Saccade-Related Neurons in the Primate Fastigial Nucleus: What Do They Encode?

J. F. Kleine, Y. Guan, U. Büttner


The cerebellar fastigial oculomotor region (FOR) and the overlying oculomotor vermis (OV) are involved in the control of saccadic eye movements, but nature and function of their saccade-related neuronal signals are not fully understood. There is controversy in at least two major aspects: first, lesion studies in OV/FOR reported eye-position-dependent dysmetria—with FOR lesions, centripetal saccades became more hypermetric than centrifugal saccades—suggesting that the cerebellum may compensate for orbital mechanics. However, single-unit studies failed to reveal corresponding eye-position dependencies in FOR saccade-related discharge patterns. Second, some single-unit studies reported precise correlation between burst and saccade duration in the FOR. However, others stated that FOR bursts were only weakly related to saccade properties. In an attempt to resolve these discrepancies, we recorded single FOR units in monkeys that made horizontal saccades (16°) from different starting positions. Sampling saccades of one fixed amplitude and application of an objective, computer-based burst-detection-routine allowed us to correlate burst parameters (onset latency, peak latency, peak amplitude, number of spikes, duration) and kinematic properties of individual saccades. FOR bursts were found to start and peak earlier and exhibit higher peak burst amplitudes for faster than for slower saccades of the same amplitude. While these correlations between FOR bursts and saccade properties were statistically significant for a minority of ∼20–25% of individual units, the same effects were also predominant in the remainder of the neuronal sample and statistically significant on the population level. Neuronal activity was not significantly modulated by eye position itself. However, reflecting differences in saccade velocities but not an actual influence of eye position per se, FOR bursts for centripetal and centrifugal saccades exhibited subtle but systematic differences, which closely paralleled, and hence probably explain, the eye-position dependency of deficits observed after FOR inactivation. Our findings indicate that FOR signals reflect much of the kinematic properties of the saccade. Moreover, they are consistent with the idea that the FOR output is purposefully modified according to these kinematic properties to maintain saccadic accuracy.


Clinical, anatomical, and electrophysiological data show that the cerebellum plays a fundamental role in the control of saccadic eye movements. Although lesions to cerebellar midline structures do not destroy the basic ability to generate saccades, they may lead to loss of saccadic accuracy by inducing saccadic step size dysmetria (Leigh and Zee 1999; Optican and Robinson 1980). The caudal part of the fastigial nucleus (cFN), the most medial of the deep cerebellar nuclei, and the overlying part of the cerebellar vermis have been identified to be the anatomical substrates of this effect. Electrical stimulation of vermal lobuli VI and VII evokes saccades at low thresholds (Noda and Fujikado 1987a), and neurons in this area, consequently labeled the “oculomotor vermis” (OV) (Noda and Fujikado 1997b), are modulated with saccadic eye movements (Helmchen and Büttner 1995; McElligott and Keller 1982; Ohtsuka and Noda 1995). The Purkinje cell (PC) output of the OV is relayed almost exclusively to a circumscribed region in the cFN (Yamada and Noda 1987), which itself contains saccade-related neurons (Fuchs et al. 1993; Helmchen et al. 1994; Ohtsuka and Noda 1991), and has been designated the “fastigial oculomotor region” (FOR) (Yamada and Noda 1987). The FOR, in turn, projects to various eye-movement-related structures in the brain stem, notably the immediate premotor structures for horizontal and vertical saccade generation, i.e., the paramedian pontine reticular formation (PPRF) and the rostral interstitial nucleus of the MLF (riMLF) (Noda et al. 1990).

What the cerebellum precisely contributes to make saccades more accurate is unclear. There are at least two major unresolved problems: first, there are profound discrepancies between the findings of lesion studies on OV and FOR on the one hand and of single-unit recordings on the other hand. Second, there is disagreement about the essence of the signal generated by these cerebellar areas, which has not yet been fully understood.

Early investigators, who made extensive, bilateral lesions to the cerebellar vermis, which mostly also affected the deep cerebellar nuclei, observed that the induced saccadic dysmetria was influenced by the initial position of the eye in the orbita: “centripetal” saccades (from the periphery toward the center of the oculomotor range) became more hypermetric than “centrifugal” saccades of the same amplitude and direction (Optican and Robinson 1980; Ritchie 1976). Qualitatively similar differences between centripetal and centrifugal saccades were also reported by inactivation studies specifically targeting the FOR (Robinson et al. 1993; Vilis and Hore 1981). However, the derived concept (Ritchie 1976) that cerebellar signals serve to balance the eye plant's passive viscoelastic forces, which act in synergy with centripetal and counteract centrifugal eye movements, was not supported by unit recording studies from the FOR. While for OV PCs notable eye-position-dependent changes in discharge patterns had been reported (McElligott and Keller 1982), single-unit studies on the FOR consistently found no or only negligible influence of eye position on saccade-related discharges (Fuchs et al. 1993; Ohtsuka and Noda 1991; Ohtsuka et al. 1994) and hence do not explain the centripetal/centrifugal asymmetries seen after OV or FOR lesions (Robinson and Fuchs 2001).

These studies, however, opened another perspective on FOR saccade-related bursts by revealing direction-dependent latency differences: FOR neurons were found to be active during all saccades, but bursts occurred earlier for contralateral than for ipsilateral saccades (Fuchs et al. 1993; Helmchen et al. 1994; Ohtsuka and Noda 1991). Ohtsuka and Noda, in addition, found fairly close correlations between burst duration and saccade duration for contralateral saccades. They concluded that the cerebellum provided accurately timed signals that effectively encode saccade amplitude, thus transforming spatial into temporal information, which serves to control saccade metrics (Ohtsuka and Noda 1991). Fuchs and colleagues, in contrast, who presented otherwise fairly similar data, found the correspondence between burst and saccade duration much less precise (Fuchs et al. 1993). Measuring various burst parameters, including burst duration, number of spikes per burst, and burst peak amplitude, these authors emphasized, contrary to Ohtsuka and Noda's proposal, that “discharge properties of fastigial saccade neurons” were “generally only weakly related to saccade metrics.” They suggested a more general role for the cerebellum according to which the early and late FOR burst are “involved in helping to accelerate contralateral saccades and in helping to decelerate ipsilateral ones” (Fuchs et al. 1993). Both conceptual schemes could in principle account for the basic effect seen after unilateral FOR inactivation, which consists in making contralateral saccades hypometric and ipsilateral saccades hypermetric (Robinson et al. 1993).

What then does the fastigial oculomotor region encode? Do FOR bursts specifically reflect the temporal properties of the saccade? Or do they provide an auxiliary accelerative and decelerative drive that is required to bring the eye close to the target but has otherwise little to do with saccade properties? How do FOR discharge patterns relate to the eye-position-dependent saccadic dysmetria seen after cerebellar lesions?

To answer these questions, we recorded saccade-related neuronal activity in the FOR of alert monkeys while they made horizontal saccades to a 3 × 3 grid of target positions. We found the temporal properties and also the peak discharge rates of saccade-related FOR signals to change systematically with saccade velocity. Furthermore, this saccade-velocity sensitivity of FOR bursts relates to previously unrecognized systematic differences between bursts for centripetal and centrifugal saccades, as a possible basis of the eye-position dependency of deficits seen after FOR lesions.


Two monkeys (Macaca mulatta) were prepared for chronic single-unit recordings. Prior to surgery, they were accustomed to the laboratory environment and trained to sit in a primate chair. Under general anesthesia and aseptic conditions, a head holder and a recording chamber were attached to the skull (see Boyle et al. 1985 for details). The recording chamber was positioned 7.0 mm posterior and 0.0 mm mediolateral, allowing a vertical approach to the left and right FOR in the stereotactic plane. Horizontal and vertical eye positions were recorded with the search-coil method (Robinson 1963) using previously described calibration routines (Bartl et al. 1996). Single-unit activity was recorded by means of varnished tungsten microelectrodes and conventional amplification techniques. After a suitable training period, the recording sessions began in which the monkeys, while sitting upright in a primate chair with the head fixed, followed a small light spot (1°) displayed by a video monitor. Target positions were arranged on a 3 × 3 square grid (16° distance between neighboring points), centered around the straight ahead position (Fig. 1,Fig. 1). Targets jumped horizontally between neighboring positions at unpredictable time intervals (1.3–2.2 s) to induce horizontal saccades ipsi- and contralaterally to the recording site as well as centripetally and -fugally at different vertical starting positions. All experimental and surgical procedures were in accord with the American Physiological Society guiding principles in the care and use of animals and approved by the responsible governmental institution.

fig. 1.

Activity of a saccade-related neuron in the fastigial oculomotor region (FOR) during ipsilateral (A) and contralateral (B) 16° horizontal saccades. The different horizontal and vertical starting positions are marked by dots in the insets on the right of each panel, which show “screen views,” with horizontal and vertical eye position plotted as 2-dimensional traces for each saccade. On the left, each panel shows (from above) horizontal eye velocity (gray lines: individual trials; black line: average), eye position, raster diagrams with spike occurrences for each saccade, and averaged neuronal acitivity as spike histogram. Traces are aligned with saccade onset, as indicated by the long vertical lines in each panel. “Burst spikes,” as identified by the described computer algorithm (see appendix), are highlighted by circles in the raster diagrams. For clarity, each raster diagram is restricted to a maximum of 30 trials, which were randomly selected from the total number actually sampled at each starting position, as indicated on top of the “screen views”. Ipsilateral bursts (A) start and peak later than contralateral bursts (B) and have, in this case, higher peak burst activity. Note the considerable variability of the discharge patterns even for saccades of the same amplitude and starting position.

fig. 1.

Activity of a saccade-related neuron in the fastigial oculomotor region (FOR) during ipsilateral (A) and contralateral (B) 16° horizontal saccades. The different horizontal and vertical starting positions are marked by dots in the insets on the right of each panel, which show “screen views,” with horizontal and vertical eye position plotted as 2-dimensional traces for each saccade. On the left, each panel shows (from above) horizontal eye velocity (gray lines: individual trials; black line: average), eye position, raster diagrams with spike occurrences for each saccade, and averaged neuronal acitivity as spike histogram. Traces are aligned with saccade onset, as indicated by the long vertical lines in each panel. “Burst spikes,” as identified by the described computer algorithm (see appendix), are highlighted by circles in the raster diagrams. For clarity, each raster diagram is restricted to a maximum of 30 trials, which were randomly selected from the total number actually sampled at each starting position, as indicated on top of the “screen views”. Ipsilateral bursts (A) start and peak later than contralateral bursts (B) and have, in this case, higher peak burst activity. Note the considerable variability of the discharge patterns even for saccades of the same amplitude and starting position.

Data analysis

The microelectrode signal passed through a Schmitt trigger generating standard pulses for each neuronal discharge that were recorded with a temporal resolution of 20 μs. Eye and target position were digitally sampled at 1,000 Hz and stored on computer hard disk along with the neuronal signal for off-line analysis. Eye-movement signals were digitally low-pass filtered (200 Hz) and screened for saccades with a computer algorithm based on combined acceleration- and velocity-threshold criteria. Only saccades that were on target, had the appropriate size, direction, and amplitude, were reasonably fast (>200–800°/s peak velocity) and started within a limited time window after the target jump (80–500 ms) were included for further analysis. To evaluate eye-position influences, the velocity criterion was increased to a minimum peak velocity of 300°/s because saccades of different starting positions were sometimes inhomogeneously distributed during periods of reduced alertness, which might have interfered with the statistical analyses.

Saccades and corresponding neuronal discharges were sorted according to direction, horizontal, and/or vertical starting position and peak velocity and further analyzed by self-developed computer routines written in MatLab (The MathWorks). Neuronal responses were evaluated by computing perisaccadic spike histograms (bin width: 5 ms). In units that exhibited saccade related bursts, burst on- and offset as well as peak burst amplitude and peak latency (re saccade onset) were automatically determined from these histograms, which were smoothed by a (phase neutral) Gaussian low-pass filter (cut off-frequency: 50 Hz) to reduce random errors in the determination of peak latency and peak burst amplitude. The last (first) bin before (after) the time when the neuronal activity exceeded (fell below) the mean +1 SD of the background rate (determined from the spike histogram in a time interval of 500 to 250 ms prior to saccade onset), was defined as burst on- and offset, respectively. The large “safety margin” relative to saccade onset was chosen for the computation of background rates because long-lead (»100 ms) activity decreases were observed in several units (see following text).

Although the perisaccadic spike histogram certainly is a valid representation of the neuronal activity, it does not provide information about the variance of burst parameters, which is required for statistical comparison of different groups of saccades (e.g., saccades from different starting positions, saccades with different velocities). We therefore applied a computer routine, a modified version of the “Poisson spike train analysis” originally described by Hanes et al. (1995) and recently applied by Thier et al. (2000) to Purkinje cell discharges in OV, to objectively determine burst parameters (number of spikes/burst, burst latency, burst duration, peak burst amplitude, peak latency) in individual trials (see appendix). This computer routine performed well in the large majority of bursting units (see following text, and, for representative examples, Figs. 1,1 2, 4, 6, and 9). The remainder, ∼20% of neurons, in which the quality of the burst detection suffered more from the large variability of the spontaneous and saccade-related discharge patterns, leading to a larger proportion of missed or, according to the investigator's subjective judgement, misdetected bursts, was nevertheless included to keep the analysis as bias-free as possible.

fig. 2.

Example to illustrate the performance of the computer-based analysis of individual bursts. See Fig. 1,Fig. 1 for a general description of the figure layout. The spike histogram is constructed from the responses to all ipsilateral saccades available for this sample neuron (n = 112), the extended raster diagram shows all individual responses. ▵, mean burst onset, peak time, and burst offset, obtained by averaging the values from the individual trials (parameters from “spurious” bursts were discarded, see appendix); ▾, the corresponding values obtained from the spike histogram. Burst onset is later and burst offset earlier when determined from the parameter average from individual bursts (see text). Note the pronounced activity decrease preceding the saccade.

fig. 4.

Sample neuron exhibiting a pronounced long-lead activity decrease prior to ipsilateral saccades, which commences at >100 ms before saccade onset. Panel layout as in Fig. 1,Fig. 1, except that neuronal activity is averaged across all starting positions.

fig. 6.

Comparison of burst activity for slow (left, 200–400°/s) and fast (right, 500–800°/s) saccades for 1 sample neuron. Panel layout as in other figures. For both ipsi- and contralateral saccades, bursts start and peak earlier and peak burst activity is higher for fast as compared with slow saccades.

fig. 9.

Burst activity of an FOR neuron during centrifugal (A) and centripetal (B) ipsilateral saccades. Panel layout as in preceding figures. Burst and peak latency are later and peak discharge rate is higher for centrifugal than for centripetal saccades.

As the subsequent statistical analyses depended strongly on our computerized burst detection routine, it needed to be validated. To this end, the averages of the burst parameters from the individual bursts for all trials that were available for a given cell were compared with the corresponding parameter obtained by evaluation of the perisaccadic spike histogram computed from the same data set. Not surprisingly in view of the nonidentical definitions and statistical criteria, there were systematic differences. Typically, burst onset was earlier, and burst offset later (burst duration, hence, longer) when determined from the perisaccadic spike histogram than when derived from single-trial analysis (Figs. 2 and 3), an effect that can be attributed to the greater sensitivity of the spike histogram (which constitutes an averaged and, therefore less noisy representation of the neuronal response in comparison with a single-spike train) to detect a significant alteration in discharge rate. In addition, peak burst amplitude was generally higher when determined from single-trial analysis (Fig. 3), indicating a lesser degree of smoothing implicit in this measure in comparison with the spike histogram representation. The amount of smoothing, which mainly depended on the bin width of the histogram and on the number of consecutive intervals averaged for burst peak determination (see appendix), respectively, had to be chosen, for each measure, to reach the respective best trade-off between noise reduction and information preservation. The systematic differences between the two analytical approaches are therefore plausibly explicable and do not disqualify the one or the other method. Indeed, the mean burst parameters obtained from individual burst analysis and the corresponding parameters from the perisaccadic spike histograms were tightly correlated (Fig. 3), indicating that either method provides reasonable quantitative measures of neuronal discharge properties.

fig. 3.

Comparison of histogram (abscissa) and individual burst (ordinate) analysis for burst onset latency (A), burst peak latency (B), and peak burst amplitude (C). ○, data from ipsilateral saccades from one neuron; •, data from contralateral saccades. For each ○ and •, individual burst data represent the average of the burst parameters determined for the individual bursts of the respective cell (“spurious” bursts were discarded, see appendix). —, slope of 1. There is a good match for peak latency, whereas values for burst latencies are systematically later and for peak burst amplitudes systematically higher for individual burst analysis (see text).

Localization of neurons

At selected recording sites, tracer substances 1-1′-dioctadecyl-3,3,3′,3′-tetramethylindocarbocyanine perchlorate (DiI) (Snodderly and Gur 1995) were placed to aid the reconstruction of electrode tracks. At the end of all experiments, one monkey was deeply anesthetized with barbiturate and perfused transcardially with 10% formalin. The brain was removed and blocked in the stereotaxic plane. Coronal sections taken every 50 μm were processed for the tract-tracing substance and counterstained with cresyl violet to confirm that the recording sites were confined to the caudal part of the fastigial nucleus.

As the second monkey was still alive during the composition of the manuscript, a detailed histological reconstruction of the recording sites is not yet available for this animal. However, on the basis of our previous, extensive experience in recording from the FOR, the recording patterns in surrounding cerebellar structures, and the stereotaxic coordinates and the discharge patterns of the saccade-related neurons, which correspond very closely to those reported in previous FOR single-unit studies (Fuchs et al. 1993; Helmchen et al. 1994; Ohtsuka and Noda 1991) (see following text), we believe that neurons were also recorded in the FOR in this monkey. Rostral to the FOR, in the rostral fastigial nucleus (rostral FN), “vestibular-only” neurons (Büttner et al. 1991; Siebold et al. 1999) were recorded. Dorsal to the rostral FN, electrodes clearly passed first through PC layers and then through several millimeters of white matter, where typically no neurons were recorded. More caudally, the white matter above the FOR was reduced in depth, and PCs below the FOR were encountered (Büttner et al. 1991). FOR saccade-related neurons were clustered in areas of 2 mm in diameter on the left and right sides, separated by 2–3 mm across the midline.


General characteristics

Data from 75 neurons from the left and right FOR of two monkeys are included in the present study. No preselections were made. Thus regardless of the specific characteristics and the prominence of the saccade-related discharge patterns, all units showing a discernible change in their neuronal activity during saccades were included, provided that they were well isolated and located within the anatomical borders of the FN. All neurons were spontaneously active [45.4 ± 24.2 (SD) imp/s, range: 8.4–104.3 imp/s). The spontaneous neuronal activity was quite irregular as noted previously (Fuchs et al. 1993; Helmchen et al. 1994). The saccade-related activity patterns, summarized in Table 1, were also highly variable. One neuron exhibited a saccade-related activity decrease during ipsilateral and no modulation during contralateral saccades. All other neurons (74/75) showed a burst of activity with saccades. Fifty-two neurons exhibited bursts for saccades in either direction (ipsi-/contralateral; e.g., Fig. 1,Fig. 1). Eighteen neurons showed a burst for ipsilateral saccades only and 4 for contralateral saccades only. The lack of a burst in certain saccade directions has been described previously (Ohtsuka and Noda 1991). The “ipsilaterally only” bursting neurons mostly exhibited, during contralateral saccades, no discernable activity change, whereas the “contralaterally only” bursting units showed, for ipsilateral saccades, a clear decrease in activity or pause [Table 1, pause (p)]. Activity decreases also occurred before ipsilateral bursts [pause-burst (pb), pause-burst-pause (pbp)]. Not infrequently, these activity decreases (p, pb, pbp) commenced early before the saccade (Figs. 2 and 3), at latencies exceeding 100 ms in ∼25% of cases (e.g., Fig. 4). Activity decreases following a burst, as part of a burst-pause or a complex pbp sequence (bp, pbp), were rarely observed during ipsilateral saccades but not uncommon for contralateral saccades. Obviously, the respective discharge patterns for ipsi- and contralateral saccades differed in most neurons. The most frequent combinations of discharge patterns were a pb sequence for ipsilateral and a “pure” burst (b; n = 23), no activity change (n = 10), or a pbp sequence (n = 8) for contralateral saccades.

View this table:
table 1.

Frequencies of occurrence of different discharge patterns for ipsi- and contralateral saccades

Ipsi- versus contralateral saccades

Overall the timing of the saccade-related bursts was quite variable. However, in general, bursts started and peaked earlier for contra- than for ipsilateral saccades (Fig. 1,Fig. 1). For all ipsi- and contralateral bursts, the mean burst latencies were determined from the perisaccadic spike histograms to be –3.0 ± 18.8 (SD) ms (ipsilateral; negative signs indicate a lead relative to saccade onset) and –11.1 ± 19.8 ms (contralateral). The difference in burst latencies was statistically significant (Wilcoxon, P < 0.05). Similarly, peak activity was reached significantly earlier in contralateral than in ipsilateral bursts (ipsilateral: 29.2 ± 19.9 ms; contralateral: 24.4 ± 22.0 ms, P < 0.05, Wilcoxon). The sample neuron of Fig. 1,Fig. 1 illustrates these effects. For this neuron, the contralateral and particularly the ipsilateral bursts occurred relatively late, resulting in fairly pronounced latency differences [overall averages across all starting positions were for burst latencies: 17.5 ms (ipsilateral) and –2.5 ms (contralateral); for peak latencies: 52.5 ms (ipsilateral) and 22.5 ms (contralateral)]. Burst and peak latencies of bilaterally bursting units were statistically indistinguishable from those of unilaterally bursting units. [Burst latencies: bilateral units, –3.3 ± 20.4 ms (ipsilateral), 9.6 ± 19.4 ms (contralateral); unilateral units, –2.0 ± 13.7 ms (ipsilateral only), –30.0 ± 16.6 ms (contralateral only). Peak latencies: bilateral units, 30.2 ± 20.7 ms (ipsilateral), 25.4 ± 22.5 ms (contralateral); unilateral units, 26.4 ± 17.6 ms (ipsilateral only), 10.0 ± 5.0 ms (contralateral only)]. Although both burst onset and burst peak may appear to occur considerably earlier in the latter group than the corresponding contralateral burst and peak latencies in bilaterally bursting neurons, these differences were statistically not significant given the variability of the data and the small number of neurons in this subgroup (n = 4, see Table 1).

Finally, average peak burst amplitude and burst duration, as determined from the spike histograms, were similar for ipsi- and contralateral saccades and did not differ between unilaterally and bilaterally bursting neurons (peak burst amplitude: ipsilateral 98.2 ± 42.2 imp/s, contralateral 102.4 ± 48.6 imp/s; burst duration: ipsilateral 89.1 ± 38.6 ms, contralateral 86.9 ± 42.5 ms). Note that the numerical values assigned to peak discharge rates are substantially influenced by the particular method used to measure them. When calculated according to the criteria applied in individual burst analysis, the peak rates obtained were considerably higher (see methods).

Relation between burst parameters and saccade kinematics

Although the amplitude of the eye movements was restricted to ∼16° (the target step amplitude), the velocity profiles of the saccades varied considerably for both animals with peak velocities extending across the entire permitted range of 200–800°/s [monkey 1: 422.7 ± 106.2°/s (mean ± SD), n = 8,170 saccades; monkey 2: 423.3 ± 117.1 imp/s, n = 5,883 saccades]. This variability, which was presumably largely due to fluctuations in the alertness level of the monkey during the recordings, enabled us to compare saccade properties and burst parameters for individual trials. To this end, we computed correlation coefficients for the relation between saccade peak velocities and durations and the various burst parameters derived from the computer-based burst-detection algorithm, including values between the 2.5th and 97.5th percentile to remove possible outliers, and evaluated them for statistical significance (α = 5%, Bonferoni-corrected for multiple comparisons). According to this procedure, significant correlations between saccade and burst properties were detected in the ipsilateral bursts of 36/70 units and in the contralateral bursts of 17/56 units (51 and 30%). Figure 5 plots burst and peak latency and peak burst amplitude versus saccade peak velocity for the neuron of Fig. 1,Fig. 1, which exhibited impressive but not exceptionally strong correlations. It also illustrates the effects typically observed: with increasing saccade velocity, the peak burst amplitude grows, and the burst starts and peaks at shorter latencies.

fig. 5.

Relation of saccade peak velocity to burst latency, peak latency, and peak burst amplitude for the sample neuron of Fig. 1,Fig. 1. ○, 1 ipsilateral (left) or contralateral (right) saccade. Sample sizes were n = 461 (ipsilateral), and n = 372 (contralateral). Burst and peak latencies decrease and peak discharge rates increase systematically with saccade velocity. —, linear regression line with 95% confidence bands. Numerical values of the slopes of the regression lines (sl) and correlation coefficients (r) are given in each panel. All correlations were statistically highly significant (P < 10–6).

For both ipsi- and contralateral saccades, these effects are congruent, although they appear to be more pronounced for ipsilateral saccades. The same effects are clearly apparent in Fig. 6, which shows for another sample neuron the perisaccadic spike histograms for ipsilateral saccades from the lower (200–400°/s) and upper end (500–800°/s) of the velocity range. Strikingly, almost all, namely 34 of the 36 ipsi- and 16 of the 17 contralaterally bursting neurons, in which saccade velocity was significantly correlated with burst parameters according to the applied threshold (α = 5%, corrected), showed at least one of these three “typical” effects: positive correlation with peak burst amplitude, negative correlation with burst latency, negative correlation with peak latency (Fig. 7B). In many cases, as indicated in the overlapping circles of the intersection diagrams of Fig. 7B, these “typical” correlations occurred in combination. Thus there was clear predominance of a specific subset, or pattern, of correlations, which emerged even more clearly (see following text), when the individually nonsignificant correlation coefficients were included in these considerations.

fig. 7.

A: correlation coefficients of all burst neurons for the relation of saccade peak velocity to various burst parameters plotted against their associated P values (for clarity, P values <5 · 10–9 were set to this value). Negative correlation coefficients preponderate for burst and peak latency, positive coefficients for peak burst amplitude, and spikes/burst. The asymmetry of the distributions is evident even for those neurons, which did not by themselves fall into the range of statistical significance, indicated by the shaded areas [P < 0.05, corrected (darker shading) or uncorrected (lighter shading), for multiple comparisons]. B: pattern of correlations between saccade peak velocity and burst parameters (burst latency, peak latency, peak discharge rate, number of spikes/burst, burst duration). The intersection diagrams refer to those neurons, which exhibited at least 1 or a combination of the 3 “typical” correlations: positive (+) with peak burst amplitude, negative (–) with burst latency or peak latency. For instance, 12 neurons exhibited statistically significant positive correlation between saccade velocity and peak burst amplitude. In 8 of these, saccade velocity was also significantly correlated with peak latency, and in 5 of these, all 3 typical effects were present and statistically significant. Other effects only refers to neurons showing any other statistically significant correlation between saccade velocity and any burst parameter that did not, at the same time, exhibit a significant typical effect (see text). C: sample sizes. For each cell, the number of sampled saccades is plotted versus the P value of the correlation between saccade peak velocity and peak burst amplitude for comparison with the corresponding panel in A (as in A, P values lower than 5 · 10–9 were set to this value for clarity). Filled symbols indicate units exhibiting negative correlation.

The preceding evaluation, based on rigid statistical threshold values applied on data from single neurons, is rather conservative and certainly underrates the true relevance of the effects. This is indicated by several observations. Figure 7A shows, for all bursting neurons, the correlation coefficients for the relation between saccade peak velocity and the various burst parameters, plotted against their associated P values. Clearly, the statistically significant correlation coefficients were all negative for burst- and peak latency and almost all positive for peak burst amplitude. However, in addition, these asymmetries also held for those correlation coefficients, which, by themselves, did not reach statistical thresholds. For both ipsilateral and contralateral bursts, the leftward or rightward shifts of the distributions of correlation coefficients were highly significant for these three parameters (P < 0.001) even when all significant correlation coefficients (α = 0.05, corrected) were disregarded. Furthermore, when only the statistically significant correlation coefficients were considered, five neurons (ipsilateral) and two neurons (contralateral) combined all three typical effects: negative correlations with burst and peak latency and positive correlations with peak burst amplitude (Fig. 7B). When the individually nonsignificant coefficients were also considered, the proportion of neurons exhibiting this specific combination rose dramatically. If the correlation coefficients and their combinations were all random, the expected value would be 2–3 = ⅛ of the population, corresponding to ∼9/70 (ipsilateral) and 7/56 (contralateral) units, respectively. The observed values were 43 (ipsilateral) and 28 (contralateral). The probability that these numbers could be due to chance is practically zero. Finally, the slopes of the regression lines for statistically nonsignificant correlations were not infrequently as large or even larger than those of some of the statistically significant correlations as demonstrated in Fig. 8. This figure shows the regression lines for burst and peak latency and peak burst amplitude for every bursting neuron of our sample. Although the slopes of the “nonsignificant” regression lines, by definition, would have to be considered as statistically indistinguishable from zero for the individual neuron, the predominance of a set of typical correlations, which is apparent in Figs. 7 and 8 and cannot reasonably be attributed to chance (see preceding text), indicates that at least some of the nonsignificant neurons contribute saccade-velocity sensitivity to the FOR population. Indeed, for burst and peak latency and peak burst amplitude, the average slopes of the regression lines were significantly different from zero even for the nonsignificant neurons (Table 2). With respect to peak burst amplitude, the average nonsignificant slopes were ∼0.11 (imp/s) · (°/s)–1 (ipsi- and contralateral). For a neuron with a (typical) peak burst amplitude of ∼100 imp/s at the lower end of the velocity range (cf. Fig. 8), this sensitivity would amount to a considerable response increase of ∼50%, as saccade velocity rises from 200 to 700°/s. Analogously, the sensitivities of the significant neurons and of the entire sample would correspond to a 130 and a 70% increase, respectively (Table 2). Remarkably, the response increase with saccade velocity seems to be caused mainly by larger numbers of spikes per burst as indicated by the predominance of positive slopes for this parameter (Fig. 7; Table 2). Forty- two of the 43 units (ipsilateral) that combined all typical correlations with peak and burst latency and peak burst amplitude also showed a positive correlation with spikes/burst. In contrast, there was not much consistency in the relations between saccade velocity and burst duration.

fig. 8.

Linear regression lines for the relation between saccade peak velocity and burst latency, peak latency, and peak burst amplitude for all neurons. Bold lines represent units in which the individual correlation was statistically significant (P < 0.05, corrected). Regression lines for the sample neurons of Figs. 5 and 6 are highlighted by dark and light gray, respectively. See text for further explanation.

View this table:
table 2.

Mean slopes of the regression lines for the relation between saccade peak velocity and burst parameters

These findings show that statistical nonsignificance of individual neurons cannot generally be attributed to quantitative irrelevance of their signals. Rather, the occurrence of notable, but nonsignificant eye-velocity sensitivities reflects the considerable “noisiness” of FOR responses and the extrinsic data variance introduced by the lack of an ideal method for quantification of burst properties. This latter factor has to be considered in spite of the overall reasonable performance of the applied burst detection algorithm (see methods). Sample-size limitations may be of additional relevance in some cases but generally did not seem to play a decisive role, as most sample sizes were actually quite large (Fig. 7C). Typically, correlations were based on 50 to ≥150 saccades.

Influence of eye position on saccade-related activity

The influence of initial eye position on the saccade-related bursts during 16° saccades was evaluated by statistical comparison (Kruskal-Wallis nonparametric ANOVA) of the burst parameter distributions derived from individual burst analysis at the different starting positions, sorted according to their horizontal or vertical components. Eye-position-related effects on burst discharges were not pronounced. Nevertheless, a distinct pattern emerged. In the following, it will be shown that there are relatively subtle but clearly recognizable and statistically significant differences between bursts for centripetal and centrifugal saccades. However, it will be argued and shown that these differences relate to systematic differences in the kinematic properties of centripetal and centrifugal saccades and do not reflect a direct effect of eye position itself.

centripetal versus centrifugal saccades. This part of the analysis was restricted to those units in which ≥10 centripetal and ≥10 centrifugal saccades were available. This held for 61 (ipsilateral) and, respectively, 42 neurons (contralateral; see Fig. 10B for detailed information about sample sizes). When a significance level of α = 0.05 (Bonferoni-corrected for multiple comparisons) was applied, 8/61 neurons were found to exhibit significant relations to the horizontal component of initial eye position during ipsilateral saccades: for six of these, bursts started and/or peaked earlier and/or showed higher peak burst amplitudes for centripetal than for centrifugal saccades. Figure 9 illustrates these typical effects and is also representative of the magnitude, or, rather, the subtlety, of the observed differences. For contralateral saccades, only 2/42 neurons exhibited statistically significant differences between centripetal and centrifugal bursts, a number that is within chance level for the statistical threshold applied.

fig. 10.

A: comparison of centripetal (cp) and centrifugal (cf) saccades for various burst parameters. Each symbol represents ipsilateral (left) or contralateral bursts of 1 neuron. The differences between the respective cp and cf parameter (cf subtracted from cp) are plotted vs. the P values of the associated nonparametric ANOVA. Signs were chosen to reveal the congruence between the influence of cp vs. cf. saccade direction (this figure) and of saccade velocity (Fig. 7A) on burst properties. Ipsilateral centripetal saccades tend to show shorter burst and peak latencies, higher peak burst amplitudes, and more spikes/burst. The asymmetry of these parameter distributions is, in each case, statistically significant. For contralateral saccades, no systematic effects can be recognized. See text for further explanation. B: sample sizes. For each cell, the number of centripetal saccades that were sampled is plotted vs. the number of centrifugal saccades.

The described pattern, in which influences of eye position become manifest mostly in shorter burst and peak latencies and higher peak burst amplitudes for centripetal as compared with centrifugal saccades, clearly resembles the asymmetries in the distributions of correlation coefficients (for saccade peak velocity) of these parameters (see preceding section). This similarity becomes more obvious with Fig. 10, which plots for each ipsilaterally or contralaterally bursting neuron and for each of the various burst parameters the difference of the respective parameter mean of centripetal and centrifugal saccades against the P value of the corresponding nonparametric ANOVA performed on the parameter distribution. There is no evidence for any consistent pattern in the plots for contralateral saccades. For ipsilateral saccades, however, there are recognizable asymmetries in the distributions of difference values, with a predominance of negative values for burst and peak latency and of positive values for peak burst amplitude and spikes/burst, which parallel the asymmetries observed in the corresponding plots of correlation coefficients (Fig. 7A). Although the clustering appears less pronounced, largely owing to the relative paucity of units exhibiting significant P values by themselves, the leftward and rightward shifts of the distributions were statistically significant for all four parameters (P < 0.05, t-test), even when the units exhibiting significant P values for the respective parameter were disregarded. The correspondence between saccade peak velocity-related and start-position-related effects is manifest not only in the congruence of Figs. 7A and 10 but also in individual neurons: all six units that exhibited typical and statistically significant differences between (ipsilateral) centripetal and centrifugal bursts showed the corresponding typical correlation with saccade peak velocity.

These strong parallels evidently suggested that the observed differences in burst discharges for centripetal and centrifugal saccades could be due not to an influence of saccade starting position per se but rather be causally related to systematic differences between centripetal versus centrifugal saccade kinematics. Indeed, comparison of saccade peak velocities revealed systematic differences between these groups: centripetal saccades were on the average faster than centrifugal saccades (Fig. 11). These differences were highly significant for both monkeys (monkey 1: centripetal: 462.1 ± 92.4°/s, centrifugal: 430.2 ± 85.0°/s, P « 10–5 (t-test); monkey 2: centripetal: 463.7 ± 102.3°/s, centrifugal: 439.1 ± 95.2°/s, P « 10–5). That centripetal saccades are on the average significantly faster than centrifugal saccades of the same amplitude is not a peculiarity of our data but has also been observed in humans (Pelisson and Prablanc 1988) and can be deduced from several previous studies on the cerebellar influences on saccadic eye movements (see discussion).

fig. 11.

Comparison of centripetal and centrifugal saccade peak velocities for ipsi- and contralateral saccades. Mean ± SE (box) and SD (error bars) are shown for both monkeys (1 and 2). Centripetal saccades are significantly faster than centrifugal saccades (P « 10–5).

Given this dependence of saccade velocity on saccade starting position, the observed differences between centripetal and centrifugal saccade-related discharges during ipsilateral saccades could obviously relate to the strong correlation between burst properties and saccade velocity described above. On the other hand, this correlation could also mask additional, truly eye-position-related effects. To control for the influence of saccade velocity, two approaches were taken: first, we recalculated the Kruskal-Wallis-ANOVA for all bursting units, restricting saccade peak velocities to mid-range values (400–600°/s). Second, we performed a multivariate analysis of covariance on the various burst parameters for centripetal and centrifugal saccades, taking into account saccade peak velocity as covariate, and compared it with the results of the corresponding (parametric) ANOVA (which disregards the covariate). With respect to the typical effects, both control measures had similar consequences, reducing the number of units exhibiting statistically significant differences between centripetal and centrifugal saccades to up to three and increasing the P values of the respective statistics (“controlled” as compared with “uncontrolled” condition) by at least one order of magnitude in those remaining. Neither control measure unmasked other specific differences between bursts for centripetal and centrifugal saccades which occurred with any consistency beyond that of chance.

vertical component. Applying analogous statistical procedures and thresholds as for the comparison of centripetal versus centrifugal saccades (Kruskal-Wallis ANOVA) on the burst parameter distributions at different vertical start positions did not reveal a consistent pattern of influence of initial vertical eye position on saccade-related discharges among individual neurons. Evaluation of the distributions of burst parameter means also did not yield evidence for any subtle monotonous or nonmonotonous trend or pattern in the neuronal sample.

Influence of eye position: background discharge

Tonic background activity at the nine different starting positions was taken from the saccade-free time interval 500 to 250 ms before saccade onset. These samples were evaluated for eye-position-related influences by means of multiple linear regression, which indicated statistically significant effects (α = 0.05) in a considerable proportion of neurons (31/76). The calculated regression coefficients could be quite high in these neurons, ranging from –0.7 to 1.0 (imp · s–1/°) (horizontal), and from –0.92 to 0.63 (imp · s–1/°) (vertical). However, both horizontal and vertical regression coefficients were fairly symmetrically distributed, with means statistically indistinguishable from zero (t-test). The combinations of horizontal and vertical regression coefficients were also quite evenly distributed and did not cluster in any specific direction so that the neurons could not be divided into clearly separable subgroups. This held for the entire neuronal sample as well as for the subset of cells in which the individual regression was statistically significant. The net effect of all units, calculated by multiple regression across all neurons and corrected for the different number of saccades obtained from each cell, was statistically nonsignificant.


One of the properties that make the deciphering of FOR signals difficult is the large variability of the discharge patterns in individual saccade-related neurons and across the neuronal population (Fuchs et al. 1993; Helmchen et al. 1994; Hepp et al. 1982; Ohtsuka and Noda 1991). As the source and the functional significance of this variability are unknown, it must currently be treated as “noise.” This noisiness not only blurs the relation of saccade to burst parameters in FOR units but also causes methodological problems in the quantitative evaluation of burst discharges, a fact acknowledged by previous investigators (Fuchs et al. 1993; Helmchen et al. 1994; Ohtsuka and Noda 1991).

To cope with these problems, we took three measures in the present study. First, we studied only fixed-amplitude horizontal saccades from a small number of different starting positions. This reduced the overall variance of the data set and avoided complications that may arise from the interrelations between saccade amplitude, velocity, and duration, all of which are linked by the main sequence. Second, we adopted a previously published computer-based burst-detection algorithm (Hanes et al. 1995) and modified it to suit the large variability in spontaneous and saccade-related FOR discharge patterns, providing objective, bias-free, and validated (see results), quantitative analysis of burst discharges. Third, instead of focusing on only the most prominent effects in single units, we also strove to recognize patterns of more subtle but recurrent correlations between bursts and saccades among different neurons. This approach let us identify a common pattern of burst/saccade correlations that is predominant within the FOR population. Moreover, it allowed us to uncover as yet unrecognized differences between bursts for centripetal and centrifugal saccades, which may reconcile the puzzling discrepancies between previous single unit and lesion studies. Before turning to these main findings of the present study, we will briefly consider the more general characteristics of FOR saccade-related activity.

General characteristics of FOR discharge patterns

The general properties of saccade-related discharge patterns observed in our sample agree well with those described in previous reports on FOR neurons (Fuchs et al. 1993; Helmchen et al. 1994; Ohtsuka and Noda 1991). The most salient feature of these patterns is a burst discharge of action potentials that commences and peaks earlier for contralateral than for ipsilateral saccades. In addition, bursts related to ipsilateral saccades are quite consistently preceded by an activity decrease or a pause. For contralateral saccades, activity decreases occur less frequently and, if there is any, mostly after the burst. In general, the present neuronal sample corresponds well with data described by earlier authors. Like Fuchs and colleagues, we also did not find notable differences in burst characteristics between those units showing only a burst and those showing also activity decreases or pauses. Moreover, there were no significant differences between bursts from neurons that burst for both ipsi- and contralateral saccades and those from neurons that burst only unilaterally. Thus there is no definitive evidence that neurons exhibiting these different types of discharge patterns belong to functionally distinct subgroups. Rather, we agree with Fuchs et al. (1993) who concluded that they “could functionally act as one population.”

Relation between saccade-related activity in FOR and in Purkinje cells of the oculomotor vermis

A previously unrecognized feature of FOR responses is the long-lead (>100 ms) gradual activity decrease that accompanies ipsilateral saccades in a considerable number of units. This novel observation may be due to the relatively large number of saccades we sampled per neuron because the activity decrease becomes evident only in averaged responses. More detailed analysis shows that the onset of the long-lead activity decrease is temporally linked to saccade onset and not to the target step (not shown). This indicates that it is indeed related to the oculomotor output and does not represent some sort of visual response.

This long-lead activity decrease is most likely related to signals conveyed by OV Purkinje cells. According to Ohtsuka and Noda's data (1995), Purkinje cells in OV typically show for ipsilateral saccades a gradual build-up of activity, which may commence at lead times >100 ms and peaks near saccade onset. These authors reported mean onset times for this “burst” of 29.3 ± 24.5 ms; however, this value almost certainly underestimates the actual lead time of the activity increase: in their study, as in their previous work on FOR saccade-related signals (Ohtsuka and Noda 1991), onset time was defined as the time when firing frequency, starting from the cell's baseline level, reached half the peak discharge rate, a criterion that by definition underestimates the true lead time (e.g., by 50%, if discharge rate increases linearly with time). This may be of little relevance if the increase in discharge rate is brisk and lasts only several milliseconds. However, the absolute error becomes increasingly severe if firing frequency rises more gradually, as in the several examples given by Ohtsuka and Noda (1995; see, for instance, their Figs. 2, 3, and 13, where the firing frequency clearly rises above baseline at ∼100 ms or earlier before ipsilateral saccades). Also Thier et al. (2000) reported an average lead of 67.9 ms for the saccade-related burst of the PC population, whereas Helmchen et al. (1994) found a mean onset time of 15.3 ms for long-lead bursts in OV during spontaneous saccades. Although definite data on the time course of PC responses are lacking, we suspect that the long-lead activity decrease observable in FOR units may well correspond to the gradual build-up of inhibitory signals, which occurs in at least some OV Purkinje cells. A decrease in some steady input from excitatory mossy fiber collaterals is unlikely because most mossy fibers ascending to OV do not exhibit tonic activity (Ohtsuka and Noda 1992).

The long-lead activity increase in OV PCs and the corresponding decrease in FOR neurons are intriguing phenomena for two reasons. First, they reveal that oculomotor- related changes in neuronal activity are present in the cerebellar loop long before the actual eye movement, i.e., at latencies in the order of magnitude of saccade-related burst discharges in the frontal eye field (Seagraves 1992) or of activity changes in the so-called build-up neurons of the superior colliculus (e. g. Wurtz and Optican 1994). Onset times of long-lead activity changes recorded from mossy fibers projecting to OV range up to ∼80 ms (Ohtsuka and Noda 1992). Second, these signal changes provide direct evidence for a strong inhibitory influence of the cerebellar cortex on the FOR, raising the unresolved issue of the opposed driving force that makes FOR units burst. Ohtsuka and Noda (1991, 1995) proposed that this excitatory input to the FOR comes from mossy fiber collaterals ultimately projecting to OV, while the cerebellar cortex itself modulates and fine-tunes FOR output through its inhibitory PC signals. Our present data in principle agree with this concept, but more data are needed on the saccade-related signals in OV and its mossy and climbing fiber input to clarify the available, partly inconsistent information about their temporal characteristics, and to understand their dynamic interplay in the generation of cerebellar saccade-control signals.

Correlation between bursts and saccades

Previous studies disagreed on how well saccade properties correlate with FOR burst characteristics. Ohtsuka and Noda (1991) reported a fairly robust positive correlation between saccade and burst duration. Strongly emphasizing this finding and focusing on the temporal properties of FOR signals, they proposed that the cerebellum might participate in oculomotor control by providing “precise temporal information” about saccade duration and suggested that the cerebellum “is a domain where the spatial information could be transformed into temporal information” (Ohtsuka and Noda 1991). This is an attractive idea as it links FOR signals to an important global concept of the basis of cerebellar function, namely, measurement of time (e. g. Braitenberg 1961), Moreover, it seems to be supported by the recent finding of Thier and colleagues (2000) that the summed activity of PCs in OV constitutes a population signal that is strongly correlated with saccade duration. There is, however, a substantial problem with Ohtsuka and Noda's correlation analyses: suffering from the inherent noisiness of FOR signals, they were performed in only a subset (19/45) of the fraction (45/96) of neurons in which a simple computer-based burst detection routine produced homogeneous results. This probably unrepresentative subsample may have led to an overestimation of the robustness and the specificity of the reported correlations as legitimately criticized by Fuchs et al. (1993). These authors proposed that the contralateral and ipsilateral FOR bursts provide supplementary accelerative and, respectively, decelerative signals for the burst generator but emphasized that these signals were only “weakly” related to the properties of the saccade. By judging the “strength” of burst-saccade correlations by the numerical values of their correlation coefficients, however, without considering statistical significance levels, Fuchs et al. relied on a somewhat arbitrary criterion, as even a low correlation coefficient may of course indicate a functionally important signal in a large and noisy sample. Indeed, a correlation coefficient of 0.6–0.65, as reported by Fuchs et al. as the average value for the relationships between burst and saccade duration and between number of spikes per burst and saccade amplitude, would be statistically highly significant (P < 0.001) for sample sizes ≥27, a number that seems to be easily exceeded for most of their neurons (Fuchs et al. 1993). Thus although the lower correlation coefficients of Fuchs et al. (1993) are certainly more realistic than those reported by Ohtsuka and Noda (1991), their data seem to support the finding of notable and presumably significant, although not particularly precise, correlations between burst and saccade properties for saccades of different amplitudes.

Our data, which pertain to different burst parameters and are derived from a different experimental protocol (see following text), additionally show that the FOR contains signals that reflect the kinematic properties of saccades. Faster saccades are typically accompanied by bursts that start and peak earlier, have a larger number of spikes that tend to occur in a shorter period of time, and, consequently, show higher peak discharge rates than slower saccades of the same amplitude. These effects are statistically significant for a minority of individual units and are often not readily apparent among the considerable discharge variability of FOR responses. On the other hand, they are clearly not restricted to a specific small subset of cells, as the same pattern of correlations is also dominant, and on the group level statistically significant, for the remainder of the neuronal population. These signals are present in the FOR and principally extractable and usable by the subsequent processing stages in the saccade generating circuitry.

Contrary to Ohtsuka and Noda's proposal (Ohtsuka and Noda 1991), then, FOR signals do not seem to make use of a specifically temporal code. They do not only, in their shorter burst onset and peak latencies, reflect the more rapid development of the acceleration-deceleration sequence of faster saccades, they also reflect, in their larger firing frequencies, the increase in acceleratory drive and in braking force required to steer them. This corroborates the notion that the cerebellum helps to accelerate contralateral and to decelerate ipsilateral saccades by means of the early and late bursts, respectively, provided by FOR neurons (Fuchs et al. 1993; Robinson et al. 1993). However, it also suggests that the FOR may provide more than an unspecific supplementary input to the burst generator as these velocity-related signals may support functional mechanisms, by which the FOR could contribute to the control of saccade accuracy (see following text).

At first glance, our observation that burst discharge rates are correlated with saccade velocities seems to be at odds with previous studies on the FOR. However, closer inspection reveals that this observation may depend on differences in the experimental protocol and hence represent an independent novel finding. Irrespective of their discrepant view on the “precision” or “weakness” of the correlation between burst and saccade duration, both Ohtsuka and Noda (1991) and Fuchs et al. (1993) agreed that burst peak discharge rate was not significantly influenced by saccade size. As saccade velocity increases with size because of main sequence effects, this would seem to exclude a strong correlation between saccade velocity and peak discharge rate. However, these previous studies dealt with saccades of different amplitudes, whereas the present study sampled saccades of one fixed amplitude. The data of Helmchen et al. (1994), who compared spontaneously generated saccades in light and darkness, indicate that this difference may indeed be decisive. When these authors matched saccades of about the same amplitude, saccades in darkness were slower than in the light, and this effect was paralleled by a dramatic decline in burst amplitude (see Fig. 5 in Helmchen et al. 1994). In contrast, for sets of spontaneous saccades with a broad range of amplitudes, peak discharge rates and peak velocities were not correlated, and, consequently, these authors concluded that saccade velocity had no significant influence on peak discharge rate in FOR units. It may seem surprising that changes in saccade velocity should have different effects on neuronal discharge patterns depending on whether they are related to variations in saccade amplitude or caused by other factors, e.g., fluctuations in vigilance. However, it is not implausible that the underlying neuronal control mechanisms must be distinct. Indeed, the systematic changes in saccade velocity with size actually constitute the “main sequence,” whereas variations in saccade velocity at a given amplitude do the opposite: they blur it.

In this context, the findings of Takagi and colleagues (1998), who performed a detailed analysis of saccadic eye movements after OV lesions, are of interest. These authors made the unexpected observation that lesions to OV (which spared the deep cerebellar nuclei) not only induced saccadic dysmetria but also led to changes in saccade dynamics. Moreover, Takagi et al. (1998) showed that the effects of OV lesions on saccade metrics and on saccade dynamics could be dissociated, both in the severity of the induced alterations and in the time course of recovery, suggesting involvement of different neural control mechanisms. The present observation that FOR burst discharge rates can change systematically with saccade velocity without concomitantly affecting saccade amplitude, whereas changes in saccade amplitude—in spite of the correlated changes in saccade velocity—are not accompanied by prominent changes in burst discharge rate (Fuchs et al. 1993; Helmchen et al. 1994; Ohtsuka and Noda 1991), represents an analogous dissociation, suggesting that FOR signals may change in diverse ways to control different aspects of the saccade.

Cerebellum may compensate for orbital mechanics

An early hypothesis on the function of the oculomotor-related cerebellar midline areas (OV and FOR) was that they might provide signals that compensate for the elastic restoring forces of the orbital tissue that tend to keep the eye close to the central position (Ritchie 1976). This was based on the observation that (bilateral) lesions to these areas resulted in eye-position-dependent saccadic dysmetria: centripetal saccades became more hypermetric than centrifugal saccades for the same target step amplitude (Ritchie 1976). With the exception of Noda and colleagues, who found no difference between centripetal and centrifugal saccades after pharmacological inactivation of the FOR (Ohtsuka et al. 1994), these results were essentially confirmed by all subsequent studies investigating this aspect, in which the FOR was affected by lesions or temporally inactivated (Optican and Robinson 1980; Robinson et al. 1993; Vilis and Hore 1981).

Unit recording studies, in contrast, did not reveal a corresponding clear-cut eye position dependency in FOR saccade-related discharges: Ohtsuka et al. (1994) reported basically identical responses in FOR units for centripetal and centrifugal saccades, and Fuchs et al. (1993) found only weak influences of eye position on FOR discharge patterns, which could not, as they believed, explain the centripetal/centrifugal differences observed in their FOR-inactivation experiments (Robinson and Fuchs 2001; Robinson et al. 1993). These discrepancies between lesion experiments and previous unit recording studies are certainly among the most intriguing problems as regards the potential function of OV and FOR in saccade control (Robinson and Fuchs 2001).

Our data now reveal systematic differences in FOR discharge patterns for centripetal and centrifugal saccades that can account for the position-dependent saccadic dysmetria seen after cerebellar lesions. Bursts for centripetal saccades exhibit shorter onset and peak latencies and show more spikes per burst and therefore higher peak discharge rates. These differences are not large, so that only a few neurons exhibit statistically significant effects by themselves; however, the pattern is clearly present and statistically significant among the neuronal population. The pattern relates to the correlations between burst properties and saccade kinematics discussed in the preceding text, and—rather than being due to a direct influence of eye position itself—it reflects the fact that centripetal saccades are on the average faster than centrifugal saccades of the same amplitude. When we statistically controlled the influence of saccade velocity, the differences between bursts for centripetal and centrifugal saccades became negligible. As already mentioned, the difference between centripetal and centrifugal saccade velocity is not peculiar to our data. The same effect, which is very probably due to the viscoelastic forces arising in the orbita (which are acceleratory for centripetal and deceleratory for centrifugal eye movements) has also been demonstrated in humans (Pelisson and Prablanc 1988) and can be deduced from the study of Ritchie (1976), which indicates (see his Fig. 6) longer duration of centrifugal as compared with centripetal saccades of the same size. Similarly, in their lesion study on OV, Takagi et al. (1998) measured—in the still intact monkeys—significantly larger accelerations for centripetal as compared with otherwise identical centrifugal saccades. Although previous lesion studies (Optican and Robinson 1980; Ritchie 1976; Robinson et al. 1993; Vilis and Hore 1981) did not explicitly compare centripetal and centrifugal saccade velocities, one must assume that the same difference existed and influenced these experiments.

These systematic differences between centripetal and centrifugal saccade velocities are, of course, present for both ipsi- and contralateral saccades. However, the differences between centripetal and centrifugal FOR bursts were observed for ipsilateral saccades only. This relates to a corresponding asymmetry seen after FOR lesions: after unilateral FOR inactivation, ipsilateral centripetal saccades become distinctly more hypermetric than ipsilateral centrifugal saccades, whereas contralateral centripetal and centrifugal saccades become about equally hypometric (Robinson et al. 1993). These parallel asymmetries of lesion effects and of neuronal discharge patterns indicate that FOR signal changes are not merely passive but functionally relevant reflections of saccade kinematics. This holds, of course, not only for the temporal aspects of FOR saccade-related bursts but also for the observed systematic changes in firing frequency.

How then do the eye-position-dependent FOR discharge patterns relate to the eye-position-dependent hypermetria caused by cerebellar lesions? Stopping saccades precisely at some targeted location requires an adaptable braking signal that compensates for changes in saccade velocity. If the late ipsilateral bursts of FOR units contribute to this function, they should occur earlier and be stronger for the faster centripetal saccades. Our data show that they indeed behave in this way. Removing this braking signal will have stronger effects on faster than on slower saccades, leading to more severe hypermetry for centripetal than for centrifugal saccades as was observed after FOR lesions. The more the system depends on adaptive capabilities of the decelerative signal contributed by the FOR, the larger this effect will be. To use the everyday analogy of driving a car: driving downhill, one has to use the brakes more vigorously than when driving uphill, and a brake failure will have more deleterious consequences, due to the larger increase in braking distance. In this way, our data are consistent with the idea that FOR signals are purposefully modified according to the actual kinematic properties of the saccade in a functionally meaningful way that would help to maintain saccade accuracy.

Apparently, the early FOR burst during contralateral saccades is normally not being modified to compensate for orbital elastic forces and level the difference between centrifugal and centripetal saccade velocities. This observation has two further corollaries. First, in contrast to saccade accuracy, saccade velocity itself does not seem to be an important controlled variable that the FOR strives to keep constant. Second, there is no evidence from our data that the cerebellum would specifically compensate for the different elastic forces during centripetal and centrifugal saccades by monitoring orbital position and resorting to some built-in model of eye plant mechanics. Rather, the observed appropriate modification of ipsilateral FOR-bursts could also be based on feedback information about actual saccade kinematics, and thus be of general relevance in the context of adaptive saccade control mechanisms.

Eye position influences on FOR background discharge

Unilateral inactivation of the FOR not only leads to saccadic dysmetria, it also causes a tonic gaze deviation toward the injected side (e.g., Goffart et al. 1998; Ohtsuka et al. 1994; Robinson et al. 1993). Previous single-unit studies did not reveal a corresponding eye-position dependency of the tonic background discharge rate in FOR saccadic burst neurons (Fuchs et al. 1993; Ohtsuka and Noda 1991). In our sample, a considerable proportion of units exhibited statistically significant correlations between eye position and background rate, and the calculated regression coefficients could be quite high, indicating differences in background firing frequency of up to ∼32 imp/s for a shift in eye position from –16 to 16°. However, the horizontal and vertical regression coefficients did not show a significant directional bias, and the net effect of eye position on the population background rate was insignificant. Overall, therefore the present data also cannot account for the tonic gaze deviation induced by (unilateral) FOR lesions.

Concluding remarks

Our data provide, for the first time, explicit evidence that temporal properties (onset and peak latency), and also firing rates of FOR bursts change systematically with saccade velocity. Accordingly, the FOR does not seem to make use of a specifically temporal code, as proposed by Ohtsuka and Noda (1991)—altogether, the FOR output reflects and may influence much of the kinematic profile of the saccade. In addition, our data are consistent with the idea that FOR bursts are appropriately modified in response to changes in saccade kinematics in a purposeful way that may help to maintain saccadic accuracy and may constitute a mechanism for adaptive saccade control. As these signals are often not readily apparent in the highly variable discharge patterns of individual units, they can more clearly be recognized, if one considers the FOR population. Whether FOR signal variability is just noise, or whether it serves an as yet unrecognized function—e. g. to compensate for instabilities in the brain stem burst generator (cf. Robinson and Fuchs 2001)–comprehending the cerebellar role in saccade control may require an even closer look at the FOR population output.


The applied burst detection algorithm is based on the same principles as the original version described by Hanes et al. (1995) but was modified to suit the large variability of saccade related bursts and the irregular background discharge of FOR units. Briefly, the modified algorithm starts at spikes j and k in the individual spike train that are closest to the peak of the associated spike histogram. It then proceeds by successively sampling the consecutive spikes k + i = {1, 2, 3}, each time calculating, according to the Poisson cumulative probability density function with parameter λbg derived from the background rate, the associated P values Math, which correspond to the probabilities that a spike train with spikes jk + i would originate from a Poisson process characterized by λbg. The algorithm continues, by setting spike k + 1 to k and reiterating, as long as any Math and Math > 0.05, where Math is the probability that the interval between spikes k and k + 1 would occur in a Poisson process characterized by λburst, derived from the preliminarily determined average burst discharge rate. Otherwise the process stops, designating spike k as burst end, reverses and analogously proceeds toward the burst start. Bursts with an overall P value > 0.05 were discarded as spurious. The main differences in comparison with the original “Poisson spike train analysis” of Hanes et al. are 1) the decrease of the overall P value resulting from the successive inclusion of burst spikes is, in our modified version, not necessarily monotonous and 2) spike intervals were evaluated by an additional parameter, pburst (see preceding text), to enhance specificity. Note that the correctness of the calculated P values, but not the basic principle itself, depends on the assumption that the spike generating processes are truly Poisson. Peak burst amplitude was defined as the reciprocal of the mean of the shortest three consecutive interspike intervals within the burst (of the shortest 2 or 1 consecutive interspike intervals, if the burst consisted of <4 or 3 spikes, respectively), and the center of the thus determined third (2nd, 1st)-order interval was defined as burst peak location.


This work was supported by the Deutsche Forschungsgemeinschaft.


The authors thank S. Langer for technical assistance, J. Benson for editing the English text, and B. Pfreundner and I. Wendl for preparing the manuscript. We also thank Dr. E. Anagnostou for important contributions in the initial phase of these experiments.


  • The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.


View Abstract