We observed that two or more afferent fibers de-fasciculate from the main nerve towards each neuromast in HGn39D (Figure 1 D, N, Q). The tight fasciculation of the axons does not allow their individual resolution at the level of the main nerve (Figure 1 D). Thus, it is possible that neurons project bifurcated fibres towards each neuromast. However, we speculated that since the afferent neurons outnumber neuromasts by a factor of two, more than one neuron could innervate each neuromast. To test this hypothesis, we took advantage of the fact that plasmid DNA injected into eggs is often expressed in a variegated manner. We generated larvæ expressing a membrane-localized red-fluorescent protein in individual afferent neurons, by injecting a cDNA coding for mem-TdTomato under the transcriptional control of the neuronal HuC promoter. Mem-TdTomato decorated the entire neuron, including the somata (Figure 2 A). In the background of the HGn39D transgenic line, this strategy allowed us to resolve individual neurons along their entire length (Figure 2 A–D), and also the thin neurites that form the dendritic arbors beneath the neuromasts (Figure 2 C–G). All the lateralis neurons that expressed mem-TdTomato in HGn39D were also positive for GFP and showed identical dynamics when compared to non-expressors. This indicates that HuC directs expression in afferents and that mem-TdTomato does not affect their behavior. We selected specimens expressing mem-TdTomato in single neurons for detailed analysis by confocal microscopy to find single fibers in 100% of the cases (N=42) (Figure 2 D), indicating that neurons do not project branched fibers towards any individual neuromast. Together, these results show that each neuromast is contacted by more than one afferent neuron.

Our previous observations predict that each neuron innervates only a subset of the constituent hair cells of each lateral-line organ. Because neuromasts harbor two populations of hair cells of opposite polarities, we asked whether there was a relationship between neuronal contacts and the orientation of hair cells. Hair cells in the zebrafish lateral line can be visualized in vivo by virtue of specific GFP expression in the SqET4 or brn3cGFP transgenic lines [33]–[35]. To determine the contacts of afferent neurons with hair cells of a particular orientation, we performed triple labeling of 7 dpf SqET4 larvæ with fluorescent phalloidin to reveal stereocilia in blue and scatter-labeled afferents with mem-TdTomato (Figure 3 A–L). We performed three-dimensional live imaging analyses in a large number of specimens (N=25), and found that afferent neurons formed two types of neurites. One type was thin along its entire length, whereas the second type had bulged termini. Thin neurites were long and very motile, extending and retracting very rapidly (Supplementary Movie S1). These neurites often did not contact any hair cell (Figure 2 C). When they established contacts with a hair cell, however, these were short-lived (Figure 3 E–J). In contrast, bulged neurites were shorter, less motile and established stable associations with hair cells (Figure 3 E–K). We found no direct relationship between thin neurites and hair cells of any orientation (Supplementary Movie S1). However, in 12 of 14 cases we observed that bulged neurites associated exclusively with hair cells of identical polarity (Figure 3 A–D and K–L) (Table 1) (Supplementary Movie S2). In the two exceptional cases (4 and 11 of Table 1), a neuron contacted hair cells of both polarities. However, the rare neurites that contacted one or two hair cells of the opposite polarity did not show a prominent bulging at their ends (Supplementary Movie S2). Interestingly, we documented an instance when a bulged neurite with a stable contact with one hair cell (Figure 3 E) projected an extension to contact a second target (arrowhead Figure 3 G). The second contact in this case is with an opposite-polarized hair cell (Figure 3 L). Over time the second contact destabilized for the bulged neurite to retain a stable contact with a single hair cell (Figure 3 K). This analysis highlights the essential need to perform long-time lapse sequence at high temporal resolution to define the dynamics of contacts between hair cells and neurons. Our observations suggest that only bulged neurites form synapses with hair cells. We further corroborated this conclusion by the establishment of an imaging method using DiASP to directly reveal synaptic communication between hair cells of a given polarity and the afferent neurons [11]. When applied in the medium, DiASP enters hair cells by passing directly through mechanically gated transduction channels on the stereocilia and subsequently permeates to the afferent neurons (Figure 1 R–S). DiASP entry into hair cells is blocked when the mechanotransducing channels are closed. To mechanically open transduction channels in hair cells of a selected orientation and close those of the opposite, we applied a fluid stream directed at the neuromasts along the anteroposterior axis of the fish (Supplementary Figure S1). After a brief period of directed water jet to deflected stereocilia, we applied DiASP to the stream. In contrast to the normal loading to all transducing hair cells in the neuromast, fluid jet application of DiASP labeled hair cells whose stereocilia were deflected positively, whereas the cells whose stereocilia were deflected negatively were not labeled (Figure 4 A–D). We reasoned that if each neuron synapses with hair cells of both orientations, DiASP should permeate to all neurons irrespectively of the orientation of the fluid jet. However, when we applied a polarized fluid jet to HGn39D specimens, we observed that DiASP did not label all fibers, suggesting that the unlabelled neurons did not establish synaptic contacts with hair cells of both polarities (Figure 4 E–H). This result is consistent with our previous observations, which together support the conclusion that each afferent neuron forms stable synaptic connections with hair cells of only one orientation.

Mosaic expression of mem-TdTomato showed that some neurons innervate simultaneously two or three adjacent neuromasts (Figure 5). These neurons tended to be the longest: i.e., those that targeted the most caudal organs. We suspected that these neurons would contact hair cells of identical orientation in each neuromast. To test this hypothesis, we examined SqET4 and brn3cGFP stable transgenics expressing mem-TdTomato in single neurons. Because hair cells are born in pairs of opposite planar polarity along a single axis creating a line of mirror symmetry, all the cells located anterior to the line of symmetry are polarized posteriorly, whereas the posterior cells adopt an anterior orientation (Figure 4 D) [34]. Thus, the orientation of the hair cells can be predicted from their position with respect to the line of mirror symmetry. When the line of mirror symmetry was evident, we observed that bulged neurites established stable associations exclusively with hair cells on the same side of each neuromast. This indicates that each afferent neuron synapses with hair cells of identical orientation in adjacent neuromasts (Figure 5). Therefore, lateralis afferents can link several organs to form multi-neuromast sensory units.

Hair cells of the superficial lateral line have their mechanosensitive stereocilia embedded into a gelatinous cupula. Thus, the cupula coordinates the deflection of the stereocilia in a manner analogous to that of the tectorial membrane in the mammalian cochlea. Consequently, all hair cells in the neuromast are stimulated simultaneously by water current. The organization of hair cells in opposite polarities suggests that the lateral line can both discriminate a mechanical signals' polarity and its vectorial component. However, a movement of the cupula will depolarize half of the hair cells and hyperpolarize the remaining half simultaneously, creating ambiguity in the resolution of stimulus polarity. How does the system resolve the vector of a mechanical stimulus? One possible mechanism would involve the division of the neuromast into compartments, each including hair cells of identical orientation in analogy to the compartmentalization created by the striola in some inner ears. An apical line of mirror symmetry separating anatomical planar-polarity compartments in neuromasts is evident only transiently during the first few hours of hair-cell development or regeneration [34]. However, hair cells of opposite polarities are intermingled in a mature and fully functional neuromast. Therefore, the neuromast's epithelium appears to lack any anatomical compartmentalization at the apex. The spatial organization of hair-cell apices mirrors that of their bases, suggesting that an anatomical compartmentalization at the basal part of the neuromast is unlikely [34]. An alternative possibility would involve a differential connection between hair cells of each polarity and processing areas in the brain. Such functional compartmentalization would necessitate the innervation of neuromasts by at least two sensory neurons. By scatter-labeling afferent neurons with mem-TdTomato in the HGn39D transgenic line, we were able to dissect the lateral-line neuronal network at the single-cell level. We discovered that each neuromast is contacted by two or more afferent neurons. By performing extensive live imaging recordings of hair cells and neurons in three dimensions, we also showed that neurons arborize extensively and establish synaptic connections with the hair cells. We found no case in which a neuron contacted all the hair cells in a neuromast. By contrast, the majority of the neurons contacted 50% or fewer hair cells, indicating that neuromasts are partitioned into synaptic sub-groups. Therefore, the lateral-line system meets the first essential requirement for functional compartmentalization.

When labeling individual neurons with mem-TdTomato, we observed neurons that synapsed with hair cells in adjacent neuromasts. This could group several neuromasts into a single sensory unit. However, for these multi-neuromast units to work efficiently a condition sine-qua-non is that each neuron forms synapses with hair cells of identical orientation in all neuromast. Indeed, single-neuron imaging in SqET4 transgenics showed that each afferent that innervated at least two neuromasts contacted hair cells of identical orientation. These results suggest that while the cupula coordinates the mechanical stimulation of all the hair cell in each neuromast, the neurons decouple hair cells based on orientation in each neuromast and, simultaneously, couple hair cells of identical orientation in adjacent neuromasts. Our results demonstrate striking similarities with the innervation of identical populations of hair cells in all the neuromasts forming a stitch in Xenopus, implying that the mechanisms governing the formation of multi-neuromast sensory units is conserved across species [21]–[22]. We wondered about the selective pressure for the evolution of multi-neuromast units. It has been proposed that the resting activity (summation of spontaneous and evoked activities) of an afferent neuron increases with the number of the neuromasts it contacts [42]. But because the spontaneous activity is independent of the number of neuromasts, only the evoked activity would increase by the recruitment of more neuromasts by the afferent neurons [43]. Therefore, multi-neuromast sensory units should permit both the amplification of the mechanical signal and a reduction of noise in the system without the need to increase the number of hair cells per neuromast, which would otherwise reduce sensitivity due to the larger resistance to cupula bending by more numerous hair bundles [44]. This thesis takes support in the observation that while most organs increase their size as the fish grows, neuromasts increase in number but remain small.

For vital labeling of hair cells, zebrafish larvæ were immersed in a 500 µM solution of DiASP for three minutes at room temperature in the dark. Treated larvæ were washed briefly to remove excess fluorophore, anæsthetized in 3-aminobenzoic acid ethyl ester solution (Sigma), mounted on a glass slide, and aligned using a hair loop. For immunohistochemistry, larvæ were fixed overnight at 4°C in a solution of 4% paraformaldehyde in phosphate-buffered saline (PBS) solution containing 1% Tween-20. After fixation, samples were washed in the same solution without fixative and blocked at room temperature with 10% bovine serum albumin. Primary- and secondary-antibody incubations were conducted overnight at 4°C in PBS with 0.2% Tween-20. Labeling with fluorescent phalloidin was done as published using Alexa-Fluor 350 or Alexa-Fluor-647 phalloidin 125 (Invitrogen) [12]. Primary antisera and monoclonal antibodies were used at the following dilutions: rabbit anti-Tyrosine hydroxylase, 1/1000 (Pel-Freeze); mouse monoclonal antibody HCS-1, 1/20; and rabbit anti-Cav1.3a (Gem25.2), 1/500. Texas Red-labeled donkey anti-mouse, and -rabbit immunoglobin secondary antibodies (Jackson Labs) were used at a dilution of 1/150. Fluid jet application of DiASP was done on anæsthetized zebrafish larvæ placed into a 1 mm width well in an agarose disk. The tail of the fish was attached to the agarose by a 5 mm long hair. A 10–15 seconds stream of E3 medium was applied to the fish from its tail along its anteroposterior axis using a wash bottle (Nalgene) with a 1 mm diameter aperture. During the continuous E3 medium stream, DiASP was applied during 1–3 seconds near the tail of the fish to the directional fluid stream. DiASP was applied at a 50 mM concentration using a pipette tip (Eppendorf).

We created genetic mosaics to scatter-label afferent neurons by injecting a plasmid coding for a membrane-targeted TdTomato construct under the transcriptional control of the neuronal HuC promoter (HuCmem-TdTomato) into fertilized eggs of stable transgenic HGn39D, SqET4 or brn3cGFP animals. We screened embryos that expressed the GFP with a stereomicroscope under ultraviolet light, and selected those expressing mem-TdTomato in single neurons. Embryos were let to grow in system water at 28.5°C in 85 mm Petri dish for four or five days in the dark. Neuromasts contacted by a single red neuron were selected for long term live imaging by laser-scanning confocal microscopy. For this purpose, zebrafish larvæ were mounted on the coverslip of a glass-bottom dish (Mattek), covered with 1% low-melting-point agarose in a 610 µM solution of the anæsthetic 3-aminobenzoic acid ethyl ester, and imaged immediately. Images were acquired with a Leica TCS SPE confocal microscope with a 40× oil immersion objective. Z-stacks were acquired at 0.8 mm intervals, imaging GFP (488 nm excitation, 500–550 nm emission) and TdTomato (532 nm excitation, 570–630 nm emission). We generated three-dimensional images over time (4-D images) for periods ranging from 3 to 24 hours and different temporal resolutions.