Brain Advance Access originally published online on November 21, 2006
Brain 2007 130(1):10-35; doi:10.1093/brain/awl309
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Review Article |
What clinical disorders tell us about the neural control of saccadic eye movements
1 University of Pavia Pavia, Italy 2 Case Western Reserve University, Cleveland OH 3 Johns Hopkins University Baltimore, MD, USA 4 National Eye Institute Bethesda, MD, USA
Correspondence to: Lance M. Optican, PhD, Laboratory of Sensorimotor Research, NEI, NIH, DHHS, Building 49, Room 2A50, Bethesda, MD 20892-4435, USA E-mail: LanceOptican{at}nih.gov
 |
Summary |
|---|
Top
Summary
IntroductionSaccades are rapid eye movements that redirect the fovea from one object to another. A great deal has been learned about the anatomy and physiology of saccades, making them an ideal system for studying the neural control of movement. Basic research on normal eye movements has greatly increased our understanding of saccadic performance, anatomy and physiology, and led to a large number of control system models. These models simulate normal saccades well, but are challenged by clinical disorders because they often do not incorporate the specific anatomical and physiological substrates needed to model clinically important abnormalities. Historically, studies of saccadic abnormalities in patients have played a critical role in understanding the neural control of saccades because they provide information that complements basic research and thus restricts hypotheses to those that are biologically plausible. This review presents four examples of clinical disorders (slow saccades, interrupted saccades, high-frequency saccadic oscillations and macrosaccadic oscillations) that have provided insights into the neurobiology of saccades, have driven the development of new models, and have suggested an explanation or treatment for these disorders. We raise general questions for both scientists and clinicians that will assist in their efforts to understand the neural control of movement, improve diagnostic criteria and develop new treatments.
Key Words: cerebellum; macrosaccadic oscillations; opsoclonus; saccadic palsy; superior colliculus
Abbreviations: cFN, caudal FN; cMRF, central mesencephalic reticular formation; EBN, excitatory PBN; FN, fastigial nucleus; FNN, FN neuron; IBN, inhibitory PBN; IN, internuclear neuron; INC, interstitial nucleus of Cajal; LLBN, long-lead burst neuron; LR, lateral rectus muscle; MedRF, medullary reticular formation; MN, motor neuron; MR, medial rectus; MVN, medial vestibular nucleus; NMDA, n-methyl D-aspartate; NPH, nucleus prepositus hypoglossi; NRTP, nucleus reticularis tegmenti pontis; OPN, omnipause neuron; PBN, premotor burst neuron; PG, pulse generator; PMT, paramedian tract; PPRF, paramedian pontine reticular formation; riMLF, rostral interstitial nucleus of the medial longitudinal fasciculus; RIP, raphe interpositus nucleus; SC, superior colliculus; SCBN, SC burst neurons; SCBUN, SC build-up neurons; SO, superior oblique; SR, superior rectus; T-channel, T-type Ca2+ channel; VIn, sixth nerve
Received July 17, 2006. Revised October 5, 2006. Accepted October 6, 2006.
|
Introduction |
|---|
Saccades are rapid eye movements used to redirect the fovea from one object to another. They must be fast and accurate to support clear vision. A great deal has been learned about the anatomy and physiology of saccades since they were first reliably recorded over 100 years ago by Dodge and Cline (1901
; Wade and Tatler, 2005). This knowledge makes them an ideal system for studying the neural control of movement. Basic research on normal saccades, especially neurophysiological recordings from identified neurons during behaviour in monkeys, has revealed a great deal about saccadic performance, anatomy, and physiology. Clinical studies of patients with saccadic disorders, and studies of the effects on saccades of brain lesions in animals, have led to the development of quantitative hypotheses (models) of the neural control of saccades. These models simulate normal saccades well, but are challenged by clinical disorders because they often do not incorporate the specific anatomical and physiological substrates needed to model clinically important abnormalities. The interaction between research on animals and patients thus leads to a better understanding of how the brain controls movements than either alone. This interaction has made saccades one of the best understood of all vertebrate movements.
The goal of theoretical studies of eye movements is to develop models that realistically represent neurobiological processes, i.e. those that are isomorphic with the brain. Such models elucidate the neuronal mechanisms underlying motor control (Girard and Berthoz, 2005). The purpose of a model is to summarize knowledge, support insights, make hypotheses explicit and quantitative, and predict or explain new phenomena. Although each aspect of a movement can be explained by many models, the requirement that a single model account for as much normal and abnormal behaviour as possible constrains the choice of models and reveals isomorphisms that contribute to our understanding of brain function. Thus, a key factor in modelling the neural control of saccades was the interaction of clinical and basic science. Nonetheless, although clinician-scientists interested in abnormal eye movements have used saccadic models to explain human disorders, the full impact of the basic research effort on saccades has yet to be translated into better biomedical care, i.e. to the bedside.
What might be impeding this translation? First, most scientists are not familiar with clinical disorders of human saccades, and most clinicians are not familiar with experimental studies of saccades in animals. Second, experimental and clinical studies often differ, in the experimental paradigms, method of training or instruction given to the subject, motivation or rewards, number of subjects involved, ability to characterize the deficit, and availability of controls. Third, clinical studies tend to be inclusive, attempting to describe features common among a group of patients with similar symptoms, whereas scientific studies tend to be exclusive, focusing on one or a few differences between experimental and control subjects. Finally, basic scientists often study saccades in several species (e.g. monkey or cat) that are both anatomically and physiologically different from human patients. These different approaches have sometimes led the two groups to emphasize different aspects of motor control. For example, the superior colliculus (SC, a midbrain structure that is involved in making eye and head movements) has been the focus of intense interest by basic scientists for over thirty years, but discussion of the effects of SC lesions is conspicuously absent from clinical texts.
Our goal here is to stimulate more interdisciplinary interactions by reviewing selected disorders of human saccades that pose specific questions for current models. In each case we describe key features of the abnormality, and discuss the issues that the disorder poses for saccadic models. We limit the scope of this review to brainstem and cerebellar contributions to the generation of saccades, because these are better understood than more cognitive functions, such as detecting and selecting targets in a visual scene.
|
Characteristics of saccades |
|---|
Saccades are the rapid eye movements used to voluntarily move gaze from one target of interest to another (Fig. 1). Human saccades follow a target jump within
250 ms, are fast (up to 600°/s), brief (typically 30–100 ms), accurate, and stop abruptly (i.e. with little subsequent ocular drift). Saccades made to target jumps >10° in amplitude often undershoot the target by 10% (Kapoula, 1985), and, after a short latency (150 ms), are followed by a corrective saccade. Saccadic waveforms are characterized by the main sequence, a set of relationships between saccade amplitude and peak velocity (Fig. 1E), and between amplitude and duration (Fig. 1G) (Bahill et al., 1975). The first is logarithmically compressed for large amplitudes, whereas the latter is linear down to <5° with a non-zero Y-intercept. The main sequence plots are similar in monkeys and humans, except that monkey eye movements are 30–50% faster. Patients with certain diseases make saccades that deviate in specific ways from the normal main-sequence plots, making them a useful diagnostic tool. Other useful measures of saccades are their reaction time (latency), symmetry of their velocity waveform, and trajectory in space or on a phase plane (discussed further below under Clinical disorders of the saccadic system and the development of models).
|
During visual search, the point of fixation is moved between features that lie in different directions and at different depths. Thus, saccadic movements generally have both conjugate (both eyes rotate in the same direction, called version) and disconjugate (the eyes rotate in opposite directions, called vergence) components.
This review summarizes current knowledge about the generation of saccades, and examines how the study of different clinical disorders has led to advances in saccadic models. First, we will review the neurobiology of saccades, using a bottom-up approach to identify brainstem and cerebellar components that can be incorporated into models for saccade generation. Second, we will discuss, from a historical perspective, how current models for the generation of saccades by the brainstem and cerebellum were developed. Third, we will present four clinical disorders of saccades (slow saccades, interrupted saccades, high-frequency saccadic oscillations and macrosaccadic oscillations). We selected these disorders because they are well defined and instructive about the underlying neurobiology, and we will discuss their implications for current models. Finally, we will discuss how models may aid the understanding of saccadic disorders and lead to the development of clinically useful therapies.
Further details are documented in supplementary material at Brain online, including videos of the clinical disorders presented here.
|
Neurobiology of saccades |
|---|
A sustained research effort over more than three decades, comprising many anatomical, physiological, and behavioural reports, has led to a better understanding of how the brain generates saccades. Here we provide a brief description of important brainstem and cerebellar populations of neurons that contribute to the generation of saccades (key features of each type of neuron are summarized in tables). A schematic summary of the anatomy is shown in Fig. 2.
|
Oculomotor plant
When discussing the neural control of eye movements, it is helpful to start by considering the dynamics of the eye and orbital tissues (e.g. Tenon's capsule, fat, ligaments), extraocular muscles and pulleys. Together, these elements form the oculomotor plant (in engineering terms plant refers to whatever is controlled).
Extraocular muscles
Movements of each eye are controlled by six extraocular muscles, which originate at the back or nasal side of the orbit and travel to fibromuscular pulleys that are formed by the fascia of the orbit (Demer, 2004). The outer, orbital part of the muscle inserts partially on the pulley, and the inner global part passes through the pulley and inserts on the globe. Details of ocular movements in three dimensions (yaw, pitch and roll) depend upon the geometry of the muscle origins, pulleys and insertions (Quaia and Optican, 2003a), but are not the focus of this review. Here, we will regard the six muscles as grouped into three agonist-antagonist pairs obeying Sherrington's law of reciprocal innervation: lateral rectus (LR) and medial rectus (MR); superior rectus (SR) and inferior rectus (IR); superior oblique (SO) and inferior oblique (IO).
Ocular motoneurons
The brain innervates the extraocular muscles via three cranial nerves. The abducens nerve (VIn) innervates the ipsilateral LR, the trochlear nerve (IVn) innervates the contralateral SO muscle, and the oculomotor nerve (IIIn) innervates the ipsilateral MR, IR, IO and contralateral SR. Muscles are paired in two ways. First, for each agonist there is a corresponding antagonist muscle with almost the same axis of action for that eye (i.e. LR–MR, SR–IR, SO–IO). Second, muscles are yoked to move both eyes together (e.g. left LR and right MR, left SO and right IO). Saccades tend to follow Hering's law, with equal innervation going to muscles in a yoked pair. However, as noted above, when looking between targets at different depths, different size movements can be made in each eye. In this review, we assume separate conjugate saccade and disconjugate vergence systems (Enright, 1998; Zhou and King, 1998; Ramat et al., 1999). Although recent evidence suggests that this is an oversimplification, it will be sufficient for the types of disorders discussed here.
Final common path
The same motor neurons (MNs) and extraocular muscles are active for all types of eye movements (e.g. saccades, pursuit and vergence). Thus, systems generating innervation for different movement types are said to share a final common path. The shape of the signals entering the final common path depends on both the desired eye movement and the dynamic response of the eye plant. One of the first important insights in understanding the neural control of movement came when (Robinson 1964) showed that the dynamics of the oculomotor plant were dominated by the viscosity of the muscles themselves. Thus, if a constant torque were suddenly applied to the globe, the eye would drift from its initial orientation to its final orientation with an exponentially decaying speed (time constant 200 ms) and the eye would reach a steady position only after about three time constants (600 ms). Since the eyes can make a saccade every 200–300 ms, this has obvious disadvantages for vision. In addition, elastic forces imposed by the orbital tissues tend to return the eye to centre position.
Thus, all types of movement entering the final common path have to consist of two main components: a phasic component (pulse) giving the torque needed to overcome the viscous drag of the orbital tissues, and a tonic component (step) giving the torque needed to overcome the elasticity of the orbital tissues. For saccades, the phasic part of the innervation has a discharge rate about nine times higher than the tonic part, so the combined innervation has a pulse-step shape (Robinson, 1970; Quaia and Optican, 2003b). Thus, to make a saccade, the brain must compute two different but related components of innervation. If the pulse and step are not matched correctly, the eye will drift (for 600 ms) after each saccade from the position reached during the pulse to that corresponding to the step.
Final common integrator
Robinson (1975) suggested that the brain could calculate the step from the pulse by integrating it (in the mathematical sense), because the number of spikes in the pulse was proportional to saccade amplitude. Earlier, Skavenski and Robinson (1973) had discovered strong evidence for a neural integrator in the vestibular system, which introduced a phase lag of 90° between the vestibular signal (head velocity) and the signal encoded on ocular motoneurons (eye position). If all the eye movement systems need to share a final common path that is matched to the plant, they must also all share the same integrator (otherwise, switching from, say, pursuit to vestibular movements would cause the eye to drift). Thus, the input to the final common path for any type of eye movement (e.g. vestibular, pursuit, saccade) should be a desired eye velocity.
|
Brain stemreticular formation |
|---|
The brainstem houses the essential machinery for generating saccades. A partial functional dichotomy exists between the midbrain and pons. Neurons in the pons are mostly concerned with the horizontal component of saccades, whereas the midbrain controls vertical and torsional components.
Premotor burst neurons (PBN)
Two main types of neuron are critical for saccade generation, which we collectively refer to as PBN because they project monosynaptically to ocular motoneurons (Table 1). The excitatory PBNs (EBN) burst just before saccades and probably generate the pulse of saccadic innervation (van Gisbergen et al., 1981). These neurons are glutamatergic (McElligott and Spencer, 2000; Horn, 2006).
|
Horizontal EBN are located in the paramedian pontine reticular formation (PPRF), below and extending rostrally from the level of the abducens nucleus (Strassman et al., 1986a
; Horn et al., 1997). They project monosynaptically to the ipsilateral abducens nucleus, and to a network of neurons in the nucleus prepositus hypoglossi (NPH) and adjacent medial vestibular nucleus (MVN) that contribute to the final common integrator's horizontal component. Bilateral chemical lesions in the PPRF selectively abolish horizontal saccades (Henn et al., 1984).
Vertical and torsional EBN are located in the midbrain, in the rostral interstitial nuclei of the medial longitudinal fasciculus (riMLF) (Horn and Büttner-Ennever, 1998). They project monosynaptically to the vertical and torsional MNs, and to the interstitial nucleus of Cajal (INC), which is an important part of the final common integrator's vertical and torsional components. Bilateral chemical lesions of riMLF abolish vertical and torsional saccades (Suzuki et al., 1995).
The second group of PBN consists of inhibitory burst neurons (IBN), which have a firing pattern almost identical to that of EBN (Strassman et al., 1986a). Horizontal IBN lie in the medullary reticular formation (MedRF), below and extending caudally to the level of the abducens nucleus. The horizontal IBN are glycinergic (McElligott and Spencer, 2000), and project to several sites, but predominantly contralaterally to the abducens nucleus and horizontal EBN and IBN. They suppress activity in the contralateral PBN of the antagonist muscle, and thus may mediate Sherrington's law of reciprocal innervation.
Vertical and torsional IBN lie in the midbrain in the region of the INC and riMLF (Horn, 2006). These IBN are assumed to play a role similar to that of horizontal IBN, but for vertical and torsional components of saccades. Note that the vertical IBN are GABAergic (Spencer and Wang, 1996), whereas the horizontal IBNs are glycinergic. As discussed later, this difference may contribute to the prevalence of horizontal saccadic oscillations, because glycine is active at two receptors, an inhibitory (strychnine-sensitive) receptor and a neuromodulatory (n-methyl D-aspartate; NMDA) receptor (Miura and Optican, 2006).
Omnipause neurons (OPN)
OPN (Table 2) are an important element of the saccadic system, but their role is not fully understood. OPN lie close to the midline in the raphe interpositus nucleus (RIP) and their dendritic arborizations cross the midline (Büttner-Ennever et al., 1999). They are glycinergic, and are presumed to inhibit PBN bilaterally in both the pons and midbrain. In an awake animal, OPN are tonically active (Strassman et al., 1987). The OPN were first recognized because they pause before saccades in any direction (Keller, 1974). Robinson proposed that their function was to prevent the PBN from firing except when a saccade was called for (Robinson, 1975). Later experiments showed that OPN performed a more complicated function than just gating saccades. Kaneko and colleagues made chemical lesions of the RIP, and found that saccades became slower, but had normal latency and accuracy (Kaneko, 1996; Soetedjo et al., 2002). Miura and Optican pointed out that this could be explained if OPN were not acting as a gate, but rather as a neuromodulator to increase the responsiveness of saccade-related neurons (Miura and Optican, 2006). This role is necessary because, without it, neurons in the saccadic system would have very high gains and would be susceptible to oscillations. By increasing the gain of these neurons only before saccades, OPN could prepare the system to make a fast movement without running the risk of becoming unstable. This is also consistent with the finding that OPN are glycinergic, because their pre-saccade activity releases the agonist co-factor for NMDA channels on PBN. They do not cause a saccade, however, because the action of the glycine at the inhibitory strychnine-sensitive channels is assumed to dominate its effect at the NMDA channels.
|
Excitatory input to the OPN arrives from the SC and vestibular nuclei (Strassman et al., 1987
; Langer and Kaneko, 1990). The OPN pause 16 ms before saccades, and resume 15 ms before the end of saccades. Inhibitory input seems to show two time courses, a fast one associated with a trigger that shuts off the OPN to initiate a saccade, and a slower one that prevents the OPN from turning on again until the saccade is over (Yoshida et al., 1999).
Latch-circuit neurons
The original local feedback model of the saccadic system (Robinson, 1975; Zee et al., 1976) was the first that could generate the pulse of innervation needed to make a saccade and stop automatically (Fig. 3A). The main element of the local feedback model was a saturating high-gain amplifier (pulse generator; PG). A gate was needed to prevent the high-gain amplifier inside the feedback loop from oscillating. In the original model, the pulse generator neuron was inhibited by a switch cell (S in Fig. 3A), which gated the input to prevent the oscillations. To initiate a saccade, a trigger signal (Trig) of short, fixed duration would inhibit the switch cell, and the pulse generator would begin firing. As the movement progressed, the trigger input died away, but the activity of the pulse generator neuron was fed back, through an OR-gate, to keep the switch cell off until the movement was over. This gate was called a latch because once it was closed, it kept itself closed (through feedback) until the end of the saccade. When the pulse generator neuron and the trigger were both off, the suppressor cell (S) became active because of a static input Robinson called ‘tone’.
|
, corrects retinal error, e, and recreates the central percept of the spatial location of the target, 
. This signal acts as the input to a bang–bang control system consisting of a high-gain saturating amplifier, PG, followed by an integrator, NI. SJ1, 2, 3 are summing junctions; suppressor switch S is normally open, unless the OR-gate (red) is activated by a trigger signal, Trig, or the PG output. MLF carries the pre-emphasis signal (or burst). The orbital dynamics are in Laplace notation and Te is the time constant of the ocular motor plant. (B) Model to account for saccadic oscillations by a high-gain pulse generator with a negative feedback loop (Ashe et al., 1991
). A trigger signal (Trig) inhibits the OPN (red), which also receive a bias signal (TONE). The output of the EBN is a pulse of innervation, which projects to ocular motoneurons (OMN), to the common ocular motor integrator (NI) that generates a step of innervation, and to the IBN. RI is the re-settable integrator with a delay (blue) interposed between its output (an internal estimate of current eye position) and an inhibitory synapse on the EBN. The IBN inhibit OPN during the saccade (NB: IBN to OPN projections have not been found).Subsequent research into the brainstem cell types associated the pulse generator cells with the excitatory medium-lead burst neurons (EBN), and the suppressor cells with the OPNs. This still leaves four unanswered questions about the suppressor mechanism. (i) Why are the OPNs usually active (i.e. where does the tone come from)? (ii) What shuts them down before the saccade (what is the trigger)? (iii) What holds them off during the saccade? And, (iv) what reinitiates their activity at the end of a saccade?
We refer to the suppressor mechanism as a latch circuit, because a latch remembers its input even after the input has been removed. This requires some form of feedback to create two stable states, on and off, that are controlled by two inputs, usually called set and reset. Robinson's original gate circuit acts as a latch circuit because it feeds back the output of the pulse generator to the suppressor cell. However, it only has the set input, which is the trigger. The circuit unlatches automatically when the motor error estimate is reduced to the dead-zone (which stops the burst cell activity). The problem of making accurate saccades still comes down to how and when to restart the OPNs. A more recent hypothesis has been proposed that separates the feedback control of gaze from the latch circuit (Lefèvre et al., 1998). The saccade normally ends when the ipsilateral cerebellum fires the contralateral brainstem IBNs, choking off (i.e. interrupting, stopping) the drive signal to the MNs. By making the latch circuit independent of motor error, the effect of time jitter on movement accuracy is significantly reduced. Accuracy can be maintained by the cerebellar circuit, and stability can be guaranteed as long as the OPNs reactivate soon after. A more detailed hypothesis for a latch circuit is developed below in A model of the latch circuit.
Long-lead burst neurons (LLBN)
LLBN (Table 3) are found in the brainstem and receive input from the SC and cortical areas responsible for saccades, such as the frontal eye field, parietal eye field and supplementary eye field (Scudder et al., 1996a, b). Some of the LLBN probably project to the premotor burst neurons, whereas others [in the nucleus reticularis tegmenti pontis (NRTP)] project to the cerebellum. Scudder (1988) proposed a saccade model that used the LLBN as a re-settable integrator in the feed-forward path. The re-settable integrator was thought to be distinct from the common final integrator and to apply only to saccadic signals (see below, in Conceptual evolution of saccadic models). However, later evidence suggests that the re-settable integrator does not really exist, and that the cerebellum replaces its function (Quaia et al., 1999). Thus, LLBN seem to act as a summing junction for combining saccade commands from different areas involved in the preparation for a saccade.
|
|
Eye position (tonic) neurons and the neural integrator for eye movements |
|---|
Tonic (and burst–tonic) neurons carry a signal related to the step component of MNs. The step is assumed to be obtained by integrating (in the mathematical sense) velocity signals from PBN that generate the pulse of innervation (this is an oversimplification, but sufficient for our needs here). Neurons that carry an eye position, or step, signal for horizontal saccades lie in the medulla and pons in the NPH nuclei (McCrea and Horn, 2006
), and adjacent MVN. Neurons that carry a vertical and torsional eye position (step) signal are present in and around the INC. In addition, the cell groups of the paramedian tract (PMT) also probably contribute to the neural integrator (Büttner-Ennever and Horn, 1996).
The mechanism that performs the integration is not well understood, although models that integrate because of reciprocal innervation (positive or negative) across the midline have been developed (Cannon and Robinson, 1985; Cova and Galiana, 1996; Arnold and Robinson, 1997), and recent work has cast light on how networks of neurons could perform this mathematical operation (Aksay et al., 2005).
Curiously, when the cerebellum is lesioned the time constant of this integrator drops from 20 s to 2 s (Carpenter, 1972), but when the commissures connecting the left and right halves of the integrator are cut, the time constant drops below 300 ms (i.e. only the time constant of the oculomotor plant itself is left). Thus, the integration of ocular motor signals depends on both cerebellar and brainstem circuits, but their respective roles remain enigmatic.
Superior colliculus (SC)
One influence on PBN and OPN is the SC, a multi-layered structure in the midbrain. The SC is laid out in a retinotopic map (Robinson, 1972), with small contralateral movements related to activity in the rostral SC, and large contralateral movements related to caudal SC. This is true with the head fixed (eye only saccades) or free (gaze saccades, where gaze = eye + head). Neurons in the superficial layer are related to visual events, but neurons in the intermediate and deep layers are related to both visual and motor events. Schiller et al. (1980) showed that saccades remained accurate after SC ablation, and thus that it was not necessary for generating the pulse for saccades. Later it was also shown that ablations of the cerebellum caused persistent deficits in saccade amplitude (e.g. Optican and Robinson, 1980). Despite these early observations, until the early 1990s the SC was widely assumed to control saccadic amplitude, direction and trajectory. Many studies have since shown that the major role of the SC seems to be in selecting a target for foveation—either with saccadic or pursuit movements (Krauzlis, 2004)—initiating the movement (Bell et al., 2005) and contributing to its speed (Hanes et al., 2005), but does not determine amplitude, duration, direction or trajectory of the saccade.
The intermediate and deep layers of the SC contain two major types of neurons that are laid out in a retinotopic map. One type shows a target related build-up in activity (SCBUN), and a burst just before the saccade. The other bursts just before the saccade, but has no prelude of activity (SCBN). Electrophysiological studies show that both SCBN and SCBUN project from the SC to the brainstem (Munoz and Wurtz, 1993, 1995a). The SC projection is not uniform. The rostral half of the SC sends more fibres to the RIP (region of the OPN) than does the caudal half. The SC neurons project mostly to LLBN, and probably do not have a functional projection to PBN (Gandhi and Keller, 1997). The level of activity in the prelude of SCBUN corresponds to the likelihood that the target at that location will be the goal of a saccade (Basso and Wurtz, 1997). SC activity is also a function of retinal error, whether for a saccade or a pursuit movement (Krauzlis et al., 2004).
In every experiment where the saccade goal is not the same as the visual target, such as adapted saccades (FitzGibbon et al., 1986), saccades to moving targets (Keller et al., 1996b), and strongly curved saccades (Port and Wurtz, 2003), the locus of activity on the SC corresponds to the visual target, and not the ensuing movement. Ablations of the SC significantly increase saccade latency and reduce peak speed, but do not affect accuracy (Hanes et al., 2005). Thus, the most parsimonious interpretation of SC function is that it contributes to identifying the target (in retinotopic coordinates) that is to be foveated, generates a trigger signal to shut down the OPN, and sends a fixed-direction drive to LLBNs to begin the saccade.
Central mesencephalic reticular formation (cMRF)
The cMRF has strong, topographically organized, reciprocal connections with the SC (Cohen and Büttner-Ennever, 1984; Chen and May, 2000) and the region of the OPNs (Langer and Kaneko, 1983, 1990), and projects to the EBN and IBN in the brainstem (Büttner-Ennever et al., 1999). Cells in the cMRF that are related to saccades discharge for contraversive movements, and can be divided into two groups (Waitzman et al., 1996). One group has a low background rate of firing (
10 spikes/s), and the other, larger, group has a high background rate of firing (>10 spikes/s). Both groups start to burst 30 ms before the saccade. Some cells have an abrupt, or clipped, end to the burst, at the end of the saccade. Indeed, burst duration in 31% of cells in cMRF are correlated with saccade duration, 48% with saccade amplitude, and 58% with saccade velocity; some cells respond to two of the three metrics (Cromer and Waitzman, 2006). Lesions in the cMRF result in hypermetric contralateral saccades, reduced latency, and macrosaccadic square-wave jerks (Waitzman et al., 2000). These results suggest that the cMRF is important for saccade function at both the initiation and termination stages. One possible model of their contribution is described below.
Cerebellum
A second influence on PBN and OPN is the cerebellum, which plays an important role in steering and stopping saccades, thus determining their accuracy. The posterior pole of the midline cerebellum has been most intensively studied. This region is divided into a cortical part (vermis and paravermis) and the underlying deep cerebellar nuclei (fastigial, interpositus and dentate). All relevant signals (e.g. eye position and velocity, target position and velocity) needed to control saccades arrive in the cerebellar cortex on mossy fibres. Climbing fibres bring a signal from the inferior olive to the cerebellar cortex. Climbing fibre activity is related to the distance to the target (motor error) at the beginning and end of a movement (Kitazawa et al., 1998). The climbing fibre activity is assumed to be necessary to learn proper motor control, but its role is not well understood.
Dorsal vermis Purkinje cells (PC)
Stimulation in many parts of the cerebellum (lobes V–VII of the vermis, and the hemispheres, crus I and II) can evoke a saccade (Ron and Robinson, 1973). Stimulation of vermal lobule V evokes saccades that range from upward to horizontal, while stimulation of lobules VI and VII evokes saccades that range from horizontal to downward. The amplitude of the elicited saccade, and the amount of post-saccadic drift, depend upon the initial position of the eye in the orbit. Purkinje cells that discharge in relation to saccades are located in a more restricted region, lobes VI(c)–VII, called the oculomotor vermis (Noda and Fujikado, 1987). The oculomotor vermis projects to the caudal part of the fastigial nuclei. Purkinje cells in the dorsal vermis discharge 15 ms before saccades in a preferred direction (Ohtsuka and Noda, 1995). Stimulation of the vermis produces saccades with an ipsilateral component (Ron and Robinson, 1973). With currents near threshold, a topographic organization is evident (Noda and Fujikado, 1987).
Fastigial nucleus neurons (FNN)
The caudal part of the fastigial nucleus (cFN) is responsible for sending saccade commands to the region of the PBN in the contralateral brainstem. FNN fire tonically with a low rate, and burst near the time of a saccade (Ohtsuka and Noda, 1991; Fuchs et al., 1993). FNN fire for saccades in all directions, but electrical stimulation in the cFN elicits contralateral saccades (Noda et al., 1988). The latency of the burst is a function of the direction of the movement. FNN burst before saccade start for contralateral movements, but burst near saccade end for ipsilateral movements. Thus, the same neurons are firing for ipsi- and contralateral saccades, with the only difference being that they fire later for ipsilateral saccades. This issue is addressed further in the section Cerebellar models.
Cerebrum and basal ganglia
Many areas of the cerebral cortex (e.g. frontal eye fields, supplementary eye fields, lateral intraparietal cortex) are known to be involved in identifying and selecting targets for saccades (Leigh and Kennard, 2004). The basal ganglia are involved in selecting and preventing movements, and in reward (Hikosaka et al., 2000). The substantia nigra pars reticulata tonically inhibits the intermediate layers of the SC. This tonic activity can be suppressed by inhibition from the caudate nucleus, allowing a saccade. As important as they are for selecting the target for saccades, these areas do not play a role in generating the saccadic pulse itself. Thus, a discussion of their function is beyond the scope of this review.
|
Models for saccades |
|---|
Conceptual evolution of saccadic models
In this section we present a brief history of saccadic models, beginning with the first model that was physiologically plausible. Robinson proposed that the innervation needed to make a saccade had to consist of two components: a pulse to overcome the viscous drag in the orbit, allowing for a fast movement, and a step to overcome the elastic restoring force in the orbit. Robinson proposed that the step could be computed as a function of the pulse, which simplified the problem of generating saccadic innervation to that of generating the pulse (Robinson, 1973
). Furthermore, extraocular muscle proprioception, unlike skeletal muscle spindles, does not participate in a stretch reflex (Keller and Robinson, 1971). Thus, it was assumed that saccades were ballistic, i.e. their innervation was programmed before the saccade, and simply ‘played out’ during the movement.
Shortly after this, Zee et al. (1976) showed that the slow refixations made by patients with spinocerebellar disease were actually very slow saccades (see section Slow saccades). Indeed, these movements were so slow that visual re-afference returned in time to influence the saccadic system. They found that if the target jumped to a new location after the patient began a saccade, the eye would turn around in mid-flight and go to the new target. This demonstrated that the saccadic system was not ballistic and led Zee and Robinson (1979) to propose that the pulse of innervation was generated during the movement by an internal, or local, feedback loop based on an efference copy of eye position. This efference copy was obtained by integrating the pulse that came from the brainstem burst neurons (Zee et al., 1976). This model immediately found wide acceptance because it generated both normal and slow saccades that stopped automatically (i.e. without pre-programming), and included a role for the recently discovered burst and pause neurons in the pons. An important refinement of the original local feedback loop model was made by Zee and Robinson in 1979, stimulated by an attempt to explain microsaccadic oscillations (see Section on high-frequency saccadic oscillations). In developing that model they used a new non-linear function to represent the activity of the burst neurons for a given motor error which is still widely used today (van Gisbergen et al., 1981). To explain the small amplitude, high-frequency oscillations observed in some patients, they added a delay in the local feedback loop around the high-gain amplifier representing the burst neurons (Zee and Robinson, 1979; Ashe et al., 1991; Fig. 3B).
The original version of the model had an efference copy of the eye position signal which was fed back to a comparator that computed the instantaneous, or dynamic motor error: the difference between desired eye position and current eye position. The efference copy was also used to reconstruct an internal estimate of the target's position in spatial coordinates. However, all saccade-related neurons found to date encode retinotopic error and change in eye position, not position in space. When Jürgens et al. (1981) found evidence for a separate neural integrator for saccades, they modified the local feedback loop model so that an efference copy of eye velocity was fed to a re-settable integrator that was reset to zero before each saccade. The output of this re-settable integrator was an efference copy of instantaneous eye displacement, which could thus be compared to a desired eye displacement signal in retinotopic coordinates. No reconstruction of target position in space was needed.
The Jürgens–Robinson model is the one from which most others have descended. Saccadic models evolved slowly after the local feedback loop was introduced. Optican and Miles (1985) introduced a third component of innervation (the slide) that was necessary to explain the transition between the pulse and the step. This led to the current concept of saccadic innervation consisting of a pulse–slide–step, with the slide and the step being automatically computed from the pulse generated by the burst neurons. Based on experimental data from the cat (Munoz et al., 1991), activity in the SC during the execution of a saccade was modelled as a moving hill (Droulez and Berthoz, 1991; Lefèvre et al., 1994), which was considered to represent the tracking of the progress of the fovea toward the intended target (i.e. the dynamic motor error). This and other work (Waitzman et al., 1988) suggested that the saccadic feedback loop might be closed within the SC itself. However, subsequent efforts to show an error-related spread in monkeys have failed (Munoz and Wurtz, 1995b; Aizawa and Wurtz, 1998; Anderson et al., 1998).
More recently, Guitton and colleagues have found that in a head-free animal, the locus of activity in the SC corresponds to the retinotopic location of the target, not instantaneous motor error (Bergeron and Guitton, 2002; Bergeron et al., 2003). Furthermore, if multiple small steps are needed to acquire the target, the activity on the SC jumps from one locus to the next, always encoding the eccentricity of the target for the next saccade (Matsuo et al., 2004). Thus, despite evidence that the SC receives some feedback about gaze movements, it does not seem to be responsible for calculating motor error during a movement.
The major problem with the Jürgens–Robinson model is that it was designed from the top down to perform an eye tracking task, and thus is not isomorphic to the brain. Many structures in the brain that are clearly related to saccades, from the cerebral cortex to the SC to the cerebellum, were not included in that model, and there is no obvious way to fit them into it. The problem of incorporating anatomical and physiological evidence into a model was addressed directly by Optican and colleagues, who took descriptions of known cell activity in different brain areas and combined them (i.e. from the bottom-up) to form a working saccadic system (Lefèvre et al., 1998; Quaia et al., 1999). This neuromimetic model was able to make visually guided saccades, and provided a structure for the addition of more cell types and brain areas to account for other types of saccadic behaviour. More importantly, its isomorphic design made it completely different, in both structure and signals, from the classic engineering design of target tracking systems (Optican and Quaia, 2001). The neuromimetic model has many significant structural differences from the Jürgens–Robinson model, but maintains its key element, the local feedback loop for the control of saccade trajectory (Fig. 3). However, it has no re-settable integrator, no efference copy of eye displacement, and no motor error signal. Instead, all their functions are performed by distributed networks in the brainstem and cerebellum (Optican and Quaia, 2002). The performance of that model is very good when making normal saccades. However, it contained no mechanism for restarting the OPNs, and could not explain hypometria of saccades after lesions of contralateral FN. Future models will have to resolve these problems.
Simple saccade model that can be applied to clinical disorders
Recently we developed a model of brainstem and cerebellar control of horizontal saccades that we applied to the clinical disorders that we discuss in the next section. This model was derived from the neuromimetic model of Optican's group. It included many anatomical details of the brainstem circuitry that were not known when the Jürgens–Robinson model was proposed. The brainstem circuit was based on the anatomical findings by Strassman et al. (1986a), and first represented in a saccade model of the local feedback loop proposed by Scudder (1988). The new model includes two new feedback loops, the shortest being one between the IBNs on both sides. Although the IBNs are inhibitory, their reciprocal connection creates a positive-feedback loop. The second positive-feedback loop involves EBNs and IBNs on both sides. The IBNs are prone to oscillate, because of their high sensitivity, close proximity (which gives rise to only a short delay) and reciprocal negative connections (which result in positive feedback). In his model Scudder avoided oscillations by assuming that the strength of these connections was quite low, although there is no direct evidence about their strengths.
Our new mathematical model (Ramat et al., 2005) of the saccadic brainstem mechanism tries to reproduce the salient characteristics and projections of the different populations of neurons that are responsible for producing horizontal saccades from the OPN to the abducens nuclei (VI). Thus, the model represents the OPN, the bilateral EBN and IBN, and the VI nucleus including both the motor and internuclear neurons (INs; Fig 4A). Each population of neurons is modelled as a single element with lumped parameters. All neurons are modelled with a similar structure representing the membrane as a high-pass filter showing adaptation, followed by a non-linear saturation and a small delay (0.8 ms). The model for the EBN neurons is shown in Fig. 4B, where the output non-linearity is represented by the ‘Burst’ block implementing the soft saturation in Eq. 1 (Zee and Robinson, 1979).
|
cMRF connections form a positive-feedback loop that keeps them firing during the saccade. The output of this loop goes to latch cells (blue), which inhibit the OPN (see text for details).
 | (1) |
The same structure, although with independent parameters, was used to represent the IBN.
The consequence of the membrane adaptation property is that these neurons show post-inhibitory rebound: at the offset of inhibition there is a rebound in the membrane potential that carries it to positive values, thus allowing the cell to fire one or more action potentials spontaneously. Post-inhibitory rebound is a property of some neuron types that, at the offset of hyperpolarization, produce a discharge mediated by low-threshold Ca2+ channels (Perez-Reyes, 2003). This implies that neurons showing this property may fire a train of action potentials when they are abruptly disinhibited, even without an input actively driving the cell. Post-inhibitory rebound in the saccadic burst neurons was first introduced by Enderle and Engelken (1995) who suggested that it is the only mechanism needed for generating saccades. A later model combined the post-inhibitory rebound with excitatory input from LLBNs to generate saccades (Miura and Optican, 2006).
The circuitry that moves the eye horizontally is represented in black in Fig. 4A, including the lateral and medial rectus muscles, the adducting MNs in the III nuclei, the abducting MNs and IN in the abducens nuclei, and their projections. The input driving saccades is provided by the ipsilateral EBN (shown in blue in Fig. 4A), which project to both the abducens MN and abducens IN, the latter in turn projecting across the midline to the contralateral IIIn. The IBN, shown in red in Fig. 4A, are excited by the ipsilateral EBN and inhibit the contralateral VIn and PBN.
The interconnections between burst neurons form two positive-feedback loops: one short loop is formed by the IBN neurons reciprocally inhibiting each other on the two sides of the midline; the other involves both EBN and IBN bilateral groups. This longer loop is provided by the excitatory projection of EBN onto ipsilateral IBN, which in turn inhibit the contralateral EBN which excite the ipsilateral IBN, which cross back to inhibit the originally firing EBN. This organization, which was anatomically identified (Strassman et al., 1986a), implements Sherrington's law of reciprocal innervation (contraction of agonist while ensuring the inhibition of antagonist muscles), and favours the fast build-up of EBN discharge (Strassman et al., 1986a).
When a desired eye position signal coding for a rightward saccade excites the right EBN, it projects both to the ipsilateral VI nucleus and to the ipsilateral IBN. The excitation of ipsilateral IBN inhibits the contralateral VI, the contralateral EBN and the contralateral IBN. Therefore the inhibitory action of this contralateral (left) IBN group is reduced, both directly via the IBN–IBN projection and indirectly since the inhibition of the (left) EBN in turn reduces the (left) IBN excitation. The left IBN's inhibition disinhibits the right IBN, thus closing the short positive-feedback loop from IBN to IBN, and disinhibits the right EBN, thus closing the second positive-feedback loop involving EBN and IBN bilaterally.
The OPN (shown in green in Fig. 4A) project to all four groups of burst neurons, tonically inhibiting them except during saccades in all directions (Keller, 1974), blinks (Hepp et al., 1989, Mays and Morrisse, 1995), and to some extent during vergence (Busettini and Mays, 2003).
In addition to the described structures and connections, the suggested mathematical model includes a ‘desired eye position’ input to the saccadic mechanism, turning off the OPN when a saccade is programmed, and a local feedback loop responsible for turning off the saccades once the eyes are on target (Ramat et al., 2005).
In humans, the mechanism generating saccades is potentially unstable due to the high-gain of the output non-linearity of the burst neurons and to the positive-feedback loops coupling EBN–IBN and IBN–IBN. If the burst neurons are not inhibited by the OPN and are not driven to produce a saccade, such latent instability may lead to high-frequency, conjugate oscillations of the eyes composed of back-to-back saccades occurring without intervening periods of steady fixation (see section High-frequency saccadic oscillations). Such a condition may occur during blinks, saccade-vergence interactions, and orthogonally directed saccades.
In fact, OPN pause for saccades in any direction and thus release all populations of PBN, although the saccade may have only one component, e.g. the OPN shut off for a vertical saccade, disinhibiting both the vertical and horizontal PBN. The vertical EBN will be driven to produce a vertical saccade, while the horizontal EBN will not receive any driving input. Yet, the offset of the OPN hyperpolarization will produce post-inhibitory rebound in both horizontal IBN and EBN on both sides of the midline, causing these cells to simultaneously fire a few action potentials. Any imbalance in the circuit will allow one side to prevail and a periodic oscillation will ensue. Suppose that the right EBN produces a few spikes from post-inhibitory rebound. They will drive the right IBN which will inhibit the left EBN and the eyes will move to the right. Because of the fast decay (adaptation) of the post-inhibitory rebound in the right EBN, the left EBN will be disinhibited and will in turn show a post-inhibitory rebound, driving the left IBN, in turn inhibiting the right EBN and moving the eyes to the left. As the post-inhibitory rebound in the left EBN is extinguished, the process repeats itself until the OPN are turned back on after the vertical saccade is over.
Cerebellar models
All fibres coming to the cerebellar cortex also deliver a branch to the corresponding deep nuclei (e.g. both vermis and FN). Currently, there are two types of cerebellar models. The first type learns (because of association with climbing fibre activity) to activate an output fibre at some time after an input fibre discharges. These models are always 1D, and cannot be extended to deal with multidimensional problems. The second type is pre-wired to perform certain hypothetical functions, such as initializing the locus of activity on the vermis and updating that locus based on velocity feedback information (Optican and Quaia, 2002). These models are helpful in understanding control of saccade trajectory, but do not provide insights into how the cerebellum could develop a network with such a function.
The mossy fibre inputs to the cerebellum carry signals from the brainstem LLBNs in the NRTP. The NRTP receives inputs from the SC and the FN itself. Neurons in the cFN discharge about 8 ms prior to onset of saccades with contralateral components, but generally towards the end of saccades with ipsilateral components. A major mystery of the cerebellum is how does the output from the ipsilateral FN fire later than its inputs, i.e. near the end of the saccade? One interpretation (Lefèvre et al., 1998; Quaia et al., 1999) is that the cerebellum, through the FNN, controls the trajectory and endpoint of a saccade. The difference in latency between ipsi- and contralateral FNNs breaks the symmetry between the projections from the cFN to the PBN in the brainstem. For example, for a rightward saccade, the left FNN excite both the IBN and EBN on the right side (Fig. 4C). The IBN on the right then inhibit the EBN and IBN on the left side, allowing the eye to move to the right. At the end of the movement the right-side FNN fire, exciting both the IBN and EBN on the left side. However, the IBN on the right side are already inhibiting the PBN on the left. If we assume that the FNN excitation to the IBN, but not the EBN, overcomes the inhibition from the right IBN, then the IBN on the left will shut down the EBN on the right, stopping the saccade.
Optican and colleagues have proposed a theory to account for this experimental result (Lefèvre et al., 1998; Quaia et al., 1999; Optican and Quaia, 2002; Optican, 2005). In this theory the role of the cerebellum in movement control is 2-fold. First, it recognizes from the sensory, motor and behavioural context what its contribution to the movement should be (determined by learning from experience). Second, it integrates velocity feedback information to obtain instantaneous displacement information (thus performing a role equivalent to the resetting integrator in the Jürgens–Robinson model) and modifies the saccade drive signal to compensate for errors. The integration in their model is performed by updating the locus of activity in the cerebellar cortex, giving the appearance of a wave of activity that spreads from the contralateral to the ipsilateral side during a saccade. This causes the difference in latency of the ipsi- and contralateral FNN observed during eye and gaze saccades (Optican, 2005). No evidence for this wave has been found yet, but the wave idea is not central to the cerebellar hypothesis. Any mechanism, such as a population of neurons with varying thresholds, that can perform the integration with an appropriate delay on both sides of the midline would be sufficient.
A model of the latch circuit
Having described features of models for PBN, OPN, SC and cerebellum, we are now in a position to return to the latch circuit for saccades. A schematic diagram of a proposed latch circuit model (Rucker et al., 2005) for a rightward saccade is shown in Fig. 4C (the model is symmetric, but many elements and pathways have not been drawn to emphasize crossover of activity from one side to the other during the movement). This new proposal differs from previous latch circuits because it is bistable, requiring both a set and reset command. The set command turns on reciprocal excitation between a subset of SC and cMRF neurons, and the reset command breaks that positive-feedback loop. We thus need two types of neurons in the SC, one which serves as part of the latch circuit, and one that sends information about the selected target to the brainstem. In Fig. 4C, the ‘When’ and ‘Where’ cells are two hypothetical sub-classes of SCBN. They correspond to two groups of SC cells, one that loses activity (blue versus red trace for ‘When’ cells), and one that does not (red and blue trace overlap for ‘Where’ cells), after OPN lesions (Soetedjo et al., 2002). Correspondingly, cMRF cells can be divided into two groups, one that receives ‘When’ cell activity and participates in the latch (which may be the cMRF cells related to saccade duration, presumably those having a low background rate), and one that relays target information to the brainstem, which may be the cMRF cells related to saccade amplitude and/or velocity, presumably those having a high background rate (Waitzman et al., 1996; Cromer and Waitzman, 2006). The SC fix cells and the high background cMRF cells excite the OPNs, keeping them on between saccades.
In this model, the rapid offset of the OPN before a saccade is caused by a burst in the EBN-like latch cell (Keller and Missal, 2003