Vestibular, locomotor, and vestibulo-autonomic research: 50 years of collaboration with Bernard Cohen

Abstract

My collaboration on the vestibulo-ocular reflex with Bernard Cohen began in 1972. Until 2017, this collaboration included studies of saccades, quick phases of nystagmus, the introduction of the concept of velocity storage, the relationship of velocity storage to motion sickness, primate and human locomotion, and studies of vasovagal syncope. These studies have elucidated the functioning of the vestibuloocular reflex, the locomotor system, the functioning of the vestibulo-sympathetic reflex, and how blood pressure and heart rate are controlled by the vestibular system. Although it is virtually impossible to review all the contributions in detail in a single paper, this article traces a thread of modeling that I brought to the collaboration, which, coupled with Bernie Cohen’s expertise in vestibular and sensory-motor physiology and clinical insights, has broadened our understanding of the role of the vestibular system in a wide range of sensory-motor systems. Specifically, the paper traces how the concept of a relaxation oscillator was used to model the slow and rapid phases of ocular nystagmus. Velocity information that drives the slow compensatory eye movements was used to activate the saccadic system that resets the eyes, giving rise to the relaxation oscillator properties and simulated nystagmus as well as predicting the types of unit activity that generated saccades and nystagmic beats. The slow compensatory component of ocular nystagmus was studied in depth and gave rise to the idea that there was a velocity storage mechanism or integrator that not only is a focus for visual-vestibular interaction but also codes spatial orientation relative to gravity as referenced by the otoliths. Velocity storage also contributes to motion sickness when there are visual-vestibular as well as orientation mismatches in velocity storage. The relaxation oscillator concept was subsequently used to model the stance and swing phases of locomotion, how this impacted head and eye movements to maintain gaze in the direction of body motion, and how these were affected by Parkinson’s disease. Finally, the relaxation oscillator was used to elucidate the functional form of the systolic and diastolic beats during blood pressure and how vasovagal syncope might be initiated by cerebellar-vestibular malfunction.

INTRODUCTION

The past 50 years have seen an explosion in research on the role of the vestibuloocular reflex (VOR) in stabilizing gaze in space, the maintenance of spatial orientation during rotations and translations of the head and body in space, and how peripheral and central vestibular mechanisms contribute to motion sickness. There has also been considerable research on the mechanisms of animal and human locomotion and how balance and gaze are maintained during natural locomotion. More recently, there has been considerable research into the role the vestibular system plays in the maintenance of blood pressure (BP), heart rate (HR), and orthostasis and controlling syncope. The collaborative work that I have done with Bernard Cohen since 1972 spanned all of these areas, and Bernie Cohen’s fundamental contributions in elucidating the workings of all of these sensory-motor systems was foundational. Because of the vast number of contributions to a wide range of topics, it is not feasible to review the details of the contributions in a single paper. Rather, this review considers from a historical perspective the contributions made by Dr. Cohen that we collaborated on. Much of this review is found in reviews that Dr. Cohen and I have written over the years (Cohen and Raphan 2004; Raphan and Cohen 2002), and the reader is referred to these reviews for a more comprehensive treatment. There were also numerous collaborators who worked with us over the years who contributed to these studies. This review attempts to give the reader a flavor of the breadth and depth of Dr. Cohen’s contribution to a wide range of topics and the threads and theoretical underpinnings that have driven the research.¹

SPATIAL AND TEMPORAL PROCESSING OF THE SEMICIRCULAR CANALS

One of Dr. Cohen’s singular contributions to the vestibular field, which has now taken on considerable significance in application to vestibular prosthetics by Della Santina and colleagues (Hageman et al. 2016; Johnson et al. 2016; Rabbitt et al. 2016), was to define the role of the semicircular canals in the generation of head and eye movements. From earlier work on vestibular processing it was known that the performance of the VOR in three dimensions was determined by the combined activation of the semicircular canals and the otolith organs as well as by the coordinated function of central integrative processes that drive the oculomotor system (Szentágothai 1950). Bernard Cohen contributed to the understanding of the relationship of the semicircular canals to eye movements by showing how the head and eyes move in canal planes in response to stimulation of the individual canal nerves (Cohen et al. 1963, 1964a, 1964b, 1965; Cohen and Bender 1966). There are six extraocular muscles (3 pairs) that rotate the eye about the pitch, roll, and yaw axes of the head: the lateral rectus, medial rectus, superior rectus, inferior rectus, superior oblique, and inferior oblique muscles. Pairs of muscles work in push-pull and rotate the eye about a specific axis. For example, the torque axis of the lateral and medial rectus muscles rotates the eye about an approximate yaw axis, tilted back relative to the yaw axis of the head. Similarly, the superior and inferior recti rotate the eye about the pitch axis. The superior oblique and inferior oblique muscles rotate the eye about an approximate roll axis. When the head is restrained, electric stimulation of the left anterior canal nerve causes the eyes to move up and roll clockwise from the animal’s perspective. Left lateral canal stimulation causes the eyes to move to the right, and left posterior canal stimulation causes eye movements down and counterclockwise. When the head is free and allowed to move, stimulation of individual canal nerves moves the head in the same direction as the eyes had moved when the head was fixed (see Cohen and Raphan 2004, Figs. 6 and 7). During large saccades made during lateral gaze shifts, however, the VOR is inhibited and the eyes and head move together (Laurutis and Robinson 1986). By combining the canal nerve stimulations, eye and head movements could be produced about any head axis. The fact that any combination of eye and/or head movements can be induced by stimulation of the individual canal nerves is verification that all head and eye movements are coded as a superposition of the vectors along the axes that are normal to the planes of the semicircular canals and that these normals form a coordinates basis for representing head and eye movements (Cohen and Raphan 2004; Raphan and Cohen 2002).

Because the eyes were shown to move so precisely in canal coordinates, this work engendered considerable interest in understanding how the torque axes of the eye muscles are related to the canal axes. The pulling directions of the primary activated and inhibited muscles were shown to be closely aligned with the planes of the canals (Ezure and Graf 1984a, 1984b; Simpson 1984), and there is a close correspondence between the canal planes and the torque axes of the eye muscles that are primarily activated or inhibited by those canals. This association between canal planes and muscle axes was shown to be across a wide range of species from lateral-eyed to frontal-eyed animals. These studies also established that there was a common coordinate frame for vestibular input and oculomotor output that allows the torque axes for reciprocal muscles to work directly in synergy with the canals without having to compute a different set of muscle activations for every eye position (Cohen et al. 1966b; Ezure and Graf 1984a, 1984b; Graf and Simpson 1981). It should be emphasized, however, that the “primary” activation or inhibition is accompanied by activation and inhibition of each of the other four muscles in each eye (secondary and tertiary activation) (Cohen et al. 1964a, 1964b) and that every movement is made by all six eye muscles.

Visual information that reaches the vestibular system directly through the subcortical visual system (i.e., through the accessory optic system) is also coded in canal coordinates (Simpson 1984; Simpson et al. 1988). Activity related to movement of the visual fields in the plane of the lateral canals is coded in the nucleus of the optic tract and the dorsal terminal nuclei, whereas activity related to vertical and oblique movement of the visual field is coded in or close to canal planes in the medial and lateral terminal nuclei (Hoffmann and Schoppmann 1975; Mustari and Fuchs 1990; Schiff et al. 1988, 1990; Simpson 1984; Simpson et al. 1988; Yakushin et al. 2000a, 2000b). The synchrony between visual and vestibular inputs and oculomotor output in canal and/or eye muscle coordinates is also extensively coded in the vestibular nuclei and vestibulocerebellum (Cohen and Henn 1988). The fact that this canal-based organization is so widely represented throughout the brain stem and cerebellum and in the anatomic organization of the eye muscles is an indication of the importance of vestibular stabilization of gaze for the functioning of the organism. Information about eye and head movement and position in other spatial frames must also be embedded in central structures, however. It is now known that the central vestibular system codes vestibular information in three-dimensional head and body coordinates as well as in spatial coordinates, which are realized by convergence of otolith- and canal-based signals (Yakushin et al. 2017). This information can also come from the visual system and from graviceptor sensors such as the otolith organs as well as from somatosensory receptors (Jian et al. 2002; Solomon and Cohen 1992a, 1992b; Yates et al. 2000). Thus, although much has been learned about vestibular organization over the past 50 years, a considerable amount of that information is based on fundamental work done on semicircular canal organization in the Cohen laboratory.

NYSTAGMUS DURING VESTIBULAR AND OPTOKINETIC STIMULATION

My first interaction with Bernard Cohen was in 1972 to study nystagmus, which is an ocular repetitive oscillation comprised of slow and quick phases. The clinical literature at that time was focused on the quick phases of nystagmus, and the reflex was used to determine the integrity of the brain stem (Bender et al. 1965; Cohen and Bender 1966; Cohen and Henn 1972a). Bender and Shanzer (Bender et al. 1965; Bender and Shanzer 1964; Shanzer and Bender 1959; Shanzer et al. 1964) showed that lesions of the pontine tegmentum in monkeys caused paralysis of ipsilateral conjugate gaze. It had been also known from the work of Lorente de No (Lorente de No 1933) that nystagmus was affected by pontine reticular formation (PRF) lesions, and he postulated that quick phases of nystagmus are produced in this region.

At that time, Dr. Cohen was focused on elucidating the role of the PRF in generating horizontal saccades, using lesion, stimulation, and single-cell recording techniques. Cohen and colleagues (Cohen et al. 1966a; Cohen and Komatsuzaki 1972) found that all ipsilateral eye movements, including horizontal saccadic and quick phases of nystagmus, were lost after PRF lesions. In further studies by Cohen and Komatsuzaki (Cohen et al. 1966a), positions of fixation and all types of horizontal eye movements were induced by stimulation of a functionally specific area of the PRF known as the paramedian zone of the pontine reticular formation (PPRF), close to the midline of the brain stem. When this area was stimulated, the eyes moved in the horizontal plane to the ipsilateral side. The eye movements induced by PPRF stimulation were conjugate, of approximately constant velocity, and similar in amplitude and velocity in both eyes. Eye movements with velocities in the range of saccades were also induced by PPRF stimulation.

Cohen and Henn (Cohen and Henn 1972b) also recorded single-unit activity in the PPRF. They characterized various unit types in this region that were shown to be importantly related to the generation of saccades (Cohen and Henn 1972a, 1972b; Henn and Cohen 1973, 1976). One class of units, called the medium- or short-lead burst units, had bursts of activity whose instantaneous frequency rose rapidly to ~1,000 spikes/s and preceded the quick phase or saccadic eye movement by ~15 ms (Henn and Cohen 1976). The instantaneous burst frequency declined and was abruptly terminated. There were also long-lead burst units whose instantaneous frequency of firing rose gradually with varying lead times and then had burst activity in synchrony with the medium-lead bursts (Henn and Cohen 1976). There were also pause units that paused their activity in synchrony with the medium-lead bursts (Henn and Cohen 1976). See Fig. 2.2 of Raphan (1976) for an illustration of these unit types that were taken from Henn and Cohen (1976). The stimulation studies together with the single-cell recordings suggested that pulses generated in the PPRF were integrated to produce position changes in the eyes. This conclusion from stimulation studies was in accord with the work of Skavenski and Robinson (1973). These investigators showed that there was an integrator that converted velocity information to position information in the vestibuloocular reflex (VOR) to drive abducens motoneurons, which innervate the lateral rectus muscle. The possibility that this “final integrator” would be common to both the VOR and the saccadic system was intriguing and led to extracellular studies of this region by a number of investigators to determine the nature of the neural coding in the PPRF (Henn and Cohen 1976; Keller 1974).

As described above, the various types of units recorded by Henn and Cohen (1976) have different types of dynamic characteristics. One class of long-lead burst units had a slow buildup in frequency of activity that reached a peak and reduced its activity and then was cut off. Another type of long-lead burst unit rose more quickly to a peak frequency, and then the frequency decayed slowly. A third class of units called medium-lead burst units instantaneously rose to high frequency, and then the frequency declined almost linearly. I worked with Dr. Cohen formulating a model of how these units could produce the quick phases of nystagmus (Raphan 1976). In this model, it was assumed that the system generating the quick phases or saccades was a second-order state-determined system, which could be activated from the vestibular and optokinetic systems to induce oscillations in the output when driven by a constant-velocity optokinetic or vestibular input. Because the state-space approach to modeling systems identifies specific states that determine the behavior of the system, this approach was a “microscopic” approach rather than the “macroscopic” or input-output approach that was prevalent at the time. The question was whether a realization of the model system could be identified that would not only predict the quick phases but also predict the different classes of units as states of the system. It was also assumed that the input to the quick phase generator was a neural representation of a constant-velocity signal coming from the vestibular system and optokinetic systems. Choice of state variables for a system is not unique, and the particular realization that best fit the frequency of firing of the unit data gave us clues regarding the structure of the saccade generator. The realization that was chosen was a controllable form realization with nonlinear feedback, which transformed the system from being overdamped to underdamped when a quick phase was triggered and generated relaxation oscillations (van Der Pol 1926). By adaptively training the system (presently referred to as machine learning), parameters were learned so that activation from the vestibular and visual systems could generate nystagmus as relaxation oscillations. It also identified state variables that predicted the classes of units found in the PPRF (Raphan 1976). It is of interest that the concept of a relaxation oscillator became important in our subsequent studies of locomotion that generate stance and swing phases (Osaki et al. 2007, 2008) and vestibulo-autonomic and vasovagal responses, which has given insight into neurogenic syncope (Raphan et al. 2016).

As we were studying vestibular and optokinetic nystagmus, a fundamental question arose as to how the vestibular system activated the quick phase generator. This consideration led into a study of the central organization of vestibular system processing and velocity storage. This work with Bernard Cohen was expansive. It led to a fundamental understanding and modeling of the three-dimensional compensatory and orienting properties of the vestibular system behaviorally and the central neuronal implementation (Cohen and Raphan 2004; Raphan and Cohen 2002; Yakushin et al. 2017).

CENTRAL PROCESSING OF VESTIBULAR SYSTEM SIGNALS (VELOCITY STORAGE)

Early work on the vestibular system assumed that the temporal processing of the head acceleration signal by the semicircular canals was governed by a second-order system with the long time constant of vestibular nystagmus (Steinhausen 1933), implying that additional central processing was simply relayed to the oculomotor system. The view that the main processing of vestibular signals was done in the semicircular canal dynamics was supported by the generation of short-latency eye movements (Baker et al. 1969; Cohen et al. 1964a, 1965; Krejcova et al. 1971). Eye movements that compensated for head movements of short duration are generally smooth and are not often interrupted by saccades. The compensatory nystagmus generated by head rotations of longer duration are interrupted by saccades but were known to give information about sensory central processing of movement signals. Over the period from 1976 to the present, Bernie Cohen and I studied a central mechanism that we referred to as velocity storage and uncovered the extent to which this mechanism governs the compensatory and orienting properties of the vestibular system (Cohen et al. 1977; Cohen and Raphan 2004; Raphan et al. 1979; Raphan and Cohen 2002).

A major impetus for the in-depth studies of central vestibular system processing was that only angular acceleration is sensed by the receptors in the semicircular canals while the motoneurons code position of the eyes. It was therefore concluded that two integrations are necessary to convert head acceleration to eye position (Robinson 1977; Skavenski and Robinson 1973). One of these integrations is done mechanically at the hair cells and cupula of the semicircular canals so that the afferent activity induced in the canal afferents in the vestibular nerve closely follows the angular velocity of the head (Goldberg et al. 1987). Thus, for frequencies of head movement between 0.05 and 8 Hz, which is the normal physiological range of head movements, activity in the vestibular nerve from the semicircular canals is predominantly related to angular head velocity, not to head position or to head acceleration. A signal related to head velocity is also the major information used by the canal-recipient portions of the central vestibular system to produce compensatory eye movements over the angular VOR (aVOR) and head movements over the angular vestibulocollic reflex (aVCR). At lower frequencies of head rotation (<0.05 Hz), the phase of the afferent firing rate is shifted closer to angular acceleration and eye velocities induced over the aVOR are greatly diminished.

A second integration was therefore inferred for the aVOR to function properly (Skavenski and Robinson 1973). This integration converts the neural signal related to angular velocity of the head, observed in the vestibular nuclei (Waespe and Henn 1977a, 1978), to the signal observed in the motoneurons, which is closely related to eye position (Skavenski and Robinson 1973). Because of the overdamped nature of the oculomotor plant (i.e., the eye and surrounding tissue), direct activation of the eye muscles is necessary to reduce delays and maintain the dynamics of the aVOR, saccades, and smooth eye movements (Robinson 1981). The time constant of the velocity-position integrator is ~30 s, but its value is under cerebellar control (Robinson 1974, 1976). In addition to being an important component of the aVOR, velocity-position integration is required by a number of subsystems related to oculomotor control, such as the saccadic and smooth pursuit systems, and is common to each of these oculomotor subsystems (Robinson 1964, 1965, 1968, 1971, 1975). There was considerable work and controversy about how the velocity-position integrator is organized in three dimensions (Raphan 1998; Tweed and Vilis 1987), but these three-dimensional approaches are beyond the scope of this review. For a more complete treatment of this issue, see Raphan (1998) and Raphan and Cohen (2002).

ADDITIONAL INTEGRATION REQUIRED FOR VOR FUNCTIONING

During rotation in darkness for short periods (3–10 s) eye velocity compensates well for head velocity, and when the rotation is stopped there is little or no after-response. If the head is rotated for a longer period, however, the eye velocity of the induced nystagmus declines steadily to zero, and at the cessation of rotation there is a substantial after-response in the opposite direction. During constant-velocity rotation, the hair cells are initially deflected but soon return to their resting position. The return to the resting position and the decline in activity in the vestibular nerve have been modeled by a second-order system (Steinhausen 1933), although the dominant time constant was not established. Recordings in the nerve established that the time constant in monkey was ~3.5–5 s (Goldberg and Fernandez 1971a, 1971b).The fact that there was good ocular compensation for up to 10 s after the onset of rotation indicates that a central mechanism had prolonged the canal response. This central mechanism has been termed “velocity storage” because it holds or stores activity from the receptor and discharges it over a longer time course (Raphan et al. 1979). One functional consequence of this central mechanism is to enhance compensation for lower frequencies of head movement than would occur from cupula deflection alone. With data modeling, it was established that the time constant of the canal in humans was also ~4 s (Dai et al. 1999).

Other insights about velocity storage arose from studies of the compensatory eye velocity responses during optokinetic stimulation of a rotating textured visual surround and visual-vestibular interaction during long-duration rotation in light (Cohen et al. 1977; Raphan et al. 1979). Long durations of continuous surround motion at a constant velocity generate optokinetic nystagmus (OKN). Eye velocity initially jumps quickly to ~0.6 of the velocity of the visual surround and then climbs toward the actual velocity of the visual surround to generate compensatory eye velocity throughout rotation (Cohen et al. 1977). In humans, the initial jump in velocity is larger, bringing eye velocity close to that of the visual surround immediately (Cohen and Raphan 1981). An important aspect of the OKN response is that the visual input activates the velocity storage network (Cohen et al. 1977). When there is rotation in light, the dynamics of the responses were shown to be governed by the same velocity storage mechanism and explained the process of optokinetic after-nystagmus when lights were extinguished. Thus velocity storage helped maintain the response to rotation in light, when the aVOR was no longer driving the system, and canceled any after responses when rotation was stopped, which would have produced a postrotatory response (Cohen et al. 1977; Raphan et al. 1979) to cancel vestibular after-responses (Cohen 1974; Raphan et al. 1979; Ter Braak 1936). The demonstration that activation of optokinetic stimulation and input from the semicircular canals are combined in a common velocity storage neural circuit was an important contribution to our understanding of visual-vestibular interaction. These studies, however, gave an incomplete picture of the overall organization of velocity storage in the vestibular system, since the VOR is organized to sense and respond to movement and orientation in three dimensions. Because of inherent complexities of describing combined rotations and translations in three dimensions, questions regarding the function of the VOR can most effectively be addressed by model-based approaches, which Bernie Cohen and I collaborated on for 40 years. One important outcome was the demonstration that velocity storage was tied to spatial orientation and to otolith function (Dai et al. 1991; Raphan and Sturm 1991). This led to not only a better understanding of VOR spatial orientation but also identification of syndromes associated with conflicting motion orientation environments (Dai et al. 2009) and motion sickness (Dai et al. 2003, 2007, 2011, 2014, 2017) and how cross-axis activation could be adapted to alleviate these clinical syndromes. It also led to studies of how the absence of gravity affects spatial orientation (Dai et al. 1994, 1998). In addition, this led to an understanding of how the neurons in the vestibular nuclei are organized to implement this three-dimensional structure (Yakushin et al. 2017).

VELOCITY STORAGE IN THREE DIMENSIONS AND ITS CENTRAL VESTIBULAR CORRELATES

Early studies of the VOR were done in one or two dimensions, and eye position was represented by horizontal and vertical displacements relative to the subject. Even when three-dimensional eye movement recording techniques became available (Robinson 1963), studies were generally restricted to two dimensions because the schemes for representing and understanding the eye movements could most easily be visualized in two dimensions. A major breakthrough in studying the VOR in three dimensions over the last 10–15 years was that improved formalisms for representing eye orientation and eye velocity in three dimensions were developed (Crawford and Vilis 1991; Raphan 1998; Schnabolk and Raphan 1994; Thurtell et al. 2000; Tweed and Vilis 1987, 1990). Analysis of eye movements based on these representations has shown that aVOR compensation for head movements is more complex than previously thought. In both humans and animals, eye velocity axes during head movement are not fixed but are rather dependent on the position of the eyes in the orbit (Misslisch et al. 1994; Thurtell et al. 1999, 2000). Moreover, the gain of the roll component is substantially less than that of the horizontal and vertical components (≈0.6 vs. 1.0, respectively) (Crawford et al. 1991; Tweed et al. 1994a, 1994b). As a consequence of the kinematic constraints and lower roll gains, the head and eyes do not generally rotate around the same axes. How ocular compensation is coded centrally, given these constraints, is still somewhat controversial (Crawford et al. 1991; Raphan 1998; Smith and Crawford 1998; Thurtell et al. 2000). During the collaboration in elucidating velocity storage, we uncovered organizational principles of the vestibular system in three dimensions, which focused new studies. It also consolidated and modeled our understanding of semicircular canal- and otolith-induced behavior of the VOR, culminating in a recent study relating this model to the neural organization in the vestibular nuclei (Yakushin et al. 2017).

The model-based study also predicted the effects of plugging by simply modifying the time constant of the plugged canals (Yakushin et al. 1998). Ramps of angular acceleration demonstrated that the cupula time constant fell from ≈4 s to 70 ms (Yakushin et al. 1998). Since the time constant is inversely proportional to the high-frequency response of the canal, the three-dimensional model predicted the spatial and dynamical properties of a wide range of canal-plugged conditions and frequencies of rotation. Angelaki and Hess (Angelaki and Hess 1996a, 1996b, 1996c) initially concluded that the high-frequency aVOR response of animals with plugged canals indicated that the central vestibular system adapted spatially for the loss of canal input after canal plugging. There is now a general consensus from studies in the toadfish (Rabbitt et al. 1999, 2001), squirrel monkey (Lasker et al. 1999), and rhesus and cynomolgus monkeys (Hess et al. 2000; Yakushin et al. 1995, 1998), however, that there is no spatial adaptation following plugging. Although the visual system complements the responses of the lost planes by shifting the phase of the spatial gain curves back to normal after plugging (Yakushin et al. 1995), the aVOR in dark does not spatially adapt. Thus sensory loss of a plane by plugging abolishes the response in that plane but leaves the remaining coordinate axes of the canals intact in the processing of aVOR signals (Yakushin et al. 1995, 1998). The implication of this conclusion is that the ability of the aVOR to adapt spatially in response to visual motion orthogonal to head rotation, i.e., cross-axis adaptation (Baker et al. 1987; Schultheis and Robinson 1981), may require intact semicircular canals and afferent input from appropriate canals.

To better understand the VOR in three dimensions, it was necessary to better elucidate how the angular VOR (aVOR) interacts with the linear VOR (lVOR) (Snyder and King 1992; Telford et al. 1998; Viirre et al. 1986; Viirre and Demer 1996) in three dimensions, since rotations of the head reoriented the head relative to gravity, which activates the otoliths. Early work had shown that the lVOR responded to the net linear acceleration of the head by activation of the otolith organs the utricle and saccule. On Earth, the imposed linear accelerations summate with the equivalent acceleration of gravity to form gravito-inertial acceleration (GIA). The lVOR produces essentially two modes of response to changes in the GIA. We have defined these modes as orienting and compensatory (Imai et al. 2001a; Raphan et al. 1996; Raphan and Cohen 1996), but they have also been termed tilt and translation (Angelaki et al. 2001; Mayne 1974; Paige and Seidman 1999; Paige and Tomko 1991a, 1991b). In the compensatory mode, the lVOR rotates the eyes to maintain fixation on a particular locus or point in space. Stabilization of a fixation point implies that only the foveal image of the point is stabilized and not the entire visual field, since the eyes rotate in the head and do not translate. Thus the lVOR compensatory response depends critically on viewing distance and accommodation (Paige 1991; Paige and Seidman 1999; Paige and Tomko 1991a, 1991b; Schwarz et al. 1989; Schwarz and Miles 1991; Skipper and Barnes 1989) and accounts for the dramatic changes in gain of the lVOR as viewing distance is changed. Because these gain changes function to maintain the fovea on target, this may explain why the compensatory lVOR is particularly prominent in primates.

The lVOR can also change the orientation of the eyes in space (Schoene 1964; Schöne 1984). This component of the lVOR responds to tilts of the GIA relative to the head, by generating ocular tilts that align the eye vertical with the GIA (Angelaki 1991; Arrott and Young 1986; Benjamins 1918; Clément et al. 2001; Curthoys and Halmagyi 1992; MacDougall et al. 1999; Magnus 1924, 1927; Merfeld 1995; Merfeld and Young 1992; Miller and Graybiel 1970; Moore et al. 2001; Telford et al. 1998, 1997; Van Der Houve and De Kleijn 1918; Wearne et al. 1999; Woellner and Graybiel 1959). As a result, the yaw (vertical) axis of the eyes tends to align with the GIA while upright, during sustained head tilts (Benjamins 1918; Magnus 1924, 1927), and during low-frequency linear accelerations on a sled (Angelaki and Anderson 1991a, 1991b; Arrott and Young 1986; Telford et al. 1996, 1997, 1998) or a centrifuge (Clément et al. 2001; Curthoys and Halmagyi 1992; Graybiel 1975; MacDougall et al. 1999; Merfeld 1995; Merfeld and Young 1992; Moore et al. 2001; Wearne et al. 1999).

Orientation information arising in the otoliths also interacts with the aVOR centrally through velocity storage. This orientation effect on the aVOR rotates the eye velocity vector toward the GIA when there is misalignment between the GIA and the axis of rotation (Dai et al. 1991; Raphan et al. 1996; Raphan and Sturm 1991; Wearne et al. 1999). A major conclusion from this work was that while orientation information in the central vestibular system is governed by a coordinate frame related to the semicircular canals, which is widely represented in the central vestibular system, the vestibulocerebellum, and the subcortical visual systems that support vestibular function and visual-vestibular interactions (Cohen and Henn 1988; Simpson 1984; Simpson et al. 1979; Yakushin et al. 1995, 1998), an additional spatial coordinate frame is imposed on the canal frame through velocity storage (Dai et al. 1991; Raphan and Sturm 1991). This change in coordinate frames is interesting because unlike the afferent canal signals, which are essentially confined to three spatial planes, otolith afferents have polarization vectors coding equivalent linear head acceleration over the entire spatial span (see Wilson and Melvill Jones 1979 for a complete description). In addition, no specific coordinate frame has been identified in which otolith afferent information is processed centrally, although studies of otolith-ocular reflexes have generally considered how ocular compensation is generated to maintain stable gaze and how the central vestibular system codes tilts of the body and head with regard to gravity. In this central frame encoded by orientation vectors associated with velocity storage, a combination of the direction of the GIA and an internal estimate of the body vertical rather than the head yaw axis is taken as the upright. The dependence of this spatial frame on velocity storage is shown by the fact that when velocity storage is abolished by midline section spatial orientation of the aVOR is lost (Wearne et al. 1997).

Spatial orientation has also been shown to influence adaptation of the aVOR (Yakushin et al. 2000c, 2000d, 2000e). The aVOR gain can be adapted for a specific head orientation with regard to gravity. The adapted gain is expressed only when the head is in the position in which the aVOR gain was adapted, but the gain of the aVOR is unchanged in other head positions. This behavior points to the close association that must exist between otolithic and canal-based coordinate frames. Although it is likely that this gravity-specific gain adaptation involves otolith-canal convergence in the vestibular nuclei, the neural basis for this adaptive behavior and the role of velocity storage in this orientation specific adaptation is unknown.

The work I did with Bernard Cohen and colleagues in the laboratory on elucidating the orientation of velocity storage in three dimensions was fundamental in the identification of the coordinate frames that the central vestibular system utilizes to process information about head and eye movements and spanned the 40 years of our collaboration. It impacted understanding of motion sickness (Dai et al. 2010, 2011, 2014), which is presently being utilized in the treatment of clinical syndromes such as mal de debarquement syndrome (MdDS) that may be related to orientation disparities (Dai et al. 2014, 2017).

NEURAL BASIS OF VELOCITY STORAGE

Despite the considerable amount of work done in elucidating the behavioral characteristic of the three-dimensional structure of velocity storage, it is only recently that it was related to neurons in the vestibular nuclei (Yakushin et al. 2017). This was the culmination of studies on velocity storage coming from the Cohen laboratory, and the description here is essentially taken from that work. In one dimension, the long dominant time constant of the aVOR had been shown to be reflected in the neural activity of all central vestibular neurons. and no units reflect the shorter peripheral time constant in their firing rates (Waespe et al. 1977; Waespe and Henn 1977a, 1977b, 1978, 1979). This indicates that the activity related to velocity storage is widely distributed in the vestibular nuclei and has a strong effect on processing. The reason for this may be that the body must know how fast it is moving, and the characteristics of the cupula and hair cells in the semicircular canals can only supply the high-frequency component. Therefore, other systems, such as the visual and somatosensory systems as well as the otolith organs, must have access to a mechanism that maintains and extends the sense of movement (Bles and Kotaka 1986; Solomon and Cohen 1992a, 1992b).

“Vestibular-only” (VO) and vestibular-pause-saccade (VPS) neurons in the rostral and medial vestibular nuclei have the appropriate characteristics to be responsible for its generation (Reisine and Raphan 1992a, 1992b). The first studies on the neural basis of velocity storage had shown the relationship of VO and VPS neurons to the time constant of velocity storage induced by angular rotation and OKN (Waespe and Henn 1977a, 1977b). It was further demonstrated that velocity storage is implemented by cross-connections between the rostral and medial vestibular nuclei on the left and right sides (Katz et al. 1989; Wearne et al. 1997). The cross-connections are GABAergic (Holstein et al. 1999a, 1999b, 1999c) and are under control of the cerebellar nodulus, which is critical for controlling spatial orientation of velocity storage as well as its temporal properties (Sheliga et al. 1999; Solomon and Cohen 1985; Waespe et al. 1985). It is likely that VO neurons generate velocity storage, whereas VPS neurons are located in the path from the velocity storage integrator to the oculomotor plant, since their activity ceases during periods of drowsiness (Reisine and Raphan 1992b). These studies of velocity storage suggested that VO and VPS neurons should code all aspects of velocity storage including spatial orientation and habituation (see Cohen et al. 2008; Cohen and Raphan 2004; Raphan and Cohen 2002 for review). However, because of technical limitations in recording of single neurons in three dimensions, a variety of tests that explore various aspects of velocity storage could not be done, limiting the understanding of how velocity storage is realized in three dimensions. It was only recently that this was made possible (Yakushin et al. 2017) and long-term recordings of the VPS and VO neurons provided information that helped show the relationship of neural recordings to the three-dimensional structure of velocity storage.

The study demonstrated that VO and VPS neurons are not a uniform group of type I and type II neurons, consistent with earlier findings (Reisine and Raphan 1992a). Rather, the neurons recorded in this study consisted of a variety of neural subgroups that implement the long time constant as well as the neurons that likely implement the direct vestibular pathway that goes around velocity storage. The neurons that are involved in the realization of velocity storage probably project across the midline implementing a bilateral integrator, similar to that proposed for the velocity-position integrator (Anastasio 1991; Galiana et al. 1984). Neurons that respond to rotation only in one direction also code velocity storage only for that direction. The bilateral model of velocity storage could be formed as two unilateral integrators connected to each other with cross-projections. The implementation of the velocity storage integrator in this fashion would be consistent with the abolition of velocity storage when the midline is cut (Katz et al. 1989; Wearne et al. 1997). This would also explain why the time constants of rotations in opposite directions are almost never identical. It is also probable that the nodulus projects to these neurons (Meng et al. 2014), carrying otolith-related information (Meng et al. 2015) to implement the orientation properties of these velocity storage-related neurons (Sheliga et al. 1999; Waespe et al. 1985; Wearne et al. 1996). We also demonstrated that some neurons coded velocity storage in head or canal coordinates. The firing rates of such neurons correlated with the horizontal or the vertical components of the VOR in one or both directions, but they were not related to the cross-coupled components of velocity storage. Such neurons could be involved in implementing the long time constant of the aVOR but probably would not receive otolith information from the nodulus to be involved in spatial orientation of velocity storage. Moreover, this indicates that velocity storage may be implemented in canal coordinates and transformed to head coordinates within the vestibular nuclei. Thus this study has demonstrated that the neural components in the vestibular nuclei not only have capability to implement the three-dimensional properties of velocity but also are part of the direct pathway around it as predicted by the model.

There are other direct links of our model to neural activity in the central vestibular system in a number of studies. We have demonstrated that otolith polarization vectors of VO neurons can be adapted and that the polarization vectors of VO neurons adapt toward the axis of gravity if animals are positioned away from spatial vertical (Eron et al. 2008, 2009). Interestingly, eye-head-velocity (EHV) and position-vestibular-pause (PVP) neurons as well as central otolith neurons have a very limited adaptive capability of their polarization vectors (Eron et al. 2009). The fact that we related unit activity recorded in the central vestibular system to the mechanistic three-dimensional model of velocity storage presented means that we can reject the notion that velocity storage is an outcome of noise in differentiating tilt from translation computations (Laurens and Angelaki 2011). Moreover, we do not see how the latter notion can be reflected in the central VO neurons that we have studied (Yakushin et al. 2017). In contrast, the head vertical and its relationship to the spatial vertical are features of the Raphan–Cohen model, are reflected in the eigenvalues and eigenvectors of the system matrix, and are a function of differential otolith activation (Dai et al. 1991; Raphan and Sturm 1991).

CLINICAL IMPLICATIONS OF VELOCITY STORAGE

Clinically, the time constant of vestibular nystagmus is an important measure when evaluating vestibular abnormalities (Bárány 1906a, 1906b; Brandt 1998). Typically, the time constant of velocity storage is reduced with unilateral and bilateral vestibular lesions. It can also be habituated by repeated rotations (Baloh et al. 1982; Cohen and Cohen 1989). Such reduction reduces the susceptibility to motion sickness (Bos et al. 2002; Cohen et al. 2008; Dai et al. 2011). This shows that velocity storage is not only critical for spatial orientation with regard to gravity but also serves as an input to the sympathetic system and that motion sickness susceptibility can be reduced by shortening the VOR (velocity storage) time constant (Dai et al. 2010). Studies in monkeys also demonstrate that prolonged oscillation in roll while rotating about a spatial vertical axis induces oscillatory modulations of nystagmus (Dai et al. 2009) similar to those in patients with mal de debarquement syndrome (MdDS) (Dai et al. 2014, 2017). Thus changes in velocity storage are postulated to be responsible for the postural instability induced by prolonged travel on water (Dai et al. 2014). Of particular significance is the fact that MdDS impacts the body postural system, resulting in rocking, or swaying at 0.2 Hz, showing that the velocity storage integrator is associated not only with spatial orientation, eye movements, and activation of the sympathetic system but also with descending vestibulo-spinal projections that are associated with strong postural instability during MdDS. Thus further shortening of the yaw time constant or adapting its orientation clearly reduces susceptibility to motion sickness in that rolling the head at the frequency of rocking/swaying in MdDS while velocity storage is activated by OKN significantly reduces MdDS symptoms (Dai et al. 2017).

MECHANISMS OF LOCOMOTION

Our studies of locomotion began when Bernie Cohen did his studies on nystagmus generated by circular monkey locomotion (Solomon and Cohen 1992a, 1992b). These studies showed that after monkeys were trained to walk on a circular treadmill there was compensatory horizontal eye velocity that maintained horizontal gaze velocity (head + eye velocity) in space interspersed with gaze shifts in the direction of motion as the animals moved or nystagmus (Solomon and Cohen 1992a, 1992b). A key point of this work was that this nystagmus was continuous in darkness as well as in light. Rotation with the head fixed resulted in a decline of the nystagmus and suggested for the first time that the active movement of the limbs through somatosensory input to velocity storage had maintained the compensatory gaze velocity.

Following these studies, Dr. Cohen and I collaborated on a more in-depth study of animal and human locomotion. Because locomotion is associated with coordinated limb, body, head, and ocular movements (Crane and Demer 1997; Hirasaki et al. 1999; Inman 1966; Inman et al. 1981; Maurer et al. 1997; Mergner and Rosemeier 1998; Moore et al. 1999; Winter et al. 1993), at that time quantitative results about the body, head, and eyes during walking had largely come from two-dimensional studies on linear overground and treadmill locomotion. These studies had shown that the body, head, and eyes rotate in response to up-down and side-to-side motion to maintain stable head pointing and gaze in space (Crane and Demer 1997; Hirasaki et al. 1999; Moore et al. 1999; Pozzo et al. 1990). Grasso et al. (Grasso et al. 1996, 1998) studied horizontal head and eye movements during circular walking. They showed that there was anticipatory control of the horizontal head direction so that the head led the trajectory motion in the horizontal plane by up to 20°. There was also horizontal ocular nystagmus with quick phases that moved the eyes and gaze ahead of the body heading, steering the turn (Grasso et al. 1998). Relatively little was known about the underlying mechanisms that coordinated movements of the upper body, head, and eyes in three dimensions during either straight walking or turning. One reason for the lack of information is that there are complex relative rotations and translations between the body, head, and eyes that were not easily related to each other or to the gait trajectory. This complicated both the representational schemes for describing the kinematics of the relative motions of the body, head, and eyes and their quantification. We therefore developed a three-dimensional representational scheme (Imai et al. 2001a; Osaki et al. 2007, 2008) for describing motion of the body in space and clarified many aspects of compensatory head and eye movements during straight walking on a treadmill as well as turning (Imai et al. 2001b).

During straight walking, there is a characteristic monotonic relationship between stride length and walking speed for velocities ranging from 1.2 to 1.8 m/s (Andriacchi et al. 1977; Cappozzo 1981; Hirasaki et al. 1999; Murray et al. 1966). Over this range of walking velocities, there is both translation and rotation of the head in the sagittal plane, reaching translational frequencies close to 2 Hz, peak vertical linear accelerations of 0.3–0.5 g, and peak pitch velocities of 15–22°/s (Hirasaki et al. 1999; Moore et al. 1999). The magnitude and frequency of the pitch angular and vertical translational head movements are sufficient to activate both the angular and linear vestibuloocular (aVOR and lVOR) and vestibulocollic (aVCR and lVCR) reflexes (Paige 1991; Paige and Tomko 1991a, 1991b). In support of this, unilateral vestibular disease is associated with instability of gait (Grossman and Leigh 1990; Ito et al. 1995), and astronauts, whose vestibular function has been adapted to microgravity, also experience difficulty when walking a curved path after flight (Bloomberg et al. 1997).

The concepts of compensatory and orienting responses developed for the VOR with the head fixed were therefore generalized to describe movements of the head and eyes in three dimensions during straight locomotion and turning (Imai et al. 2001b). The compensatory aVOR had been inferred to stabilize horizontal and vertical fixation of far targets during straight walking (Moore et al. 1999) and to compensate for yaw head movements during circular locomotion (Grasso et al. 1998; Solomon and Cohen 1992a, 1992b). The aVCR appears to compensate for pitch body rotation when walking at low velocities (Hirasaki et al. 1999). Compensatory lVCR/lVOR responses would be induced by linear translations that rotate the head and/or eyes to maintain an invariant point relative to the subject’s forward motion. Compensatory responses have been observed in vertical head and eye movements that stabilize the head and gaze during linear treadmill locomotion at optimal walking velocities. These head movements maintain an approximate fixed point at the intersection of all head pointing directions in the sagittal plane over the gait cycle (Hirasaki et al. 1999; Moore et al. 1999; Pozzo et al. 1990). An invariant point of head fixation is presumably present during overground walking but had not been systematically studied. Orienting responses during locomotion would tend to align the yaw axes of the head and eyes, which are normally along the spatial vertical when standing, with the tilted GIA. During turning, there are significant linear accelerations that produce dynamic tilts of the GIA that induce orienting head, eye, and body movements (Imai et al. 2001a, 2001b).

During the course of studying locomotion, we presented a novel representational scheme for describing the linear and angular motions of the body, head, and eyes to analyze roll, pitch, and yaw of the body, head, and eyes relative to any arbitrary coordinate frame, which translates and rotates in space. The scheme was based on the theory of finite rotations and the representation of these rotations by axis-angle (Altman 1986; Raphan 1998; Rodrigues 1840). Utilizing this, we determined how the body, head, and eyes compensate and orient to variations in the GIA during the locomotion trajectory to maintain stable posture and gaze in three-dimensional space.

The main findings of these studies were that yaw, pitch, and roll movements of the body, head, and eyes maintained gaze in the direction of forward motion during straight walking and direct gaze in advance of the heading during turning. Grasso et al. (Grasso et al. 1996, 1998) showed that turns are anticipated by directing gaze about a yaw axis. Our results extend these findings and demonstrate that changes in the GIA are anticipated by tilting the head about the pitch and roll axes relative to heading. The results also showed that the eyes are driven by tilts in the GIA during turning to help point gaze in the direction of turning.

Although the characterization of body, head, and eye movements was important in defining the role of the vestibular system in maintaining forward gaze while walking and turning, it became clear that the motion of the foot in three dimensions was important for understanding and modeling a critical interface of the body and the supporting surface during walking and how these foot movements were propagated through the body, head, and eyes so that neural mechanisms could be identified. Although there were numerous studies that defined how the foot moved between stance and swing phases, there was no coherent framework at the time for modeling foot rotation and translation as a function of walking velocity (Osaki et al. 2007). It had been noted that there is a striking similarity between the stance and swing phases of locomotion in the forward (x-axis) direction of monkeys and the slow and rapid phases of nystagmus when similar analyses of the dynamics were applied to locomotion data (Xiang et al. 2007). Stimulus velocity is a critical input to produce both the slow and rapid components of nystagmus (Cohen et al. 1977; Raphan et al. 1979). We therefore utilized dynamical system analysis techniques and the analogy with oculomotor behavior to develop a mathematical model that predicted the timing and phase plane trajectories of the gait cycle and their relationship to walking velocity for the forward motion of the foot during locomotion. By adapting methodologies similar to those used in analyzing oculomotor system behavior, i.e., phase plane and main sequence plots (Bahill et al. 1975; Stark 1971; Stark et al. 1980), we determined that there was an active feedback control law that determined the relationships between walking velocity, stride length, and frequency of stepping (Hirasaki et al. 1999; Winter 1983) as well as the dynamic motor-related responses.

This was the beginnings of the development of a coherent mathematical framework on which to model the dynamics of the stance and swing phases (Osaki et al. 2007, 2008). When such an analysis was done, it led us back to modeling the stance and swing phases as a relaxation oscillator as was done for oculomotor nystagmus (Raphan 1976). The analysis of the phase plane trajectories gave important information about how the swing phases were organized. The concentric and circular properties of the trajectories support the idea that the dynamics of the forward motion of the foot during the swing phases are undamped. The model suggests that this undamped mode is under feedback control of position and that the model’s dynamics are governed by the desired walking velocity and stride frequency. As the model predicts, this control would appropriately position the foot at heel contact. Peak velocity along the forward motion of the foot was linearly related to walking velocity, which is the predominant driver during the stance phase. We infer that walking velocity also determines the dynamics and the underlying time course of the swing movements dependent on the state of the foot at the initiation of the swing phase. There was no similar dependence of the roll and vertical movements of the toe on walking velocity. We speculated that the goals of toe movement in the roll and vertical directions have other functions, such as to maintain stability (Donelan et al. 2004; MacKinnon and Winter 1993).

The character of the trajectory and its relation to dynamical systems also gave insight into how foot orientation at the end of the stance and beginning of the swing contributes to the phase plane trajectory. Although there are undoubtedly stochastic aspects to the foot trajectory, locomotion can be viewed as a dynamical system, i.e., that the trajectory and terminal state of the system are determined by its initial state, its state transition function, and its input (Zadeh and Desoer 1963). The initial state of the swing phase was the foot orientation at toe-off, which included toe position, the pitch angle of the foot, and the consequent heel elevation at toe-off. We postulate that this initial state, together with the feedback control law, determined the phase plane trajectory at any given walking velocity, thereby allowing the swing phases to be goal-directed to match increments in the distance covered during the stance phase. Foot orientation was also an initial state determinant that influenced the lateral, upward, and three-dimensional rotation of the foot during the swing phase and determined how the foot would be oriented at heel contact. Such precise control of toe elevation at heel contact and heel elevation at toe-off had been postulated to increase the efficiency of locomotion by extending the effective length of the leg (Della Croce et al. 2001; Inman et al. 1981; Saunders et al. 1953). It should be noted that whereas each phase plane is unique for a given walking velocity, the stance and swing phases are separated by a switch, which is nonlinear, so walking velocity and the state at toe-off determine the approximately circular trajectory of the swing phase. Because of the nonlinear switching, any point on the swing trajectory cannot completely specify the whole trajectory, which includes the stance phase. Neither can any other point on the stance trajectory determine the whole trajectory. Of interest was that the pitch orientation of the swinging foot was linearly related to forward foot position. The spatial position and alignment of the foot at toe clearance was also independent of the phase plane trajectory and walking velocity. These findings are consistent with the idea that the foot is under end-point control (Grasso et al. 2004; Redfern and Schumann 1994) and are consistent with studies showing that the limb is orientated relative to a spatial reference frame rather than to joint angles for the arm (Soechting and Ross 1984) or leg (Borghese et al. 1996; Grasso et al. 2004). The findings are also consistent with the idea of motor equivalence, which is the control hierarchy used to accomplish a given motor goal when an invariant task goal can be achieved with variable means (Hebb 1949; Lacquaniti 1989; Lashley 1933).

Further work, using this relaxation oscillator model, examined the relative contribution of walking velocity and stepping frequency to the neural control of locomotion (Osaki et al. 2008). This was done by maintaining frequency of stepping with a metronome constant frequency (isofrequency) and varying walking velocity on a treadmill or maintaining walking velocity constant (isovelocity) on the treadmill and varying the frequency of stepping with the metronome. The work showed that in the isovelocity condition peak forward toe velocity during the swing phases was related to walking velocity and did not vary with alterations in stride frequency. In the isofrequency condition, in contrast, stepping frequency altered the relationship between toe acceleration and toe position in the fore–aft direction. The cycle frequency, main sequence (peak velocity vs. amplitude) relationships, and the shape of the phase plane trajectories of the swing phases also reflected this relationship. The data were modeled by decoupling stepping frequency from walking velocity while maintaining active feedback control dependent on frequency. The latter predicted both the dominant shape of the phase plane trajectories and the main sequence relationships. Thus, according to the model, walking velocity and stride frequency are independent central variables that control the dynamics of the swing phases and stepping. The ability to decouple stride frequency from walking velocity may help in navigating over uneven terrain or when executing curved trajectories while maintaining a constant velocity. This approach helped us to better understand the gait dysfunction in patients with advanced Parkinson’s disease (PD) (Cho et al. 2010, 2006).

CLINICAL EVALUATION OF LOCOMOTOR FUNCTION IN PARKINSONIAN GAIT

It had been presumed that the fundamental defect in PD gait was an inability to generate normal stride length. Our data suggested, however, that the basic problem in PD gait is an impaired ability to match step frequency to walking velocity. At low walking velocities, PD stepping had a reduced or absent terminal toe lift, which truncated swing phases, producing shortened steps. Auditory pacing was not able to normalize step frequency at these lower velocities. Peak forward toe velocities increased with walking velocity, and PD subjects could initiate appropriate foot dynamics during initial phases of the swing. They could not control the foot appropriately in terminal phases, however. Increased treadmill velocity, which matched the natural PD step frequency, generated a second toe lift, normalizing step size. Levodopa increased the bandwidth of step frequencies but was not as effective as increases in walking velocity in normalizing gait.

The collaboration on locomotion, which began through a top-down evaluation of how head and eye movements behave during walking and turning, evolved into a bottom-up evaluation of gait about how the foot is controlled during walking and turning. This approach allowed us to model the foot trajectory as a relaxation oscillator, which was the model developed for nystagmus, the starting point of our collaboration. This model allowed us to give new insights into Parkinson gait and how l-DOPA corrects certain aspects of the gait (Cho et al. 2010). It also supported the idea that the dynamics of locomotion is governed by a velocity signal that drives the foot through end-point control governing the movements of the lower limbs. This motion generates bottom-up drive on the body, head, and eyes, which through compensatory mechanisms in the vestibular system maintain gaze along the direction of movement.

VESTIBULAR-AUTONOMIC INTERACTIONS

In early 2005, Dr. Cohen began a project to study how the vestibular system impacts muscle sympathetic activity to control blood pressure (BP) and heart rate (HR) (Kaufmann et al. 2002). The studies were based on evidence that suggested that the vestibular system modulated sympathetic activity in response to changes in posture relative to gravity (Yates et al. 2000). With the use of a miniature microneurography device developed for the NASA Neurolab Space flight (STS-90), it was possible to measure BP and nerve activity in the peroneal nerve of normal subjects while they were rotated about an Earth vertical axis (EVAR) and an off-vertical axis (OVAR) (Kaufmann et al. 2002). During constant-velocity EVAR, there was no discernible pattern in muscle sympathetic nerve activity (MSNA) associated with the frequency of rotation (Kaufmann et al. 2002). In contrast, during OVAR bursts of MSNA appeared at the frequency of rotation at different velocities, with peak activity close to or just after the subjects reached the “nose-up” position, when positive linear acceleration was maximal along the nasooccipital axis. Changes in MSNA (ΔMSNA) rose slowly and then had a sharp inflection as the head approached the nose-up position. The peak ΔMSNA occurred shortly after the nose-up position, and then ΔMSNA fell slowly over the rest of the cycle. Respiration, HR, and diastolic BP were also closely linked to the rotation rate. Diastolic BP was lowest close to the nose-up and rose as the subjects approached the nose-down position. Because OVAR in the steady state is mainly driven by otolith activation (Dai et al. 2003, 2010; Fanelli et al. 1990; Raphan and Schnabolk 1988), it suggested that sympathetic outflow to the legs is modulated at short latencies during forward linear acceleration, which is the equivalent linear acceleration when the nose is pointing up during OVAR (Kaufmann et al. 2002). These conclusions were supported by giving trains of high-frequency pulses over the mastoids delivered to the vestibular nerve at an appropriate latency following the R wave of the ECG (Voustianiouk et al. 2006). The finding that ocular roll but not nystagmus is induced by galvanic stimulation (MacDougall et al. 2002, 2003, 2005; Watson et al. 1998) has provided strong evidence that the major effect of the galvanic stimulus is on the otolith organs, not the semicircular canals, and is likely the feedforward driving mechanism to increase sympathetic outflow similar to the response during OVAR in response to head and body acceleration (Cohen et al. 2012; Kaufmann et al. 2002; Voustianiouk et al. 2006).

An important observation during these studies was that sinusoidal galvanic stimulation (sGVS) evoked BP and HR changes in addition to changes in MSNA (Kaufmann et al. 2002; Voustianiouk et al. 2006). In studies on rats, there was an unexpected finding that vasovagal responses (VVRs) were elicited in many of the rats exposed to sGVS (Cohen et al. 2010). VVRs are characterized by sudden reductions in BP and HR, which is frequently associated with dizziness and syncope (Lewis 1932). This raised the specter that the role of the vestibular system in generating vasovagal syncope could be studied in the anesthetized rat (Cohen et al. 2011). Because BP and HR signals contained a complex mixture of frequencies in response to a sGVS, it became advantageous to perform a wavelet analysis of the BP and HR waveforms to determine the behavior of this waveform at a multiresolution scale with Daubechies wavelets (Cohen et al. 2011). BP and HR signals were examined in four frequency bands that were associated with the stimulus and its harmonics. The wavelet analysis was particularly useful in identifying the frequency bands that were elicited in BP and HR and in comparing the energies in these bands to that of the stimulus. The changes in BP and HR induced by sGVS in the rat were similar to the changes in BP and HR observed in a wavelet analysis of a “syncopal youth” during a faint by Nowak et al. (2009). The slow component in the rat and the wavelet-based frequency band distribution bore a striking resemblance to the human data. As in humans, the slow response was not elicited in all rats, nor could it always be elicited in susceptible rats, but it appeared many times in a susceptible animal. The wavelet analysis showed that the transient component of the alterations in BP and HR was confined to low frequency bands, although the energy distributions for BP and HR were somewhat different. Since sGVS mainly activates the otolith organs, the data suggest that the otolithic system can exert a powerful inhibitory effect on both BP and HR whenever activated, independent of direction of activation. These frequency bands also demonstrated that there was significant energy at four times the stimulus frequency when the stimulus was 0.17 Hz, at the upper limit of frequencies that could activate the simultaneous inhibitory modulation in BP and HR. Whether the energies distributed in the high frequency band represented dispersion of the responses that were more coherent at the lower frequencies of sGVS (0.025 Hz) or were due to activity generated by the higher stimulus frequencies is not known. Regardless, the wavelet analysis clearly demonstrated that, despite the diffuse nature of the sGVS stimulus to the vestibular end organs, the changes in BP and HR were associated with the stimulus frequency and that the stimulus was channeled through otolith activation (Cohen et al. 2011).

Moreover, it had been suggested that a close association exists between the initiation of VVRs in human fainters and BP oscillations in the Mayer wave frequency range (Bent et al. 2006; Cohen et al. 2013). This possibility was explored by determining whether tilts in the susceptible rats, i.e., rats that had developed VVRs from tilt, were associated with significant changes in the amplitude of BP oscillations in the Mayer wave range, comparable to those induced by sGVS. The wavelet analysis showed that sGVS, tilting the animal, and translation while rotating, which tilted the gravito-inertial acceleration (GIA) relative to the animal, all had the maximum power in frequency bands associated with Mayer waves (Cohen et al. 2013; Julien 2006; Mayer 1876; Nowak et al. 2009). On the basis of findings in the anesthetized rat, we proposed that there is a specific low frequency band that contains activity that triggers the cardiovascular system into oscillation and that these oscillations are critical for the generation of VVRs in rats, and presumably VVR and syncope in humans. We further hypothesize that the otolith system is the major pathway from the vestibular to the autonomic system that is responsible for vestibular-induced neurogenic syncope (Yakushin et al. 2017).

Key questions that arose throughout these studies were how a model could be developed to explain the behavior of VVR and how the vestibular system initiates a VVR and syncope (Raphan et al. 2016). Other models of BP modulation that had been developed to address these questions generally have used input-output linear models (Julien 2006). Such models do not apply to oscillating systems, however, and cannot explain their control, which is inherently nonlinear (Ocon et al. 2011; Ottesen 1997, 2000; Ottesen and Olufsen 2011). The methodological approach we used was a multiresolution mathematical model. The first level at the highest resolution modeled the systoles and diastoles, which could simulate those measured in the anesthetized rat. This enabled a modeling framework for studying the modulations in peak BP and HR and provided the ability to define the parameters that might be important for generating VVRs. The conceptual framework was that of a relaxation oscillator that produces a nonsinusoidal repetitive output (Noble and Noble 2011; van der Pol and van der Mark 1927, 1928). The second-order system that realized the relaxation oscillator had two states, nonlinear thresholding and saturation, consistent with feedback mechanisms characteristic of the baroreflex feedback. The model oscillations displayed systolic and diastolic behavior. The systolic and diastolic transitions and oscillations were sustained by an input related to Desired BP. Interestingly, the model had the same structure that was used to model vestibular and optokinetic nystagmus (Raphan 1976; Raphan et al. 2016).

The frequency and amplitude of the oscillations were modulated by introducing low-frequency vestibular input into the feedback loop to test the hypothesis that vestibular control of the systolic-diastolic oscillations was accomplished through the baroreflex feedback. Simulations were then performed to identify how specific parameters of the model affected the frequency and amplitude of the oscillations in BP and HR, as well as how modification of the parameters could simulate the behavior of the VVR. The model was further tested by comparing its predictions to experimental data on pulse pressure during normal and VVR. An important point of the model was that the shape of the systolic/diastolic waveform is not determined strictly by the heart. It is determined by an internal model (relaxation oscillator) through a feedback neural network, which mimics the oscillation features of the heart. The feedback mechanisms implement closed-loop control and then activate actuators through nonlinear mechanisms that control the constriction of the vascular beds, which has been referred to as peripheral vascular resistance. The feedback from the baroreceptors is compared with the output of the internal model to implement model-reference control. At a larger timescale (10 s) or lower resolution, oscillations in systolic amplitude related to breathing can be observed. There are also small-amplitude very low-frequency oscillations in the systolic and diastolic BP that have been termed Mayer waves (Julien 2006; Mayer 1876; Myers et al. 2001) as well as low-frequency oscillations, which are associated with VVRs, which we have termed vasovagal oscillations (Yakushin et al. 2014). It was interesting that the model simulated the data at all levels of resolution (Raphan et al. 2016). The model predicted the diastoles and systoles during baseline BP. It also predicted the derivatives of BP and consequently the phase plane trajectories of the diastolic-systolic pulse.

From this model-based analysis, the following hypotheses were formulated about the parameters of the system that caused a cessation of oscillations leading to VVRs and the structures in the central nervous system that might be involved in generating a VVR:

• 1) If the central signal that generates Desired BP is maintained, compensatory mechanisms associated with baroreflex sensitivity could maintain adequate blood flow.
• 2) If there is a drop in Desired BP, then a combined fall in BP and HR sends the oscillator into a reset state, i.e., a VVR. Where the signal related to Desired BP is generated in the central nervous system and how it functions are as yet not known.
• 3) The vestibular input can alter the amplitude of the systoles, thereby modulating BP and HR to implement the vestibulo-sympathetic reflex (VSR).
• 4) The model also suggests that continuous oscillation induced by this input can generate a precipitous drop in in Desired BP, simulating the behavior during VVRs, presumably through cerebellar mechanisms (Grubb 2005; Julu et al. 2003).

CONCLUSIONS AND FUTURE DIRECTION OF RESEARCH ON THE VESTIBULAR SYSTEM AND ITS MULTIMOTOR CONTROL

Through the nineteenth and twentieth centuries, vestibular function was largely studied from a reductionist viewpoint by concentrating on the responses around one or two axes, with the subject’s head and body restrained. This approach has been particularly useful in determining canal function, and the central organization of the aVOR, since the spatial planes of action are well defined. Studying otolith function is more challenging, since the otoliths sense acceleration via polarization vectors that are broadly distributed in space. Moreover, no apparent central segregation into activity related to specific spatial planes or axes has yet been clearly identified. Thus the studies of the lVOR have been confined to using body- and head-fixed paradigms, where the linear acceleration on the head can be precisely controlled.

It is becoming increasingly clear, however, that the vestibular system detects and controls body, head, and eye movements in three-dimensional space by integrative action of canal, otolith, visual, somatosensory, and motor activity. It also plays an important role in locomotion and controlling autonomic behavior and neurogenic syncope. The neural networks that process this information must take into account the different reference frames and the dynamics, which govern the sensory and motor apparatus. Future studies of vestibular function, therefore, must necessarily consider not only how three-dimensional movements are sensed but also how central coordinate frames formed by the neural networks that process vestibular signals are coordinated to maintain stability in space as well as ensuring that BP and HR are maintained while performing a host of functions. To tackle these problems, the model-based approaches described in this review should become increasingly more important, especially as technical advances permit studies relating head and eye movements to body movements during complex motions with the head free and during natural locomotor tasks. These three-dimensional techniques will also become important in understanding adaptive strategies in microgravity, in gravity-dependent context-specific behavior, and in adapting to diseases that affect balance and locomotion. It is extraordinary that Dr. Bernard Cohen was at the forefront of almost every aspect of vestibular function and we had the opportunity to collaborate on so many topics.

GRANTS

The preparation of the current review was supported in part by reassigned time to T. Raphan for research as a Distinguished Professor of CUNY.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author.

AUTHOR CONTRIBUTIONS

T.R. drafted manuscript; edited and revised manuscript; and approved final version of manuscript.