All experimental procedures were reviewed and approved by the Queen’s University Health Sciences and Affiliated Teaching Hospitals Research Ethics Board. Healthy individuals between the ages of 5–93 were recruited from the greater Kingston area via newspaper and online advertisements. All participants reported no history of neurological or psychiatric illness and had normal or corrected-to-normal vision. A subset of participants aged 18 and older completed a Montreal Cognitive Assessment (MoCA), a brief screening tool shown to be sensitive in the detection of mild cognitive impairment (Nasreddine et al., 2005). Here, participants were excluded if they scored <20 on the MoCA. This cut-off score was determined based on the range of MoCA scores from the available subset of adult participants (aged 18–93) in our study cohort prior to outlier rejection (see Section “Pre-processing”). The use of a cut-off score lower than the recommended 26 (Nasreddine et al., 2005) is consistent with more recent studies suggesting that lower thresholds may decrease the false positive rate for mild cognitive impairment in large, diverse cohorts including older adults and individuals with lower education levels (Rossetti et al., 2011; Carson et al., 2018). Written informed consent was obtained from all individuals aged 18 and older. Written informed assent, in addition to a parent or guardian’s written informed consent, was obtained from all individuals under the age of 18. Study sessions took approximately 1 h each. Participants were compensated $20 CAD for their time.

During the eye tracking portion of the study, participants were seated in a dark room with their heads resting comfortably in a head rest. Participants were seated 60 cm away from a 17 inch 1280×1024 pixel resolution LCD computer monitor. An infrared video-based eye tracker (Eyelink 1000 Plus, SR Research Ltd., ON, Canada) was used to track monocular eye position at a sampling rate of 500 Hz. A 9-point array calibration and validation procedure was performed for each participant prior to beginning the task to map raw pupil position into gaze position. Eyelink 1000 measures validation accuracy as the average error in degrees between gaze and validation target positions. Here, participants had to have an average validation accuracy < 1.5° in order for their eye tracking data to be considered sufficiently accurate for further analysis.

The IPAST (Figure 1A) consisted of two blocks of 120 trials each, lasting approximately 20 min in total. Each trial began with the appearance of a colored fixation point (FP; 0.5° in diameter, 44 cd/m²) in the center of a black screen (0.1 cd/m²) for 1000 ms. The color of the FP indicated the trial condition (green = PRO, red = ANTI). Following a 200 ms gap during which the FP was removed (GAP), a gray stimulus (STIM; 0.5° in diameter, 62 cd/m²) appeared 10° to the left or right of the FP position and remained on screen for an additional 1000 ms. On PRO trials, participants were instructed to look at the STIM as soon as it appeared. On ANTI trials, participants were instructed to look away from the STIM (i.e., to its diametrically opposite position) as soon as it appeared. An inter-trial interval (ITI) consisting of a black screen (0.1 cd/m²) was presented for 1000 ms before the start of each new trial. Drift checks occurred every 40 trials to confirm the accuracy of eye tracking or to allow for re-calibration, if necessary. Trial condition (PRO/ANTI) and STIM location (left/right) were pseudo-randomly interleaved with equal frequency throughout each block. Verbal task instructions and 10–20 practice trials were provided to each participant prior to beginning the task in order to ensure comprehension.

631 individuals (409 F, 222 M, 5–93 years of age) were recruited to participate in this study from 2015 to 2022. Of the 430 individuals aged 18 and older, 346 (80%) completed a MoCA. We note that our original study cohort is skewed toward young (i.e., >25 years old) female participants due to extensive participant recruitment from local university and college student bodies, as well as our lab’s efforts to match control participants to various neuropsychiatric patient cohorts within this demographic. From this original study cohort, 26 participants were excluded as a result of our three-step outlier rejection procedure. Nine participants were excluded on the basis of a PRO or ANTI viable trial count < 30, 14 participants were excluded on the basis of a PRO or ANTI eye loss or non-compliance trial count > 20%, and three participants were excluded on the basis of a MoCA score < 20. One additional participant was excluded on the basis of not making a single ANTI viable correct response. The final study cohort therefore consisted of 604 participants (393 F, 211 M, 5–93 years of age). Of the 419 individuals aged 18 and older, MoCA scores were available for 335 (80%). The majority of individuals for which MoCAs were not available were between the ages of 18–25, a demographic for which scores in large population cohorts have been found to be well above the cut-off used here (Rossetti et al., 2011). Age and MoCA score distributions for male and female participants included in the final study cohort are shown in Figures 2A,B, respectively. Additional demographic information (i.e., education level, average MoCA score, where applicable) for all included and excluded participants is provided in Supplementary Table 1.

Table 1 displays the GAM fit parameters (ref df, F, p, R², deviance explained) for each of the seven IPAST measures of interest (see Section “Materials and Methods”). All measures exhibited significant (p < 0.05) age-related changes. The amount of deviance of the IPAST measures explained by age ranged from 7.18 to 31.2%. In order to investigate if GAM fits differed as a function of participant sex, a second GAM (Model 2) was defined by a smoothed fixed effect of age split by sex and also performed for each measure of interest. Bayesian information criterion (BIC) was used to compare the goodness-of-fit of Model 2 with the originally specified GAM (Model 1; smoothed fixed effect of age only), with lower values indicating a superior fit. Supplementary Table 2 displays the fit parameters for both models. For Model 2, each smooth fit of age remained significant when split by participant sex, and a similar amount of deviance of the measures was explained by age (7.31–31.5%). However, BIC values were lower (indicating a superior fit) for Model 1 for all measures investigated. We therefore describe the characteristics of these GAMs for the remainder of our results.

Significant periods of change for the fitted GAMs were identified at the ages where the confidence intervals of the first derivative did not contain zero (p < 0.05). At least one significant period of change was identified for each measure. With the exception of PRO express- and regular-latency correct response ratios, which exhibited significant change across the entire lifespan, all periods of change beginning before the age of 23 captured improvements in task performance (i.e., decreases in SRT/VOT or direction errors), while all periods of change beginning after the age of 23 captured declines in task performance (i.e., increases in SRT/VOT or direction errors). We therefore refer to periods of change as being either “developmental-related” or “aging-related,” depending on whether the period began before or after the age of 23.

Standardized residuals derived from each of the GAMs described above were input into Spearman’s correlations to investigate the pairwise relationships between measures after controlling for age (Figure 7). PRO and ANTI viable correct SRT were positively correlated with one another. PRO and ANTI SRT were also both positively correlated with PRO regular-latency correct response ratio and ANTI regular-latency direction error ratio, and negatively correlated with PRO express-latency correct response ratio and ANTI express-latency direction error ratio. PRO express-latency correct response ratio was negatively correlated with PRO regular-latency correct response ratio, and positively correlated with ANTI express-latency direction error ratio. Finally, VOT was positively correlated with both ANTI viable correct SRT and ANTI regular-latency direction error ratio, as expected. All reported correlations were statistically significant (p < 0.05).

GAM fits of PRO and ANTI viable correct SRT demonstrated that PRO SRT decreased from the ages of 5–17 and increased from the ages of 23–33 and 52–58, while ANTI SRT decreased from the ages of 5–18 and increased from the ages of 25–29, 50–57, and 74–86 (Figure 3). The initial dramatic decrease in PRO and ANTI SRT captured here is consistent with past group- and regression-based studies, reporting that saccade latencies decrease significantly from the ages of 5–25 (Munoz et al., 1998; Fukushima et al., 2000; Klein and Foerster, 2001; Luna et al., 2004; Kramer et al., 2005; Velanova et al., 2008; Ordaz et al., 2010; Bucci and Seassau, 2012; Alahyane et al., 2014). It has previously been proposed that PRO and ANTI SRT reach adult-levels by the ages of 12–14 (Fukushima et al., 2000; Klein and Foerster, 2001; Irving et al., 2006; Bucci and Seassau, 2012). Using a large cohort of 8–30 year olds, blocked PRO/ANTI paradigm, and change point analysis based on fitted piecewise linear models, Luna et al. (2004) identified 15 as the age at which saccade latencies reached adult-levels. Our findings of saccade latencies maturing later in adolescence (i.e., 17 and 18) likely reflects our use of: (1) a larger study cohort spanning the entire lifespan, (2) a more cognitively demanding eye tracking paradigm (i.e., interleaved vs. blocked design), and (3) more flexible analytical approaches, relative to past work.

Our findings of gradual increases in PRO and ANTI SRT during aging is consistent with numerous group-based studies describing significant differences in SRT between adults in the third and fourth decades of life relative to those in the seventh, eighth, and ninth (Munoz et al., 1998; Klein et al., 2000; Sweeney et al., 2001; Abel and Douglas, 2007; Fujiwara et al., 2010; Bonnet et al., 2013; Noiret et al., 2016). Two recent, well-powered regression-based studies further determined that PRO and ANTI SRT increased significantly from the ages of 30–95 (Coors et al., 2021) and 51–84 (Mack et al., 2020). Our change point analysis revealed that PRO and ANTI SRT begin to increase as early as the ages of 23 and 25, respectively. While there is a paucity of studies examining PRO and ANTI behavior in adults from the third through fifth decades of life, studies investigating the effect of age on other behavioral paradigms of inhibitory control and processing speed have similarly suggested that decline may begin as early as the 20–30s (Salthouse, 2009; Ferguson et al., 2021).

As previously introduced, a considerable advantage to using the IPAST to investigate inhibitory control across the lifespan is that behaviors can be linked to well-characterized frontal-parietal and frontal-striatal mechanisms. Monkey neurophysiology, human lesion, and human neuroimaging work have jointly indicated that the frontal, supplementary, and parietal eye fields (FEF, SEF, PEF) are critical in the planning, sensorimotor mapping, and execution of saccades (Connolly et al., 2002; Pierrot-Deseilligny et al., 2002; Amador et al., 2004; Munoz and Everling, 2004). Notably, preparatory FEF activity prior to STIM appearance on PRO and ANTI trials predicts subsequent saccade latencies in both monkeys (Everling and Munoz, 2000) and humans (Connolly et al., 2005; Alahyane et al., 2014; Fernandez-Ruiz et al., 2018). Previous studies employing event-related fMRI during IPAST performance have shown that children have significantly lower preparatory FEF, SEF, and PEF activity relative to adolescents and adults (Alahyane et al., 2014), while older adults have significantly lower SEF activity relative to younger adults (Fernandez-Ruiz et al., 2018). Our findings of PRO and ANTI SRT maturation at the ages of 17 and 18, as well as their subsequent decline beginning in the mid-20s may therefore be attributed to changes in the structural and functional integrity of the FEF, SEF, and PEF.

PRO express- and regular-latency correct responses exhibited opposing linear trends that were significant across the entire lifespan (Figure 4). Using a blocked PRO/ANTI paradigm, we have previously reported higher express-latency saccades in individuals younger than 40 (Munoz et al., 1998), and others have found express-latency saccade rates to be significantly higher in 6–7 and 10–11 year olds relative to 18–26 year olds (Klein and Foerster, 2001), and in 20–35 year olds relative to 59–73 and 74–88 year olds (Klein et al., 2000). Most comparable to the present results, Klein et al. (2005) also found a weak negative linear relationship between age and express-latency saccade rate in a cohort of individuals aged 9–88.

PRO express- and regular-latency correct responses are both triggered by the STIM visual transient. Express-latency responses result from a direct sensory-motor transformation of the transient (Edelman and Keller, 1996; Dorris et al., 1997; Sparks et al., 2000) that occurs in the SC (Schiller et al., 1987), while regular-latency responses result from the propagation of a well-learned, automated motor command that may involve cortical areas such as the PEF (Munoz and Everling, 2004; Coe and Munoz, 2017). Our findings suggest that with age, cortically mediated automated processes may increasingly dominate subcortically mediated reflexive processes, resulting in a continuous decrease in the proportion of express- to regular-latency responses.

ANTI express- and regular-latency direction errors exhibited distinctly non-linear trends across the lifespan; both error types decreased until the mid-20s, but only regular-latency errors subsequently increased beginning at the age of 63 (Figure 5). It is well-established that children make more direction errors than adolescents, who make more direction errors than young adults (Fukushima et al., 2000; Klein and Foerster, 2001; Luna et al., 2004; Kramer et al., 2005; Velanova et al., 2008; Ordaz et al., 2010; Bucci and Seassau, 2012; Alahyane et al., 2014), and that older adults make more direction errors than younger adults (Klein et al., 2000; Sweeney et al., 2001; Abel and Douglas, 2007; Fujiwara et al., 2010; Peltsch et al., 2011; Bonnet et al., 2013; Noiret et al., 2016; Fernandez-Ruiz et al., 2018; Mack et al., 2020; Coors et al., 2021). The latency of these direction errors, however, is rarely characterized. In one of the few studies to do so, Klein and Foerster (2001) found that the difference in the proportion of express- to regular-latency direction errors was significantly greater in 6–7 year olds relative to 10–11 and 18–26 year olds. While not explicitly examining express- and regular-latencies, other studies have described longer direction error latencies in older relative to younger adults (Bowling et al., 2012; Noiret et al., 2016).

On ANTI trials, pre-emptive, global inhibition provided by regions such as the DLPFC, FEF, SEF, BG, and SC is required prior to STIM appearance to suppress the express-latency direction error (Coe and Munoz, 2017; Coe et al., 2019). Subsequently, coordinated activity between these regions is required to drive the voluntary, location-specific motor command for an ANTI to overcome the regular-latency direction error. Support for these claims stems from monkey (Everling et al., 1999; Everling and Munoz, 2000; Amador et al., 2004; Johnston and Everling, 2006; Johnston et al., 2007; Watanabe and Munoz, 2010) and human (Connolly et al., 2002; Curtis and D’Esposito, 2003; DeSouza et al., 2003; Ford et al., 2005; Brown et al., 2007) work demonstrating differential preparatory activity in these regions on ANTI vs. PRO trials. Preparatory activity in the DLPFC and anterior cingulate cortex (ACC) has been specifically associated with the monitoring and suppression of direction errors on ANTI trials (Pierrot-Deseilligny et al., 2002, 2003; Ford et al., 2005; Johnston and Everling, 2006; Johnston et al., 2007).

Neuroimaging studies of PRO and ANTI behavior in development suggest that functional activity and effective connectivity of frontal-parietal and frontal-striatal regions become more widely distributed from childhood through to adulthood, supporting a reduction in direction error rates (Luna et al., 2001; Velanova et al., 2008; Hwang et al., 2010; Ordaz et al., 2013; Alahyane et al., 2014). Although neuroimaging studies of PRO and ANTI behavior in aging cohorts have reported mixed findings to-date (Raemaekers et al., 2006; Nelles et al., 2009; Alichniewicz et al., 2013), Fernandez-Ruiz et al. (2018) recently found that older adults had significantly lower preparatory SEF and ACC activity relative to younger adults, potentially contributing to a reduced ability to monitor task performance and drive the voluntary motor command for a correct ANTI. Considering the present work, the ability to suppress both express- and regular-latency direction errors by the mid-20s may be mediated by mature activation and integration of frontal-parietal and frontal-striatal regions, including the DLPFC, FEF, SEF, BG, and SC. By the 60s, however, the ability of these regions to drive the voluntary motor command for a correct ANTI may begin to deteriorate, resulting in an increase in regular-latency direction errors.

We characterized the time at which voluntary processes begin to overcome automatic processes on ANTI trials, or the VOT (Coe et al., 2019), for the first time across the lifespan. VOT decreased dramatically from the ages of 5–19, remained relatively stable through the third and fourth decades of life, then exhibited gradual increases from the ages of 40–52 and 76–85 (Figure 6A). Recent work from our group describes a generative model of saccadic action selection, inspired by known signal components of neural activity, capable of producing PRO and ANTI behaviors similar to those observed here (Coe et al., 2019). Post-hoc investigations into the VOT generated by this model revealed that simulating a type of inhibitory “crosstalk” between voluntary (based on activity in the FEF, SEF, and SC) and automated (based on activity in the PEF) signals, such that voluntary signals could override automated signals, produced behavior that better approximated human data relative to prior simulations (Coe et al., 2019). The present findings indicate that the ability of voluntary signals to override automated signals matures at the age of 19 (i.e., after maturation of saccade latency but before maturation of direction error suppression), and begins to decline at the age of 40 (i.e., after decline of saccade latency but before decline of direction error suppression). Taken together, the animal, human, and computational literature described here provide insight into potential frontal-parietal and frontal-striatal mechanisms underlying the behavioral changes revealed in our analysis. We suggest that structural and functional maturation of these circuits mediates decreased saccade latencies and direction error rates throughout adolescence, while their subsequent decline mediates increased saccade latencies in the mid-20s and increased regular-latency errors in the 60s. Increasing input from cortical, relative to subcortical regions of the brain mediates a continuous decrease in the proportion of PRO express- to regular-latency correct responses across age.

The generalizability of our study cohort is somewhat limited by the overrepresentation of young female participants. Recent work has highlighted the importance of having sufficient numbers of middle-aged adults in lifespan cohorts in order to more comprehensively characterize cognitive changes that occur with age (Ferguson et al., 2021). As there is a paucity of studies investigating PRO and ANTI behavior in adults from the third through fifth decades of life, our findings of PRO and ANTI SRT beginning to decline as early as the mid-20s requires replication. The uneven distribution of female and male participants in our study cohort prevented us from conducting an in-depth analysis of sex differences in IPAST behavior. However, BIC values indicated that GAMs specified with a smoothed fixed effect of age provided a superior fit for all measures relative to GAMs specified with a smoothed fixed effect of age split by sex. Previous studies examining sex differences in PRO and ANTI tasks have been mixed; some failing to identify any differences (Sweeney et al., 2001; Fujiwara et al., 2010; Bonnet et al., 2013), others reporting shorter latencies in adolescent females relative to males (Luna et al., 2004), and others reporting longer latencies in adult females relative to males (Bargary et al., 2017; Mack et al., 2020). Although sex differences in PRO and ANTI behavior remains to be clarified, it is clear that any sex effects that do exist are substantially weaker relative to those of age (Ordaz et al., 2018; Coors et al., 2021).

A second limitation of the present work is the reliance on a cross-sectional study design rather than a longitudinal one, as the latter is more sensitive to age-related changes and inter-individual variability (Casey et al., 2005; Goh et al., 2012; Sørensen et al., 2021). Indeed, the GAM and change point analyses described here are highly amenable to longitudinal study designs (Simmonds et al., 2014; Wierenga et al., 2019; Calabro et al., 2020; Danielsen et al., 2020). Regarding inter-individual variability, there are likely many biological, environmental, and psychosocial factors contributing to variability in IPAST behavior that have yet to be fully characterized (Bargary et al., 2017). As the frequency and timing of express-latency saccades can be influenced by various task parameters (Dorris and Munoz, 1998; Bibi and Edelman, 2009; Marino and Munoz, 2009), it is likely that these behaviors also vary on an individual participant basis. Investigation into the factors mediating inter-individual variability of the frequency and timing of express-latency saccades is therefore a promising direction of future work.

The GAM fits and significant periods of change presented here have significant implications for improving our understanding of the vulnerable windows of brain maturation and decline in which the risk for developing neurological disorders may be elevated. Attention-deficit hyperactivity disorder and Parkinson’s disease are two neurological disorders with onset at either ends of the lifespan that can be differentiated from healthy age-matched controls based on ANTI SRT and ANTI express- and regular-latency direction errors (Cameron et al., 2012; Hakvoort Schwerdtfeger et al., 2013; Coe and Munoz, 2017). As VOT indexes both of these ANTI behaviors and is explained in large part (31%) by age, this measure presents as an ideal “cognitive growth curve” for differentiating normative aging and neurological disease across the lifespan (Figure 6C). We propose that this curve could be used to identify individuals in early childhood and late adulthood who may be at an elevated risk for developing these disorders, by virtue of their VOT values falling outside of a specified percentile for normative aging.