The vergence-mediated gain increase: Physiology and clinical relevance

1 Introduction

The vestibulo-ocular reflex (VOR) evolved to drive compensatory eye-rotations to provide clear vision by stabilizing images on the retina during rapid head movements. Vestibular semi-circular canal primary afferents predominantly encode rotational head velocity (20–30% of canal afferents carry a signal closer to rotational head acceleration), while otolith afferents encode linear head acceleration [24]. The VOR is the only oculomotor system that can stabilize vision during rapid (high frequency and high velocity => high acceleration) head motion [25]. Other oculomotor systems, such as fixation and smooth pursuit, operate at lower frequencies (<1 Hz) and velocities (<80°/s (e.g., [54]). Consequently, the VOR is the main vision-stabilizing system during everyday activities that involve rapid head movement, such as walking, running, looking before crossing a busy street, and driving on a bumpy road. A physiologically relevant stimulus to examine the VOR is the head impulse, defined as a transient (∼200 ms), unidirectional, high acceleration (∼3000°/s²) head rotation. The function of the VOR during head impulses is characterized by ‘gain’. The angular VOR (AVOR) gain is usually defined as angular eye velocity / angular head velocity [26]. The linear/translational VOR (LVOR) gain can be calculated as ‘actual’ eye position (or velocity) / ‘ideal’ eye position (or velocity), where ‘ideal’ eye position (or velocity) is the eye movement required to perfectly fixate the visual target during the head movement (e.g., [16]). For far viewing, the AVOR gain calculation is straightforward, because the velocity of eye movements must simply correspond to the inverse head velocity, i.e., the gain = 1, whereas the LVOR gain is close to zero. For near viewing, the AVOR gain becomes greater than 1, and the LVOR gain becomes greater than zero, both increasing with decreasing target distance. However, the gain change also depends on the axis of head rotation. When the axis is behind the nasion, which is the typical case due to the anatomical position of the eyes with respect to the spine, the eyes translate in the same direction as the head rotation, and so during near viewing the gain must increase from unity to compensate for that translation. However, when the axis is in front of the nasion, the eyes translate in the opposite direction as the head rotation, which during near viewing is compensated by decreasing the gain from unity. For example, when the axis of rotation coincides with the target, stable vision requires VOR suppression, i.e., a gain equal to 0. (For methods of calculations of ideal eye movements/position during AVOR and LVOR see model computations in Supplementary Material).

Fixation of visual targets / scenes during passive rapid high acceleration (i.e., high frequency and high velocity) head movements is demanding due to the short latency required of the compensatory eye movement. Therefore, the reflex neural arc is necessarily short: the direct pathway consists of three neurons with corresponding latency of 5–6 ms [32]. During natural movements, head perturbations can have frequency content approaching 20 Hz, and, partly due to higher stages of neural processing, the VOR is compensatory for head velocity as high as 300–500°/s [32, 64].

During horizontal linear accelerations (e.g., along the interaural axis) the nearer the target, the more the eyes must turn to compensate for the target movement (relative to the head, i.e., the target is not moving in space). Even when the human head only rotates about its natural axis (the cervical spine), both the eyes and otoliths have linear (sideways, fore-aft) components in addition to angular components because the eyes are anatomically positioned anterior to the axis of head rotation about the spine. The vestibular system modifies the VOR gain, depending on-target distance and position (elevation, azimuth), to make eye movements fully compensatory, even during high acceleration head rotations. One adaptive mechanism during binocular near viewing is convergence. Vergence eye movements are disconjugate horizontal eye movements that allow binocular fixation of visual targets at different viewing distances and consist of convergence (the simultaneous adduction of both eyes) and divergence (the simultaneous abduction of both eyes) [7]. The ‘near triad’ consists of: the eyes converging to minimize binocular disparity; the eye lenses increasing in power to accommodate; and the pupils constricting to increase depth of focus [48].

During near-viewing (convergence) the gain increases to above unity to compensate for the relatively larger translation of the eyes with respect to the target. This has been shown for both the AVOR [82] and LVOR [5, 67]. In 1986 Viirre et al. [82] examined the horizontal AVOR while modifying the fore-aft position of the axis of head rotation but keeping it behind the eyes, during fixation of a near target. They found that the AVOR gain magnitude increased with increasing radius of head rotation (i.e. with an axis of rotation increasingly posterior to the otoliths causing increasing [mostly] sideways acceleration) and with decreasing target distance. This gain increase is a ‘vergence-mediated gain increase’ (VMGI) because binocular near viewing involves vergence. When the distances from the two eyes to the target were different, the instantaneous velocities and AVOR gains of the eyes were also different, so the authors noted that the healthy AVOR was capable of simultaneously driving both eyes with different gains. The need for different gains arises from the geometry. For near target viewing, the ideal response of the abducting eye has a larger gain than that of the adducting eye [82]. For example, according to our model calculations (for formula see Supplemental Material), during angular head acceleration around the spinal column, when the fixation target is at 120 cm (far viewing) and the head is straight-ahead, the starting position of the left eye is 1.7° to the right. When the head turns 10° left, the left eye should turn 12.5° right (adducting), a 10.8° degree change. Similarly, the right eye turns from 1.7° left to 9.1° right (abducting), i.e., a change of 10.8°. In contrast, when the fixation target is at 15 cm (near viewing) and the head is straight-ahead, the left eye's starting position is 13.5° right and when the head turns 10° left, the left eye should turn to 28.5° right, a 15° degree change. However, when the right eye turns from starting position 13.5° left to 2.9° right, there is a slightly larger change of 16.4°. So, provided that the leftward head turn has the same velocity profile during far and near viewing, the near target adducting (left) eye velocity gain should roughly increase to 15°/10.8°= 1.39, and the abducting (right) eye gain should be ∼16.4°/10.8°= 1.52 compared to a gain of close to 1 during far viewing.

Purely linear high frequency interaural translations are measured by the utricular organs, which are also stimulated by low frequency or static lateral head tilts. It is this difference in frequency content, in addition to semi-circular canal and proprioceptive input (using a rotational feedback mechanism and the somatogravic feedback loop to derive an estimate of gravity [1]), which the central nervous system uses to differentiate between the two movements. Each of these movements require different compensatory responses, i.e., for tilt - the eyes must move torsionally (counter-roll), whereas for interaural translation - the eyes must move horizontally (laterally). During far viewing, lateral movement of the head does not require significant compensatory eye movements. However, during near viewing, the amplitude of these lateral compensatory eye movements increase in a frequency dependent way, with high pass dynamics [78]. It has to be mentioned that for most of the context of this review we are referring to eye movements in the head coordinate system. The position of the eyes (in head coordinates) should have minimal effect on the axis of eye rotation when measured in head coordinates, though there is some effect due to the VOR / Listing’s Law compromise strategy mentioned later in this review. The situation, however, is different when measured in eye coordinates. For example, in the situation where the eyes are looking up during a purely interaural head translation, the eyes will rotate (almost) purely horizontally in head coordinates, but horizontally and torsionally in eye coordinates.

Over the last decades, several research groups have investigated the VMGI, especially for the AVOR because of its possible clinical relevance. The far viewing AVOR gain is unity in healthy individuals, whereas in pathological cases it decreases and the VMGI can be expressed as a ratio of the near to far viewing gain. In contrast, the far viewing LVOR gain is almost immeasurably small and only has meaning when the target is near. Measurement of LVOR gain is clinically more difficult (and therefore clinically less common) because a purely linear head acceleration is difficult to deliver due to constraints by the cervical spine. Sophisticated equipment is often required to deliver the head acceleration stimulus and data analysis is complex, so unlike the AVOR, the LVOR has been less studied.

In the vestibular nerve, larger diameter irregular afferent fibers carry information from either type I hair cells located at the center of the vestibular sensory neuroepithelium or both type I and type II hair cells [24]. In patients with ototoxic lesions that preferentially affect type I hair cells, such as Meniere’s patients treated with intratympanic gentamicin, the VMGI is missing, suggesting that the irregular afferent pathway, which carries the type I hair cell signal, plays an important role in the VMGI [55]. It has also been suggested that in cases with a unilateral vestibular lesion, rehabilitation may be more effective when patients are exposed to retinal slip during gaze stability exercises that demand vergence (due to near viewing) because this increases the retinal slip signal that drives VOR gain change [12].

Anticipating further developments concerning VMGI as a clinical tool we think it timely to summarize recent findings in this field. In this review, our aim will be to present the available literature on its testing, physiology and pathology. While LVOR VMGI delivers useful scientific information concerning the involved mechanisms and pathways, it is not easy to implement in the clinic, so almost exclusively the AVOR VMGI has been studied. We will see how the vestibular system drives compensatory eye movements at every target distance, and at every physiologically relevant head acceleration and direction. The analysis of the stimulus starts peripherally and using a whole spectrum of information created during the sensory process, a three-dimensional representation is centrally processed and updated to provide the appropriate compensatory output, even in the most demanding situations. We ask if measurement of the VMGI in pathological states (such as after vestibular neuritis) delivers clinically useful information, such as prognosis or effectiveness of rehabilitation that targets compensation and adaptation mechanisms.

2 VMGI testing

Ideally, the VOR is examined during the first 100 ms of a high frequency/high acceleration head movement stimulus, because after 100 ms the other longer-latency vision-stabilizing oculomotor systems, such as smooth pursuit, start contributing to the eye movement response. The VMGI can be examined for the: AVOR using purely angular high acceleration head impulses in the plane of the horizontal or vertical canals; and LVOR using purely translational (fore-aft, interaural, up-down) head impulses or during combined (linear with angular) stimuli. Even unnatural stimuli, e.g., tone bursts administered laterally (i.e., in the horizontal plane) on the temporal bone using a shaker can be used to measure parameters related to the VMGI [76]. The gain of the VOR increases as the fixation target distance decreases. Measurement of the VMGI relies on determining head velocity to ensure that it is of sufficient magnitude to elicit an eye movement response that is predominantly mediated by the VOR. Also, the geometric demand for target fixation must be known, which will be based on the target distance and direction relative to each eye. Head and monocular eye position / velocity can be measured using commercially available head impulse goggles, although in experimental situations researchers use the more invasive, but higher resolution magnetic search coil systems (eg. [41, 51]). Measurement and monitoring of the vergence angle during testing is important, especially with older patients or those with poor vision. Reliable monitoring of the vergence angle immediately before each head impulse (such as a 100 ms window starting 150 ms before head impulse onset [9]) is best achieved using a binocular measurement system, although recently, a monocular technique has been demonstrated that allows for accurate measurement of the convergence angle immediately before a rapid translational or rotational head movement [23]. For clinical use, the VMGI measurement should incorporate a reliable assessment of the actual vergence angle to exclude poor convergence as a cause of poor VMGI. To measure the VMGI, one must measure the VOR gain during far viewing (typically straight-ahead at 120–130 cm from the nasion and at eye level) and near viewing (typically 15 cm) [9, 41]. The near target distance of 15 cm is often chosen because most people can fuse a target at that distance, i.e., the near point of convergence is on average (±SD) 8.59±4.82 cm in the normal population [57, 81]. The AVOR gain is determined by eye and head velocities, which may be measured as area under the velocity curve (e.g., from 60 ms before peak acceleration to the last value of 0°/s [46]) or simply by peak eye-velocity divided by peak head velocity at one point [83] or the median of the instantaneous VOR gains calculated during the 30 ms immediately prior to peak head velocity [47]. The AVOR VMGI is the near viewing gain divided by the far viewing gain. In contrast, the LVOR VMGI is equal to the near viewing gain, because the far viewing gain is close to 0 [39].

3 VMGI drive and behavior

There is a continuous learning process in the vestibular nuclei with long-term memory storage in both the cerebellar cortex and the brainstem [6]. Typically, during foveal fixation of an approaching target, the eyes move disjunctively in opposite directions (adduction). Does the VOR gain increase purely as a function of increasing vergence angle? The human VOR gain has been observed to decrease rather than increase during voluntary convergence, suggesting gain is not directly related to motor vergence [69]. In 1984 Jones et al. [33] showed that the VOR can be suppressed in darkness if the subject imagines looking at a head fixed target. In line with that finding, the reflex can be modified without any changes to external cues, presumably using an internal reconstruction of target and eye movement. In 2001 Han et al. [27] showed that target proximity and vergence angle were not the key determinants of the visuo-vestibular response during head rotation while viewing a near target, but that contextual cues from the perceived motion vision were more important in generating the appropriate response. Also, applying unexpected head perturbations during sinusoidal head rotations while keeping the vergence angle constant, did not increase the VOR gain [28]. These authors concluded that several separate factors contribute to eye rotations during sinusoidal yaw head rotations while viewing a near target: the VOR, visual-tracking eye movements that utilize retinal image motion, predictive eye movements, and possibly, other unidentified non-vestibular factors. Internal models of motor control, sensorimotor integration and sensory processing may also explain / resolve sensory ambiguities, e.g., during post-rotational tilt in humans a linear response component is noted even when no actual linear acceleration is present [50].

The fact that a retinal image position signal is needed to keep the VMGI sufficiently high may seem to contradict the observation that the VMGI has a very short latency (e.g., 18 ms as measured by the effect of lateral tone bursts on evoked myogenic potentials of the lateral rectus muscle [76]). No central gaze pursuit mechanism can be initiated within the VMGI latency window. Considering that contextual cues from motion vision are important in generating the appropriate VOR gain increase, we suggest that a retinal image position signal is necessary to build and maintain a three-dimensional representation of target position in the higher vestibular centers. Hypothetically, the vestibular system maintains this model and supplies the VOR with the necessary gain increase instantaneously. In 1992 Snyder at al. [72] found in monkeys that neither proprioceptive nor efference copy signals were being used exclusively to drive VOR changes and suggested that a cue for modulating the VOR response was derived from a central sensory or motor command signal linked to shifts in visual attention. This visual attention may be maintained by retinal image position signals and influence in turn the stimulus-position context. The authors compared the time-course of changes in the VOR with the time-course of changes in vergence angle in monkeys [72] and found that the changes in the amplitude of vestibular induced eye movements anticipated changes in vergence angle by 50 ms, but in some instances, up to 200 ms of anticipation was observed. For the authors this finding meant that a modifiable central command signal, rather than a fixed function of afferent or efferent copy of binocular (vergence) eye position, was used to modulate the VOR.

In 2013 Chim et al. [15] used a strobe light (0.5–15 Hz and constant light) to illuminate a target that was in otherwise darkness, to examine the role of visual feedback on the VMGI. The strobe lighting frequencies were presented in either ascending or descending order (randomly assigned to each subject), followed by constant light and constant dark (with imagined near target). The VOR gain increased during near viewing compared to far viewing during all lighting conditions, but more so at the higher strobe frequencies and during constant light. Interestingly, strobe frequency had no effect on the far target VOR gain, suggesting that the far viewing VOR was not as dependent on visual feedback as the near viewing VOR. The order of strobe frequency presentation (ascending or descending) did not affect the gain, but it did affect the vergence angle (angle between the two eye’s lines of sight). The VOR gain and vergence angles were constant during each trial. These findings showed that a retinal position error signal helps increase the vergence angle (visual feedback contributes to the VMGI) and could be invoking central mechanisms to increase the high-frequency VOR response during near viewing. However, the fact that the VOR gain at a given strobe frequency did not change depending on the presentation order (ascending or descending), suggests that time (a given frequency was presented earlier or later depending on whether it was ascending or descending) was not a significant factor for gain change. If an internal model was being built, then the VMGI would have been increasing with time, i.e., not with visual-feedback magnitude, which was the case in this study. Rather these data suggest that the immediate visual context prior to the head movement contributes most to the VMGI, i.e., eye position, plus a well-foveated target drives a greater (i.e., closer to ideal) VMGI.

The results of Ramat et al. [61], however, seem to support the ‘internal model’ theory. The authors examined the ability of the brain to enhance and suppress the LVOR response to brief high acceleration interaural head translations while viewing a near (15 cm) target. They found when subjects had a priori information about the position of the target and/or direction of the head movement, modulation of the LVOR response occurred within a period shorter than the (90–100 ms) latency needed for visual information to produce eye movements, i.e., smooth pursuit or saccades. For the authors, this finding suggested that corrective saccades were being preprogrammed due to cognitive factors. Busettini et al. [10] suggested that multiple, usually redundant cues (vergence, accommodation, target size, etc.) to target distance are employed to modulate the VOR responses to linear stimulation. G.D. Paige [58] suggested that vergence has a strong effect on the LVOR only at high frequencies and that the LVOR is influenced both by the vergence angle and by the imagined target motion [59], with the former exerting its most reliable effect only at frequencies exceeding 2 Hz. Today there is a consensus that it is not the vergence angle itself that sets the VOR gain during near viewing, but the context and stimulus conditions during testing [20, 43]. The role of perception on the VOR is not fully understood, e.g., it has been shown that perception does not scale with target distance [39]. Also, the vergence angle on a well-foveated target is not the sole determinant of the VMGI response, e.g., using diverging and converging prisms Lewis et al. [45] showed that subjects can learn to use the vergence angle as a contextual cue to retrieve adaptive changes in the AVOR. The connection between perception, an internal model and the VMGI will have to be explored in future.

Study of the head impulse vertical AVOR during near viewing also provides insights on the VMGI drive. In 2003 Migliaccio et al. [51] showed using vertical (pitch) head impulses a higher-than-predicted eye rotation axis tilt of approximately one-third from the gaze-normal plane. The authors suggested that this additional tilt in the torsional plane was because the AVOR and LVOR both contributed to the vergence-mediated eye rotation response during near viewing. The authors also showed that the vergence-mediated change in the axis of eye rotation during the VOR could occur independent of eye position, and that the AVOR and LVOR obeyed a VOR/Listing’s Law compromise strategy. (Listing’s Law defines the plane in which the axes of eye rotations are confined to during saccadic eye movements.) Taken together, these findings suggested that changes in the axis of eye rotation during convergence was not simply due to mechanical mechanisms affected by eye position, e.g., muscle pulleys, but due to central oculomotor signal changes. These signal changes seemed to occur faster than visual processing feedback allows, i.e., occurring at ∼40 ms after head impulse onset, suggesting they are pre-programmed and primed by the pre-head impulse visual-context. [51] reported that the head impulse vertical AVOR VMGI was asymmetrical, i.e., ∼1.3 for upward head impulses and ∼1.6 for downward head impulses (target placed 15 cm directly in front of the eye in primary position), which the authors explained by geometric considerations, i.e., during downward head impulses the eyes translate towards the target, whereas during upward head impulses they translate away from the target.

The head impulse torsional AVOR during near viewing behaves differently compared to both the vertical and horizontal VOR. In fact, Migliaccio et al. [53] showed that the torsional VOR gain does not change between near and far viewing due to the presence of a torsional quick-phase with onset prior to peak head impulse velocity. The onset and magnitude of this quick phase is earlier and larger during near viewing compared to far viewing. In effect, the quick phase undoes the torsional AVOR VMGI, so that the gain and resting position of the eyes at the end of the head impulse is the same for both near and far viewing. The authors argued that when the eyes are converged, a pure roll eye rotation, i.e., about the naso-occipital axes as occurs during a roll head impulse, creates a vertical skew between the eyes resulting in diplopia. The torsional quick-phase, most pronounced during near viewing, could serve at least three functions: (1) reset the retinal meridians closer to their usual orientation in the head, (2) correct for the ‘skew’ deviation, and (3) take the eyes back toward Listing’s plane (which is partially obeyed during the horizontal and vertical AVOR, but violated during the torsional VOR).

4 The VMGI is under-compensatory

Viirre et al. [82] showed that when the distance to the target was different for each eye, the VOR gain was also different for each eye. This difference in gain was observed at high frequencies of head rotation (2 Hz), suggesting that they were not mediated by smooth pursuit or visually driven vergence systems. The authors suggested that the oculomotor system makes use of a fast nonvisual estimate of current target location relative to the head. Gain differences between eyes during near viewing have also been shown by [23]. In their vergence experiments, the authors noted that there were significant differences in VOR gain calculated for each eye depending on whether the eye started in primary position or was adducted: during near viewing the eye starting in primary position had a greater gain (on average 1.25) than the eye starting in the adducted position (on average 1.14). How much greater that gain was depended on whether the head rotation caused the eye starting in primary position to adduct or abduct. These differences between eyes can be partially explained by the geometric demand to maintain binocular fixation (See Introduction and Supplemental Material), i.e., the eye that must move angularly furthest to maintain fixation has the higher gain. In four humans Lasker et al. [41] found no difference in the VOR gain values for the two eyes, but the trajectory of the responses for the adducting eye showed more prominent phasic dynamics than those recorded from the abducting eye, suggesting a difference in the signal going to each eye.

The human AVOR is frequently under-compensatory for near targets if we consider the ideal increases of 39% (adducting eye) and 52% (abducting eye) for a target at 15 cm (as mentioned in the Introduction). According to [18], the AVOR VMGI in humans was not ideal for near targets, only reaching 70% of the ideal value with the head rotation axis posterior to the head. Typically, the AVOR gain increase due to near viewing is not compared to the geometric ideal in clinical literature. According to Lasker et al. [41], the gain of the horizontal AVOR at the peak of the velocity response for a far target was 1.01; while for a near target it was 1.25. Tamas et al. [75] had similar results: the horizontal AVOR gain increase was on average 28% for impulses to the right during convergence (although using an atypical, non-physiological bone vibration stimulus applied laterally). In Ujjainwala et al. [81], healthy volunteers had a VMGI of 0.10 (nominal change in gain) for rightward and 0.16 for leftward head rotations. Similarly, Chang and Schubert [12] measured an AVOR VMGI of 0.17 (nominal change in gain) for rightward and 0.18 for leftward head rotations in a healthy population. In Chim et al. [15], the AVOR VMGI became larger with increasing strobe frequency, from 1.17±0.17 in constant dark (imagined near target) to 1.36±0.27 in constant light.

The LVOR slow phase is modulated by target distance as evaluated in comparison to the calculated ‘ideal’ slow phase. For example, Ramat et al. [62] showed that in healthy persons the slow phase response was 30–50% of the calculated ‘ideal ‘ slow phase. Similarly, Crane et al. [20] reported that the LVOR slow phase amplitude was suboptimal. Measured 100 ms after head translation onset, the mean response was 20% of ideal for the target at 15 cm, 22% at 25 cm, 31% at 50 cm, and 53% at 200 cm. Tian et al. [79] showed that the slow phase response was no less than one-half of the ‘ideal’ response for their most distant near target, but no more than one-fifth of the ‘ideal’ for their closest near target. Wei and Angelaki [84] showed that the LVOR slow phase amplitude during passive movements was under-compensatory for near targets and was only enhanced with the addition of simultaneous horizontal canal stimulation with congruent cues from the vertical canals. The authors cited a hypothesis from the literature explaining that this greater response occurs because gaze stability during actively generated head translations always involves an actively generated head rotation in the compensatory direction. Ramat et al. [62] speculated that the LVOR VMGI is less compensatory than the AVOR VMGI, because the VOR compensates for head movements with both angular and translational components (usually involving the simultaneous stimulation of the canal and otolith afferents). Purely lateral motion of the head in the absence of angular movement may be a stimulus that the vestibular system does not encounter naturally and therefore does not compensate well.

5 VMGI Latency

Unlike the other oculomotor systems, such as smooth pursuit, which have long latencies (80–120 ms) because they require multisynaptic retinal and central processing [5], the VOR is fast-reacting due to a short reflex arc with latency 5–10 ms during which the sensory signal from the vestibular organ is transformed into a motor signal that moves the eyes [3, 42]. In humans, the latency of the VOR is around 7–10 ms after onset of head rotation [18] or more precisely 7.5±1.5 ms [41]. Since the VOR VMGI probably involves scaling of the far viewing response in the brainstem, it is likely to also have a short latency [1].

In animal experiments, the horizontal AVOR VMGI has a latency of 60 ms (30 ms for combined AVOR and LVOR) [14], and less than 20 ms for vertical translations [11]. Snyder et al. [71] found during sudden off-axis rotations that the eye velocity traces of macaque monkeys recorded during near viewing diverged after 18 ms from the traces recorded during far viewing. The authors identified successive response windows: the first 10–20 ms corresponded to the short latency VOR reflex; after 20 ms viewing distance began to affect eye velocity, after 30 ms an otolith derived signal was added, and finally, between 45 and 100 ms the translation of the eyes relative to the otolith organs was calculated and a fully compensatory VOR was generated. In contrast, Angelaki and McHenry [2] found in monkeys that even the earliest components of the eye position response to translation scaled with target distance, i.e., horizontal eye velocity for viewing distances of 40 and 10 cm differed by 1 SD as early as 13 ms. Snyder and King [70] found that linear acceleration required 30 ms to produce an effect on the AVOR of monkeys compared to far viewing, but that this effect occurred as early as 20 ms for the LVOR.

In humans, the latency of the VMGI is as short as 1–11 ms after onset of eye movement according to [18]. Target distance effects depend on acceleration in a nonlinear fashion, i.e., for 1000°/s² it was 32 ms, at 2800°/s² it was 8 ms. This nonlinearity is compatible with the hypothesis that the VMGI is mediated by the highly nonlinear phasic primary vestibular afferents. The horizontal AVOR VMGI had a latency of 21.6 ms as measured during passive head impulses [41]. The horizontal LVOR gain changes 30 ms after linear head heave onset (longer than the AVOR possibly because of the greater inertia of the human head during translation) [19]. The LVOR VMGI latency was found shorter, at 18 ms, when measured by recording muscle potentials of the lateral rectus during laterally applied (i.e., utricular) short 500 Hz stimuli using a mini-shaker during both far and near viewing [76]. In humans, the 3D AVOR eye velocity axis during near viewing begins to differ from the far viewing axis 40 ms after angular head impulseonset [51].

During combined linear and angular (AVOR + LVOR) head accelerations in humans, the gain was found to be independent of both target distance and eccentricity in the first 30–50 ms [18]. An otolith mediated effect modified the VOR gain depending on both linear acceleration magnitude and target distance beginning 25–90 ms after onset of head rotation (see also [20]). Collewijn and Smeets [17] found a variable time course in the enhancement of velocity gain by the near target, but that on average, this increase in gain was manifest after 40 ms.

6 VMGI Peripheral signal processing

Primary vestibular afferents tend to fall into one of two categories: irregularly and regularly firing. Afferents with high discharge irregularity in the absence of stimulation (irregular afferents) have gains and phases that are greater than those of regular afferents over the frequency range of natural head movements and are two times more sensitive to head motion at higher frequencies (i.e., 15 Hz) compared to regularly discharging afferents. Regular afferents, on the other hand, transmit two times more information about head motion than do irregular afferents over the physiological frequency range [21, 30]. Regular and irregular afferents effectively comprise two parallel information channels: irregular afferents encode high frequency stimuli with higher gains through changes in their firing rate; and regular afferents transmit information about the detailed time course of the stimulus through precise spike timing [21]. Hullar et al. [30] showed in chinchillas that low-gain irregular afferents are well suited for encoding the onset of rapid head movements, a property that would be advantageous for initiation of the VOR with a short latency. Regularly discharging afferents, on the other hand, provide a plausible signal to drive most of the AVOR response, even during high frequency head motion, but these are not a likely source for the nonlinearities present in the VOR [31]. The nonlinear component of the reflex appears more modifiable [16]. Therefore, it is likely to be important for the VMGI. The loss of VMGI for ipsilesional rotations after intratympanic gentamicin treatment, which preferentially affects irregular afferents, is consistent with the anticipated effect of selective loss of the nonlinear component on the treated side [55] (see below in ‘Pathology’). Modelling nonlinear neural computations at the premotor level have been shown to be consistent with experimental observation [63].

7 VMGI Central pathways

Because the VMGI has a short latency, modification of the VOR gain during near viewing must be due to a primed effect on the classic three-neuron arc of the VOR [18]. Ramat et al. [61] suggested that higher-level cognitive control of the VOR could occur as early as the synapse of peripheral afferents on neurons in the vestibular nuclei, either directly from higher level centers or via the cerebellum. The sensitivities of position-vestibular-pause and eye-head-velocity secondary neurons in the vestibular nuclei increase with closer target distance [49]. Irregular primary vestibular afferents likely modify the gain and phase of these secondary neurons [13]. In addition, visual error signals due to retinal image slip project onto cerebellar Purkinje cells via the climbing fibers (e.g., [73, 74]), which in turn project onto the secondary neurons mentioned above to modify the VOR (e.g., [68]). The VOR gain anticipates increases in vergence angle by at least 50 ms, which means that the internal representation of vergence angle modulates VOR gain [48, 72]. Snyder et al. [72] showed that the VOR changes its response, when there is a change in target position, before the eye movement to acquire the target occurs. This interpretation suggests that the drive modulating the VOR response is derived from a central sensory or motor command signal that is linked to a shift in visual attention. The authors hypothesized that this central command signal could drive the vergence system via the mesencephalic vergence cells and might also be responsible for the transient response of floccular Purkinje cells to changes in vergence, which in turn could modify the VOR response. They argued for the importance of Area 7 in the posterior parietal cortex (a centrum for visual attention), which has direct projections to regions in the vestibular nuclei. This same group later measured the activity of type I gaze velocity Purkinje cells in the cerebellar flocculus and ventral paraflocculus (these cells represent a side loop of the VOR pathways) and found that these cells discharged 25 ms after the onset of head translation. Also, after an additional 15 ms, they changed their firing depending on viewing distance, thereby validating the role of the cerebellar flocculus in the VOR response to rotation-induced translation.

Khojasteh and H.L. Galiana [38] modeled the VOR using nonlinear integration and showed that the vestibular nucleus was a plausible site for context-specific VOR gain modifications. They also presented a bilateral model for the horizontal AVOR based on realistic physiological mechanisms and showed that by assigning proper nonlinear neural computations at the premotor level, the model was capable of replicating target-distance-dependent VOR responses in agreement with geometrical requirements [63]. Lasker et al. [40] found that there was an increase in the nonlinear contribution with near target viewing. In their model, the “far viewing” pathway was represented by a constant (linear) gain term, i.e., so that the far viewing VOR gain was 1. The “near viewing” pathway was represented by a first-order lead term (i.e., a nonlinear term because it increased with head rotation frequency), a gain that was dependent on viewing distance, and a delay. The authors examined a patient with unilateral vestibular hypofunction. Ipsilesional head impulse VOR responses showed no VMGI, whereas contralesional responses showed normal VMGI, as predicted by their model. Their model also showed that the near-viewing pathway was more susceptible to inhibitory cutoff than the far viewing pathway, suggesting that the nonlinear component, likely mediated by irregular primary vestibular afferents, is predominantly responsible for the VMGI. A later study showed that canal plugging in humans, which halves the far viewing head impulse AVOR gain, but should not preferentially affect linear or nonlinear components of the AVOR, did not alter the VMGI (as a percentage increase) [56]. In contrast, a peripheral lesion caused by intratympanic gentamicin, which also halves the far viewing head impulse AVOR gain, but thought to preferentially affect type I vestibular hair cells, resulted in elimination of the VMGI [55]. Type I vestibular hair cells are located in the central zone of the crista, where irregular afferents predominate. The authors hypothesized that irregular afferents provided the necessary signal for vergence-mediated modulation of the VOR. In invasive animal experiments, galvanic stimulation shown to silence irregular afferents, reduced the AVOR VMGI by an average of 64% [13]. In humans, transmastoid galvanic stimulation had minimal to no effect on the VMGI, probably because during head impulses the vestibular signal carried by irregular afferents was so large that any change in the afferent firing rate (due to non-invasive, lower level galvanic stimulation) had a comparatively negligible impact on theAVOR [52].

8 AVOR VMGI and peripheral pathologies

Crane and Demer [18] examined 4 human subjects after unilateral surgical vestibular deafferentation. When rotated to the normal unaffected side, the VMGI was normal, whereas for rotations towards the lesioned side, the VMGI was smaller and often insignificant. It is not clear to what extent the VMGI is preserved for rotations towards the lesioned side in patients with vestibular neuritis. Vestibular neuritis is thought to be elicited by a post viral neuro inflammation of the vestibular nerve [4]. Because of how the vestibular nerve branches, the different semicircular canals are not always involved uniformly. In cases with a severe initial decrease of gain, the affected canals have a 50 per cent chance of significantly recovering [8]. Büki et al. [9] showed that 11/20 patients with unilateral vestibular neuritis had a significant ipsilateral VMGI, suggesting that the effects of vestibular neuritis on irregular and regular vestibular afferents are variable.

Testing patients with unilateral peripheral vestibular lesions (17/22 with vestibular neuritis, 2/22 unknown cause, 2/22 labyrinthine infarction) Chang and Schubert [12] found that on average the VMGI during near viewing was preserved for contralateral rotation, but impaired for ipsilesional rotations. Some subjects showed a partial recovery in the VMGI concurrent with recovery of their passive far viewing VOR gain. In that study, 11 patients had a chronic unilateral peripheral vestibular lesion after vestibular neuritis that did not improve after the acute phase (i.e., gain <0.7). Of these 11, only six patients had a VMGI greater than 5% (four greater than 10%). Ujjainwala et al. [81] tested 6 patients after vestibular neuritis and showed that only 2/6 cases had a VMGI (of 26% and 55%) for rotations towards the lesioned side. Again, these results suggest that vestibular neuritis effects on vestibular irregular and regular afferents varies between individuals. Presumably, those individuals with neuritis that minimally affected irregular afferents had preserved VMGI, whereas those maximally affected had no VMGI. Another confounding factor is whether otolith afferents have been affected by the neuritis [77], which could potentially affect both the VOR gain and VMGI. For example, a recent study in mice showed that the AVOR was the same between normal (wild type) mice compared to mice with non-functioning otoliths (Otop 1), suggesting that normally the AVOR is mostly mediated by the semicircular canals [36] [Khan et al., 2019a]. However, after a unilateral lesion, the Otop 1 acute and chronic AVOR responses were significantly smaller than for the wild type, suggesting that the otoliths can sense angular head rotations and can contribute to the AVOR [37].

Büki et al. [9] examined the connection between the VMGI and VOR adaptation training, where irregular afferents are also thought to play an important role (e.g. [16]). However, the authors showed no correlation between the VMGI and VOR adaptation, suggesting they were mediated by separate mechanisms. Visual loss influences the VMGI as well. Judge et al. [34] found that patients with visual loss had lower VOR gains and lower VMGIs compared to controls. As mentioned above, intratympanic gentamicin therapy ablates the VMGI [55], whereas it is unaffected by canal plugging [56], transmastoid galvanic stimulation [52] and cold caloric irrigation [75]. This latter was examined by Tamas et al. [75] using head impulses during unilateral cold stimulation with water irrigation. This setup can be compared to canal plugging (because it hypothetically leaves the irregular and regular afferents intact, but attenuates the signal these carry) but with the difference that there is also an acute vestibular asymmetry. Also, it can be compared to galvanic stimulation because there is an acute asymmetry, but with the difference that irregular afferents are unaffected by the caloric stimulation. The authors wanted to determine whether an acute vestibular asymmetry abolished the VMGI. They showed that only the ipsilateral head impulse AVOR gain could be lowered by the cold caloric irrigation and the VMGI was not abolished by the partial peripheral hyposensitivity. Like canal plugging, but unlike low doses of intratympanic gentamicin in humans, cold caloric irrigation does not preferentially inhibit the vestibular responses of irregularly dischargingafferents.

9 Clinical considerations

In patients with unilateral lesions, the high frequency / high acceleration AVOR gain is lower for ipsilateral head rotations. However, the ipsilateral AVOR VMGI is not uniformly missing in these patients when considered in terms of percentage gain increase, i.e., not so in terms of nominal gain increase. There are two questions that should be answered in future studies to explore the clinical significance/usefulness of VMGI measurement:

The VMGI is part of every LVOR examination, because compensatory eye movements during translation are larger, and hence clinically noticeable, only during near viewing. In patients with unilateral vestibular hypofunction following intratympanic gentamicin injections, both the rotational and translational VOR is asymmetric [62]. Kessler et al. [35] showed that the head heave test (for LVOR) detects acute, asymmetric otolith diseases. However, the peripheral otolith lesion, either utricular (interaural translation, along the Y axis) or saccular (vertical translation, along the Z axis), can only be diagnosed if there is no other explanation for the VMGI loss. Purely interaural translation may be a cleaner test of utricular function than that of vestibular myogenic evoked potentials measured using electrodes under the contralateral eye (oVEMPs). A potential confounding aspect of oVEMP is that bone conducted vibrations also activate irregularly discharging canal afferents, albeit less effectively, than utricular afferents [22]. The clinical value of LVOR VMGI has yet to be determined because interaural translation excites both utricular organs, so although asymmetry is evident in patients with acute unilateral utricular dysfunction, it is not always present in patients with chronic dysfunction [19, 62]. Since the afferents that mediate the LVOR come from the mid-lateral part of the utricle, compensation may depend on the off-directional responses from hair cells in the remaining utricle [44]. Future work on sufficiently challenging stimuli, the role of central compensation and the behaviour of the LVOR VMGI (depending on the remaining function of irregular afferents) in (unilateral) peripheral vestibular dysfunction will be needed to elucidate potential clinical value. At present there are no studies comparing otolith function measures using head heaves versus oVEMPs. This may be an important topic because it has been shown that the AVOR is mediated by a combination of otolith and canal inputs, which becomes important after a unilateral labyrinthlesion [36].

As a first step towards the introduction of VMGI measurement in the clinic, there is a need for standardization. For results to be comparable, methods of ideal eye velocity calculation, target distance, target presentation (if strobe, in constant dark or light) and time window of eye velocity measurement (e.g., maximum speed, median value in a time window, or area-under-the-curve calculation) should be standardized. Also, standardized monocular [23] or binocular regular eye position monitoring should be applied, i.e., the examiner must be sure that the VMGI is not missing because the patient did not fixate the near target. Starting eye position (straight-ahead, adducting or abducting) is also important because the ideal AVOR is capable of mediating a different gain for each eye depending on its starting position and direction of rotation [82].

VOR adaptation can be induced by deliberately creating a mismatch between the vestibular stimulus and visual scene or target [66]. Further studies are needed to determine whether increasing retinal slip during head impulses while viewing a near target may have a beneficial clinical effect. Patients with unilateral peripheral lesion (after vestibular neuritis) retain their VMGI and ability to enhance their VOR gain via incremental VOR adaptation training, but there is little evidence that these are mediated by the same pathways or mechanism. The ability to increase the AVOR gain during vergence is not prognostic for successful adaptation training [9]. Perhaps in patients where incremental VOR adaptation training is not successful, VOR gain adaptation can be achieved by having them perform head impulses while looking at a near target. For example, Williams et al. [85] showed that 6 minutes of adaptation training while viewing a near viewing target results in AVOR gain change. However, more work will be needed to determine the effectiveness of this type of rehabilitation.

In conclusion, currently there are technical barriers that hinder VMGI measurement in the clinic. There are problems with interpretation because of non-uniform stimuli and protocols, and lack of clinically available equipment that allow reliable binocular tracking during high acceleration/high velocity head movements. However, VMGI may have diagnostic value, especially with regards to measuring otolith function in response to a physiologically relevant stimulus (i.e., head heaves), and further experiments may supply evidence of its eventual advantages to aid rehabilitation both indirectly and directly. Indirectly, because it may provide insight about the patient’s lesion and how to best tailor for them a rehabilitation program. Directly, because VOR adaptation training during near viewing may provide therapeuticvalue.

Conflict of interest

The authors report no conflicts of interest.

Dr Büki was supported by the grants RTO005 (“Octavus”) and SF06 of the Karl Landsteiner Private University of Health Sciences, Austria.