a Schematic of the gravity-sensing system in fish. Hair cells in the utricular macula (HC, gray) are inertial sensors of head tilt and translation, exciting the peripheral process of utricular afferents (green). These afferents, whose cell bodies are located in the utricular ganglion, project to brainstem neurons involved in escape (Mauthner cell, black), posture (vestibulospinal [VS] cell, blue), and oculomotor (VOR) reflexes (superior vestibular nucleus [SVN] and tangential nucleus [Tan], brown). Dashed line indicates midline. b Coronal section through the head of a 5.5 dpf zebrafish. The region reimaged at high resolution is visible as an L-shaped territory (dashed outline) covering the right utricle and hair cells, utricular ganglion, and ipsilateral brainstem. The reimaged territory extended across 1757 coronal sections (105 µm in the rostrocaudal axis). Scale bar, 100 µm. c Electron micrograph of two hair cells in the utricular macula, with portions of their cilia. Scale bar, 3 µm. Source data for this and all example EM images are provided as links in the Source Data file. d Section of the vestibular nerve, peripheral processes. At this developmental stage, some axons are myelinated (pseudocolored dark green) while others are not yet (light green). Scale bar, 1 µm. e Horizontal projection of reconstructed brainstem targets (Mauthner, VS, SVN, Tangential) colorized as in a. The file to generate this and all other reconstructions in this paper are provided in the Source Data file. f Sagittal projection of utricular hair cells, afferents, and brainstem targets, as in a. g Coronal projection as in f.

a Top, Electron micrograph of utricular hair cell with stereocilia (black) and kinocilium (red) marked. Right, schematic of tuning vector derived from cilia positions, viewed from above. Bottom, EM image of hair cell synaptic ribbons (arrowheads) apposed to a utricular afferent. Scale bars, 1 µm. b Horizontal projection of the utricular macula, showing tuning direction vectors for all 91 hair cells. Dashed line: line of polarity reversal (LPR). Vectors are colorized by directional tuning. Note slight asymmetry in colorization; this was chosen to ensure hair cells from the medial and lateral sides of the LPR are represented in different colors. Here and throughout: R, C = rostral, caudal = nose-down and nose-up pitch, respectively. M, L = medial, lateral = contralateral and ipsilateral roll, respectively. Source data for this and all graphs is provided in the Source Data file. c Hair cells located near each other have more similar directional tuning, which falls off sharply within ~30 µm. Data from hair cells medial to the LPR only (n = 77 hair cells) and represent mean values ± SD. d Horizontal view of reconstructions of all 105 utricular ganglion afferents including somata (larger spheres, left) and their postsynaptic contacts in the utricular macula (smaller spheres, right). Afferents are colorized by inferred direction tuning as in b. View is slightly tilted to aid in visualization of ganglion. Afferents with inferred contralateral head tilt tuning (red) form a segregated axon bundle in the brainstem (left). e Sagittal view of reconstructions shown in d. f Correlation of rostrocaudal soma position between synaptically connected utricular hair cells and afferents. Circle size reflects the number of synaptic ribbon connections (range: 1–19). Significance of linear correlation, t-test, p = 1.3 × 10⁻¹⁰². g Horizontal projection of inferred afferent tuning vectors, relative to soma position in the afferent ganglion. Each vector indicates an afferent’s tuning direction, calculated by weighting by the number of ribbon inputs it receives from each hair cell. Colors as in b. h Afferents located close to each other have similar directional tuning, but the relationship is looser than in hair cells (c). Data are from afferents innervating the macula medial to the LPR only (n = 94 afferents) and represent mean values ± SD.

a Example micrograph of utricular afferents (pseudocolored green) synapsing onto the lateral Mauthner cell dendrite (yellow). Chemical synapses are recognizable by clustered presynaptic vesicles and synaptic density. The tight apposition between the upper afferent and the Mauthner is likely an electrical synapse. Scale bar, 0.5 µm. b Coronal projection of Mauthner cell skeleton reconstruction (black, spherical soma) with utricular afferent input synapses (gray). Inset, expanded view of utricular inputs. c Horizontal projection of reconstructions of both Mauthner cells (gray, blue) and four commissural utricular neurons (red). Commissural utricular neurons make synaptic contacts on the contralateral Mauthner (small red circles). Colors indicate inferred directional tuning. d Polar plot of inferred direction tuning of utricular input to Mauthner cell and four commissural escape neurons. Directional tuning is indicated in the context of head tilt. Note that as inertial sensors, otoliths are equally sensitive to head translation in the opposite direction (e.g., ipsilateral head tilt and contralateral translation are indistinguishable). e Schematic of predicted Mauthner cell computation of head translation. A predator approaching from the right will cause a head deflection to the left. Deflection of the utricular otolith by inertia (blue arrow) would depolarize the ipsilateral tilt / contralateral translation pathway (blue; medial to the LPR). These utricular afferents excite the ipsilateral Mauthner cell, promoting an escape movement to the left. Commissural escape neurons, in contrast, will respond to rightward head movements (red hair cells and dashed lines) and are predicted to activate the contralateral Mauthner cell, promoting escapes to the right. f Example of behavioral response in a free-swimming larva subjected to rapid translation. High-speed videography captures the onset of escape and characteristic C-bend. Bottom, quantification of tail angle and translational stimulus. Scale bars, 1 rad, 100 ms, and 1 g. g Escape responses to a unidirectional translational stimulus are plotted relative to the larval heading angle at the start of the stimulus. As predicted, larvae accelerated to the right (heading direction 0–180°) escape to the right, whereas larvae accelerated to the left escape left. Escapes only occurred in the “incorrect” direction when animals were accelerated in predominantly rostral or caudal directions. h With both types of stimulus, utricle-deficient rock solo larvae escaped at approximately half the rate as their heterozygous siblings. N = 52 sibling and 55 −/− fish. Chi-squared test, p = 0.0006 (1 degree of freedom, chi-squared value = 11.65, 266 total observations).

a Horizontal projection of 12 SVN reconstructed neurons. Neurons are colorized based on their inferred directional tuning. Polar plot inset shows the directional tuning vectors of each neuron. Gray bar indicates location of coronal plane where axon trajectories are shown in boxed inset. Scale for reconstructions as in c. Electron micrograph (bottom) shows a utricular afferent (pseudocolored green) contacting two dendrites of SVN neurons (yellow). Scale bar, 500 nm. b As in a, for 11 tangential nucleus reconstructed neurons. Sagittal projection inset (bottom) shows the divergence of tangential neuron axons in the dorsoventral axis. c As in a but for 19 VS neurons. d Dendrograms of three example neurons from the SVN, tangential, and VS populations. The soma is represented by a black circle and each dendrite is represented by lines. Gray circles indicate synaptic inputs from utricular afferents, which appear disproportionately concentrated on a small number of dendrites. Axons are truncated for purposes of scale. These and all dendritic arbors are provided in the Source Data file. e Quantification of synaptic clustering, measured as distances between synapses, using three Monte Carlo models compared to actual data. Synaptic locations were modeled as randomly distributed across the arbor (unweighted model), preferentially weighted within 50 µm of afferent axons, or preferentially weighted within 5 µm of afferent axons. The observed level of synaptic clustering is shown in red. See Methods for detailed description.

a Representation of kinocilium reconstruction across successive images to obtain the total length. The neighboring tallest stereocilium was also reconstructed (not schematized). b Plot of length of the kinocilium vs. the tallest stereocilium for all 91 hair cells. Striolar hair cells are identified by their long stereocilia and low K/S ratios, whereas extrastriolar hair cells have higher K/S ratios. Hair cells characterized as immature have short kinocilia and stereocilia. c Horizontal projection of the utricular macula, showing number of synaptic ribbons in each hair cell. Circle diameter reflects synaptic ribbon count; hair cells with larger numbers of ribbons tend to be located more centrally. See also Supplementary Fig. 1. d Quantification of synaptic ribbon counts across hair cell categories. Box plot represents medians ± 25%ile; whiskers indicate 10-90%iles. Immature hair cells form fewer ribbon synapses than striolar or extrastriolar hair cells. N = 23 striolar, 52 extrastriolar, and 16 immature hair cells. Significance was tested with ANOVA (p = 4.8 × 10⁻⁵) and then pairwise with Wilcoxon–Mann–Whitney (striola—extrastriolar, p = 0.60; striolar—immature, p = 4.3 × 10⁻⁴; extrastriolar—immature, p = 1.3 × 10⁻⁴). e Horizontal projection of all utricular afferents, colorized by whether they are myelinated or not. f Horizontal projection map of all utricular ganglion somata by position; circle diameter reflects the number of hair cell ribbon synaptic inputs that each one receives. g Myelinated afferent somata are located more laterally in the utricular ganglion. Each dot represents one afferent soma. Wilcoxon–Mann–Whitney, p = 2.9 × 10⁻⁴. h Myelinated afferents are contacted by significantly more hair cell ribbons than unmyelinated afferents. Wilcoxon–Mann–Whitney, p = 1.8 × 10⁻¹⁰. See also Supplementary Fig. 3c. i Myelinated afferents receive the majority of their input from striolar hair cells, whereas unmyelinated afferents receive most of their input from extrastriolar and developing hair cells. The distributions are significantly different (chi-squared test, p < 1 × 10⁻¹⁰). See also Supplementary Fig 3d, e. j The weighted fraction of inputs each utricular afferent receives from the different hair cell classes. Afferents are ordered based on soma position from medial to lateral. Red dots identify afferents receiving input from hair cells lateral to the LPR (contralateral tilt sensitive). Laterally positioned afferents typically have predominantly striolar inputs, whereas medially positioned afferents have predominantly extrastriolar or immature inputs.

a Hair cells in the utricular macula that excite afferents connected to Mauthner escape, VOR, or VS circuits (top, middle, and bottom). Hair cells are colorized if they contribute input to a pathway or gray if they do not. Right, afferents connected to escape, VOR, or VS circuits, colorized by their myelination status (black: myelinated; gray, unmyelinated). b Quantification of the contribution of striolar, extrastriolar, and immature hair cells to these central pathways. Striolar hair cells preferentially drive rapid escape circuits, both via the direct afferent input to the ipsilateral Mauthner and the afferent input to the commissural escape pathway, whereas VOR and VS circuits receive input from a mixture of pathways, with extrastriolar inputs dominating. Numbers in parentheses indicate the number of central neurons in each category. Wilcoxon test, striolar contribution to escape vs non-escape neurons, p = 0.041. c Quantification of the contribution of synapses from myelinated and unmyelinated afferents to each central pathway. Afferents driving escape pathways are largely myelinated at this age, whereas afferents driving VOR and postural pathways are more mixed. Chi-square test for all groups, p = 6.0 × 10⁻⁸. Follow-up chi-square: VOR vs escape, 1.4 × 10⁻⁴; VS vs escape, 1.7 × 10⁻⁸. d Subsets of VS neurons, identified by axon trajectories, are predicted to exhibit different temporal kinetics. VS neurons with axons that approach the midline before descending (VS_med, greens) receive mostly extrastriolar (tonic) input with a large contribution from immature hair cells, whereas VS neurons with ventral axon trajectories (VS_vent, dark reds) receive mostly striolar (phasic) input and none from immature pathways. The skeleton reconstruction is projected at a mixed horizontal/sagittal angle to facilitate visualization of these groups. See also representation of axon trajectories in Fig. 4c and Supplementary Fig. 4.

a Summary schematic of organization by directional tuning. Hair cells in the utricular macula, left, project via afferents that maintain rostrocaudal organization but not mediolateral organization. The utricular afferent ganglion is organized rostrocaudally, but contralateral tilt sensitive afferents are intermingled. These afferents project with different patterns to distinct brainstem targets, conferring directional sensitivity in the mediolateral (escape) or rostrocaudal (VOR, posture) pathways. Colors indicate directional tuning as previously. b Summary schematic of organization by temporal kinetics. Early-born, striolar hair cells make synaptic connections to early-born afferents, whose cell bodies are positioned laterally in the utricular ganglion, and typically myelinated by the larval stage examined here. These early-born afferents, carrying phasic information about head movement, preferentially excite escape pathways, which consist of early-born, fast reticulospinal and spinal motor neurons and muscles. Postural and VOR reflex pathways rely more on the tonic and phasic-tonic signals arising from extrastriolar, slightly later-born pathways. We speculate that circuits carrying immature input, like VS_med, may project to motor circuits governing slower and more refined control of movement.