Microsaccadic suppression of peripheral perceptual detection performance as a function of foveated visual image appearance

Microsaccades are known to be associated with a deficit in perceptual detection performance for brief probe flashes presented in their temporal vicinity. However, it is still not clear how such a deficit might depend on the visual environment across which microsaccades are generated. Here, and motivated by studies demonstrating an interaction between visual background image appearance and perceptual suppression strength associated with large saccades, we probed peripheral perceptual detection performance of human subjects while they generated microsaccades over three different visual backgrounds. Subjects fixated near the center of a low spatial frequency grating, a high spatial frequency grating, or a small white fixation spot over an otherwise gray background. When a computer process detected a microsaccade, it presented a brief peripheral probe flash at one of four locations (over a uniform gray background), and at different times. After collecting full psychometric curves, we found that both perceptual detection thresholds and slopes of psychometric curves were impaired for peripheral flashes in the immediate temporal vicinity of microsaccades, and they recovered with later flash times. Importantly, the threshold elevations, but not the psychometric slope reductions, were stronger for the white fixation spot than for either of the two gratings. Thus, like with larger saccades, microsaccadic suppression strength can show a certain degree of image-dependence. However, unlike with larger saccades, stronger microsaccadic suppression did not occur with low spatial frequency textures. This observation might reflect the different spatio-temporal retinal transients associated with the small microsaccades in our study versus larger saccades.


Subjects and ethical approvals
We recruited 12 human subjects for this study.Of these, seven were female, and five were male.The subjects were aged 21-42 years, and each took part in three experimental sessions.The first and last sessions included 690 trials each, and the second session included 675 trials.Two subjects had too few baseline trials (defined explicitly in more detail below), so they each performed an addi9onal session of 600 trials.The subjects individually took up to three short breaks during each session.All subjects consented to the experiment, and the procedures were approved by ethical commi_ees of the Medical Faculty of the University of Tübingen.The subjects, who were naive to the purposes of the study, were also compensated financially for their 9me.

Laboratory setup
All experiments were performed in the same setup as that used for recent studies (Baumann et al., 2021;Idrees et al., 2020;Idrees et al., 2022).In brief, the subjects sat comfortably 57 cm in front of a CRT display spanning approximately +/-17 deg horizontally and +/-13 deg ver9cally.The display had a refresh rate of 85 Hz and a pixel resolu9on of 41 pixels/deg.The display was also calibrated and linearized for luminance, as we used a similar procedure of collec9ng psychometric curves as in our earlier studies on perceptual thresholds (Baumann et al., 2021;Idrees et al., 2020).The room was otherwise dark.We tracked eye movements using a video-based eye tracker (EyeLink 1000, SR Research), which was desk-mounted.This required stabilizing the head posi9on, which we did using a custom-built device involving a chin rest, a forehead rest, constraints around the temple of the head, and a head band wrapped behind the head (Hafed, 2013).We controlled the experiments using the Psychophysics Toolbox (Brainard, 1997;Kleiner, Brainard, & Pelli, 2007;Pelli, 1997) and Eyelink Toolbox (Cornelissen, Peters, & Palmer, 2002).

S6muli
We asked the subjects to always maintain fixa9on on an image that was presented at the center of the display in every trial.Across trials, three different images were possible.The first two were ver9cal gabor gra9ngs, and the last was a small, white fixa9on spot.The white fixa9on spot was a square of 0.12 deg and 94.91 cd/m 2 luminance.The gabor gra9ngs could have a spa9al frequency of either 0.5 cycles/deg (referred to as the low spa9al frequency gra9ng) or 5 cycles/deg (referred to as the high spa9al frequency gra9ng).These gabors had a s parameter of 1.75 deg, and they thus visually spanned approximately +/-6 deg horizontally and ver9cally on the display.The underlying sine wave luminance of each gra9ng had a contrast of 100%, and we randomly picked one phase on every trial (from 8 possible phases equally spaced between 0 and 2p).The gray background luminance was 22.15 cd/m 2 .To probe perceptual thresholds, we also presented brief flash s9muli, which were squares of 1 deg size (each trial was associated with only one single probe flash presenta9on, as described in more detail below).These probe s9muli were presented at 9.1 deg from the screen center either horizontally or ver9cally, and their luminance varied across trials.This allowed us to collect full psychometric curves of detec9on performance.Specifically, each flash probe had a luminance increment above the background screen luminance of 2, 4, 5, 6, 8, 10, 11, or 14 computer register steps, with each step causing an actual luminance increase of 0.56 cd/m 2 (which we measured a[er display calibra9on).The range of luminance steps used in any given condi9on depended on the 9me at which we presented the probe flash rela9ve to a detected microsaccade (see details below).Specifically, our experiments involved gaze-con9ngent microsaccade detec9on (Baumann et al., 2021;Chen & Hafed, 2013;Idrees et al., 2020), and we expected higher perceptual thresholds for flashes very close to microsaccade onset than for later flashes.Thus, we used luminance increment steps of 2, 5, 8, 11, and 14 for the trials with higher expected perceptual thresholds, and we used 2, 4, 6, 8, and 10 increment steps for the trials with flashes farther away from microsaccades.In post-hoc analyses (see below), we redetected microsaccades and reclassified flash 9mes, and we used all available luminance increments of a given trial type for fifng the psychometric curves.

Experimental procedures
Each session took approximately 50-60 minutes of 9me.In each trial, a central image first appeared (low spa9al frequency gra9ng, high spa9al frequency gra9ng, or white fixa9on spot).The subjects were asked to maintain fixa9on on the image.For the white fixa9on spot, this was easy because it was the only visible item on the display, and it was very small.For the gra9ngs, we instructed the subjects to look somewhere near the middle of the image.This allowed equalizing the eccentricity of the probe flashes across their four possible loca9ons (also see Results).A[er 250 ms from image onset, we started a computer process of monitoring eye posi9ons in real-9me (Baumann et al., 2021;Idrees et al., 2020).If a microsaccade was detected between 250 ms and 700 ms from the image onset, we presented a probe flash for only one display frame at one of four possible loca9ons (as described above).Moreover, the probe flash was triggered at 0, 25, or 75 ms a[er online eye movement detec9on.If no microsaccade was detected at all by 700 ms, we presented the probe flash anyway at one of the four loca9ons.Later, in post-hoc analyses, we checked for the 9me of the nearest microsaccade to flash onset in this case, and we classified the trial according to our standard 9me course analyses (see more data analysis details below).A[er the probe flash, the computer waited for the subject to press one of four bu_ons, indica9ng the perceived flash loca9on (right, le[, up, or down).If no bu_on was pressed a[er 1 second, a text message appeared on the display asking the subject to respond (and guess the flash loca9on if necessary).Our process for online microsaccadic eye movement detec9on was described earlier (Chen & Hafed, 2013) and successfully used for both microsaccades (Chen & Hafed, 2013) and larger saccades (Baumann et al., 2021;Idrees et al., 2020).Briefly, in every millisecond, we collected a running series of the latest 5 ms of eye posi9on samples.Within each collec9on, we es9mated the rate of change of eye posi9on by fifng a line to the collected samples.To reduce effects of noise, we then took the median of the latest 3 slope measurements and flagged a microsaccade occurrence if the value of the slope was larger than a user-adjusted threshold.Note that this procedure necessarily delayed our es9mate of microsaccade onset.This is why we redetected all microsaccades in later offline analyses a[er data collec9on, and we then recalculated probe flash 9mes to actual microsaccade onset 9mes.Also note that with this approach, we did not systema9cally measure pre-microsaccadic perceptual performance.This was the case because it would have required excessively more trials: without experimental control on microsaccade onset 9me, collec9ng pre-microsaccadic performance would entail presen9ng flashes at random 9mes and then collec9ng enough trials to catch ones in which the flashes occurred pre-microsaccadically; with full psychometric curves (requiring repeated presenta9ons of a given flash luminance), this requires much more data sessions per subject.In our previous work, we reached qualita9vely similar conclusions whether we used our current approach or one also including pre-saccadic perceptual suppression trials (Idrees et al., 2020).

Data analysis
We only analyzed trials with bu_on press reac9on 9mes between 300 and 3000 ms.We also only included trials in which there were no flagged eye posi9on samples within +/-250 ms from probe flash onset.Flagged eye posi9on samples could occur due to blinks (missing eye posi9on data) or eye tracker noise (for example, by interference from eye lashes if subjects started squin9ng).We detected all microsaccades using our established methods (M.E. Bellet, Bellet, Nienborg, Hafed, & Berens, 2019;Chen & Hafed, 2013).We then recalculated probe flash 9mes rela9ve to the recalculated microsaccade onset 9mes.This accounted for the fact that online microsaccade detec9on was necessarily always slightly later than actual microsaccade onset (due to the data buffering men9oned above).Trials with saccades near probe flash onset that were larger than 3 deg in radial amplitude were excluded.These were extremely rare; in fact, for all three image types, most microsaccades were less than 1 deg in amplitude (see Results).
To obtain a 9me course of microsaccade-related perceptual threshold eleva9ons (immediately around microsaccades) followed by recovery (for longer latency probe flash 9mes rela9ve to saccade onset), we then classified all trials into three different groups according to the 9me of the closest microsaccade to flash onset.The first group included all trials in which the closest microsaccade to flash onset started within +/-50 ms from the probe flash event (because of our reclassifica9on of microsaccade onset 9mes in post-hoc analyses, there could be very few trials with a flash right before a microsaccade, and that is why we included 50 ms on either side of flash onset 9me here).This group of trials was expected to be associated with impaired detec9on performance (and it was called the group containing the microsaccadic suppression 9me bin).The second group included all trials in which the flash occurred 70 to 150 ms a[er the onset of the closest microsaccade to the flash.This group of trials was expected to show recovery in perceptual thresholds (and the 9me bin associated with it was called the recovery 9me bin).Finally, the third group of trials was that in which there were no microsaccades at all within +/-250 ms from probe flash onset.These trials were called the baseline trials.
We further filtered trials according to microsaccade direc9on.Specifically, we found that the great majority of microsaccades were predominantly horizontal (see Results), consistent with earlier observa9ons (Engbert & Kliegl, 2003;Laubrock, Engbert, & Kliegl, 2005).Therefore, we only included trials in the analyses for which the closest microsaccade to probe flash onset (including the three temporal categoriza9ons described above) was predominantly horizontal (direc9onal angle within <45 deg from the horizontal direc9on).This was reasonable given the ver9cal gra9ngs, which would mean that horizontal microsaccades would be expected to cause the largest sensory transients in the brain a[er they occur (Khademi, Chen, & Hafed, 2020).To assess perceptual performance, for each subject and 9me bin, we plo_ed the propor9on of correct trials as a func9on of probe flash luminance increment above the background luminance.We then fit psychometric curves using the psignifit 4 toolbox (Schu_, Harmeling, Macke, & Wichmann, 2016); we specifically used the cumula9ve gaussian func9on for fifng.We defined the perceptual threshold as the luminance increment of the probe flash yielding 62.5% correct performance rate (given that ours was a 4-alterna9ve forced choice paradigm with a 25% chance performance rate).For each subject and 9me bin, we es9mated the threshold, and we then compared thresholds between condi9ons (e.g.low spa9al frequency versus white fixa9on spot in the suppression 9me bin; or the suppression 9me bin versus the baseline 9me bin) across the popula9on by showing mean and SEM across all subjects.We performed sta9s9cal tests using the Friedman non-parametric test.Specifically, to test for an impact of probe flash 9me within a given image condi9on, we compared thresholds across subjects by grouping the measurements into three 9me bins as the factors of the sta9s9cal analysis (suppression, recovery, and baseline 9me bins).If the Friedman test had a p-value of less than 0.05, we then performed pairwise Wilcoxon signed rank tests to check which factors (9me bins) were associated with thresholds that were different from each other.We consider an alpha value of less than 0.05 as significant in this study.For checking whether the threshold depended on the foveated image type, we again performed a Friedman test, but only on data from within a given 9me bin (e.g. the microsaccadic suppression 9me bin).This 9me, the factors of the sta9s9cal test were the three image types.We then used the same logic of post-hoc pairwise comparisons.We report all pvalues in Results.We also assessed sensi9vity at threshold by measuring the slope of the psychometric curve near the probe flash levels causing threshold performance.To do so, for each curve, we divided the difference in performance (65% minus 60% correctness rate) by the difference in luminance values yielding 65% and 60% correctness rate.This gave an es9mate of the slope of the psychometric curve at threshold.We then performed similar sta9s9cal tests on the slope measurements as we did for the detec9on thresholds.For comparing microsaccade amplitudes as a func9on of foveated image type, we first averaged the microsaccade amplitude per condi9on within each subject's data.Then, we visualized the popula9on results by averaging across subjects, and showing SEM ranges.We also did this for microsaccade peak veloci9es, in order to assess the movements' kinema9cs.Finally, we also characterized where subjects fixated their gaze at the 9me of probe flash presenta9on.To do so, we averaged eye posi9on in the interval from -25 ms to +75 ms rela9ve to probe flash 9me.Then, we plo_ed the measurements across trials for each fixated image 9me.We picked more 9me samples a[er the flash than before because the great majority of flashes were triggered (by design of the experiment) a[er a microsaccadic event, and we wanted to avoid including the microsaccadic displacement itself in the eye posi9on measurement.

Results
We asked subjects to fix their gaze near the center of one of three possible images (Fig. 1A-C).One image was a low spa9al frequency ver9cal gabor gra9ng of 0.5 cycles/deg spa9al frequency (Fig. 1A); the other was a high spa9al frequency ver9cal gabor gra9ng of 5 cycles/deg spa9al frequency (Fig. 1B); and the third was a small white fixa9on spot (Fig. 1C).We then presented a brief probe flash peripherally for just one display frame (at one of the four cardinal direc9ons; 9.1 deg eccentricity), and we asked subjects to indicate where it appeared on the display (4-alterna9ve forced choice paradigm) (Fig. 1A-C).The probe flash was designed to appear at different 9me intervals rela9ve to the occurrence of a microsaccade (Methods; examples shown in Fig. 1D-F), and the microsaccade itself was not explicitly instructed.Rather, the subjects were only told to look at the center of the image, and the computer waited for online microsaccade detec9on in order to trigger the probe flash (Methods).Across all trials, in post-hoc analyses, we searched for the nearest microsaccade to probe flash onset.We then first assessed the metric and kinema9c proper9es of these eye movements.Independent of the underlying foveated image, the majority of microsaccades that occurred were predominantly horizontal (Fig. 2).For the gra9ng images, this likely reflected the ver9cal orienta9on of the gra9ngs, since orthogonal eye movements to the luminance gradient would be expected to give rise to the most useful informa9on to the visual system about the underlying image (Rucci, Iovin, Polef, & San9ni, 2007).This is also consistent with neurophysiological signatures of microsaccade-induced visual reafferent responses at extra-foveal eccentrici9es, in which orthogonal eye movements scaled to a given spa9al frequency give rise to the clearest modula9ons (Hafed, Chen, & Khademi, 2022;Khademi et al., 2020).For the white fixa9on spot, there were slightly more ver9cal eye movements than with the gabor gra9ngs (likely reflec9ng the square appearance of the fixa9on spot, which includes both horizontal and ver9cal edges); nonetheless, the overall predominantly horizontal signature of eye movement direc9ons with the white fixa9on spot was consistent with previous reports (Engbert & Kliegl, 2003;Laubrock et al., 2005).All of these observa9ons led us to focus our remaining analyses on predominantly horizontal eye movements with an absolute direc9on from horizontal of less than 45 deg; we obtained generally similar results when we included all trials into the analyses, as expected given the large number of predominantly horizontal movements seen in Fig. 2. It is also interes9ng to note here that there were barely any downward microsaccades in our data at all (Fig. 2); this might be related to general tendencies of the oculomotor system to bias gaze upward, whether in visual or memory condi9ons (Goffart, Hafed, & Krauzlis, 2012;Goffart, Quinet, Chavane, & Masson, 2006;Khademi et al., 2024;Malevich, Buonocore, & Hafed, 2020;Snodderly, 1987;White, Sparks, & Stanford, 1994;Willeke, Cardenas, Bellet, & Hafed, 2022;Zelinsky, 1996).There was also a significant paucity of downward microsaccades in all condi>ons.
In terms of movement amplitudes, we found that predominantly horizontal microsaccades tended to be very slightly larger for the low spa9al frequency image than for the high spa9al frequency image, and microsaccades were also the smallest in size for the small white fixa9on spot.These results can be seen in Fig. 3A, and they are consistent with the above-men9oned ideas about how fixa9onal eye movement proper9es can be strategically op9mized by the visual-oculomotor system to maximize informa9on gain from the underlying images.Nonetheless, in all cases, the microsaccades that we inves9gated in this study were always significantly smaller than 1 deg in radial amplitude regardless of the underlying image type (Fig. 3A), and they also obeyed the main sequence rela9onship between peak velocity and amplitude (Fig. 3B) (Zuber et al., 1965).Interes9ngly, and as we show explicitly in more detail below, we observed the strongest microsaccadic suppression for the white fixa9on spot condi9on, which had the smallest, and thus slowest, eye movements.We return to this point later in the text, a[er describing the subjects' perceptual performance results in the task.Microsaccade amplitudes reflected the underlying spa>al scale of the viewed foveal image, as expected, but they were always significantly smaller than 1 deg.(B) Same as A but now ploSng the peak velocity of the microsaccades as a func>on of their amplitude.Error bars denote SEM across subjects.The eye movements followed the expected main sequence rela>onship between saccade size and saccade peak speed (Zuber et al., 1965).Also see Fig. 8 below for raw microsaccade amplitude distribu>ons across image types.Thus, given that we have now confirmed the occurrence of small, fixa9onal microsaccades in our experiments, we now had a situa9on rela9vely similar to that described in (Idrees et al., 2020): that is, a rapid eye movement (this 9me, small) was generated across a textured background, and the detec9on of a brief probe flash away from the saccade end point was inves9gated.We now turn to describing how the detec9on of the probe flash varied as a func9on of both its 9me rela9ve to microsaccade onset 9me as well as the underlying foveated visual image appearance.We finish by exploring the poten9al influences of gaze posi9on and microsaccade amplitude differences on the interpreta9on of our results.

Both perceptual detec6on thresholds and sensi6vity are affected in the immediate temporal vicinity of microsaccades
Figure 4 shows psychometric curves characterizing the performance of one example subject (S05) when fixa9ng the low spa9al frequency gra9ng.The baseline curve (gray in Fig. 4A, B) was obtained from all trials in which there were no microsaccades occurring within +/-250 ms from probe flash onset (Methods).For the other shown curves, a predominantly horizontal microsaccade started either within +/-50 ms from probe flash onset (that is, during the expected microsaccadic suppression 9me bin; Fig. 4A) or 70 to 150 ms before probe flash onset (that is, during a recovery 9me bin with the microsaccadic event being sufficiently far away in 9me; Fig. 4B).As can be seen, the subject's performance in the task was clearly impaired in the microsaccadic suppression 9me bin, as evidenced by the lower propor9on of correct trials in every flash level that was neither too difficult (floor effect) nor too easy (ceiling effect).For probe flashes longer a[er a microsaccade (in the recovery 9me bin), performance recovered and approached that observed in the baseline trials characterized by an absence of nearby microsaccades to the probe flash (Fig. 4B).The impairment of the example subject's performance in the microsaccadic suppression 9me bin was manifested in two ways.First, there was an increase in detec9on threshold.For example, during the microsaccadic suppression 9me bin (Fig. 4A), the probe flash needed to be almost approximately 3.1 cd/m 2 brighter than the background to result in a 62.5% correctness rate in task performance (Fig. 4C, le[most data point).This was a Weber contrast value of 0.14.On the other hand, during baseline, the probe flash needed to be only approximately 1.9 cd/m 2 brighter than the background luminance (Fig. 4C, rightmost data point); equivalent to a 0.09 Weber contrast.Second, the sensi9vity of performance to subtle luminance changes of the probe flash was also impaired.This is evidenced by the shallower slope of the psychometric curve of the subject during the microsaccadic suppression 9me bin (Fig. 4A) when compared to the recovery and baseline condi9ons (also quan9fied in Fig. 4D).In the recovery 9me bin (Fig. 4B), the slope of the psychometric curve was more similar to that in baseline, sugges9ng an expected gradual return to baseline sensi9vity with 9me (Fig. 4D).Thus, both the detec9on threshold and sensi9vity (slope of the psychometric curve at the perceptual threshold) of the subject were impaired in associa9on with microsaccades.Across all subjects, microsaccadic suppression affected both detec9on thresholds and sensi9vity (psychometric curve slopes), and for all foveated visual image appearances that we tested.This is best seen by the analyses of Fig. 5. Here, we plo_ed in the le[ column (Fig. 5A, C, E) the detec9on thresholds of all subjects as a func9on of probe flash 9me rela9ve to microsaccade onset 9me.The different panels denote the different fixated images, and the error bars denote SEM across subjects.In each panel, there was an eleva9on of perceptual detec9on thresholds in the microsaccadic suppression 9me bin, which recovered in other 9me windows.As for sensi9vity (the slope of the psychometric curve at the perceptual threshold flash level), the results are shown in Fig. 5B, D, F. Here, the microsaccadic suppression 9me bin was associated with generally reduced sensi9vity (shallower psychometric curves) rela9ve to the other two analyzed 9me windows, and this happened for all foveated image types.Sta9s9cally, these results were robust.Specifically, within each image type, there was a significant effect of flash 9me on perceptual thresholds (p=0.0061992,0.0001872, and 0.00030864 for the low spa9al frequency, high spa9al frequency, and white fixa9on spot, respec9vely; Friedman test with 9me bin as factor).There were also effects on the slopes of the psychometric curves (p= p=0.024611, 0.00050886, and 0.022274 for the low spa9al frequency, high spa9al frequency, and white fixa9on spot; Friedman test with 9me bin as factor).In post-hoc comparisons, the threshold in the microsaccadic suppression 9me bin was systema9cally different from that in the baseline 9me bin (p=0.0073,0.0005, 0.0005 for the low spa9al frequency, high spa9al frequency, and white fixa9on spot, respec9vely; Wilcoxon signed-rank test).Similarly, the slopes tended to also be different between the microsaccadic suppression and baseline 9me bins (p=0.0049,0.0063, 0.0400 for the low spa9al frequency, high spa9al frequency, and white fixa9on spot, respec9vely; Wilcoxon signed-rank test).Thus, both perceptual detec9on thresholds and sensi9vity (psychometric curve slopes) were affected in the immediate temporal vicinity of microsaccades.

Only perceptual detec6on thresholds might depend on the foveated image appearance
Despite the qualita9vely similar results in Fig. 5 across all three foveated visual image appearances, when we quan9ta9vely compared these results, we found that microsaccadic suppression of peripheral perceptual detec9on performance was strongest when viewing the small white fixa9on spot rather than when viewing a low spa9al frequency gra9ng, as we had previously observed with large saccades (Idrees et al., 2020).Consider, for example, Fig. 6A, which combines the threshold plots of Fig. 5 together into one single visualiza9on.In the microsaccadic suppression 9me bin, there was a higher threshold value for the white fixa9on spot than for the low spa9al frequency gra9ng (p=0.01389;Friedman test comparing all three image condi9ons; and p=0.0161; post-hoc Wilcoxon signed-rank test comparing the white fixa9on spot to the low spa9al frequency gra9ng condi9on).This effect also con9nued in the recovery 9me bin (p=0.0022806;Friedman test; p=0.002; post-hoc Wilcoxon signedrank text comparing the white fixa9on spot to the low spa9al frequency gra9ng condi9on), consistent with the higher threshold for the white fixa9on spot in the microsaccadic suppression 9me bin.In the baseline 9me bin, all the detec9on thresholds were sta9s9cally similar to each other (p=0.35255;Friedman test comparing all three image condi9ons in the baseline 9me bin).During microsaccadic suppression, the high spa9al frequency performance was intermediate between the two, and closer to the low spa9al frequency condi9on.Thus, in terms of detec9on thresholds, peri-microsaccadic suppression of peripheral perceptual detec9on performance was strongest for the white fixa9on spot, as opposed to either a low or high spa9al frequency gra9ng.Interes9ngly, unlike the thresholds, the slopes of the psychometric curves of the same subjects did not appear to depend on the foveated visual image appearance during the microsaccadic suppression 9me bin (p=0.59156;Friedman test comparing all three image condi9ons).This can be seen in Fig. 6B.For all image types, the shallower slope during the microsaccadic suppression 9me bin was similar in value.This was also the case for the higher slopes seen in the recovery and baseline 9me bins.Thus, it was only the detec9on thresholds, and not the slopes of the psychometric curves, that showed an imagedependence of microsaccadic suppression of peripheral perceptual detec9on performance.It would be interes9ng in future studies to inves9gate why this was the case.sugges>ng stronger microsaccadic suppression.Moreover, this stronger suppression persisted in the recovery >me bin (70-150 ms from microsaccade onset), and it was only in baseline that the thresholds for all three viewed image types were sta>s>cally similar.Quan>ta>vely, the thresholds during the microsaccadic suppression >me bin were 3.84, 3.44, and 3.36 cd/m 2 , respec>vely, for the white fixa>on spot, high spa>al frequency, and low spa>al frequency gra>ng (equivalent to 0.173, 0.155, and 0.152 Weber contrast, respec>vely).(B) For the slopes of the psychometric curves, there were no differences across viewed image types in any of the >me bins.Thus, only detec>on thresholds showed an image-dependence of microsaccadic suppression in our data.We next considered whether the results of Fig. 6A could be explained by factors other than microsaccadic suppression per se.Specifically, since there was no specific marker to fixate on in the gra9ng images, it could be the case that the subjects were biased in where they directed their gaze during the trials with gra9ng images.For example, if these subjects systema9cally fixated their gaze slightly upward rela9ve to the gra9ng center, then this could have rendered one flash loca9on (the upper one in this example) significantly closer to the gaze center than in the case of the small white fixa9on spot, with much more focused gaze direc9on.This would have made one flash loca9on easier to detect than with the white fixa9on spot.However, this logic fails since a bias in gaze posi9on with the gra9ngs towards one probe flash loca9on would render the three other flashes actually farther away from the gaze center, and therefore harder to detect.If anything, this should have made the overall task harder with the gra9ng images than with the white fixa9on spot.This was clearly not the case in our data (subjects performed worse during the microsaccadic suppression 9me bin with the white fixa9on spot).We also have three addi9onal reasons to rule out a poten9al influence of gaze posi9on (and thus probe flash visibility) on the results of Fig. 6A.First, we explicitly measured gaze posi9on at the 9me of probe flash presenta9on across all trials and image types (Methods).Figure 7 shows these measurements for each subject individually, with the blue dots showing trials with the low spa9al frequency gra9ng and green dots showing trials with the white fixa9on spot.We did not plot the high spa9al frequency gra9ng data in order to reduce clu_er in the figure, but these data were virtually iden9cal to those of the low spa9al frequency gra9ng data (consistent with Fig. 3).The insets show the mean and standard devia9ons of the shown raw data points.As can be seen, while it was certainly true that gaze posi9on was more dispersed with the gra9ng images, as expected, the subjects correctly followed our instruc9ons to maintain their gaze near the center of the image (average gaze posi9on was similar whether the subjects were viewing a gra9ng or a small white fixa9on).Quan9ta9vely, devia9ons in mean gaze posi9on between the white fixa9on spot and the gra9ng cases were always smaller than approximately 0.5 deg, and o[en significantly smaller.Given that our peripheral probes were at 9.1 deg, this small difference in average gaze posi9on was not expected to influence detectability.In fact, even at 5 deg, we found in an earlier study that such gaze posi9on devia9ons of the same magnitude as those observed here did not alter peripheral detec9on performance (J.Bellet et al., 2017).Second, in the baseline 9me bin in Fig. 6A, perceptual performance was similar for all image types.Thus, if gaze posi9on was indeed systema9cally biased for the gra9ngs rela9ve to the white fixa9on spot, then we should have also seen a difference in performance during the baseline 9me bin.This was not the case.Third, the stronger microsaccadic suppression of peripheral detec9on performance for the white fixa9on spot (as opposed to the gra9ng images) was specific for threshold values but not for the slopes of the psychometric curves (Fig. 6B).If gaze posi9on altered the visibility of the peripheral targets between the different image types, then we might have expected similar changes in both thresholds and sensi9vity across image types.Therefore, all of these observa9ons, coupled with the results of Fig. 7, suggest that systema9c gaze posi9on differences across image condi9ons likely do not fully explain the results of Fig. 6.As can be seen, all subjects fixed their gaze at very similar posi>ons between the two image types.There was clearly larger dispersion of fixa>on posi>on with the gra>ng images (due to a lack of a specific punctate marker), but this dispersion was largely symmetric in all direc>ons.Thus, all four peripheral flashes were, on average, at a similar re>notopic eccentricity when they appeared, ruling out a simple re>notopic visibility as the primary explana>on of the results of Fig. 6A.This leaves a final ques9on of microsaccade size itself; that is, it could be possible that microsaccade size could influence the results of Fig. 6A.In Fig. 3, we did indeed observe that microsaccades were smaller and slower, on average, with the white fixa9on spot when compared to the gra9ng images.However, slower movements should cause milder image transients and blurs than faster movements, which should, in principle, be associated with milder saccadic suppression.This is opposite to what we observed experimentally, with stronger microsaccadic suppression for the white fixa9on spot condi9on.Moreover, there seems to be a dissocia9on between saccade speed and saccadic suppression strength in general (Gremmler & Lappe, 2017).And, an early study with large saccades actually documented larger saccadic suppression effects with larger (and faster) saccades (Mitrani, Yakimoff, & Mateeff, 1970), again opposite of what we observed.Nonetheless, we inves9gated our microsaccade amplitude distribu9ons more closely.Despite the differences in average microsaccade sizes that we observed in Fig. 3, there was a large overlap in the raw distribu9ons of microsaccade sizes, as can be seen from Fig. 8 (with most microsaccades being smaller than 1 deg in all condi9ons).Thus, it does not seem likely that the results of Fig. 6A could be fully accounted for by microsaccade size, and we confirmed this (in a control analysis) by excluding all gra9ng image trials containing microsaccades larger than 1 deg in amplitude.This maximized the overlap in microsaccade sizes across all image condi9ons, and we observed similar trends as those shown in Fig. 6.It would be interes9ng in future experiments to relate, within a single image, microsaccadic suppression strength to the sizes of the microsaccades that are generated, in order to explicitly document whether it was the smaller microsaccades in the white fixa9on spot condi9on that fully accounted for our results or not.Therefore, our analyses, combined, suggest that microsaccadic suppression in our experiments did occur for all tested visual image types at the fovea, and that the suppression was stronger when the viewed foveal visual image was a small white fixa9on spot as opposed to either a low or high spa9al frequency texture.

Discussion
In this study, we inves9gated the dependence of peri-microsaccadic suppression of peripheral perceptual detec9on performance on the visual appearance of the images across which microsaccades were generated.This was a microsaccadic correlate of studies in which saccadic suppression was researched with larger saccades being made across textured backgrounds (Idrees et al., 2020).Unlike a poten9al expecta9on from these studies, we did not find stronger microsaccadic suppression for the low spa9al frequency gra9ng.Rather, the strongest suppression occurred when microsaccades were made across a small, foveal fixa9on spot (Fig. 6A).Moreover, only detec9on threshold eleva9ons depended on the foveal visual image appearance, but not the slope reduc9ons of the psychometric curves.Our effect on threshold eleva9ons, being slightly larger for the small white fixa9on spot than for gra9ngs (Fig. 6A), was quite different from what we observed earlier with much larger saccades (Idrees et al., 2020), in which there was stronger saccadic suppression for low than high spa9al frequency background textures.This difference in results might be a manifesta9on of the significantly slower and smaller eye movements studied here when compared to our earlier experiments.For example, with the small eye movements, shi[ing gaze on the low spa9al frequency gra9ng might be more similar to shi[ing gaze over a blank, given how gradually the gra9ng luminance changes with the diminu9ve angular displacements associated with microsaccades.Similarly, with the high spa9al frequency gra9ng, the image displacement caused by the average microsaccade size that we observed (approximately two thirds of a degree; Fig. 3) might have shi[ed the image by a whole number mul9ple of luminance cycles (e.g. 3 cycles), just averaging the luminance modula9ons across the cycles out by the 9me the eye movement was finished.On the other hand, with the broad band fixa9on spot, shi[ing gaze over it would be expected to influence mul9ple spa9al frequency visual processing channels, and might cause stronger saccadic suppression.This interpreta9on is consistent with the idea that saccades of different sizes have different spa9o-temporal profiles of re9nal image modula9ons when they occur (Mostofi et al., 2020).Thus, the interac9ons between background image spa9al frequency content and the strength of saccadic suppression (Idrees et al., 2020) should not always be iden9cal for different saccade sizes; rather, these interac9ons might reflect the specific sensory consequences of the par9cular saccades being generated.Given that microsaccades to a small spot can overshoot it slightly (Tian et al., 2016;Willeke et al., 2022;Willeke et al., 2019), making a microsaccade in our white fixa9on spot condi9on was addi9onally equivalent to crossing a luminace edge (e.g. the preferred re9nal locus went from a gray background to being over a white image patch and then to being over a gray background again by the end of a given microaccade).This is a similar situa9on to our recent observa9on that when large saccades crossed a luminance bar, we observed stronger saccadic suppression than when the saccades were made across a blank (Baumann et al., 2021).Thus, a second poten9al explana9on of the results of Fig. 6A is that the microsaccades with the white fixa9on spot were crossing a luminance discon9nuity.This would s9ll be consistent with a visual component to microsaccadic suppression, as with larger saccades.However, in the current study, the recep9ve fields experiencing the peripheral probe flashes (especially if they were small in early visual areas like re9na, lateral geniculate nucleus, and primary visual cortex) likely never crossed luminance bars when the microsaccades happened with the white fixa9on spot; these recep9ve fields presumably always experienced a gray background since the probe flashes were at a peripheral eccentricity.Thus, it is not clear whether in our earlier study (Baumann et al., 2021), it was the foveal crossing or the crossing of the recep9ve fields seeing the probes of a luminance bar that ul9mately caused the stronger saccadic suppression.It would be interes9ng in the future to inves9gate this issue further.Indeed, all descrip9ons above include an implicit assump9on that the visual condi9ons in the fovea in our current experiments could influence peripheral performance even though the peripheral probe flashes themselves occurred over a completely gray background.However, this is not the first 9me that probes over a gray background were shown to be affected by visual condi9ons far from them.For example, in some experiments in (Idrees et al., 2020), we had probe flashes over a gray background (with the same re9nal locus being s9mulated by gray both before and a[er saccades), and only the far surround having different textures.We s9ll obtained altered saccadic suppression strengths with the differing far surrounds.Thus, it is s9ll possible that our results in Fig. 6A could be affected by the foveal image even though the probes were peripheral.Indeed, the thresholds that we observed in the current study during the microsaccadic suppression 9me bin (e.g.0.173 Weber contrast for the white fixa9on spot) were slightly higher than those observed in (Baumann et al., 2021) with saccades made across a completely blank background in the same experimental setup (0.13 Weber contrast).Thus, it could be the case that the foveal visual condi9ons in the current study could s9ll influence peripheral performance over a gray background.This leads to the intriguing ques9on of how and why peripheral sensi9vity can be impaired so much when 9ny microsaccades occur.In the above example, our perceptual thresholds with the white fixa9on spot were slightly higher than those we observed earlier with much larger saccades over a gray background (Baumann et al., 2021).This also happens at the neuronal level, with even very eccentric recep9ve fields (preferring >20 deg of eccentricity) experiencing massive suppression of visual neural sensi9vity to probe onsets whenever 9ny microsaccades occur (Chen & Hafed, 2017;Hafed, Chen, & Tian, 2015;Hafed & Krauzlis, 2010).Of course, one aspect of this could s9ll be visual.For example, at least in the superior colliculus, a structure relevant for saccadic suppression (Berman et al., 2017;Lee et al., 2007;Phongphanphanee et al., 2011), eccentric recep9ve fields can be very large.Thus, moving them by microsaccadic amounts can s9ll cause luminance transients associated with the display edge moving on the re9na rela9ve to the dark surroundings of the display region.That is, a single peripheral recep9ve field (e.g. in the superior colliculus) can s9ll experience both the display and the dark background within it, and thus be exposed to a visual edge movement whenever 9ny microsaccades occur.There could also be other non-visual components for such a disparate difference between saccade size and the peripheral eccentricity that experiences suppression.For example, if microsaccade-related motor bursts in the superior colliculus were to vary as a func9on of the image appearance, as is the case with larger saccades (Baumann et al., 2023), then a dependence of microsaccadic suppression on foveal visual image appearance could emerge peripherally through extra-re9nal mechanisms (with concepts like corollary discharge).It seems likely that microsaccade-related superior colliculus motor bursts would exhibit imagedependence given the current evidence in the literature so far.For example, these motor bursts can disappear completely for microsaccades made towards a blank (Willeke et al., 2019), just like with larger saccades (Baumann et al., 2023;Edelman & Goldberg, 2001;Mohler & Wurtz, 1976;Zhang et al., 2022).However, an explicit experiment probing collicular microsaccade-related motor discharge with different underlying foveal textures is warranted.Finally, it would be interes9ng in follow-up experiments to present our probe flashes more centrally, such that they s9ll appear on the low or high spa9al frequency gra9ngs themselves, and not over the gray background.In that case, we can expect higher overall detec9on thresholds than with a blank background (Baumann et al., 2021), but it remains to be seen whether microsaccadic suppression of performance would now be stronger for the low spa9al frequency gra9ng than for the high spa9al frequency gra9ng.In such experiments, one can even parametrically change gra9ng size rela9ve to probe flash loca9on, in order to find the extent of overlap between background images and probe flashes that is needed to result in an image-dependence of microsaccadic/saccadic suppression.This can turn allow predic9ng the effec9ve sizes of recep9ve fields that would be most relevant for the visual component of saccadic and microsaccadic suppression (e.g. in re9na, lateral geniculate nucleus, primary visual cortex, superior colliculus, or elsewhere).Overall, we believe that our results demonstrate the relevance of studying saccadic suppression using different eye movement sizes and direc9ons, and also the relevance of considering both visual and motor components of this highly ubiquitous and robust perceptual phenomenon.

Figure 1
Figure 1 Experimental paradigm.(A-C) The images that were fixated in this study (Methods), along with the possible brief probe flash loca>ons (small dim squares in the periphery).In every trial, the subjects were instructed to fixate near the center of the image.At a variable >me from online microsaccade detec>on, a brief probe flash was presented at one of four peripheral loca>ons (right, leE, up, or down from the image center; we show the four probes here simultaneously only for illustra>on purposes because only one flash was presented per trial).The subjects had to indicate where they saw the brief probe flash.The luminance of the probe flash varied from trial to trial, in order to collect full psychometric curves.(D-F) Example eye posi>on traces and probe flash onset >mes (ver>cal red lines) from one example subject (S05) in each of the image condi>ons shown in A-C.An upward deflec>on in the shown eye posi>on traces indicates a rightward gaze shiE for horizontal eye posi>on and an upward gaze shiE for ver>cal eye posi>on.Our goal was to assess perceptual detec>on performance as a func>on of the >me of flash onset rela>ve to microsaccade onset, and also as a func>on of the different underlying foveated images.Note how the probe flash >me was variable rela>ve to microsaccade onset across different trials.

Figure 2
Figure 2 Predominantly horizontal nature of microsaccades in our experiments.Each panel shows the direc>on distribu>on of observed microsaccades (closest to probe flash onset >me in every trial; Methods) for each fixated image of Fig. 1 (pooled across all subjects).As can be seen, most movements were predominantly horizontal.

Figure 3
Figure 3 InteracDon between foveated visual image appearance and microsaccade amplitudes.(A) For all trials with the nearest microsaccade to probe flash onset being predominantly horizontal, we ploRed the radial amplitude of the microsaccade as a func>on of the foveated image type.Error bars denote SEM across subjects.

Figure 4
Figure 4 Impairment in both detecDon threshold and sensiDvity (slope of the psychometric curve at threshold) by microsaccades in an example subject.(A) When viewing the low spa>al frequency gra>ng (Fig. 1A), peripheral probe flashes (at approximately 9 deg eccentricity) had to be higher in luminance to be successfully detected if they occurred within +/-50 ms (blue) from microsaccade onset than if they occurred without any nearby microsaccades within +/-250 ms (gray).This figure shows the results from one example subject (S05).The con>nuous curves show psychometric curve fits to the shown data points (Methods), and each data point's size is scaled by the number of observa>ons collected with the shown probe flash luminance increment above the background luminance.The ver>cal lines show the flash levels resul>ng in threshold performance (the slight yvalue differences at the threshold indica>ons reflect the slightly different asympto>c levels of performance in the two shown condi>ons) (SchuR et al., 2016).(B) Same analyses but with the flashes now occurring 70 to 150 ms aEer microsaccade onset.Performance recovered to near baseline performance.The gray curve is the same as that in A. (C) Detec>on thresholds of this subject as a func>on of flash >me rela>ve to microsaccade onset.There was clear peri-microsaccadic suppression of performance (manifested as a threshold eleva>on).(D) Similar to C but for measures of the slope of the psychometric curves near the threshold values (Methods).Psychometric curves were shallower for probe flashes within +/-50 ms from microsaccade onset.

Figure 5
Figure 5 Summary across subjects.(A) Detec>on thresholds as a func>on of flash >me from microsaccade onset when viewing the low spa>al frequency gra>ng.This figure is similar to Fig. 4C, but averaging across all subjects.Error bars denote SEM across subjects.There was a threshold eleva>on near microsaccade onset, and recovery for larger temporal separa>on between microsaccades and flash >mes.(B) Same as A but for the slopes of the psychometric curves when viewing a low spa>al frequency gra>ng.This figure is, thus, similar to Fig. 4D.(C, D) Same as A, B, but for the case of viewing the high spa>al frequency gra>ng.Qualita>vely similar observa>ons were made.(E, F) Same as A, B, but for the case of viewing the white fixa>on spot.Again, qualita>vely similar observa>ons were made.See Fig. 6 for quan>ta>ve comparisons.

Figure 6
Figure 6 Stronger microsaccadic suppression for the small white fixaDon spot.(A) We ploRed the threshold values of Fig. 5 together in one graph.Subjects exhibited higher detec>on thresholds for the white fixa>on spot,

Figure 7
Figure 7 Lack of large systemaDc gaze posiDon biases with graDng images.For each subject (each panel), we measured eye posi>on at the >me of probe flash presenta>on (Methods).In each panel, each dot represents a single trial, and the different colors indicate which image was being viewed by the subject (the color legend in C applies to all panels).The inset in each panel shows the mean and standard devia>on values of the corresponding raw data plots of the panel, and E contains the eye posi>ons of the same example subject whose psychometric curves were shown in Fig. 4 above (S05).As can be seen, all subjects fixed their gaze at very similar posi>ons

Figure 8
Figure 8 Microsaccade amplitude distribuDons.In this figure, we ploRed the raw microsaccade amplitude distribu>ons underlying the summary sta>s>cs in Fig. 3.We placed the distribu>on for the white fixa>on spot under each of the low or high spa>al frequency gra>ngs for easier comparison.In all cases, most microsaccades were smaller than 1 deg in amplitude.Thus, there was large overlap across condi>ons.In each distribu>on, the dashed ver>cal line indicates the median, and the solid ver>cal line indicates the mean.