Muscarinic Lateral Excitation Contributes to Visual Object Segmentation during Collision Avoidance

Visual neurons specialized in tracking objects on a collision course are often finely tuned to their target stimuli as this is critical for survival. The presynaptic neural networks converging on these neurons and their role in tuning them remains poorly understood. We took advantage of well-known characteristics of one such neuron to investigate the properties of its presynaptic input network. We find a structure more complex than hitherto realized. In addition to dynamic lateral inhibition used to filter out background motion, presynaptic circuits include normalizing inhibition and short-range lateral excitatory interactions mediated by muscarinic acetylcholine receptors. These interactions preferentially boost responses to coherently expanding visual stimuli generated by colliding objects, as opposed to spatially incoherent controls, helping implement object segmentation. Hence, in addition to active dendritic conductances within collision detecting neurons, multiple layers of both inhibitory and excitatory presynaptic connections are needed to finely tune neural circuits for collision detection.


23
Both in vertebrates and invertebrates, collision-detecting neurons are typically located in higher order 24 nuclei and neuropils, several synapses away from the sensory periphery. In fish, for example, collision 25 detecting neurons are found in the optic tectum and the hindbrain (Preuss et al., 2006;Dunn et al., 2016; 26 Temizer et al., 2015). Similarly, in birds they were identified in the nucleus rotundus of the thalamus, a 27 major recipient of ascending tectal projections (Wang and Frost, 1992;Sun and Frost, 1998). In mice and 28 cats, they have been documented in the superior colliculus (Liu et Rind and Simmons, 1992;Hatsopoulos et al., 1995;Schlotterer, 1977). These 34 arrangements leave plenty of room for presynaptic networks to implement through specific patterns of 35 connectivity the neural computations required to shape the responses of collision-detecting neurons.

36
Although we know much about the response properties of collision-detecting neurons in the systems 37 mentioned above, at present the connectivity patterns of their presynaptic networks and their 38 computational roles remain largely unknown. We took advantage of a well-characterized circuit dedicated 39 to collision avoidance in locusts to investigate these issues. LGMD is exquisitely tuned to detect approaching objects on a collision course (e.g., Judge 46 and Rind, 1997;Gray et al. 2001; Gabbiani et al. 2001); a tuning that is in part mediated by active 47 conductances close to its spike initiation zone and within its dendritic tree  48 Dewell and Gabbiani 2017). Since the computation carried out by the LGMD as it tracks approaching 49 objects was first characterized (Hatsopoulos et al. 1995;Gabbiani et al. 1999 Rind et al. 2016). Further, histochemical and immunocytochemical methods demonstrated that these 63 lateral connections are cholinergic and suggested that they might mediate inhibition through muscarinic 64 acetylcholine receptors (mAChRs; Rind and Simmons, 1998;Rind and Leitinger, 2000). However, 65 mAChRs can either be excitatory or inhibitory (Brown 2010) and there is to date no direct experimental 66 evidence to support either of these alternatives for the lateral connections presynaptic to the LGMD.

77
Calcium enters the excitatory dendritic field of the LGMD exclusively through the calcium-permeable 78 nicotinic receptors associated with its afferents ). Thus, in this system calcium imaging 79 offers an accurate means of monitoring activation of the afferents and their lateral interactions. To study 80 the spatiotemporal calcium activation dynamics on the LGMD excitatory dendrites, we used the indicator 81 Oregon Green BAPTA-5N (OGB-5N), whose fast kinetics allowed us to minimize the difference between 82 calcium fluorescence and membrane potential changes. To investigate whether the lateral interactions 83 between TmAs are excitatory or inhibitory, we locally applied the mAChR antagonist scopolamine or the 84 agonist muscarine and tested responses to different visual stimuli. For each stimulus type the lateral 85 interactions proved excitatory. The excitatory nature of lateral connections raises a conundrum: since they 86 amplify an excitatory input that already grows rapidly over the course of object approach, how is this 87 input maintained within the dynamic range of the LGMD? One possibility is for the presynaptic network 88 to rely on global, normalizing inhibition, a feature present in many sensory circuits (e.g., Heeger 1992; 89 Olsen

97
Orderly spatio-temporal activation of LGMD excitatory dendritic field is evoked by looming 98 stimuli.

99
The LGMD's excitatory dendritic field is retinotopically organized ; Zhu and Gabbiani, 100 2016). We therefore started by mapping the dendritic activation pattern elicited by different regions of the 101 display used to generate visual stimuli (Methods). For this purpose, we presented square flashes at 5 102 distinct locations (Fig. 1A, top; Supp. Fig. 1A) and determined the dendritic branches activated by each 103 stimulus (Fig. 1A, bottom). Previous work showed that a single dendritic branch is activated by more than 104 one ommatidium (facet) and that the amount of overlap in dendritic activation decreases with inter-105 ommatidial distance, becoming close to zero for a separation of four ommatidia (Zhu and Gabbiani 2016, 106 Fig. 3). Since each ommatidium is receptive to approximately 2° of visual space when light adapted 107 (Wilson, 1975), we separated the stimuli by 8° to minimize dendritic overlap. Indeed, we found that the 108 branches activated by these stimuli intersected only minimally, as illustrated in Fig. 1A

110
To address this question, we presented looming stimuli with a relatively high half-size (l) to speed (|v|) 111 ratio, l/|v|=120 ms, and thus a long time of approach (Fotowat and Gabbiani, 2007), maximizing our 112 ability to resolve temporally the associated calcium signals (

115
1B shows for one animal the average calcium responses at the center branches (color-coded red in Fig.   116 1A, bottom), and two sets of branches symmetrically surrounding them (blue and green Fig. 1A, bottom).

117
The same data is presented on an expanded time scale in Fig. 1C (middle panel).

118
Unexpectedly given the retinotopic organization of the excitatory dendritic field, we found that the

129
We further analyzed the calcium responses on a finer spatial scale by considering each imaged dendritic 130 branch separately (Fig. 1D, top). We first defined a threshold dF/F activation as 5 times the baseline noise following, we will call this branch the (dendritic) looming center. As predicted from retinotopy, the rise 137 time increased in an orderly manner with the distance (in the imaging plane) of each branch from the 138 looming center, a result consistent across animals (Fig. 1E, bottom). We also found that dF/F integrated 139 over time up to its peak, an indirect measure of cumulative calcium influx, decreased as a branch lied 140 farther from the looming center (Fig. 1F, top). This result was again consistent across animals (

145
Calcium fluorescence changes are usually much slower than membrane potential (V m ) changes due to the 146 buffering properties of calcium dyes and their binding kinetics. In preliminary experiments, we compared 147 the calcium responses to looming stimuli obtained using either OGB-1 or OGB-5N and found OGB-5N to 148 have faster kinetics (Supp. Fig. 2A, B). Next, we checked how closely OGB-5N calcium responses track 149 the subthreshold membrane potential by recording them simultaneously during the presentation of 150 looming stimuli with l/|v| = 120 ms. The calcium signals from individual dendritic branches were 151 analyzed separately ( Fig. 2A) and are plotted together with V m (median filtered to eliminate spikes) in

174
To study the effects of putative muscarinic acetylcholine receptors (mAChRs) located at the presynaptic 175 terminals of the LGMD on looming responses, we compared the activation pattern on the LGMD 176 excitatory dendritic field before and after puffing the agonist muscarine. As illustrated in one example in broadening (p=0.0002, paired t-test) of these curves following muscarine application. Next, we pooled the 188 data across five animals and found that the peak dF/F decreased with increasing distance from the 189 looming center both before and after puffing muscarine (Fig. 3G). Additionally, the magnitude of the 190 slope of the peak dF/F as a function of the distance from the looming center significantly decreased after 191 puffing muscarine.

193
We then compared the activation pattern on the LGMD excitatory dendritic field in response to looming 194 stimuli before and after puffing the muscarinic antagonist scopolamine (Fig. 3H-N legend). The peak dF/F also decreased with increasing distance from the looming center after puffing 201 scopolamine (Fig. 3N). However, we did not observe a significant difference in the slope of the peak dF/F 202 as a function of the distance from the looming center before and after puffing scopolamine (Fig. 3N).

203
Thus, the effects of scopolamine are nearly opposite to those observed when applying muscarine.  (Fig. 4C, bottom). On each selected dendritic branch (Fig. 4D), the amplitude of dF/F in response to a 237 small translating stimulus was stronger than in response to the same stimulus flanked by drifting gratings 238 at a separation of 50°, consistent with activation of lateral inhibition (Fig. 4E, 4F). After puffing 239 scopolamine, the amplitude of the dF/F on each selected dendritic branch decreased (Fig. 4G, 4H).

240
However, the lateral inhibition persisted. We employed flanking gratings with separations of 70, 50, 30 241 and 7.5° and computed the amplitude of dF/F on every selected dendritic branch after puffing 242 scopolamine versus before puffing in five animals (Fig. 4I). The amplitudes of dF/F after puffing 243 scopolamine were significantly smaller than before puffing across the animals (p<10 -9 for all stimuli 244 types, paired t-test). The time at peak dF/F before and after puffing scopolamine did not significantly 245 6 change for two separations (30° and 50°) while in the remaining three cases the change was small (-1.1% 246 change at 70°; -4.2% at 7.5° and -1.8% for no grating; Fig. 4J). The mean amplitudes of dF/F for all the 247 branches in each animal decreased with closer flanking gratings, both before and after puffing 248 scopolamine (Fig. 4K). However, the subtraction of the mean amplitudes of dF/F before puffing 249 scopolamine from that after puffing showed no statistical difference in response to all five different visual 250 stimuli ( Fig. 4L and legend). This suggests that blocking mAChRs presynaptic to the LGMD with 251 scopolamine did not influence dynamic lateral inhibition.
252 253 Next, we tested the influence of muscarine on dynamic lateral inhibition using the same protocol as 254 above. In response to the small translational stimulus, activation on the LGMD excitatory dendrites 255 increased after puffing muscarine (Supp. Fig. 4A). On each selected dendritic branch (Supp. Fig. 4B), we 256 compared the amplitude of dF/F in response to the small translational stimulus without and with the 257 drifting gratings at a separation of 50° (Supp. Fig. 4C and D, respectively). We found that the overall 258 amplitude of dF/F was reduced in presence of the gratings, consistent with the activation of presynaptic 259 lateral inhibition. After puffing muscarine, the amplitude of the dF/F on each selected dendritic branch 260 increased (Supp. Fig. 4E and F). However, the responses with gratings were still reduced compared to the 261 no grating condition. Thus, lateral inhibition appeared to be unaffected by activation of the mAChRs in 262 presynaptic terminals to the LGMD. We repeated this experiment across five animals, using gratings with 263 separations of 70, 50, 30 and 7.5° and plotted the amplitude of dF/F on every selected dendritic branch 264 after puffing muscarine versus before puffing (Supp. Fig. 4G). The amplitudes of dF/F after puffing 265 muscarine were slightly but not significantly bigger than before puffing across the animals. The time at 266 peak dF/F before and after puffing muscarine did not change (Supp. Fig. 4H). The mean amplitudes of 267 dF/F across all the branches in each animal decreased as the distance between the drifting gratings 268 decreased both before and after puffing muscarine, the characteristic signature of lateral inhibition (Supp. 269 Fig. 4I). However, subtraction of the mean amplitudes of dF/F before puffing muscarine from that after 270 puffing showed no statistical difference between the five different visual stimulus conditions (Supp. Fig.   271 4J). Thus, muscarine like scopolamine did not affect dynamic lateral inhibition.

273
Taken together, lateral interactions between transmedullary afferents through mAChRs at presynaptic 274 terminals onto the LGMD do not appear to generate dynamic lateral inhibition.

276
Horizontal band-restricted looming stimuli activate lateral branches better than looming stimuli.

277
To further study the dendritic activation pattern elicited by looming stimuli, we designed looming stimuli 278 restricted to horizontal (along the dorsal-ventral eye axis due to the 90° head rotation performed during 279 the dissection; Methods) or vertical (along the posterior-anterior eye axis) bands with a width of 8° (Fig.   280 5A). As expected from the retinotopic mapping, the activation width of the vertical band-restricted 281 looming response (Fig. 5A, right) along the first principal axis (magenta line) was narrower than that of 282 the standard looming response (Fig. 5A, left). Surprisingly, we found that the activation width of the 283 horizontal band-restricted looming response (Fig. 5A, middle) was wider than that of the standard 284 looming response (Fig. 5A, left). On each selected dendritic branch (Fig. 5B), we compared the calcium 285 responses to the standard looming stimulus (Fig. 5C, left), the looming stimulus restricted to the 286 horizontal band (Fig. 5C, middle) and to the vertical band (Fig. 5C, right), respectively. We found that the 287 farther a branch was from the looming center the smaller its calcium responses were both to standard 288 looming (Fig. 5C, left) and to vertical band-restricted looming (Fig. 5C, right). The latter responses 289 decreased even more strongly with distance from the looming center. However, in response to horizontal 290 band-restricted looming, calcium responses did not decrease with distance from the looming center and 291 even increased slightly (Fig. 5C, middle). To show this more directly, we plotted the peak dF/F on each 292 selected dendritic branch, normalized to the peak dF/F on the branch that mapped to the looming center, 293 as a function of distance to the looming center across animals (n=5 animals; Fig. 5D). We found that with 294 increased distance from the looming center the normalized peak dF/F decreased fastest in response to the 295 vertical band-restricted looming stimulus, while the normalized peak dF/F increased a bit and then 296 7 decreased slowest in response to the horizontal band-restricted looming stimulus (Fig. 5D). By comparing 297 the activated width along the first principal and second principal axis in response to standard looming, to 298 horizontal, and to vertical band-restricted looming across animals (Fig. 5E), we found that the activated 299 widths along the second principal axis for the three cases were similar (blue traces; p=0.15, one-way 300 ANOVA). However, along the first principal axis the activated width in response to horizontal band-301 restricted looming was the biggest, and the activated width in response to vertical band-restricted looming 302 was the smallest (red traces). Interestingly, along the first principal axis, the activation strength at the 303 center branch is not the strongest in response to horizontal band-restricted looming as there is a clear 304 groove at the center position (Fig. 5E, middle). We computed the ratio of the full-width-at-half-maximum 305 (FWHM) of the activation along the first principal axis in response to horizontal or vertical band-306 restricted looming to that in response to standard looming (Fig. 5F). The horizontal-based ratio was larger 307 than 1 and the vertical-based ratio smaller than 1 for all the animals (Fig. 5F, n=5, p=0.0088, paired t-308 test). When we carried out a similar analysis for small translating squares motivated by the wide 309 fluorescence pattern observed in Fig. 5C, we also found increased activation for small squares relative to 310 looming stimuli (Supp. Fig. 5).

312
Presynaptic mAChRs partially counteract global inhibition upstream of the LGMD.

313
To better understand why the activation of lateral branches is weaker in response to standard looming 314 stimuli than to band-restricted horizontal looming stimuli and small translating squares, we designed 315 flicker-off stimuli with different angular sizes (Fig. 6A). The activated dendritic area initially increased as 316 the size of the stimulus increased (Fig. 6B). However, for sizes exceeding 40° the magnitude of the peak 317 dF/F decreased dramatically as angular size further increased up to 100° (Fig. 6B). This suggests the 318 existence of size-dependent inhibition upstream of the LGMD excitatory dendrites that was strongly 319 activated by stimuli larger than 40º. Next, we tested the influence of scopolamine and muscarine on this 320 novel, global type of inhibition. The peak dF/F of the center branch (Fig. 6C) and the mean peak dF/F of 321 all selected branches (Fig. 6D) decreased as the size of the stimulus exceeded 40°. After puffing 322 scopolamine, the peak dF/F of the center branch (Fig. 6C) and the mean peak dF/F of all selected 323 branches (Fig. 6D) was reduced for all stimuli, irrespective of size. After puffing muscarine, the peak 324 dF/F of the center branch (Fig. 6E) and the mean peak dF/F of all the selected branches (Fig. 6F)

325
increased as the size of the stimulus increased from 10° to 40°, but exhibited no clear decrease as the size 326 of the stimulus further increased to 100°. These results suggest that mAChRs partially counteract the 327 effect of global inhibition upstream of the LGMD that is triggered by large visual stimuli.

330
Recent research has demonstrated that the LGMD is selective for the spatial coherence of approaching 331 objects and that this selectivity is partially explained by the interaction of hyperpolarization activated,

345
To test whether mAChR-mediated lateral excitation increases coherence preference, we presented such 346 stimuli while recording the firing rate of the LGMD before and after blocking it with scopolamine. Across 347 8 stimulus conditions, the timing of LGMD firing remained similar after adding scopolamine, but with 348 much lower firing rates (Fig. 7B, C). Responses to standard looming stimuli decreased from 58±16 spikes 349 to 17±7 spikes after addition of scopolamine (mean±sd; p = 0.004). Responses to incoherent stimuli were 350 also reduced after scopolamine addition (Fig. 7C), with 30±8 spikes in control and 11±6 spikes after 351 puffing (p=0.017). For every animal tested there was a decrease in LGMD firing (p≤0.004) and the 352 coherence dependent increase in firing was reduced (Fig. 7D). The coherence preference decreased for 353 each animal (n=5) and on average there was a decrease from 0.28 spikes per percent coherence to 0.07 354 spikes per percent coherence (Fig. 7E). As scopolamine reduced responses to stimuli of all coherence 355 levels, we calculated the coherence specific reduction by subtracting from the reduction in coherent 356 looming response the reduction in incoherent response, and expressing it as a percentage of the control 357 looming response. We found a 37.1 ± 17.8 % specific reduction for responses to coherent looming stimuli 358 (p=0.01, paired t-test). These results indicate that lateral excitation indeed plays a role in shaping the

362
Little is known about how networks converging onto collision-detecting neuron are wired to shape their 363 selectivity. Using calcium imaging, we investigated experimentally (for the first time to the best of our 364 knowledge) how network connectivity among presynaptic excitatory inputs to a collision-detecting 365 neuron shapes its responses. We provide evidence for two new types of presynaptic connectivity patterns:  Drosophila's visual system, small-field directional motion-selective T4 and T5 cells, transmedullary 399 neurons that synapse onto lobula plate tangential cells, both express transcripts of nicotinic and 400 muscarinic acetylcholine receptors (Shinomiya et al. 2014). Although the role of mAChRs in this system 401 remains to be determined, our results suggest that the effects of their activation are likely to last at least 402 100 ms, a time scale that is long relative to that thought to be relevant for directional motion detection 403 (Behnia et al., 2014). In vertebrates, mAChRs have been mainly associated with modulation of attention 404 during behavior (e.g., Niell and Stryker, 2010). Our results suggest they may also contribute to visual 10 As lateral excitation is local, requires activation of several neighboring ommatidia, and occurs on a 450 relatively slow time scale, this suggested to us that it might be involved in discriminating between 451 coherently and incoherently expanding looming stimuli, a form of object segmentation. We confirmed 452 this hypothesis experimentally by showing that coherence sensitivity was diminished in the output firing 453 rate of the LGMD following block of lateral excitation (Fig. 7). Local lateral excitation is thus a 454 mechanism that helps tune the LGMD to looming stimuli and ignore spatially incoherent ones, in addition 455 to dendritic conductances located within the LGMD's dendrites (Dewell and Gabbiani 2017). Although 456 both pre-and post-synaptic mechanisms influence the spatial selectivity of LGMD responses, the 457 coherence tuning due to presynaptic mechanisms shown here plays a somewhat smaller role than that of 458 the dendritic conductances reported in our earlier work (median coherence-specific reduction of 57% after 459 blocking HCN channels, computed from Dewell and Gabbiani, 2017, vs. 34% median specific reduction 460 after scopolamine addition reported here).

462
We have summarized in Supp. Fig. 6  Electronics). The monitor was calibrated to ensure linear, 6-bit resolution control over luminance levels.

493
A 100-120 s delay was used between stimuli and locusts were repeatedly brushed and exposed to light 494 flashes and high frequency sounds to decrease habituation. 'Coarse' looming stimuli were generated as in 495 our earlier work (Jones and Gabbiani 2010;Dewell and Gabbiani 2017). Briefly, the stimulation monitor 496 was first pixelated with a spatial resolution approximating that of the locust eye (2° x 2°), referred to as 497 'coarse' pixels. Each coarse pixel's luminance followed the same time course as that elicited by the edge 498 of the simulated approaching object sweeping over its area. To alter the spatial coherence of these stimuli, 499 a random two-dimensional Gaussian jitter with zero mean was added to each coarse pixel screen location.

11
The jittered positions were rounded to the nearest available coarse pixel location on the screen to prevent 501 any coarse pixels from overlapping. The standard deviation of the Gaussian was altered between 0 and 502 80° to control the amount of shifting and thus the resulting spatial coherence of the randomized stimulus.

504
Staining with calcium indicators.

519
We employed wide-field calcium imaging during visual stimulation experiments. The calcium indicators 520 were excited by light from a 100 W mercury short arc lamp and the resulting fluorescence was measured 521 by a charge-coupled device (CCD) camera at a frame rate of 5 Hz (Rolera XR, Qimaging, Surrey, BC, 522 Canada). We used a 16 X/0.8 NA water-immersion objective lens for imaging (CFI75 LWD 16XW; 523 Nikon Instruments). Three to five trials were averaged for each visual stimulus.

525
Scopolamine, resp. muscarine were prepared at 6 mM in locust saline containing fast green (0.5%) to 526 visually monitor the affected region. They were puffed using a pneumatic picopump (PV830, WPI,

545
Data analysis was performed with Matlab. The relative fluorescence change was calculated as 546 dF/F(t)=[F(t)-F 0 ]/F 0 , with the baseline fluorescence F 0, being the average of the first 10 frames (2 s) 547 before the visual stimulus. For peak dF/F spatial maps (e.g. Fig. 2A), the value of dF/F(t) at each pixel 548 was calculated as the median over a 5 x 5-pixel area centered at the given pixel (0.9 µm/pixel). After 549 median filtering, only suprathreshold pixels were used, defined as having a peak dF/F > 180% of the 550 maximum noise of dF/F at that pixel. The maximum noise at each pixel was computed as the maximum 12 of the dF/F values within 2 s before the onset of the visual stimulus. The threshold value of 180% was 552 selected to eliminate pixels resulting in falsely positive dF/F outside the dendritic branches based on 553 earlier, two-photon imaging experiments with high spatial resolution (Zhu and Gabbiani 2016, Fig. 1D).

554
The peak dF/F values at the pixels above threshold were retained and passed through a Gaussian filter 555 with a standard deviation sigma = 5 pixels. In Fig. 1D, we selected the threshold value (dashed line) to be 556 well above noise level, so that calcium signals were unambiguously increasing at those time points. The 557 results in Fig. 1E and F are robust to changes of the threshold around that value.

559
To map the dendritic receptive fields of square flashes at 5 locations on the screen, we first measured the 560 peak dF/F spatial map in response to individual square flashes. For this purpose, we first filtered the 561 pixels that were above the noise level (see above), then computed the maximum of the peak dF/F over all 562 the pixels and retained those pixels above 4% of the maximum. This additional thresholding step was 563 necessary due to the small amplitude of peak dF/F in response to square flashes. The threshold was 564 selected to eliminate falsely positive dF/F pixels outside the dendritic branches. The branches that were 565 covered by the retained pixels were taken to be the dendritic receptive fields of that square flash.

567
The principal axes of the peak dF/F spatial map were computed by first scaling and rounding the intensity 568 of the peak dF/F at each pixel into integers. Next, we took these integers for each pixel as the number of 569 identical data points with coordinates matching those of that pixel. After that, we performed principal 570 component analysis on all these data points to extract the first principal component that had the largest 571 variance and the second principal component orthogonal to the first one. We rotated the peak dF/F spatial 572 map to make the first or second principal axis horizontal, and summed the values of peak dF/F along the 573 vertical axis to get the projected values along each principal axis.

575
To make plots, the membrane potential was first median-filtered over a time window of 30 ms and down-576 sampled 100 times (to 50 Hz).

579
For paired comparison of the control and scopolamine groups, we performed a nonparametric one-tailed 580 Wilcoxon sign-rank test. The null hypothesis was rejected and the results were judged statistically 581 significant at probabilities p<0.05. For responses to small flashes (Supp. Fig. 3G-I) a non-parametric 582 Wilcoxon rank sum test was conducted on the number of spikes in response to individual stimuli before 583 and after addition of scopolamine. For determining whether there are any statistically significant 584 differences between the means of groups stimulated with gratings of varying separations, we performed a 585 one-way analysis of variance (one-way ANOVA). The null hypothesis that the means of all groups were 586 statistically equal was rejected at significance levels p<0.05. For comparing the slopes of two regression 587 lines, we performed an analysis of covariance (ANCOVA). The null hypothesis that the slopes of the two 588 regression lines were statistically equal was rejected at significance levels p<0.05. A two-way analysis of 589 variance (two-way ANOVA) was performed to determine whether there were any statistically significant 590 differences between the control and the scopolamine/muscarine groups, each of which having different 591 levels (types of visual stimuli). A value of p<0.05 was regarded as significant. For experiments studying 592 the influence of muscarine on calcium responses to flicker-off stimuli with different sizes, statistical 593 analysis of paired comparisons for each stimulus type was not performed as the separation between 594 control and muscarine was unambiguous already after testing 3 animals (Fig. 6E, F)