The behavioral onset of the semicircular canal-related angular VOR in Xenopus laevis tadpoles was determined during passive rotation at different developmental stages (Lambert et al., 2008). Respective experiments indicated that the horizontal angular VOR becomes functional around stage 48, which corresponds to a time when tadpoles already swim freely for several days. The obvious absence of a horizontal angular VOR during early larval locomotion is surprising, given the necessity of compensatory eye movements for retinal image stabilization. The relatively late onset of the angular VOR, however, is not related to extraocular motor deficits since otolith- or optokinetically-driven eye movements can be elicited as early as stage 42 (Horn et al., 1986; Lambert et al., 2008), which coincides with the onset of longer swimming episodes (Currie et al., 2016). This suggests that all structural elements, at least for visual and otolith-driven reflexes, are functionally intact and appropriately interconnected. Given the convergence of otolith and semicircular canal signals in 2°VN (Straka et al., 2002b), the failure to transmit signals from the latter vestibular endorgans must derive from a functional incapacity of the semicircular canal sensory periphery. The integrity of the circuitry underlying horizontal semicircular canal (HC)-derived reflexes was tested by electrical semicircular canal nerve stimulation and pressure pulse-induced endolymph displacements of the cupula at stage 47, before the onset of the horizontal angular VOR. These experiments confirmed the full operational capacity of extraocular muscles and central neural pathways, including appropriate response latencies and dynamics. However, more importantly, the experiments also demonstrated the ability of semicircular canal hair cells to evoke graded responses. Thus, the absence of a horizontal angular VOR prior to stage 48 is not related to a general failure of hair cells to sense endolymph motion or to transmit the transduced neural signals onto vestibular afferents. It rather suggests that a passive element of the HC is the limiting factor for the ontogenetic onset of the reflex.

The capacity to detect angular acceleration during head rotation through cupula displacements is critically influenced by the inertia of the endolymph within the tubular structure. This, in turn is directly related to the diameter and the length of the semicircular canals. Accordingly, the endolymph movement, required to sufficiently displace the cupula and shear hair cell cilial bundles, depends on the canal lumen and circuit radius (Oman et al., 1987; Muller, 1999). The resistance for fluid motion through a tube is inversely proportional to the fourth power of the tube radius (Hagen-Poiseuille law). Thus, this parameter is highly relevant for the capability of the semicircular canals to detect angular head motion. Accordingly, the physical dimensions of the semicircular canals are the likely determinant for the functional onset of the angular VOR. Comparison of HC lumen and circuit radii before and after the functional onset of the angular VOR indicated that first reflex measurements coincided with a significant increase in canal dimensions between stages 47 and 48 (Lambert et al., 2008). The earliest above-threshold discharge modulation of the LR nerve was encountered at stage 48, when the average canal lumen and circuit radii were ~60 μm and ~0.44 mm, respectively (right in Figure 2B). These values comply well with theoretically predicted duct dimensions necessary for an undisturbed laminar endolymph flow (Muller, 1999).

The hypothesis that semicircular canal size determines the functional onset of the angular VOR was verified by comparing vertical semicircular canal dimensions and the onset of the vertical angular VOR in Xenopus tadpoles (Lambert et al., 2008). In fact, a modulated SO nerve activity occurred in stage 47 but not in stage 46 tadpoles, despite anatomically complete semicircular canals and fully functional VOR circuitry including extraocular motoneurons and eye muscles (Lambert et al., 2008). The lumen radii of the vertical semicircular canals at the onset of the vertical angular VOR were also ~60 μm as found for the HC at the onset of the horizontal angular VOR. Thus, independent of the developmental stage all three semicircular canals had similar lumen radii at the functional onset of the respective angular VOR (Lambert et al., 2008). This intrinsic comparison thus supports the hypothesis that semicircular canal size determines the angular VOR onset during Xenopus ontogeny, a conclusion that was confirmed in larval zebrafish (Figure 2B; Beck et al., 2004). This further suggests that very small vertebrates, such as amphibian or fish larvae (see left plot in Figure 2B) must grow rapidly to acquire sufficiently sized semicircular canals that allow the detection of angular acceleration (Muller, 1999).

The conclusion that semicircular canals in Xenopus tadpoles need to acquire a particular size to detect angular acceleration is, however, based on a test stimulus acceleration magnitude of 400°/s² (Lambert et al., 2008). This is a critical assumption, since the absence of a horizontal angular VOR in larval Xenopus at stage 46–48 depends on the fact that naturally occurring accelerations do not exceed this value. This, however, has been recently challenged by a study of ontogenetic changes in free swimming Xenopus tadpoles, demonstrating that these animals reach angular accelerations of up to 10,000°/s² during self-motion (Hänzi et al., 2015b). This is considerably higher than the value used to determine the developmental onset of the VOR in the previous study. This suggests that semicircular canals would be in fact capable of detecting naturally occurring angular accelerations during locomotor activity at much younger stages. It is therefore quite possible that semicircular canals become already functional for detecting self-motion immediately after structural completion at stage 46 (Haddon and Lewis, 1991; Bever et al., 2003; Quick and Serrano, 2005). Therefore, high stimulus acceleration magnitudes such as those generated during swimming could theoretically elicit an angular VOR. This interpretation, however, requires closer investigation, since vestibular signal processing during rhythmic swimming is considerably attenuated (Chagnaud et al., 2015). This complies with the suppression of the horizontal angular VOR by ascending spinal locomotor efference copies during actively induced body motion (Lambert et al., 2012). Thus, even though swimming-related angular accelerations could be theoretically detected by very young larvae, peripheral sensory attenuation and central cancellation during self-motion likely prevents an activation of an angular VOR. However, during externally induced passive perturbations of the head/body with lower acceleration profiles, semicircular canal size plays the dominant role in determining the effective sensitivity. This would explain the delayed functional onset of the VOR with respect to completion of the tubular structure as previously suggested (Lambert et al., 2008).

Otolith-dependent counter-rotations of the eyes have been observed in Xenopus tadpoles early after hatching at stage 42 (Horn et al., 1986; Lambert et al., 2008). However, the early appearance of roll-induced eye movements does not necessarily imply that the underlying extraocular motor commands are spatially specific. In fact, directionally appropriate compensatory eye movements require that spatially specific signals from particular regions of the utricular epithelium are mediated onto extraocular motoneurons with matching directional selectivity. The 360° horizontal directionality of the utricular hair cell epithelium, however, challenges a simple vector-oriented establishment of VOR connections between specific utricular sectors and sets of extraocular muscles (Rohregger and Dieringer, 2002). One possibility is a wiring mechanism that depends on genetically encoded molecular markers that allow utricular hair cells with specific directional preferences to be linked with respective vestibular afferent fibers; these in turn would then connect to corresponding sets of 2°VNs that project to specific sets of extraocular motoneurons and eye muscles. While this scenario is not entirely unlikely, it is difficult to imagine that such a hard-wired solution would provide a sufficiently precise spatial specificity for connections between utricular sectors and extraocular motoneurons (Rohregger and Dieringer, 2002). Alternatively, the establishment of spatially appropriate wiring between sectors of utricular hair cells and sets of eye muscles could be spatially fine-tuned along a reference frame such as the visual motion detection system in the pretectum or the orthogonal semicircular canal system (Figures 1B_1,2; Simpson and Graf, 1981).

Roll motion in Xenopus larvae elicits eye movements already at stage 42 (Horn et al., 1986). Experiments, in which the SO nerve was recorded in Xenopus tadpoles during linear translation in different directions in the horizontal plane (for two orthogonal directions see Figure 3A₁) allowed determining the utricular tuning of this extraocular motor nerve (Branoner and Straka, 2015). At stage 46–48, SO motoneurons receive omnidirectional utricular signals during horizontal translational motion (Figure 3A₂), indicating an absence of spatial tuning (Branoner and Straka, 2015). Between stage 49 and 53, SO motoneurons receive otolith inputs with a gradually increasing spatial tuning. At the end of this period, the directional selectivity of the utricular signal, mediated onto SO motoneurons, is narrowed (Figure 3A₃). In fact, the utricular sector (i.e., the optimal hair cell polarization) from which the SO nerve responses originate is now aligned with the ipsilateral posterior vertical semicircular canal (PC) plane (Figure 3B). This alignment suggests that the latter endorgan might provide the reference frame after its functional onset and assist in the directional tuning of otolith responses. This hypothesis was confirmed by selectively suppressing the formation of bilateral semicircular canals by bilateral injections of hyaluronidase into the otic capsule in Xenopus tadpoles at stage 44, i.e., before the anatomical completion of these endorgans (Haddon and Lewis, 1991). At variance with the absence of an angular VOR in these semicircular canal deficient (hyaluronidase-treated) Xenopus tadpoles at stage 52 (Branoner and Straka, 2015), cyclic sinusoidal translation caused a modulation of the SO nerve activity, indicating a functionally intact utricle. However, in the absence of semicircular canals, the modulated SO nerve activity of tadpoles at stage 53 had no preferential translational vector orientation and thus remained omnidirectional and very similar to that of young Xenopus tadpoles at stage 47 (Branoner and Straka, 2015). Accordingly, the lack of directionality of otolith-derived extraocular motor responses in contrast to those of age-matched controls demonstrates a lack of developmental tuning of otolith inputs in SO motoneurons in the absence of semicircular canals. This suggests that the maturation of otolith response directionality during translational motion in extraocular motoneurons depends on functional semicircular canals. The mechanism of this maturation process likely includes stabilization of co-activated sensory inputs from a semicircular canal and a utricular epithelial sector with spatially aligned hair cell specificity during e.g., roll/tilt head/body motion. This might occur either at the level of 2°VN or further downstream at the level of SO motoneurons (Figure 3B).

In all vertebrates including amphibians, semicircular canal and otolith organs in the inner ear develop from placodal regions of the ectoderm at the level of hindbrain segment 5 and 6 (Fritzsch and Straka, 2014). Asymmetric cell proliferation processes cause an invagination of the epithelial region into an otocyst within which distinct cellular patches transform into several vestibular endorgans (Fritzsch and Straka, 2014). The formation of sensory epithelia, primary afferent neurons and functionally appropriate connections between the two structural elements are genetically regulated by paraxial mesodermal as well as ectodermal tissue (Fritzsch et al., 2002, 2007; Fritzsch, 2003). The similarity of the underlying cellular and molecular processes between Xenopus (Quick and Serrano, 2007) and other vertebrates complies with a common regulatory scheme of inner ear formation and endorgan differentiation (Straka, 2010). This is mostly due to the combinatorial expression of evolutionary conserved genetic regulatory elements (Fritzsch, 2003). Vestibular endorgans derive from specific otic placodal areas that are controlled by regionally different genetic patterns (Kil and Collazo, 2001). However, the timing of formation and maturation differs for semicircular canal and otolith organs as shown in Xenopus (Nieuwkoop and Faber, 1994; Quick and Serrano, 2005). While the utricular macula is already well developed at larval stage 42/43 (Nieuwkoop and Faber, 1994), semicircular canals appear initially as paired axial protrusions early after hatching (stage 43). During subsequent development these protrusions fuse and finally become morphologically complete at stage 46–47 (Haddon and Lewis, 1991) at ~5 days post-fertilization (Bever et al., 2003). A differential timing of semicircular canal and otolith organ formation has also been observed in salamanders (Wiederhold et al., 1995) and zebrafish (Haddon and Lewis, 1996; Bever and Fekete, 2002; Whitfield et al., 2002). Thus, a relatively late morphological completion of semicircular canals likely represents a general feature of vestibular organ development in vertebrates. This however might be related to the different anatomical complexity of the two types of vestibular endorgans (Straka et al., 2014). In fact, otolith organs, which also precede the evolutionary appearance of semicircular canals (Fritzsch and Straka, 2014), are structurally relatively simpler organs compared to the more specialized and derived tubular structure.

Vestibular nerve afferent fibers are formed early during embryonic development and link hair cells within the sensory epithelia of the different endorgans with 2°VN (see Straka, 2010). The somata of these afferents form the ganglion of Scarpa, which in amphibians is inconspicuously hidden in the VIIIth nerve at the level of its passage through the cranial wall (e.g., Dunn, 1978; Honrubia et al., 1989; Reichenberger and Dieringer, 1994; Straka et al., 1996). Bundles of afferent fibers from individual semicircular canal and otolith endorgans as well as their respective cell bodies in the ganglion are partially segregated with respect to their peripheral origin (Fritzsch et al., 2002). This tendency of separation into endorgan-specific fiber bundles is likely related to a spatially segregated topography and temporally dispersed formation of the different endorgans. With respect to various morpho-physiological properties such as resting discharge rate and regularity, afferent fibers of both larval (Gensberger et al., 2016) and adult frogs (Honrubia et al., 1989) form a broad spectrum. Despite the apparent continuum of distinguishing features, afferent fibers in Xenopus tadpoles segregate into two distinct categorical groups (Gensberger et al., 2015) with physiological characteristics, reminiscent of phasic and tonic 2°VN of adult frogs (Straka et al., 2004). The obvious duality of the vestibular afferent population in Xenopus larvae complies with a separation into frequency-tuned pathways as a major organizational principle of the VOR (Straka et al., 2009) and its implementation already early after hatching.

In larval Xenopus and adult ranid frogs, vestibular afferents terminate within a recipient area in the dorsal hindbrain between the cerebellar peduncle and the obex (Hellmann and Fritzsch, 1996; Birinyi et al., 2001; Straka et al., 2001; Chagnaud et al., 2015). After entering the brainstem, the bundle of afferent fibers splits into a rostral and a caudal branch that each ramifies extensively within a delimited area (Suarez et al., 1985; Straka et al., 2001). In addition, a noticeable number of vestibular afferents extend into the reticular formation at the level of the abducens nucleus and into the cerebellum. Tracing studies of the vestibular nerve also delineated a small, localized group of efferent neurons, with projections to all vestibular endorgans (Hellmann and Fritzsch, 1996; Straka et al., 2001, 2006; Chagnaud et al., 2015). These neurons, present in all vertebrates are located ventro-medially to the afferent termination area close to the facial branchiomotor nucleus.

Despite the common placodal origin of the inner ear and lateral line mechanosensory systems in anamniotes, afferent fibers from the two locally separated hair cell sensory peripheries, occupy largely distinct, non-overlapping regions in the dorsal hindbrain (e.g., Chagnaud et al., 2015). This indicates the implementation of a topographical principle that is based on afferent projections into modality-specific, evolutionarily conserved central areas (Fritzsch et al., 2005). As shown in Axolotl, the developmental segregation of mechanosensory termination areas is a function of time and space. Later forming inner ear endorgans are innervated by later forming afferent connections that project centrally to more dorsal hindbrain areas (Fritzsch et al., 2005). However, while afferent projections from different mechanosensory systems are segregated, afferents from individual vestibular endorgans overlap to a large extent within the vestibular-recipient area of the hindbrain in amphibians as in other vertebrates (Figure 4A; Birinyi et al., 2001). This pattern, at variance with the well-established retinotectal map in Xenopus (e.g., Debski and Cline, 2002), indicates that the vestibular nuclei lack an obvious structured or layered topographic representation of the sensory periphery (Straka et al., 2000).

The rather negligible central afferent segregation suggests a considerable semicircular canal and otolith signal convergence at the level of individual 2°VN. In contrast with the highly overlapping termination areas of afferents from different inner ear endorgans (Figure 4A; Birinyi et al., 2001; Straka et al., 2003), adult frog 2°VN demonstrate a remarkably high selectivity for monosynaptic inputs from the sensory periphery (Straka et al., 1997, 2002b, 2003). As demonstrated in ranid frogs, most 2°VN select monosynaptic afferent inputs from only one semicircular canal (Figure 4B) and/or from only one otolith organ, despite the availability of inputs from all endorgans (Straka et al., 2002b). This indicates that the decomposition of angular and linear motion vectors by the spatial arrangement of the vestibular endorgans is largely maintained at the first neuronal level of the CNS. However, in many 2°VN, semicircular canal and otolith afferent inputs are combined monosynaptically in a specific manner (Figures 4C,D) that is related to the spatial orientation of the respective endorgans (Straka et al., 2002b). Accordingly, inputs from the horizontal otolith endorgan (utricle) combine preferentially with those from the HC in 2°VN (Figure 4C). In addition, afferent inputs from particular utricular sectors (see color-coded areas in Figure 4C) converge monosynaptically with signals from vertical semicircular canals, likely related to the co-stimulation of utricular and semicircular canal hair cells during tilt movements in the respective canal planes. In a complementary fashion, inputs from the vertical otolith organ (lagena in amphibians) converge in 2°VN exclusively with those from the anterior and PCs (Figure 4D). The ontogenetic establishment of this spatially specific convergence pattern likely requires an activity-dependent mechanism that allows a synaptic refinement based on co-activated sensory inputs during body motion in space. Accordingly, fine-tuning of the precise spatial specificity is likely based on a similar super-numerous production of exploratory axon collaterals as during the development of retinotectal pathways (Witte et al., 1996). Whether synapses from inappropriate connections are lost entirely or only silenced with the possibility of reactivation, as previously suggested after partial lesions of the VIII nerve, remains to be determined (Goto et al., 2000, 2002).

Classical descriptions of central vestibular nuclei in the vertebrate hindbrain are based on anatomical features such as cell density, size and projection patterns (Figure 5; see Büttner-Ennever, 1992, 1999). These features allowed a separation into five subnuclei: a superior (SVN), medial (MVN), lateral (LVN) and descending vestibular nucleus (DVN) and a group Y. The entity of these nuclei forms the vestibular nuclear area and as such coincides with the brainstem recipient area for vestibular afferents (Figure 5F; Büttner-Ennever and Gerrits, 2004). This nomenclature is traditionally used in all vertebrates (see Straka and Dieringer, 2004; Straka et al., 2005) including amphibians (Matesz, 1979; Birinyi et al., 2001). However, the distinction into different subnuclei based on morphological features and hindbrain location prevents conceptualizing underlying organizational principles. More desirable would be a differentiation of vestibular neurons according to developmental origin or a compartmentalization into specific groups of functional phenotypes. An alternative reference frame that allows mapping spatial locations of defined subgroups of central vestibular neurons (Auclair et al., 1999; Cambronero and Puelles, 2000; Glover, 2000, 2003; Straka et al., 2001) is offered by the hindbrain segmentation (Vaage, 1969). While morphological visualization of the rhombomeric framework is usually restricted to vertebrate embryos, most species with a long larval development such as most amphibians retain a visible patterning (Figure 5A; Straka et al., 2002a). This scaffold therefore provides a link between hindbrain position of specific groups of vestibular neurons and their segmental origin, which governs the combinatorial expression of genetic regulatory elements (Gilland and Baker, 1993; Straka and Baker, 2013).

As a unique feature, hindbrain segments of larval frogs are often plainly visible as a rostro-caudal series of up to eight transversely oriented pigmented regions (Figure 5A; Straka et al., 1998, 2002a). However, segments can also be visually defined by glial cell accumulation at the boundaries (Yoshida and Colman, 2000), by labeling with streptavidin Alexa 546 (Figure 5B; Hänzi et al., 2015a) or simply by illumination with a wavelength between 610–650 nm (Figure 5C; Chagnaud et al., 2015). These segments correspond to the rhombomeric scaffold described in Xenopus embryos (Papalopulu et al., 1991). Independent of the visualization method, rhombomeres 2–6 (r2–6) appear as approximately similar-sized rostro-caudal brain regions (Figure 5A; see Straka et al., 2006). In contrast, r1 and r7/8 span across longer regions, respectively, and might in fact consist of multiple segments as suggested earlier (Gilland and Baker, 1993; Cambronero and Puelles, 2000). The visibility of the rhombomeric structure in amphibians, even at older larval stages or in adults, along with the absence of a post-embryonic longitudinal migration at least in anurans (Straka et al., 2006) facilitates a direct linkage between hindbrain location and ontogenetic origin of vestibular pathways (Straka, 2010).

Extraocular motoneurons are located within the oculomotor, trochlear and abducens nucleus (Straka et al., 2014). Within the anuran oculomotor nucleus, situated immediately rostral to the midbrain-hindbrain boundary, individual eye muscle-specific populations (MR, SR, IR, IO motoneurons) are arranged in a partially segregated manner (Matesz and Székely, 1977). This spatial organization is reminiscent of the birth date-related pattern recently described for zebrafish oculomotor and trochlear motoneurons (Greaney et al., 2016). In contrast to the midbrain origin of oculomotor motoneurons, trochlear and abducens motoneurons in anurans originate from rostral r1 and from r5, respectively (Straka et al., 1998, 2006). The mono-segmental origin of the latter nucleus is similar to that reported for mammals but differs from the bi-segmental pattern of fish and birds, where abducens motoneurons derive from both r5 and r6 (see Straka et al., 2006; Gilland et al., 2014). In addition, abducens internuclear neurons with crossed ascending projections to contralateral MR motoneurons in the oculomotor nucleus are intermingled with abducens motoneurons in r5 (Straka and Dieringer, 1991; Straka et al., 2001). These neurons ensure an activation of bilaterally synergistic horizontal eye muscles in order to execute conjugate eye movements in anurans as in all other vertebrates (Straka and Dieringer, 1993). In a classical push-pull organizational scheme, all extraocular motoneurons receive reciprocal excitatory and inhibitory vestibular inputs from the bilateral vestibular nuclei that are spatially matched to the pulling direction of the eye muscles (see e.g., Figure 1C; Straka and Dieringer, 2004).

Eye muscles and extraocular motoneurons are generated early during amphibian embryogenesis and axonal bundles of all six groups of motoneurons appear morphologically complete and identifiable in Xenopus and Axolotl at the time of hatching (Sonntag and Fritzsch, 1987; Matesz, 1990a,b). However, the shape of the motoneuronal dendritic tree undergoes further post-embryonic maturation, visible as regionally restricted extension of the dendrites and an increase in branch number (Matesz, 1990a,b). If the earlier appearance of neuroblasts in the oculomotor compared to the trochlear and abducens nucleus is due to a rostro-caudal developmental gradient, or rather due to a methodological bias remains unclear (Matesz, 1990a,b). Even though extraocular motoneurons are formed by the time of hatching, their number increases during larval development in both anuran and urodeles (Sonntag and Fritzsch, 1987; Matesz, 1990a,b). The gradual augmentation of motoneurons parallels the concurrent increase in the number of extraocular muscle fibers in each eye muscle (Faust and Straka, unpublished observations). This suggests that the ratio for extraocular motor innervation is genetically determined and maintained into adulthood.

The early embryological establishment of eye muscles and motoneurons in amphibians as in other vertebrates (Glover, 2003; Straka, 2010) is compatible with the idea that the formation of the extraocular motor system and its wiring is one of the first steps in VOR circuit establishment and essential in creating the specific framework for mediating vestibular sensory signals onto spatially matching extraocular motor targets (Straka et al., 2014). The morphological completion of all structural elements and interconnections necessary for vestibular reflexes by the time of hatching, suggests that amphibian larvae have a fully functional VOR as soon as the tadpoles start to swim freely (Boothby and Roberts, 1992; Jamieson and Roberts, 2000). This notion fully complies with the interpretation derived from the different behavioral studies on VOR development in that the wiring between the vestibular sensory periphery and the extraocular motor plant has been established at the embryonic stage prior to locomotor onset. However, a further refinement and maturation occurs during subsequent larval stages.