Saccade direction modulates the temporal dynamics of presaccadic attention

Abstract

Presaccadic attention enhances visual perception at the upcoming saccade target location. While this enhancement is often described as obligatory and temporally stereotyped, recent studies indicate that its strength varies depending on saccade direction. Here, we investigated whether the time course of presaccadic attention also differs across saccade directions. Participants performed a two-alternative forced-choice orientation discrimination task during saccade preparation. Tilt angles were individually titrated in a fixation baseline condition to equate task difficulty across the upper and lower vertical meridians. Sensitivity was then assessed at different time points relative to saccade onset and cue onset, allowing us to characterize the temporal dynamics of attentional enhancement. We found that presaccadic attention built up faster and reached higher levels preceding downward than upward saccades. Linear model fits revealed significant slope differences but no differences in intercepts, suggesting that the observed asymmetries reflect differences in attentional deployment during saccade preparation rather than preexisting differences in sensitivity. Saccade parameters did not account for these asymmetries. Our findings demonstrate that the temporal dynamics of presaccadic attention vary with saccade direction, which may be a potential mechanism underlying previously observed differences in presaccadic benefit at the upper and lower vertical meridians. This temporal flexibility challenges the view of a uniform presaccadic attention mechanism and suggests that presaccadic attentional deployment is shaped by movement goals. Our results provide new insights into how the visual and oculomotor systems coordinate under direction-specific demands.

Introduction

The close relation between visual attention and saccadic eye movements has long been a cornerstone of oculomotor and perceptual research. Foundational studies by Eileen Kowler and colleagues demonstrated that attention shifts reliably to the future saccade target before movement onset, enhancing perceptual processing at that location (Kowler, Anderson, Dosher, & Blaser, 1995; Zhao, Gersch, Schnitzer, Dosher, & Kowler, 2012). This presaccadic shift of attention is thought to support both accurate eye movement execution and the integration of visual information across saccades (Deubel & Schneider, 1996; Hoffman & Subramaniam, 1995; Rolfs, Jonikaitis, Deubel, & Cavanagh, 2011).

While early studies emphasized the spatial specificity of the presaccadic attention shift, more recent work has revealed that its spatial profile is not fixed. For example, the extent and location of the attentional benefit depend on the visual structure of the saccade target: Saccades directed toward blank or sparsely textured locations elicit broader, less focal enhancements (Hanning & Deubel, 2022; Szinte, Puntiroli, & Deubel, 2019). Moreover, presaccadic attention has been typically described as mandatory—automatically enhancing perception at the location of the saccade target at the cost of sensitivity elsewhere, including the presaccadic center of gaze (Hanning & Deubel, 2022), even if detrimental to current task goals (Hanning, Wollenberg, Jonikaitis, & Deubel, 2022; Li, Pan, & Carrasco, 2019). However, our group's recent findings indicate that this enhancement does not hold equally for all saccade directions. Specifically, we were the first to show that presaccadic attention shifts are markedly weaker before saccades directed upward than for other directions (Hanning, Himmelberg, & Carrasco, 2022; Hanning, Himmelberg, & Carrasco, 2024), challenging the prevalent view that these attentional benefits are universal and uniform.

Consistent with this notion that not all attentional shifts operate in an identical manner, such direction-dependent asymmetries in the magnitude of presaccadic enhancement may also be task-dependent or feature-specific. Whereas contrast sensitivity improves more before horizontal and/or downward saccades than before upward saccades (Hanning et al., 2024; Hanning, Himmelberg, et al., 2022; Kwak, Zhao, Lu, Hanning, & Carrasco, 2024), acuity or spatial frequency sensitivity appears largely unaffected by saccade direction (Kwak, Hanning, & Carrasco, 2023; Kwak et al., 2024). Therefore, not only does the shift of presaccadic attention differ as a function of saccade direction, but it also interacts with the specific visual dimension it modulates.

Temporal aspects of presaccadic attention have also received growing attention in recent years. Studies characterizing the time course of attention deployment typically report a gradual increase in perceptual sensitivity at the target location beginning approximately 150 to 100 ms before saccade onset, peaking just prior to execution (Castet, Jeanjean, Montagnini, Laugier, & Masson, 2006; Deubel, 2008; Hanning, Deubel, & Szinte, 2019; Kroell & Rolfs, 2021; Li, Barbot, & Carrasco, 2016; Montagnini & Castet, 2007; Ohl, Kuper, & Rolfs, 2017; Rolfs & Carrasco, 2012). Importantly, this temporal profile has been measured almost exclusively for horizontal saccades. Even when it is measured at different locations (Hanning et al., 2019; White, Rolfs, & Carrasco, 2013), it is merely assumed to generalize across saccade directions.

While the aforementioned studies have begun to characterize directional differences in magnitude (i.e., overall modulations of visual sensitivity within ∼100 ms before saccade execution), the temporal dynamics of presaccadic attention across directions remain largely unexplored. This represents a critical gap: The reduced behavioral benefits at the upper vertical meridian compared to other locations could be explained by (a) delayed onset and/or saturation of attentional buildup, (b) a slower or shallower rise of attentional buildup, or (c) their interaction. Yet, no prior study has directly compared how presaccadic attention unfolds over time for different saccade directions.

In the present study, we address this gap by measuring presaccadic orientation discrimination performance at upward, downward, and horizontal saccade targets throughout saccade preparation. As visual sensitivity during fixation varies profoundly as a function of polar angle—higher for horizontal than vertical meridian, as well as for the lower than upper vertical meridian, at matched eccentricity (e.g., Abrams, Nizam, & Carrasco, 2012; Barbot, Xue, & Carrasco, 2021; Lee & Carrasco, 2025b; review: Himmelberg, Winawer, & Carrasco, 2023)—we calibrated tilt angles at these locations for each participant to equate baseline perceptual difficulty around the visual field. This adjustment allowed us to isolate the dynamics of presaccadic attention from baseline differences in sensitivity, ensuring a fair comparison across these peripheral (saccade target) test locations.

Note that in the horizontal condition, sensitivity at the very beginning of or even before saccade preparation was elevated compared to the vertical condition, which precludes a meaningful comparison of the temporal dynamics of presaccadic attention directed toward the horizontal and vertical meridians (see Discussion). We therefore focus our analyses and discussion on comparing presaccadic perceptual modulations along the vertical meridian (i.e., upward and downward saccades) and report results from horizontal saccades in the Supplementary Materials.

Our results revealed a steeper buildup of attention preceding downward compared to upward saccades, despite matched baseline sensitivity and saccade parameters. These findings show that the temporal profile of presaccadic attention varies systematically across directions, which may account for previously observed perceptual and physiological asymmetries: The reduced presaccadic benefit at the upper vertical meridian in previous studies (Hanning et al., 2024; Hanning, Himmelberg, et al., 2022; Kwak et al., 2024) can be explained by the speed/rate of the perceptual enhancement. Although presaccadic attentional processes are tightly linked to the oculomotor system, the current study indicates that the direction of the upcoming saccade modulates the dynamics of attention and provides further support of the notion that this relation is flexible.

Methods

Participants

Ten observers (6 females, including author YK; aged 22–32 years) with normal or corrected-to-normal vision participated in the experiment. All participants, except for the author, were naive to the experimental hypothesis, and the experimental procedures were approved by the Institutional Review Board at New York University. All participants provided informed consent and were paid $12 per hour. All procedures were in agreement with the Declaration of Helsinki.

Setup

The experiment was conducted in a dark room with participants’ heads stabilized by a chin and forehead rest. All stimuli were generated and presented using MATLAB (MathWorks, Natick, MA, USA) and the Psychophysics Toolbox (Brainard, 1997; Pelli, 1997). We used a gamma linearized 20-inch ViewSonic G220fb CRT monitor at a viewing distance of 57 cm. The CRT screen had a resolution of 1,280 by 960 pixels and a refresh rate of 100 Hz. Gaze position was recorded and monitored online using an EyeLink 1000 Desktop Mount eye tracker (SR Research, Osgoode, Ontario, Canada) at a sampling rate of 1 kHz and the Eyelink Toolbox (Cornelissen, Peters, & Palmer, 2002).

Experimental procedure

Experimental task and structure

Participants viewed dynamic streams of vertically oriented Gabor and noise patches alternating every 30 ms (Hanning et al., 2019; Rolfs et al., 2011), presented at the four polar angle locations (Figure 1). The task was to judge the orientation of a test Gabor that appeared at one stream for 30 ms, which was rotated counterclockwise or clockwise in reference to the vertical axis. The tilt angle was determined by a thresholding session conducted before the start of the main experiment (see “Titration procedures”). Gabor patches in all other streams remained vertical (distractor Gabors).

Figure 1.

Experimental design. (A) Example trial sequence. After stable fixation has been verified, dynamic streams consisting of vertically oriented distractor Gabors alternating with noise patches were presented throughout the trial. Embedded in one of the dynamic streams was a test Gabor tilted counterclockwise or clockwise to vertical, and participants judged its orientation by pressing the left/right key after the response tone. In the saccade condition illustrated here, a central line cue pointed toward one of the two relevant locations. The tilted test Gabor always appeared at the cued location, between 0 and 220 ms after cue onset (i.e., during saccade preparation). In the fixation condition, no cue was presented, and participants maintained fixation throughout the trial. (B) Stimulus timing relative to cue onset. Solid gray lines indicate that the respective event was present. Dashed gray lines indicate jittered delays (the corresponding event could begin anytime between the start and end of the dashed line). In red is the hypothetical distribution of saccade onset (i.e., latency). We only included trials in which the test offset was before the saccade onset. (C) Group average visual sensitivity as a function of location (upper/lower) and eye movement condition (fixation/saccade). Error bars indicate ± 1 SEM.

The experiment consisted of 10 blocks of testing along the vertical meridian (vertical condition) and 6 blocks of testing along the horizontal meridian (horizontal condition). Because it is well established that sensitivity in the left and right horizontal meridians does not differ (e.g., Cameron, Tai, & Carrasco, 2002; Carrasco, Talgar, & Cameron, 2001), measurements for horizontal saccades (at the left and right locations) were combined for analyses, which is why we collected less data for the horizontal meridian. We focus on the vertical meridian in the main text, and the results at the horizontal meridian are reported in the Supplementary Materials. Each block consisted of 220 trials (the number of trials collected differed across participants and blocks due to repetition of trials that did not meet the eye-tracking criteria; see “Eye movement analysis”). The starting condition (vertical or horizontal) was counterbalanced across participants, and all blocks of one condition were completed before switching to the other.

In both the vertical and horizontal conditions, stimuli appeared at all four polar angle locations, but participants were explicitly instructed at the start of each block that only two locations were task-relevant in a certain block (e.g., only the upper and lower locations in the vertical condition). To reinforce this, bold placeholders were displayed around the relevant locations.

Within each block, we randomly interleaved saccade and fixation trials. In a saccade trial, participants made a saccade to one of the two relevant locations (i.e., upper or lower for the vertical blocks), indicated by a central cue. The tilted test Gabor always appeared at the cued, saccade target location, and participants judged its orientation in a nonspeeded manner after performing a saccade. In a fixation trial, we ensured via eye tracking that participants maintained fixation throughout the trial, and no information was given on which of the two locations the test Gabor would appear. Therefore, participants had to first identify the location of the test Gabor and report its orientation. The ratio of saccade to fixation trials was 4.5:1.

Trial sequence

The trial sequence and stimuli timing are illustrated in Figure 1. Each trial started with a black fixation circle (0.175° radius) on a gray background (∼26 cd/m²). Four placeholders, each composed of four corners (black lines, 0.2° length), indicated the locations of the upcoming stimuli, 9° left, right, above, and below fixation. The two locations relevant for the current block were marked with bold placeholders (e.g., upper and lower locations for the vertical condition).

After detecting a minimum of 300 ms stable fixation (eye coordinates within a 1.75° radius virtual circle centered on fixation), the dynamic stimuli streams—vertically oriented Gabors patches (2.5 cpd, 3.6° diameter, 100% contrast) alternating with noise patches (bandpass-filtered noise, 3.6° diameter, 50% contrast) at 33.3 Hz (30 ms for each stream)—appeared and remained on the screen throughout the trial. After fixating for 300 to 400 ms (Figure 1B), a central line cue indicated one of the two relevant locations (saccade trial). This line was 100% informative, as the target always appeared at the location it indicated. Participants made a saccade to the center of the cued location as fast and as precisely as possible. A test Gabor, tilted counterclockwise or clockwise relative to vertical, appeared at the saccade target location during eye movement preparation. The timing between cue onset and the test Gabor onset varied on each trial –0 to 220 ms in steps of 20 ms (Figure 1B, Test). Together with the variability in saccade latencies, this enabled a continuous measurement of presaccadic attention at the saccade target during saccade preparation. After the presentation of the test Gabor, the streams continued with alternating noise patches and blanks, instead of vertically oriented distractor Gabors, to avoid apparent motion effects (Hanning et al., 2019; Rolfs et al., 2011).

After all the streams had been presented, a tone informed participants to indicate their response, and they pressed the left/right arrow key to indicate that the test Gabor was tilted counterclockwise/clockwise, respectively. No feedback was provided on the accuracy of their response. The trial sequence was identical for fixation trials, except that no line cues were presented.

Gaze position was monitored online to ensure fixation within a 1.75° radius virtual circle from the central fixation until the response phase (fixation trials) or until cue onset (saccade trials). Trials in which gaze deviated from fixation beforehand were aborted and repeated at the end of each block. In addition, for the saccade condition, we repeated trials in which saccade latency was too short (<150 ms) or too long (>350 ms), or in which saccades landed outside of a 2.25° radius virtual circle around the saccade target center. Moreover, because we are interested in the presaccadic interval, we only included trials in which saccades were initiated after the test Gabor offset.

Titration procedures

A titration session was conducted before starting the main blocks of the horizontal and vertical conditions to account for sensitivity differences around polar angle locations, which are known to be present without any eye movements (e.g., Abrams et al., 2012; Barbot et al., 2021). Our goal was to equate baseline performance (during fixation), to enable a fair comparison of the attentional buildup at the different test locations. The trial sequence was the same as in the main fixation condition, with the exception that the onset of the test Gabor was always 140 ms after the onset of the dynamic stream.

We titrated the tilt angle of the test Gabor stimulus separately for each location with best PEST, a maximum likelihood adaptive procedure implemented in the Palamedes toolbox (Prins & Kingdom, 2018). Each location was blocked, and we targeted the tilt angle from the vertical orientation, resulting in 60% orientation discrimination accuracy. For the vertical condition, there were four independent staircase procedures for each location. For the horizontal condition, there were three staircase procedures for each location, taking into account the lack of systematic differences in sensitivity between the left and right locations (e.g., Cameron et al., 2002; Carrasco et al., 2001) and that the measurements at the two locations would be combined for further analysis. Each staircase procedure consisted of 45 trials. To derive the final tilt thresholds, we averaged the estimated thresholds (the last value each staircase converged on) for all staircases in each location.

After the titration sessions, we performed an initial check of the thresholded values. The trial sequence was identical to that of the fixation trial in the main experiment (see “Trial sequence”). There were 35 trials per each location, randomly intermixed, resulting in a total of 70 trials per condition (vertical/horizontal). If average performance at a location deviated by more than 5% from the target of 60% correct discrimination performance, we either increased (performance < 55%) or decreased (performance > 65%) the tilt angle (depending on how much the performance deviated from the target performance, matching the log-scale increments/decrements used in the staircase procedure). Moreover, as the pilot study confirmed that participants’ performance improved over the course of the experiment, we adjusted the tilt angle after the first half of the experiment (after 5 out of 10 blocks for the vertical condition and after 3 out of 6 blocks for the horizontal condition), also according to a predetermined criterion based on participants’ previous performance in the fixation trials.

The average tilt angle used in the main experiment was 4.048° ± 0.883° at the upper vertical, 3.248° ± 0.447° at the lower vertical, and 6.051° ± 1.088° at the horizontal meridian (mean ± 1 SEM). These threshold results are consistent with the well-established vertical meridian asymmetry (better performance at the lower than the upper vertical meridian) and with previous pilot data from our lab showing that orientation tilt thresholds relative to vertical are lower along the vertical meridian than the horizontal meridian. This pattern aligns with the radial bias reported in the literature: Sensitivity to gratings is higher for radial than tangential orientations (Ezzo, Winawer, Carrasco, & Rokers, 2023; Lee & Carrasco, 2025b; Ryu & Lee, 2024; Sasaki et al., 2006). Consequently, the horizontal–vertical asymmetry (i.e., the perceptual advantage at horizontal locations) is less pronounced for perceptual judgments with reference to vertical than horizontal orientations (Ezzo et al., 2023; Lee & Carrasco, 2025b; Ryu & Lee, 2024; Sasaki et al., 2006).

Eye movement analysis

In addition to the gaze monitoring and detection of saccades online (see “Trial sequence”), we also performed an offline eye movement analysis using an established algorithm for saccade detection (Engbert & Mergenthaler, 2006). Saccades were detected based on their velocity distribution using a moving average over 20 subsequent eye position samples. Saccade onsets and offsets were detected based on when the velocity exceeded or fell below the median of the moving average by 3 standard deviations for at least 20 ms. This offline analysis takes into account the entire distribution of eye position samples and velocity, allowing for a more detailed evaluation of both fixation and saccadic eye movements.

Behavioral data analysis

To derive sensitivity measures as a function of test offset relative to saccade and cue onset, we performed a sliding window analysis (Hanning et al., 2019) and grouped the trials into 60-ms bins. For our analyses relative to saccade onset, the bin centers were 180, 150, 120, 90, 60, 30, and 0 ms before saccade onset; for analyses relative to cue onset, the bin centers were 0, 30, 60, 90, 120, and 150 ms after cue onset (all bins are equidistant and have the same width, except for the last bin before saccade onset and first bin after cue onset, which are half the width; see below). For each time point, visual sensitivity (d′ = z(hit rate) − z(false alarm rate)) was computed based on trials that belonged to the 60-ms bin centered on that time point (e.g., for −150 ms, trials that fell between −180 and −120 ms). However, note that for 0 ms before saccade onset (last time bin), the bin size was 30 ms because we excluded trials in which saccade onset was after test offset as we were interested in the presaccadic interval. Similarly, the bin size was 30 ms for 0 ms after cue onset (first time bin), because the earliest that the test Gabor could appear was 0 ms after cue onset. Note that the results reported here did not depend on the particular step size and width of the temporal bins chosen (see Supplementary Materials). The number of trials within each time bin is reported in Supplementary Table S1.

Model fitting

We fit two models—linear and nonlinear—on d′ as a function of time (t). The linear regression model had two parameters—slope (s) and intercept (b):

$$\begin{array}{c}\hfill d^{\text{'}}\left(t\right)=s\times t+b.\end{array}$$

The nonlinear model was a sigmoidal function with four parameters—amplitude (a), time of inflection (t₀), steepness (τ), and baseline (b):

$$\begin{array}{c}\hfill d^{\text{'}}\left(t\right)=\frac{a}{1+e^{-\left(t-t_0\right)/\tau }}+b.\end{array}$$

We performed a weighted model fit in which each datapoint was weighted by the proportion of trials in its bin (i.e., $\frac{numberoftrialswithinthebin}{totalnumberoftrials}$), and the weighted sum of squares was minimized. In addition, the model fitting was simplified by setting t to the number of bins (1 to 7 for locked to saccade onset and 1 to 6 for locked to cue onset). Note that all bins were equidistant (i.e., 30 ms) and had the same width (i.e., 60 ms), at the upper and lower vertical locations, and for both analyses locked to saccade and cue onset.

For comparing between the linear and nonlinear models, we used the Akaike information criterion (AIC), which evaluates the relative quality of different statistical models for a given data set, penalizing for model complexity (i.e., the number of parameters in the model).

Statistical analysis

We performed permutation testing over 1,000 iterations with shuffled data. The p value is the proportion of a metric (F score and t score) in the permuted null distribution greater than or equal to the metric computed using intact data. The minimum p value achievable with this procedure is 0.001. The p values were corrected for multiple comparisons using the false discovery rate, when applicable (Benjamini & Hochberg, 1995).

Results

In a dual-task protocol, we cued participants to perform horizontal or vertical saccades and presented dynamic orientation discrimination signals at various time points throughout saccade preparation to precisely quantify the buildup of presaccadic visual sensitivity at the saccade target (Figure 1). We focus our study on characterizing and comparing the temporal dynamics of presaccadic attention during upward and downward saccade preparation. Results from horizontal saccades (tested in separate experimental sessions) are reported in Supplementary Materials (see Discussion).

As in previous studies reporting a difference in performance as a function of saccade direction (Hanning et al., 2024; Hanning, Himmelberg, et al., 2022), we first conducted a two-way repeated measures analysis of variance, with eye movement condition (fixation/saccade) and location (upper/lower vertical meridian) and as within-subject factors (Figure 1C). Robust effects of presaccadic attention are consistently observed around 100 ms before saccade onset (Hanning et al., 2024; Hanning, Himmelberg, et al., 2022; Kroell & Rolfs, 2021; Kusunoki & Goldberg, 2003; Li et al., 2016, Li et al., 2019; Li, Pan, & Carrasco, 2021; Ohl et al., 2017; Rolfs & Carrasco, 2012; Szinte, Carrasco, Cavanagh, & Rolfs, 2015). Therefore, sensitivity in the saccade condition was calculated using all trials that fell within the desired presaccadic window of test offset 100 to 0 ms before saccade onset. There were main effects of both eye movement condition (F(1, 9) = 27.279; p < 0.001) and location (F(1, 9) = 6.857; p = 0.025), as well as an interaction between them (F(1, 9) = 5.234; p = 0.049).

Post hoc t tests showed no difference in sensitivity at the two locations during fixation (t(9) = 0.051, p = 0.961, corrected), confirming that the tilt angle titration (see “Titration procedures”) led to comparable performance along the vertical meridian. Interestingly, sensitivity in the saccade condition was significantly higher in the lower than the upper vertical meridian (t(9) = 3.337, p = 0.018, corrected), which accounted for the interaction effect. Consistent with these findings, the magnitude of the presaccadic benefit (i.e., the difference between sensitivity in the saccade condition [within 100 ms before saccade onset] and the fixation condition) was larger at the lower than the upper vertical meridian (t(9) = 2.288, p = 0.048).

To investigate the temporal dynamics of presaccadic attention for upward and downward saccades, rather than evaluating overall performance within a relatively large presaccadic window (i.e., within 100 ms prior to saccade onset), we binned trials in the saccade condition (a) as a function of the interval between test Gabor offset and saccade onset using a 60-ms moving average (step size of 30 ms) and (b) as a function of the interval between cue onset and test Gabor offset in a 60-ms moving average (step size of 30 ms), for test Gabor offsets between 0 and 220 ms after cue onset (see “Behavioral data analysis”). Figure 2 shows orientation discrimination sensitivity (d′) throughout the time course of saccade preparation. The first approach is commonly used to examine the presaccadic attentional benefit, which peaks close to and thus is time-locked to saccade onset (e.g., Li et al., 2016; Rolfs & Carrasco, 2012). We also evaluated the presaccadic benefit relative to cue onset (e.g., Hanning et al., 2019) as a second approach to investigate whether the initial selection of the saccade target may differ across locations. Whether the two approaches yield similar results may depend on interindividual differences in participants’ saccade latencies.

Figure 2.

Presaccadic visual sensitivity at the upper and lower vertical meridians, averaged across participants, time-locked to (A) saccade onset and (B) cue onset. Each datapoint is computed from trials that fall within a 60-ms sliding time window. Shaded error regions are ± 1 SEM. For horizontal results, see Supplementary Figure S1.

To characterize the buildup of presaccadic attention over time, we fit two models to the binned data (see “Model fitting”): (a) a linear regression model with a slope and an intercept parameter (Figure 3) and (b) a nonlinear sigmoidal model with amplitude, inflection point, slope, and intercept parameters. Model comparison with the AIC, which selects the better model taking into account the trade-off between goodness of fit and model complexity, revealed no significant difference between the two models (relative to saccade onset: t(9) = 1.836, p = 0.318; relative to cue onset: t(9) = 8.739, p = 0.083, corrected). We therefore selected the more parsimonious, two-parameter linear model for further analysis, which provided a good explanation for our data irrespective of the distance between time bins (i.e., number of bins) and the width of each bin.

Figure 3.

Linear regression model. Average slope and intercept estimates for the upper and vertical meridians, time-locked to (A) saccade onset and (B) cue onset. (A, B) Gray lines show individual data. *p < 0.05, corrected; n.s., not significant. (C, D) Fitted model on an example participant, time-locked to (C) saccade onset and (D) cue onset. The size of the colored dots is proportional to the number of trials in that 60-ms bin, and the black line shows the model fit.

When fitting the linear model, we observed a significant difference in the slope between the upper and lower vertical meridians, as a function of both saccade onset (Figure 3A, left panel; t(9) = 3.001, p = 0.030, corrected) and cue onset (Figure 3B, left panel; t(9) = 3.145, p = 0.024, corrected). These results indicate that presaccadic attention rose more steeply and reached a higher maximum at the lower than the upper vertical meridian. Importantly, there was no difference in the intercept parameter, both as a function of saccade onset (Figure 3A, right panel; t(9) = 0.854, p = 0.415, corrected) and cue onset (Figure 3B, right panel; t(9) = 0.410, p = 0.706, corrected). The intercept parameter corresponded to the sensitivity at the earliest time point, soon after cue onset and much (∼213.735 ms) before saccades were initiated. At this point, the saccade target was about to be selected, and presaccadic attention was only about to begin shifting to the target (Deubel, 2008). The comparable intercepts for the upper and lower vertical meridians reflecting this very early presaccadic period were consistent with the absence of any performance difference between these locations during fixation. Therefore, the slope difference in the time course of presaccadic attention between the upper and lower vertical meridians was not due to differences in baseline sensitivity, which we equated via titration. For an example participant's model fit, see Figures 3C, 3D. Note that we obtained the same pattern of results in Figure 3 with different bin step sizes and widths (see Supplementary Figure S2 for an example). We report the number of trials in each time bin in Supplementary Table S1.

We next evaluated the possible contribution of differences in eye movement parameters to the observed perceptual effect (Table 1). There was no significant difference in saccade amplitude (t(9) = 1.718, p = 0.180, corrected) or precision (i.e., Euclidean distance between saccade offset location and center of the saccade target; t(9) = 1.170, p = 0.272, corrected) between the upper and lower vertical meridians, consistent with previous studies (e.g., Hanning, Himmelberg, et al., 2022; Kwak et al., 2023). In addition, although saccade latencies were numerically faster for upward than downward saccades, this difference was not significant (t(9) = 1.785, p = 0.108, corrected), unlike in previous studies that reported significantly slower saccade latencies for saccades toward the lower than the upper vertical meridian (e.g., Hanning, Himmelberg, et al., 2022; Honda & Findlay, 1992; Kwak et al., 2023; Tzelepi, Laskaris, Amditis, & Kapoula, 2010; but see Irving & Lillakas, 2019). This difference from previous results may in part be due to the use of dynamic visual streams, for which latencies have not been compared for saccades directed toward the upper and lower vertical meridians.

Table 1.

Saccade parameters as a function of saccade direction, averaged across participants. Mean ± 1 SEM. Precision is the mean Euclidean distance of saccade landing positions from the center of the saccade target location.

Parameters	Upper vertical meridian	Lower vertical meridian
Amplitude (°)	8.484 ± 0.198	8.732 ± 0.115
Precision	1.221 ± 0.082	1.090 ± 0.078
Latency (ms)	210.504 ± 3.872	216.967 ± 6.060

Although at a group level, saccade latencies were comparable along the vertical meridian, we nonetheless evaluated whether the slope difference observed between presaccadic attention buildup at the upper versus lower locations could result from latency differences at an individual level (Supplementary Figure S3). This was crucial, as prolonged latencies would give room for longer/higher presaccadic improvement, which could compromise a fair comparison of the temporal dynamics of presaccadic attention as a function of test offset relative to saccade onset. We observed no correlation between individual participants’ saccade latency and slope across participants and locations (r = −0.111, p = 0.641, corrected). The slope difference is thus not a by-product of saccade latency but instead reflects the property of saccade direction.

Discussion

We have investigated whether the temporal dynamics of presaccadic attention differ across saccade directions—specifically, between upward and downward saccades along the vertical meridian. Building on previous findings that the magnitude of presaccadic enhancement is weaker for upward saccades (Hanning et al., 2024; Hanning, Himmelberg, et al., 2022), we asked whether these asymmetries could be explained by differences in the buildup of attention over time. Using individually titrated orientation discrimination thresholds to equate baseline performance, we tracked sensitivity as a function of both saccade and cue onset.

Our results show that presaccadic attention builds up faster and reaches higher levels before the movement for downward saccades than for upward saccades. This directional difference in attentional dynamics was robust across different time-locking and time-binning schemes, and it was present despite comparable saccade latencies, amplitudes, and endpoint precision. Interestingly, we found that the direction-specific time course of presaccadic attention is no better described by a nonlinear sigmoidal function than a simpler linear function, suggesting that differences in the time of onset and/or saturation of the attentional benefit are not needed to explain the reduced benefit at the upper than the lower vertical meridian reported previously. If the reduced enhancement in sensitivity were due to a delayed onset of attentional buildup and/or saturation, the nonlinear sigmoidal model would have provided a better explanation of the data.

Our findings provide compelling evidence that the time course of presaccadic attention is not fixed but varies systematically across saccade directions. Although presaccadic attention is traditionally described as a stereotyped, obligatory process (Deubel & Schneider, 1996; Kowler et al., 1995), our results indicate that its temporal deployment is more flexible. The difference in the slope of the attentional time course across directions suggests that the benefit is not simply delayed or fully absent before upward saccades but rather builds up more gradually. This temporal asymmetry may account for previously observed direction-dependent differences in perceptual sensitivity (e.g., Hanning et al., 2024; Hanning, Himmelberg, et al., 2022; Kwak et al., 2024). Together, these results highlight the need for models of saccade programming and presaccadic attention to incorporate directional factors to comprehensively explain how they benefit perception. They also raise new questions about how the visual and oculomotor systems coordinate for saccades in different directions. For example, at which stage in the processing hierarchy does this directional asymmetry arise? Do the temporal dynamics of key oculomotor regions—the frontal eye fields, parietal eye fields, and the superior colliculus—differ during saccade programming to different directions, and if so, how?

Importantly, this difference in presaccadic attention buildup cannot be explained by low-level sensitivity differences between upper and lower visual field locations. In the fixation condition—when participants maintained gaze and no saccade was prepared—orientation discrimination performance was equivalent at the upper and lower vertical meridians. This confirms that the observed directional difference emerged only under saccade preparation. Likewise, intercept parameters in the time course models (reflecting performance at early, preattentional time points) did not differ between these two locations, further ruling out baseline asymmetries as a source of the effect.

We also ruled out polar angle differences in saccade latencies as a potential driver of the observed performance difference. In some studies, downward saccades have been associated with longer latencies (e.g., Hanning, Himmelberg, et al., 2022; Honda & Findlay, 1992; Kwak et al., 2023; Tzelepi et al., 2010; but see Irving & Lillakas, 2019), potentially allowing more time for presaccadic attention to build up. We, however, found no evidence for such a latency effect here; saccade latencies were statistically equivalent across directions. Moreover, there was no correlation between saccade latency and the slope of attentional buildup. If anything, one might expect a negative relation—longer latencies giving attention more time to accumulate, thus producing shallower slopes. This would have resulted in a shallower slope for downward saccades, which is the opposite of what we found here. Therefore, we conclude that the observed differences in time course reflect genuine directional differences in how attention is allocated during saccade preparation.

Crucially, the observed performance difference cannot be due to covert attention. Across several studies using the same timing as in the present experiment, we have directly compared the effects of presaccadic and covert attention and found robust effects of presaccadic attention but no effect of covert attention (Li et al., 2016; Li et al., 2019; Rolfs & Carrasco, 2012). Moreover, both endogenous (Purokayastha, Roberts, & Carrasco, 2021; Tünçok, Carrasco, & Winawer, 2024) and exogenous (Cameron et al., 2002; Carrasco et al., 2001; Roberts, Ashinoff, Castellanos, & Carrasco, 2018; Roberts, Cymerman, Smith, Kiorpes, & Carrasco, 2016) covert spatial attention improve discriminability to a similar degree around polar angle. Without eye movements, only adaptation has been shown to alter the shape of the performance fields: The adaptation effect is stronger along the horizontal than the vertical meridian, making perception more uniform across the visual field (Lee & Carrasco, 2025b). But notwithstanding this adaptation difference, covert attention benefits are uniform across the visual field and—unlike presaccadic attention—do not vary between the horizontal and vertical meridians (Lee & Carrasco, 2025a).

How might such directional asymmetries in the time course of presaccadic attention arise? One possibility is that the oculomotor system prioritizes certain directions due to biomechanical or ecological factors. On the one hand, downward saccades may be associated with more urgent or salient information (e.g., food, terrain), consistent with a higher probability of downward saccades being the initial saccade during free viewing of natural scenes (Foulsham & Kingstone, 2010). This could lead to more efficient or more strongly reinforced attentional deployment over time. However, we do note that this possibility remains to be further tested as there are mixed reports on the relative frequency of upward and downward saccades and fixations (Najemnik & Geisler, 2005; Tatler & Vincent, 2008). On the other hand, studies using visual tasks during natural scene viewing have shown that downward saccades require more effort to be executed—as indicated by greater pupil dilation when preparing downward than upward saccades—which may explain why they are less preferred than upward saccades in such more controlled settings (Koevoet, Strauch, Naber, & Van der Stigchel, 2025; Koevoet et al., 2024). One speculation is that the greater motor effort for downward saccades may increase activation in oculomotor control regions, which in turn accelerates the buildup of presaccadic attention before they are executed. Moreover, downward saccades involve distinct extraocular muscle coordination compared to saccades in other directions, including upward saccades (Goffart, 2025). Both rely on synergistic yoke muscles to produce pure vertical movements while preventing unwanted rotation; upward saccades are driven primarily by the superior rectus and inferior oblique, whereas downward saccades depend on the inferior rectus and superior oblique.

Neural anisotropies may also play a role in these directional asymmetries. There are vertical asymmetries in the representation of visual space in the superior colliculus in nonhuman primates (Hafed & Chen, 2016). There are also visual field asymmetries at the retinal level (Kupers, Benson, Carrasco, & Winawer, 2022; Kupers, Carrasco, & Winawer, 2019) and within early visual cortex (Benson, Kupers, Barbot, Carrasco, & Winawer, 2021; Himmelberg et al., 2021; Himmelberg, Winawer, & Carrasco, 2022; Silva et al., 2018). These neural differences could affect the temporal evolution of target selection signals across saccade directions. For instance, assuming constant neural density (Rockel, Hiorns, & Powell, 1980), the larger cortical surface corresponding to the lower than the upper vertical meridian for a given eccentricity could yield a higher signal-to-noise ratio and faster accumulation of evidence, leading to a steeper presaccadic buildup of attention.

Our findings can also help to clarify and extend the interpretation of previous studies that have documented weaker presaccadic perceptual performance at the upper vertical meridian. The vertical meridian asymmetries have often been attributed to differences in low-level visual factors such as perceptual contrast sensitivity and spatial resolution (Abrams et al., 2012; Barbot et al., 2021; Carrasco et al., 2001; Talgar & Carrasco, 2002). A computational observer model has shown that optical factors and photoreceptor density do not contribute to this effect, whereas midget retinal ganglion cells contribute moderately (Kupers et al., 2022; Kupers et al., 2019). Importantly, the present study reveals that the temporal dynamics of presaccadic attentional deployment is a critical factor previously overlooked. Specifically, a slower buildup of perceptual sensitivity during presaccadic attentional selection may leave less time for perceptual enhancement to reach full strength before the eye movement begins. This consideration can help reconcile recent psychophysical and physiological findings.

For example, the onset latency of the pupil light response (PLR), which has been used to track covert and presaccadic attention (Mathôt, van der Linden, Grainger, & Vitu, 2013; Mathôt & Van der Stigchel, 2015), is reduced for both upward and downward saccades, suggesting that presaccadic attention shifts to both the upper and lower vertical meridians (Koevoet, Naber, Strauch, & Van der Stigchel, 2025). However, the behavioral benefit of presaccadic attention has generally been found to be more pronounced for downward saccades (Hanning et al., 2024; Hanning, Himmelberg, et al., 2022). Based on the current study, we suggest that because attention builds up slowly before upward saccades, the perceptual consequence at the upper vertical meridian could be diminished even with deployment of presaccadic attention at that location.

To account for the discrepancy between their physiological findings and our previous psychophysical results, Koevoet and colleagues (2025) propose that presaccadic attention itself shifts equally upward and downward, as directly captured by the PLR, and that only the perceptual consequences—as measured in discrimination tasks—differ across directions. On this basis, they propose that the PLR may serve as a better indicator of presaccadic attention because it does not rely on manual reports. Whereas the PLR is certainly informative and interesting, presaccadic attention—defined as the selective allocation of processing resources and enhanced perception at the impending saccade target—has been assessed primarily through perceptual detection and discrimination tasks (review: Li, Hanning, et al., 2021). Moreover, as saccade onset approaches, presaccadic attention builds up gradually, with increasingly stronger effects on perceptual performance (Castet et al., 2006; Deubel, 2008; Hanning et al., 2019; Kroell & Rolfs, 2021; Rolfs & Carrasco, 2012) and representation (Li et al., 2016; Ohl et al., 2017), alongside concomitant costs at other peripheral locations (Deubel & Schneider, 1996; Kowler et al., 1995; Li, Pan, et al., 2021; Montagnini & Castet, 2007) and even at the fovea (Hanning & Deubel, 2022). Whether the PLR can capture this gradual temporal buildup or the associated costs remains an open question. Future research should examine this potential dissociation between perceptual effects and the pupillary response, which would add to the list of intriguing perception–action dissociations (Naber, Frässle, & Einhäuser, 2011; review: Carrasco & Spering, 2024; Spering & Carrasco, 2015).

Finally, our findings raise the question of why presaccadic attention builds up more slowly before upward than downward saccades. One possible contributing factor relates to polar angle asymmetries in the speed of information accrual—the rate at which sensory information becomes available for perceptual decisions. Spatial aspects of polar angle asymmetries in visual perception are well documented (review: Himmelberg et al., 2023) and have been investigated more extensively than their temporal counterparts. Notably, it has been shown that information accrual is slower at the upper than at the lower vertical meridian (Carrasco, Giordano, & McElree, 2004). This relatively slower accrual may limit the ability of presaccadic attention to exert its effect at the upper vertical meridian. Interestingly, exogenous (involuntary) covert spatial attention speeds information accrual more at the upper than at the lower vertical meridian, thereby eliminating this temporal vertical meridian asymmetry (Carrasco et al., 2004). This effect contrasts with our current findings with presaccadic attention, which builds up more rapidly at the lower than the upper vertical meridian—a pattern that may underlie the exacerbated polar angle asymmetries reported previously (Hanning et al., 2024; Hanning, Himmelberg, et al., 2022; Kwak et al., 2024). These distinct and opposite effects of covert and presaccadic attention contribute to the growing body of research documenting dissociations between their effects (review: Li, Hanning, & Carrasco, 2021).

In this study, we focused primarily on the vertical meridian, reporting results for horizontal saccades in the Supplementary Materials. This decision was driven by an unexpected inconsistency in baseline sensitivity across locations, not a lack of interest. Specifically, performance in the earliest time bin in the saccade condition—when presaccadic attentional benefits are not yet expected (Deubel, 2008)—was significantly higher at the horizontal meridian than at either the upper or lower vertical meridian. This was reflected in a significantly higher intercept in the linear fits to the horizontal data (Supplementary Table S2), despite the fact that performance in the fixation condition was matched across all locations. These elevated starting values much before saccade execution are puzzling and do not align with the baseline performance observed during titration or fixation. We cannot offer a definitive explanation for this discrepancy, but it prevents a valid comparison of temporal slopes across the horizontal and vertical meridians. For this reason, we focused our main analyses and interpretations on the vertical meridian, where titration based on fixation was effective for equating starting sensitivity and the resulting time courses could be meaningfully compared. However, if performance in the early time bins were matched across all saccade directions, we would expect the temporal buildup preceding horizontal saccades to resemble that of downward saccades or be even faster, rather than resembling or lagging behind that of upward saccades. This expectation is based on evidence that during the same time window, presaccadic attentional benefits are larger at horizontal than upward (Hanning et al., 2024; Hanning, Himmelberg, et al., 2022) or vertical saccade targets (Kwak et al., 2024), implying a relatively faster buildup.

Conclusions

In conclusion, our results demonstrate that the time course of presaccadic attention differs across saccade directions and reveal a possible mechanism of why the effect is weaker at the upper compared to the lower vertical meridian. Our findings challenge the notion of a uniform attentional deployment mechanism and instead show that the visuomotor system modulates not only where but also how quickly attention is allocated to the movement goal. This temporal flexibility in attentional dynamics offers new insight into how perception and action are coordinated in a direction-sensitive manner and points to the need to better understand how different temporal signatures of performance—linked to action planning versus decision formation—interact across the visual field. As such, our findings echo Eileen Kowler's influential view that attentional and oculomotor processes are tightly linked but shaped by context and intent—not rigid or reflexive. By showing that even the timing of attentional buildup depends on the direction of an upcoming saccade, we contribute to a growing perspective in vision science that not all saccades—and not all attentional shifts—are equal.