Saccadic suppression by way of retinal-circuit image processing

Saccadic suppression by way of retinal-circuit image processing 1 2 Saad Idrees, Matthias-Philipp Baumann, Felix Franke, Thomas A. Münch, 3 Ziad M. Hafed 4 5 1 Werner Reichardt Centre for Integrative Neuroscience, University of Tübingen, Tübingen, 6 Germany. 7 2 Hertie Institute for Clinical Brain Research, University of Tübingen, Tübingen, Germany. 8 3 Bio Engineering Laboratory, ETH Zürich, Basel, Switzerland. 9 4 Institute for Ophthalmic Research, University of Tübingen, Germany 10 * Correspondence to: ziad.m.hafed@cin.uni-tuebingen.de and thomas.muench@cin.uni11 tuebingen.de 12 † Contributed equally 13 14 Abstract 15


Introduction 28
Saccades are a prominent feature of visual behavior because they allow successive sampling 29 of visual information from the environment. However, from the perspective of the flow of 30 visual information into the brain, these rapid eye movements constitute "disruptive" events, 31 introducing spurious motions that should normally go unnoticed. The question of how and 32 why such perceptual cancelation takes place has intrigued philosophers and scientists for 33 many decades (O'Regan and Noë, 2001). It has been repeatedly found, using a multitude of 34 experiments, that visual sensitivity to brief visual "probes" is strongly impaired around the 35 time of saccades, in a phenomenon known as saccadic suppression (Zuber and Stark, 1966; the suppression with real saccades (Fig. 3E, F). This last point, in particular, suggests that 142 motor-related signals associated with real saccades (Fig. 1) may act to shorten the perceptual 143 interruption resulting from visually-induced saccadic suppression, while maintaining the 144 putatively retinally-determined (Fig. 2) dependence on image statistics. 145 In humans, we observed perceptual suppression also prior to saccade-like texture 146 displacements (Mackay, 1970;Mateeff, Yakimoff and Mitrani, 1976) (Fig. 3). This was again 147 consistently dependent on texture statistics ( Fig. 3B-D). Critically, making real saccades also 148 shortened such pre-saccadic perceptual suppression relative to when saccades were only 149 simulated using texture displacements (Fig. 3E). Even in our retinal data, we found very slight 150 "pre-saccadic" suppression. However, the effect size was much smaller than for flash 151 responses after texture displacements: the strongest "pre-saccadic" retinal effect occurred at -152 67 ms with a median population modulation index of -0.024 (p = 6 x 10 -8 , Wilcoxon signed-153 rank test) compared to -0.55 (p = 3 x 10 -82 ) for "post-saccadic" suppression at 150 ms delay 154 ( Fig. 2E, Fig. S5B). It is therefore likely that this particular phenomenon of perceptual pre-155 saccadic suppression (Fig. 3B-E) arises not in the retina, but from visual (not movement-156 command-related) processing further downstream, perhaps through backwards masking 157 (Macknik and Livingstone, 1998;Breitmeyer, 2007). 158 Our results so far suggest that visual contributions go a long way in explaining properties of 159 perceptual saccadic suppression. We therefore wondered whether such contributions can also 160 explain classic suppression phenomena when uniform, rather than textured, backgrounds are 161 used. One such robust phenomenon has been the selective suppression of low spatial 162 frequencies. In a classic study (Burr, Morrone and Ross, 1994), subjects viewed briefly 163 flashed gratings over a uniform background. Around the time of saccades, visibility of low-164 spatial frequency gratings was suppressed much more strongly than of high-frequency 165 gratings, and this was interpreted as a motor-related influence on magnocellular pathways 166 (Diamond, Ross and Morrone, 2000;Ross et al., 2001). Still, convincing neural mechanisms 167 for this phenomenon remain elusive (Castet, S Jeanjean and Masson, 2001;Castet, Sébastien 168 Jeanjean and Masson, 2001;Ramcharan, Gnadt and Sherman, 2001;Reppas, Usrey and Reid, 169 2002;Kleiser, Seitz and Krekelberg, 2004;Royal et al., 2006;Hass and Horwitz, 2011;Chen 170 and Hafed, 2017). Can our results so far, highlighting the prominence of visual mechanisms 171 underlying saccadic suppression, also account for this classic phenomenon? The answer lies in 172 considering this phenomenon from the perspective of visual input during such experiments: 173 saccades across a uniform background invariably involve moving the image of the video 174 monitor (or other form of display) across the retina. Therefore, the image of any edge 175 discontinuity associated with the display monitor (or with the surrounding cardboard paper 176 around it (Burr, Morrone and Ross, 1994)) will invariably move because of the saccade. This 177 allows us to ask if one can replicate selective suppression of low spatial frequencies (Burr,178 Morrone and Ross, 1994)  The presence of the textured surround allowed us to next isolate the effects of visual flow 192 during these experiments. In separate trials, we asked subjects to fixate, and we presented 193 saccade-like image motion. For example, the virtual monitor moved together with its textured 194 surround from the top left corner towards display center (Fig. 4C), in order to simulate a real 195 saccade from the lower right corner to display center (Fig. 4A). We then briefly presented the 196 same Gabor gratings as in Fig. 4A, B. Relative to fixation position, this experiment was 197 comparable to the situation with real saccades: there was a uniform background against which 198 a brief Gabor grating was flashed. And, indeed, we observed the same selective suppression of 199 low spatial frequencies despite the absence of saccades (Fig. 4D). Moreover, again consistent 200 with our results from Figs. 1-3, the suppression with simulated saccades lasted longer than 201 with real saccades (robust selective suppression in Fig. 4D occurred even 84 ms after 202 simulated saccades; Fig. S7). Similar results were obtained with a uniform black surround 203 around the virtual monitor, as might be the case in typical laboratory settings (Fig. S8). 204 Therefore, visual mechanisms account even for the results of (Burr, Morrone and Ross, 1994) 205 and similar experiments (Chen and Hafed, 2017) using uniform backgrounds, without the 206 need to invoke non-visual (motor-related) mechanisms. 207 Motivated by the differences between coarse and fine textures in Figs. 1-3, we next replaced 208 the coarse texture around the virtual monitor (Fig. 4A, C) with a fine texture, and we repeated 209 the same experiments with simulated saccades (Fig. 4F). In this case, surprisingly, we 210 observed uniform suppression of gratings of all spatial frequencies (Fig. 4F). This led us to 211 make a strong prediction: if saccadic suppression properties do indeed rely on visual 212 processing, then suppression during real saccades should depend mainly on visual context, 213 and one should be able to easily violate the classic phenomenon (namely, the specific 214 suppression of low spatial frequencies (Burr, Morrone and Ross, 1994)). This is exactly what 215 we found (Fig. 4E): for real saccades across the virtual monitor, and with the surrounding 216 visual context being a fine rather than coarse texture, we observed perceptual suppression for 217 all gratings, abolishing suppression selectivity for low spatial frequencies (Burr, Morrone and 218 Ross, 1994). In all cases, the effects were not explained by motor variability across surround 219 texture conditions (Fig. S3E, F). 220 Therefore, either with or without real saccades, perceptual saccadic suppression always 221 occurred, simply as a function of visual flow (Figs. 1,3,4). Such suppression quantitatively 222 depended on scene statistics, both for full-field textures (Figs. 1, 3) in a manner predicted by 223 retinal processing (Fig. 2), and for textures limited to the surround (Fig. 4). Even the 224 selectivity of suppression for low spatial frequencies (Burr, Morrone and Ross, 1994) was 225 determined by visual context (Fig. 4). 226

Conclusion 227
Taken together, our results indicate that visual image processing accounts for a large 228 component of classic perceptual demonstrations of saccadic suppression, and that such image 229 processing occurs as early as in the very first stage of visual processing, the retina. Motor-230 related mechanisms seem to be equally important, though, since they appear to shorten pre-231 and post-saccadic suppression originating from visual mechanisms (Fig. 3), and therefore 232 minimize the duration of saccade-induced disruptions. Furthermore, information contained in 233 the motor command is likely critical for adjustments of spatial receptive fields across saccades 234 in parietal and frontal cortices (Duhamel, Colby and Goldberg, 1992;Sommer and Wurtz, 235 2006  was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint electric fields applied to brain slices', Biosensors and Bioelectronics, 24(7), pp. 2191-2198. was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint Acknowledgements: We thank Andreas Hierlemann for providing the HiDens CMOS MEA 403 system and helping us to establish our high-density MEA recordings. We also thank Roland 404 Diggelmann for help in setting up the pipeline (including providing code) for automatic spike 405 sorting of high-density MEA recordings. This work was supported by funds of the Deutsche 406 show individual subject results, as well as controls for flash visibility (in the absence of 433 saccades) and saccade motor variability. 434 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made repeatedly translated in a saccade-like manner (red or blue "scan paths"), and we presented 438 brief visual flashes at different times relative to "saccades" (similar to Fig. 1 was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made suppression", and the population data underlying panel E. 454 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made  peri-, and post-displacement perceptual suppression occurred for both coarse (B) and fine (C) 460 textures without "real" saccades. (D) As with real saccades (Fig. 1), suppression started earlier 461 and lasted longer with coarse textures (also compare to similar retinal effects in Fig. 2E). 462 Notably, pre-displacement suppression depended on texture statistics, just like with real 463 saccades ( Fig. 1). (E, F) Texture displacements were associated with significantly longer 464 suppression than real saccades. For coarse textures (which were most effective in causing 465 suppression overall), flashes presented before real or simulated saccades were suppressed 466 earlier in the simulated saccade condition than in the real saccade condition; thus, prolonged 467 suppression with texture displacements was not restricted to post-displacement flashes only. 468 Error bars denote s.e.m. across individual subjects' curves. Asterisks denote significant 469 differences between coarse and fine textures (D) or between real and simulated saccades (E, Morrone and Ross, 1994), low spatial frequencies were associated with the strongest 482 suppression relative to baseline. Right: ratio of peri-saccadic to baseline performance (highest 483 spatial frequency not shown because it was at chance performance even in baseline  textures by convolving random binary pixel images with a Gaussian blurring filter. We varied 506 the σ parameter of the Gaussian blurring filter (Methods) to define a so-called "spatial scale" 507 for a given texture (indicated as yellow circles in the examples shown). For each species, we 508 picked the spatial scale to result in dark or bright image blobs that approximated the sizes of 509 either retinal ganglion (coarse) or bipolar (fine) cell receptive fields, and we then set σ to half 510 the "spatial scale" value (Methods). (B) We computed radially-averaged power spectra for 511 textures like in A, normalized to the maximum average power. Low-pass characteristics in all 512 spatial scales were clear, as expected: less than 5% of the total average power was above the 513 spatial frequency corresponding to the specific spatial scale of a given texture (vertical dashed 514 lines). The inset x-axes in the first two spectra (used for human perceptual experiments) show 515 units of cycles per degree (cpd) in addition to cycles per µm on the retina. (C) Histograms  516 showing the distributions of receptive field diameters (Methods) in mouse (left) and pig 517 (right) retinal ganglion cells that we recorded. Since the distributions were generally similar, 518 we used the same spatial scale parameter for the retinal recordings in both species. Human 519 spatial scale parameters were estimated based on human receptive field diameters from the 520 literature (Methods). 521 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made  Identical analyses to Fig. 1D, but now shown separately for each individual subject. Error bars 524 in this case denote s.e.m. across trials. All subjects experienced strong saccadic suppression, 525 going from perfect localization performance to near-chance performance at peak suppression. 526 Moreover, using strict statistical criteria (indicated in the figure and described in detail in 527 Methods), all subjects had significant time clusters during which perception was different 528 between saccadic suppression for saccades generated across coarse or fine textures. (B) Same 529 analyses as in Fig. 3D, but now showing individual subject results when saccades were 530 replaced by saccade-like texture displacements during fixation. All subjects showed prolonged 531 suppression after coarse texture displacements than after fine texture displacements; all 532 subjects also showed earlier and stronger "pre-saccadic" suppression for coarse textures. Note 533 that this "pre-saccadic" effect is purely visual, since the subjects never made saccades in this 534 condition. Also, note that all subjects who participated in this experiment had also participated 535 in the version with real saccades in A. Therefore, whether with or without saccades, 536 perceptual suppression depended on image statistics. (C, D) Comparisons of perceptual 537 suppression between real and "simulated" saccades across coarse (C) and fine (D) textures, as 538 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint in Fig. 3E, F but now separating data from individual subjects. Note how even pre-saccadic 539 suppression was prolonged in simulated relative to real saccades (i.e. started earlier in 540 simulated saccades) in the coarse texture condition, which was most effective in causing 541 suppression overall. Error bars in all cases denote s.e.m. across trials. Asterisks in B denote a 542 significant difference between coarse and fine conditions at the indicated flash time ( χ² tests 543 with Bonferroni corrections; * p<0.005, ** p<0.001, *** p<0.0001). Asterisks in C, D denote 544 significant differences (* p<0.007, ** p<0.0014, *** p<0.00014) between real and simulated 545 saccades, comparing perception of a flash at the indicated time delay after simulated saccades 546 to the corresponding time bin (+/-25 ms) from the real saccade condition. 547 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made   Figs. 1, 3, we asked subjects to maintain fixation, and we 550 eliminated both real saccades (Fig. 1) as well as saccade-like texture displacements (Fig. 3). 551 At a random time during fixation, a luminance pedestal appeared exactly as in the main 552 experiments, but this time, we varied its contrast from background across trials (Methods).

553
We ensured that no microsaccades occurred near the time of flash onset (Methods). 554 Psychometric curves of localization performance indicate that, at the flash contrast used in 555 Figs. 1, 3 (highlighted by the black arrow), subjects could easily detect flashes during fixation 556 and without visual transients associated with saccade-like texture displacements. Importantly, 557 flash visibility was identical for coarse or fine textures at all flash contrasts. Therefore, flash 558 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint visibility alone (or lack thereof) did not explain the main experiments' results (Figs. 1, 3), 559 including differences in perception caused by coarse and fine textures. The strong suppression 560 observed in Figs. 1, 3

was instead likely a function of interaction between visual transients 561 associated with saccades or texture displacements and the flashes. (B) This idea is further 562
supported by the fact that all individual subjects that performed the present control experiment 563 showed consistent results. All of these subjects had also participated in the experiments of 564 Figs. 1, 3 (with the exception of subject ZH who only performed the control experiment). 565 Psychometric curves were fit using the psignifit 4 toolbox (Schütt et al., 2016), and error bars 566 denote 95% confidence intervals. (C) We also checked for potential effects of motor 567 variability on perceptual performance, in order to rule out the possibility that differences in 568 performance between textures (Fig. 1) were due to differences in eye movement kinematics. 569 For the experiments of Fig. 1, we plotted average radial eye velocity (top) and average radial 570 eye position (bottom) across subjects (error bars denote 95% confidence intervals across the 571 individual subjects' curves). There was no effect of background texture on eye movement 572 kinematics. (D) This was also true for each individual subject. In this case, error bars denote 573 s.d. across trials, further supporting that saccade kinematics were not different when saccades 574 were made across coarse or fine textures. (E, F) Same kinematic analyses, but now for the 575 saccades of the experiment of Fig. 4. 576 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made  suppression also did not depend on the velocity properties of texture displacement. We 584 observed very similar suppression of flash-induced neural responses when the texture jumped 585 from the pre-to post-displacement positions in one display frame (right) compared to our 586 standard paradigm, in which the same texture displacement occurred over 100 ms in 6 equal 587 successive steps (left). This means that retinal "saccadic suppression" was mainly governed 588 by stimulus-stimulus interactions, with the first stimulus influencing the responses to the 589 second. (C) Examples of two RGCs for which we presented a flash before the saccade-like 590 texture displacement. The response to the second stimulus (texture displacement) was 591 suppressed because of the first stimulus (flash), supporting the notion that stimulus-stimulus 592 interactions are the main drive for retinal "saccadic suppression". 593 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made  RGCs at different flash times relative to texture displacement onset. Red and blue denote 598 coarse and fine textures, respectively. Black numbers in each panel indicate the numbers of 599 RGCs analyzed for each condition; gray numbers show the logarithm (base 10) of the exact p-600 value (two-tailed Wilcoxon signed-rank test to determine if the population median was shifted 601 away from 0). Asterisks additionally indicate the level of significance. 602 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made  (the baseline curves apply to both time points since they were collected without any saccades).

621
At around 70 ms after saccade onset, perceptual recovery from saccadic suppression emerged, 622 but the selectivity of suppression across different spatial frequencies was still present (there 623 was a main effect of spatial frequency on suppression ratio; χ² =11.4, p=0.022, df=4, Kruskal-624 Wallis test). All other conventions are as in Fig. 4B  was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint A and B, but with a fine texture surrounding the virtual monitor (Fig. 4E, F). (E) Time courses 633 of suppression from Fig. 4A, B with a coarse surround around the virtual monitor. We used 634 similar binning procedures to Fig. 1. Peak suppression was strongest when 0.41 cpd gratings 635 were flashed and progressively weakened for higher spatial frequency gratings (horizontal 636 colored dashed line across panels). (F) With simulated saccade-like virtual monitor and 637 texture displacements, we sampled two grating flash times relative to displacement onset. 638 Recovery at the later time point for each grating spatial frequency was evident. Moreover, 639 selectivity of suppression as a function of grating spatial frequency was evident (horizontal 640 colored dashed line across panels demonstrating the peak suppression for the lowest spatial 641 frequency). The faint curves show time courses from e for comparison. Note how simulated 642 saccades caused longer-lasting suppression than real saccades, exactly as in the experiment of 643 analyses for fine texture surrounds around the virtual monitor. In this case, suppression was 646 the same across all spatial frequencies (horizontal colored dashed lines across panels). (I) For 647 real saccades, and for low spatial frequencies of gratings (i.e. when both coarse and fine 648 surround contexts were associated with strong saccadic suppression), the coarse surround was 649 associated with longer lasting suppression than the fine surround. This is consistent with the 650 results of Fig. 1 when saccades were generated across full-screen textures. (J) This texture-651 dependence was also true with "simulated" saccades (* p<0.05, random permutation test 652 comparing coarse and fine textures at a given grating flash time). Error bars in all panels 653 denote s.e.m. All other conventions are as in Figs. 1, 3, 4. 654 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint and all conventions are similar to Fig. 4 and Fig. S7. 672 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint

Ethical approvals 674
We performed electrophysiological experiments on ex vivo mouse and pig retinae as well as 675 non-invasive perceptual experiments on human subjects. 676 Animal use was in accordance with German and European regulations, and animal 677 experiments were approved by the Regierungspräsidium Tübingen. 678 Human subjects provided written, informed consent, and they were paid 8 Euros per session 679 of 45-60 minutes each. Depending on the experiment, each subject was measured for 2-10 680 sessions (detailed trial and session numbers are provided below). Human experiments were 681 approved by ethics committees at the Medical Faculty of Tübingen University, and they were 682 in accordance with the Declaration of Helsinki. 683

Data availability 684
All data are stored and archived on secure institute computers, and they are available upon 685 reasonable request. 686

Retina electrophysiology laboratory setup 687
We used retinae extracted from PV-Cre x Thy-S-Y mice (B6;129P2-Pvalb tm1(cre)Arbr /J × 688 C57BL/6-tg (ThystopYFPJS), which are functionally wild type (Münch et al., 2009;Farrow et 689 al., 2013;Tikidji-Hamburyan et al., 2015). 19 retinae from 6 male and 12 female mice (3-12 690 months old) were used. We also replicated experiments on pig retinae obtained from domestic 691 pigs after they had been sacrificed during independent studies at the Department of 692 Experimental Surgery in our Medical Faculty. We used 9 pig retinae. 693 We housed mice on a 12/12 h light/dark cycle, and we dark adapted them for 4-16 h before 694 experiments. We then sacrificed them under dim red light, removed the eyecups, and put them 695 in Ringer solution (in mM: 110 NaCl, 2.5 KCl, 1 CaCl2, 1.6 MgCl2, 10 D-glucose, and 22 696 NaHCO3) bubbled with 5% CO2 and 95% O2. We removed retina from the pigment 697 epithelium and sclera while in Ringer solution. 698 Pigs were anesthetized using atropine, azaperone, benzodiazepine (midazolam), and ketamine, 699 and then sacrificed with embutramide (T61). Before embutramide administration, heparin was 700 injected. The pigs were dark-adapted for 15-20 min before sacrifice. Immediately after 701 sacrifice, the eyes were enucleated under dim red light, and the cornea, lens, and vitreous were 702 removed. Eyecups were kept in CO2-independent culture medium (Gibco) and protected from 703 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint light. We transported eyecups to our laboratory and cut pieces from mid-peripheral or 704 peripheral retinae. 705 We recorded retina ganglion cell (RGC) activity using either low or high-density multi-706 electrode arrays (MEAs). The low-density setup consisted of a perforated 60-electrode MEA 707 (60pMEA200/30ir-Ti-gt, Multichannel Systems, Reutlingen, Germany) having a square grid 708 arrangement and 200 μm inter-electrode distance. We mounted an isolated retina on a 709 nitrocellulose filter (Millipore) with a central 2x2 mm hole. The mounted retina was placed 710 with the RGC side down into the recording chamber, and good electrode contact was achieved 711 by negative pressure through the MEA perforation. We superfused the tissue with Ringer with the aid of a micromanipulator. We recorded extracellular activity at 20 kHz using FPGA 727 signal processing hardware and custom data acquisition software. 728 In total, we performed 32 recordings, 20 from mouse and 12 from pig retina. 15 of the 32 729 recordings were done using low-density MEAs. Once a basic experimental protocol was 730 established, we shifted to HiDens CMOS MEA providing much higher throughput. 12 731 experiments were done using this setup. We upgraded to the MaxOne MEA for even higher 732 throughput and did our final 5 recordings using this setup. 733 We presented light stimuli to the retinal piece that was placed on the MEA using a DLP 734 projector running at 60 Hz (Acer K11 for low-density MEA experiments and Lightcrafter 735 4500 for high-density MEA experiments). Acer K11 had a resolution of 800x600 pixels 736 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ;https://doi.org/10.1101/562595 doi: bioRxiv preprint covering 3 x 2.25 mm on the retinal surface. Lightcrafter 4500 had a resolution of 1280x800 737 pixels, extending 3.072 x 1.92 mm on the retinal surface. We focused images onto the 738 photoreceptors using a condenser. The light path contained a shutter and two motorized filter 739 wheels with a set of neutral density (ND) filters (Thorlabs NE10B-A to NE50B-A), having 740 optical densities from 1 (ND1) to 5 (ND5). Light intensity was adjusted to be in the mesopic 741 range. 742 We measured the spectral intensity profile (in µW cm -2 nm -1 ) of our light stimuli with a 743 calibrated USB2000+ spectrophotometer (Ocean Optics) and converted the physical intensity 744 into a biological equivalent of photoisomerizations per rod photoreceptor per second 745 (R*rod -1 s -1 ), as described before (Tikidji-Hamburyan et al., 2015). Light intensities of the 746 projector output covered a range of 3 log units (i.e. 1,000-fold difference between black and 747 white pixels, over an 8-bit range). We linearized the projector output, and we used only

Human psychophysics laboratory setup 756
We used a similar laboratory setup to (Hafed, 2013;Bellet, Chen and Hafed, 2017;Grujic et 757 al., 2018). Briefly, subjects sat in a dark room 57 cm in front of a CRT monitor (85 Hz refresh 758 rate; 41 pixels/deg resolution) spanning 34.1 x 25.6 deg (horizontal x vertical). Head fixation 759 was achieved with a custom head, forehead, and chin rest (Hafed, 2013), and we tracked eye 760 movements (of the left eye) at 1 kHz using a video-based eye tracker (EyeLink 1000, SR 761 Research Ltd, Canada). Gray and texture backgrounds (e.g. Figs. 1, 3, 4) were always 762 presented at an average luminance of 22.15 cd/m 2 , and the monitor was linearized (8-bit 763 resolution) such that equal luminance increments and decrements were possible around this 764 average for textures and gratings. 765 Human Experiment 1 (Fig. 1) was performed by 8 subjects (2 female) who were 21-25 years 766 old. All subjects were naïve to the purposes of the experiment, except for subject MB (an 767 author). For Human Experiment 2, the "simulated saccade" version of Human Experiment 1 768 (Fig. 3), 6 of the same subjects participated. A control experiment for testing visibility of 769 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint flashes without saccades and without saccade-like texture displacements (e.g. Fig. S3A, B) 770 was performed by 6 of the same subjects plus one non-naïve subject, ZH (another author). 771 Human Experiment 3 tested suppression selectivity for low spatial frequencies (Fig. 4). Six 772 subjects (3 females, 23-25 years old) participated, and only subject MB was non-naïve. Three 773 subjects had also participated in Human Experiments 1 and 2 and their control version. A 774 control version of Human Experiment 3 was also performed with black surrounds (Fig. S8). 775 This control experiment was performed by the same subjects that participated in Human 776

Experiment 3. 777
Across all experiments, we ensured that the same subjects performed real and "simulated" 778 saccade versions of a given paradigm so that we could make meaningful comparisons between 779 these two eye movement conditions. 780

Coarse and fine textures used in retina electrophysiology and human psychophysics 781 experiments 782
We created coarse and fine textures ( The parameter σ of the kernel influenced the amount of blurring. This resulted in textures 787 having effectively low-pass spectral content (Fig. S1B) with a cutoff frequency (fc) depending 788 on σ. As we describe below, we picked cutoff frequencies for coarse and fine textures that 789 resulted in dark and bright image blobs approximating the receptive field sizes of RGCs (for 790 coarse textures) and retinal bipolar cells (for fine textures). In other words, for a given species, 791 coarse textures matched the resolution of RGCs, and fine textures matched the resolution of 792 one processing stage earlier, the retinal bipolar cells. 793 For the ex-vivo experiments with mouse and pig retinae, we assumed receptive field 794 diameters for RGCs of at least 150 µm ( Fig. S1C; the parameter σ of the Gaussian blurring 795 filter would be half that value), and diameters for bipolar cells of 25 µm (see (Zhang et al., 796 2012)). For human psychophysics experiments, we estimated, from the literature (Dacey and 797 Petersen, 1992), the sizes of human parasol RGC receptive fields at eccentricities >6 deg from 798 the fovea (our flash eccentricities were 7 deg) to be around 200 µm. This translated into a 799 cutoff frequency of ~0.68 cycles per deg (cpd) (Fig. S1B). Bipolar cell receptive field sizes at 800 this eccentricity were estimated to be 10 µm (corresponding to a cutoff frequency of ~13.7 801 cpd), based on sizes of human midget RGC receptive fields in the fovea (Dacey and Petersen, 802 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint 1992). When calculating the textures, the actual value of the parameter σ (in pixel-803 dimensions) always incorporated the specific experimental magnification factor between the 804 stimulation screen and the retinal projection of the image. 805 Calculating power spectra for coarse and fine textures confirmed that cutoff frequencies for a 806 given species were consistent with our aimed designs described above (Fig. S1B). 807 For both retinal and perceptual experiments, we normalized pixel intensities in the textures to 808 have uniform variations in luminance around a given mean. In the retinal experiments, we 809 used pixel intensities (from our 8-bit resolution scale) ranging from 0 to 60 around a mean of 810 30, or ranging from 30 to 90 around a mean of 60 (see flash conditions below for when each 811 paradigm was used). For the human experiments, textures had a mean luminance of 22.15 812 cd/m 2 with undulations in luminance in the texture within the range of 7.5-35.5 cd/m 2 . 813 Because each texture, particularly when coarse, could have patterns of dark and bright blobs 814 that human subjects can remember or interpret as potential shapes/objects/figures, we varied 815 the displayed texture images from trial to trial. This was also necessary to avoid afterimages. 816 We generated sets of 20 coarse and 20 fine textures, which we randomly interleaved across 817 trials. Moreover, the textures themselves were designed to be larger than the viewable display 818 area, allowing us to jitter the displayed sub-rectangle of each texture (within the viewable area 819 of the display) from trial to trial (we jittered the displayed sub-rectangle within a range of 0.6 820 x 0.6 deg in steps of 0.024 deg). This way, even fine patterns at foveal fixation locations could 821 not be memorized by the subjects. 822

Retina electrophysiology experimental procedures 823
To simulate saccades in our ex vivo retina electrophysiology experiments, we displaced the 824 texture across retina in 6 display frames (100 ms at 60 Hz refresh rate). For easier readability, 825 we sometimes refer to these saccade-like texture displacements as "saccades". The textures 826 were displaced in each frame by a constant distance along a linear trajectory. While each 827 "saccade" lasted 100 ms, displacement direction was varied randomly for each "saccade" 828 (uniformly distributed across all possible directions) and "saccade" amplitude could range 829 from 310 μm to 930 μm (corresponding to a velocity range of 3,100-9,300 μm/s on the retinal 830 surface). In visual degrees, this displacement corresponds to 10-30 deg, well in the range of 831 saccade amplitudes observed in mice (Sakatani and Isa, 2007) and velocities of 100-300 deg/s 832 and 10-32 deg/s for mouse and pig eyeball sizes, respectively. 833 Each "trial" consisted of 39 successive sequences that each combined a "saccade" with a 834 probe flash, as follows: there was first a "pre-saccade" fixation of 2 seconds, then a 100 ms 835 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint "saccade", followed by "post-saccade" fixation. At a certain time from "saccade" onset (delay 836 d, range: -177 ms to 2,100 ms), we presented a probe flash. In most cases, the probe flash had 837 a duration of 1 frame (~16 ms), we used 2 frames (~33 ms) in a subset of experiments (mouse: 838 161/688 cells analyzed for "saccadic suppression"; pig: 112/228 cells). Results were pooled 839 across these paradigms as they were indistinguishable. For sequences containing no probe 840 flash, the next "saccade" happened 4 s after the previous one. The probe flash was a full-841 screen positive ("bright") or negative ("dark") stimulus transient. In different experiments, 842 only a subset of possible delays was used within a given set of trials, depending on total 843 recording time for a given retina (see below). 844 Bright or dark probe flashes could happen in two different ways across our experiments. The 845 results were indistinguishable between the two ways, so we pooled results across them. 846 Briefly, in one manipulation, the probe flash was a homogeneous bright (pixel intensity of 60 847 in our 8-bit projectors) or dark (pixel intensity of 0) full-screen rectangle replacing the 848 background texture (in these experiments, the textures themselves had intensities ranging 849 from 0 to 60 pixel intensity; see coarse and fine texture descriptions above). This way, the 850 flash contrast from the underlying background luminance was variable (e.g. a bright flash on a 851 bright portion of a texture had lower contrast from the underlying texture than the same flash 852 over a dark portion of the texture). In the second manipulation, the bright and dark flashes 853 were simply luminance increments or decrements (by pixel values of 30 on our 8-bit 854 projectors) over the existing textures (like in our human perceptual experiments). This way, 855 local contrast relationships in the background textures were maintained. In these experiments, 856 the textures themselves had a range of 30-90 pixel intensities and a mean pixel value of 60 (on 857 our 8-bit projectors). 332/688 cells we analyzed for "saccadic suppression" experienced such 858 probe flashes whereas the rest (356 cells) experienced the homogenous probe flash. For pig 859 retina recordings, we always used the homogenous framework. However, in the subset of pig 860 experiments where the 2-frame probe flash was employed (112/228 RGCs), we used a high-861 contrast probe flash such that a bright flash would be achieved by first going completely dark 862 in the first frame followed by the bright flash in the next frame and vice versa for a dark flash. 863 Again, all data was pooled across these different paradigms because their outcome was 864

indistinguishable. 865
The number of trials required during a physiology experiment depended on the number of 866 conditions that we ran on a specific day. For example, testing 7 different flash delays required 867 15 trials (7 with bright probe flashes, 7 with dark probe flashes, and 1 without probes). In a 868 given experiment, we always interleaved all conditions; i.e. in any one of the 15 necessary 869 geometry, this meant a slightly larger displacement (of 12.4 deg) when compared to the 936 saccade sizes in Human Experiment 1. However, we chose this translation because it resulted 937 in a sufficiently fast average speed of the displacement (average speed in the real saccades of 938 Human Experiment 1 was 160 deg/s). This choice is not problematic because our retinal 939 experiments revealed that visual mechanisms related to saccadic suppression were not 940 sensitive to parameters of individual motion patterns (Fig. S4B). 941 In this experiment, the texture displacement happened in a diagonal direction to simulate the 942 directions of saccadic displacements of Human Experiment 1 (and also to dissociate the 943 direction of motion flow from the locations of the flashes, again as in Human Experiment 1). 944 For example, the texture could move globally down-right, as might be expected (in terms of 945 image motion) if subjects made upward-leftward saccades in Human Experiment 1. Also, 946 flash times were chosen relative to the onset of texture displacement (from among the 947 following values : -35, -24, 24, 47, 84, 108, 141, 200, 259, 494 ms). 948 All subjects participated in 10 session each in this experiment. 949 We also performed a control experiment, in which there was neither a real saccade (Human 950 Experiment 1) nor a texture displacement (Human Experiment 2), but otherwise identical to 951 these 2 experiments. Subjects simply fixated display center, and we presented (after 1,200 to 952 2,400 ms from trial onset) a luminance pedestal exactly as in Human Experiments 1 and 2. To 953 obtain full psychometric curves, we varied the luminance increment from among 6 values 954 (e.g. Fig. S3A, B). Subjects performed two sessions each of this experiment (600 trials per 955 session). 956 In Human Experiment 3 (Fig. 4), the flashes of Human Experiments 1 and 2 were replaced by 957 vertical Gabor gratings having one of five different spatial frequencies (0.41, 0.85, 1.71, 3.42, 958 4.56, or 6.8 cpd). Spatial phase was randomized from trial to trial, and the parameter of the 959 Gaussian envelope was 0.49 deg. Also, a virtual monitor of 20 deg diameter was present at 960 display center at the time of grating flashes. The virtual monitor had a uniform gray 961 luminance equal to the average of the textures used in Human Experiments 1 and 2. 962 Surrounding the virtual monitor, a coarse or fine texture could be visible. 963 In one block of trials, subjects generated saccades towards display center using the same 964 procedures as in Human Experiment 1. Grating flash times were similar to Human 965 Experiment 1, and the subjects performed 6 sessions each (576 trials per session). 966 In another block of trials, subjects maintained fixation at display center. In one third of the 967 trials, the virtual monitor and surrounding texture did not move. These trials provided us with 968 . CC-BY-NC 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted February 27, 2019. ; https://doi.org/10.1101/562595 doi: bioRxiv preprint