Foveal vision accounts for the highest acuity in the human eye, and as such, we move our eyes roughly three times per second to bring items of interest onto this high-acuity region (Kowler, 2011). High-velocity eye movements—saccades—are interspersed with periods of relative stability—fixations. We say ‘relative stability’ because during fixations, both ocular drift and microsaccades occur, and these micromovements share much of the same neuronal mechanisms as saccades of higher amplitude (Krauzlis et al., 2017). Fixations do not eliminate all oculomotor activity associated with the preparation of saccades. For example, when a cue is presented, and observers have to wait until they make a saccade, there is considerable oculomotor activity in the cued direction even before the saccade to that target is made (Engbert & Kliegl, 2003a).

Early literature placed the amplitudes of microsaccades at 12 to 20 arcmin (Zuber et al., 1965) but more recent literature tends to define microsaccades as high-velocity movements of the eyes with amplitudes of less than 1° (Ditchburn, 1973; Engbert & Kliegl 2004; and see Collewijn & Kowler, 2008, for a historical overview). Rolfs (2009), however, has recently argued for a lower amplitude boundary with a maximum amplitude of 30 min-arc. Rolfs adds that most microsaccades are horizontal and concurrent for both eyes during the performance of fixation tasks and show a binocular correlation of velocity and amplitude. Since we are typically unaware of these small saccades, it is tempting to think of microsaccadic eye movements as involuntary, but there is actually strong evidence for a shared executive mechanism with voluntary saccadic eye movements (Willeke et al., 2019). This overlap in neural mechanisms is one of the reasons that microsaccades can be difficult to define, with microsaccades sometimes considered to be on the same continuum as saccadic eye movements, which complicates the identification of any cutoff between the two (Hafed et al., 2009).

In addition to the difficulty of agreeing on a single strict definition of microsaccades, the functional purpose of these micromovements is still debated despite 60 years of research. Cornsweet (1956) studied both drift and microsaccades during fixation concluding that the primary role of microsaccades was to return the eyes to a fixation point after excess drift. Martinez-Conde et al. (2009) provided an extensive review of microsaccade research, arguing that they may serve the role of improving parafoveal and peripheral vision by preventing Troxler fading (Clarke, 1957, 1960). Microsaccades may also play a role in foveal vision, in line with evidence that the fovea may contain small subareas of fine spatial vision (Poletti et al., 2013). Microsaccades may maintain optimal positioning within this small (5 arcmin) area of the fovea (Kagan & Hafed, 2013). Others suggest, however, that other types of eye movements fulfil this role, such as drift and saccades, especially in nonexperimental, natural viewing conditions (Collewijn & Kowler, 2008). Steinman and colleagues (1973) suggested that the maintenance of eye fixations and error reduction may be achieved through slower micromovements like drift and tremor when microsaccades are voluntarily suppressed. Nevertheless, according to Martinez-Conde et al. (2009), this does not mean that microsaccades play no role in the prevention of retinal image fading.

Ditchburn (1980) argued that microsaccades perform an error-correcting function during fixation. After the viewer’s gaze moves outside the boundaries of the target area via naturally occurring drift and tremor movements to prevent image fading, microsaccades return the gaze to the location of the target. In response, Kowler and Steinman (1980) argued against this error-correcting role of microsaccades and against any clear function of microsaccades, calling them an “evolutionary puzzle.” Bridgeman and Palca (1980) tested performance on a needle-threading task, finding suppressed microsaccade activity preceding the moment high-acuity information was to be received. Nevertheless, similar to Kowler and Steinman (1980), they did not find any functional explanation for microsaccades. Later, Engbert and Kliegl (2004) studied microsaccadic eye movements during fixation. They concluded that while microsaccades keep the eyes stable during fixation, they also serve both error-producing and error-reducing roles, depending on the temporal scale. In the former role they decrease the effects of retinal adaptation during short fixations (around 20 ms), whereas in the latter role they correct fixation errors and aid in binocular coordination during long fixations (between 100 and 400 ms; see also review by Martinez-Conde et al., 2013). More recently, Ko et al. (2010) and Kowler and Collewijn (2010) concluded that micromovements differ by type, distinguishing between fixational and exploratory microsaccades. Microsaccade rates were higher under strict fixation conditions than during high-acuity task performance, and only the fixational microsaccades could be voluntarily suppressed.

Another key question is the degree to which microsaccades are binocular, and whether they are binocularly coordinated (see Collewijn & Kowler, 2008, for a review). Cornsweet’s (1956) studies of microsaccades were carried out monocularly and only horizontal movements were recorded. Krauskopf et al. (1960) recorded eye movements binocularly and determined that, unlike drift, saccadic eye movements in the two eyes were correlated in direction and amplitude. They found that during binocular fixation, 98% of the saccades were binocularly synchronized and unidirectional. They suggested that once a specific amount of error has accumulated, the eyes correct this fixation error in each eye independently. However, this process is driven by a single central mechanism that initiates the microsaccades in both eyes simultaneously. St-Cyr and Fender (1969) reported that even though microsaccades occurred simultaneously in both eyes, their amplitudes could differ between the eyes leading them to suggest that microsaccades correct fixation errors in both eyes on average and that they serve to minimize the disparity between the eyes. Since then, most research on microsaccades has been conducted with binocular recordings. Møller and colleagues (2002) found that the amplitude and direction of microsaccades in both eyes matched almost perfectly. Most authors therefore recommend using binocular recordings (Engbert & Kliegl, 2003a, b).

Using microsaccades to understand visual cognition

Microsaccades can provide insights into various aspects of cognition. For example, microsaccade direction and rate can provide clues about the allocation of attention. Changes in salience at fixation can lead to reductions in microsaccade rates (Engbert & Kliegl, 2004), and studies of microsaccade dynamics have provided strong evidence for a connection between microsaccades and covert visual attention, such as response enhancement for pre-microsaccadic stimuli (Chen et al., 2015). Zuber and Stark (1966) found that microsaccades may influence saccade generation with each microsaccadic eye movement delaying saccade onset by approximately 50–100 ms. This observation supports the idea of a shared neurophysiological mechanism responsible for the generation of saccadic and microsaccadic eye movements (see Hafed et al., 2015, for a review). Rolfs et al. (2008) concluded that the mechanisms involved in the inhibition of saccades and microsaccades are shared. Since inhibition is one of the characteristics of executive control (Johnston & Everling, 2006; Paneri & Gregoriou, 2017), one approach to studying microsaccades could be within the context of using executive control to suppress the activity of the oculomotor system.

Antisaccades

Antisaccades are voluntary eye movements made to the opposite location to where a stimulus is presented (initially described by Hallett, 1978; see Kristjánsson, 2011, for a review) and are a useful tool for studying executive control. In a typical prosaccade task, participants fixate on a central fixation point, and should make a prosaccade towards a peripheral stimulus onset. For antisaccades they should suppress reflexive saccades towards the stimulus onset and instead make a saccade of equal amplitude in the opposite direction. Antisaccades therefore require the successful suppression of stimulus-driven prosaccades before the saccadic stimulus appears (Coe & Munoz, 2017), followed by deliberate programming of the opposite saccade vector (Kristjánsson et al., 2001; Munoz & Everling, 2004). If this top-down suppression builds-up throughout the trial, then it might be found in prestimulus microsaccade rates.

Another experimental manipulation requiring executive control is task-switching (Monsell, 2003). Performance costs associated with switching between two or more tasks are seen in a large variety of paradigms and may interact with saccades. For example, task-switching involving prosaccades with a short suppression of response may introduce interference and delay the planning of subsequent saccades (Tari et al., 2019; Tari & Heath, 2019). When prosaccades are less probable within a trial block, prosaccade errors increased (Pierce et al., 2015). Also, within mixed blocks of pro- and antisaccades, the percentage of correct responses was lower than in fixed blocks. Moreover, a preceding antisaccade trial increased saccade latency on subsequent trials, irrespective of type. The authors suggested that the repeated antisaccades inhibit saccade-generating neurons in the frontal eye fields and superior colliculus, increasing their thresholds.

The oculomotor system is well-suited to studying executive control because of solid theoretical foundations for gaze control mechanisms (Kristjánsson, 2011; Munoz & Everling, 2004). Hermens et al. (2010) studied microsaccades in the antisaccade paradigm, conducting two experiments. In the first part of Experiment 1, participants fixated for 1,000 ms, delaying their response during the display of a peripheral target for 1,500 ms until the fixation cross disappeared and then made an eye movement towards or away from the target. The second part had no saccade onset delay, but the fixation period was randomized between 1,000 and 1,400 ms. Experiment 2 was similar except that a central arrow was used to indicate the direction of the saccade on both delayed and nondelayed trials. All blocks consisted of either prosaccades only or antisaccades only. Microsaccade rates were analyzed 600 ms before target onset until 1,500 ms and 50 ms after onset in the delayed and nondelayed conditions, respectively. Hermens and colleagues (2010) found lower microsaccade rates for antisaccades than prosaccades, but notably only on delayed trials with peripheral targets.

Watanabe et al. (2013) hypothesized that the preparation of voluntary actions can be predicted by looking at fixational saccades. In their primary paradigm participants had to make a saccade either towards or away (depending on fixation point colour) from a peripheral stimulus appearing for 1,000–1,500 ms on 80% of the trials. The remaining 20% of trials were catch-trials, where the fixation point disappeared for 50 ms, and participants had to maintain fixation for another 1,000–1,500 ms when it reappeared. Fixational saccade rates were analyzed with a temporal window spanning 400 ms before and 70 ms after stimulus onset. With longer fixations preceding stimulus onset, the frequency of fixational saccades was reduced. The frequency was also reduced on correct antisaccade trials compared with prosaccade trials. This differs from Hermens et al. (2010), where there were no differences in microsaccade rates preceding pro- and antisaccades in the nondelayed condition.

Dalmaso et al. (2020) compared the Hermens et al. (2010), and Watanabe et al. (2013) studies, combining the methods by investigating microsaccades during antisaccades within fixed and random blocks. Participants fixated a ring at screen centre for 1,500 ms, followed by a cue (a square or diamond). A target followed 2,000–2,500 later (on the left or right) and participants were to make a saccade toward or away from it. Pupil-size dynamics were analyzed in addition to microsaccade rates. Dalmaso et al. (2020) found fewer microsaccades and larger pupil sizes before an antisaccade, but only within mixed blocks. Dalmaso et al. (2020) argued that this reflected differences in cognitive load causing decreased microsaccade rates during more difficult tasks. Dalmaso et al. (2019) investigated the effects of cognitive load on microsaccade rates in a flanker task (Eriksen & Eriksen, 1974), finding reduced microsaccade rates before trials associated with so-called ‘cognitive conflict’, where the tasks were incongruent with the preceding cues.

Current aims

Building on previous research, we studied microsaccade dynamics in the context of task switching and pro- versus antisaccades. We hypothesized that oculomotor inhibition would be higher when observers perform antisaccades than prosaccades because they have to suppress reflexive saccades to the peripheral onset (Kristjánsson et al., 2001, 2004). We therefore expected participants to begin active suppression of the oculomotor system in anticipation of the antisaccade response and thus to observe fewer microsaccades on antisaccade than prosaccade trials. Another important question involves the temporal profiles of any microsaccade modulation. We hypothesized that the microsaccade reduction would be most pronounced closer to the critical signal indicating whether a pro- or antisaccade was to be made. We also expected higher microsaccade rates on trials in the randomized mixed block because of larger uncertainty associated with task-switching. Less executive control resources for inhibiting saccadic movements would be available since those limited resources are shared with multiple response encodings needed for successful performance of the two tasks. All the trials in our paradigm involved immediate, nondelayed saccadic responses. We used both mixed and fixed blocks with no auditory cues and the precue period was fixed across all trials. We additionally compared microsaccade rate dynamics for binocular and monocular eye tracking.

Methods

Overview

Experiment 1 involved monocular eye tracking, while in Experiment 2 we used binocular recording to avoid noise and random eye movements during monocular recording that might be mistaken for microsaccades.

Data analysis

The pipeline for data analysis was similar in both experiments (see Supplementary Material, Fig. S1). We used the microsaccade detection toolbox in R (Engbert et al., 2015) with the following parameters: the sampling rate of the eye tracker was 1000 Hz; the minimum duration and the velocity threshold scaling factor parameters were both set at 4. The analyses for both experiments included calculations of saccade latency, the estimation of microsaccade rates (the average rate per second using generalized linear mixed effects (GLME) models and a running rate analysis using a binned sliding window algorithm) and a cluster permutation analysis of the data.

Stimuli and procedure

Stimuli and procedure were the same for both experiments. Each participant contributed to four conditions, involving both saccade types (prosaccade and antisaccade) and both block types (saccade type randomly mixed within a block and saccade type fixed within a block; Fig. 1). For prosaccade trials, a green fixation cross was initially presented at screen centre along with empty placeholder boxes (size = 3° × 3°) 7° to the right and left of fixation. After 800 ms, a response signal in the form of a white flash appeared for 100 ms, after which participants were to make a saccade towards the highlighted box. For antisaccade trials, a red X for fixation appeared at screen centre, and participants had to make a saccade towards the box opposite the highlighted location. The mixed trial blocks consisted of randomly mixed pro- and antisaccade trials. The four conditions were presented across three blocks (two fixed, one mixed) with block order counterbalanced.

Fig. 1
figure 1

The General Design Used in Both Experiments. Note. a Fixed prosaccade trials; b fixed antisaccade trials; c randomly mixed block of pro- and antisaccade trials. On all trials fixation lasted for 800 ms, after which a cue appeared for 100 ms, signalling the direction of the saccade to be made. Up to 3,000 ms were allowed for an eye movement before the trial ended. (Colour figure online)

Full size image

Participants had 3,000 ms on each trial to make a response, and if they did not respond, the trial ended and was registered as a “no response” trial. Each trial was followed by a refreshed screen with a black fixation cross. Once the display refreshed, participants could initiate the new trial by pressing the space bar. There was also a 2-minute resting period in between blocks.

Participants were encouraged to make saccades as quickly and accurately as possible. The trial did not end if a saccade was made towards the wrong direction, unless a saccade toward the correct side was initiated after the incorrect one within the 3,000-ms period. However, only the first high-amplitude saccade (>1°) was included in the analysis.

Experiment 1: monocular microsaccade detection

Participants

In Experiment 1 there were twenty-two participants (10 male, 12 female). Participants with fewer than 50% of correct responses and over 50% fixation errors, blinks, missing samples, or poor calibration (error >1°) were excluded from analyses (one participant was excluded). All participants had normal or corrected-to-normal eyesight. They were not rewarded for their participation and provided their voluntary consent for participation in written form. The experiment was approved by the ethical committee of HSE University.

Recording

The stimuli were presented on a 24-in. Asus VG248 LCD monitor with a resolution of 1,920 x 1,080 pixels and a refresh rate of 144 Hz. An EyeLink 1000 Plus eye tracker (SR Research Ltd, Osgoode, Ontario, Canada) was used to track eye position. Recordings were carried out monocularly on the dominant eye (determined by the Porta test) at 1000 Hz in a slightly darkened room. Participants sat with their head position fixed with an adjustable chinrest at a position comfortable for each participant. The distance from the monitor to the chinrest was 58 cm. Stimuli were presented using the Psychophysics toolbox (Brainard, 1997) for MATLAB (The MathWorks Inc., Natick, MA).

Procedure

Participants were instructed to make saccadic eye movements in response to a peripheral spatial onset for each trial (see Fig. 1). The three blocks (run in counterbalanced order) included ( a) a block of 50 prosaccade trials, (b) a block of 50 antisaccade trials, and (c) a block of 200 randomly mixed trials (100 of each saccade type) with a break after each 50 trials. Before starting, participants performed 10–15 prosaccade trials for training.

Results

Mean latency

Latency was calculated as the time from the peripheral stimulus onset until a saccade was initiated. A generalized linear mixed effect model (GLMER, using the lme4 package in R) was used to assess the results. Fixed effects included saccade type (pro and anti) and block type (mixed and fixed) with both slopes and intercepts included. A gamma distribution was used to control for the skew typically seen in reaction time data (Kristjánsson & Jóhannesson, 2014; Lo & Andrews, 2015), and subject was included as a random factor. Fixed effects were tested for significance using a chi-squared (χ2) test provided by the ANOVA function in the CAR package. Expected marginal means, error bars and pairwise comparisons were calculated using the emmeans package.

There was a significant main effect of saccade type, χ2(1) = 33.622, p < .001, and an interaction between saccade and block types, χ2(1) = 8.282, p < .005, but no significant effect of block type, χ2(1) = 0.064, p > .05, on saccade latency. Latencies were longer for antisaccades than prosaccades (282 ms, SE = 0.0042 and 260 ms, SE = 0.0046, respectively; Fig. 2).

Fig. 2
figure 2

Expected Saccade Latency Marginal Means by Saccade Type and Block Type in Experiment 1. Note. The interaction of saccade and block types are plotted. Only the main effect of saccade type was significant. Block boundaries represent standard error and whiskers are 95% CIs. (Colour figure online)

Full size image

Microsaccade velocity-amplitude correlations

To ensure that the microsaccade detection algorithm worked correctly, we plotted the observed microsaccades by saccade amplitude versus velocity. The “main sequence” where velocity is a linear function of amplitude (Bahill et al., 1975) is generally assumed to be a characteristic feature of both saccades and microsaccades (Zuber et al., 1965). We say “generally” because of reported exceptions (Jóhannesson & Kristjánsson, 2013; Rolfs et al., 2008; also see General Discussion). Saccades fall into the space between 0.5° and 40° (Wong, 2014), and microsaccades typically have an amplitude of 1’–25’ (Ciuffreda & Tannen, 1995). The data in Fig. 3 confirm that this was the case in our data.

Fig. 3
figure 3

The Main Sequence Correlation of Velocity and Amplitude of the Detected Microsaccades in Experiment 1. Note. The correlation between the amplitude and velocity of microsaccades established that microsaccades were detected correctly. r = .881, df = 7891, p < .001. (Colour figure online)

Full size image

Microsaccade amplitude

We assessed microsaccade amplitude separately for both correct and incorrect pro- and antisaccades by condition (see Fig. 4). There were significant effects of saccade type, χ2(1) = 3.979, p = .046, and saccade correctness, χ2(1) = 133.027, p < .001, and there was an interaction between block type and saccade correctness, χ2(1) = 8.355, p < .01. Microsaccade amplitudes on prosaccade trials were generally larger than on antisaccade trials (0.311°, SE = 0.022 against 0.3°, SE = 0.022, respectively). This difference was strongest on incorrect trials in the fixed condition (0.418°, SE = 0.032 on prosaccade trials, 0.386°, SE = 0.034 on antisaccade trials). However, the opposite was the case for incorrect trials in the mixed condition, where microsaccade amplitudes were larger preceding antisaccades than prosaccades (0.341°, SE = 0.021 against 0.33°, SE = 0.022, respectively).

Fig. 4
figure 4

Microsaccade Amplitudes in Experiment 1. Note. The figure depicts the mean amplitudes of microsaccades preceding correct and incorrect pro- and antisaccades in fixed and mixed conditions. The two top panels show the mean amplitude figures with a 95% CI for correct (left panel) and incorrect (right panel) trials. The two bottom panels show the density distribution of the same data. (Colour figure online)

Full size image

Microsaccade direction

The graph of microsaccade direction (by correct versus incorrect saccades) revealed a horizontal bias across all conditions (Fig. 5), in line with the literature (Rolfs, 2009).

Fig. 5
figure 5

Microsaccade Direction by Correct and Incorrect Trials in Experiment 1. Note. Directions of microsaccades are plotted as polar density plots separately for correct (top panel) and incorrect (bottom panel) trials with regard to saccade and block types. (Colour figure online)

Full size image

Average microsaccade rate

The average microsaccade rate was calculated as the number of microsaccades per second in each condition (Fig. 6). The microsaccade rate was subjected to a linear mixed-effect model, with saccade type and block type as fixed effects and subject as a random factor. Neither the effects of saccade type, χ2(1) = 1.182, p = .276, nor block type, χ2(1) = 1.236, p = .266, were significant, but their interaction was significant, χ2(1) = 4.447, p = 0.035, primarily caused by the difference between prosaccades and antisaccades in the fixed block (meanpro = 1.41, SE = 0.083 and meananti = 1.56, SE = 0.086) with little differences in the mixed condition (meanpro = 1.57, SE = 0.085 and meananti = 1.52, SE = 0.084). The microsaccade rate for antisaccades was numerically higher in the fixed (mean = 1.56, SE = 0.086) than the mixed condition (mean = 1.52, SE = 0.084), but the difference was not significant (z ratio = 0.572, p = .567). The results pattern is consistent (though not significant) with our expectations in the mixed condition, where microsaccade rates for antisaccades were numerically lower than for prosaccades. However, the direction of the average microsaccade rate was opposite to what we predicted for the fixed condition, with higher microsaccade rates for antisaccades than prosaccades.

Fig. 6
figure 6

The Microsaccade Rate per Second in the Different Trial Blocks in Experiment 1. Note. Mean values are the expected marginal means with errors including SE (blocks) and CI (whiskers). (Colour figure online)

Full size image

Microsaccade running rate

We calculated the microsaccade running rates over the course of trials to assess the temporal dynamics of microsaccade rates in the 800-ms period between the onset of the central fixation signal at trial start and the peripheral onset signalling the saccadic response. To calculate the microsaccade running rate, we divided the time interval prior to the response signal at 800 ms into 5-ms bins and used a 10-ms sliding window on the data. This was then plotted out (Fig. 7) using “ggplot” (Wickham, 2016). While we observed no clear differences between block nor saccade type, we performed a cluster permutation analysis to determine whether microsaccade rates differed significantly between saccade types within any ranges in the two blocks.

Fig. 7
figure 7

The Running Rate of Microsaccades Plotted Using a Moving Window from Experiment 1. Note. The figure depicts a total period of 850 ms: 800 ms between the onset of the central fixation signal and the peripheral onset that signalled the saccadic response and 50 ms post onset. The left panel contains clusters of significant differences in microsaccade activity between pro- and antisaccade trials in the fixed condition marked by red vertical lines at the following time points: (1) 147.5 ms, (2) 197.5 ms, (3) 312.5 ms, and (4) 507.5–517.5 ms. The width of the lines corresponds to the cluster mass statistics specified in black bold font over the lines. The right panel contains the plot of the MS running rate in the mixed condition. No significant differences in MS activity were found. (Colour figure online)

Full size image

Cluster permutation

We used the “Cluster Permutation” package in R (Barr, 2020) to test the binned microsaccade data obtained during the pre-processing step before the calculation of the microsaccade running rate. The data were divided by block type. A bin-by-bin analysis of variance was performed on the running rate for each saccade type. We then attempted to detect clusters of significant differences between saccade types, separately by block (Fig. 7, in red). For the fixed condition there were four clusters of significant rate differences between pro- and antisaccades, at: (1) 147.5 ms, (2) 197.5 ms, (3) 312.5 ms, and (4) 507.5-517.5 ms, while no such clusters were found in the mixed condition (see detailed test results in Table 1, Supplementary Material section).

Discussion

We hypothesized that microsaccade rates would be smaller for antisaccades than prosaccades, reflecting suppression of the oculomotor system during antisaccade execution. We also calculated the microsaccade running rate to assess the temporal dynamics from the onset of fixation and the onset of the peripheral saccade signal.

Antisaccade latencies were slower than prosaccade latencies in both the fixed and mixed conditions, consistent with previous literature (Edelman et al., 2006; Hallett & Adams, 1980; Olk & Kingstone, 2003). There were no significant effects of either saccade or block types upon microsaccade rates, but there was a significant interaction between the two. For antisaccades, microsaccade rates were numerically higher in the fixed than the mixed condition. This is consistent with the common finding of decreased microsaccade rates in more demanding cognitive tasks (e.g., Dalmaso et al., 2017; Hermens et al., 2010; Krejtz et al., 2018; Lange et al., 2017; Pastukhov & Braun, 2010; Siegenthaler et al., 2014; Watanabe et al., 2013; but see also Benedetto et al., 2011; Chen et al., 2008; Di Stasi et al., 2013).

The analysis of the dynamic microsaccade rates showed an overall reduction in rates following the fixation onset before the response signal in all conditions. In the mixed condition, microsaccade activity did not differ between antisaccade and prosaccades. In the fixed condition, there were more points of variability between prosaccade and antisaccade trials, manifesting in an almost cyclic pattern of alternating stronger and weaker levels of inhibition on antisaccade trials, with each “cycle” having more suppression than the previous one until a gradual release before a saccade. The “cycles” occurred at approximately the following periods: (1) 0–100, (2) 100–300 ms, (3) 300–500 ms, (4) from 500 ms onward, with the fourth “cycle” having a lesser level of suppression right before the response signal onset. This larger difference between pro- and antisaccades in the fixed block was confirmed by the cluster permutation analysis. A total of four significant clusters were identified by the algorithm at the following time points: (1) 147.5 ms, (2) 197.5 ms, (3) 312.5 ms, and (4) 507.5–517.5 ms, but notably, these windows of significance were extremely short.

These results could be a consequence of a higher control levels necessary for suppressing reflexive saccades in the antisaccade condition (Coe & Munoz, 2017) but could also reflect the monocular recording. Most existing studies agree that microsaccades occur in both eyes simultaneously, so monocular data may contain artifacts. So, while the results of Experiment 1 are suggestive, we further addressed these issues in Experiment 2, this time using binocular eye recording.

Experiment 2: binocular microsaccade detection

Participants

Twenty-three participants (11 male, 12 female) took part. Participants with less than 50% of correct responses and over 50% fixation errors, blinks, missing samples, and other tracker messages potentially indicating unreliable data or poor calibration (error >1°) were excluded from the analyses (two participants were excluded due to too many fixation errors and one due to poor calibration). All participants had normal or corrected-to-normal eyesight and provided written consent for participation. The experiment was approved by the ethical committee of HSE University and was in accordance with the requirements of the Icelandic Bioethics Committee.

Recording

The methods in Experiment 2 were identical to Experiment 1, except for the following differences: the monitor was a 24-in. BenQ XL2411Z screen with a resolution of 1,920 × 1,080 pixels and a refresh rate of 144 Hz. Recordings were carried out binocularly at a frequency of 1000 Hz in a slightly darkened room.

Procedure

As in Experiment 1, participants were instructed to respond with anti- or prosaccades during two block types (mixed and fixed). Trials were delivered across three blocks with one block of 50 prosaccade trials, a block of 50 antisaccade trials and a block of 100 randomly mixed trials (50 of each saccade type) divided into two sets of 50 trials to equate the fixed and mixed conditions. The number of trials in the mixed block was reduced to 100 in Experiment 2 to match the trial number in the fixed blocks. The blocks were run in counterbalanced order across participants. Experiment 2 was performed in accordance with the internationally standardized antisaccade protocol (Antoniades et al., 2013). Before recording, participants performed 10–15 prosaccade training trials.

Results

Mean latency calculation

Latencies were analyzed with the same general linear model as in Experiment 1, with saccade type and block type as fixed effects and subject as an independent variable. There were significant main effects of saccade type, χ2(1) = 61.597, p < .001, and block type, χ2(1) = 14.607, p < .001), as well as an interaction between them, χ2(1) = 15.691, p < .001. Latencies for antisaccades were longer than for prosaccades (296 ms, SE = 0.0053 and 279 ms, SE = 0.0052, respectively). Latencies were also longer in the mixed than the fixed blocks (296 ms, SE = 0.0051 and 278 ms, SE = 0.0054, respectively). The interaction was caused by larger latency costs for antisaccades in the fixed than the mixed condition (Fig. 8).

Fig. 8
figure 8

Expected Saccade Latency Marginal Means for Saccade Type and Block Type in Experiment 2. Note. The interaction of the two is plotted, although only the main effects were significant. Block boundaries represent standard error, and whiskers are 95% CIs. (Colour figure online)

Full size image

Microsaccade velocity versus amplitude

Velocity and amplitude are plotted in Fig. 9, showing that the peak velocities of the detected microsaccades were consistent with previous literature and in accordance with the main sequence.

Fig. 9
figure 9

The Main Sequence Correlation of Velocity and Amplitude of the Detected Microsaccades in Experiment 2. Note. The correlation between the amplitude and velocity of microsaccades established that microsaccades were detected correctly. r = .811, df = 1718, p < .001. (Colour figure online)

Full size image

Microsaccade amplitude

We analyzed microsaccade amplitudes separately for correct and incorrect pro- and antisaccades in fixed and mixed conditions (Fig. 10). Saccade correctness significantly affected preceding microsaccade amplitudes (p < .001, χ2 = 25.659). Microsaccades had higher amplitudes and higher variance on incorrect than correct trials (mean amplitude of 0.456°, SE = 0.045 and 0.352°, SE = 0.029, respectively). There was an interaction between saccade type and saccade correctness (p < .01, χ2 = 8.550), with higher amplitudes on erroneous prosaccade trials, compared with erroneous antisaccade trials (mean amplitude of 0.5°, SE = 0.048 and 0.412°, SE = 0.043, respectively).

Fig. 10
figure 10

Microsaccade Amplitudes in Experiment 2. Note. Mean microsaccade amplitudes are plotted separately for correct (top left panel) and incorrect (top right panel) trials by saccade and block types. The bottom panels illustrate density plots for microsaccade amplitudes on correct (bottom left panel) and erroneous trials (bottom right panel). (Colour figure online)

Full size image

Microsaccade direction

We plotted microsaccade direction by saccade correctness (Fig. 11), finding a horizontal bias across all conditions, in line with the literature (Rolfs, 2009). The direction graphs also reflect that microsaccades on incorrect trials had higher amplitudes than correct trials, especially in the horizontal plane.

Fig. 11
figure 11

Microsaccade Direction by Correct and Incorrect Trials in Experiment 2. Note. Directions of microsaccades are plotted as polar density plots separately for correct (top panel) and incorrect (bottom panel) trials with regard to saccade and block types. (Colour figure online)

Full size image

Microsaccade-saccade direction congruency

We further analyzed microsaccade direction by their congruency with the direction of consecutive correct and incorrect saccades (Fig. 12). Looking at the frequency of microsaccades that had the same (congruent) or the opposite (incongruent) direction as the following saccade, we observed a main effect of saccade correctness, χ2 = 8.650, p = .003, an interaction between saccade correctness and congruency, χ2 = 5.537, p = .018, and an interaction between congruency, saccade correctness, saccade type and block type, χ2 = 4.203, p = .040. On correct fixed antisaccade trials incongruent microsaccades were more frequent than congruent ones (57.273, SE = 4.038 and 50.035, SE = 3.634). On correct fixed prosaccade trials congruent microsaccades were more frequent than incongruent directions (53.624, SE = 3.711 and 46.577, SE = 3.248). The pattern in the mixed condition was the same as in the mixed condition on correct antisaccade trials: more incongruent than congruent microsaccades (55.190, SE = 3.838 and 52.328, SE = 3.731). On correct prosaccade trials the frequency of incongruent microsaccades was slightly higher than on congruent ones (50.855, SE = 3.458 and 49.198, SE = 3.346). For more details see estimated marginal means, Table 2, Supplementary Material section.

Fig. 12
figure 12

Congruency of Microsaccade Direction With Consecutive Saccade by Error in Experiment 2. Note. The frequencies of microsaccade directions are illustrated with regard to their congruency with consequent saccade directions. The plots show the direction congruency separately for correct and incorrect trials with regard to saccade and block types, with fixed trials on the left and mixed trials on the right. (Colour figure online)

Full size image

Average microsaccade rate

Microsaccade rates were analyzed with the same mixed model as in Experiment 1. As we hypothesized, the main effect of saccade type was significant, χ2(1) = 7.908, p = .005, with reduced microsaccade rates prior to antisaccades than prosaccades (mean = 0.077, SE = 0.026). The effect of block type was not significant, χ2(1) = 1.884, p = .169, and there was no interaction between the two, χ2(1) = 1.014, p = .313 (Fig. 13). There was a visible trend towards lower rates for antisaccades (meanpro = 0.421, SE = 0.057 and meananti = 0.370, SE = 0.056; z ratio = 1.943, p = .052). The difference in microsaccade rates in the mixed condition was significant (meanpro = 0.555, SE = 0.074 and meananti = 0.451, SE = 0.069; z ratio = 2.271, p = .023). Numerically, microsaccade rates were lower within fixed (mean = 0.395, SE = 0.055) than mixed blocks (mean = 0.503, SE = 0.067).

Fig. 13
figure 13

The Microsaccade Rates per Second in the Different Trial Blocks in Experiment 2. Note. Mean values are the expected marginal means with errors including SE (blocks) and CI (whiskers). (Colour figure online)

Full size image

Microsaccade rates by saccade error

We additionally assessed microsaccade rates preceding correct and incorrect saccades (Fig. 14). The analysis showed an interaction between saccade type and saccade correctness, χ2(1) = 3.951, p = .046. For estimated marginal means, see Table 3, Supplementary Material section.

Fig. 14
figure 14

Microsaccade Rates by Saccade Error in Experiment 2. Note. The figure illustrates the rate of microsaccades per second preceding correct and incorrect pro- and antisaccades. (Colour figure online)

Full size image

Microsaccade running rate

The same approach as in Experiment 1 was used for calculating microsaccade running rates in Experiment 2: We implemented a sliding window across the binned data for all conditions and participants. The time interval prior to the response signal at 800 ms was divided into 5-ms bins, after which a 10-ms window was slid over these bins to calculate an average at each bin point. The resulting data were plotted using “ggplot” (Fig. 15). We next performed a cluster permutation analysis to determine if any ranges contained significantly different microsaccade rates for prosaccades and antisaccades for the two block types.

Fig. 15
figure 15

The Running Microsaccade Rates in Experiment 2, Plotted Using a Moving Window. Note. The figure depicts a total period of 850 ms: 800 ms between the onset of the central fixation signal and the peripheral onset that signalled the saccadic response and 50 ms post onset. The left panel shows clusters of significant differences in microsaccade activity between pro- and antisaccade trials in the fixed condition with red vertical lines at the following time points (1) 2.5–12.5 ms, (2) 392.5 ms, (3) 787.5–792.5 ms, (4) 802.5–827.5. The right panel shows microsaccade activity between pro- and antisaccade trials in the mixed condition (red vertical line at 152.5ms.) The width of the lines corresponds to the cluster mass statistics figures specified in black on the lines. (Colour figure online)

Full size image

Cluster permutation

We used the “Cluster Permutation” package in R (Barr, 2020), testing the binned microsaccade data obtained during the preprocessing step before the calculation of the microsaccade running rates. The data were divided by block type (fixed vs. mixed). A bin-by-bin analysis of variance was performed on the running rate for each saccade type. We then attempted to detect clusters of significant differences between saccade types, separately by block (Fig. 15). For the fixed condition, the analysis showed four clusters of significant rate differences between the pro- and antisaccade trials at: (1) 2.5–12.5 ms, (2) 392.5 ms, (3) 787.5–792.5 ms, (4) 802.5–827.5. Only one significant cluster was found for the mixed condition at 152.5 ms (see detailed test results in Tables 4 and 5, Supplementary Material section).

Discussion

Consistent with our central hypothesis, in Experiment 2 we again found evidence for suppressed microsaccade activity during antisaccades, this time with binocular eye tracking. As in Experiment 1, we used GLME models to calculate the main effects of saccade and block types for saccadic latencies and microsaccade rates, using a sliding window to calculate the microsaccade running rate over bins of data, as well as checking for clusters of significant differences between saccade types.

Latencies were lower in the antisaccade and the mixed conditions, as in Experiment 1 and consistent with previous results. The microsaccade rates were also reduced prior to the generation of antisaccades in both the fixed and mixed conditions. The effect of block type was not significant, although the rate for fixed blocks was numerically lower than for mixed blocks.

The running rate analysis showed a consistent high rate of microsaccades immediately after the onset of the fixation instructions followed by a reduction in all experiment conditions.

We may notice that the suppression in the fixed antisaccade condition is more gradual and stable from around 100 ms to the strongest point of suppression around 530 ms, followed by a release around 100 ms before the response signal onset. Specifically, microsaccade rates on antisaccade trials were suppressed at trial start, with a peak at around 110 ms. After this increase, microsaccade activity dropped notably, especially around the 530–680 ms time frame where a dip was followed by a large increase right before the peripheral onset. This might reflect released suppression towards the end of the fixation period in anticipation of the onset. No such effect was seen for prosaccades presumably because suppression of reflexive saccades is not needed or weaker. Notably, the cyclic pattern from Experiment 1 was not seen in Experiment 2.

In the mixed condition, we again found initial suppression with a rise in microsaccade rates before antisaccades, but not as strong as in the fixed block. This initial rise was followed by a large drop starting at around 180 ms and lasting until 750 ms, but not as stable and gradual as in the fixed antisaccade block. At this point we again see an increase in microsaccade activity preceding the peripheral onset, though this rebound effect was again not as strong as in the fixed block.

The additional cluster permutation test revealed a larger difference in microsaccade activity between pro- and antisaccade trials in the fixed than the mixed condition, with four clusters against one cluster, respectively, at the following time points: (1) 2.5–12.5 ms, (2) 392.5 ms, (3) 787.5–792.5 ms, and (4) 802.5–827.5 ms in the fixed condition and (1) 152.5 ms in the mixed condition.

General discussion

We conducted two experiments to assess microsaccade activity prior to the generation of prosaccades and antisaccades, during either fixed blocks (only pro- or antisaccades) and mixed blocks where both saccade types were to be performed. We expected suppression of microsaccade activity during antisaccade trials and within the fixed trial blocks compared with prosaccades and mixed trial blocks. We assessed the mean microsaccade rates for each trial, the mean saccadic latency for each trial and the microsaccade running rates within each trial with both monocular (Experiment 1) and binocular (Experiment 2) eye tracking. The binocular data were more reliable as seen in the main sequence plots, the average microsaccade rates and the running rates. The main sequence plot for Experiment 1 shows larger dispersion than the plot for Experiment 2, while microsaccade rates in Experiment 1 were higher than in Experiment 2. There were also more outliers in Experiment 1 (30%) than Experiment 2 (20%).

Microsaccade suppression: Mean rates

Experiment 1 did not reveal significant effects of saccade type nor block type upon microsaccade rates, but there was a significant interaction. This pattern contrasted with our predictions with lower microsaccade rates for prosaccades than antisaccades within fixed blocks, and a slight reversal of this pattern within mixed blocks.

Conversely, the binocular recordings in Experiment 2 supported our predictions of suppressed microsaccade rates in the antisaccade condition. Additionally, the microsaccade rates were lower in the fixed than mixed condition, although this trend was not significant.

The main effect of saccade type in Experiment 2 was consistent with Dalmaso et al. (2020), but our trend for fewer microsaccades within fixed blocks contrasts with their finding of reduced microsaccade rates during mixed trials. They concluded that lower microsaccade rates during mixed trials may be due to a higher cognitive load than within fixed blocks. One explanation for the trend in our study could be the top-down control mediated by the frontal cortex required by the oculomotor system (Everling & Johnston, 2013; Ploner et al., 2005). In the mixed block, participants may have had less time to fully implement the top-down control for microsaccade suppression, a task-switching effect, in other words.

The increased microsaccade rates for antisaccades during fixed blocks in Experiment 1 runs counter to our main hypothesis and the findings of Dalmaso and colleagues (2020). This may reflect that the mechanisms for saccade and microsaccade generation overlap (Hafed et al., 2009; Rolfs et al., 2008; Willeke et al., 2019; see Hafed et al., 2015, for a review). Since the production of saccades may be temporarily suppressed due to microsaccade activity (Rolfs et al., 2005), reduced microsaccade rates in the fixed prosaccade condition might reflect an attempt by the oculomotor system to avoid the cost of inhibited saccade initiation (Melloni et al., 2009; Zuber & Stark, 1966). However, a simpler explanation could be that fixations recorded monocularly contained more spurious microsaccades (Engbert & Kliegl, 2003a, b).

Saccadic latencies

The results for saccadic latencies in both experiments were consistent with existing results with main effects of both saccade and block types (Everling & Fischer, 1998; Hallett, 1978; Hallett & Adams, 1980; Jóhannesson & Kristjánsson, 2013; Kristjánsson, et al., 2001, 2004). Latencies were higher for antisaccades, which may reflect inhibition of the oculomotor system to prevent automatic saccades towards a cued location (Berggren et al., 2011; Hsu et al., 2020; Kristjánsson, 2011; Munoz & Everling, 2004; Stuyven et al., 2000). Latencies were also higher in the mixed than the fixed condition. This may reflect the additional load associated with maintaining two response codings for both prosaccades and antisaccades (Chan & DeSouza, 2013; Pierce et al., 2015; Tari & Heath, 2019).

To summarize, saccade latencies were in line with previous results. Additionally, longer saccade latencies along with lower microsaccade rates in the antisaccade conditions may reflect oculomotor inhibition. Pro- and antisaccades in Experiment 2 had even longer latencies in the mixed than the fixed condition, but this was not the pattern for the suppression of microsaccade rates. However, the longer response times and lesser rate suppression may also reflect uncertainty related to task switching in the mixed blocks, resulting in difficulties maintaining executive control over the oculomotor system and longer processing times of the constantly updating task.

Running rate analyses

We also assessed microsaccades using a “running rate” analysis since this technique has successfully been used in various tasks to observe microsaccade characteristics during temporal periods of particular interest (Abeles et al., 2020; Engbert & Kliegl, 2003a, b; Laubrock et al., 2005; Piras et al., 2015; Rolfs et al., 2005). To our knowledge, this is the first study of microsaccade running rates prior to saccades in the antisaccade paradigm. The running rate analyses allows the observation of microsaccade dynamics throughout the entire trial period before the onset of the response signal and an additional 50 ms after it. In Experiment 1 we observed a consistent decrease of microsaccade rates over time in all conditions.

The cluster permutation test confirmed a larger difference in microsaccade rates between pro- and antisaccades in the fixed blocks, with four significant clusters of differences in microsaccade activity, while no such evidence was found in the mixed block. There was also an interesting pattern where running rates showed a cyclic pattern within fixed antisaccade blocks. One possible explanation for this could be higher levels of executive control required for antisaccades aimed at suppressing reflexive saccades (Coe & Munoz, 2017). Another (perhaps more probable) explanation is increased noise due to monocular tracking, as these significant clusters were extremely short (the largest was 15 ms in duration) and this pattern was not observed in Experiment 2. Less strict filtering conditions to eliminate trials without microsaccades might result in more numerous spurious microsaccades.

The running rate analyses in Experiment 2 revealed a significant suppression of microsaccade rates on antisaccade trials. This occurred after the onset of the central cross followed by a reduction in microsaccade rate for both types of saccades and is consistent with similar microsaccade suppression seen with peripheral onsets (Dalmaso et al., 2019; Engbert, 2006; Engbert & Kliegl, 2003a, b; Laubrock et al., 2005; Shinn et al., 2020). This pattern was seen in both experiments but was more pronounced during fixed blocks. The suppression also differed in magnitude and timing: In the fixed block in Experiment 1 it was smaller due to increased fluctuations. Nevertheless, for antisaccade trials the time points at around 450 and 720 ms where lower rates are followed by their rebounds seem to be of interest; in Experiment 2 the global dip on antisaccade trials began sooner, at around 280 ms. Before that dip there was a small activity peak around 100 ms, which is close to the period when erroneous express saccades occur in humans when early suppression fails (Coe & Munoz, 2017; Kristjánsson, 2007). The rebound in Experiment 2 started around 600–680 ms, just prior to the saccade onset signal.

The pattern on antisaccade trials was similar in both experiments. The suppression of microsaccade activity was released immediately before the onset of the peripheral cue. This effect was larger on fixed trials, possibly reflecting that the additional load of switching between saccade types in the mixed condition either affected the strength of the suppression or delayed its release to a point that happened outside the 850-ms observation period.

Cluster permutation

Rather than test saccade type differences over time with preset bins, we used a cluster permutation analysis for grouped times determined by the data itself. The cluster permutation approach allows increased power requiring fewer comparisons (Barr, 2020). Again, this is, to our knowledge, the first use of such cluster permutation in estimating the temporal profiles of pre-response microsaccadic activity in the antisaccade paradigm. These analyses confirmed larger differences in microsaccade rates between pro- and antisaccade trials within fixed blocks. In Experiment 1, we found four significant difference clusters for the fixed trials and none for the mixed trials, where the difference between saccade types was nonsignificant. In Experiment 2, the rate differences were quite robust for the fixed condition with four clusters, but smaller for the mixed condition with one small cluster. Two clusters in the fixed condition—Cluster (1) at the beginning and Cluster (4) at the end—are notable. Since participants knew that suppression of saccades was needed, they seemed prepared for this quite early within the trial.

Overall, we observed significant differences in Experiment 1 in both latency and microsaccade rate, though we suspect the results may not be entirely reliable due to monocular recording. In Experiment 2 microsaccade rates prior to antisaccades were reduced, consistent with the hypothesis that microsaccade rates are influenced by which saccade type is being prepared. Specifically, suppression of the reflexive oculomotor system to prevent incorrect express saccades to the peripheral stimulus (Coe & Munoz, 2016) may have reduced overall microsaccade rates in the pre-stimulus period.

Interestingly, our results suggest that participants may have better control over microsaccadic eye movements than previously suggested (Buonocore et al., 2017). Our design may have encouraged participants to suppress most microsaccades prior to saccade generation since we enforced fixation with an error signal and a reminder to fixate and started each trial with drift correction. Both manipulations may have resulted in stronger ocular inhibition than otherwise. Nevertheless, in both experiments the suppression during fixed antisaccade trials was released more strongly right before the cue, while this release was not as strong in the mixed condition. This probably reflects the load associated with switching between randomly appearing pro- and antisaccade cues. No such release effect occurred for prosaccades in either the fixed or mixed conditions, probably because less suppression is required.

The cluster permutation test additionally showed larger differences in microsaccade activity on fixed trials in both experiments. On mixed trials, however, there were either no significant clusters of rate differences or they were very small. Since the saccade was executed immediately following the cue and the precue interval was 800 ms, there simply may not have been enough time to observe larger clusters. Participants may have been anticipating the cue flash and, combined with saccade preparation, they may have suppressed all oculomotor activity (including microsaccades, Hermens et al., 2010; see also evidence from monkey studies: Hafed et al., 2011). If there were any larger changes over time in microsaccade rate difference preceding the two types of eye movements, they may have been too subtle to be detected by our methods.

Future directions

We chose the microsaccade detection parameters adapted for a typical antisaccade paradigm. In future studies any effects on microsaccade rates might be increased by adding an intertrial interval instead of the manually controlled onset of a trial, as there have been reports of microsaccade suppression due to motor task preparation (Betta & Turatto, 2006; Olmos-Solis et al., 2017). Other possibilities may include increasing the pre-cue period (e.g., Dalmaso et al., 2020) and eliminating enforced fixation and drift correction to include fixational microsaccades in the processed data, as the rate of such microsaccades drops closer to the onset of stimulus presentation, especially on antisaccade trials (Watanabe et al., 2013). In addition to microsaccade rate, pupil size dynamics could also be a measure of cognitive load during observations of microsaccade rates (Dalmaso et al. (2020), and to measure superiocollicular neural activity (Corneil & Munoz, 2014; Wang et al., 2015, 2017). However, careful control for luminance differences for pro- and antisaccades would be required (Mathot, 2018).

Conclusions

Our results suggest that microsaccades can be used to measure oculomotor suppression during antisaccades. This may be important for current theories that suppose that suppression of the reflexive oculomotor system may gate other attentional mechanisms. For example, Redden and colleagues (2021) have suggested a model of Inhibition of return that changes form (input or output) based on whether the reflexive oculomotor system is in a state of inhibition. Further, our cluster permutation analyses may set a new direction for investigating the role of these micromovements and may provide a framework for their study.