Eye Movements: Saccades, Pursuit, and the Circuits That Aim the Fovea

01Foveation and acuity

Saccades and pursuit exist to defend a patch of retina about 1.25 degrees wide

Saccades and smooth pursuit both exist to hold a target on the fovea, the only patch of retina that resolves fine detail. In human donor retinas, peak foveal cone density averages 199,000 cones per square millimeter and ranges from 100,000 to 324,000 between individuals Curcio 1990. Density falls by roughly an order of magnitude one millimeter away from the center, and the rod-free zone measures 0.350 mm, about 1.25 degrees. A target that drifts off that zone blurs.

Two voluntary conjugate movements defend that patch. Saccades reposition gaze in rapid jumps between targets. Smooth pursuit tracks motion, keeping the image of a moving object steady on the retina.

Aim alone does not buy the acuity

Human observers gain more than 0.15 logMAR from fixational behavior with the fovea already on target, which Intoy and Rucci report as at least two lines of the Snellen chart Intoy 2020. Gaze was localized precisely and retinal stimulation was held under experimental control. The gain comes from tuning motion, because observers adapt both microsaccades and ocular drift to place the image on the receptors and to set how fast it travels across them.

Peak cone density varies threefold across healthy human eyes Curcio 1990. The same half degree of gaze error therefore costs one person more acuity than another.

02Findings

What the research shows

03The saccade generator

Burst neurons in the pons set both the distance and the speed of a horizontal saccade

A complex of premotor neurons called the saccade generator produces horizontal saccades. Four nodes make it up. The paramedian pontine reticular formation sits rostral to the abducens nucleus and holds the burst neurons themselves. The other three are the contralateral medullary reticular formation caudal and ventral to it, the bilateral nucleus prepositus hypoglossi, and the medial vestibular nucleus Zee 1986.

Spike count sets saccade amplitude and firing rate sets velocity

Excitatory burst neurons in the caudal pontine reticular formation of the alert squirrel monkey begin firing 5 to 15 ms before the saccade starts Strassman 1986. Three linear relations run through that discharge. Burst duration sets saccade duration, spike count sets amplitude, and instantaneous firing frequency sets instantaneous velocity. The same cells project to the ipsilateral abducens nucleus, the medial vestibular nucleus, and the nucleus prepositus hypoglossi, so one burst reaches the motoneurons and the gaze-holding circuit together.

A second population exists to shut the other eye's puller off. Single-unit recording caudal to the abducens nucleus in alert rhesus monkeys yielded 80 short-lead and long-lead burst neurons Scudder 1988. In the 33 analyzed for direction, the preferred direction lay 1 degree off horizontal with a standard deviation of 12 degrees. Short-lead cells made up 45 percent of the sample. Microstimulation at those recording sites produced potentials in the contralateral lateral rectus of the sign and latency of a monosynaptic inhibitory projection onto abducens motoneurons. One lateral rectus is driven in the same millisecond the opposite lateral rectus goes quiet.

Conjugacy is wired into the abducens nucleus

Internuclear neurons of the abducens nucleus send axons across the midline and up the medial longitudinal fasciculus to medial rectus motoneurons, so a single horizontal command reaches both eyes Zee 1986. The yoking is close and it is not exact. Human saccades peak at 502 degrees per second with a standard deviation of 32, for amplitudes of 40 degrees and above Collewijn 1988. They undershoot the target by about half a degree. The two eyes also diverge transiently by as much as 3 degrees in flight, because the abducting eye moves faster than the adducting eye. That divergence is corrected after the eyes land, which is the work of ocular alignment.

Vertical and torsional saccades come from a different nucleus. Their excitatory burst neurons sit in the rostral interstitial nucleus of the medial longitudinal fasciculus, and each projects to motoneurons so the two eyes stay yoked through a vertical movement Bhidayasiri 2000. That review also names the asymmetry. Each rostral interstitial nucleus carries up and down burst neurons but only ipsilateral torsional ones, so a unilateral lesion shows at the bedside as absent torsional quick phases toward one side while vertical saccades survive.

04The collicular motor map

The superior colliculus holds fixation and saccades at two ends of one map

The superior colliculus carries a motor map of saccade direction and amplitude. The rostral colliculus forms a fixation zone corresponding to the foveal region of the retina, holding neurons that discharge continuously during steady fixation. The caudal colliculus forms a saccadic zone, holding movement neurons that begin firing before the saccade itself. Stimulating the caudal colliculus evokes a saccade whose vector matches the stimulated site.

In the monkey, the active fixation zone spans the most rostral 0.72 mm of a 5 mm colliculus Munoz 1995. That is about 15 percent of the cells lying along the horizontal meridian of the buildup layer. The zone of active burst cells runs about 1.4 mm across, roughly 28 percent of the colliculus along a line drawn through it. The width of that burst zone does not change with saccade amplitude across 0.5 to 60 degrees, so a larger saccade moves the active population along the map instead of enlarging it. Buildup activity starts at least 100 ms before the movement, and fixation cells stop firing about 25 ms before it begins.

Fixation is a lock that has to be released

Raising fixation-cell activity with bicuculline in the rostral pole of the monkey colliculus delays saccades to visual and to remembered targets. Lowering it with muscimol releases express saccades, and the latency histogram peaks below 100 ms Munoz 1993. Stimulating the fixation zone during a movement interrupts the saccade in midflight at a latency as short as 12 ms.

The basal ganglia hold a tonic brake on the collicular map

The colliculus is governed in turn by the basal ganglia. Projection neurons of the substantia nigra pars reticulata release GABA onto collicular targets and hold a tonic nigrocollicular brake over the saccade map in the monkey Hikosaka 1985. That brake, and what lifting it does to movement generally, is set out in the basal ganglia.

In one patient with spasm of fixation, saccade latency ran 369 ms with a fixation target present, 197 ms with it absent, and 122 ms after a gap interval Johnston 1992. Johnston and colleagues postulated that hemispheric damage disinhibited the substantia nigra pars reticulata and so inhibited the colliculus. The latencies show fixation and saccade as two states of one gated circuit.

05Smooth pursuit pathways

Smooth pursuit converts retinal slip into matched eye velocity through two cortical streams

Retinal slip, the residual motion of the image across the retina, is the error signal pursuit works on. One stream descends through the dorsolateral pontine nuclei to the cerebellar flocculus and ventral paraflocculus Zee 1986. A second originates in the frontal eye field and passes through the nucleus reticularis tegmenti pontis. Both reach the output motor nuclei by way of the vestibular nucleus, which is also where head velocity signals arrive.

The pontine relay carries a pursuit command

Chemical lesions of the dorsolateral and lateral pontine nuclei in the monkey cut initial eye acceleration to less than half of normal May 1988. The deficit follows the direction of target motion rather than the visual hemifield the target sits in, which marks the relay as motor and not sensory. Lidocaine deficits cleared within 30 minutes and ibotenic acid deficits over 3 to 7 days, so pursuit machinery re-tunes around a loss on a timescale of days.

In the flocculus and ventral paraflocculus, gaze velocity Purkinje cells fire during pursuit and during smooth gaze tracking with head motion, and stay nearly silent during the vestibulo-ocular reflex. Among the gaze velocity cells in a sample of 187 horizontal eye-movement Purkinje cells recorded in squirrel monkeys, 76 percent preferred ipsilateral eye and head velocity Belton 2000. The head velocity signal appeared at a latency near 40 ms. Muscimol in that region cut pursuit velocity and the suppression of the reflex by roughly 50 percent together. One patch of cerebellum drives the tracking of a target and the canceling of a reflex, which is why pursuit signs and vestibular signs travel in company.

The firing pattern matches the mechanics it has to beat. Gaze velocity Purkinje cells in the monkey show a large transient rise at pursuit onset, a smaller sustained rise through maintenance, and a smooth return to baseline at offset with little undershoot Krauzlis 1994. Passing that population average through a computational model of the brainstem final common pathway and the oculomotor plant reproduced the measured eye velocity, so the transient compensates for the lagging mechanics of the eyeball.

The frontal stream sets pursuit gain

In macaque rostral nucleus reticularis tegmenti pontis, 78 percent of pursuit-related cells encode eye velocity and 22 percent encode eye acceleration Suzuki 2003. Velocity sensitivity averages 0.81 spikes per second per degree per second, and acceleration sensitivity 0.20 spikes per second per degree per second squared. Directional tuning is broad, with half-maximal responses still present 85 degrees away from the optimal direction, so a population reading sets velocity and one cell firing hard leaves the target's direction undetermined. Inactivating the monkey frontal pursuit area with muscimol lowers pursuit velocity and acceleration, and microstimulation there evokes pursuit outright, while the choice of which target to follow is untouched Mahaffy 2011. How fast the eye goes and which object it goes after are set in separate places, and this node holds only the speed.

06Fixational eye movements

The fixating eye never rests, and that motion is worth more than 0.15 logMAR of acuity

Fixation carries three movements: microsaccades, drift, and tremor. Microsaccades are the largest and fastest of them. Drift wanders slowly between microsaccades. Tremor, the smallest of the three, oscillates at high frequency on top of the drift.

Ocular microtremor has been measured without touching the eye. Laser speckle correlation metrology in 20 human subjects put the mean frequency at 78 Hz, with a standard deviation of 3.86 Hz Kenny 2014. Peak-to-peak amplitude averaged 21.42 microradians with a standard deviation of 7.01 microradians, about 4.4 arcseconds. Contact and imaging methods report larger figures, from about 11 to 60 arcseconds, because they sample different components of the motion. Adaptive optics scanning laser ophthalmoscopy in six human subjects places the tremor band at 50 to 100 Hz Bowers 2019. McCamy 2013 describes tremor as about one photoreceptor width and greater than 0.5 arcmin, with dominant frequencies of 70 to 103 Hz.

The tremor is brainstem output. Of 32 patients recorded with a piezoelectric probe resting on the anesthetized sclera, the 28 with confirmed brain stem death produced no ocular microtremor at all Bolger 1999. The smallest motion the eye makes stops when brainstem activity stops.

The perceptual work belongs to microsaccades and drift

Tremor moves the retinal image about one photoreceptor width, while a microsaccade shifts it across dozens to hundreds of photoreceptors. In simultaneous piezoelectric and video-oculography recordings in human observers, microsaccades counteracted perceptual fading and tremor did not McCamy 2013. Retinal image motion measured over the tremor band came to just over 5 arcseconds of standard deviation, rarely departing more than 1 arcsecond from the 1/f spectrum of ocular drift Bowers 2019. Those authors concluded that tremor is too small to shape the percept. Tremor is a real motion of the eyeball, and the image is kept alive by the larger movements riding above it.

The oculomotor system triggers microsaccades in part as a response to deviations from desired gaze position, and those corrective movements help high acuity tasks. Cancel fixational behavior in the laboratory and acuity drops by more than 0.15 logMAR Intoy 2020.

07The neural integrator

A separate circuit turns the velocity command into the position signal that holds the eye

Burst neurons deliver velocity, and the eye needs position. The neural integrator performs that conversion. It runs through the nucleus prepositus hypoglossi for horizontal holding and through the interstitial nucleus of Cajal for vertical and torsional holding.

Recordings from 100 units in the nucleus prepositus hypoglossi and adjacent medial vestibular nucleus of behaving macaques found a class of neurons responding during eye movements alone McFarland 1992. Of those, 73 percent discharged in a burst-tonic pattern resembling abducens motoneurons. The burst preceded the saccade by 7.6 ms with a standard deviation of 1.7 ms, and 86 percent of those eye-movement neurons had ipsilateral on-directions. Firing rate rose with eye position, which is the position signal itself. A third population in the same territory carried eye and head velocity for pursuit.

Silence the integrator and the eye slides back

A unilateral muscimol injection into the nucleus prepositus hypoglossi of the alert cat produced bilateral gaze-holding failure, with every saccade followed by a centripetal postsaccadic drift Mettens 1994. The eye still reached the target and could not stay on it. Pulse and step are separate operations run by separate cells.

Muscimol in the interstitial nucleus of Cajal of the alert monkey cut the time constant of gaze-holding decay to 330 to 370 ms for vertical and torsional positions Helmchen 1998. The vertical oculomotor range fell by up to 50 percent while saccade velocity and the main sequence stayed normal, which separates the integrator from the burst generator. Bilateral injections halved vertical and torsional vestibulo-ocular reflex gain, and unilateral injections left that gain intact despite severe holding failure. One integrator does not serve every eye movement equally.

The integrator is organized around behavior

Inactivating the interstitial nucleus of Cajal in four macaques produces exponential, position-dependent decay of vertical and torsional eye position Crawford 1994. The axis of that torsional drift tracked the normal to Listing's plane at r = 0.85, aligning with it to within 0.06 degrees on average. The same drift showed no such relation to the stereotaxic coordinates of the skull, at r = 0.10. The coordinate frame the integrator defends is the one behavior uses, and the skull's geometry does not set it.

08Cervical input and gaze

Cervical afferent input changes smooth pursuit gain with the head and the eyes untouched

Rotating the trunk under a stationary head lowers smooth pursuit gain in people with neck disorders, with the head, the eyes, and the inner ear left where they were. In two whiplash-associated disorder groups, 50 patients with dizziness and 25 without, neck torsion reduced pursuit gain at p below 0.001 Tjell 1998. Twenty patients with vertigo of central origin, 20 with Meniere's disease, and 30 healthy subjects showed no such drop. The smooth pursuit neck torsion test ran at 90 percent sensitivity and 91 percent specificity in the whiplash patients with dizziness, and at 56 percent sensitivity in those without it. The neck feeds the pursuit pathway specifically, and a general vestibular or central fault does not produce the same result.

Neck pain lowers pursuit gain without whiplash

Across 55 patients with neck pain, 11 of them with whiplash, and 20 healthy controls, pursuit gains were reduced and fell further under neck torsion Janssen 2015. The protocol used seven static trunk-to-head rotations from 45 degrees left to 45 degrees right. Healthy controls tracked predictable targets better than unpredictable ones while the neck pain patients tracked both alike, and the neck was identical across those conditions. Cervical traffic is one term in the pursuit computation and not the whole of it.

The neck torsion test is itself a cervical input, applied by a clinician and delivered without touching the eye. The Unified Model of Tone reads the cervical spine as one of the channels that sets pursuit gain, and the channel these two studies isolated is the one a clinician can load directly. Chiropractic care enters through the cervical afferents that feed pursuit, at a grade the clinician chooses instead of the fixed 45 degrees of the test. Pursuit gain is a number that can be watched moving while the neck is the only thing that changed.

Why only the dizzy half showed the deficit

One injury, one test, and the pursuit deficit showed up at 90 percent sensitivity in the whiplash patients who were dizzy while it missed almost half of those who were not, at 56 percent. Cervical traffic was reaching the pursuit pathway in both groups. What differed was the condition that traffic arrived into. The Unified Model of Tone reads that split as an input meeting a state. Eye velocity is assembled from retinal slip, neck afferents, cerebellar drive and expectation, and the neck term is only as loud as the rest of that assembly allows.

That reading is testable with the instrument these two studies already built. Neck torsion gives every patient an internal control, because gain is measured twice, once with the trunk square and once rotated under a stationary head. The model predicts that repeated cervical input narrows the difference between those two gains before it lifts the neutral reading. An input that loads the joint without changing the afferent traffic those segments send would do the reverse, moving the neutral reading and leaving the torsion difference where it was. The torsion difference is the neck's own contribution to eye velocity, and it is the number to watch.

Gain that can be varied is tone inside its healthy range. A bird crossing a window and a line of print demand different velocities within the same minute, and a healthy eye keeps the freedom to match either. Gain stuck low, or gain that will not settle, is tone pushed outside that range. It is met as the world lagging on a head turn, and as losing the line while reading.

09Tone

How this system expresses tone

Gaze is read to a fraction of a degree without laying a hand on the person. Tone is legible here in the position the eye holds, the velocity it matches, and the gain that relates the two.

A single trace of eye position carries the rest. Prediction: pursuit gain rises when target motion is predictable, so eye velocity is built partly from expectation and not from slip alone. Oscillation: ocular microtremor runs at a mean 78 Hz and never stops while the brainstem lives. Time course: the burst leads its saccade by 7.6 ms, the integrator decays over hundreds of milliseconds, and pursuit machinery re-tunes over 3 to 7 days after a pontine lesion in the monkey. Load: holding the eye still costs continuous firing in the rostral colliculus and in the integrator. Constraint: saccade velocity saturates near 502 degrees per second, a ceiling the muscles and motoneurons impose no matter what the map commands. Coupling: one patch of flocculus serves pursuit and the suppression of the vestibulo-ocular reflex, and inactivating it halves both at once.

10Across the library

How this page relates to the rest of the library

11Frequently asked

Questions about this topic

12The sources

References

Sources: primary literature, linked inline.