Two male adult macaque monkeys, monkey ME and monkey MB participated in the experiment. Both animals were well trained by other oculomotor experiments and used to the test conditions. All procedures had been approved by the regional ethics committee and were in accordance with the published guidelines on the use of animals in research (European Communities Council Directive 2010/63/EU).

Directional precision of saccades and SPEM was measured with their heads freely moveable and unrestricted (Figure 1). The monkey was seated in a conventional primate chair in front of a monitor. Eye movements of the right eye were recorded with an infrared video-oculographic system (EyeLink 1000, SR Research Ltd., Osgoode, Canada) running at 1000 Hz. Testing only could be performed when the monkey’s head was in the appropriate position for the video camera system to detect the monkey’s right eye. To encourage the monkey to place its head in a suitable position for eye movement measurements and execution of the oculomotor task, we combined our reward system with a custom made mouth piece as shown in Figure 1. The mouth piece contained a photoelectric barrier and only when the monkey`s mouth interrupted the light beam, an experimental trial was started. This mouthpiece allowed for some variability of head orientations that could move by about +/-15° around the yaw axis and about +/- 10° around the pitch axis without interrupting the trial (a drawing of the mouthpiece can be obtained from the corresponding author on reasonable request).

Stimuli were presented on a color monitor (Sony Trinitron GDM F520, resolution 1280 x 1024 pixels) placed 60 cm in front of the monkey. The display subtended 37 x 28 degrees of the central visual field. All stimuli were generated using the Psychophysics Toolbox (38, 39, 40).

We tested the directional precision of initial saccades and SPEM to ramp target movements very similar to two paradigms recently tested in human subjects (30). In the ramp paradigm first a small red fixation spot was presented in the center of a uniform gray screen (38 cd/m²) for a randomized duration between 500-1000 ms (see Ramp paradigm, Figure 2). When the monkey kept its head in the appropriate position, i.e. interrupting the beam of the light barrier with its mouth, and fixated the central spot for 500-800 ms, the fixation spot was replaced by a pursuit target that moved immediately at a constant speed of 10°/s randomly either leftward or rightward across the screen for 1 s. One out of nine different vertical components of 0°, ±2°, ±5°, ±10° and ±20° was added unpredictably to the horizontal direction. In this paradigm, the monkey made first a target directed initial saccade and followed then the moving target with SPEM.

The second paradigm was the classical step-ramp paradigm developed by Rashbass (41) to elicit pure SPEMs without initial saccades. Here, the only difference compared to the ramp condition was, that the pursuit target was displaced by a small step in the direction contraversive to the direction of the upcoming target motion (Step-Ramp paradigm, Figure 2). In this paradigm, the contraversive step eliminates the necessity for an initial saccade. The step size of the pursuit target was adjusted for each monkey to minimize the occurrence of saccades during the initiation phase of pursuit. For both monkeys the best step size was 1.5 deg. After each trial, the monkey was rewarded for keeping the eye position within a 7° window around the fixation and the pursuit target by a drop of water. The ramp and step-ramp paradigms were presented in separate blocks. A single block lasted for approximately 1 hr and the monkey usually achieved 400-700 trials in each block. The order of the blocks was pseudo-random.

Typically, several hundred trials were collected for each paradigm (ramp and step-ramp) and each vertical ramp component. All data processing was done using the MATLAB (MathWorks, Natick, USA) programming package. Trials were excluded if any saccade was detected in a time-window from 200 ms before to 500 ms after stimulus motion onset during step-ramp trials. In ramp trials only one saccade was permitted, the latency of which had to be 100 ms to 300 ms from the onset of the stimulus motion. Due to these strict exclusion criteria, ~50% of the trials were rejected from further analysis in each of the conditions. On average, 657 successful trials remained in the Ramp paradigm for each vertical component in monkey MB (altogether 5912 trials) and 224 trials for each vertical component for monkey ME (altogether 2015 trials). In the Step-Ramp paradigm on average 426 trials were successful in each condition for monkey MB (3831 trials) and 311 trials for monkey ME (2796 trials).

To quantify directional precision during SPEM as well as during initial saccades, we constructed oculometric functions for each point in time (42, 43) using similar methods as described in more detail in Braun & Gegenfurtner (30). This method allowed us to compute the temporal profile of the directional precision of pursuit to step-ramp targets and of initial saccades to ramp targets.

In short, for pure SPEM to step-ramp stimuli, we first calculated the vertical and horizontal eye velocities in 1 ms time bins and smoothed the resulting speed profiles using a running average with a window size of 40 ms. We aligned the velocity traces for the step-ramp trials to the SPEM onset. SPEM onsets were calculated in each trial using the velocity profile in a time window of 10 ms that was centered on the point in time when the eye speed first exceeded 5°/s. A regression line was fitted to the velocity trace in that time window and the intersection of the regression line with x-axis was used to determine the SPEM onset (44, 45). The algorithm failed in ~9% of the cases in each monkey and those trials were excluded from further analysis.

To remove any directional bias, we used the median vertical eye velocity in response to purely horizontal ramp movements (black line in Figure 3a) as baseline for data from each monkey. For the eight step-ramp directions that had a non-zero vertical component we subtracted the vertical eye speed component from this baseline. In a next step, we calculated for each of the eight step-ramp directions the proportion of trials with an upward vertical eye direction for each 1 ms time bin (Figure 3a). Then we fitted a cumulative Gaussian to the proportions of upward trials (Figure 3b) to estimate the directional precision of the eye responses at the selected point in time. We chose the difference between the 69% and the 31% points of the Gaussian, irrespective of lapse and guess rates, as our estimate of the directional oculometric thresholds. This procedure led to more robust threshold estimates than simply taking the standard deviation of the Gaussian. Note that due to this calculation, low numeric values of the thresholds indicate high directional precision, i.e. better performance. To estimate the time course of the increase of directional precision for pure SPEM, we used the least squares method to fit an exponential decay function to the time course of threshold data (function ‘nlinfit’ in MATLAB):

Where x₁ is the asymptote of the function, x₂ is the scaling factor and x₃ represents the time constant of the decay function.

We calculated the directional precision for the whole time course of SPEM in the step-ramp paradigm. For data from the ramp paradigm we mainly focused on the directional precision during the initial saccade since our aim was to compare the directional precision of pure SPEM responses to the precision of the first initial saccade during the same time interval. To measure the directional precision of initial saccades to ramp targets we aligned the eye traces of saccades to their onsets and constructed oculometric functions based on the average vertical component added to the median direction of the eye in response to purely horizontal ramps. Peri-saccadic direction thresholds were averaged in a 30 ms time window beginning from the saccade onset. A time window of 30 ms was chosen because it was the average duration of the initial saccades.

Our results largely confirm previous reports on human oculomotor behavior showing that during saccades directional precision is higher than during SPEM (30). This is the case in particular when the peri-saccadic directional precision is compared to the Saccade Time Equivalent Pursuit thresholds (STEP) – these are the SPEM thresholds that were measured at the same time relative to the onset of stimulus motion. The degree to which saccadic direction thresholds were lower than STEP was dependent on the latency of the saccades. In our current study, the saccade latency of monkey ME was particularly short (130 ms, Figure 5) which resulted in particularly high STEP since the time-course of the precision follows a decay function. Since the peri-saccadic thresholds did not seem to be dependent on the saccadic latency (in (30), the correlation between saccade latency and peri-saccadic direction threshold was not significant) the difference between STEP and saccadic thresholds was much higher in ME than MB (for illustration see Figure 6). The conclusion for the human subjects as well as for head unrestrained monkeys is that the saccadic system receives quite accurate directional input very early so even saccades with very short latencies show low directional thresholds. In contrast, the SPEM-system either accumulates directional information over longer periods of time or is slower in translating the accurate sensory information in an equally accurate motor representation.

Our results largely confirm data of Osborne et al. (7) on the dynamics of directional precision during pursuit initiation in head-fixed monkeys. We found a similar overall time course of decreasing directional thresholds after pursuit initiation (compare our data shown in Figure 6 with Figure 10 A of (7). However, the absolute values of directional precision were different as the directional SPEM thresholds of the six monkeys in Osborne’s study were considerably lower. This difference in results probably originates from several sources. Firstly, and most importantly, Osborne and colleagues employed a head-fixed preparation, while we used a head-unrestrained approach. This was done on purpose since we aimed to employ an approach as similar as possible with typical human oculomotor studies. Since we measured eye-in-head-position rather than gaze direction, a part of the increased thresholds observed in our study might be caused by eye movements intended to compensate for (unregistered) head-movements.

Secondly, during our measurements we observed some noise in the eye position signal that was most likely a property of the experimental apparatus than of the monkey’s oculomotor behavior. It may have been caused by a relatively large distance between camera and eye as necessitated by our experimental setup. This noise obviously induced a lower absolute precision in our study compared to Osborne et al. (7).

A third difference between our and Osborne’s study was the apparatus used for the measurement of eye position. Osborne and colleagues (7) employed an invasive approach, i.e. they used implanted scleral search coils that are considered to be the gold standard regarding the precision of eye movement measurements. In our study, we employed a non-invasive approach, i.e. we used an infrared, video based eye-tracker (EyeLink 1000 Plus), as in Braun and Gegenfurtner (30). The accuracy and precision of these video-based eye trackers are potentially also very high (average accuracy ~0.5° as described in the manual) but also more dependent on the specifics of the experimental setup. Kimmel et al. (48) monitored simultaneously in two macaque monkeys the eye position with a sclera-embedded search coil and an optical tracker (Eyelink 1000) while they performed simple eye movement tasks, i.e. saccades and fixation but not SPEM. Their comparison of the two eye tracking techniques revealed a broad agreement and correlation in eye position, but also differences such as higher peak velocities for saccades and stronger post-saccadic oscillations for the optical eye-tracker.

The neural substrate for the generation of eye movements involves largely the same processing stages in humans and macaque monkeys (e.g. 49). However, the exact properties like the latency, selectivity and sensitivity of the neurons at each of these stages may be different between the species reflecting anatomical constraints and ecological demands.

Studies that have compared oculomotor behavior of monkeys and humans (e.g. 46, 47, 50, 51, 52) often concluded that there is a ‘qualitative similarity’ between the results from humans and monkeys. This term generally means that the overall structure of oculomotor behavior and types of eye movements (e.g. catch-up saccades, express-saccades, pursuit initiation, steady-state SPEM) can be found in both species, although their specific parameters and absolute values (like latencies, peak velocities and precision) often differ. These discrepancies can be partly attributed to different neuronal substrates that allow monkeys to make eye movements with shorter latencies (47) and higher peak velocities (46) than humans. While we believe that this interpretation is reasonable in cases where monkeys outperform even well trained and motivated human subjects, we also think that the situation is not as clear cut in situations where monkeys show a degraded performance compared to humans as it is the case with our monkeys when compared to the results from Braun & Gegenfurtner (30). Human and monkey subjects may differ, e.g. in the degree of training in oculomotor tasks in preference between speed and accuracy or in motivation. While both of our monkey subjects were involved in similar tasks for a longer period of time, they were never forced to perform as precisely as possible. Hence, their preference may have been rather on speed than on accuracy. This, however, can’t fully explain the different performance before they reached steady-state pursuit since speed likely doesn’t play a major role in this late pursuit component. Bourrelly et al. (53) showed in NHPs how by training the quality of pursuit (eye velocity, gain) evolved and improved while the accuracy of interceptive saccades showed no difference.

NHPs in the experimental setting often show large inter-individual differences in performance as well as performances that are vastly inferior to performance of human subject under similar conditions. As an example, Liu & Newsome (54) found that the speed discrimination thresholds of two trained monkeys differed by a factor of 2 and did not change throughout the training period.

Our results support the observation made in a number of studies, i.e. that oculomotor behavior from humans and NHPs shows the same general patterns but not always the same absolute values of the investigated parameters (46, 47, 50, 51, 52, 53). To make matters more complicated the oculomotor behavior of NHPs may show a superior performance in some aspects, e.g. saccadic latency, but inferior performance in others, e.g. precision. Accordingly, it is not always easily possible to predict the behavior of NHPs based on human data. Thus, caution should be applied when behavioral data from humans and electrophysiological recordings from NHPs are directly combined. The human data can be used as a source of information about what kinds of phenomena should be expected in NHPs, however, to link behavior directly to measurements of the underlying neurophysiological substrate in NHPs it still is required to investigate the behavior of NHPs directly.

In summary, while our findings in general support the feasibility of the monkey model for SPEM studies there are restrictions that can arise from subtle differences in the neural substrate as well as in differences in the experimental conditions as reported here. This should be kept in mind while modelling predictions of human behavior based on neuronal activities recorded in NHPs.