Learning to silence saccadic suppression

Abstract

Perceptual stability is facilitated by a decrease in visual sensitivity during rapid eye movements, called saccadic suppression. While a large body of evidence demonstrates that saccadic programming is plastic, little is known about whether the perceptual consequences of saccades can be modified. Here, we demonstrate that saccadic suppression is attenuated during learning on a standard visual detection-in-noise task, to the point that it is effectively silenced. Across a period of 7 days, 44 participants were trained to detect brief, low-contrast stimuli embedded within dynamic noise, while eye position was tracked. Although instructed to fixate, participants regularly made small fixational saccades. Data were accumulated over a large number of trials, allowing us to assess changes in performance as a function of the temporal proximity of stimuli and saccades. This analysis revealed that improvements in sensitivity over the training period were accompanied by a systematic change in the impact of saccades on performance-robust saccadic suppression on day 1 declined gradually over subsequent days until its magnitude became indistinguishable from zero. This silencing of suppression was not explained by learning-related changes in saccade characteristics and generalized to an untrained retinal location and stimulus orientation. Suppression was restored when learned stimulus timing was perturbed, consistent with the operation of a mechanism that temporarily reduces or eliminates saccadic suppression, but only when it is behaviorally advantageous to do so. Our results indicate that learning can circumvent saccadic suppression to improve performance, without compromising its functional benefits in other viewing contexts.

Keywords: microsaccades; perceptual learning; saccadic suppression.

Fig. 1.

Robust learning on a conventional visual detection-in-noise task. (A) Subjects fixated on a dot and were required to indicate the orientation (±10° of vertical) of a brief peripheral grating embedded in dynamic white noise. The timing of the target and noise was identical on each trial. Individuals spontaneously made small fixational saccades (see eye position trace), allowing sorting of trials depending on the time of the target relative to the nearest saccade (Δt). (B) Mean contrast thresholds across subjects decreased as a function of training day. Each subject completed five contrast staircases of 100 trials on each day. Error bars are 95% CIs across subjects. (C) All subjects (represented as dots) demonstrated learning, having lower contrast thresholds on day 7 than on day 1.

Fig. 2.

Saccadic suppression reduces over the course of training. (A–G) Perisaccadic thresholds (solid black lines) were raised in a broad time window around the saccade on day 1 (relative to no-saccade baseline trials) but systematically decreased as the week progressed. Thresholds were derived by binning trials according to the timing of the target onset relative to the nearest saccade (sliding 150-ms window) and fitting the resulting psychometric functions. Fitted proportion correct is indicated by the background color, with “warmer” colors signifying better performance. Baseline performance, derived from all trials without a saccade within 600 ms of the target stimulus, is indicated by the dashed horizontal lines. (H) Suppression magnitude, computed as the threshold from a broad suppression window (−100 ms < Δt < 100 ms) minus the baseline threshold (|Δt| < 0.6 s), as a function of training day. Error bars indicate bootstrapped 95% CIs.

Fig. 3.

Suppression reduction is not due to learning-related changes in saccade parameters. (A and B) Saccade rate fluctuates as a function of noise and target onsets, both of which are highly predictable. This behavior arises quickly within the first day of training and continues to develop as training progresses. (C) The number of trials with saccade timings falling within the suppression interval (−100 ms < Δt < 100 ms) drops after day 1. (D) Accounting for the change in trial numbers, suppression is predicted to be flat across the week, indicating that the reduction in suppression is not due to rate changes (see Materials and Methods for further details). (E and F) The majority of saccades were oriented horizontally either toward or away from the target. (G) The proportion of target-oriented saccades during the suppression interval increased modestly during training (Top) and suppression was slightly lower for target-oriented saccades (Bottom). (H) Suppression predicted from the change in orientation ratio decreased only slightly across the week and did not account for the magnitude of suppression reduction that we observed. (I and J) Histograms of saccade amplitude on days 1 and 7. (K) Median saccade amplitude during the suppression interval dropped slightly across the week (Top), and there were differences in suppression magnitude depending on saccade amplitude (Bottom). (L) Suppression predicted from the change in saccade amplitude was flat across the week, indicating that the reduction in suppression is not due to changes in amplitude. All error bars show 95% CIs.

Fig. 4.

Suppression reduction transfers to untrained spatial locations. (A) The Spatial transfer task differed from the main task in two respects: The target was presented to the left (rather than the right) of fixation and subjects indicated whether the grating was ±10° of horizontal (rather than vertical). (B) Mean thresholds were lower after training, indicating some transfer of learning. (C) All but one subject (represented as dots) had lower thresholds on day 7 than on day 1. (D and E) Saccadic suppression was still apparent on day 7 but was less than that observed on day 1. (F) In the window −100 ms < Δt < 100 ms, suppression was reduced on day 7 relative to day 1. All error bars show 95% CIs.

Fig. 5.

Suppression reduction is tuned to the expected time window. (A) The Temporal transfer task differed from the main task in only one respect: The target appeared at a random time within a 1 s interval centred on the expected target onset. (B) Mean thresholds were lower after training, indicating some transfer of learning. (C) All subjects (represented as dots) had lower thresholds on day 7 than on day 1. (D and E) Saccadic suppression was of a similar magnitude on day 7 as it was on day 1. (F) In the window −100 ms < Δt < 100 ms, suppression marginally increased on day 7 relative to day 1. (G) Considering only those trials in which the target occurred near to the expected time, suppression reduction was apparent on day 7 relative to day 1. (H) Conversely, suppression increased on day 7 relative to day 1, if only those trials in which the target deviated from the trained onset time are considered. All error bars show 95% CIs.