Abstract
When presented with a periodic stimulus, humans spontaneously adjust their movements from reacting to predicting the timing of its arrival, but little is known about how this sensorimotor adaptation changes across development. To investigate this, we analyzed saccade behavior in 114 healthy humans (ages 6–24 years) performing the visual metronome task, who were instructed to move their eyes in time with a visual target that alternated between two known locations at a fixed rate, and we compared their behavior to performance in a random task, where target onsets were randomized across five interstimulus intervals (ISIs) and thus the timing of appearance was unknown. Saccades initiated before registration of the visual target, thus in anticipation of its appearance, were labeled predictive [saccade reaction time (SRT) < 90 ms] and saccades that were made in reaction to its appearance were labeled reactive (SRT > 90 ms). Eye-tracking behavior including saccadic metrics (e.g., peak velocity, amplitude), pupil size following saccade to target, and blink behavior all varied as a function of predicting or reacting to periodic targets. Compared with reactive saccades, predictive saccades had a lower peak velocity, a hypometric amplitude, smaller pupil size, and a reduced probability of blink occurrence before target appearance. The percentage of predictive and reactive saccades changed inversely from ages 8–16, at which they reached adult-levels of behavior. Differences in predictive saccades for fast and slow target rates are interpreted by differential maturation of cerebellar-thalamic-striatal pathways.
SIGNIFICANCE STATEMENT From the first moments of life, humans are exposed to rhythm (i.e., mother's heartbeat in utero), but the timeline of brain development to promote the identification and anticipation of a rhythmic stimulus, known as temporal prediction, remains unknown. Here, we used saccade reaction time (SRT) in the visual metronome task to differentiate between temporally predictive and reactive responses to a target that alternated at a fixed rate in humans aged 6–24. Periods of age-related change varied little by target rate, with matured predictive performance evident by mid-adolescence for fast and slow rates. A strong correlation among saccade, pupil, and blink responses during target prediction provides evidence of oculomotor coordination and dampened noradrenergic neuronal activity when generating rhythmic motor responses.
Introduction
From an early age, we are exposed to rhythms in our environment. In the womb, we listen to our mother's heartbeat, and as an infant, we find comfort in the swaying of a rocking chair. Attending to stimuli with periodicity has been shown to induce physiological relaxation (Johnson and Trawick, 1938; Brauchli et al., 1995). Identifying rhythmicity and shifting motor behavior from reacting to predicting an upcoming stimulus, known as temporal prediction, occurs spontaneously and is fundamental for adaptive sensorimotor behavior (Fitch, 2013; van der Steen and Keller, 2013). Neural maintenance of an interstimulus interval (ISI) and coordination of endogenously driven motor commands to anticipate its arrival are required to predict rhythmic stimuli (Repp and Su, 2013). Brain regions important for timing-related signals include the frontal cortex (Maimon and Assad, 2006; Jazayeri and Shadlen, 2015), basal ganglia (Lee and Assad, 2003; Turner and Anderson, 2005), cerebellum (Ashmore and Sommer, 2013), and thalamus (Matsuyama and Tanaka, 2021). Neuronal firing in these areas correlates with the ISI of a periodic target, and pharmacological silencing impairs predictive movements (Buhusi and Meck, 2005; Merchant et al., 2013; Matsuyama and Tanaka, 2021; Tanaka et al., 2021). Additionally, an impaired ability to anticipate rhythmic stimuli to respond predictively is evident in multiple brain-related disorders, including developmental coordination disorder (Debrabant et al., 2013), developmental dyslexia (Lukasova et al., 2016), Huntington's disease (Vaca-Palomares et al., 2019), and spinocerebellar atrophy (Bares et al., 2007). Despite the important role temporal prediction plays in essential sensorimotor learning, and its clinical relevance to various neurologic conditions, its developmental timeline in normal controls remains unknown.
One commonly used method of measuring temporal prediction is the visual metronome task (Stark et al., 1962), where participants are asked to move their eyes in time with a square-wave target that alternates at a fixed rate between two known locations. Saccades initiated before neural registration of the visual target, in anticipation of its upcoming temporal appearance, are categorized as predictive [saccade reaction time (SRT) < 90 ms; Fischer et al., 1997; Munoz et al., 1998; Lee et al., 2016). Saccades initiated in response to the visual target are considered reactive (SRT > 90 ms). Analysis of saccade behavior in this task is particularly useful for estimation of brain functioning of areas relevant to temporal prediction given that they overlap with brain regions required for saccade initiation (i.e., brainstem, cerebellum, superior colliculus (SC), thalamus, basal ganglia and frontal cortex). Based on a handful of studies using the visual metronome task in children [2–53 child participants per study (age range: 4–15 years)], it is clear that compared with adults, children make fewer predictive saccades and on average, have SRTs that are on average, 100–150 ms slower (Kowler and Martins, 1982; Ross and Ross, 1987; Ross et al., 1994; Lukasova et al., 2018). However, given the small sample sizes of these studies and paucity of adolescent participant data, normal predictive saccade performance from early childhood to young adulthood has yet to be characterized.
To investigate temporal predictive performance across developmental ages, we measured predictive saccades in 114 healthy participants, aged 6–24 years, performing the metronome task at five different target rates. Because pupil size has reliably been shown to correlate with noradrenergic modulation by the locus coeruleus norepinephrine (LC-NE) system (Alnæs et al., 2014; Murphy et al., 2014), we sought to test whether previous reports of autonomic relaxation while attending to rhythm (Johnson and Trawick, 1938; Brauchli et al., 1995) could be replicated based on evidence of a decreased pupil size when participants predict versus react to targets, indicative of dampened LC-NE signaling. Finally, we tested whether eyeblink timing varied based on participant prediction or reaction to targets. To ensure that age-related changes in performance were not driven by improvements in processing speed or oculomotor kinematics, participants also performed a random task, which is identical to the metronome task, except the timing of the target is randomly selected for each target step, making the target steps unpredictable.
Materials and Methods
Study participants
The research protocol was approved by Queen's University Faculty of Health Sciences (protocol ID: PHYS-007-97). A total of 118 participants completed the study (mean age: 14.9 ± 3.8 years; 55.1% female). Adults aged ≥18 years provided their written consent. Children aged <18 years gave their verbal assent and parents/legal guardians provided written informed consent. All participants were free of neurologic, psychiatric or ocular diagnoses and were not taking psychotropic medications, and participants had normal or corrected-to-normal vision.
Recording of eye movements
Participants were seated 60 cm away from a computer screen with their head position stabilized by a fixed head mount and chin rest at a viewing angle of 32 × 26°. Experimental stimuli were presented as 0.5° diameter sized circular red targets (luminance measured 44 cd/m2 with an optometer for LCD monitors) on a 17-inch LCD iiYama Prolite monitor at a screen resolution of 1280 × 1024 pixels with a 60-Hz refresh rate. Experiments were completed in the dark with the only light source being the experimentally controlled stimuli. Monocular eye tracking was conducted (right eye position was measured) using the video-based eye tracker EyeLink 1000 Plus in 83 participants and the EyeLink II in 35 participants (unpublished data previously collected in the lab that used the same experimental task; SR-Research Ltd.). Both cameras of the EyeLink systems had a sampling rate of 500 Hz and a mean eye position accuracy of ≤0.5°. Eye movements to a nine-point calibration grid were performed by each subject before each experiment, with eye position accuracy within 1° of the visual target considered acceptable. Eye position was reassessed for drift after every five trials of target presentation and calibration was performed again if needed.
Experimental design
Participants were cued with a red central fixation point (FP) on a blank screen that had a random offset (i.e., the FP disappeared) between 1 and 1.5 s. Coincident with the disappearance of the FP, a peripheral target appeared 10° right or left from the central FP, on the horizontal axis. The target alternated between the two fixed locations on the horizontal plane for a total of 12 target steps (i.e., six in each direction; see Fig. 1A for task design). In the visual metronome task, a square-wave target alternated consistently at one of the five target rates (0.66, 0.8, 1, 1.33, and 2 Hz). These five target rates correspond to the following ISIs: 1500, 1250, 1000, 750, and 500 ms, respectively. Participants were instructed to move their eyes in time with the jumping target given both the target location and timing of the next target were predictable after the first presentation. In the random task, one of the five ISIs used in the metronome task was randomly used for each target step, such that the target location was predictable, but the timing of the next target appearance was unpredictable. The randomization was set so that the same ISI could not appear more than three times in a row. Thus, the only difference between the metronome and random task was the temporal predictability of target appearance. Participants were offered a break in between the two saccade tasks to assure alertness. The order of the metronome and random tasks was counterbalanced across participants. All stimulus timings were verified independently with a photosensor.
Eye-tracking analysis
Offline analyses of eye-tracking data were performed using MATLAB version R2019b (MathWorks). Saccades were identified based on their instantaneous velocity which was calculated on a three-point moving average of x and y eye positions in degrees. When the instantaneous velocity exceeded 2 SD above the mean fixation velocity (defined as <50°/s) for at least five continuous points, a saccade was labeled.
SRT
SRT was computed by subtracting the time of peripheral target appearance from the time of primary saccade onset. Saccades were labeled as predictive, express, and regular based on their SRTs, being <90, 90–120, and >120 ms, respectively (Fig. 1B,C; Fischer and Ramsperger, 1984; Fischer and Weber, 1993; Dorris and Munoz, 1998; Munoz et al., 1998). The 90-ms cutoff for predictive saccades was based on data by Munoz et al. (1998) which demonstrated that pro-saccades to one of two potential target locations were correct only 50% of the time when initiated before 90 ms after target appearance, whereas those with a SRT >90 ms were correct >95% of the time. Therefore, 90 ms is the lower limit of time to allow visual processing and motor reaction to a peripheral external target, and saccades with SRTs below this cutoff are internally generated. Express and regular saccades are both reactive (e.g., made in response to a visual target), but have unique SRT distributions. Express saccades are the fastest visually evoked eye-movements and are generated when the incoming visual transient signal to the SC is transformed directly into a saccade motor command (Edelman and Keller, 1996; Dorris et al., 1997). Previous research supports that the express saccade epoch varies based on the specific parameters of an eye-tracking task, for example variations to target luminance can alter the timing of visual response latencies in the SC (Marino et al., 2015). Moreover, it is known that variables including target eccentricity, target predictability, previous training, and the presence or absence of a gap between fixation and target appearance all influence the probability of express saccades in a given behavioral task (Weber et al., 1992; Fischer and Weber, 1993; Paré and Munoz, 1996; Dorris and Munoz, 1998). Thereby we sought to quantify the SRT range of express saccades in the metronome and random task, respectively, and later assess the effect of target properties (e.g., rate) on their occurrence. We plotted the distribution of saccades with a SRT >90 ms made in the metronome and random task in Figure 1C. A distinct dip in the SRT distribution is seen at 120 ms for both experimental tasks, supported by a bimodality coefficient (BC) of 0.63 in the metronome task and 0.54 in the random task. Correspondingly, the express epoch range for saccades in the current paper was defined as 90–120 ms, inclusive, and regular saccades were defined as those with reaction times >120 ms. The percentage of predictive, express, and regular saccades were computed for each participant for comparison across trial conditions (metronome task's five target rate conditions and random task; six conditions in total), as well as analyzed across participant age.
Saccade metrics
Saccadic amplitude and peak velocity were analyzed for predictive, express, and regular saccades to all target steps made in the metronome task. To ensure quality of saccade data, we removed saccades with a maximum peak velocity value above 1000°/s (N = 81/34 497; 0.23% of saccades), while still including any remaining saccades in the trial if they had a peak velocity below 1000°/s. To compare the main sequence for predictive, express, and regular saccades, peak velocity versus amplitude were plotted for each saccade type. A square root model (
Pupil size
To evaluate whether physiological arousal varied based on the reaction time to the alternating target, we analyzed pupil size in the metronome task. Pupil size, recorded every 2 ms, was averaged over a 200 ms epoch following the end of a saccade, provided the participant remained fixated on that location for a minimum of 200 ms. This window length was selected to avoid any pupil size changes induced by the pupillary light reflex if the target appeared postsaccade (i.e., a source of light stimulates pupil constriction ∼300 ms following its appearance on a dark background; Ellis, 1981; Wang et al., 2018). Pupil trials that met the minimum fixation length criterion were excluded from analysis if a blink or saccade occurred during peripheral target fixation, if the timing of fixation onset exceeded the minimum gap (100 ms) between fixation and next target appearance, and if pupil velocity fell outside the range of −5000°/s to 5000°/s. This left 19175 viable pupil trials in the metronome task to analyze (2 Hz: 4093 trials; 1.33 Hz: 3904; 1 Hz: 3845 trials; 0.8 Hz: 3794 trials; and 0.66 Hz: 3539 trials), and these trials were divided into three categories based on SRT: predictive (6094 trials), express (2384 trials), and regular (10697 trials). To query whether pupil size varied based on prediction or reaction to the rhythmic target, two analyses were performed. First, a one-way ANOVA was computed to test for a main effect of saccade type (predictive, express, and regular) on pupil size. Second, saccade types were combined to test whether SRT correlated with pupil size using a Spearman rank correlation. To test whether pupil size varied across the developmental age range, baseline pupil size for predictive, express, and regular saccades were analyzed per individual subject.
Blink metrics
Inspection of loss of eye-tracking revealed a consistent profile of blinks with a duration of 50–300 ms, which is supported by the literature (Caffier et al., 2003; Betke and Chau, 2005). Frequency of blinks during the presence of right and left peripheral targets in the visual metronome task were compared to examine whether participants exercised a spatial bias. No bias was observed (1876 and 1864 blinks made during right and left targets, respectively; χ2 test of proportions = 0.06, p = 0.80) and blinks made during either target direction were collapsed for analysis. Blink rate was recorded for every trial starting 1 s before the appearance of the first target, continuing to the trial's end. Blinks that were not yet completed by the end of the trial were excluded from analysis as their duration was unknown. The number of blinks was summed per trial and divided by the trial length to compute blink rate (blinks/min) per participant.
To evaluate the timing of blinks in the metronome task, blink reaction times were characterized relative to target appearance for each target rate. The probability that participants made a blink on a given trial was computed for three epochs relative to target appearance (0 ms): −1000 to 1000, −1000 to 0, and 0 to +1000 ms. Blink probabilities were then categorized by the type of saccade (predict; express; regular) generated within −1000 and +1000 ms of target appearance and analyzed to understand whether the timing of blinks on a given trial varied based on whether the participant predicted or reacted to the target.
Statistical analysis and model selection
Multivariate tests revealed no difference in parameters between the Eyelink tracking systems [multivariate test of participant mean saccade SRT, amplitude and peak velocity (F(18,94) = 1.19, Wilk's λ = 0.815, p = 0.288), participant mean pupil size for predictive, express, and regular saccades by system (F(3,108) = 0.90, Wilk's λ = 0.975, p = 0.442), and mean blink rate for the five target rate conditions by system (F(5,108) = 1.50, Wilk's λ = 0.935, p = 0.195)]. Therefore, eye-tracking data were collapsed for analysis. Generalized additive models (GAMs; Wood, 2017) were used to estimate the effect of age on eye-tracking parameters. GAMs were selected for their semiparametric nature, robustness to overfitting, and ability to query age-related effects across adolescence without assuming the shape of the developmental trajectory (Wierenga et al., 2019; Luna et al., 2021). Age was entered as a smooth function in the GAM model
To identify periods of statistically significant developmental change and age of maturation, the uncertainty of each estimated model was calculated using confidence intervals. Confidence intervals (95%) were computed via posterior simulation, a process described by Wood (2017) and previously implemented to identify periods of age-related changes across adolescence in white matter and brain functional connectivity (Simmonds et al., 2014; Calabro et al., 2020). During posterior simulation, 1000 random draws from a multivariate normal distribution whose vector of means and covariance corresponded to the fitted GAM parameters were taken, therefore each random draw represented a new trend that was compatible with the fitted trend but also reflected the uncertainty in the estimated trend (Simpson, 2018). The first derivative of the GAM fit was computed (at 0.1-year age intervals) for the 1000 random draws to identify time points
In general, statistical analyses were performed on population-level (i.e., averaged participant responses) data for saccade, blink, and pupil responses and on an individual trial level (i.e., trials with predictive saccades, for example). Before any group-level analysis was performed, normality of response parameters was assessed using the Kolmogorov–Smirnov test. Parametric tests were performed for normally distributed data and nonparametric tests were performed for skewed data. Means, SDs, and confidence intervals were reported for statistically significant main effects. All main effects were further assessed using t tests and reported significant p values were corrected for multiple comparisons using the Bonferroni method. t tests and regression analyses were two-tailed.
Results
Data were excluded from three participants in the random task and four participants in the visual metronome task because of poor quality eye-tracking. Therefore, statistical analyses were performed on 115 participants in the random task and 114 participants in the metronome task (see Fig. 1D for the distribution of participants' age and gender).
A, In the metronome task (first developed by Stark et al., 1962), participants were cued with a central FP with a random interval offset (1000–1500 ms) followed by 12 targets that alternate 10° right (R) and left (L) from center at a fixed target rate. Five target rates (0.66, 0.8, 1.0, 1.33, 2.0 Hz) were delivered over five blocks of trials with 12 targets each (60 targets per stimulus rate), with a pseudorandom trial order. These target rates correspond to an ISI of 1500, 1250, 1000, 750, and 500 ms, respectively. Participants were instructed to move their eyes in time with the targets. In the random task, target rates were randomized such that the participant could not anticipate the timing of the upcoming target while all other aspects of the task were held constant. B, Schematic of the characterization of saccades by SRT: predictive (SRT < 90 ms) in dark blue, express (SRT: 90–120 ms) in light blue, and regular (SRT > 120 ms) in red. C, Example eye position data collected from a single subject over a single trial. Colored bars correspond with the type of saccade generated according to its SRT toward the alternating target. D, The distribution of saccades with a SRT > 90 ms in the metronome and random task is bimodal, with an express saccade epoch from 90 to 120 ms (solid light blue line) and regular saccade SRTs being >120 ms in both task conditions. The BCs for the distribution of SRTs >90 ms in the metronome and random task were 0.63 (skewness = 4.7; kurtosis = 33.5) and 0.54 (skewness = 4.3; kurtosis = 33.6), respectively. E, Histogram of participants' ages in the metronome and random task that were included for experimental analysis (bin width = one year). White boxes represent female participants and dark blue boxes represent males.
Task metrics
SRT
Figure 2 shows all participants' SRTs to target steps 1–12 for each trial of the visual metronome task (∼7000 saccades per target frequency; 34497 total saccades) and the random task (35811 saccades). In the metronome task, the first three target steps largely consisted of regular and express saccades (SRT > 90 ms), whereas saccades to target steps 4–12 showed an increased incidence of predictive behavior (SRT < 90 ms). Participants therefore used the first approximately three target steps to identify the rate of the alternating target before launching predictive saccades to anticipate the timing of the next target's appearance, in agreement with previously published predictive saccade results (Joiner and Shelhamer, 2006; Zorn et al., 2007). Nonparametric kernel density curves of SRTs in Figure 2A (predictive in red; express in light blue; regular in dark blue) show the probability of a given reaction time in the three saccade categories. Although the SRT range for each category was unique, the kernel nature of the density curves displays an artificial overlap among distributions. As the rate of the alternating target slowed from 2 to 0.66 Hz, the probability peak of predictive saccades correspondingly decreased and became broader (Fig. 2, red traces).
All subjects' SRTs in the metronome and random tasks. Individual subjects' SRT to each target step is plotted from the top to the bottom according to the trial number. For example, an individual subject's SRT to target step 1 in trial 1 is plotted on top of the figure, followed by that same individual's SRT to trial 2's target step 1 and so forth until trial 5's target step 1; then the next subject's SRT data to target step 1 is plotted. This is continued for all subjects across the five blocks of 12 target steps. Frequency of saccade type is visualized with a kernel density estimator. The bandwidth of the kernel smoothing window for SRTs was optimized by MATLAB's scatterhist function. SRT in ms. Predictive saccades (red; SRT < 90 ms), express saccade (light blue; SRT: 90–120 ms), regular saccade (dark blue; SRT > 120 ms).
A main effect of metronome target rate was observed for percentage of regular (F(4,565) = 25.70, p = 1.39e-19, effect size η2 = 0.154) and predictive saccades (F(4,565) = 22.16, p = 5.18e-17, effect size η2 = 0.136) but not express saccades (F(4,565) = 1.26, p = 0.283). t tests revealed that the percentage of predictive saccades significantly varied among the following frequencies: 2 Hz [mean percentage: 55.6% (SD = 22.3)] and 0.8 Hz [42.2% (20.5); p = 3.10e-05], 2 and 0.66 Hz [34.9% (20.0); p = 1.09e-11], 1.33 Hz [57.5% (22.9)] and 0.8 Hz (p = 1.00e-06), 1.33 and 0.66 Hz (p = 9.89e-14), and 1 Hz [50.1% (21.1)] and 0.66 Hz (p = 1.00e-06). Target rates that differed by the percentage of regular saccades included: 2 Hz [33.9% (19.1)] and 0.8 Hz [47.7% (21.3); p = 6.00e-06], 2 and 0.66 Hz [56.7% (21.0); p = 8.13e-15], 1.33 Hz [33.6% (21.3)] and 0.8 Hz (p = 4.00e-06), 1.33 and 0.66 Hz (p = 3.86e-15), and 1 and 0.66 Hz (p = 2.25e-08). In summary, as target rates slowed, participants generated fewer predictive saccades and more regular saccades, while the frequency of express saccades remained unchanged.
Age and SRT
GAM models, corrected for multiple comparisons (required p value for significance = 0.05/5 target rate conditions; p = 0.01), revealed a significant relationship among age and percentage of predictive saccades for the following target rates in the metronome task (Fig. 3, red data points and traces; adjusted r2 values and period of age-related changes are reported in the figure): 2 Hz (deviance explained = 15.5%, p = 0.0011), 1.33 Hz (22.5%, p = 2.29e-05), 1 Hz (10.1%, p = 0.00843), and 0.66 Hz (8.58%, p = 0.01). Age of maturation for predicting short and medium target rates (2, 1.33, 1 Hz) were highly similar (∼14 years), whereas maturation for the longest target rate (0.66 Hz) was delayed until age 16. There was no significant age × percentage of predictive saccade effect observed for 0.8 Hz (p = 0.10).
Age-related changes of temporal saccadic prediction according to each of the metronome task's target frequencies (A–E) and in the random task (F). GAMs were applied (% of predict, express, regular × age) for each target condition and corrected for multiple comparisons. Solid lines represent significant interactions between percentage of saccade type and age (p < 0.01). 95% confidence intervals are shown as semi-transparent banding along developmental curves of statistical significance. Lines without shaded confidence bands reflect models of nonsignificance. Adjusted r2 values and periods of significant age-related change are reported for each significant model in the upper right panel of the plots. The maturation point is the upper limit of the significant period of age-related change where task performance reached adult-level performance. Predictive saccades (red; SRT < 90 ms), express saccade (light blue; SRT: 90–120 ms), regular saccade (dark blue; SRT > 120 ms).
The paucity of predictive saccades to rhythmic targets in young children across all presentation rates coincided with a higher prevalence of regular saccades. GAM models revealed a significant inverse relationship between age and percentage of regular saccades in all target rates of the metronome task (Fig. 3, dark blue data points and traces). Significant periods of age-related change for percentage of regular saccades in the 2-Hz condition were observed from 7.5–14.1 years (22.8%, p = 1.87e-05), 7.6–13.8 years for 1.33 Hz (30.5%, p = 3.91e-07), and 7.4.−16.7 years for 1 Hz (14.0%, p = 5.12e-04), 6.5–23 years for 0.8 Hz (7.51%, p = 0.00319), and 7.8–17.3 years for 0.66 Hz (13.2%, p = 6.51e-04). The percentage of express saccades × age GAM models revealed a significant effect of age on express saccades for targets that alternated at 0.8 Hz (Fig. 3D, light blue data points and traces), with 8.98% deviance explained (p = 0.00121). In the random task, a significant interaction between age and the percentage of express saccades was observed (14.1%, p = 0.00914).
Metronome saccade metrics and age
A multivariate ANOVA was run with individuals' mean saccade amplitude and peak velocity toward all target steps entered as dependent variables and saccade type in the metronome task as the independent variable. A main effect of saccade type was observed on the saccade metrics [F(4,674) = 28.89 (p = 3.49e-22), partial η2 = 0.146]. As expected (Bronstein and Kennard, 1987; Evans et al., 1999; Shelhamer and Joiner, 2003; Wong and Shelhamer, 2011) predictive saccades were hypometric [mean = 17.7° (SE = 0.16)] relative to the 20° targets, likely because the target appearance was not yet visually processed when the saccade was initiated, resulting in increased neural noise in the brainstem for the saccadic command. Regular saccades were also hypometric [16.4° (0.16)], although we attribute this to the abundance of regular latency saccades to the first target step 10° away from FP (see Fig. 4D). Amplitudes differed among predictive and regular saccades (p = 8.70e-09), express [18.0° (0.16)] and regular saccades (p = 2.90e-12), but not among predictive and express saccades. Peak velocity differed among all saccade types (mean peak velocity for predictive saccades = 433.9°/s; express = 496.4°/s; regular = 459.1°/s; p < 0.001).
To evaluate the slope of the main sequence by saccade type, a square root model (
A, Square root model (
Square root model fits were derived to estimate the main sequence of individual subjects' predictive, express, and regular saccades (including all amplitudes) in Figure 4B–D and a Kruskal–Wallis test was performed to test whether they shared the same underlying distribution. A main effect of saccade type was observed on the model fits (χ2(2,336) = 47.76, p =4.26e-11), where model coefficients varied between predictive and express saccades (p < 0.001) and predictive and regular saccades (p < 0.001), but not between express and regular saccades (p = 0.81; Fig. 4A). Spearman ρ correlations were run for age and the main sequence model fits for the three saccade types, and as expected (Fischer et al., 1997; Munoz et al., 1998; Luna et al., 2001), there was no observed association with age (Fig. 4E).
Pupil size, temporal prediction, and age
All participant pupil size values were combined (Fig. 5C) and categorized according to their saccade (e.g., predictive, express, regular) to target on given metronome trial (Fig. 5A). A Kruskal–Wallis test supported a main effect of saccade type on pupil size in the metronome task (χ2(2) = 369.82, p = 4.94e-81), with post hoc t tests supporting differences among all saccade types (p < 0.001; see Fig. 5A,B). In Figure 5C, participants' corresponding pupil size for all metronome trials that met the pupillary analysis criteria are plotted against the RT of the saccade to target, showing an increased density of bigger pupil sizes at around the time that the visual transient signal is converted to a saccade motor command (SRT > 90 ms). A significant positive Spearman's rank-order correlation was observed between pupil size and SRT (
On metronome trials when participants fixated for a minimum of 200 ms following the completion of saccade to target, pupil size was recorded at 2-ms intervals (Eyelink camera sampling rate: 500 Hz) to test whether sympathetic arousal, indicated by pupil size, varied based on the participant's reaction time to square-wave target. Metronome trials that met the criteria for pupillary analysis were combined across subjects and target frequencies are plotted in A–C. Mean pupil size (A), from the time of the saccade completion (time point: 0 ms) to 200 ms following the saccade, significantly differed according to saccade type (predictive in red; express in light blue; regular in dark blue), with smaller pupil sizes observed for predictive versus reactive saccades [mean pupil size (in pixels) for predictive saccades = 2.67e+03, express = 2.937e+03, and regular = 2.9702e+03]. B, Variability in pupil sizes averaged in A is shown in a cumulative probability plot. Analysis of the SD of participants' pupil size revealed that they did not significantly differ by saccade type. C, Participants' pupil sizes positively correlated with the reaction time of their saccade to target (
Blinks
Inspection of blink and saccade probability density plots in Figure 6A revealed an anti-correlated relationship between saccades and blinks. During visual fixation, participants show the highest rate of blinking, which is reduced during target steps 1 and 2, and increased again for target steps 3–12. Participants' mean blink rate did not vary according to the metronome task's target frequency (p = 0.323; Fig. 6B), nor did it vary among metronome target rates and the random task. On the other hand, the timing of blink onsets relative to target appearance in the metronome task significantly varied (χ2(4) = 203.73, p = 5.91e-43; see Fig. 6C). Multiple comparison tests showed differences in blink reaction times (relative to target appearance) between all target frequencies at the p < 0.001 level (participants' median blink reaction times: 262.2 ms for 2 Hz, 417.5 ms for 1.33 Hz, 501.3 for 1 Hz, 584.7 for 0.8 Hz, and 678.8 for 0.66 Hz), with the exception of frequencies 1.33 and 1 Hz (p = 0.0155), 1 and 0.8 Hz (p = 0.0108), and 0.8 and 0.66 Hz (p = 0.0569). The relationship among participants' age and their corresponding mean blink rate and average blink reaction time to target was queried tested using Spearman's rank correlation coefficient tests across the metronome target frequencies and corrected for multiple comparisons. No correlation was observed among participants' age and blink reaction time (range of ρ values: −0.005–0.20). Similarly, no relationship was observed among participants' age and mean blink rate with the exception of target frequency 0.66 Hz (ρ = 0.34; p = 5.15e-04).
A, Diagram of blinks (in purple) and saccades (in gray) made toward the alternating target of the metronome task at target frequency 0.66 Hz. Blinks and saccades are displayed such that each horizontal line marks a unique trial from a participant, making up 570 occurrences (114 participants × 5 trials of the alternating target at a rate of 0.66 Hz), and gray borders mark the timing of the target appearance. Histogram of blink probability (in purple) is scaled up by a factor of 5 so blink and saccade probabilities can be viewed on the same plot. B, Moving average of blink rate for each of the five metronome frequencies and the random task across trial lengths. C, A main effect of mean blink reaction time by target frequency was observed, with gray bars indicating the frequencies that were significantly different at the post hoc level. Crosses show outliers (±2 SD away from the mean). D–F, Probability of a blink or saccade 1000 ms before and after target appearance (0 ms) collapsed across all metronome trials. To visualize blink and saccade probabilities on the same graph, blink probabilities are multiplied by 20. Color of blink traces reflect the saccade type color scheme [predictive (red), early express (light blue), and regular (dark blue)]. G, Unscaled blink probability categorized by saccade type. A main effect of saccade type was observed on blink probability across the 2000-ms epoch (1000 ms before and 1000 ms after target appearance); χ2(2) = 18.99, p = 7.53e-05.
Mean probabilities of participants making a saccade or blink during the 1000 ms before and after target appearance are plotted in Figure 6D–G, categorized by saccade type. Participant mean blink probability differed by saccade type across the −1000- to +1000-ms epoch relative to target appearance at 0 ms (2000-ms epoch total; χ2(2) = 18.99, p = 7.53e-05]. Multiple comparisons showed that participants were more likely to make a blink during this epoch if they generated a regular saccade to target as opposed to a predictive (p = 0.0059) or express saccade (p = 0.0001). In the 1000-ms period before target appearance, blink probability significantly varied by saccade type (χ2(2) = 32.35, p = 9.43e-08), with post hoc differences among all saccade types (p < 0.001). In the 1000-ms period following target appearance, blink probability also significantly varied by saccade type (χ2(2) = 10.54, p = 0.005), with all saccade types differing at the post hoc level [predictive and express (p = 0.0135), predictive and regular (p < 0.001), and express and regular (p = 0.0135)]. These data support a novel finding in that when participants generated a predictive saccade, blinks were more likely to be inhibited in the 1000 ms prior to target appearance. Indeed, compared to visually triggered saccades, blinks on predictive saccade trials were more likely to occur in the 1000 ms following target appearance. Spearman's rank correlation coefficient tests revealed that blink probability across these tested epochs did not significantly vary with participant age in any of the saccade categories.
Discussion
The goal of this study was to better understand temporal prediction of visual metronome targets in typically developing children, adolescents, and young adults. Temporal prediction significantly correlated with age in four of the five metronome target rates with adult-level performance being reached by age 16. Smaller pupil sizes were observed following a predictive versus reactive saccade, supporting autonomic relaxation when anticipating rhythmic stimuli. Blink rate did not differ among metronome and random targets, however, blink timing varied with target frequency and saccade type; blinks occurred at the middle of the ISI and were suppressed before target appearance on trials when participants predicted the target. Together, these results support a strong age effect on the temporal prediction of periodic targets and a coordination among saccade, pupil, and blink metrics.
Maturation timeline of saccadic temporal prediction
Predictive saccades increased with age for nearly all target rates in the metronome task (Fig. 3A–E, red curves). This was not observed in the random task (Fig. 3F), indicating that changes in performance were specific to conditions that had rhythmic stimuli, rather than any changes in anticipatory saccades toward visual targets in general. Developmental curves for predictive and regular saccades across target rates were inversely related, illustrating a shared maturation timeline; while the percentage of express saccades changed only slightly with age (Fig. 3D,F). These data are in agreement with previous reports that young children (<12 years) made fewer predictive saccades than adults to target frequencies 0.5–1.33 Hz (Kowler and Martins, 1982; Ross and Ross, 1987; Lukasova et al., 2018). Similar to adult studies (Shelhamer and Joiner, 2003; Isotalo et al., 2005; Lee et al., 2016), we found that the highest rates of saccadic prediction occurred toward target frequencies of 1.33 and 2 Hz. Likewise, in subjects aged 4–38 years, the preferred motor tempi for spontaneous finger tapping was found to range from 1–2 Hz (McAuley et al., 2006). (McAuley et al., 2006). This frequency range closely aligns with the human heart rate (80 BPM = 1.33 Hz), raising interesting questions about whether individuals can more adeptly predict tempos that synch to their heart rate, or if tempo preference is shaped by exposure to a heartbeat in utero.
The prolonged period of age-related change for low-frequency (i.e., 0.66 Hz) versus high-frequency (i.e., 1–2 Hz) targets may reflect additional signaling of the striatum, thalamus and/or cerebral cortex during middle adolescence. The cerebellum and basal ganglia have unique temporal sensitivity for high and low frequencies (Ivry and Spencer, 2004; Buhusi and Meck, 2005; Allman et al., 2014). For example, when listening to auditory stimuli, patients with cerebellar lesions had reduced surface level EEG amplitude for beats with high frequencies, whereas patients with basal ganglia lesions demonstrated reduced EEG activity for low frequencies (Nozaradan et al., 2017). Additionally, pharmacological silencing of the cerebellar dentate in monkeys impaired self-timed SRTs to fast versus slow-paced periodic targets (Kunimatsu et al., 2018). In a combined eye-tracking and fMRI study, Debrabant et al. (2013) observed that compared with age-matched controls, children diagnosed with developmental coordination disorder made fewer predictive saccades to a periodic target (ISI: 1200 ms) and exhibited less BOLD activity in cerebellar Crus I. The maturational timelines of predictive saccades in our participants toward target frequencies ≥1 Hz were remarkably similar, varying by only 1.3 years from the onset of performance-related change to its conclusion (Fig. 3A–C). These data, combined with previous reports of the specificity of cerebellar signaling for coordinating motor movements toward target frequencies ≥1 Hz (see Ivry and Spencer, 2004), together imply that in healthy children, the period of age-related cerebellar maturation to support saccades toward fast-paced rhythmic tempos occurs from ages 8 to 14.
Previous nonhuman primate research has causally demonstrated that the caudate closely maintains neuronal activity to match the delay interval of an ISI, while, neuronal activity in the dentate is relatively constant for various target frequencies, showing characteristic ramping activity 500 ms before the self-timed saccade and peaked activity at saccade execution (Ohmae et al., 2017). These results may explain why we observed fewer predictive saccades at the lowest frequency of 0.66 Hz, given low frequencies demand prolonged neuronal firing in the striatum to accurately signal the timing of saccade execution. Longitudinal cohort studies support age-related change of caudate gray matter volume from early childhood into young adulthood (Wierenga et al., 2014; Larsen and Luna, 2015). Furthermore, gray matter atrophy in the putamen in Huntington's disease patients was negatively correlated with predictive saccade SRT (Vaca-Palomares et al., 2019). Other potential neural correlates for the prolonged maturation of predicting targets at low frequencies include the thalamus, which acts as a relay region for a disynaptic connection between the dentate and striatum (Hoshi et al., 2005; Bostan et al., 2013). The thalamus integrates timing-related subcortical signals with cortical signals (Matsuyama and Tanaka, 2021) and has strong buildup activity for self-timed saccades (Tanaka, 2006, 2007). Additional cortical regions that modulate their firing during timing-related tasks and may further contribute to young adolescents' delayed temporal performance at 0.66 Hz include the Frontal Eye Fields (FEF) (O'Driscoll et al., 2000; Gagnon et al., 2002) and parietal cortex (Maimon and Assad, 2006; Jazayeri and Shadlen, 2015).
Pupil size and temporal prediction across age
Previous studies have reported a physiological relaxation effect when subjects were presented with rhythmic stimuli (Johnson and Trawick, 1938; Brauchli et al., 1995; Berger, 2012). Indeed, Figure 5A,B shows that this relaxation effect was replicated using pupil size, which was smaller for predictive versus reactive saccade trials. NE has been shown to have a neuromodulatory effect on the parasympathetic oculomotor complex via LC inputs to SC (Edwards et al., 1979) and SC's corresponding projection through Edinger–Westphal nucleus (Harting et al., 1980) to the ciliary ganglion, which can be both excitatory and inhibitory (Barnerssoi et al., 2017), providing a potential pathway for a decreased pupil size and thereby arousal for predictive saccades. This pupil effect is complementary to a previous fMRI study by Lee et al. (2016) that demonstrated strong activation of the default mode network (DMN), a neural network that increases its metabolic activity during relaxation, when adults made predictive saccades in the metronome task. Pupil size has been shown to negatively correlate with BOLD activity of the posterior cingulate cortex and parahippocampal gyrus of the DMN and positively correlate with LC (Alnæs et al., 2014). Pupil size positively correlated with SRT (Fig. 5C), in accordance with a previous study that reported this coordination via shared signaling pathways in the SC (Wang and Munoz, 2021). Yet while differences in mean pupil size by saccade type were evident when all metronome trials were collapsed, this effect was not maintained when analyzed at the subject-level (Fig. 5D). This may be because of variability in the number of individual participants' pupil trials that successfully met criteria for analysis for each saccade type. To address this, the field may benefit from incorporating biometric sensors that are additional to pupil size recording (e.g., respiration rate; skin conductance; heart rate) to measure sympathetic tone while individuals predict or react to rhythmic stimuli.
Pupil size reliably decreased with age for all saccade types (Fig. 5D). Previous studies have described age-related pupil size declines starting in later adolescence (for a detailed description, see Loewenfeld, 1999, p 501); however, our results suggest that this phenomenon may be observed earlier, beginning in childhood (onset range 7–12 years) and declining in size until age 24, the oldest we tested. Presumably this reduction in pupil size with age is driven by a diminished constriction capacity of the dilator pupillae muscle (Loewenfeld, 1999).
Coordination of blinks and predictive saccades
Despite not being provided with instructions regarding when to blink, participants exhibited a clear pattern in the timing of their blinks relative to metronome target appearance, with median blink reaction times occurring at the halfway point of the ISI (Fig. 6C). Blink rate did not vary according to the rhythmicity of targets (Gagnon et al., 2002), but rather, blink timing varied according to frequency, and according to whether subjects predicted or reacted to targets (Fig. 6). It has been shown previously that blink reaction time variability decreases when participants view videos that have clear temporal event structures (Nakano et al., 2009). Plotting the blink probability leading up to target appearance allowed us to reasonably estimate the SRT category of the subsequent saccade (Fig. 6G), with increased blink probability for saccadic categories with slower RTs. Given that dopaminergic signaling within the basal ganglia has been shown to modulate blink excitability (Groman et al., 2014; Jongkees and Colzato, 2016) and adjust the precision of interval-based timing in mammals and humans (Kunimatsu et al., 2016; Soares et al., 2016; De Corte et al., 2019), the coordination we observed between blinks and temporal prediction may involve dopaminergic phasic activity. Kaminer et al. (2011) describe a neural mechanism whereby blinks are modified by striatal dopamine levels via substantia nigra pars reticulata inhibition of the SC, and the SC's excitation of the nucleus raphe magnus and the subsequent inhibition of the spinal trigeminal complex. However, more research is needed to elucidate the exact mechanisms underlying the coordination between saccade and blink reaction times toward periodic targets.
Footnotes
This work was supported by the Southeastern Ontario Academic Medical Organization AFP Innovation Fund Award SEA-17-004 (to O.G.C., L.B., S.K.-K., and D.P.M.) and the Canadian Institutes of Health Research Grant MOP-FDN-148418 (to D.P.M.). O.G.C. is supported by an Ontario Graduate Scholarship. D.P.M. is supported by the Canada Research Chair Program. We thank E. Robertson, H. Riek, and M. Kilmade for help with data collection. We also thank A. Lablans and M. Lewis for their outstanding technical assistance and Dr. B. Calancie for his helpful comments on an earlier version of the manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to Douglas P. Munoz at doug.munoz{at}queensu.ca or Olivia G. Calancie at olivia.calancie{at}queensu.ca
