reveals its intimate association with an attracted sensory arbor (magenta).

Video 5. Sensory axonal projections remain in proximity to myofibers expressing Sema7A^sec. A through-focus scan of a 3 dpf larva depicts mature myofibers expressing exogenous Sema7A^sec deep in the myotome (orange) and the sensory axonal projection (magenta) that remains nearby.

Video 6. Volumetric surface reconstruction of a sensory arbor and the mature myofibers expressing Sema7A^sec. Reconstruction of mature myofibers expressing ectopic Sema7A^sec (orange) at 3 dpf shows their close association with the attracted sensory arbors (magenta).

Video 7. Aberrant axonal projections reenter the lateral-line nerve while following an ectopic Sema7A^sec source. A through-focus scan of a 3 dpf larva depicts cells expressing exogenous Sema7A^sec (orange) that guides the aberrant sensory axonal projection (magenta) back into the lateral-line nerve. Scale bar, 20 µm.

Video 8, 9, and 10. Ectopic Sema7A^sec expression induces aberrant neurite projections from the lateral-line nerve. Through-focus scans of 3 dpf larvae depict three individual incidents of ectopic sema7a^sec (orange) integrations that occurred distant from the neuromasts. In each case, myofibers express ectopic Sema7A^sec protein that attracts neurite extensions (arrowheads) from the lateral-line nerve (magenta). Scale bars, 20 µm.

Figure 1—figure supplement 1.

Figure 2—figure supplement 1.

Figure 2—figure supplement 2.

Figure 3—figure supplement 1.

Figure 3—figure supplement 2.

Figure 4—figure supplement 1.

Figure 5—figure supplement 1.