Saccade Preparation Reshapes Sensory Tuning

Abstract

Human observers make large rapid eye movements-saccades-to bring behaviorally relevant information into the fovea, where spatial resolution is high. In some visual tasks [1-4], performance at the location of a saccade target improves before the eyes move. Although these findings provide evidence that extra-retinal signals evoked by saccades can enhance visual perception, it remains unknown whether and how presaccadic modulations change the processing of feature information and thus modulate visual representations. To answer this question, one must go beyond the use of methods that only probe performance accuracy (d') in different tasks. Here, using a psychophysical reverse correlation approach [5-8], we investigated how saccade preparation influences the processing of orientation and spatial frequency-two building blocks of early vision. We found that saccade preparation selectively enhanced the gain of high spatial frequency information and narrowed orientation tuning at the upcoming saccade landing position. These modulations were time locked to saccade onset, peaking right before the eyes moved (-50-0 ms). Moreover, merely deploying covert attention within the same temporal interval without preparing a saccade did not alter performance. The observed presaccadic tuning changes may correspond to the presaccadic enhancement [9-11] and receptive field shifts reported in neurophysiological studies [12-14]. Saccade preparation may support transaccadic integration by reshaping the representation of the saccade target to be more fovea-like just before the eyes move. The presaccadic modulations on spatial frequency and orientation processing illustrate a strong perception-action coupling by revealing that the visual system dynamically reshapes feature selectivity contingent upon eye movements.

Figure 1. Experimental procedure and stimulus

(A) Experimental procedure. Each trial started with a 300 ms fixation period followed by the cue presented at the fixation. In the saccade condition, the cue pointed toward the aperture (10° to the left or right of fixation) where the test stimulus would be presented in that trial. In the neutral condition, the cue pointed toward both locations. After a variable delay (SOA uniformly sampled from 12 to 224 ms), the test stimulus (35 ms in duration) was presented at one of the apertures. Observers were instructed to saccade to the cued aperture as fast as possible in the saccade condition and to maintain their fixation at the center throughout each trial in the neutral condition. Observers reported whether the target was present or absent. (B) Stimulus. In half of the trials only the noise was presented; in the other half, the target was presented with noise. The test stimulus was the target embedded in random noise. The noise was filtered white noise and the target was a vertical Gabor with a randomly chosen phase.

Figure 2. Presaccadic enhancement

(A) Top: Observers’ performance, indexed by d´. For the saccade condition, the d´ was plotted as a function of the time of test stimulus offset relative to saccade onset. The error bars represent ±1 s.e.m. The neutral baseline was the d´ computed from all the trials from the neutral condition, and the blue shading represents ±1 s.e.m. Bottom: The median of SOA (temporal interval between cue onset and test stimulus onset) distribution of each time bin. The error bars represent ±1 s.e.m. The dark background represents the trials in which the saccade onset occurred before the offset of the test stimulus. (B) Covert attention did not improve performance. In the covert attention condition, the 100% valid cue was presented, just like in the saccade condition, but observers had to maintain steady fixation. Because there was no saccade onset time in this condition, we simulated five time bins corresponding to the five time bins in Figure 2A. We aimed to have the five data points here with the same SOA distribution as the corresponding data points in Figure 2A. We extracted the SOA distribution for each data point in Figure 1A and used the distribution as the constraint to sample the trials from the covert attention condition. For each observer, we computed one d´ for each time bin by averaging over 1000 resampled d´. The error bars denote ±1 s.e.m. See also Figures S1 and S2.

Figure 3. Saccade preparation modulates sensitivity kernel

(A and B) Group-averaged presaccadic (-50~0 ms from saccade onset) and neutral sensitivity kernels. Each pixel in the kernels is the beta weight estimated from a general linear model, indicating the degree to which the noise at each SF-and-orientation component correlated with the behavioral judgments. (C) The difference kernel computed by subtracting the neutral kernel from the presaccadic kernel. The contours in the difference kernel denote the clusters of the components that showed a significant difference between the presaccadic and neutral kernels. (D) Correlation between the original kernel and reconstructed kernels plotted for each observer. A correlation close to 1 represents high separability between SF and orientation (see Supplementary Experimental Procedures).

Figure 4. Saccade preparation reshapes sensory tuning

(A) Group-averaged SF tuning functions (left panel). The data points are plotted with best-fit raised Gaussians. Error bars denote ±1 s.e.m. The vertical lines represent the peak frequency (the SF where the fitted function reached its peak). The dashed line represents the target Gabor. The orange shaded area represents the cluster of SF channels that showed a significant difference between the two conditions. (B) The peak frequency of the presaccadic interval was plotted against the peak frequency of the neutral condition. The dark-green data point and error bars represent group mean and ±1 s.e.m. The light data points represent individual parameter fits. (C) The peak frequency of spatial frequency tuning of three presaccadic time bins (with equal numbers of trials). The data points are the means of individual parameters and the error bars denote ±1 s.e.m. The horizontal position of each data point is determined by the representative time mark of each time bin averaged across observers (see Supplementary Experimental Procedures) (D) Group-averaged orientation tuning functions (left panel). The data points are plotted with the best-fit Gaussians. Error bars denote ±1 s.e.m. The three horizontal bars at the bottom denote the fitted tuning widths (±1; the dashed line denotes the width of the target Gabor). (E) The presaccadic tuning width was plotted against the neutral tuning width. Each data point corresponds to one observer. The dark data point and error bars represent group mean and ±1 s.e.m. The light data points represent individual parameter fits. (F) The width of orientation tuning of three presaccadic time bins (with equal numbers of trials). The data points are the means of individual parameters and the error bars denote ±1 s.e.m. The horizontal position of each data point is determined by the representative time mark of each time bin averaged across observers (see Supplementary Experimental Procedures). See also Figures S1 and S3.