A transient signal in foveal superior colliculus neurons for jumpstarting peripheral saccadic orienting

27 28 The superior colliculus (SC) both senses the environment and orients gaze within it. While 29 the SC’s sensory and motor bursts appear qualitaDvely similar to each other, populaDon 30 acDvity structure in the two processing regimes is very different, necessitaDng a hitherto 31 unexplored rapid representaDonal transformaDon, occurring on the scale of only tens of 32 milliseconds. Here, using male rhesus macaque monkeys, we first show that when a planned 33 saccade is released with a go signal, peripheral SC neurons represenDng the saccade target 34 locaDon exhibit a transient, short-latency pause right before their motor bursts eventually 35 erupt. This pause starts within ~50 ms from the go signal, and it is sDmulus-dependent. It is 36 also absent in the primary visual cortex, and is significantly weaker in purely visual SC 37 neurons than in saccade-related ones. Foveal SC neurons, on the other hand, burst, and 38 their bursts lead the peripheral neurons’ pauses by ~10 milliseconds. Remarkably, during 39 immediate visually-guided saccade tasks, requiring a transformaDon from visual to motor 40 peripheral bursts in <50-100 ms, the transient foveal SC bursts sDll occur, resulDng in 41 simultaneous short-latency bursDng at two disparate SC loci: one foveal; and one eccentric 42 and responding to the visual appearance of the saccade target. Our results suggest that in 43 classic saccade tasks used to invesDgate visual, motor, and cogniDve processes in primate 44 brains, a transient foveal SC signal may jumpstart peripheral saccadic orienDng by facilitaDng 45 a necessary rapid representaDonal transformaDon needed for SC saccade motor bursts to 46 ensue. 47 48 49 50 Significance 51 52 Studies of eye movement control ohen involve behavioral paradigms involving sensing, 53 deliberaDng, and ulDmately releasing an instructed rapid eye movement (saccade). While 54 the sensing, deliberaDon, and movement aspects of saccades have been well studied in the 55 past, the releasing phase is less understood. Using classic saccade behavioral paradigms, we 56 discovered that foveal superior colliculus neurons exhibit transient acDvity bursts during 57 instructed saccade releasing, which likely enables rapid representaDonal transformaDons 58 from a visual to a motor regime in the collicular neurons driving the eye movements. 59 60 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 3 Introduc3on 61 62 AcDve vision entails a conDnual cycling between sensing, deliberaDon, and acDon (Findlay, 63 1982; Wolpert et al., 1995; Findlay and Walker, 1999; Buonocore and Hafed, 2023). In the 64 oculomotor system, the neuronal mechanisms underlying the different subcomponents of 65 acDve vision have classically been studied using behavioral paradigms that aiempt to 66 dissociate these subcomponents from one another as much as possible. For example, in the 67 delayed visually-guided saccade paradigm (Hikosaka and Wurtz, 1983; Munoz and Wurtz, 68 1995b), a visual sDmulus first appears, but the iniDally fixated spot remains visible. This 69 prevents a reflexive saccade to the appearing sDmulus and allows invesDgaDng visual 70 sensory processing mechanisms independent of an overt motor output. Then, aher some 71 delay, during which cogniDve processes related to deliberaDon and motor planning may be 72 invesDgated, the iniDally fixated spot disappears, allowing a saccade towards the eccentric 73 sDmulus to be subsequently triggered. In this case, motor-related processes associated with 74 saccade generaDon can be studied under a constant, steady-state visual appearance of the 75 environment. 76 77 Among the many insights gleaned from the classic delayed saccade paradigm, it was recently 78 recognized that this paradigm addiDonally highlights a fundamental problem in sensory-79 driven motor behavior, namely the need to transform neuronal representaDons from a 80 sensory regime to a motor regime, ohen in the very same neurons and within very short 81 Dme intervals (Jagadisan and Gandhi, 2022; Baumann et al., 2023; Bourrelly et al., 2023; 82 Heusser et al., 2023). For example, populaDon acDvity subspaces in the superior colliculus 83 (SC) are different from each other at sDmulus onset and saccade generaDon (Jagadisan and 84 Gandhi, 2022; Baumann et al., 2023). And, at the individual neuron level, some SC neurons 85 can prefer a parDcular visual image feature in the sDmulus onset phase of trials, but a 86 different image feature in the saccade generaDon phase (Baumann et al., 2023). This rapid 87 alteraDon in operaDng regimes of idenDcal neurons likely necessitates a switch-like 88 mechanism at some point during delayed saccade task trials, and this is what we 89 invesDgated here. 90 91 We specifically hypothesized that the go, or release, signal for saccades may be associated 92 with neuronal dynamics that were not previously characterized in sufficient detail. Indeed, 93 past studies generally focused on neuronal acDvity for (peripheral) neurons represenDng the 94 eccentric locaDons of the saccade targets. These studies uncovered well known sensory, 95 cogniDve, and motor processes associated with the neurons driving the eye movements. 96 However, it was not known what specifically happens to foveal representaDons in the SC at 97 the Dme of the go signal; since acDve vision ulDmately starts and ends with the fovea, 98 understanding foveal SC representaDons when releasing saccades is important. 99 100 To address this gap, we studied the acDvity of peripheral SC and primary visual cortex (V1) 101 neurons during the delayed saccade paradigm, with a parDcular focus on neuronal acDvity 102 around the Dme of the go signal. We found that releasing an instructed eye movement in 103 this paradigm is associated with a short-latency transient pause in SC, but not V1, acDvity 104 right before SC saccade-related motor bursts ensue. Importantly, this pause is preceded by 105 foveal SC bursts ~10 ms earlier, and the foveal burst properDes suggest that they could be a 106 generalized trigger signal rather than a sDmulus-dependent phenomenon. Remarkably, 107 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 4 when we switched our paradigm to instead employ immediate visually-guided saccades, SC 108 foveal bursts sDll happened, and thus temporally coincided with peripheral visual responses 109 to target onsets in other SC neurons. 110 111 Our results reveal novel neuronal dynamics in the foveal SC at the Dme of the go signal in 112 classic eye movement paradigms that have been used for several decades to study 113 percepDon, cogniDon, and acDon. These results also moDvate invesDgaDng mechanisms of 114 foveal-to-peripheral, and vice-versa, neuronal modulaDons in the SC and other brain 115 structures, especially when these modulaDons may not always be trivially explained by 116 classic lateral inhibitory mechanisms (Munoz and Istvan, 1998; Trappenberg et al., 2001; Isa 117 and Hall, 2009; Marino et al., 2011; Satel et al., 2011; Phongphanphanee et al., 2014). 118 119 120 121

122 123 Research animals and ethical approvals 124 This study involved a re-analysis of data collected previously for other publicaDons 125 invesDgaDng other research quesDons. In all cases, we analyzed data collected from male 126 rhesus macaque monkeys. For analyzing peripheral SC and V1 acDvity in the delayed visually-127 guided saccade task (see below for the details of the different behavioral tasks), we used the 128 same database as the one in (Baumann et al., 2023). For analyzing foveal SC acDvity in the 129 delayed visually-guided saccade task, we used the control condiDons of (Zhang et al., 2025). 130 And, for analyzing peripheral and foveal SC acDvity in the immediate visually-guided saccade 131 task, we employed the control condiDons of (Zhang et al., 2026). 132 133 In all cases, all experiments were approved by ethics commiiees at the regional 134 governmental offices of Tübingen. 135 136 137 Behavioral tasks 138 139 Saccades-to-gra-ngs task for peripheral SC and V1 neurons. For peripheral SC and V1 acDvity 140 in the delayed saccade condiDon, the behavioral task was the “saccades-to-graDngs” 141 paradigm (Baumann et al., 2023). This task was a slight modificaDon of the classic delayed 142 visually-guided saccade paradigm. Briefly, the monkeys first saw a white fixaDon spot. Aher 143 they fixated it by a few hundred milliseconds, a visual target appeared eccentrically, 144 centered on the recorded neurons’ response field (RF) locaDons. The target consisted of a 145 disc of 3 deg radius, the inside of which had a sine wave graDng. We always placed a white 146 spot (like the fixaDon spot) at the center of the disc, surrounded by a small gray disc (to 147 avoid visibility loss of the spot due to the background sine wave graDng). This placement 148 allowed the saccades to be accurately made towards the disc’s center, which was criDcal for 149 the previous study (but it did not influence the scienDfic quesDons of the present one). In 150 the spaDal frequency version of the task, the sine wave graDng was verDcal and had 100% 151 contrast, but it could have different spaDal frequencies across different trials. In the contrast 152 version of the task, we fixed the spaDal frequency to 1 cycle/deg (1 cpd) and instead varied 153 the contrast of the verDcal graDng from trial to trial. And, finally, in the orientaDon version of 154 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 5 the task, both the spaDal frequency (1 cpd) and contrast (100%) were fixed, and the 155 orientaDon of the graDng was varied from trial to trial. Aher maintaining fixaDon for another 156 few hundred milliseconds aher target onset, the fixaDon spot was removed, and this was the 157 cue for the monkeys to generate a visually-guided saccade towards the eccentric target. 158 Thus, the fixaDon spot removal was the go signal in this task, as is the case in classic 159 instanDaDons of it. Trials from each version of the task were collected in separate blocks. 160 161 Saccades-to-gra-ngs task for foveal SC neurons. For foveal SC acDvity in the delayed saccade 162 condiDon, the behavioral task was similar to the one above (Zhang et al., 2025). Specifically, 163 aher fixaDng the fixaDon spot, an eccentric target appeared. Again, it consisted of a disc of 3 164 deg radius, the inside of which had a verDcal sine wave graDng of 100% contrast and either 1 165 cpd (low) or 4 cpd (high) spaDal frequency. There was no spot placed at the center of the 166 graDng in this case. The target appeared at 8 deg eccentricity either to the right or leh of 167 fixaDon (except in two sessions in which it was placed at an oblique posiDon of similar 168 eccentricity; ~10 deg). Aher maintaining fixaDon for a few hundred milliseconds aher target 169 onset, the fixaDon spot was removed, instrucDng the generaDon of the saccade towards the 170 eccentric graDng. Aher fixaDng the graDng for another 500 ms, this graDng was removed, 171 and the monkeys were rewarded. A short inter-trial interval (with a blank gray screen) then 172 ensued. 173 174 Immediate visually-guided saccade task for perihpheral and foveal SC neurons. For 175 peripheral and foveal SC acDvity in the immediate saccade condiDon, we employed a classic 176 reflexive visually-guided saccade paradigm; the control condiDon of (Zhang et al., 2026). 177 Specifically, the monkeys fixated a white fixaDon spot. Aher a few hundred milliseconds, the 178 fixaDon spot was removed and a simultaneous saccade target (white disc of 0.51 deg radius) 179 appeared either at the peripheral neurons’ RF posiDons or at an eccentricity >3.5 deg from 180 the fovea for the foveal neurons. 181 182 RF mapping tasks. The delayed saccade tasks above allowed us to invesDgate neuronal 183 dynamics at the Dme of transiDoning from a visual regime (aher sDmulus onset and waiDng 184 for saccade instrucDon) to a motor regime (aher the go signal). The reflexive saccade task, 185 on the other hand, allowed us to explore what happens when the transiDon from visual to 186 motor regimes needed to happen much more urgently, as quickly as possible aher 187 peripheral target onset. In all cases, we also mapped neuronal RF’s in the same sessions, and 188 this was done in order to idenDfy the RF’s posiDons relaDve to the eccentric saccade target. 189 Our RF mapping tasks (Chen et al., 2015; Chen and Hafed, 2017; Chen et al., 2018; Chen et 190 al., 2019) involved fixaDng a fixaDon spot, like in all tasks above. Aher a few hundred 191 milliseconds of fixaDon, a white spot appeared at some locaDon on the display, and it 192 typically remained on for 300-500 ms, before being turned off again. Across trials, we 193 changed the posiDon of the appearing spot, in order to construct the RF maps of the 194 neurons. 195 196 197 Data analysis and sta6s6cal tests 198 199 All saccadic eye movements were previously detected for the purposes of the earlier studies 200 (Baumann et al., 2023; Zhang et al., 2025; Zhang et al., 2026). Similarly, all neuronal 201 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 6 preprocessing (such as spike sorDng) was performed earlier. Our subsequent analyses were 202 geared for the specific scienDfic purposes of the present study. 203 204 To characterize visual RF’s of neurons (e.g. Figs. 1A, C, 5A, C in Results), we averaged firing 205 rate in an interval 50-150 ms aher sDmulus onset, and we ploied it as a funcDon of the 206 horizontal and verDcal posiDon of the appearing sDmulus (including interpolaDon in 207 between sampled posiDons). For some analyses, we also characterized the sDmulus-offset 208 responses of the recorded neurons from the RF mapping data (parDcularly for the foveal 209 neurons in the delayed saccade paradigm). In this case, we simply measured average firing 210 rate 50-150 ms aher the offset of the target in the same RF mapping task. As we explain in 211 Results, this analysis was aimed at checking whether the removal of the fixaDon spot in our 212 main saccade paradigms caused offset responses in the foveal neurons. Thus, for measuring 213 offset responses from the RF mapping task, we picked sDmulus locaDons in the RF mapping 214 data that were consistent with the eccentricity of the fixaDon spot (in the main tasks) from 215 the foveal neuron’s RF hotspot locaDon. In other words, because foveal SC neurons are 216 strongly lateralized (Chen et al., 2019), fixaDon spot removal in the main tasks involved an 217 offset of a sDmulus that was at some non-zero distance, r, from the hotspot of the recorded 218 foveal neurons. Therefore, from the RF mapping data, we defined a ring around the hotspot 219 locaDon of each foveal SC neuron with radius of r deg (+/- 0.25 deg), where r was the 220 eccentricity of the RF hotspot from the fixaDon spot. Then, we picked all sampled RF 221 mapping trials in this ring of locaDons that were also in the contralateral hemifield relaDve to 222 the recorded SC side (we picked contralateral locaDons to maximize the likelihood of 223 observing an offset response if it did exist), and we measured offset responses from these 224 trials in parDcular. We then compared these offset responses to those associated with the go 225 signal in the main saccade tasks (which are described in Results). 226 227 For the main saccade tasks, we ploied firing rates as a funcDon of Dme from target onset, 228 saccade onset, and go signal onset (fixaDon spot removal). The first two kinds of plots are 229 similar to what was done previously, by us (Baumann et al., 2023; Zhang et al., 2025; Zhang 230 et al., 2026) and others; we included them here for providing the context associated with 231 the third kind of plot, which was of interest for us in the current study. As we show in 232 Results, this third plot gave either transient pauses or transient bursts in acDvity (shortly 233 aher the go signal), depending on the neuron locaDon and behavioral task. To characterize 234 these transient responses, we measured average firing rate in the interval 50-150 ms aher 235 go signal onset. For a reference, we also measured firing rate in the final 100 ms before go 236 signal onset. This gave a baseline to which we compared go-signal responses when assessing 237 whether there was a burst or pause occurring aher the go signal. Such a comparison was 238 made either by plotng the two measures directly against each other for each neuron, or by 239 obtaining a neuronal modulaDon index. When assessing transient pauses (see Results), we 240 defined the modulaDon index as the firing rate aher the go signal (50-100 ms) minus the 241 firing rate before the go signal (final 100 ms), divided by the sum of the two firing rates. 242 NegaDve modulaDon indices meant reducDons in firing rate aher the go signal. 243 244 We also someDmes classified trial condiDons (e.g. different spaDal frequencies in the spaDal 245 frequency task) as a funcDon of sustained acDvity level they gave in a given neuron at the 246 end of the delay period (right before go signal onset). Specifically, we looked for the 247 condiDon that gave the highest sustained acDvity and called it the most preferred sustained 248 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 7 feature. Similarly, we looked for the image condiDon that gave the lowest sustained acDvity, 249 and we called it the least preferred sustained feature. This allowed us to check whether 250 transient pauses aher the go signal that we characterize in Results depended on the starDng 251 firing rate that was present at the Dme of the go signal. 252 253 Similarly, we divided SC neurons according to their funcDonal type (such as visual or visual-254 motor neurons). We used classificaDons that we used previously (Baumann et al., 2023). 255 Thus, we did not reclassify the neurons here. 256 257 To compare transient pauses and bursts directly (e.g. Fig. 7 in Results), we normalized 258 individual neuron firing rates and then combined neurons by averaging their normalized 259 firing rate curves. The normalizaDon was achieved by measuring, for each neuron, the 260 average firing rate in the final 100 ms before go signal onset, and then dividing the enDre 261 firing rate curve by this value. Thus, the normalized firing rate was 1 at the go signal, and 262 bursts/pauses were higher/lower than 1. SomeDmes, we also kept firing rates unnormalized, 263 but we baseline-subtracted them based on the final 100 ms before the go signal. Thus, in 264 this case, pauses were associated with negaDve baseline-subtracted firing rates. 265 266 Similarly, to assess the relaDve Dming between transient pauses and transient bursts in the 267 delayed saccade paradigm (e.g. Fig. 7 in Results), we relied on individual spike Dmes. Across 268 all trials and all neurons (either foveal or peripheral), we binned spike Dmes around the go 269 signal into 2-ms non-overlapping Dme bins. Then, we ploied a histogram of all spike Dmes. 270 For the peripheral neurons, they paused aher the go signal (see Results). Thus, starDng from 271 10 ms aher the go signal, we searched for the first Dme point at which the histogram of 272 spike Dmes had 3 consecuDve drops in spike likelihood aher the go signal. This was 273 considered the populaDon latency of the pause. For the foveal neurons, which burst instead 274 (see Results), we searched for the first Dme point at which the histogram of spike Dmes had 275 3 consecuDve increases in spike likelihood aher the go signal. This was considered the 276 populaDon latency of the burst. 277 278 For foveal neurons, we also someDmes looked at task context. Specifically, we compared 279 foveal bursts that we got from the go signal (see Results) to potenDal bursts that might be 280 associated with the offset of a foveal target sDmulaDng the recorded neurons’ RF’s. We 281 obtained the laier from the ends of the trials in the delayed saccade paradigm used to study 282 the foveal neurons. As menDoned above, aher the monkeys fixated the graDng with a 283 saccade, the graDng covered the fovea, and thus sDmulated the recorded foveal neurons’ 284 RF’s (Zhang et al., 2025). When the graDng disappeared, signaling the end of the trial, the 285 monkeys were without any task-instrucDon and could look wherever they wanted. We 286 measured firing rates aher such graDng disappearance, to look for potenDal sDmulus-offset 287 responses in the neurons. To ensure that there were no influences of potenDal saccades, we 288 only included cases in which there were no saccades aher graDng disappearance for at least 289 150 ms. This way, we had an offset of a visual sDmulus in the recorded RF’s, but there was no 290 instructed saccade context (as was the case with the fixaDon spot offset of the go signal in 291 the main task). 292 293 For the immediate saccade task, we used similar analyses to the ones above. For populaDon 294 firing rates, we normalized the firing rate of each neuron by the peak visual or foveal burst 295 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 8 response (see Results). Specifically, we first subtracted the average acDvity of each neuron in 296 the final 100 ms before sDmulus onset from all firing rates at all Dmes in the trials (Zhang et 297 al., 2026). Then, we divided the firing rate curves by the peak value occurring 50-100 ms 298 aher sDmulus onset (Zhang et al., 2026). Then, we averaged across all neurons. This allowed 299 us to focus on the Dming dynamics between foveal and peripheral bursts in this task, as well 300 as on the relaDonships of the bursts to saccadic reacDon Dmes. 301 302 For such saccadic reacDon Dme relaDonships, for each neuron, we split the trials into ones 303 with saccadic reacDon Dmes faster than the median value of the session and ones with 304 longer reacDon Dmes than the median value of the session. Then, we evaluated the 305 peripheral and foveal bursts for both groups of trials separately across neurons. 306 307 StaDsDcally, we always showed standard error of the mean ranges in all plots. For 308 comparisons between condiDons, we used signrank tests, rank sum tests, or t-tests. 309 SomeDmes, we split foveal SC neurons as ones with or without a foveal burst at the go 310 signal. This classificaDon was made staDsDcally. If the neuron’s firing rate 50-150 ms aher the 311 go signal was significantly higher than in the final 100 ms before the go signal with an t-test, 312 then the neuron was classified as having a foveal burst; otherwise, it was not. This allowed 313 us to compare foveal bursts to sDmulus-offset responses from RF mapping tasks. 314 315 316 317

318 319 We first invesDgated the dynamics of switching between fixaDon and saccade generaDon in 320 the standard delayed saccade paradigm. In this paradigm, monkeys fixate an iniDal fixaDon 321 spot, and an eccentric saccade target appears. The monkeys withhold saccadic orienDng 322 towards the eccentric target unDl a go signal arrives, which comes in the form of fixaDon 323 spot disappearance. Classic SC recordings of peripheral neurons in this task demonstrate 324 visual responses to target onset followed by motor bursts at saccade onset, ohen in the very 325 same neurons (Mohler and Wurtz, 1976; Munoz and Wurtz, 1995b; Massot et al., 2019; 326 Jagadisan and Gandhi, 2022; Baumann et al., 2023). Here, we were interested in the 327 neuronal dynamics that take place in between these two phases, and parDcularly at the Dme 328 of the go signal. Moreover, from the same task, we also recorded from foveal SC neurons, 329 which presumably should not burst for either target or saccade onset because of the 330 disparity between their RF locaDons and the saccade target locaDon (Munoz and Wurtz, 331 1993a, b; Dorris and Munoz, 1995; Dorris et al., 1997; Everling et al., 1998; Munoz and 332 Istvan, 1998; Everling et al., 1999; Munoz et al., 2000; Hafed and Krauzlis, 2008; White et al., 333 2013). 334 335 In what follows, we first start by characterizing peripheral SC and V1 neuronal acDvity in the 336 delayed saccade paradigm, demonstraDng a transient reset event in the SC, but not V1, 337 leading up to saccade-related motor bursts. We then document how foveal SC neurons 338 behave at the Dme of the go signal in the same task, revealing a transient bursDng signal 339 that precedes the peripheral SC reset event. Finally, we demonstrate that our observaDons 340 about foveal bursts hold even in reflexive saccade tasks (without an enforced delay), 341 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 9 resulDng in the occurrence of two simultaneous short-latency neuronal acDvity bursts (aher 342 saccade target onset) in two disparate SC loci, one foveal and one peripheral. 343 344 345 Peripheral superior colliculus neurons exhibit transient ac6vity pauses before 346 instructed-saccade motor bursts emerge 347 348 We analyzed peripheral SC neuron acDvity from a “saccades-to-graDngs” task that we 349 recently designed (Baumann et al., 2023) (Methods). Figure 1A, C shows the visual RF’s of 350 two example SC neurons from this task. The saccade target (schemaDzed by the white circle) 351 appeared over the neurons’ RF’s, which expectedly elicited visual responses by the neurons 352 (Fig. 1B, leh and Fig. D, leh); these visual responses were different for different image 353 appearances of the saccade target, consistent with the presence of feature tuning properDes 354 in SC neurons (Chen and Hafed, 2018; Chen et al., 2018; Bogadhi and Hafed, 2023; Hafed et 355 al., 2023). Since the two neurons were visual-motor neurons, at the Dme of saccade onset 356 later on in the same trials (Fig. 1B, right and Fig. 1D, right), both neurons also emiied a 357 saccade-related motor burst, which could again vary in strength depending on the visual 358 appearance of the saccade target (Zhang et al., 2022; Baumann et al., 2023; Hafed, 2025). 359 Remarkably, when we measured the acDvity of the two neurons around the Dme of the go 360 signal (Fig. 1B, middle and Fig. 1D, middle), both neurons exhibited a short-latency transient 361 pause in their acDvity (downward gray arrows). This pause occurred with a latency of ~50 ms 362 from the go signal occurrence, and it was apparently also all-or-none. That is, if the 363 sustained acDvity of the neuron was high for one image feature (e.g. 0.5 cpd for the neuron 364 of Fig. 1A, B or 11 cpd for the neuron of Fig. 1C, D), then the reducDon in firing rate aher the 365 go signal was more marked than if the sustained acDvity was low (e.g. for 11 cpd in the 366 neuron of Fig. 1A, B and 2 cpd in the neuron of Fig. 1C, D); this resulted in more similar firing 367 rates aher the go signal than before it, regardless of the iniDal firing rate that was influenced 368 by the image appearance of the saccade target. Immediately aher each pause, the strong 369 motor burst of each neuron erupted, as classically expected from saccade-related SC 370 neurons. 371 372 373 374 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 10 375 376 Figure 1 Transient ac/vity pauses before the occurrence of saccade-related motor bursts in the SC. (A) Visual 377 response field of an example SC neuron. The white circle indicates the saccade-target loca<on and size; monkeys 378 were only allowed to generate a saccade to the target upon a go signal (consis<ng of the removal of the fixa<on 379 spot; Methods) (Baumann et al., 2023). (B) The ac<vity of the same example neuron aligned to s<mulus onset in 380 the main saccade task (leJ) , go signal onset (middle), and actual saccade onset (right). Two visual target 381 appearances of the saccade target are shown (each target had a small central spot to aid in saccade accuracy) 382 (Baumann et al., 2023) . Expected differences in visual (leJ) or motor (right) burst proper<es as a func<on of 383 image appearance were observed (Chen and Hafed, 2018; Chen et al., 2018; Zhang et al., 2022; Baumann et al., 384 2023; Hafed et al., 2023; Hafed, 2025). Cri<cally, shortly aJer the go signal (middle), the neuron paused its 385 ac<vity before the strong saccade-related burst could erupt. (C, D) Similar observa<ons from a second example 386 neuron, showing a <me -locked ac<vity pause aJer go signal before the motor bursts finally occurred (middle 387 panel in D). Numbers of trial repe<<ons per panel can be seen from the individual trial spike rasters shown. 388 389 390 391 Across the populaDon of SC neurons, and for all image features that we tested, we measured 392 neuronal acDvity in the final 100 ms before the go signal. Then, we subtracted the acDvity of 393 each neuron around the go signal from this baseline measurement, to obtain a baseline-394 subtracted firing rate (Methods). When we aligned the baseline-subtracted firing rate of all 395 neurons to the go signal, we observed a short-latency reducDon in acDvity right before the 396 elevaDon associated with saccade-related motor bursts (Fig. 2A; SC data; trials with all image 397 features were combined together). This was consistent with what we observed from the two 398 example neurons of Fig. 1. We also measured the raw firing rate of each neuron, both in 399 baseline (final 100 ms before the go signal) as well as 50-150 ms aher the go signal, and we 400 compared the two measurements (Fig. 2B; trials with all image features were combined 401 together). Once again, across the populaDon, there was a robust reducDon in firing rate aher 402 the go signal (p= 7.7693 x 10-23 for the spaDal frequency task; p=1.4893 x 10-19 for the 403 contrast task; p=6.4765 x 10-18 for the orientaDon task; signrank test). 404 405 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 11 Thus, we observed a short-latency pause in SC acDvity when releasing an instructed saccade 406 in the classic delayed saccade paradigm. We think that this pause is funcDonally useful in the 407 SC, especially because it represents a perfect and Dmely-suitable opportunity to transform 408 SC representaDons from being in a visual regime to being in a motor regime (Jagadisan and 409 Gandhi, 2022; Baumann et al., 2023; Heusser et al., 2023). 410 411 412 413 414 415 416 417 Figure 2 Ubiquity of SC ac/vity pauses across task contexts, and a lack of a correlate for them in the primary 418 visual cortex (V1) in similar behavioral tasks. (A) Popula<on firing rates aligned on the go signal from the same 419 task as in Fig. 1. Here, we baseline-subtracted the ac<vity of each neuron based on firing rate in the final 100 ms 420 before go signal onset (Methods; numbers of neurons can be inferred from B, C). Across all three task variants 421 (Methods), SC ac<vity pauses were present to the same extent (downward arrow), but there were no such 422 pauses in V1. The later increase in the SC curves is the saccade-related burst, and the even later V1 increase is 423 visual reafference due to eye movement. (B) For each SC neuron, we measured ac<vity in the final 100 ms before 424 go signal onset and ploWed it on the x-axis; on the y-axis; we measured average ac<vity 50-150 ms aJer the go 425 signal. For each task, there was a reduc<on aJer the go signal. (C) There was no such reduc<on in V1. Numbers 426 of neurons are indicated in the figure, and sta<s<cal results are men<oned in the text. 427 428 429 430 431 We also explored the SC pause properDes further, by linking them to the sustained acDvity 432 associated with the image impinging on the visual RF’s of our recorded neurons. Different 433 image features caused different levels of sustained SC acDvity (Chen and Hafed, 2018; Chen 434 et al., 2018) (Figs. 1B, D, 3A). If the SC pause is all-or-none, then image features with higher 435 sustained acDvity should be associated with a bigger firing rate drop than image features 436 with lower sustained acDvity; this would allow converging all SC firing rates to the same 437 (low) level regardless of the image condiDon driving the peripheral neurons’ acDvity. We 438 confirmed this by finding, for each neuron, the image condiDon that gave rise to the highest 439 firing rate in the final 100 ms before the go signal, and we called this the most preferred 440 sustained feature. We also found the image condiDon associated with the lowest sustained 441 firing rate, and called it the least preferred sustained feature. Across neurons, the magnitude 442 of the drop in firing rate between the baseline (final 100 ms before the go signal) and post-443 go (50-150 ms aher the go signal) intervals was larger for the most preferred than least 444 preferred sustained feature (Fig. 3B for the spaDal frequency task); there was a significant 445 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 12 difference between the two image features (p= 3.9233 x 10-29; t-test). Results from the 446 contrast and orientaDon tasks were virtually idenDcal, as also seen from Fig. 2A, B. 447 448 Thus, we idenDfied a systemaDc short-latency pause in SC acDvity at the Dme of releasing a 449 task-instructed saccade in the classic delayed saccade paradigm. 450 451 452 453 454 455 Figure 3 Dependence of peripheral SC ac/vity pauses on the viewed s/mulus appearance. (A) Ac<vity of a third 456 example SC neuron at the <me of the go signal for three image appearances of the saccade target. During the 457 delay period leading up to the go signal, the neuron had ongoing ac<vity that was different for the different 458 images, reflec<ng a feature tuning property of the neuron (Hafed et al., 2023). As a result, the amount of ac<vity 459 reduc<on aJer the go signal was s<mulus-dependent: the pause was stronger for the condi<on in which the 460 delay-period ac<vity of the neuron was the highest, resul<ng in the same low ac<vity level regardless of ini<al 461 state. (B) Across all neurons, we measured each neuron’s ac<vity in the final 100 ms before the go signal, and 462 picked the saccade-target image appearance that gave either the highest or lowest such ac<vity. Then, we 463 measured the pause strength (difference between ac<vity aJer the go signal and ac<vity before; Methods) for 464 the most or least preferred image feature. The pause was stronger for the most preferred image feature, 465 consistent with A. Note that only data from the spa<al frequency task is shown, but the other tasks revealed 466 similar results (also seen in Fig. 2). 467 468 469 470 Peripheral superior colliculus ac6vity pauses are absent in the primary visual 471 cortex 472 473 We also wondered whether the transient pause that we saw in the SC was a general 474 property of visually-responsive neurons in other brain areas. Therefore, we collected a 475 smaller number of V1 neurons from the same task (Methods). There was no transient pause 476 aher the go signal, as can be seen from the V1 data in Fig. 2A. Given the fact that a large part 477 of visual responses in the SC derives from V1 (Wilson and Toyne, 1970; Lund, 1972; Schiller 478 et al., 1974; May, 2006; Cerkevich et al., 2014; Zheng et al., 2024), this observaDon confirms 479 that SC visual responses are funcDonally transformed and not merely inherited from the 480 cortex (Wurtz and Mohler, 1976; Hafed et al., 2023; Troienberg et al., 2026). 481 482 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 13 Also note that in the V1 neurons, there was a later elevaDon of neuronal acDvity well aher 483 the go signal (Fig. 2A), almost 100 ms later than the elevaDon seen for the SC neurons. This 484 later elevaDon reflects visual reafferent responses in V1 as a result of eyeball rotaDons, 485 whereas the earlier SC increase in firing rate (aher the pause) represents the SC saccade-486 related motor bursts. Figure 2C also shows that in V1, there was no staDsDcally significant 487 difference between acDvity right before and right aher the go signal (p=0.6165 for the 488 spaDal frequency task; p=0.4209 for the contrast task; p=0.5876 for the orientaDon task; 489 signrank test), again confirming the absence of a transient pause in V1 neurons. 490 491 Therefore, our results so far indicate that we observed a transient pause in SC acDvity at the 492 Dme of the go signal in a classic delayed saccade paradigm. This pause was not a general 493 property of other brain areas that might be recruited by the same paradigm, and that might 494 modulate SC acDvity, such as V1. 495 496 497 Peripheral superior colliculus ac6vity pauses are stronger in saccade-related 498 neurons 499 500 As menDoned above, the presence of an SC pause at the go signal might be funcDonally 501 parDcularly useful, because it is only in the SC, and not in V1, that the very same neurons 502 would be engaged in both visual processing and saccade-related modulaDons (Jagadisan and 503 Gandhi, 2022; Baumann et al., 2023; Hafed et al., 2023; Heusser et al., 2023; Hafed, 2025). If 504 this is true, then one might expect that this pause should be more relevant for saccade-505 related SC neurons than for purely visual ones; this is because it is the saccade-related 506 neurons that would exhibit both sensory and motor responses requiring a representaDonal 507 transformaDon on a rapid Dme scale. To check this, we classified our SC neurons as being 508 either purely visual or motor-related. The visual neurons included visual-burst and visual-509 delay neurons, and the motor-related neurons included visual-motor and purely motor 510 neurons, as defined previously (Baumann et al., 2023; Bogadhi and Hafed, 2023; Hafed et al., 511 2023). 512 513 We found that there were stronger pauses for the motor-related SC neurons than for the 514 visual ones. For example, in Fig. 4A, B, we reploied the data of Fig. 2B but aher first 515 classifying the neurons into the two groups menDoned above. Across all tasks, there were 516 bigger differences between the acDvity before and aher the go signal in the motor-related 517 neurons (Fig. 4B) than in the visual neurons (Fig. 4A). To quanDfy this further, we calculated a 518 modulaDon index for each neuron (acDvity aher the go signal minus before the go signal, 519 divided by the sum; Methods). This index was negaDve for acDvity pauses and zero for no 520 pauses. In all tasks, the median value of the modulaDon index was more negaDve in the 521 motor-related neurons than in the visual ones (Fig. 4C, D, E), and the differences between 522 distribuDons of modulaDon indices across the two classes of neurons were staDsDcally 523 significant, but only marginally so for the orientaDon task (the results of the staDsDcal tests 524 are included in the legend of Fig. 4). Thus, the transient pause that we observed was 525 stronger for motor-related neurons. 526 527 Of course, the presence of sustained acDvity at the Dme of the go signal is a prerequisite for 528 successfully seeing a pause, if one is present at all. Thus, it might be suggested that the 529 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 14 differences in pause strengths that we saw in Fig. 4 between purely visual and motor-related 530 neurons could reflect a lack of sustained acDvity in the visual neurons. However, this was not 531 the case. For example, inspecDon of the raw firing rates on the x-axis of Fig. 4A reveals that 532 the purely visual SC neurons had similar sustained acDvity levels, in general, to the visual-533 motor and motor neurons (x-axis in Fig. 4B). This is expected because visual neurons 534 (especially visual-delay ones) are known to conDnuously represent the presence of a visual 535 sDmulus over their RF’s. Thus, the results in Fig. 4 are not explained by the purely visual 536 neurons having no (or less) sustained acDvity at the Dme of the go signal than the motor-537 related neurons. 538 539 540 541 542 543 544 Figure 4 Specificity of SC ac/vity pauses for neurons related to the motor genera/on of saccades. (A) Similar 545 analysis to Fig. 2B, but only purely visual SC neurons. Visual neurons were visual-burst and visual-delay neurons 546 (Hafed et al., 2023). There was minimal ac<vity reduc<on at the go signal in these neurons despite the presence 547 of delay-period ac<vity (p=9.8480 x 10 -4 for the spa<al frequency task, p= 3.9107 x 10-5 for the contrast task, 548 p=3.0530 x 10-7 for the orienta<on task; signrank test); also see C-E. (B) Neurons that needed to emit a saccade-549 related motor burst aJer the go signal underwent clearer ac<vity reduc<ons (p=4.5489 x 10-21 for the spa<al 550 frequency task, p=8.0457 x 10-16 for the contrast task, p=1.0135 x 10-12 for the orienta<on task; signrank test). (C, 551 D, E) Neuronal modula<on indices (Methods) for the results in A, B across the different task contexts. Visual-552 motor and motor neurons had generally stronger reduc<ons in their ac<vity aJer the go signal than visual and 553 visual-delay neurons (p=0.000532 for the spa<al frequency task, p=0.004 for the contrast task, p=0.0966 for the 554 orienta<on task; Wilcoxon rank sum test comparing the distribu<on of visual-motor/motor modula<on indices 555 to the distribu<on of visual/visual-delay modula<on indices). The median values of the shown distribu<ons were: 556 -0.1069 and -0.0453 for visual-motor/motor neurons and visual/visual-delay neurons, respec<vely in C; -0.0981 557 and -0.036 in D; -0.0862 and -0.0537 in E. 558 559 560 561 562 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 15 563 Foveal superior colliculus neurons exhibit transient bursts instead of pauses 564 565 Having established the presence of a short-latency acDvity pause in SC neurons right before 566 saccade-related bursts erupt, we next asked what happens in foveal SC representaDons 567 during the delayed saccade paradigm. Figure 5A, C shows the visual RF’s of two example 568 foveal SC neurons. In both cases, the RF’s were contained within a reDnotopic eccentricity of 569 2 deg (dashed circle) (Chen et al., 2019), and the saccade target was at an eccentricity of 8-570 10 deg (Methods). At sDmulus onset (Fig. 5B, D, leh) and saccade onset (Fig. 5B, D, right), the 571 two neurons behaved as expected from foveal SC neurons when the saccade target is placed 572 outside of their RF’s: at sDmulus onset, there was either no modulaDon or a reducDon in 573 acDvity; and, at saccade onset, there was a strong pause when the peripheral neurons were 574 emitng their motor bursts. Both of these observaDons were documented before in the SC 575 literature (Munoz and Wurtz, 1993b, a; Hafed and Krauzlis, 2008; Zhang et al., 2025). 576 Remarkably, right aher the go signal, both foveal SC neurons showed a very strong, short-577 latency acDvity burst, which almost doubled or tripled the firing rate relaDve to its level at 578 the Dme of the go signal (Fig. 5B, D, middle; upward gray arrows). This foveal burst occurred 579 right before the saccade-related pause that was seen in the foveal neurons’ acDvity at the 580 Dme of saccade onset. Thus, unlike the peripheral SC neurons (Figs. 1-4), these two example 581 foveal SC neurons showed acDvity bursts, rather than pauses, when releasing instructed 582 saccades (Fig. 5). 583 584 585 586 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 16 587 588 Figure 5 Transient ac/vity bursts, rather than pauses, in the foveal representa/on of the SC prior to saccade 589 triggering. (A, B) Similar to the formadng of Fig. 1 but now for an example foveal SC neuron. The RF was 590 constrained in the central 2 deg of visual angle (A), and the saccade target was at a farther eccentricity (Zhang 591 et al., 2025). In B, the neuron behaved as expected at the <mes of s<mulus (leJ) and saccade (right) onset. AJer 592 the go signal (middle), the neuron showed a strong transient burst in ac<vity, and at a qualita<vely similar <me 593 to the peripheral pauses of Figs. 1-4. Note that this neuron is the same example neuron of Fig. 1 in (Zhang et al., 594 2025), but here we addi<onally showed the ac<vity at the <me of the go signal. (C, D) Similar observa<ons for a 595 second example foveal SC neuron. There was again a transient burst right aJer the go signal. All other 596 conven<ons are like Fig. 1. Error bars denote SEM across trial repe<<ons. Panel A was adapted from (Zhang et 597 al., 2025). 598 599 600 601 602 Just like with the peripheral neurons, we also had mulDple visual features of the eccentric 603 saccade target in the version of our delayed saccade paradigm that we used when recording 604 our foveal SC neurons (Methods). This allowed us to invesDgate whether there was any 605 sDmulus dependence of the foveal acDvity bursts. In Fig. 6A, the foveal burst of a third 606 example foveal SC neuron is shown, but this Dme for the two visual appearances of the 607 eccentric saccade target. In both cases, the strength of the foveal burst was the same. This 608 was a general property across our populaDon of foveal SC neurons. Specifically, we 609 measured average firing rate 50-100 ms from go signal occurrence, and we did so for either 610 the low or high spaDal frequency saccade target. There was no difference in foveal burst 611 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 17 strength (p=0.8509; signrank test across the populaDon; n=33). Therefore, unlike in the 612 peripheral neurons, the foveal SC burst was not sDmulus-dependent. As we will show later 613 with our Dming analyses, this burst might thus be a general trigger signal to jumpstart 614 peripheral saccadic orienDng, regardless of the peripheral target’s image appearance. 615 616 617 618 619 620 621 622 Figure 6 Lack of s/mulus or direc/on dependence of the foveal bursts. (A) The foveal burst of an example SC 623 neuron for two visual appearances of the peripheral saccade target. The foveal burst was similar in strength for 624 the two s<muli. Error bars denote SEM across trials. (B) We also checked whether the foveal burst depended on 625 congruence between the recorded foveal SC hemifield and the saccade direc<on. There were no or moderate 626 dependencies on saccade direc<on (signrank test; n=33). 627 628 629 630 631 632 We also checked whether the foveal SC burst depended on saccade direcDon. Specifically, 633 foveal SC neurons are lateralized (Chen et al., 2019), just like peripheral ones are. Thus, if a 634 foveal neuron was recorded from the right SC, then it represented the leh foveal space. In 635 this case, a lehward eccentric saccade target would elicit motor bursts in the peripheral 636 neurons of the same SC as the recorded foveal neuron, whereas a rightward saccade would 637 elicit motor bursts in the opposite SC. As can be seen from Fig. 6B, there was no clear 638 dependence of foveal burst strength on saccade direcDon; for the high spaDal frequency 639 saccade target, there was a staDsDcally significant result, but the magnitude of the 640 difference between hemifield direcDons was not qualitaDvely different from that seen with 641 the low spaDal frequency target (for which there was no staDsDcally significant difference 642 between saccade direcDons). Thus, foveal SC bursts (Figs. 5, 6A) also did not show a 643 systemaDc dependence on saccade direcDon relaDve to the recorded neurons’ RF hemifields 644 (Fig. 6B). 645 646 647 648 649 650 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 18 Foveal superior colliculus ac6vity bursts lead peripheral pauses 651 652 Our results so far indicate that releasing an instructed saccade in the classic delayed saccade 653 paradigm is associated with a short-latency (foveal) burst in one part of the SC topographic 654 map and a (more peripheral) pause in another (Fig. 7A). To understand the potenDal links 655 between these two types of neuronal responses further, we summarized SC populaDon 656 acDvity in the two groups of neurons, aher normalizing the acDvity of each neuron to its 657 acDvity in the final 100 ms before go signal onset. At the Dme of saccade onset (Fig. 7B), the 658 peripheral neurons emiied strong saccade-related motor bursts, and the foveal neurons 659 decreased their acDvity, consistent with the example neurons of Fig. 5 and with the prior 660 literature (Munoz and Wurtz, 1993b; Hafed and Krauzlis, 2008; Zhang et al., 2025). However, 661 this relaDonship was completely reversed in the go signal epoch (Fig. 7C), with foveal 662 neurons now exhibiDng an approximately three-fold increase in their acDvity immediately 663 aher the go signal and right before saccade triggering; this is consistent with the foveal burst 664 strengths seen in the example neurons of Fig. 5. The peripheral neurons, on the other hand, 665 paused. InteresDngly, plotng the peripheral and foveal neurons together in Fig. 7C revealed 666 that the foveal burst actually led the peripheral pause by some Dme. To beier quanDfy this 667 Dming relaDonship, we binned spike Dmes in all neurons into 2 ms non-overlapping Dme 668 bins, and we esDmated spike likelihood in each Dme bin aher go signal onset. The onset of 669 the foveal burst (defined as the first Dme point to have at least three successive increases in 670 spike likelihood; Methods) led the onset of the peripheral pause (defined as the first Dme 671 point to have at least three successive decreases in spike likelihood; Methods) by 10 ms. 672 673 674 675 676 677 678 Figure 7 Temporal sequencing of foveal bursts and ac/vity pauses in the SC prior to saccade triggering. (A) 679 Schema<c of the SC topographic map (Hafed and Chen, 2016; Chen et al., 2019; Hafed et al., 2021), with an 680 example indica<on of the rela<ve posi<ons of neurons that were recruited in our popula<ons of recordings from 681 either the experiments of Figs. 1 -4 or those of Figs. 5 -6. (B) At the <me of saccade onset, peripheral neurons 682 exhibited a motor burst, and foveal neurons reduced their ac<vity, as expected (Mohler and Wurtz, 1976; Munoz 683 and Wurtz, 1993b, 1995b; Hafed and Krauzlis, 2008, 2012; Zhang et al., 2025). Error bars denote SEM across 684 neurons, and the normaliza<on factor for each neuron was the ac<vity level in the final 100 ms before go signal 685 onset; see C. (C) At the <me of the go signal, foveal neurons exhibited strong bursts, which started slightly earlier 686 than the peripheral SC pauses. (D) We quan<fied this <ming differences by having 2 ms bins in which we 687 measured spike likelihood across all neurons. Then, we iden<fied the onset of the burst or the pause as the first 688 <me point at which the spike likelihood changed in the same direc<on (increase for bursts and decrease for 689 pauses) for at least three more upcoming <me bins (Methods). The peripheral pauses lagged the foveal bursts 690 by 10 ms. 691 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 19 Thus, foveal SC bursts at the Dme of releasing instructed saccades in the classic delayed 692 saccade paradigm occur systemaDcally earlier than peripheral SC pauses, and may thus aid 693 in triggering the peripheral pauses. 694 695 696 Foveal superior ac6vity bursts reflect instructed saccade context 697 698 A parsimonious explanaDon for our results so far could relate to intrinsic RF properDes of SC 699 neurons (Hafed et al., 2023), combined with potenDal long-range lateral inhibiDon 700 mechanisms (Munoz and Istvan, 1998; Trappenberg et al., 2001; Isa and Hall, 2009; Marino 701 et al., 2011; Satel et al., 2011; Phongphanphanee et al., 2014). Specifically, it could be 702 possible that the removal of the fixaDon spot in the delayed saccade paradigm may act as an 703 offset sDmulus to foveal SC neurons. Since SC neurons are expected to someDmes have 704 offset responses (i.e. acDvity bursts in response to an offset of a visual sDmulus that was 705 previously present in their RF’s) (Humphrey, 1968; Schiller and Koerner, 1971; Cynader and 706 Berman, 1972; Goldberg and Wurtz, 1972b; Hafed et al., 2023), it is conceivable that the 707 fixaDon spot removal in our saccade paradigm could trigger an offset response in foveal SC 708 neurons. Aher such a response, lateral inhibiDon in the SC might, in turn, cause peripheral 709 neuron acDvity pauses (as a result of the foveal bursts). To invesDgate this potenDal cascade 710 of events, we first compared our foveal bursts to real offset responses as measured by our 711 RF mapping tasks. We then explored other situaDons with offsets of visual sDmuli in our 712 paradigms, but without an explicit task instrucDon to release a saccade. And, finally, we 713 explored cases in which we created a compeDDon between putaDve lateral inhibiDon and 714 the occurrence of peripheral SC acDvity bursts (rather than pauses), to understand whether 715 the peripheral pauses are due to lateral inhibiDon from the foveal bursts or not. We describe 716 the results of these three successive invesDgaDons next. 717 718 To explore whether foveal bursts at the go signal reflect offset responses to the removal of 719 the fixaDon spot, we measured offset responses in our foveal SC neurons from our RF 720 mapping tasks (Methods). During RF mapping, we presented a small white spot (similar to 721 the fixaDon spot) at different locaDons near the fixaDon spot that the monkeys were looking 722 at. Aher a few hundred milliseconds, the white spot was removed, allowing us to measure 723 offset responses. Figure 8A, B shows the offset responses of the two example foveal neurons 724 of Fig. 5. Because foveal SC RF’s are strongly lateralized (Chen et al., 2019) (Fig. 5A, C), 725 fixaDon spot removal in our main task was equivalent to an offset response for a sDmulus 726 that was visible at a non-zero distance, r, from the RF hotspot eccentricity (assuming, rightly, 727 that the monkeys properly centered gaze on the fixaDon spot, on average); here, r would be 728 the distance between the RF hotspot and the central preferred reDnal locus of fixaDon. 729 Therefore, from the RF mapping data, we picked a ring of sampled sDmulus locaDons at a 730 distance r from the RF hotspot (+/- 0.25 deg; Methods). We took all of these sDmulus 731 locaDons in the hemifield of the RF hotspot (to maximize the likelihood of seeing offset 732 responses), and we ploied the RF offset responses in Fig. 8A, B. As can be seen, at the same 733 eccentricity from the RF hotspot as the fixaDon spot in the main task, the two example 734 neurons did not emit substanDal offset responses at all. For comparison, the insets in Fig. 8A, 735 B replicate the foveal bursts of Fig. 5 from the same neurons, showing that the neuronal 736 responses at the go signal in the main task were very different from their responses for the 737 offset of small white spots near their RF hotspots. Thus, for these two example neurons, 738 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 20 foveal bursts were not trivially explained by offset responses to the removal of the fixaDon 739 spot. 740 741 742 743 744 745 746 Figure 8 Dissocia/on between foveal bursts aTer the go signal and s/mulus-offset responses in the foveal SC 747 neurons. (A) For the example neuron of Fig. 5A, B, we measured the response of the neuron to the offset of a 748 small spot during the RF mapping task (Methods); we took care to define a band of eccentrici<es and direc<ons 749 in which the s<mulus loca<on was in a similar por<on of the RF as the disappearing fixa<on spot in the main task 750 (Methods). The neuron did not exhibit a substan<al offset response; for comparison, the inset shows the foveal 751 burst of the same neuron from Fig. 5B. Thus, the foveal burst was not an offset response to the disappearance 752 of the fixa<on spot in our saccade task. Error bars denote SEM across trial repe<<ons. (B) Similar observa<ons 753 for the second example neuron of Fig. 5C, D. (C) For each neuron that exhibited a foveal burst in the main saccade 754 task (Methods), we ploWed the foveal burst strength against the offset response of the neuron from the RF 755 mapping task (Methods). There was a moderate rela<onship between offset responses and foveal bursts (r = 756 0.6026, p = 0.0005 for the low spa<al frequency; r = 0.5924, p = 0.0007 for the high spa<al frequency). Each color 757 shows the foveal burst for a one spa<al frequency of the saccade target. (D) For neurons with no foveal burst in 758 the main saccade task (Methods), they could s<ll exhibit offset responses in the RF mapping task (r = 0.2662, p 759 = 0.1166 for the low spa<al frequency ; r = 0.2890, p = 0.0874 for the high spa<al frequency) . Thus, C and D 760 combined suggest that foveal bursts were not always explained by offset responses due to the disappearance of 761 the fixa<on spot in the delayed saccade paradigm. (E) In the main saccade task, aJer a successful saccade to the 762 gra<ng, the gra<ng eventually disappeared, and the monkey was free to make any saccades in the inter-trial 763 interval. The foveal neurons did not burst for the disappearance of the gra<ng, but they s<ll burst for the 764 disappearance of the fixa<on spot during the instructed-saccade phase of the trial (p = 6.6424 x 10-8 for the low 765 spa<al frequency, p = 1.5453 x 10-7 for the high spa<al frequency; signrank test). Thus, foveal bursts were task-766 dependent. 767 768 769 770 771 Across the populaDon of foveal SC neurons, we then collected offset responses and 772 compared them to foveal bursts. In Fig. 8C, we took the neurons that staDsDcally exhibited 773 foveal bursts at the go signal (Methods), and we invesDgated how they behaved in terms of 774 offset responses during the RF mapping task. The neurons did generally exhibit offset 775 responses, as might be expected (Hafed et al., 2023). Importantly, we also looked at foveal 776 SC neurons that did not staDsDcally emit foveal bursts in the main delayed saccade task 777 (Methods). If our foveal bursts of Figs. 5-7 were fully explained by RF offset responses, then 778 these neurons should not have exhibited any offset responses at all in the RF mapping task. 779 In reality, this was not the case at all (Fig. 8D). There were clear offset responses in the RF 780 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 21 mapping task for the neurons that did not exhibit strong foveal bursts at the go signal in the 781 main delayed saccade task. Thus, foveal bursts at the go signal in the classic delayed saccade 782 paradigm could be disDnct from simple offset responses associated with fixaDon spot 783 removal. 784 785 Perhaps the strongest evidence that we had for a dissociaDon between offset responses and 786 the foveal SC bursts at the go signal came from our main delayed saccade task itself. 787 Specifically, in this task, aher the monkeys foveated the instructed saccade target (the spaDal 788 frequency graDng), this target eventually disappeared aher a few hundred milliseconds of 789 fixaDon (Methods). Since the monkeys had successfully foveated the saccade target prior to 790 that, the removal of this target in this case was equivalent to an offset sDmulus for the 791 neurons’ RF’s. CriDcally, the monkeys had no explicit task instrucDon associated with the 792 target removal; this disappearance of the visual sDmulus was the onset of the inter-trial 793 interval in which the monkeys were free to look wherever they wanted (Methods), and we 794 analyzed neuronal acDvity of the foveal neurons upon target removal only when there were 795 no saccades for at least 150 ms aher the removal. As can be seen from Fig. 8E, the neurons 796 that did show foveal bursts at the go signal (y-axis) did not show substanDal acDvity 797 elevaDons upon target removal (x-axis). 798 799 Thus, all of the analyses of Fig. 8 suggest that foveal SC bursts are dependent on the task 800 context, and may not trivially reflect simple sDmulus-offset responses due to fixaDon spot 801 removal in the classic delayed saccade paradigm. 802 803 804 Foveal bursts cooccur with peripheral visual responses in reflexive saccade tasks 805 806 If foveal bursts are indeed not explained by sDmulus offset responses, then are the 807 peripheral SC pauses at least sDll explained via long-range lateral inhibiDon mechanisms in 808 the SC (Munoz and Istvan, 1998; Trappenberg et al., 2001; Isa and Hall, 2009; Marino et al., 809 2011; Satel et al., 2011; Phongphanphanee et al., 2014)? We believe that this may not 810 necessarily be the case, and our evidence for this comes from another classic saccade-811 related task, now not enforcing a delay period between fixaDon spot removal and eccentric 812 saccade target appearance (Fig. 9A). When recording from peripheral SC neurons in this task, 813 short-latency visual responses appear, before the saccade-related motor bursts quickly 814 evolve (Bell et al., 2006; Boehnke and Munoz, 2008; Marino et al., 2012; Marino et al., 2015; 815 Troienberg et al., 2026). If lateral inhibiDon was the sole determinant of complementary 816 transient pauses and bursts at disparate SC loci (Fig. 7), then the peripheral visual bursts in 817 this version of the saccade task might be expected to eliminate the foveal bursts that we 818 observed above, and maybe even replace them with pauses instead. Thus, we recorded not 819 only peripheral SC neurons, but also foveal ones in the classic reflexive, visually-guided 820 saccade task (Fig. 9B). 821 822 Figure 9C shows the acDvity of an example peripheral SC neuron. Here, we ploied the 823 neuron’s acDvity aligned on eccentric saccade target onset (which coincided with fixaDon 824 spot removal in this task; Fig. 9A), and aher subtracDng pre-sDmulus acDvity like we did 825 recently (Zhang et al., 2026). The neuron exhibited a short-latency visual response, followed 826 <50-100 ms later by a second volley of spiking. This second volley of spiking was the 827 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 22 saccade-related motor burst, as can be seen when aligning the same data to saccade onset 828 (Fig. 9D). Our prior work with this task has taught us that the same rapid transformaDon 829 from a visual to a motor regime also does take place in it (Baumann et al., 2023), suggesDng 830 that a foveal trigger signal might sDll be recruited in such a task. Remarkably, this was indeed 831 the case. When we recorded from an example foveal SC neuron in this task, it exhibited a 832 strong foveal burst (Fig. 9E), before it expectedly paused at saccade onset (Fig. 9F). The net 833

was that the two example neurons of Fig. 9C-F provided suggesDve evidence for the 834 existence of two simultaneous short-latency acDvity bursts in the SC at two very disparate 835 loci: one peripheral represenDng the saccade target appearance, and one foveal associated 836 with the go signal for generaDng a saccade. 837 838 839 840 841 842 843 Figure 9 Persistence of foveal bursts even in immediate, visually-guided saccade tasks. (A) In this task, the go 844 signal (fixa<on spot disappearance) coincided with the onset of the peripheral saccade target. (B) Thus, we could 845 ask whether at the <me of the peripheral visual burst in response to target appearance, we could s<ll observe a 846 foveal burst. (C, D) Example peripheral SC neuron showing expected ac<vity discharge: there was an ini<al visual 847 burst (C), followed immediately by a saccade-related motor burst (D). (E, F) Example foveal SC neuron from the 848 same task. Remarkably, the foveal neuron burst (E) at the same <me as the peripheral visual burst in C. It then 849 reduced its ac<vity at saccade onset (F), as expected from foveal SC neurons (Munoz and Wurtz, 1993b; Hafed 850 and Krauzlis, 2012; Zhang et al., 2025). Trial numbers in C-F can be inferred from the spike rasters, and error bars 851 denote SEM across trials. (G) Popula<on results for all peripheral neurons collected during this task. Error bars 852 denote SEM across neurons. Visual and motor bursts were evident. (H) For the foveal neurons, they exhibited a 853 foveal burst even when the peripheral neurons were burs<ng for the visual onset inside their RF’s (the faint blue 854 curve is a replica<on of the curve in G to illustrate the similarity of the <ming of the two bursts). The neurons 855 then reduced their ac<vity when the peripheral neurons were burs<ng at saccade onset (post-saccadic reafferent 856 responses also emerged). Thus, in the immediate, visually-guided saccade task, there are two simultaneous 857 bursts at two different loci in the SC aJer target appearance. Error bars denote SEM across neurons. 858 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 23 859 Across the populaDon, our results were consistent with those seen from the two example 860 neurons. Specifically, peripheral SC neurons showed expected biphasic responses consisDng 861 of a first visual burst followed by a later saccade-related motor burst (Fig. 9G). And, 862 remarkably, foveal SC neurons showed a transient burst at the same Dme as the peripheral 863 visual bursts (Fig. 9H), before reducing their acDvity at the Dme of saccade onset. 864 865 Thus, the complementary nature of transient pauses and bursts at two disparate SC loci that 866 we saw in the delayed saccade paradigm above (e.g. Fig. 7C) was not an obligatory 867 observaDon dictated by lateral inhibiDon mechanisms. Clearly, simultaneous short-latency 868 bursts are possible in the SC (Fig. 9). InteresDngly, we also previously saw that it is possible 869 to observe simultaneous acDvity increases in two different locaDons on the SC map 870 (Buonocore et al., 2021), but in that case, it was a simultaneity between a peripheral visual 871 burst and a foveal microsaccade-related motor burst. Here, it was two transient non-872 saccadic bursts that we observed. Similarly, prior experiments with peripheral SC transient 873 pauses concluded that they were not fully explained by lateral inhibiDon from foveal SC 874 responses to foveal image changes (Li et al., 2006). 875 876 877 Peripheral visual responses, but not foveal bursts, correlate with saccadic 878 reac6on 6me 879 880 The immediate saccade paradigm of Fig. 9A also allowed us to further explore addiDonal 881 properDes of foveal SC bursts. Specifically, we asked to what extent these bursts might relate 882 to saccade Dming variability. In peripheral neurons, we replicated the expected finding that 883 SC visual responses are significantly stronger (and earlier) for faster saccadic reacDon Dmes 884 (Bell et al., 2006; Boehnke and Munoz, 2008; Marino et al., 2012; Marino et al., 2015; Chen 885 et al., 2018; Troienberg et al., 2026). This can be seen from the populaDon results shown in 886 Fig. 10A, C: here, we split the trials for each neuron according to whether the saccadic 887 reacDon Dme was faster or slower than the median of the session, and we found that the 888 faster trials had stronger peripheral SC visual bursts (Fig. 10A, C). No such relaDonship 889 emerged for the simultaneously occurring foveal SC bursts (Fig. 10B, D), again suggesDng 890 that these bursts might be a switch-like trigger signal that is independent of saccade-target 891 appearance (Fig. 6A), saccade direcDon (Fig. 6B), or saccade Dming (Fig. 10B, D). 892 893 894 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 24 895 896 Figure 10 Dissocia/on between foveal bursts and saccadic reac/on /mes in immediate, visually -guided 897 saccade tasks. (A) In peripheral SC neurons, we replicated expecta<ons from the previous literature (Bell et al., 898 2006; Boehnke and Munoz, 2008; Marino et al., 2012; Marino et al., 2015; Chen et al., 2018; TroWenberg et al., 899 2026): visual bursts in response to target onset were weaker and slightly later for later saccadic reac<on <mes. 900 In this figure, we split trials by the median saccadic reac<on <me of each session. Then, we ploWed the popula<on 901 firing rates. Trials with late saccades (light blue; later saccade bursts than in the saturated blue) had much 902 weakened and slightly delayed visual bursts. Error bars denote SEM across neurons. (B) In the foveal neurons, 903 we did not observe a clear difference in foveal bursts between trials with fast and slow saccadic reac<on <mes 904 (the different <mes of post-burst ac<vity reduc<ons in the two curves reflect the different <mes of saccade 905 onsets in the two sets of trials per neuron). Error bars denote SEM. (C) Individual neuron results from the analysis 906 of A. Visual bursts were stronger on fast trials (p = 2.6741 x 10-35; signrank test). n=253 neurons. (D) For foveal 907 neurons, there was no difference (p=0.059; signrank test). n=41 neurons. 908 909 910 911 912 913 Finally, and for completeness, we also revisited the foveal bursts from our delayed saccade 914 paradigm. When we analyzed these bursts as a funcDon of saccadic reacDon Dme, we again 915 found that the bursts had the same strength for either the fast or slow saccadic reacDon 916 Dme trials (Fig. 11). Peripheral SC pauses also did not appear to have a systemaDc 917 relaDonship to saccade Dming in the same task. 918 919 920 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 25 921 922 Figure 11 Dissocia/on between foveal bursts and saccadic reac/on /mes also in delayed saccade contexts. (A) 923 For the delayed saccade task of Figs. 5-8, we split trials based on the median saccadic reac<on <me of a given 924 session (just like we did for Fig. 10), and we then ploWed popula<on firing rates for the two split groups. Foveal 925 bursts were similar in both cases. (B) Similar results for the other spa<al frequency of the peripheral saccade 926 target. Thus, the foveal bursts behaved similarly in both versions of our visually -guided saccade paradigm in 927 terms of subsequent saccade <ming. Error bars denote SEM across neurons. 928 929 930 931 932

933 934 We invesDgated neuronal response dynamics at the Dme of “releasing” instructed saccades 935 in two classic visually-guided saccade paradigms. Such paradigms are rouDnely used in the 936 study of primate sensaDon, cogniDon, and acDon. We specifically idenDfied a transient 937 foveal SC signal that may jumpstart peripheral saccadic orienDng. In delayed saccade 938 paradigms, this foveal signal leads peripheral SC acDvity pauses by ~10 ms, is not explained 939 by sDmulus-offset responses, does not depend on sDmulus appearance or saccade direcDon, 940 and is independent of eventual saccade Dming. Remarkably, in immediate saccade 941 paradigms, this transient foveal signal sDll occurs, resulDng in two simultaneous acDvity 942 bursts at two disparate loci in the SC topographic map. 943 944 Peripheral SC acDvity pauses have been reported before. For example, Li and colleagues 945 employed a foveal sDmulus change as a cue in a target selecDon paradigm. Peripheral SC 946 neurons exhibited transient pauses in their acDvity at the Dme of the foveal sDmulus change 947 (Li et al., 2006). Similar observaDons were made in a variety of related tasks, most of which 948 involving the appearance or modificaDon of a sizeable foveal visual sDmulus, and in mulDple 949 brain areas like the SC, frontal eye fields (FEF), and lateral intraparietal area (LIP) (Li and 950 Basso, 2005; Dorris et al., 2007; White et al., 2013; Grimaldi et al., 2018; Cho et al., 2021; 951 Shinn et al., 2022; SDne et al., 2023). Here, we explicitly recorded foveal SC neurons without 952 presenDng a new foveal sDmulus and demonstrated that peripheral SC pauses might be 953 related to a transient foveal trigger signal. Importantly, this signal is likelu not a simple 954 sensory response, because it clearly depended on the task context, and also because it was 955 dissociated from sDmulus-offset responses observed during RF mapping (Fig. 8). Moreover, 956 the foveal trigger signal need not directly mediate peripheral pauses via lateral inhibiDon 957 (Munoz and Istvan, 1998; Trappenberg et al., 2001; Isa and Hall, 2009; Marino et al., 2011; 958 Satel et al., 2011; Phongphanphanee et al., 2014). This is because we saw peripheral visual 959 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 26 bursts at the same Dme as the foveal bursts in the immediate saccade paradigm (Figs. 9, 10); 960 in this case, it would be difficult for lateral inhibiDon mechanisms alone to account for such 961 simultaneity of bursts. 962 963 We believe that there are at least two good reasons for transient peripheral SC pauses to 964 occur in our invesDgated paradigms. The first is biophysical. SC saccade-related bursts are 965 explosive, and can exceed 1000 spikes/s (Goossens and van Opstal, 2012). Thus, a transient 966 pause right before motor bursts could aid in the biophysical processes leading up to burst 967 generaDon (Aizawa et al., 1999; Saito and Isa, 2003; Moren et al., 2013). 968 969 The second reason why peripheral SC pauses may be funcDonally useful is that a large 970 number of saccade-related SC neurons are also sensory neurons as well (Massot et al., 2019; 971 Hafed et al., 2023). This requires a representaDonal transformaDon between visual and 972 motor regimes (Jagadisan and Gandhi, 2022; Ayar et al., 2023; Baumann et al., 2023; 973 Bourrelly et al., 2023; Heusser et al., 2023), and this transformaDon even needs to happen 974 within only a few tens of milliseconds in the immediate saccade paradigm (Baumann et al., 975 2023). For example, Jagadisan and Gandhi (Jagadisan and Gandhi, 2022) showed that while 976 visual and motor bursts in the SC can appear qualitaDvely similar to each other and reach 977 similar firing rates, the temporal structure of SC populaDon acDvity is altered at the Dme of 978 saccade generaDon. In our subsequent confirmaDon of this observaDon, we also noted that 979 acDvity subspaces could be orthogonal to each other in the two neuronal regimes, and that 980 individual neuron preferences for specific images can change between visual and motor 981 epochs (Baumann et al., 2023). Thus, peripheral SC pauses right before the motor bursts 982 would consDtute a perfect resetng mechanism for rapid representaDonal transformaDons 983 to be implemented. It would be interesDng in future studies to understand how these 984 transformaDons themselves emerge. 985 986 In terms of the fovea, prior work has recorded from the deep rostral SC during saccades. 987 Consistent pauses in foveal SC acDvity were observed at the Dme of saccade generaDon 988 (Munoz and Wurtz, 1993b, a). InteresDngly, there were sDll some hints in some of these 989 studies for subtle elevaDons in foveal SC acDvity when releasing saccades by certain task 990 events (Dorris and Munoz, 1995; Dorris et al., 1997; Everling et al., 1998; Munoz and Istvan, 991 1998; Everling et al., 1999; Munoz et al., 2000; Hafed and Krauzlis, 2008; White et al., 2013). 992 However, there was nothing reported that was as strong as we saw, and the foveal 993 elevaDons were not the focus of these earlier studies (and thus not exhausDvely 994 characterized). 995 996 From the perspecDve of the peripheral reset alluded to above, our observed foveal bursts 997 are parDcularly interesDng. This is because they may act as the trigger signal jumpstarDng 998 peripheral pausing, and thus acDvely parDcipate in the orienDng process. Consistent with 999 this, our foveal bursts were not sDmulus or saccade-direcDon dependent. They were also not 1000 related to subsequent saccade Dming, unlike (peripheral) visual bursts (Bell et al., 2006; 1001 Boehnke and Munoz, 2008; Marino et al., 2012; Marino et al., 2015; Troienberg et al., 1002 2026). And, most importantly, they sDll occurred in the immediate saccade paradigm, which 1003 sDll requires a representaDonal transformaDon between visual and motor regimes 1004 (Baumann et al., 2023). Of course, the quesDon is: what drives these bursts? In our work, we 1005 tried to dissociate them from simple offset responses due to the removal of the fixaDon 1006 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 27 spot. InteresDngly, we found that target removal at the ends of trials in the main delayed 1007 saccade paradigm was not associated with foveal bursts, suggesDng that the bursts were 1008 indeed dependent on the context of releasing a planned, instructed saccade. 1009 1010 More broadly, we think that it is not unheard of for a foveal signal in the SC to be relevant 1011 for peripheral SC processes. For example, we recently found, in the context of peri-1012 microsaccadic changes in peripheral visual sensiDvity, that exclusive experimental control 1013 over foveal SC state is sufficient to modulate peripheral visual sensiDvity (Zhang et al., 2026). 1014 Similarly, in the opposite direcDon, we also found that foveal SC state can be modulated by 1015 peripheral SC state across saccadic eye movements (Zhang et al., 2025). These observaDons 1016 may relate to wider concepts like traveling waves. While such waves are presently gaining 1017 more research interest, classic SC work has indeed demonstrated a potenDal role of such 1018 waves in acDve vision (Munoz and Wurtz, 1995a; Nakahara et al., 2006). In the future, it 1019 would be interesDng to invesDgate such waves in more detail. For example, we can aiempt 1020 to understand the links between foveal bursts and peripheral pauses by blocking the Inputs 1021 to the SC from the cortex and invesDgaDng whether signatures of an impact of foveal bursts 1022 sDll appears in the peripheral SC representaDon aher losing corDcal drive. 1023 1024 We are also especially intrigued by our observaDon of simultaneous bursts in the foveal and 1025 peripheral SC in the immediate visually-guided saccade paradigm. These simultaneous bursts 1026 are significant because they suggest that the peripheral pauses in the delayed saccade 1027 paradigm may not necessarily result from lateral inhibiDon mechanisms. In fact, in the 1028 immediate saccade paradigm, it was known for many decades that peripheral visual bursts 1029 are actually enhanced, rather than suppressed, relaDve to visual bursts during fixaDon 1030 (Goldberg and Wurtz, 1972a). Since it is very likely that foveal bursts sDll occurred in these 1031 classic experiments (Goldberg and Wurtz, 1972a), had they been invesDgated, it remains to 1032 be seen whether the foveal bursts may actually aid in the peripheral enhancement. 1033 Certainly, in our recent work, we found that enhancing pre-sDmulus acDvity in the foveal SC 1034 can indeed help in enhancing peripheral visual bursts (Zhang et al., 2026). We should also 1035 note that there are other observaDons in the literature for which direct evidence of lateral 1036 inhibiDon between the foveal and peripheral SC representaDons was absent. For example, 1037 when we invesDgated microsaccade generaDon in the foveal SC representaDon, we found 1038 that there could sDll be microsaccade-related foveal SC motor bursts at the exact same Dme 1039 as visually-driven peripheral SC bursts (Buonocore et al., 2021), and we again argued that 1040 lateral inhibiDon would not explain these observaDons. It would be interesDng in the future 1041 to understand the condiDons under which physiological correlates of a lack of lateral 1042 inhibiDon would be most likely to occur. 1043 1044 Finally, the cauDon raised by our present work here is that in the classic study of saccade 1045 tasks, we may have generally assumed that the fixaDon spot was not influenDal. However, it 1046 clearly maiers. While we believe that the foveal bursts that we observed are not direct 1047 responses to fixaDon spot removal in our tasks, it is sDll imperaDve to next ask whether they 1048 would conDnue to happen when there is no foveal sensory transient. For example, one could 1049 use a more abstract go instrucDon (that is also not represented foveally). One possibility 1050 could be to maintain the fixaDon spot visible throughout the trials, and instead use a subtle 1051 spaDally-uninformaDve auditory cue as the instrucDon to generate a saccade. Under certain 1052 circumstances, such sounds only minimally affect visually-driven effects (Malevich et al., 1053 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 28 2026), and as long as they do not drive foveal SC neurons, one can record from the foveal SC 1054 and check whether the foveal bursts that we observed would sDll happen. 1055 1056 1057 1058

1059 1060 We were funded by the German Research AssociaDon (Deutsche Forschungsgemeinschah; 1061 DFG) under the Special Priority Programme “SPP 2411 Sensing LOOPS” (project numbers 1062 520617944, 520283985; HA6749/11-1). We were also supported by the DFG-funded 1063 InternaDonal Research Training Group IRTG2804. 1064 1065 1066 1067 Compe3ng interests statement 1068 1069 The authors declare no compeDng interests. 1070 1071 1072 1073

1074 1075 1076 1077 Aizawa H, Kobayashi Y , Yamamoto M, Isa T (1999) InjecDon of nicoDne into the superior 1078 colliculus facilitates occurrence of express saccades in monkeys. J Neurophysiol 1079 82:1642-1646. 1080 Ayar EC, Heusser MR, Bourrelly C, Gandhi NJ (2023) DisDnct context- and content-dependent 1081 populaDon codes in superior colliculus during sensaDon and acDon. Proc Natl Acad 1082 Sci U S A 120:e2303523120. 1083 Baumann MP , Bogadhi AR, Denninger AF, Hafed ZM (2023) Sensory tuning in neuronal 1084 movement commands. Proc Natl Acad Sci U S A 120:e2305759120. 1085 Bell AH, Meredith MA, Van Opstal AJ, Munoz DP (2006) SDmulus intensity modifies saccadic 1086 reacDon Dme and visual response latency in the superior colliculus. Exp Brain Res 1087 174:53-59. 1088 Boehnke SE, Munoz DP (2008) On the importance of the transient visual response in the 1089 superior colliculus. Curr Opin Neurobiol 18:544-551. 1090 Bogadhi AR, Hafed ZM (2023) Express detecDon of visual objects by primate superior 1091 colliculus neurons. ScienDfic reports 13:21730. 1092 Bourrelly C, Massot C, Gandhi NJ (2023) Rapid Input-Output TransformaDon between Local 1093 Field PotenDal and Spiking AcDvity during SensaDon but not AcDon in the Superior 1094 Colliculus. J Neurosci 43:4047-4061. 1095 Buonocore A, Hafed ZM (2023) The inevitability of visual interrupDon. J Neurophysiol 1096 130:225-237. 1097 Buonocore A, Tian X, Khademi F, Hafed ZM (2021) Instantaneous movement-unrelated 1098 midbrain acDvity modifies ongoing eye movements. eLife 10. 1099 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 29 Cerkevich CM, Lyon DC, Balaram P , Kaas JH (2014) DistribuDon of corDcal neurons projecDng 1100 to the superior colliculus in macaque monkeys. Eye and brain 2014:121-137. 1101 Chen CY , Hafed ZM (2017) A neural locus for spaDal-frequency specific saccadic suppression 1102 in visual-motor neurons of the primate superior colliculus. J Neurophysiol 117:1657-1103 1673. 1104 Chen CY , Hafed ZM (2018) OrientaDon and Contrast Tuning ProperDes and Temporal Flicker 1105 Fusion CharacterisDcs of Primate Superior Colliculus Neurons. Front Neural Circuits 1106 12:58. 1107 Chen CY , Ignashchenkova A, Thier P , Hafed ZM (2015) Neuronal Response Gain Enhancement 1108 prior to Microsaccades. Curr Biol 25:2065-2074. 1109 Chen CY , Hoffmann KP , Distler C, Hafed ZM (2019) The Foveal Visual RepresentaDon of the 1110 Primate Superior Colliculus. Curr Biol 29:2109-2119 e2107. 1111 Chen CY , Sonnenberg L, Weller S, Witschel T, Hafed ZM (2018) SpaDal frequency sensiDvity in 1112 macaque midbrain. Nat Commun 9:2852. 1113 Cho SH, Crapse T, Grimaldi P , Lau H, Basso MA (2021) Variable StaDsDcal Structure of 1114 Neuronal Spike Trains in Monkey Superior Colliculus. J Neurosci 41:3234-3253. 1115 Cynader M, Berman N (1972) RecepDve-field organizaDon of monkey superior colliculus. J 1116 Neurophysiol 35:187-201. 1117 Dorris MC, Munoz DP (1995) A neural correlate for the gap effect on saccadic reacDon Dmes 1118 in monkey. Journal of neurophysiology 73:2558-2562. 1119 Dorris MC, Pare M, Munoz DP (1997) Neuronal acDvity in monkey superior colliculus related 1120 to the iniDaDon of saccadic eye movements. The Journal of neuroscience : the official 1121 journal of the Society for Neuroscience 17:8566-8579. 1122 Dorris MC, Olivier E, Munoz DP (2007) CompeDDve integraDon of visual and preparatory 1123 signals in the superior colliculus during saccadic programming. The Journal of 1124 neuroscience : the official journal of the Society for Neuroscience 27:5053-5062. 1125 Everling S, Pare M, Dorris MC, Munoz DP (1998) Comparison of the discharge characterisDcs 1126 of brain stem omnipause neurons and superior colliculus fixaDon neurons in monkey: 1127 implicaDons for control of fixaDon and saccade behavior. J Neurophysiol 79:511-528. 1128 Everling S, Dorris MC, Klein RM, Munoz DP (1999) Role of primate superior colliculus in 1129 preparaDon and execuDon of anD-saccades and pro-saccades. J Neurosci 19:2740-1130 2754. 1131 Findlay JM (1982) Global visual processing for saccadic eye movements. Vision Res 22:1033-1132 1045. 1133 Findlay JM, Walker R (1999) A model of saccade generaDon based on parallel processing and 1134 compeDDve inhibiDon. Behav Brain Sci 22:661-674; discussion 674-721. 1135 Goldberg ME, Wurtz RH (1972a) AcDvity of superior colliculus in behaving monkey. II. Effect 1136 of aienDon on neuronal responses. J Neurophysiol 35:560-574. 1137 Goldberg ME, Wurtz RH (1972b) AcDvity of superior colliculus in behaving monkey. I. Visual 1138 recepDve fields of single neurons. J Neurophysiol 35:542-559. 1139 Goossens HH, van Opstal AJ (2012) OpDmal control of saccades by spaDal-temporal acDvity 1140 paierns in the monkey superior colliculus. PLoS computaDonal biology 8:e1002508. 1141 Grimaldi P , Cho SH, Lau H, Basso MA (2018) Superior colliculus signals decisions rather than 1142 confidence: analysis of single neurons. J Neurophysiol 120:2614-2629. 1143 Hafed ZM (2025) Superior colliculus peri-saccadic field potenDals are dominated by a visual 1144 sensory preference for the upper visual field. iScience 28:112021. 1145 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 30 Hafed ZM, Krauzlis RJ (2008) Goal representaDons dominate superior colliculus acDvity 1146 during extrafoveal tracking. J Neurosci 28:9426-9439. 1147 Hafed ZM, Krauzlis RJ (2012) Similarity of superior colliculus involvement in microsaccade 1148 and saccade generaDon. J Neurophysiol 107:1904-1916. 1149 Hafed ZM, Chen CY (2016) Sharper, Stronger, Faster Upper Visual Field RepresentaDon in 1150 Primate Superior Colliculus. Curr Biol 26:1647-1658. 1151 Hafed ZM, Hoffmann KP , Chen CY , Bogadhi AR (2023) Visual FuncDons of the Primate 1152 Superior Colliculus. Annu Rev Vis Sci 9:361-383. 1153 Hafed ZM, Chen CY , Tian X, Baumann M, Zhang T (2021) AcDve vision at the foveal scale in 1154 the primate superior colliculus. J Neurophysiol. 1155 Heusser MR, Jagadisan UK, Gandhi NJ (2023) Drihing populaDon dynamics with transient 1156 resets characterize sensorimotor transformaDon in the monkey superior colliculus. 1157 bioRxiv. 1158 Hikosaka O, Wurtz RH (1983) Visual and oculomotor funcDons of monkey substanDa nigra 1159 pars reDculata. I. RelaDon of visual and auditory responses to saccades. J 1160 Neurophysiol 49:1230-1253. 1161 Humphrey NK (1968) Responses to visual sDmuli of units in the superior colliculus of rats and 1162 monkeys. Exp Neurol 20:312-340. 1163 Isa T, Hall WC (2009) Exploring the superior colliculus in vitro. Journal of neurophysiology 1164 102:2581-2593. 1165 Jagadisan UK, Gandhi NJ (2022) PopulaDon temporal structure supplements the rate code 1166 during sensorimotor transformaDons. Current Biology 32:1010-1025. 1167 Li X, Basso MA (2005) CompeDDve sDmulus interacDons within single response fields of 1168 superior colliculus neurons. J Neurosci 25:11357-11373. 1169 Li X, Kim B, Basso MA (2006) Transient pauses in delay-period acDvity of superior colliculus 1170 neurons. J Neurophysiol 95:2252-2264. 1171 Lund RD (1972) SynapDc paierns in the superficial layers of the superior colliculus of the 1172 monkey, Macaca mulaia. Exp Brain Res 15:194-211. 1173 Malevich T, Baumann MP , Yu Y , Hafed ZM (2026) Vision dominates sound in mediaDng classic 1174 cue-induced microsaccadic eye movement modulaDons in rhesus macaque monkeys. 1175 Journal of Neurophysiology 135:777-796. 1176 Marino RA, Levy R, Munoz DP (2015) Linking express saccade occurance to sDmulus 1177 properDes and sensorimotor integraDon in the superior colliculus. J Neurophysiol 1178 114:879-892. 1179 Marino RA, Trappenberg TP , Dorris M, Munoz DP (2011) SpaDal InteracDons in the Superior 1180 Colliculus Predict Saccade Behavior in a Neural Field Model. Journal of cogniDve 1181 neuroscience. 1182 Marino RA, Levy R, Boehnke S, White BJ, It L, Munoz DP (2012) Linking visual response 1183 properDes in the superior colliculus to saccade behavior. Eur J Neurosci 35:1738-1184 1752. 1185 Massot C, Jagadisan UK, Gandhi NJ (2019) Sensorimotor transformaDon elicits systemaDc 1186 paierns of acDvity along the dorsoventral extent of the superior colliculus in the 1187 macaque monkey. Commun Biol 2:287. 1188 May PJ (2006) The mammalian superior colliculus: laminar structure and connecDons. Prog 1189 Brain Res 151:321-378. 1190 Mohler CW, Wurtz RH (1976) OrganizaDon of monkey superior colliculus: intermediate layer 1191 cells discharging before eye movements. J Neurophysiol 39:722-744. 1192 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 31 Moren J, Shibata T, Doya K (2013) The mechanism of saccade motor paiern generaDon 1193 invesDgated by a large-scale spiking neuron model of the superior colliculus. PLoS 1194 One 8:e57134. 1195 Munoz DP , Wurtz RH (1993a) FixaDon cells in monkey superior colliculus. II. Reversible 1196 acDvaDon and deacDvaDon. J Neurophysiol 70:576-589. 1197 Munoz DP , Wurtz RH (1993b) FixaDon cells in monkey superior colliculus. I. CharacterisDcs of 1198 cell discharge. J Neurophysiol 70:559-575. 1199 Munoz DP , Wurtz RH (1995a) Saccade-related acDvity in monkey superior colliculus. II. 1200 Spread of acDvity during saccades. J Neurophysiol 73:2334-2348. 1201 Munoz DP , Wurtz RH (1995b) Saccade-related acDvity in monkey superior colliculus. I. 1202 CharacterisDcs of burst and buildup cells. J Neurophysiol 73:2313-2333. 1203 Munoz DP , Istvan PJ (1998) Lateral inhibitory interacDons in the intermediate layers of the 1204 monkey superior colliculus. J Neurophysiol 79:1193-1209. 1205 Munoz DP , Dorris MC, Pare M, Everling S (2000) On your mark, get set: brainstem circuitry 1206 underlying saccadic iniDaDon. Can J Physiol Pharmacol 78:934-944. 1207 Nakahara H, Morita K, Wurtz RH, OpDcan LM (2006) Saccade-related spread of acDvity 1208 across superior colliculus may arise from asymmetry of internal connecDons. J 1209 Neurophysiol 96:765-774. 1210 Phongphanphanee P , Marino RA, Kaneda K, Yanagawa Y , Munoz DP , Isa T (2014) DisDnct local 1211 circuit properDes of the superficial and intermediate layers of the rodent superior 1212 colliculus. Eur J Neurosci 40:2329-2343. 1213 Saito Y , Isa T (2003) Local excitatory network and NMDA receptor acDvaDon generate a 1214 synchronous and bursDng command from the superior colliculus. J Neurosci 23:5854-1215 5864. 1216 Satel J, Wang Z, Trappenberg TP , Klein RM (2011) Modeling inhibiDon of return as short-term 1217 depression of early sensory input to the superior colliculus. Vision Res 51:987-996. 1218 Schiller PH, Koerner F (1971) Discharge characterisDcs of single units in superior colliculus of 1219 the alert rhesus monkey. J Neurophysiol 34:920-936. 1220 Schiller PH, Stryker M, Cynader M, Berman N (1974) Response characterisDcs of single cells 1221 in the monkey superior colliculus following ablaDon or cooling of visual cortex. J 1222 Neurophysiol 37:181-194. 1223 Shinn M, Lee D, Murray JD, Seo H (2022) Transient neuronal suppression for exploitaDon of 1224 new sensory evidence. Nat Commun 13:23. 1225 SDne GM, Trautmann EM, Jeurissen D, Shadlen MN (2023) A neural mechanism for 1226 terminaDng decisions. Neuron 111:2601-2613 e2605. 1227 Trappenberg TP , Dorris MC, Munoz DP , Klein RM (2001) A model of saccade iniDaDon based 1228 on the compeDDve integraDon of exogenous and endogenous signals in the superior 1229 colliculus. Journal of cogniDve neuroscience 13:256-271. 1230 Troienberg C, Yu Y , Zhang T, Baumann MP , Malevich T, Prasad S, Hafed ZM (2026) Much 1231 higher covariaDon with foveaDon Dming by superior colliculus than primary visual 1232 corDcal neuronal acDvity. iScience 29:115432. 1233 White BJ, Marino RA, Boehnke SE, It L, Theeuwes J, Munoz DP (2013) CompeDDve 1234 integraDon of visual and goal-related signals on neuronal accumulaDon rate: a 1235 correlate of oculomotor capture in the superior colliculus. J Cogn Neurosci 25:1754-1236 1768. 1237 Wilson ME, Toyne MJ (1970) ReDno-tectal and corDco-tectal projecDons in Macaca mulaia. 1238 Brain Res 24:395-406. 1239 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint 32 Wolpert DM, Ghahramani Z, Jordan MI (1995) An internal model for sensorimotor 1240 integraDon. Science 269:1880-1882. 1241 Wurtz RH, Mohler CW (1976) Enhancement of visual responses in monkey striate cortex and 1242 frontal eye fields. J Neurophysiol 39:766-772. 1243 Zhang T, Bogadhi AR, Hafed ZM (2025) Foveal neurons of the monkey superior colliculus 1244 signal trans-saccadic predicDon errors. PLOS Biology. 1245 Zhang T, Malevich T, Baumann MP , Hafed ZM (2022) Superior colliculus saccade motor bursts 1246 do not dictate movement kinemaDcs. Commun Biol 5:1222. 1247 Zhang T, Tian X, Malevich T, Baumann MB, Hafed ZM (2026) Foveal acDon for the control of 1248 extrafoveal vision. BioRxiv. 1249 Zheng YJ, Adams DL, Gentry TN, Dilbeck MD, Economides JR, Horton JC (2024) ReDnal Input 1250 to Macaque Superior Colliculus Derives from Branching Axons ProjecDng to the 1251 Lateral Geniculate Nucleus. J Neurosci 44. 1252 1253 1254 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 8, 2026. ; https://doi.org/10.64898/2026.04.06.716792doi: bioRxiv preprint