Faster detection of “darks” than “brights” by monkey superior colliculus neurons

Visual processing is segregated into ON and OFF channels as early as in the retina, and the superficial (output) layers of the primary visual cortex are dominated by neurons preferring dark stimuli. However, it is not clear how the timing of neural processing differs between “darks” and “brights” in general, especially in light of psychophysical evidence; it is also equally not clear how subcortical visual pathways that are critical for active orienting represent stimuli of positive (luminance increments) and negative (luminance decrements) contrast polarity. Here, we recorded from all visually-responsive neuron types in the superior colliculus (SC) of two male rhesus macaque monkeys. We presented a disc (0.51 deg radius) within the response fields (RF’s) of neurons, and we varied, across trials, stimulus Weber contrast relative to a gray background. We also varied contrast polarity. There was a large diversity of preferences for darks and brights across the population. However, regardless of individual neural sensitivity, most neurons responded significantly earlier to dark than bright stimuli. This resulted in a dissociation between neural sensitivity and visual response onset latency: a neuron could exhibit a weaker response to a dark stimulus than to a bright stimulus of the same contrast, but it would still have an earlier response to the dark stimulus. Our results highlight an additional candidate visual neural pathway for explaining behavioral differences between the processing of darks and brights, and they demonstrate the importance of temporal aspects in the visual neural code for orienting eye movements. Significance statement Objects in our environment, such as birds flying across a bright sky, often project shadows (or images darker than the surround) on our retina. We studied how primate superior colliculus (SC) neurons visually process such dark stimuli. We found that the overall population of SC neurons represented both dark and bright stimuli equally well, as evidenced by a relatively equal distribution of neurons that were either more or less sensitive to darks. However, independent of sensitivity, the great majority of neurons detected dark stimuli earlier than bright stimuli, evidenced by a smaller response latency for the dark stimuli. Thus, SC neural response latency can be dissociated from response sensitivity, and it favors the faster detection of dark image contrasts.

Visual processing is segregated into ON and OFF channels as early as in the retina, and the 48 superficial (output) layers of the primary visual cortex are dominated by neurons preferring 49 dark stimuli. However, it is not clear how the timing of neural processing differs between 50 "darks" and "brights" in general, especially in light of psychophysical evidence; it is also 51 equally not clear how subcortical visual pathways that are critical for active orienting 52 represent stimuli of positive (luminance increments) and negative (luminance decrements) 53 contrast polarity. Here, we recorded from all visually-responsive neuron types in the 54 superior colliculus (SC) of two male rhesus macaque monkeys. We presented a disc (0.51 deg 55 radius) within the response fields (RF's) of neurons, and we varied, across trials, stimulus 56 Weber contrast relative to a gray background. We also varied contrast polarity. There was a 57 large diversity of preferences for darks and brights across the population. However, 58 regardless of individual neural sensitivity, most neurons responded significantly earlier to 59 dark than bright stimuli. This resulted in a dissociation between neural sensitivity and visual 60 response onset latency: a neuron could exhibit a weaker response to a dark stimulus than to 61 a bright stimulus of the same contrast, but it would still have an earlier response to the dark 62 stimulus. Our results highlight an additional candidate visual neural pathway for explaining 63 behavioral differences between the processing of darks and brights, and they demonstrate 64 the importance of temporal aspects in the visual neural code for orienting eye movements. 65 66 67 Significance statement 68 69 Objects in our environment, such as birds flying across a bright sky, often project shadows 70 (or images darker than the surround) on our retina. We studied how primate superior 71 colliculus (SC) neurons visually process such dark stimuli. We found that the overall 72 population of SC neurons represented both dark and bright stimuli equally well, as 73 evidenced by a relatively equal distribution of neurons that were either more or less Introduction 82 83 Early visual processing is segregated into parallel pathways conveying information about 84 either luminance increments or decrements in visual scenes (Hartline, 1938;Schiller et al., 85 1986). Such segregation starts in the retina and persists in the early retino-geniculate visual 86 pathway (Hubel and Wiesel, 1961;Schiller et al., 1986). Interestingly, such segregation is also 87 accompanied by asymmetries with which dark and bright stimuli are processed. For 88 example, primate retinal ganglion cells possess asymmetric spatial and temporal properties 89 depending on whether they are part of the ON or OFF pathway (Chichilnisky and Kalmar, 90 2002). Similarly, in the primate lateral geniculate nucleus (LGN), neurons with OFF-center 91 response fields (RF's) are more sensitive to their preferred stimuli (dark contrasts) than 92 neurons with ON-center RF's experiencing bright contrasts (Jiang et al., 2015). OFF-center 93 neurons also have higher spontaneous activity and more sustained responses during visual 94 stimulation (Jiang et al., 2015). Ultimately, signals reach the primary visual cortex (V1), 95 where ON/OFF asymmetries are amplified even more. For example, primate V1 is dominated 96 by "black" responses, especially in the superficial cortico-cortical output layers (Yeh et al.,97 2009). 98 99 Asymmetries in the processing of dark versus bright stimuli might make ecological sense. For 100 example, the incidence of dark contrasts in natural scenes is not necessarily uniform. 101 Instead, there is a coincidence of dark contrasts with regions of low spatial frequency, high 102 contrast, and far binocular disparities in natural images (Cooper and Norcia, 2015). As a 103 result, rhesus macaque V1 neurons having far preferred binocular disparities tend to also 104 prefer dark contrasts (Samonds et al., 2012). Similarly, in cat V1, there is a systematic 105 contrast-dependent OFF-dominance, matching natural scene statistics (Liu and Yao, 2014), 106 and cat V1 neurons are more strongly driven by luminance decrements than luminance 107 increments at low spatial frequencies . Interestingly, cat studies 108 revealed that ON and OFF domains in the LGN also exist in the V1 projections (Jin et al.,109 2008), with area centralis representations being dominated by dark preferences, and that 110 OFF-dominated V1 neurons respond earlier than ON-dominated ones . 111 This last observation on the timing of OFF and ON V1 channels is consistent with a large 112 body of psychophysical literature for better and faster processing of dark stimuli (e.g. 113 Komban et al., 2011;Komban et al., 2014). 114 115 Having said that, whether monkey superior colliculus (SC) neurons differentially process dark 116 stimuli remains unclear. In the mouse SC, the majority of superficial layer neurons prefer 117 dark stimuli (De Franceschi and Solomon, 2018), consistent with the RF subfield structure of 118 these neurons (Wang et al., 2010). Yet, it is not clear whether such dark preference still 119 exists in the deeper SC layers, and whether it is accompanied by differences in visual 120 response latencies of the neurons. Moreover, differences in the ecological environments and 121 neuroanatomical organizations of mice and other species do not trivially predict how 122 primate SC neurons might behave with respect to luminance contrast polarity. Therefore, we 123 exhaustively characterized all visually-responsive rhesus macaque monkey SC neurons (that 124 is, also including intermediate and deeper layer neurons). We were particularly motivated by 125 our recent observations of differential effects of luminance contrast polarity on 126 microsaccades . 127 128  In contrast to LGN, V1, and SC results from other species, we did not find a dominant  129  preference for dark stimuli in the primate SC. Rather, there was a large diversity, with  130  approximately half of the neurons being more sensitive to bright stimuli. Despite such large  131 diversity, what we did find was that the great majority of SC neurons had significantly 132 shorter visual response latencies to dark stimuli. Thus, there was a strong dissociation 133 between visual response latency and visual response sensitivity, reminiscent of a similar 134 dissociation that we observed in the case of spatial frequency tuning (Chen et al., 2018). 135 Such a dissociation was sufficient to account for at least some saccadic reaction time 136 dependencies on stimulus luminance polarity in our experiments. 137 138

140 141
Experimental animals and ethics approvals 142 We recorded superior colliculus (SC) neural activity from two adult, male rhesus macaque 143 monkeys (M and A) aged 9 and 10 years, and weighing 9.5 kg and 10 kg, respectively. We 144 also measured saccadic reaction times from the same two animals plus a third one (F; aged 145 11 years and weighing 14 kg). The experiments were approved by ethics committees at the 146 regional governmental offices of the city of Tübingen. 147 148 149 Laboratory setup and animal preparation 150 The experiments were conducted in the same laboratory as that described in our recent 151 studies (Bogadhi et al., 2020;Bogadhi and Hafed, 2022 Behavioral tasks 173 For the recording data in monkeys M and A, we employed a gaze fixation task in which we 174 presented static disc of 0.51 deg radius within the visual response field (RF) of a recorded 175 neuron. Each trial started with the onset of a black (0.11 cd/m 2 ) fixation spot at screen 176 center. After 550-800 ms of stable fixation on the spot, the disc appeared and remained on 177 for at least ~500 ms. In each trial, the disc could have a Weber contrast of 5%, 10%, 20%, 178 50%, or 100%. We defined Weber contrast as |Is-Ib|/Ib, where Is is the disc's luminance value 179 and Ib is the gray background's luminance value. We often described the contrast as a 180 percentage for convenience (e.g. 5% contrast). Importantly, across trials, the disc could have 181 either positive or negative luminance polarity relative to the gray background, meaning that 182 Is could be either higher (positive polarity) or lower (negative polarity) than Ib. The gray 183 background had a luminance (Ib) of 25.09 cd/m 2 . We collected approximately 50 trials per 184 condition per neuron. 185 186 For some neurons in both monkeys (sometimes in the very same sessions as in the above 187 task), we also ran an immediate orienting version of the stimulus polarity task. That is, at the 188 time of disc onset, we extinguished the fixation spot (which was now white instead of black) 189 simultaneously. This instructed the monkeys to generate an immediate orienting saccade 190 towards the disc. We used this task to confirm that initial visual responses in the main task 191 above were not dictated by the black fixation spot at display center, since the current task 192 had a white fixation spot and showed similar observations (see Results), and also to obtain 193 saccadic reaction time data for additional behavioral analyses (see below). Also note that, 194 for neurophysiological analysis purposes, we only analyzed the initial visual response in this 195 task. Saccade-related responses were deferred to another unrelated project focusing on SC 196 motor bursts, and they are not described here. Finally, to reduce trial counts in this task, we 197 only tested three contrast levels (10%, 50%, and 100% Eye movement data analysis 241 We detected saccades and microsaccades as described previously ( To identify a response saccade and subsequently analyze its reaction time, we required that 252 it had a latency of 50 to 500 ms from stimulus onset, and that it was directed towards the 253 stimulus (this latter criterion was easy to achieve because we used computer-controlled 254 reward windows around the target to allow rewarding the monkeys based on successful 255 saccade generation towards the target). In all neural and behavioral analyses, we also 256 excluded trials with blinks or other movement artifacts near stimulus onset. Statistically, we 257 were interested in whether contrast or luminance polarity affected saccadic reaction times. 258 Therefore, we performed a 1-way non-parametric ANOVA ( Neural data analysis 266 We analyzed a total of 221 SC neurons (109 from monkey M and 112 from monkey A) from 267 the fixation task. We also analyzed 225 neurons from the immediate saccade version of the 268 task (113 from monkey M and 112 from monkey A). Ninety of the neurons in the second task 269 (all in monkey A) were also recorded from the fixation variant of the task. 270 271 The bulk of our analyses was on neurons exhibiting a positive visual response to stimulus 272 onset (that is, an increase in firing rate relative to baseline shortly after stimulus onset baseline, rather than an increase. We performed analyses of these neurons as well, from the 294 fixation task only. To assess the neurons as having a transient decrease in firing rate that was 295 time-locked to stimulus onset, we repeated the same procedure above, but we now checked 296 for a statistically significant decrease in firing rate in the response interval, rather than an 297 increase. We analyzed 15 neurons with transiently decreasing firing rates immediately after 298 stimulus onset (9 from monkey M and 6 from monkey A). 299 300 To obtain contrast sensitivity curves from the neurons with visual bursts, we measured the 301 peak value of the average firing rate curve in a response interval after stimulus onset. used the same measurement intervals for all neurons and also for both positive and negative 308 polarity stimuli. Even though we found a difference in response latency between positive 309 and negative polarity stimuli (as described in Results below), our measurement intervals 310 were large enough to encompass (and exceed) any such latency differences. Therefore, our 311 estimates of contrast sensitivity for brights and darks were not biased by using similar 312 measurement intervals for both types of stimuli (especially because we were searching for 313 only the peak firing rate). After measuring firing rates in the above intervals for each 314 contrast, we plotted the measured firing rates as a function of absolute contrast. We then fit 315 contrast sensitivity curves using the following equation: 316 317 where f is the estimated firing rate, c is stimulus contrast, C50 is semi-saturation contrast, R 320 is the dynamic range of the response, N is the sensitivity/slope of the contrast sensitivity 321 curve, and B is the baseline firing rate (which we just measured across all trials from the 322 same baseline interval mentioned above; final 50 ms before stimulus onset). We then 323 compared the fit parameters R, C50, and N for either bright or dark stimuli to assess whether 324 there were differences in contrast sensitivity between them in the SC. We did this by 325 computing parameter modulation indices as a function of luminance contrast polarity. For 326 example, to compare how R was modulated by luminance polarity, we calculated the R 327 parameter for bright stimuli minus the R parameter for dark stimuli, and we divided this 328 difference by the sum of R values for bright and dark stimuli. This gave us a value between -1 329 and +1. We then plotted histograms of parameter modulation indices across the population. 330

331
For assessing the time course of changes of contrast sensitivity in the sustained interval long 332 after stimulus onset (that is, after the initial visual burst), we again obtained fits of equation 333 1 but now based on measurements after the initial visual burst. To do so, we exploited the 334 fact that our firing rate estimates were already averaging across time (because of a 335 convolution of spike times with a gaussian of s 40 ms). Therefore, for each time after 80 ms 336 after stimulus onset and before 220 ms, we used instantaneous average firing rate as a 337 measure to input to the fit of equation 1 for a given contrast. This allowed us to obtain time 338 courses of semi-saturation contrast (C50), sensitivity/slope (N), and dynamic range (R) during 339 the sustained interval long after stimulus onset. Note that even though the intervals that we 340 chose for the sustained response analysis slightly overlapped with the intervals that we 341 picked up for the initial visual burst analyses mentioned above, the latter analyses were 342 performed on the peak values of the average firing rates, which definitely belonged to the 343 earlier phases of the neural responses (and typically occurred earlier than 80 ms); that is, the 344 initial visual burst intervals were just ranges meant to catch the peak response. Also note 345 that the above contrast sensitivity fits were only performed on the fixation version of the 346 task because we could obtain a longer period of sustained response than in the immediate 347 saccade version of the task. 348 349 For estimating visual response latency in both tasks, we measured the firing rate of a given 350 neuron in a baseline interval (final 50 ms before stimulus onset) across all trials. Then, for 351 each condition (e.g. bright luminance polarity; 20% contrast), we marched forward in time 352 after stimulus onset until 300 ms (typically, the algorithm converged on a visual burst much 353 earlier, of course). As soon as the lower bound of the 95% confidence interval around the 354 average firing rate of the neuron across trial repetitions exceeded the average baseline 355 activity (and continued to do so for at least 30 ms), we flagged the time as the response 356 latency of the neuron. Whenever this algorithm failed to detect the response latency in at 357 least one of the luminance polarities for a given stimulus contrast level, we excluded the 358 neuron from further analysis in that contrast level. This explains the varying numbers of 359 neurons reported in some figures (e.g. the different panels of Fig. 3 in Results). We then 360 compared such response latency across contrasts and stimulus polarities. Note that we 361 focused on relative latency differences across luminance polarities in our analyses. This is 362 important to note because firing rate estimates (in our case, convolution of spike times with 363 a Gaussian kernel) necessarily blurs the exact response onset times of the neurons. 364 However, our approach of estimating response latencies described above still captured the 365 latency differences that we were interested in documenting, and it simplified the detection 366 of visual response latencies for neurons with non-zero baseline firing rates. 367 368 To statistically test for differences in latencies between luminance polarities at a given 369 contrast level, we used non-parametric permutation tests on the pairwise mean latency 370 differences, with 10000 permutations. That is, we obtained the permutation distribution by 371 shuffling the polarity labels of the latencies for 10000 times while maintaining their pairwise 372 relationship and calculating their pairwise difference. Monte Carlo p-values were obtained 373 by assessing the probability of getting larger than or equal to absolute latency differences in 374 the permutation distribution than the absolute latency difference of the original data. We 375 ran the tests separately for each monkey to ensure that our pooling of data in figures for 376 visualization purposes was justified. 377 378 We also used a similar approach to test for statistically significant effects of upper versus 379 lower visual field RF location on the latency differences between luminance polarities. This 380 time, we obtained the latency differences between the responses to bright and dark stimuli, 381 and then subtracted this measurement for the upper visual field neurons from the 382 measurement for the lower visual field neurons. We defined lower and upper visual field 383 neurons based on the location of the stimulus (which was placed close to the location of the 384 RF hotspot location). Thus, negative values in the final measurement would indicate a larger 385 difference between dark and bright stimuli in the upper visual field than in the lower visual 386 field. After that, we ran permutation tests by shuffling the labels of the upper and lower 387 visual field neurons for 10000 times. To assess the significance of the results, we calculated 388 the Monte Carlo p-value. The same procedure was applied to assess the absolute values of 389 sensitivity differences (see next paragraph) between dark and bright stimuli in the upper and 390 lower visual fields. 391 392 To compare visual response latency to sensitivity, we also measured peak firing rate in the 393 initial visual response interval (as defined for each contrast above) of the neuron. We then 394 checked whether there was a dissociation between response latency and sensitivity (i.e. 395 response strength) for black and white stimuli, as we previously saw for spatial frequency 396 stimuli (Chen et al., 2018). We did so by sorting the neurons according to the difference in 397 response latencies between brights and darks, and then checking whether the same sorting 398 applied to the difference in response sensitivities (see Results). 399 400 For the immediate saccade version of the task, we only analyzed initial visual bursts (50-130 401 ms after stimulus onset) and not sustained intervals. This was because the response saccade 402 occurred too soon after the initial visual bursts. We assessed both response sensitivity and 403 response latency (as described above) to confirm that we got similar results to those from 404 the fixation task. 405 406 For the neurons with transient decreases in firing rate, we assessed response latency in a 407 similar way to the neurons with visual bursts, but we looked for statistically significant 408 decreases in firing rate after stimulus onset, rather than increases. 409 410 In some figures, for illustration and visualization purposes, we elected to show example 411 population firing rates from individual monkeys. For example, we did this in Fig. 6A, B in 412 Results. To obtain such population firing rates, we obtained the normalized average firing 413 rate of each neuron, per monkey and condition. That is, for each neuron, we found the peak 414 visual response in the interval 0-100 ms after stimulus onset for the 100% contrast stimuli, 415 regardless of the stimulus polarity. Then, for each contrast and polarity, we normalized the 416 neuron's average firing rate by that peak visual response value. This resulted in a series of 417 average normalized curves for the neuron across conditions. After that, we averaged all of 418 the normalized firing rate curves of each monkey's neurons in a given condition. This gave us 419 a population summary of responses, maintaining the relative changes in responses across 420 conditions. We used a similar approach in Fig. 9B Experimental design and statistical analyses 424 We recorded neurons in an unbiased manner by collecting data in parallel (with linear 425 electrode arrays) in most sessions and then sorting the neurons offline. This allowed us to 426 minimize sampling bias. In each variant of the task, we also analyzed >80 neurons per 427 monkey. This provided a large enough sample to assess the reliability of our interpretations. 428 Within each neuron, we ensured collecting approximately 35-50 repetitions per condition 429 (after filtering out bad trials and so on) to allow robust within-neuron statistics. Similarly, in 430 our behavioral analyses, we collected thousands of saccades. In all cases, we randomly 431 interleaved stimulus presentations across trials, to avoid any blocking effects. 432 433 We showed the full distributions of data points that we had. 436 437 Since the replicate of interest was neurons, our numbers of sampled neurons were 438 sufficient. The use of two monkeys in recording was valuable to increase neuron counts, and 439 to also demonstrate repeatability across individuals. Our results were highly similar in the 440 two animals (e.g. Fig. 6A, B in Results). When they did differ, we showed each individual 441 monkey's results separately (e.g. Fig. 9 in Results), and this was highly useful for us to 442 interpret the behavioral results. Moreover, we collected behavior from a third monkey 443 exactly to improve our interpretation of our individual monkey behavioral phenomena. 444 445 All statistical tests are reported and justified in Results at appropriate points in the text. As 446 stated above, we statistically analyzed each monkey's data individually, confirming that each 447 monkey showed the same effects (unless otherwise stated; for example, in Fig. 9 in Results). 448 449

451
We investigated how monkey superior colliculus (SC) neurons respond to dark and bright 452 visual stimuli. In our primary task, the monkeys fixated while we presented a small disc that 453 was either higher or lower in luminance than the gray background of the display. We varied 454 the contrast of the disc from the background luminance, and we assessed contrast 455 sensitivity curves separately for positive and negative luminance polarities. We first analyzed 456 the neurons that exhibited visual bursts (that is, increases in firing rate) after stimulus onset, 457 and we investigated visual burst strength, visual burst latency, as well as sustained response 458 dynamics for dark and bright stimuli. The results for these neurons are described next, 459 followed by an analysis of a smaller number of neurons for which stimulus onsets caused 460 transient decreases in firing rates, rather than increases. 461 462 463 Diverse preferences for darks and brights across SC neurons 464 We first asked whether neurons tended to be more sensitive to darks or brights across the 465 population. For each recorded neuron, we plotted firing rate as a function of time from 466 stimulus onset, and we assessed the strength of the visual burst as a function of luminance 467 contrast polarity. Figure 1A-C shows the responses of three example neurons (from the 468 same monkey, A) to a 100% contrast stimulus. The black lines show responses to the 469 negative polarity stimulus (darker than background), and the light gray lines show responses 470 to the positive polarity stimulus (brighter than background). In all cases, the negative and 471 positive polarity stimuli were of the same size and presented at the same location. They also 472 had the same absolute Weber contrast, and their presentation sequence was randomly 473 counterbalanced across trials. As can be seen, there was a diversity of neural preferences 474 across the three neurons: neuron 1 (Fig. 1A) was more sensitive to the positive polarity 475 stimulus than to the negative polarity stimulus; neuron 2 was, more or less, equally sensitive 476 to the two stimuli (Fig. 1B); and neuron 3 was clearly more sensitive to the dark stimulus 477 (Fig. 1C). We also plotted full contrast sensitivity curves for the same neurons ( Fig. 1D-F   optimizing three parameters characterizing how the neuron altered its visual response with 507 Weber contrast: R reflected the dynamic range of the response, C50 characterized the semi-508 saturation contrast of the neuron, and N characterized the steepness of the contrast 509 sensitivity curve (slope parameter). We performed such a fit for either positive or negative 510 luminance polarity stimuli. We then obtained a parameter modulation index, describing, for 511 each neuron, to what extent each parameter of the fit was different between positive and 512 negative luminance polarity stimuli. For example, for dynamic range (parameter R in 513 equation 1), we obtained the R value for bright stimuli minus the R value for dark stimuli in 514 each neuron, and we then divided this difference by the sum of R values for the two stimulus 515 types (Methods). This gave us an index in which 1 meant that the neuron responded 516 maximally only to bright stimuli and -1 meant that the neuron responded maximally only to 517 dark stimuli. An R parameter modulation index value of 0, instead, indicated equal visual 518 response dynamic ranges for bright and dark stimuli. We then plotted histograms of the 519 parameter modulation indices across the population. As can be seen in Fig. 2, all three 520 modulation indices of the contrast sensitivity function fits had distributions straddling 0, and 521 with large diversity across the population. Some neurons clearly preferred bright stimuli, 522 others clearly preferred dark stimuli, and yet others were equally sensitive to darks and 523 brights (near 0 in the histograms of Fig. 2). The vertical lines in Fig. 2

553 554
Earlier detection of darks by SC neurons, regardless of preference 555 Unlike response sensitivity, for which we saw diverse preferences for brights and darks (Figs.  556 1, 2), SC neurons exhibited systematically shorter visual response latencies for dark stimuli, 557 independently of their visual response strengths. Consider, for example, the same three 558 neurons of Fig. 1A-C. In each of them, visual responses occurred earlier for the dark stimuli 559 than for the bright stimuli, as can be visually assessed from the spike rasters and the firing 560 rate density plots below them. This happened even for neuron 1, which clearly preferred 561 bright stimuli (Fig. 1A). Thus, there was a dissociation between visual response sensitivity 562 and visual response latency in the neuron, not unlike what we recently observed when we 563 presented different spatial frequencies to SC neurons (Chen et al., 2018). 564 565 To investigate this dissociation further, we estimated, for each neuron, the onset of the 566 visual burst as the first time point at which the lower bound of the 95% confidence interval 567 of the neuron's average firing rate was elevated for a prolonged period of time above its 568 baseline activity (Methods). Even though estimating neural response latencies from firing 569 rate measures like we did might blur the actual absolute values of the response latencies, 570 due to convolution kernels with spike times, this approach was still sufficient to capture the 571 latency differences across luminance polarities that we were interested in (Methods). 572 Therefore, for each contrast, we subtracted each neuron's visual response latency for dark 573 stimuli from its visual response latency for bright stimuli, and we sorted the neurons 574 according to this difference. An example of such sorting can be seen in the top panel of  We also made similar observations for the other stimulus contrasts that we tested ( Fig. 3B-611 E). Note that for each panel in Fig. 3, we indicated the total number of neurons included into 612 each analysis, which varied across panels (that is, across contrast levels). This happened 613 because some neurons may not have met our inclusion criteria for estimating visual 614 response latencies, resulting in slightly different neuron counts across the different panels 615 (Methods). For example, for the particularly low contrast stimuli (e.g. 5% and 10%), some 616 neurons did not even exhibit any significant visual bursts at all (Methods), so they were 617 obviously not included in the figure. Having said that, in all contrasts, there was a majority of 618 neurons responding earlier to dark than to bright stimuli (top row in each panel of Fig. 3 Of course, our results do not deny that high visual response sensitivity is normally associated 635 with short visual response latencies. For example, with the sorting of neurons shown in Fig. 3  636 based on their response latency differences (top row), there was still a hint of an additional 637 trend: neurons with a smaller latency difference between dark and bright stimuli tended to 638 be the neurons preferring bright stimuli (bottom row). For example, compare the first and 639 last quartiles in the bottom panel of Fig. 3A: more bright-preferring neurons occurred in the 640 first quartile (having 0 or negative latency differences) than in the last quartile (having 641 positive latency differences). This suggests that there were divergent forces influencing 642 visual response latency: a neuron strongly preferring bright stimuli might have had its high 643 response strength for bright stimuli counterbalance the normally earlier detection of dark 644 stimuli. Indeed, when we evaluated response latency as a function of both stimulus contrast 645 (a proxy for visual response strength in the neurons) and luminance polarity, we found that 646 both factors clearly influenced the neurons' visual response latencies. This is shown in Fig.  647 4A, B for an example neuron, and in Fig. 4C for the population. In also that visual response latencies were always systematically shorter, at a given contrast, 656 for dark than bright stimuli (consistent with Fig. 3). This effect was the largest in magnitude 657 for high contrast stimuli; response latencies were anyway increased with the weaker low 658 contrast stimuli, making differences between dark and bright stimuli milder. Therefore, both 659 stimulus contrast (a proxy for response sensitivity) and stimulus polarity (conferring a 660 temporal advantage for darks) dictated our SC neurons' visual response latencies. 661 662 The above results indicate that there is faster detection of dark than bright stimuli by SC 663 neurons, in general. However, it is also known that SC visual responses preferentially process 664 the upper visual field (Hafed and Chen, 2016), consistent with the notion that eye 665 movements support orienting towards or away from extra-personal stimuli largely occupying 666 the upper visual field (Previc, 1990). If that is indeed the case, then it might be expected that 667 differential temporal processing of dark versus bright stimuli might be magnified in the SC's 668 upper visual field representation. For example, birds of prey, or other threats, across a 669 daylight sky would normally cast dark contrasts on retinal images, and they need to be 670 detected efficiently by SC neurons. We, therefore, next asked whether the results of Fig. 3  671 could depend on the visual field locations of our recorded neurons.

724
Interestingly, the lower panel shows that the absolute value of visual response sensitivity difference (between 725 brights and darks) was also higher for upper visual field neurons than for lower visual field neurons (compare the 726 dynamic ranges of the two curves). Therefore, both visual response latency and visual response sensitivity 727 effects, in terms of luminance contrast polarity, were magnified in the upper visual field.

729 730 731
Note also that the same dissociation between response latency and response sensitivity 732 occurred in Fig. 5 as in Fig. 3: the bottom panel in Fig. 5 shows that with the same sorting of 733 neurons as in the top panel, response sensitivity was not systematically sorted in either the 734 upper or lower visual fields, consistent with the results of Fig. 3 Finally, we wondered whether the black fixation spot at display center (Methods) might have 746 dictated our results above. Such an effect would be unlikely because our neurons were 747 extra-foveal; our stimuli were, therefore, generally far from the fixation spot (Methods). 748 However, to unambiguously rule such an effect out, we repeated the same experiment with 749 two slight modifications. First, the fixation spot was now white instead of black (Methods). If 750 the black fixation spot was the reason for the faster detection of dark stimuli in the results of 751 Figs. 1, 3-5 above, then this effect should be altered with a white fixation spot. Second, the 752 fixation spot was now removed at the same time as stimulus onset, allowing the monkeys to 753 generate immediate, visually-guided saccades. We analyzed 213 neurons recorded with this 754 variant of the task (81 were also recorded in the original fixation task). We will describe 755 saccadic reaction times as a function of contrast and luminance polarity in more detail 756 below. However, for now, our aim was to replicate the visual burst results shown above. For 757 each neuron, we normalized the neuron's average firing rate by the peak visual response for 758 100% contrast stimuli in the interval 0-100 ms after stimulus onset (Methods). We then 759 averaged all of the normalized firing rate curves of each monkey's neurons (Fig. 6A, B). We 760 separated the neurons of each monkey in this analysis to demonstrate the repeatability of 761 our results across the animals, and also because subsequent saccadic behavior later in the 762 trials differed between them, as we clarify in more detail below. 763 764 Both animals had clear visual responses in the task, consistent with the results of the fixation 765 variant (Figs. 1, 3-5). Most importantly, these responses were also clearly still occurring 766 earlier for dark stimuli than for bright stimuli (Fig. 6A, B). To summarize these results on an 767 individual neuron basis, we replicated the same analyses of Fig. 3 (Fig. 6C-E; we combined 768 the neurons of both monkeys here because of how similarly they behaved in Fig. 6A, B). We 769 did this for all three stimulus contrast levels that we tested in this variant of the task 770 (Methods). For all contrasts, most neurons still detected dark contrasts earlier than bright 771 contrasts, irrespective of neural sensitivity (Fig. 6C-E  Different temporal dynamics of firing rates in the sustained interval for darks 804 and brights 805 The results so far have focused on initial visual bursts. However, with prolonged fixation (as 806 in our primary task of Figs. 1-5), we also observed significant differences in SC neural 807 response dynamics in the sustained interval (long after stimulus onset) for bright and dark 808 stimuli. In particular, bright stimuli were generally associated with secondary elevations of 809 firing rate above those of dark stimuli. To illustrate this, Fig. 7A, B shows the responses of 810 four example neurons to high contrast stimuli (100%). Two neurons are from monkey A (Fig.  811 7A), and two neurons are from monkey M (Fig. 7B). In the first three neurons (neurons 5-7 in 812 Fig. 7A, B), after the initial visual bursts, bright stimuli evoked stronger sustained activity 813 than dark stimuli (see pink intervals highlighting the sustained interval). The stimuli were still 814 present within the RF's of the neurons in all cases, but there was an altered response 815 dynamic after the initial visual bursts, particularly for bright stimuli. Even the fourth neuron 816 (neuron 8 in Fig. 7B), which showed relatively weak sustained activity, still showed a 817 subdued secondary peak in firing rate after the initial visual burst for bright stimuli (also see 818 To characterize this altered dynamic of neural responses as a function of time in more detail, 842 we took each firing rate curve after 80 ms from stimulus onset (that is, after the initial visual 843 bursts). We then estimated contrast sensitivity curves at each time point. Each time sample 844 of a firing rate curve is already a kind of average over some discrete measurement interval 845 (due to the convolution of spike times with a gaussian kernel to generate firing rates). 846 Therefore, we took each sample of the firing rate curve of a neuron in the sustained interval, 847 and we used it to fit contrast sensitivity curves from equation 1 at each time point. This gave 848 us a series of contrast sensitivity curves as a function of time. We then plotted the time 849 courses of the parameters R, C50, and N of the fits during the sustained interval, and we did 850 this for either bright or dark stimuli. The results across the entire population of neurons are 851 shown in Fig. 7C. As can be seen, all parameters were varying differently between darks and 852 brights in the interval around approximately 100-200 ms after stimulus onset (that is, during 853 sustained visual response intervals), consistent with the example neurons of Fig. 7A, B. The 854 biggest effect was in the R parameter, which was stronger for brights than darks, suggesting 855 higher sustained firing rates for brights after the ends of the initial visual bursts. Both C50 856 and N gradually changed in a manner that was consistent with higher thresholds and 857 shallower contrast sensitivity functions in the sustained interval. That is, the contrast 858 sensitivity of the neurons was generally the highest in the initial visual burst intervals, and it 859 gradually degraded in sustained intervals (other than the R parameter elevation for bright 860 stimuli). This makes sense given that sustained intervals were generally associated with 861 much lower firing rates than in the initial visual burst intervals. In any case, during the 862 sustained interval, and unlike in the initial phases of SC visual responses, there was a 863 generalized elevation of firing rates for bright stimuli compared to dark stimuli for all 864 contrasts. As we show later, this effect was strong enough in monkey M, to the extent that it 865 appeared to dominate this monkey's saccadic reaction time patterns in the immediate, 866 visually-guided saccade version of the task. 867 868 869 Earlier detection of dark stimuli also by inhibited neurons 870 In all of the above analyses, we focused solely on neurons exhibiting positive visual 871 responses (that is, increases in firing rates above baseline). However, with our offline sorting 872 pipelines (Methods), we also sorted a fewer number of neurons that exhibited transient 873 decreases in activity after stimulus onset rather than increases. These neurons were 874 obtained from similar recording sites to those from which we isolated neurons with visual 875 bursts (we used linear electrode arrays primarily orthogonal to the SC surface; Methods). 876 The neurons were, therefore, from similar topographic locations to those associated with 877 the neurons reported in Figs. 1-7. When we analyzed these inhibited neurons in more detail, 878 we found that their transient, stimulus-induced decreases in firing rates were still sensitive 879 to luminance polarity. For example, in Fig. 8A, B, we show two sample neurons from one of 880 our electrode penetrations in monkey M. The two neurons showed classic visual responses 881 (to black stimuli); moreover, their RF's (shown in the insets for data collected with the 882 presentation of black small spots during fixation) were spatially localized and overlapping 883 with each other. From the very same electrode penetration, Fig. 8C shows a third sample 884 neuron that was recorded simultaneously with the two other neurons; it was thus in the 885 same SC topographic region as the two neurons of Fig. 8A, B. The neuron of Fig. 8C was  886 inhibited instead. Most interestingly, this neuron clearly "responded" to a high contrast dark 887 stimulus earlier than to a bright stimulus of the same contrast, with the only difference from 888 the results of Figs. 1-7 being that the response in this case was a transient reduction from 889 baseline activity rather than an increase. Across the population of such inhibited neurons 890 (n=15 neurons), we repeated the same latency analyses of Fig. 3 above. That is, we assessed 891 the relative time of "response" between bright and dark contrasts (Methods). As can be 892 seen from Fig. 8D, the majority of such inhibited neurons also reacted to dark stimuli earlier 893 than to bright stimuli, just like with the neurons possessing visual bursts. Similar 894 observations were also made for the lower contrast stimuli. Therefore, even inhibited 895 neurons in the SC detected dark contrasts faster than bright contrasts. 896 897 898 899 900  Saccadic reaction times can be significantly shorter for dark stimuli 916 Finally, prior work has demonstrated a tight relationship between SC visual response 917 properties and saccadic reaction times ( predict such reaction times. Therefore, given the faster response latencies of SC neurons for 922 dark stimuli that we found, we wondered whether this effect was sufficient to be associated 923 with faster saccadic reaction times to such stimuli (even when response sensitivity was, on 924 average, similar for darks and brights, as shown in Fig. 2). We tested our two monkeys and a 925 third one on the immediate visually-guided saccade task; we used the same sessions as in 926 Fig. 6 for monkeys M and A, and we ran separate behavior-only sessions for monkey F. The 927 monkeys simply generated a saccade as soon as the target appeared (the fixation spot also 928 disappeared at target onset, as mentioned above for Fig. 6  exploring the effect of contrast on reaction time, when collapsing across luminance 935 polarities). Interestingly, two out of the three monkeys (A and F) also showed consistently 936 faster reaction times for the darker stimuli, like with the SC visual bursts. These results are 937 shown in Fig. 9A, C, E. As can be seen, monkeys A and F were faster to react to dark stimuli 938 at all contrasts. Monkey M, on the other hand, had faster reaction times for the bright 939 stimuli (Fig. 9C). All of these results (of luminance polarity effects on reaction times) were 940 significant (p < 2.5x10 -8 in each monkey individually, Kruskal-Wallis test exploring the 941 relationship between luminance polarity and reaction time, when collapsing across 942 contrasts); the effect sizes (shown for each condition in Fig. 9A, C, E) were also substantial. 943 944 We were particularly intrigued by the discrepancy in the reaction times of monkey M with 945 respect to dark and bright stimuli. On the one hand, it might suggest that SC visual response 946 latency (e.g. Fig. 3) is not the only determinant of saccadic reaction times, which is indeed 947 plausible. For example, we earlier found that SC visual response latency and visual response 948 sensitivity together provided a better correlate of reaction times than either parameter 949 alone (Chen et al., 2018). Therefore, since about half of the neurons in our population were 950 more sensitive to bright stimuli anyway (Fig. 2), despite the faster detection of darks, it could 951 be that this particular monkey's reaction times were more dictated by SC visual response 952 sensitivity than by visual response latency. On the other hand, it could additionally be the 953 case that the later elevation of responses for bright stimuli that we saw in Fig. 7 was more 954 pronounced in this monkey, potentially suggesting stronger top-down control for bright 955 stimuli. In that case, bright stimuli could be preferentially processed by this monkey. Indeed, 956 in a previous behavioral study in which we investigated the properties of saccadic inhibition 957 as a function of luminance contrast polarity, this monkey reacted differently to full field 958 white versus black visual flashes from the two other monkeys in the very initial oculomotor 959 response to flash onset, again seeming to react faster for bright than dark flashes (Malevich 960 et al., 2021) (see their Fig. 3). Therefore, we decided to check how this monkey's neurons, in 961 particular, reacted to white stimuli long after the initial visual bursts, and we were able to do 962 so from our fixation variant of the task. 963 964 We plotted each monkey's population visual responses for dark and bright stimuli in the 965 fixation variant of the task, allowing us to explore the longer sustained interval. These results 966 are shown in Fig. 9B, D. Even though monkey M's neurons still detected dark stimuli earlier 967 than bright stimuli in the initial visual response period (consistent with all of our results 968 shown earlier), this monkey's elevation of sustained visual activity for the bright stimuli (e.g. 969 Fig. 7) was particularly pronounced when compared to monkey A (note the secondary peak 970 in population firing rate for bright stimuli in Fig. 9D for monkey M, which was stronger than 971 the same peak in monkey A). We also even saw hints of this secondary elevation in Fig. 6B in 972 the immediate, visually-guided saccade variant of the task, with a sharper elevation for 973 bright stimuli right after the initial visual burst and leading up to the saccade-related burst; 974 however, of course, in this task, this sharper elevation for brights was harder to properly 975 analyze in the saccade task because of how quickly the motor burst came.  Therefore, the results of Fig. 9 suggest that saccadic reaction times can indeed be faster for 1001 dark than bright stimuli, consistent with the faster detection of dark stimuli by SC neurons, 1002 and that even violations of such an observation (as in the case of monkey M) are still related 1003 to the SC visual responses (in this case, the sustained responses after the initial visual bursts 1004 subside). 1005 1006 In all, our results in this study indicate that SC neurons robustly detect dark stimuli faster 1007 than bright stimuli; that sustained visual responses in the SC instead favor bright stimuli; and 1008 that saccadic reaction times can reflect the faster detection of dark stimuli in the SC's initial 1009 visual bursts and/or the later elevation for bright stimuli. 1010 1011 1012