Intrasaccadic perception triggers pupillary constriction

Materials and availability

Participant data, the experimental script, and analysis scripts are available from https://github.com/smathot/materials_for_P0018. On-line materials include easy-to-use demonstration software that allows anyone with a sufficiently fast display (> = 150 Hz) to experience intrasaccadic perception.

Participants

Ten naïve observers (age range: 18–21 y; 3 men) participated in the experiment. Participants reported normal uncorrected vision. Participants provided written informed consent prior to the experiment, and received € 20 for their participation. The experiment was conducted with approval of the Comité d’éthique de l’Université d’Aix-Marseille (Ref.: 2014-12-03-09).

Software and apparatus

Eye position and pupil size were recorded monocularly with an EyeLink 1,000 (SR Research, Mississauga, Ontario, Canada), a video-based eye tracker (1,000 Hz; ellipse mode). Stimuli were presented on a gamma-calibrated 21″ ViewSonic p227f CRT monitor (1,024 × 768, 150 Hz). Testing took take place in a dimly lit room. The experiment was implemented with OpenSesame (Mathôt, Schreij & Theeuwes, 2012) using the PsychoPy back-end (Peirce, 2007) for display control, and PyGaze (Dalmaijer, Mathôt & Van der Stigchel, 2014) for eye tracking.

General stimuli and procedure

At the beginning of each session, a nine-point eye-tracker calibration was performed. Before each trial, a small dot was presented 8.5° left of, right of, above, or below the display center. When a stable fixation was detected, a one-point recalibration (drift correction) was performed. Next, the trial started with the presentation of a green fixation dot (36.8 cd/m²) at the same location as the drift-correction stimulus. At the same time, another green dot (the saccade target) was presented at the location opposite from the fixation dot (Fig. 1). After 1 s, an auditory cue (a 100 ms, 440 Hz sine wave) instructed the participant to make a saccade from the initial fixation dot to the saccade target. The trial ended 3 s after the onset of the auditory cue. Participants were instructed not to move their eyes before the cue, and to fixate on the saccade target until the end of the trial. In total, the experiment lasted about 2.5 h. For some participants, data collection was spread across two days, in which case peak saccade velocities (as described below) were determined again on the second day.

Part 1: Peak saccade velocity

In the first part of the experiment, we determined the median peak velocity of saccades for each direction (left, right, down, up) and participant. The background was static and uniformly gray (51.1 cd/m²). Gaze position was sampled on every frame (i.e., every 6.67 ms). Blinks were treated as missing data. On every trial, the maximum of the peaks of the horizontal or vertical velocity profiles for the entire trial was taken as the peak saccade velocity; that is, peak velocity was the highest speed of the eyes at any moment and in any direction. When peak saccade velocity was unrealistic (above 664°/s or below 220°/s, based on a peak velocity of around 400°/s for 17° saccades, see Baloh et al., 1975), the trial was discarded and repeated at a random moment during the remainder of the session. For every saccade direction, peak saccade velocity was based on the median of 40 trials.

Part 2: Intrasaccadic perception

In the second part of the experiment, we investigated the effect of intrasaccadic perception on pupil size. The background was a 22.6° × 22.6° full-contrast sinusoid (5.2 cd/m² to 95.1 cd/m²) grating that reversed in polarity on every frame (Fig. 2). During fixation, the background appeared a homogeneous gray surface, because the frame rate (150 Hz) exceeded the flicker fusion threshold (see Fig. 3A). Furthermore, the display appeared uniformly gray, because the monitor was gamma calibrated: A pixel that alternated between minimum (5.2 cd/m²) and maximum brightness (95.1 cd/m²) had the same mean (fused) luminance as a pixel that was constantly at 50% brightness (50.2 cd/m²). The spatial frequencies of the gratings were set for each participant and saccade direction separately, based on the peak saccade velocities estimated during the first part of the experiment. (Due to a technical issue, for three participants the gratings were set to a default spatial frequency of 0.17 cycles/°. However, this default value was based on pilot testing, and was close to the ideal spatial frequency.) The spatial frequency was such that the maximum distance that the eyes traveled between two frames was equal to half a cycle. The logic behind this is that the flickering background contains ambiguous motion: It can be perceived as moving leftward or rightward with a speed of half a cycle per frame (Fig. 2). Therefore, the peak velocity of the eyes matched the velocity of the motion. As a result, the background approximated a retinotopically stabilized grating when saccade velocity was maximal (Fig. 3B).

In the Intrasaccadic-Percept condition, the orientation of the grating was perpendicular to the direction of the saccade. Therefore, in this condition, participants should briefly perceive a static grating during the saccade, as described above (Fig. 3B). In the No-Percept condition, the orientation of the grating was parallel with the direction of the saccade. Therefore, in this condition, no static grating should be perceived, because the saccade did not stabilize the grating on the retina (Fig. 3C).

Part 3: Subjective report

At the end of the experiment, participants provided a subjective report of how strongly they had perceived ‘something odd’ while making saccades. They provided separate ratings for each saccade direction on a 1–5 scale. Participants did not report the nature of the percept, i.e., whether they had perceived static or moving gratings.

Preprocessing of pupil and eye-position signal

All signals were locked to the moment that the eyes crossed the vertical (for horizontal saccades) or horizontal (for vertical saccades) meridian (from now on: mid-saccade point). Mean pupil size from 105 to 95 ms before the mid-saccade point was takes as a baseline, and all pupil size measures are reported in area (i.e., not diameter) relative to this baseline (cf. Mathôt et al., 2015). We analyzed pupil size from 300 ms before until 1,200 ms after the mid-saccade point. Pupil size during blinks was reconstructed using cubic-spline interpolation (Mathôt, 2013). Pupil size was not smoothed. To obtain an acceptable noise level in eye-velocity profiles, eye position was smoothed using an 11 ms Hanning window.

Trial-exclusion criteria

Trials were discarded when any of the following criteria was met: saccade latency was below 0 ms (i.e., anticipation) or above 2,000 ms; peak saccade velocity could not be determined or was unrealistically high (>1,000°/s; usually due to data loss); or the eyes deviated more than 3.3° from the fixation dot before the saccade (excluding a 200 ms around the mid-saccade point). The eyes deviated more than 3.3° from the saccade target after the saccade (again excluding a 200 ms around the mid-saccade point). No participants were excluded. In total, 4,001 trials (83.3%) remained for further analysis.

Position artifacts in pupil size

In video-based eye trackers, changes in eye position cause artifactual changes in measured pupil size. That is, when the eyes move, the angle from which the camera records the eye changes, and this may lead the eye tracker to register changes in pupil size, even when pupil size remains constant. A common way to correct for position artifacts is to determine a linear regression that predicts (baseline) pupil size from horizontal (X) and vertical (Y) eye position. The X and Y slopes can then be used to ‘regress out’ eye position from pupil size (e.g., Brisson et al., 2013).

In our data, position artifacts are visible as rapid changes in pupil size during the saccade (Figs. 5A, 5D, 5G and 5J); these changes are too rapid to be real. But when we tried to eliminate position artifacts through linear regression, we noticed the following: When fixating the top of the screen, the pupil was larger than when fixating the bottom of the screen (data not shown), which suggests that pupil size is overestimated when fixating the top of the screen. But when participants made a top-to-bottom saccade, pupil size increased even further (Fig. 5J), which suggests that pupil size is underestimated when fixating the top of the screen. In other words, the direction of the position artifact seems to depend on whether you look at baseline pupil size, or at the rapid change that occurs during the saccade. This implies that there are two distinct effects of eye position on pupil size as measured by the eye tracker: The first is a real effect, so that (in our case) the pupil slightly dilates when fixating the top of the screen, perhaps due to effort or the eye’s musculature; the second is an artifact, so that (in our case) the size of the pupil is underestimated when fixating the top of the screen. We are currently investigating this issue further, but for now we note that: not all effects of eye position on pupil size are artifactual (as generally assumed); and any corrective technique that assumes this is problematic.

Therefore, we decided not to correct our pupil-size measurements for position artifacts, and absolute changes in pupil size that occur while the eyes are moving (as shown in Figs. 5A, 5D, 5G and 5J) should be interpreted with caution.

Statistical analyses

Unless otherwise specified, we used linear mixed-effects models (LME) with by-participant random intercept and by-participant random slopes for all predictors. Identical analyses were conducted separately for each 1 ms sample. We did not estimate p-values, but considered effects reliable if they correspond to t > 2 for at least 200 consecutive 1 ms samples (cf. Mathôt et al., 2014). However, we emphasize effect sizes and overall patterns.

Intrasaccadic perception induced pupillary constriction

The main result is shown in Fig. 4A, in which pupil size is plotted over time as a function of condition (Intrasaccadic Percept vs No Percept). The pupil initially dilated in both conditions, until about 200 ms after the mid-saccade point, reflecting motor preparation (Jainta et al., 2011; Wang, Brien & Munoz, 2015). Next, the pupil constricted in both conditions from about 220 ms after the mid-saccade point. Such a constriction is typically observed after saccades (Mathôt et al., 2015; Zuber, Stark & Lorber, 1966). Crucially, as predicted, this constriction was more pronounced in the Intrasaccadic-Percept Condition than in the No-Percept Condition. Based on an LME with pupil size as dependent measure, and Condition (reference: No Percept), Saccade Direction (reference: Horizontal), and their interaction as fixed effects, the effect of condition was reliable from 306 ms after the mid-saccade point until the end of the analysis period. As can be seen in the indivual-participant data (Fig. 4B), this effect was highly consistent across participants.

Differences between saccade directions

Based on the LME described above, there was a reliable interaction between Condition and Saccade Direction from 426 ms until the end of the analysis period. This interaction indicated that the Condition effect was driven mainly by horizontal saccades. We confirmed this by analyzing each saccade direction separately, based on four separate LMEs with Condition as fixed effect and pupil size as dependent measure (Fig. 5). There was a strong and reliable Condition effect for leftward and rightward saccades (Figs. 5A and 5D). There was also a reliable Condition effect for downward saccades (Fig. 5G), but it was much weaker. For upward saccades, there was no reliable Condition effect (see Fig. 5J). In addition, there was considerable sample-to-sample variability in the model estimates for upwards saccades (as shown by the jittery error bands in Fig. 5J), suggesting that this subset of data was especially noisy.

Although there are known differences between saccades in different directions (Collewijn, Erkelens & Steinman, 1988; Van der Stigchel & Theeuwes, 2008), we had not expected saccade direction to interact with the effect of intrasaccadic perception on pupil size. Therefore, we conducted several post-hoc analyses to better understand these differences.

First, peak velocities are more variable for vertical than horizontal saccades (compare Figs. 5C, 5F, 5I and 5L). This was the case for all participants (two-sided paired-samples t-test using per-participant standard deviation of peak velocity as dependent measure: t(9) = 6.42, p = .0001). This variability is important, because the intrasaccadic percept was optimized for a peak velocity of around 400°/s (depending slightly on the participant and saccade direction). Therefore, the intrasaccadic percept may have been less salient for vertical than horizontal saccades. To control for this, we determined the difference between the actual and the optimal peak velocity (from now on: peak-velocity error) for each trial. Next, we selected trials on which peak-velocity error was less than the median peak-velocity error (separately for each participant and saccade direction).

Second, vertical saccades are more curved than horizontal saccades (Van der Stigchel & Theeuwes, 2008). Therefore, the velocity component that is perpendicular to the saccade direction (i.e., horizontal velocity during vertical saccades, and vertical velocity during horizontal saccades; from now on: orthogonal velocity) was higher for vertical than horizontal saccades (Fig. 6). This was the case for all participants (two-sided paired-samples t-test using per-participant mean peak orthogonal velocity as dependent measure: t(9) = 10.25, p < .0001). Peak orthogonal velocity during vertical saccades often approached 150°/s. In the No Percept condition, this would have led to partial retinal stabilization of the grating, which is sufficient to trigger an intrasaccadic percept (Garcıa-Pérez & Peli, 2001). In other words, intrasaccadic perception may have triggered a pupillary constriction in both conditions, thus reducing the Condition effect. To control for this, we selected trials on which peak orthogonal velocity was less than the median peak orthogonal velocity (seperately for each participant and saccade direction).

Next, using vertical saccades from this subset of data, selected as described above, we performed the same analysis as before. Based on an LME with Condition as fixed effect and pupil size as dependent measure, we now also observed a reliable Condition effect for vertical saccades (Fig. 7). Therefore, it is likely that the Condition effect for vertical saccades was reduced by variability in peak velocity, and pronounced curvature.

Subjective ratings

Figure 8 shows the participants’ subjective ratings of how strongly they had perceived ‘something odd’ during saccades, separately for each saccade direction. These ratings suggest that all participants experienced an intrasaccadic percept, despite the fact that all participants were naïve and intrasaccadic perception was not mentioned until the end of the experiment. There is a tendency for the intrasaccadic percept to be rated more salient for horizontal than vertical saccades, but this tendency is weak and unreliable (F(3, 27) = 0.918, p = .446; based on a Repeated Measures Analysis of Variance with saccade direction as independent variable and rating as dependent measure).