Dependence of the stimulus-driven microsaccade rate signature on visual stimulus polarity

Microsaccades have a steady rate of occurrence during maintained gaze fixation, which gets transiently modulated by abrupt sensory stimuli. Such modulation, characterized by a rapid reduction in microsaccade frequency followed by a stronger rebound phase of high microsaccade rate, is often described as the microsaccadic rate signature, owing to its stereotyped nature. Here we investigated the impacts of stimulus polarity (luminance increments or luminance decrements relative to background luminance) and size on the microsaccadic rate signature. We presented brief visual flashes consisting of large or small white or black stimuli over an otherwise gray image background. Both large and small stimuli caused robust early microsaccadic inhibition, but only small ones caused a subsequent increase in microsaccade frequency above baseline microsaccade rate. Critically, small black stimuli were always associated with stronger modulations in microsaccade rate after stimulus onset than small white stimuli, particularly in the post-inhibition rebound phase of the microsaccadic rate signature. Because small stimuli were also associated with expected direction oscillations to and away from their locations of appearance, these stronger rate modulations in the rebound phase meant higher likelihoods of microsaccades opposite the black flash locations relative to the white flash locations. Our results demonstrate that the microsaccadic rate signature is sensitive to stimulus polarity, and they point to dissociable neural mechanisms underlying early microsaccadic inhibition after stimulus onset and later microsaccadic rate rebound at longer times thereafter. These results also demonstrate early access of oculomotor control circuitry to sensory representations, particularly for momentarily inhibiting saccade generation. New and noteworthy Microsaccades are small saccades that occur during gaze fixation. Microsaccade rate is transiently reduced after sudden stimulus onsets, and then strongly rebounds before returning to baseline. We explored the influence of stimulus polarity (black versus white) on this “rate signature”. We found that small black stimuli cause stronger microsaccadic modulations than white ones, but primarily in the rebound phase. This suggests dissociated neural mechanisms for microsaccadic inhibition and subsequent rebound in the microsaccadic rate signature.


Abstract 30
Microsaccades have a steady rate of occurrence during maintained gaze fixation, 31 which gets transiently modulated by abrupt sensory stimuli. Such modulation, 32 characterized by a rapid reduction in microsaccade frequency followed by a stronger 33 rebound phase of high microsaccade rate, is often described as the microsaccadic rate 34 signature, owing to its stereotyped nature. Here we investigated the impacts of 35 stimulus polarity (luminance increments or luminance decrements relative to 36 background luminance) and size on the microsaccadic rate signature. We presented 37 brief visual flashes consisting of large or small white or black stimuli over an otherwise 38 gray image background. Both large and small stimuli caused robust early 39 microsaccadic inhibition, but only small ones caused a subsequent increase in 40 microsaccade frequency above baseline microsaccade rate. Critically, small black 41 stimuli were always associated with stronger modulations in microsaccade rate after 42 stimulus onset than small white stimuli, particularly in the post-inhibition rebound phase 43 of the microsaccadic rate signature. Because small stimuli were also associated with 44 expected direction oscillations to and away from their locations of appearance, these 45 stronger rate modulations in the rebound phase meant higher likelihoods of 46 microsaccades opposite the black flash locations relative to the white flash locations.

47
Our results demonstrate that the microsaccadic rate signature is sensitive to stimulus 48 polarity, and they point to dissociable neural mechanisms underlying early 49 microsaccadic inhibition after stimulus onset and later microsaccadic rate rebound at 50 longer times thereafter. These results also demonstrate early access of oculomotor 51 control circuitry to sensory representations, particularly for momentarily inhibiting 52 saccade generation.  Reingold and Stampe 1999;2004;. 85

86
The neural mechanisms behind the microsaccadic rate signature, and saccadic 87 inhibition in general, are still being investigated. Neurophysiological perturbation 88 studies in the superior colliculus (SC), frontal eye fields (FEF), and primary visual 89 cortex (V1) have resulted in initial informative steps towards clarifying these 90 mechanisms. First, using a paradigm involving peripheral stimulus onsets, Hafed and 91 colleagues demonstrated that monkeys exhibit the same microsaccadic rate 92 signature as humans (Hafed et al. 2011). These effects persisted even after 93 thousands of trials performed by the same animals in the same tasks, confirming the 94 and colleagues investigated the impacts of luminance and color contrast, as well as 121 auditory stimulation, on microsaccadic inhibition (Rolfs et al. 2008; White and Rolfs 122 2016). Similarly, contrast sensitivity was related to the microsaccadic rate signature 123 in other recent studies (Bonneh et al. 2015; Scholes et al. 2015). In all of these 124 investigations, the general finding was that the strength of both inhibition and 125 subsequent rebound increases with increasing stimulus strength. This suggests that 126 expected sensory neuron properties (e.g. increased neural activity with increased 127 stimulus contrast) must act rapidly on the oculomotor system to mediate inhibition, 128 and potentially also influence subsequent rate rebounds. Here, we add to such 129 existing descriptive studies about the microsaccadic rate signature. We document 130 new evidence that visual stimulus polarity matters. We presented localized as well as 131 diffuse visual flashes that were either white or black, relative to an otherwise gray 132 background. We found that black localized stimuli were particularly effective in 133 modulating the microsaccadic rate signature when compared to white stimuli, 134 especially in the rebound phase, even when the white stimuli had higher contrast 135 relative to the background. cueing effects, to our knowledge, has not been explicitly addressed. This question 146 might be of special interest, since "darks" / "blacks" seem to have temporal and 147 sensitivity advantages over "whites" in visual perception, and there are perceptual 148 asymmetries in processing of low and high luminances (Chubb and  Animal preparation 190 We collected behavioral data from 2 adult, male rhesus macaques (Macaca Mulatta). 191 Monkeys M and A (aged 7 years, and weighing 9-10 kg) were implanted with a 192 scleral coil in one eye to allow measuring eye movements (sampled at 1KHz) using 193 the electromagnetic induction technique (Fuchs and Robinson 1966;Judge et al. 194 1980). The monkeys were also implanted with a head holder to stabilize their head 195 during the experiments, with details on all implant surgeries provided earlier (Chen 196 and 100-1400 ms after flash onset, the fixation spot disappeared, and the monkeys were 213 rewarded for maintaining gaze fixation at the fixation spot throughout the trial. Note 214 that this paradigm is the fixation variant of the paradigm that we used earlier during 215 smooth pursuit eye movements generated by the same monkeys (Buonocore et al. 2012), we aimed to ensure that such stronger influences would be independent of 226 stimulus contrast relative to the background. That is, because stimulus contrast can 227 affect the microsaccadic rate signature (as detailed above in Introduction), we 228 avoided a potential confound of stimulus contrast by having our background gray 229 luminance level being closer to black than to white. Thus, relative to the background 230 luminance, the contrast of black flashes was lower than that of white flashes. Yet, as 231 we report in Results, black flashes often still had significantly stronger impacts on the 232 microsaccadic rate signature, especially with the localized stimuli. depending on the specific signal noise levels in the digitized signals. We manually 240 inspected each trial to correct for false alarms or misses by the automatic algorithms, 241 which were rare. We also marked blinks or noise artifacts for later removal. In scleral 242 eye coil data, blinks are easily discernible due to well-known blink-associated 243 changes in eye position.  We then moved the window in steps of 5 ms to obtain full time courses. The mean 252 microsaccade rate curve across all trials of a given condition was then obtained by 253 averaging the individual trial rate curves, and we obtained the standard error of the 254 mean as an estimate of the dispersion of the across-trial measurements. Since some 255 trials ended before 500 ms after flash onset (see Monkey behavioral tasks above), 256 the across-trial average and standard error estimates that we obtained for any given 257 time bin were restricted to only those individual trials that had data in this time bin; 258 this was a majority of trials anyway. Also, because of the window duration and step 259 size, the time courses were effectively low-pass filtered (smoothed) estimates of 260 microsaccade rate (Bellet et al. 2017). We did not analyze potential higher frequency 261 oscillations in microsaccade rate time courses. These tend to come later after the 262 rebound phase anyway (Tian et al. 2016). We also confirmed that pre-stimulus 263 baseline microsaccade rate in a given monkey was similar in the separate blocks of 264 white and black flashes, therefore allowing us to compare and contrast polarity 265 effects on the rate signature after flash onsets. we used similar procedures to the rate calculations described above. That is, we 286 interval from -100 ms till +500 ms relative to stimulus onset, we compared two given 299 conditions (e.g. localized versus full-screen flashes) by calculating the mean 300 difference in their microsaccade rate. In order to obtain the null experimental 301 distribution, we collected the trials from both conditions into a single set and, while 302 maintaining the initial ratio of numbers of trials in each of the conditions, we randomly 303 permuted their labels; we repeated this procedure 1000 times and recalculated the 304 test statistic (i.e. the difference in rate curves between the two conditions) on each 305 iteration. Second, we selected the bins of the original data whose test statistics were 306 either below the 2.5 th percentile or above the 97.5 th percentile of the permutation 307 distribution (i.e. significant within the 95% confidence level). For adjacent time bins 308 having significant differences (i.e. for clusters of significance), we classified them into 309 negative and positive clusters based on the sign of the difference in rate curves 310 between the two conditions (i.e. clusters had either a negative or positive difference 311 between the two compared microsaccade rate curves). We also repeated this 312 procedure for each random permutation iteration by testing it against all other 999 313 random permutation iterations. This latter step gave us potential clusters of 314 significance (positive or negative) that could arise by chance in the random 315 permutations. Third, for both the observed and permuted data, we calculated the 316 cluster-level summary statistic; this was defined as the sum of all absolute mean 317 differences in any given potentially "significant" cluster. After that, we computed the 318 Monte Carlo p-values of the original data's clusters by assessing the probability of 319 getting clusters with larger or equal cluster-level statistics under the null distribution 320 (i.e. by taking the count of null data clusters with test statistics equal to or larger than 321 the test statistic of any given original data cluster and dividing this count by the 322 number of permutations that we used). A p-value of 0 indicated that none of the 323 clusters of the null distribution had larger or equal cluster-level statistics than the real 324 experimental data. 325

326
When testing either the localized or full-screen flash conditions against the control 327 condition, the test was two-sided (i.e. looking for either positive or negative clusters) 328 to avoid mutual masking of the expected inhibition and rebound effects. In this case, 329 positive and negative clusters (i.e. clusters with positive and negative mean rate 330 differences, respectively) in the experimental data were compared with positive and 331 negative clusters in the permuted data, respectively; the clusters whose p-values 332 exceeded the critical alpha level of 0.025 were considered as significant. All other 333 comparisons were done with a one-tailed test, whereby the clusters were compared 334 in their absolute value regardless of their sign; the critical alpha level was set to 0.05 335 in this case. The same algorithm was applied to the time course analyses of 336 microsaccade amplitudes, except that here, all tests were one-sided. 337 338 When comparing magnitudes of the effects in different phases of the microsaccadic 339 rate signature across conditions, we ran additional non-parametric permutation tests 340 on the differences in minimum microsaccade rates during the inhibition phase or 341 differences in peak microsaccade rates in the rebound phase, as well as in their 342 latencies. To that end, based on the observations across monkeys, we predefined 343 time intervals of interest for both microsaccadic inhibition (i.e. 70-180 ms after 344 stimulus onset) and post-inhibition (i.e. 180-340 ms after stimulus onset) periods. For 345 each experimental condition, we computed the mean microsaccade rate within such 346 a predefined interval and found its extreme value (i.e. the minimum mean inhibition 347 rate or the maximum mean rebound rate) and its latency relative to stimulus onset. 348 Then, we calculated the difference in these values between two given conditions. In 349 order to obtain the null experimental distribution, we did the same procedure as 350 described above: we collected the trials from both conditions into a single data set 351 and randomly permuted their labels, while keeping the initial ratio of the numbers of 352 trials across conditions. We repeated this procedure 1000 times and, on each 353 iteration, we recalculated the test statistics (i.e. the differences between the rate 354 values and their latencies, when applicable). Finally, we computed the Monte Carlo test that we used to investigate the properties of microsaccadic inhibition (Methods) 450 revealed a rate difference between the localized and diffuse conditions during the 451 interval 50-140 ms after stimulus onset (p = 0.017), consistent with a slightly later 452 inhibition for diffuse flashes (Fig. 1C). Monkey A showed no difference in inhibition 453 between localized and diffuse black flashes (Fig. 1F). In both monkeys, the time to   After the microsaccadic inhibition phase, there was a dramatic difference in the 506 rebound phase of the microsaccadic rate signature between localized and diffuse 507 flashes. In Fig. 1B, E, it can be seen that with full-screen flashes, post-inhibition 508 microsaccade rate just returned to the baseline control rate without a clear "rebound" 509 going above baseline. Targeted permutation tests revealed no difference in peak 510 microsaccade rate (relative to control) in a predefined rebound interval (Methods) in 511 monkey M (p = 0.098) and even showed an opposite effect in monkey A (mean peak 512 rate difference = -0.489 microsaccades/s, p = 0.019). This is very different from how 513 microsaccade rate strongly rebounded after the inhibition that was caused by 514 localized flashes (Fig. 1A, D, indicated by red horizontal bars); peak rate was almost 515 3 times the baseline control rate in monkey M mean (peak rate difference = 3.04 516 microsaccades/s, p = 0; permutation test) and almost 2 times the baseline control 517 rate in monkey A (mean peak rate difference = 1.288 microsaccades/s, p = 0; 518 permutation test) (Fig. 1A, D; compare colored to gray curves). 519 520 We also compared the rate curves obtained with diffuse and localized flashes with 521 each other by plotting them together (Fig. 1C, F). Cluster-based permutation tests 522 revealed a significant difference between conditions in the rebound phase, starting at 523 170 ms after stimulus onset, for both monkeys (p = 0). As can be seen from Fig. 1C, 524 F, peak microsaccade rate after the inhibition phase with localized flashes was more 525 than 2 times stronger than peak microsaccade rate after the inhibition phase with 526 diffuse flashes in both monkeys. We quantified these effects by running permutation 527 tests on the peak rate values and their latencies. In monkey M, the mean peak rate 528 difference between localized and diffuse flashes was 2.491 microsaccades/s (p = 0), 529 and the latency difference was -30 ms (p = 0). These values were 1.777 530 microsaccades/s (p = 0) and -45 ms (p = 0.026), respectively, for monkey A. 531 532 With white flashes, similar conclusions could also be reached (Fig. 1G, H). In this 533 case, significant differences between diffuse and localized conditions in the post-534 inhibition period emerged 170-175 ms after stimulus onset. Moreover, once again, 535 with localized flashes, microsaccade rate reached its peak earlier (latency difference 536 = -20 ms, p = 0.003 for monkey M and latency difference = -45 ms, p = 0.014 for 537 monkey A; permutation tests) and rose higher (mean peak rate difference = 1.36 538 microsaccades/s, p = 0 for monkey M and mean peak rate difference = 0.643 539 microsaccades/s, p = 0.004 for monkey A; permutation tests) than with diffuse 540 stimuli. However, note that the peak in microsaccade rate after localized flashes was 541 notably lower than that with black flashes, as we describe in more detail later. The above results, so far, suggest that diffuse visual stimuli are as effective as 558 localized visual stimuli in causing robust microsaccadic inhibition in rhesus macaque 559 monkeys (Fig. 1). However, post-inhibition microsaccade rates are much lower with 560 diffuse stimuli (Fig. 1). Moreover, these effects with diffuse stimuli are largely 561 independent of stimulus polarity (Fig. 2). There were also no clear effects on With localized flashes, we saw above that the microsaccadic rate signature looked 585 more similar to classic literature descriptions. That is, there was a strong post-586 inhibition rebound in microsaccade rate, reaching levels significantly higher than 587 baseline microsaccade-rate during steady-state fixation (colored versus gray curves 588 in Fig. 1A, D, indicated by red horizontal bars on the x-axes). However, comparing 589 the different y-axis scales used in Fig. 1C, F and Fig. 1G, H additionally revealed an 590 influence of stimulus polarity. Unlike in Fig. 2, there was a substantial effect of black 591 flashes in particular on the microsaccadic rate signature with localized stimulus 592 onsets. This effect can be seen clearly in Fig. 3; black localized flashes were 593 particularly effective in modulating the post-inhibition rebound phase of the 594 microsaccadic rate signature, as was also confirmed by cluster-based permutation 595 tests (the red horizontal bars on the x-axes in Fig. 3  white flashes (Fig. 4B). That is, the difference between the congruent and 676 incongruent curves was smaller overall than in Fig. 4A, and the post-inhibition 677 rebound rate was also weaker. In fact, with white flashes, the early difference in 678 inhibition between congruent and incongruent microsaccades was virtually absent 679 ( Fig. 4B). Similarly, the difference in maximal microsaccade rebound rate was now

706 707
With monkey A, all of the effects described above were significantly weaker overall 708 (Fig. 4C, D). Nonetheless, consistent with monkey M, black localized flashes were 709 always associated with stronger trends (Fig. 4C; also see which were still present but muted due to the large baseline directional bias. It is 714 intriguing that even for a monkey like this one, for whom the "cueing effects" with 715 white flashes were weak (Fig. 4D), they were still amplified with black flashes (Fig.  716   4C). 717 718 Therefore, not only were black localized flashes associated with stronger 719 microsaccadic rate modulations in both monkeys (Fig. 3), these stronger effects had 720 a directional component, the largest of which was on enhancing the post-inhibition 721 rebound of incongruent microsaccades (Fig. 4). So-called cueing effects on 722 microsaccades were, thus, stronger with black than white localized flashes (at least 723 in monkey M), even though the contrast of the black flashes relative to background 724 luminance was lower. 725

726
To further explore this incongruent microsaccade effect in more detail, and to confirm 727 that it was still present in monkey A despite the baseline directional bias alluded to 728 above, we plotted the rates of only incongruent microsaccades, now separated 729 based on whether the localized flash was white or black (Fig. 5). In both monkeys, 730 the post-inhibition rate of incongruent microsaccades was significantly higher (and 731 rose earlier) with black localized flashes than with white localized flashes (   (Fig. 7G, H). 844 In fact, direct evaluation of stimulus polarities under the different stimulus size 845 conditions showed that amplitude effects did not strongly depend on stimulus polarity 846 (Fig. 8). If anything, there was a trend for white flashes, small or large, to be 847 associated with stronger overall amplitude modulations as a function of time after 848 stimulus onset (Fig. 8). We investigated the effects of stimulus polarity and size on the microsaccadic rate 940 signature after stimulus onsets. We exploited the fact that even subtle and highly 941 fleeting flashes of only ~8 ms duration are sufficient to cause rapid microsaccadic 942 inhibition after their occurrence followed by a rebound in microsaccade rate. We 943 found that the inhibition was similar for small, localized flashes and large, diffuse 944 ones. However, the subsequent rebound was completely absent with the latter 945 flashes. In terms of stimulus polarity, it had the biggest effects with localized flashes. 946 For these localized flashes, black stimuli caused more substantial changes in the 947 microsaccadic rate signature than white ones, and particularly in the rebound phase 948 after the initial microsaccadic inhibition had ended. 949

950
Our results can inform hypotheses about the neural mechanisms for microsaccadic 951 and saccadic inhibition. In (Hafed and Ignashchenkova 2013), we hypothesized that 952 the rate signature reflects visual neural activity in oculomotor areas like, but not 953 exclusively restricted to, the SC. We specifically hypothesized that the dissociation 954 between rate and direction effects (also present in our own data; e.g. Fig. 6) might 955 reflect spatial read out of SC visual activity for the direction effects (Buonocore et al. 956 2017a) but additional, and potentially different, use of visual activity by the 957 oculomotor system to inhibit saccades for the rate effects (Hafed and 958 Ignashchenkova 2013). Consistent with this, in our current experiments, the similarity 959 that we observed for microsaccadic inhibition between small and large stimuli (Fig. 1)  960 suggests that the early rate effect (i.e. microsaccadic inhibition) is an outcome of 961 early sensory activity that is not necessarily strictly spatial in organization. We