Extraocular motor neurons for vertical/torsional gaze stabilization are located in cranial nuclei III (nIII) and IV (nIV). To eliminate nIII/nIV motor neurons and by extension, any secreted signals, we used a genetic loss-of-function approach (Figure 1A). A single highly-conserved transcription factor, phoxa, specifies nIII/nIV fate ^34–37. In the vestibulo-ocular reflex circuit, phox2a is exclusively expressed in nIII/nIV motor neurons but not its upstream partners (Figure 1B). Therefore, phox2a is an ideal genetic target to eliminate motor-derived signals without compromising evaluations of upstream functional development. Prior mutagenesis established a phox2a loss-of-function allele in zebrafish ³⁴, but the line has since been lost. Here, we generated three new phox2a loss-of-function alleles using CRISPR/Cas9 mutagenesis (Figure 1C) (one allele shown here; additional alleles described in Methods). Consistent with prior reports and human mutations ³⁸, both eyes in phox2a null mutants were exotropic (rotated towards the ears) reflecting a loss of motor neurons in nIII/nIV. phox2a mutants failed to hatch from their chorions without manual intervention and did not inflate their swim bladders by 5 days post-fertilization (dpf) (Figure 1D), phenotypes not previously reported ³⁴. Consequently, null mutants do not survive past 7 dpf. We did not observe these morphological phenotypes in wildtype and heterozygous siblings (Figure 1D). As vestibulo-ocular reflex circuit architecture and behavior is established by 5 dpf ^24,25,29, premature lethality did not preclude further measurements of circuit development.

To validate phox2a loss-of-function, we leveraged a downstream transcription factor: isl1 ³⁹. The Tg(isl1:GFP) line ⁴⁰ labels all nIII/nIV motor neurons except inferior oblique neurons ³³, which comprise one of four pools for upwards eye rotations. We first quantified changes in the number of labeled nIII/nIV neurons (Figure 1E-Figure 1F). In phox2a mutants, we observed an expected and near-total loss of isl1 expression (WT: 298±19 neurons across both hemispheres; null: 19±11 neurons; Wilcoxon rank sum test, p=2.5×10⁻⁴) at 5 dpf, well-after nIII/nIV differentiation is complete ³³. Unexpectedly, we also observed slightly fewer neurons in phox2a heterozygotes (heterozygote: 229±20 neurons; Wilcoxon rank sum test against WT, p=6.7×10⁻⁴). In heterozygotes, loss of isl1 fluorescence was restricted to the medial domain of dorsal nIII, which contains some of the earliest-born neurons in nIII/nIV (Figure S1A-Figure S1C) ³³. Globally, this manifested as a rostral and ventral shift in the positions of all neurons mapped (Figure S1D) (two-sample, two-tailed KS test, WT vs. heterozygotes: mediolateral axis, p=0.13; rostrocaudal: p=4.0×10⁻²⁹; dorsoventral: p=2.5×10⁻⁹). This region contains two motor pools that control the inferior (IR) and medial rectus (MR) muscles ³³. We conclude that phox2a acts in a dose- and birthdate-dependent manner to specify nIII motor pool fate.

Together, these observations validate our phox2a loss of function alleles as a selective means to disrupt nIII/nIV motor neuron fate specification and vertical eye rotation behavior.

Vertical gaze stabilization requires that (1) peripheral VIII^th nerve sensory afferents relay tilt sensation (nose-up/nose-down) directly to projection neurons in the tangential nucleus, and (2) projection neurons innervate appropriate nIII/nIV pools (eyes-up/eyesdown). For proper circuit function, appropriate connectivity must first develop across up/down circuit subtypes. The “retrograde” model predicts that motor partners specify circuit assembly. Therefore, in the absence of motor neurons, projection neurons should fail to respond selectively to directional tilt sensations – either due to loss of their fate, the fate of upstream sensory afferents, or sensory-to-central connectivity.

To evaluate upstream circuit formation, we measured tilt-evoked responses in projection neurons using Tilt-In-Place Microscopy (TIPM) ⁴¹ (Figure 2A-Figure 2B). Peripherally, tilts activate utricular VIII^th nerve sensory inputs to projection neurons ^29,41. We used a galvanometer to deliver tonic nose-up and nose-down pitch tilts to phox2a null larvae and sibling controls. We then measured the activity of a calcium indicator, GCaMP6s ⁴², in projection neurons. We performed experiments at 5 dpf, when nearly all projection neurons are selective for one tilt direction ²⁹, circuit architecture is stable ²⁵, and gaze stabilization behavior is directionally-appropriate ²⁴.

Projection neuron responses and topography were strikingly unchanged in phox2a mutants compared to controls. We recorded the activity of n=297 neurons from N=16 phox2a mutants and n=440 neurons from N=21 sibling controls (Methods and Table 1 split by genotype). We observed comparable ratios of projection neuron subtypes (sib: 46% nose-down, 47% nose-up, 7% untuned; phox2a: 49% nose-down, 44% nose-up, 7% untuned) (Figure 2C). Next, we evaluated their topography (Figure 2D-Figure 2E). Projection neurons are topographically organized along the dorso-ventral axis by their directional selectivity ²⁹. Global spatial separation between subtypes remained significant in phox2a mutants (one-way multivariate analysis of variance, p=0.004). We also compared the topography of nose-up and nose-down neurons separately across phox2a genotypes. Nose-down neurons were comparably distributed between null and control larvae (one-way multivariate analysis of variance, p=0.15). We observed a minor lateral shift to nose-up neurons in null mutants (median mediolateral position, sib: 15.2µm from medial edge; phox2a, 13.2µm; twotailed, two-sample KS test, p=3.0×10⁻⁴) but no changes in other spatial axes (dorsoventral: p=0.16; rostrocaudal: p=0.56). The small medial deviation (2µm across a 40µm space) is within the limits of our registration error. We conclude that projection neuron to-pography is established independently of motor partners.

Projection neuron sensitivity and selectivity also developed comparably between phox2a mutants and siblings (Figure 2F-Figure 2K). Projection neurons responded to tilt sensations with comparable magnitudes (Figure 2G-Figure 2J) (nose-down mean ΔFF, sib: 1.86±1.69; phox2a: 2.07±1.48; two-tailed Wilcoxon rank sum test, p=0.98; nose-up mean ΔFF, sib: 1.24±1.23; phox2a: 1.02±0.89; p=0.18). Previously, we defined a metric to describe a neuron’s selectivity for one tilt direction (0 = equal responses to up/down; 1 = maximally selective) ²⁹. Directional selectivity remained unchanged in phox2a mutants (Figure 2H-Figure 2K) (nose-down mean index, sib: 0.73±0.29; phox2a: 0.68±0.29; two-tailed Wilcoxon rank sum test, p=0.85; nose-up mean index, sib: 0.85±0.26; phox2a: 0.81±0.29; p=0.12). Collectively, this demonstrates that the functional responses of projection neurons and, by inference, connectivity with utricular afferents are not shaped by motor partners.

Ventral projection neurons receive additional input from the semicircular canals ²⁹, which encode phasic (fast) tilt sensation. To activate sensory afferents from the semicircular canals, we used TIPM to deliver two impulses of angular rotation (Figure 3A-Figure 3B) ^29,41. We observed no changes in phox2a mutants. Projection neurons responded to impulses in comparable ratios (Figure 3E) (sib: 58% responsive; phox2a: 71% responsive). Responsive projection neurons remained localized to the ventral nucleus (dorsoventral axis: two-tailed, two-sample KS test, p=0.99). Lastly, the functional properties of projection neurons were unchanged (Figure 3F-Figure 3G). We observed no change in calcium response magnitudes (Figure 3F) (mean ΔFF, sib: 0.33±0.29; phox2a: 0.36±0.40; two-tailed Wilcoxon rank sum test, p=0.85) or lack of directional selectivity (Figure 3G) (mean index, sib: 0.07±0.41; phox2a: 0.06±0.40; p=0.39). Therefore, fate and connectivity between phasic sensory afferents and projection neurons must not require motor partners.

Lastly, we considered whether loss of one subtype of nIII/nIV neurons might alter connectivity. For example, loss of eyes-down motor pools could impair wiring between their corresponding nose-up sensory and projection neuron partners. Here, we leveraged phox2a heterozygotes, which lack a subtype of nIII neurons (IR/MR) that contribute to downwards eye rotations (Figure S1). We observed no differences in tonic tilt responses between phox2a wildtype, heterozygote, and null larvae, though we did note a minor decrease in response strength to impulses (statistics in Table 1). We note that phox2a heterozygotes do not lack all motor pools for downwards eye rotations. Nevertheless, we conclude that individual motor pools do not meaningfully contribute to connectivity between sensory and projection neurons.

Taken together, these experiments demonstrate intact directional selectivity for two peripheral sensory inputs – utricular and semicircular canal VIII^th nerve afferents – and appropriate connectivity with projection neurons. We conclude that functional sensory-to-central circuit formation is established independently of motor partners.

Motor partners could secrete signals that initiate pre-motor axon outgrowth, target arriving axons to specific motor pools, or trigger synaptogenesis ³. Motor pool topography in nIII/nIV reflects ocular muscle targets: dorsal pools innervate downward-rotating muscles (superior oblique and inferior rectus), while ventral pools target the converse (eyes-up, superior rectus and inferior oblique) ^33,43. In turn, projection neuron somatic and axonal organization mirrors motor pool topography ^28,29, which could facilitate directionally-selective circuit assembly. We reasoned that projection neurons may fail to initiate axon outgrowth, target spatially-appropriate motor pools, and/or form synapses in phox2a mutants. To test this hypothesis, we measured changes in projection neuron anatomy at 5 dpf, when axonal arbors are established and stable ²⁵.

To test whether projection neurons establish gross, long-range (hindbrain to midbrain) axonal outgrowths, we performed optical retrograde labeling ⁴⁴ using a photolabile protein, Kaede. We targeted the medial longitudinal fasciculus at the midbrain-hindbrain boundary, which contains projection neuron axons ^25,29 (Figure 4A). In both phox2a mutants and sibling controls, we observed retrograde photolabeling of projection neuron soma (Figure 4B), supporting that initial axon outgrowth does not require motor partner-derived signals.

Next, we evaluated whether projection neuron axons remain capable of wiring with spatially-appropriate motor partners. Projection neuron axons segregate along the dorsal (nose-up) and ventral (nose-down) axes according to their birth order (early/late born, respectively) ^28,29 and the pool topography of their motor targets ³³. To test whether projection neurons retain this topography, we optically labeled the axons of early-born (before 30 hpf) projection neurons ²⁹. In phox2a mutants, axons remained dorsoventrally segregated at midbrain targets (Figure 4C, inset). Typically, projection neurons robustly collateralize to nIII/nIV targets at the midbrain-hindbrain boundary. We did not observe collaterals to nIII/nIV in phox2a mutants (Figure 4C). However, projection neurons still robustly arborized to more rostral, spinal-projecting targets in the nucleus of the medial longitudinal fasciculus, suggesting they retain the machinery necessary to collateralize. Consistent with this hypothesis, we observed that projection neurons formed occasional, small collaterals in phox2a mutants with few (1-5%) nIII/nIV neurons remaining (Figure 4D). We conclude that projection neurons remain competent to assemble with spatially-appropriate targets.

If motor neurons are required to initiate synaptogenesis, then projection neurons should fail to develop pre-synaptic machinery. To test this hypothesis, we performed fluorescent in situ hybridization against common pre-synaptic transcripts: synaptophysin a (sypa), synaptic vesicle glycoprotein (sv2), and synapsin I (syn1). In both phox2a mutants and controls, we observed robust transcript expression in projection neuron somata at 5 dpf (Figure 4E), well-after synaptogenesis onset in wildtype larvae ²⁹. Motor partner-derived signals are thus not required for projection neurons to develop the necessary components for synaptogenesis.

Though motor neurons may play later roles in selecting and/or refining pre-motor input specificity, our data supports that projection neurons remain anatomically and molecularly poised to assemble with appropriate targets. We predict that absent collaterals and synapses reflect a lack of adhesive contact necessary to stabilize ^45,46, but not instruct the formation of nascent structures.

Functional and anatomical connectivity, from peripheral sensors to motor targets, develop independently of motor partners. Fate in the vestibulo-ocular reflex circuit follows from connectivity ²¹, but neuronal fate can also be defined with respect to unique transcriptional signatures. Previously, we developed a sequencing pipeline to discover transcription factors that specify functional subtypes of spinal motor neurons and evaluate the consequences of perturbations on transcriptional fate ⁴⁷. We adapted this approach to determine if loss of motor-derived signals changed the transcriptional profiles of projection neurons.

We compared the transcriptional profiles of projection neurons in phox2a mutants and sibling controls (Figure 5A) using bulk RNA sequencing. We performed sequencing experiments at 72 hours post-fertilization (hpf), after projection neuron differentiation is complete and synaptogenesis to motor targets has peaked ²⁹. We sequenced projection neurons labeled by Tg(-6.7Tru.Hcrtr2:GAL4-VP16);Tg(UAS-E1b:Kaede) ^24,25,48,49 (Methods, Figure 5B, Figure S2). Neurons labeled in this line include, but are not exclusive to the projection neurons in the tangential nucleus used for vertical gaze stabilization. Therefore, we evaluated our bulk RNA sequencing dataset in the context of a single-cell reference atlas derived from the same transgenic line (Methods, Figure S3) to minimize noise from other labeled populations. We used evx2 ⁵⁰ as a reference, as it was expressed in all projection neurons (Figure S3D) and highly detected (50%) in singly-profiled projection neurons.

There were strikingly few differentially-expressed genes in projection neurons between phox2a siblings and null mutants (Figure 5C, Table 2). All candidate differentially-expressed genes were lowly-expressed (detected in <10% of reference projection neurons, Figure 5D). To determine if any candidates were differentially expressed in projection neurons, we used a fluorescent in situ hybridization method ⁵¹ in which fluorescence intensity correlates with detected transcript expression reliably across individual larvae (Figure S4). We evaluated 8 candidate genes (Figure 5E-Figure 5F). Qualitatively, we observed no differences in expression patterns between phox2a mutants and siblings, neither in candidates with significant differential expression in the bulk RNA sequencing dataset or in a highly-expressed control markers, evx2.

We considered that our inability to detect differentially-expressed genes could arise from our exclusion of candidates based on their expression in our reference single-cell atlas. Therefore, we repeated our analyses in unfiltered bulk sequencing data. The top 50 highest-expressed genes in phox2a siblings were highly detected in singly-profiled neurons labeled by Tg(-6.7Tru.Hcrtr2:GAL4-VP16);Tg(UAS-E1b:Kaede), including projection neurons (Table 3). This suggests that our dissections adequately captured our target population. However, we again identified few differentially-expressed genes in our unfiltered data (Figure S5A-Figure S5C), with substantial decreases as significance stringency increased. In situ hybridization validated that top candidates remained lowly expressed in projection neurons in both phox2a siblings and mutants (Figure S5D), Importantly, nearly all candidates had low detection across all neurons in our reference single-cell atlas and had predicted expression in populations such as glia and the caudal hindbrain (Methods, Table 4). Notably, some candidates were highly expressed in the medial vestibular nucleus, which lies on the medial edge of the tangential nucleus and expressed phox2a (Figure S6). Together, we conclude that the any differential gene expression in our data either reflects noise or contamination from other labeled populations, but not projection neurons in the tangential nucleus.

We acknowledge the possibility that our in situ method is insufficiently quantitative to detect subtle differences in expression. Similarly, despite using both bulk and single-cell RNA sequencing approaches, we may lack the resolution to uncover differential gene expression within projection neurons. Nevertheless, consistent with functional and anatomical characterization, our sequencing data argues that projection neurons acquire the correct transcriptional profiles in the absence of motor partner-derived signals. Our findings are reminiscent of recent reports that the molecular signatures of spinal interneurons develop independently of motor partners ¹⁵.

While the primary focus of our work was circuit assembly, we found that, unexpectedly, phox2a acts in a dose-dependent manner to specify extraocular motor pool fate. Key evidence comes from phox2a heterozygotes, in which the earliest-born dorsal neurons in nIII are lost but later-born neurons in nIII/nIV are intact. This observation extends prior characterizations of phox2a mutations in zebrafish ³⁴, chick ³⁶, and human ^37,38. Prior work hypothesized that phox2a dosage may regulate midbrain motor neuron differentiation into visceral and somatic types ³⁶. In other systems, transcription ^52–54, growth ⁵⁵ and axon guidance factors ⁵⁶ can act in such a graded manner to regulate coarse cell type specification and wiring specificity. We extend these ideas to show that phox2a dose-dependency acts both over closely-related subtypes (pools within a single cranial nucleus) and along a temporal axis, where partial dosage preferentially targets the earliest-born neurons ³³. Specifically, if phox2a is expressed in neural progenitor cells that give rise to nIII/nIV, then the earliest-born motor neurons would have the shortest exposure to phox2a.

Molecular insight into ocular motor neuron pool specification is sparse but would be welcome given the strong links between genetic development and ocular motor disease ^38,57–59. For example, subpopulation markers could resolve the topography of pools within dorsal nIII; whether IR/MR pools are spatially segregated or intermingled ^33,60; whether the medial/lateral axis reflects functional differences among motor neuron subtypes; and whether/how local interactions between motor neuron pools contributes to fate specification ⁶¹. In spinal circuits, the rich molecular understanding of motor pool specification ^47,62–65 has enabled targeted perturbations of pool identity, allowing for major discoveries of their roles in circuit assembly ^9,13,15,66. Our findings thus represent a step forward towards understanding how developmental deficits may contribute to ocular motor disorders ⁶⁷.

Our discoveries advance outstanding controversies over whether motor neurons actively or passively shape pre-motor connectivity. We find that extraocular motor neuron axons do not serve as “pioneers” ^68–72, with pre-motor axon targeting following passively from motor-derived pathfinding signals ². Such a model predicts that projection neuron targeting would be entirely ablated after constitutive loss of extraocular motor neurons and their secreted signals ^73–77. Instead, we observed that projection neurons still establish long-range (hindbrain to midbrain) axonal projections, with appropriate spatial segregation that matches the topography of their motor partners ^28,29,33. Our findings complement reports in spinal circuits that pre-motor targeting is grossly appropriate after manipulating the spatial source of, but not ablating, potential pathfinding signals ¹³, and that the transcriptional fate of premotor projection neurons similarly develops independently ¹⁵. We point to the late development of ocular musculature ^78,79 compared to spinal musculature ² as a potential source of the dispensability of muscle-derived signals.

Our work is also inconsistent with the strongest form of the “retrograde” hypothesis for vestibulo-ocular reflex circuit assembly. Originally, the retrograde model posited that motor neurons release a diffusable or cell-surface available signal that instructs premotor collaterals to sprout and then innervate specific pools, enabling behavioral specificity ^3,31. Here, the proper spatial and temporal segregation of projection neuron axons suggests they remain poised to wire with spatially-appropriate (dorsal/ventral pools) targets. Additional evidence comes from incomplete phox2a knockouts (1-5% of nIII/nIV remaining), where projection neurons still form collaterals, though not robustly or reliably. We predict that projection neuron axons do not require a target-derived cue to grow, search, and synapse onto motor targets, and simply lack the adhesive contact necessary to stabilize nascent structures (reviewed in ^45,46).

Nevertheless, extraocular motor neurons might still play an active or passive role in selecting and/or refining input specificity from their projection neuron partners. In spinal circuits, motor pool position passively imposes geometric constrains on pre-motor axon targeting ^13,14, and manipulating the dendritic structure of motor neuron axons transforms input specificity ^9,11. Genetic perturbations of nIII/nIV motor neuron position selectively compromise ocular responses to directional visual stimuli ⁶¹, though the circuit-level origin of such impairments is unclear. For the vestibulo-ocular reflex circuit, transforming all motor pools to the same fate or genetically “scrambling” pool position could resolve whether motor input specificity is truly hard-wired in projection neurons, or whether projection neurons instead target gross spatial domains irrespective of partner identity ¹³. Motor neuron-derived signals are of course capable of shaping their input by strengthening/weakening their inputs. Importantly, our results suggest that such signals will not define the fate of projection neurons, and by extension, circuit architecture.

We note that our study does not eliminate one additional source of post-synaptic partner signals to projection neurons. As in primates ⁸⁰, projection neurons also contact neurons in the interstitial nucleus of Cajal, also known as the nucleus of the medial longitudinal fasciculus (INC/nMLF) ²⁴. INC/nMLF neurons project early in development ⁸¹ to spinal circuits used for postural stabilization during swimming ^82–86. Notably, ablation of projection neurons disrupts postural stability ⁵⁰. As we did not observe postural deficits in phox2a mutants, we infer that projection neuron connectivity to INC/nMLF targets is present and functional. Correspondingly, the development of projection neuron collaterals and synapses to INC/nMLF neurons appeared qualitatively normal in phox2a mutants, supporting our interpretation that projection neurons retain the capacity to properly assemble with post-synaptic targets even though similar structures to extraocular motor neurons are absent. In the future, if a similarly specific marker like phox2a is identified that labels the INC/nMLF, it will be possible to test whether these neurons play a role in vestibulo-ocular reflex circuit development.

What is the origin of signals that govern projection neuron fate and sensory input specificity, if not motor-derived? In comparable systems, fate signals can be intrinsically-expressed or originate from extrinsic sources. For example, intrinsic genetic mechanisms assemble laminar connectivity in visual circuits ^87,88 and facilitate sensorimotor matching in spinal circuits ^9,89–91. In directionally-selective retinal circuits, subtype fate is established in a similar manner ^92,93. In “intrinsic” models, synaptic specificity arises from molecular matching between subtypes ^94–96. Alternatively, in somatosensory and auditory circuits, transcriptional fate depends on extrinsic signals such as growth factors ⁹⁷ and sensation ⁹⁸, respectively. In spinal circuits, positional fate, which constrains connectivity ¹³, is established by extrinsic codes such as morphogen gradients in early development ⁹⁹ and Hox factors ^63,64. In “extrinsic” models, early inputs are often erroneous and refined by activity ^100,101 or molecular factors ¹⁰². Collectively, these findings offer two alternative models for how vestibulo-ocular reflex circuit assembly emerges.

The tight links between birth order, somatic position, and stimulus selectivity ^{28,29,33,85,103,104} across vestibulo-ocular reflex circuit populations support an “intrinsic” determination model. Further, neurogenesis and initial axon targeting develops contemporaneously for sensory afferents ^{55,103,105,106}, projection neurons ^{3,29,107–109}, and extraocular motor neurons ^{3,33,39,110–112}, suggesting that neurons are poised to assemble with targets as early as their time of differentiation. Importantly, an “intrinsic specification” model makes a testable prediction about how and when sensory selectivity should emerge across the circuit: projection neurons and extraocular motor neurons should be directionally selective as soon as pre-synaptic input is established. Such evidence would justify future molecular inquiries into the underlying genetic factors, expanding early characterizations of the mechanisms that shape hindbrain topography ^113,114, recent molecular profiling of the zebrafish hindbrain ^115,116, and reports of molecular matching between extraocular motor neurons and muscle ^117,118. Operationally, the present study lays a foundation for molecular explorations of projection neuron subtype determinants by establishing bulk- and single-cell transcriptomic profiling and in situ validation pipelines.

Conversely, evidence that stimulus selectivity emerges gradually would suggest that sensory afferents and/or projection neurons initially wire indiscriminately and that circuit connectivity is refined in time by extrinsic forces. Prior work in the vestibuloocular reflex circuit has proposed developmental roles for sensory-derived trophic factors ¹¹⁹ and activity-dependent refinement ^120,121, though sensory afferents develop typically in the absence of utricular input ¹²² and ocular motor behavior does not depend on stimulus-driven activity ¹²³. Here, an “extrinsic” determination model would predict that connectivity is established by an anterogradely-transmitted signal – that is, from sensory afferents to ocular muscles. If so, then future investigations might constitutively ablate sensory afferents to eliminate activity-driven, diffusible, or cell-surface instructional signals, similar to the present study. The directional bias in opsin-evoked activity in projection neurons ²⁵, together with their transcriptional profiles established here, offer a clear readout of the role of sensory-derived factors. However, genetic targets exclusive to vestibular sensory afferents for gaze stabilization have not been identified, and tissue-specific genetic ablations remain limited in zebrafish. Looking ahead, resolving when and how stimulus selectivity emerges across the vestibulo-ocular reflex circuit will be key to understanding whether connectivity with pre- and/or post-synaptic partners instructs subtype fate, or whether subtype fate instructs connectivity.