Synergy of color and motion vision for detecting approaching objects in Drosophila

Color and motion are used by many species to identify salient moving objects. They are processed largely independently, but color contributes to motion processing in humans, for example, enabling moving colored objects to be detected when their luminance matches the background. Here, we demonstrate an unexpected, additional contribution of color to motion vision in Drosophila. We show that behavioral ON-motion responses are more sensitive to UV than for OFF-motion, and we identify cellular pathways connecting UV-sensitive R7 photoreceptors to ON and OFF-motion-sensitive T4 and T5 cells, using neurogenetics and calcium imaging. Remarkably, the synergy of color and motion vision enhances the detection of approaching UV discs, but not green discs with the same chromatic contrast, and we show how this generalizes for visual systems with ON and OFF pathways. Our results provide a computational and circuit basis for how color enhances motion vision to favor the detection of saliently colored objects.

enabled behavioral responses to ON-motion to be more sensitive to UV, as compared for OFF-motion 201 ( Fig. 3a, b), and silencing R7 cells reduced this difference (Fig. 3c). 202 FDR correction, NL3, flies = 9, NL1, flies = NL2, flies = NL4, flies = NL5, flies = 10). We also measured responses in 294 the lamina cell C3 because it provides direct GABAergic, presumed inhibitory, input to T4 cells 26,41 , as 295 well as feedback from the medulla to the lamina where it synapses onto L1, L2, and L3 23 . C3 is an ON 296 cell, and the isoluminance levels of L5 and C3 were indistinguishable ( Fig. 5f; p > 0.05, two-sample t-test 297 with FDR correction; NL5, flies = 10, NC3, flies = 9), as were their increasing calcium responses ( Fig. 5e; p > 298 0.05, two-sample t-test with FDR correction). These results indicate that lamina cells providing the 299 primary inputs to the motion pathways have UV-green isoluminance levels that differ in their sensitivity 300 to UV, covering a range much broader than the T4-T5 isoluminance difference (Fig. 5f). 301 We next examined the T4 inputs cells that receive prominent inputs from the L1-5 LMCs. The Mi1 302 and Tm3 cell types are the principal excitatory inputs to T4 cells 26,41,42 and they receive major inputs from 303 L1, with a contribution from L5 26 (Fig. 5c). The increasing calcium responses of Mi1 and Tm3 to 304 different intensities of UV discs were not significantly different from each other ( Fig. 5e; p > 0.05, two-305 sample t-test with FDR correction, NMi1, flies = NTm3, flies = 10), nor were their isoluminance levels ( Fig. 5f; 306 p > 0.05, two-sample t-test with FDR correction). Their isoluminance levels were not significantly 307 significantly outside the values of T4 and T5. 328 To compare the calcium responses of the cells presynaptic to T4 with those of a cell that receives 329 much of its inputs from R7 cells, we measured the responses to expanding UV discs of Dm9, an UV-OFF 330 cell type and prominent R7 target 13,26,36,44 . Dm9 had a lower UV-green isoluminance level than all the 331 cells in the lamina and T4 motion pathway we recorded, including Mi9 ( Fig. 5f; e.g., Dm9-Mi9, p = 332 0.003; two-sample t-test, NMi9, flies = 9). Although Dm9 provides a minor input to Mi4 (Fig. 5c), its 333 calcium response was the slowest of the cells we measured (Fig. 5d), indicating it is not likely to drive 334 rapid responses in Mi4. 335 Together, these results indicate that T4 input cells differ in their sensitivity to UV (Fig. 5f), 336 differences that are consistent with the UV-sensitivity of their primary LMC inputs (Fig. 5g). In 337 particular, L5 drives Mi4, and both cell types have a greater isoluminance level than T4. 338 Complementarily, L3 drives Mi9, and both cell types have lower isoluminance levels than T4 or T5. 339 340 Behavioral responses to UV-Green and Green-UV edges are asymmetric 341 intermediate intensities (4.5 < UV < 9.2), we predicted that an intensity-matched green disc approaching 362 on a UV background does not generate motion contrast. 363 In agreement with this stringent prediction, flies did not turn away from an intensity-matched 364 green disc expanding on a UV background over the expected range of UV levels between 4 and 9 ( Fig.  365 6ci, di; T4 + T5 > DL, p > 0.05 for 2 £ UV £ 15; Enhancerless split GAL4 > kir , p > 0.05 for 4 £ UV £ 9; 366 one tailed t-test that the mean is greater than zero, with FDR correction, N = 10). These results revealed 367 an asymmetry in responses to moving color edges, that the response of a fly to an expanding color edge 368 depends on its color polarity, that is whether UV expands into green, or green expands into UV. 369 To clarify the role of the primary motion pathway in these experiments, we silenced the T4 and T5 370 cells by expressing Kir2.1. Under these conditions, flies no longer responded to the direction of the 371 expanding discs (Fig. 6cii, dii; p > 0.05, t-test that the mean is zero, with FDR correction, N = 10). The 372 responses of the control flies indicated that the discs were visible even when the flies didn't turn away 373 from them. After the discs had fully expanded, the flies reliably turned towards the side the disc came 374 from, for intensities of UV ³ 2 ( Fig. 6ci; t = 1 -2 s, example indicated by a purple arrow for UV = 7). 375 These responses were not just an attraction to a green disc, because they turned away from the discs when 376 the background was dark ( Fig. 6ci; UV = 0). Therefore, the turning towards the location of the disc 377 depended on seeing the background UV, and may involve multiple pathways, for example those 378 supporting phototaxis or object vision. 379 By verifying an unexpected prediction -that green discs do not evoke turning responses over a 380 large range of background UV levels -these experiments support our hypothesis, that a difference in the 381 spectral sensitivity of ON-and OFF-motion underlies the contribution of color to motion vision in flies. 382 383 Enhanced motion detection for approaching objects of selected colors 384 The mechanisms we have identified in the fly for detecting UV objects could be adjusted to enhance the 385 motion detection of objects of other selected colors too. For example, we expect that the detection of an 386 approaching red object would be enhanced by an augmented red-sensitivity of ON-motion, relative to 387 OFF-motion. To illustrate this hypothesis, we considered the image of an orange in a tree, as seen through 388 a hexagonal lattice of the fly's compound eye, and estimated the ON-and OFF-motion at every hexagonal 389 pixel as the viewer approached the center of the orange or receded from it (Fig. 7a) These results support our hypothesis and demonstrate how motion detection for approaching 405 colored objects in an artificial algorithm can be enhanced by introducing asymmetries in the spectral 406 sensitivity in ON-and OFF-motion detection. The gain in motion detection for the approaching object is 407 tied to a drop in the detection of the receding object, a trade-off that may be acceptable in many situations, 408 for example in automated harvesting systems tailored for specific fruits, or collision avoidance systems. 409 410

411
We have shown that color contributes to motion vision for UV-green edges in Drosophila (Fig. 1). 412 Behavioral responses to ON-motion were much more sensitive to UV than responses to OFF-motion ( Fig.  413 2), a difference requiring the R7 photoreceptors (Fig. 3). The T4 and T5 cells that process ON-and OFF-414 motion, showed a corresponding difference in their sensitivity to UV (Fig. 4), and in the cells linking the 415 R7 photoreceptors and T4 cells, there were consistent spectral differences between lamina monopolar 416 cells and the medulla T4 input cells: L5 and Mi4 were less sensitive to UV, and L3 and Mi9 were more 417 sensitive to UV, compared to the L1 driven Mi1 and Tm3 (Fig. 5g). Finally, we correctly predicted that if 418 the augmented UV-sensitivity of ON-motion processing explained the contribution of color to motion 419 vision, then green discs should not be visible against a UV background that was neither bright nor dark 420 ( Fig. 6). We have shown that the contribution of color to motion vision is not just a mechanism for 421 resolving low-contrast UV edges (Fig. 1), but can also be organized to preferentially support the motion 422 detection of objects of specific colors (Fig. 7). 423 In vertebrates, differences in the spectral sensitivity to ON-and OFF-motion have not been 424 thoroughly investigated, to our knowledge, and if present they could support a contribution of color to 425 motion vision as we have found for Drosophila. Larval zebrafish use UV-ON processing to detect 426 paramecia while foraging 45 , and the mechanism we have described has the potential to operate in wavelengths, and since they have the greatest chromatic sensitivity in the visual circuits viewing the described is in theory directly applicable to the mouse visual system: it predicts that if OFF-motion 431 responses were more sensitive to UV than for ON-motion, this would favor the detection of an 432 approaching object seen against a UV-rich sky. In mice, Khani and Gollisch 47 recently reported ON and 433 OFF retinal ganglion cells that nonlinearly integrate UV and green to allow an OFF cell, for example, to 434 have different isoluminance levels for UV-OFF and green-ON and for green-OFF and UV-ON. In this cell 435 the spectral divergence of ON and OFF processing supports the detection of a light decrement, whether 436 the decrement is in green or UV wavelengths (UV-OFF and green-OFF), and in other cells the 437 nonlinearities were UV-selective (UV-ON and UV-OFF). These nonlinearities are algorithmically very 438 similar to our proposed mechanism and indicate a potential platform for motion-sensitive cells 439 downstream to be sensitive to isoluminant motion, and to preferentially detect objects rich in UV or 440 green, relative to the background. In primates, motion-sensitive cells in area MT contribute to the smooth 441 pursuit tracking of objects and frequently retain some degree of chromatic sensitivity, such that around the 442 isoluminance level the response to motion is decreased, but not to zero 48 . Our results indicate that, in 443 addition to allowing the primates to view isoluminant edges, an individual cell's chromatic sensitivity 444 may be organized to enhance following the motion of targets of a particular color. 445 We established that augmented ON-motion UV-sensitivity is not limited to D. melanogaster but is 446 also displayed by other drosophilids (Fig. 2g). Among invertebrates, color has been reported to contribute 447 to motion vision in other insects including the honeybee 49 and the butterfly Papilio xuthus 50 , whose 448 behavioral responses to moving colored ON-and OFF-edges indicated that responses to ON-motion were 449 more sensitive to red, compared to responses to OFF-motion that were more sensitive to blue and green. If 450 Papilio implements the chromatic mechanism that we have proposed for Drosophila, then this would 451 predict that its ON-and OFF-motion pathways support seeing red objects against green backgrounds, for 452 example red flowers set against foliage.
developing methods to accurately display wide-field UV-green stimuli. Prior work also indicated that 456 color might contribute to motion vision by broadening the spectral sensitivity of the luminance channel 457 through unidentified cellular mechanisms 14 , and subsequent EM reconstructions indicated that the R7 and 458 R8 photoreceptors form synapses in the medulla with cells specifically presynaptic to T4 12,26,27 . Our 459 results indicate that UV-sensitivity is maintained along the R7-L3-Mi9-T4 pathway (Fig. 5g), and predict 460 that R7 cells innervate L3. Indeed, we recently demonstrated in an EM reconstruction study that R7 cells 461 form substantial numbers of previously unreported synapses with L3 and other cells in the optic chiasm 462 between the lamina and medulla 13 . To understand how sensitivity to UV propagates from R7 through the spectral tuning, and recurrent connections along the pathway. To focus on just one example, the Mi4 and 465 Mi9 cell types, which are inhibitory to T4 cells 42 , heavily synapse onto each other 26 and reciprocal 466 inhibition between these cell types may amplify their chromatic differences. 467 In summary, we have shown how UV contributes to the processing of ON-motion in Drosophila 468 in a way that enhances responses to expanding UV discs. We have identified key cellular components of 469 how color contributes to motion vision in flies, the R7 and T4 cells, and how cells linking them show 470 consistent differences in their spectral tuning. We have shown how a spectral divergence in ON-and 471 OFF-motion processing can be used to favor objects of a specific color, an insight that is directly 472 applicable to many vertebrate and invertebrate sensory systems.   Rel. Irradiance (a.u.)              Figure 6. Behavioral responses to UV-Green and Green-UV edges are asymmetric. a. Illustration of how a difference in UV-sensitivity of behavioral responses to ON-and OFF-motion enable the detection 757 of isoluminant UV discs expanding on a green background. We hypothesized that when the UV disc is 758 darker than the OFF-motion isoluminance level (UV = 9.2; Fig. 2d), the edge of the disc drives OFF-759 motion responses (black bar), and when the UV disc is brighter than the ON-motion isoluminance level 760 (UV = 4.5; Fig. 2d), the edge of the disc drives ON-motion responses (purple bar). This hypothesis 761 predicts that when the disc has a UV intensity in the isoluminance band (UV = 4.5 -9.2), the edge of the 762 disc is simultaneously bright enough (UV > 4.5) to drive ON-motion and dark enough (UV < 9.2) to drive 763 OFF-motion responses (overlap of black and purple bars). b. In contrast to panel (a), our hypothesis 764 predicts a lack of responses to a green disc expanding from UV, whose intensity is matched to the green 765 background used in our prior experiments with UV discs. When the UV background is darker than the 766 ON-motion isoluminance level, the edge of the intensity-matched green disc is bright enough to generate 767 ON-motion responses (black bar), and when the UV background is brighter than the OFF-motion 768 isoluminance level, the edge of the disc is dark enough to drive OFF-motion responses (purple bar). When 769 the disc has a UV intensity in the isoluminance band (UV = 4.5 -9.2), we predicted that the edge of the 770 disc would be neither bright enough (UV > 9) nor dark enough (UV < 9) to drive ON-or OFF-motion 771 responses (gap between black and purple bars). c. Turning responses of flies of control genotypes (ci) and 772 T4 and T5 silenced flies (cii) to an intensity-matched green disc expanding on a UV background, with the 773 UV intensity of the background given in purple above panels. The timing of the stimulus is as for the UV 774  Table 1.    Table 1, and those used for imaging are listed in Table 2. For all behavioral results, all the primary data 827 were from enhancerless split GAL4 crossed with wild type DL flies unless otherwise stated. The 828 enhancerless split GAL-4 flies have transgenes in the same genomic location as the other split-GAL4 829 drivers, and so match the general genotype, but these transgenes lack the enhancer-containing cis-830 regulatory sequences that determine the specific patterns of the other driver lines. 831 Rh3-and Rh6-GAL4 lines with insertions on the second chromosome were obtained from the 832 Bloomington stock center (BDSC #7457 and #7459, respectively). To generate additional Rh1-, Rh2-, 833 Rh3-, Rh4-, Rh5-and Rh6-GAL4 driver lines with the transgenes inserted in the attP2 landing site on the 834 third chromosome, we PCR-amplified previously characterized promoter regions 52-57 from genomic 835 DNA, TOPO-cloned the PCR products into pENTR-D-TOPO and transferred to pBPGUw (Addgene 836 #17575) using standard Gateway cloning. Primer sequences were as described 13 (Rh3, Rh5, Rh6) or as 837 listed below (Rh1, Rh2, Rh4). Transgenic flies were generated by phiC31-mediated integration (Genetic 838 Services, Inc.). 839 Rh2R GCT CAG CTA CCC GCA ACC CCT T  Immunohistochemistry 852 To visualize the expression pattern of the L1 split-GAL4 driver line (Extended Data Fig. 5a), we 853 used pJFRC51-3XUAS-IVS-Syt::smHA in su(Hw)attP1 and pJFRC225-5XUAS-IVS-myr::smFLAG in 854 VK00005 58 . The images were generated by the Janelia FlyLight project team. Sample preparation was as 855 previously described 59 and a full protocol is available online, https://www.janelia.org/project-856 team/flylight/protocols under "IHC -Anti-GFP", "IHC -Polarity Sequential" and "DPX mounting". 857 Images were acquired on Zeiss LSM 710 or 800 confocal microscopes with 20x 0.8 NA or 63x 1.4 NA 858 objectives. For display, we generated resampled views from three-dimensional image stacks using the 859 Neuron annotator mode of V3D 60 and exported these images as TIFF format screenshots. 860 value roughly in the middle of the intensity range of 0 and 15. As a result, the green illumination pattern 880 is fixed in all experiments where there is green light. The UV intensity varies slightly across the screen 881 (Extended Data Fig. 1a), and as we could not create a luminance mask for the UV-channel and maintain 882 the ability to change the UV intensity, the UV intensity is expressed by the intensity value (0-15) and not 883 the irradiance (but we note that the ratio of UV to Green at each location is tightly controlled after the 884 calibration, Extended Data Fig. 1e). The effectiveness of this approach was validated by the motion 885 isoluminance shown by colorblind norpA 36 mutants with norpA function restored in R1-6 using Rh1-886 GAL4 (Fig. 1d-e, 3b). 887 Two projector systems were used to collect the behavioral data, created and calibrated identically. 888 All the data presented in the main figures were collected on the same projector system. Data from the 889 second system is used for Extended Data Fig. 2a  Corp., Santa Clara, CA, USA). We organized stimuli into trials of 8 s duration. The first 6 s were closed-918 loop stripe fixation, with the fly's turning response (DWBA) controlling the position of a 10° azimuth 919 wide black bar, moving on a green background. The stimulus was presented in the last 2 s of the trial. 920 Two types of stimuli were used, expanding discs and competing ON-and OFF-motion, described below. 921 In total, four protocols were used: 1) expanding UV discs with a green background; 2) expanding Green 922 discs with a UV background; 3) expanding UV discs with a green background of different speeds; 4) 923 competing ON-motion edges over the range UV = 0 -15; 5) competing OFF-motion over the range UV = 924 0 -15; 6) competing ON-and OFF-motion over the range UV = 3-9. 925 926 Expanding UV discs with a green background. 927 For flies viewing UV discs expanding from a green background (Fig. 1d-e, Extended Data Fig. 1g-h), the 928 disc appeared at 6 s and expanded as though moving towards the fly with a constant velocity until it had 929 expanded to fill the visual display after one more second, at 7 s. The size of a disc expanding with 930 apparent constant motion (displayed with an accelerating angular size) can be parameterized by the ratio In every set of trials, expanding green discs were shown with UV intensities of the background of 950 {0, 2, 4, 5, 6, 7, 8, 9 10, 12, 15}. In all other respects, the stimuli were organized as for the expanding UV 951 discs. 952

953
Expanding UV discs with a green background of different speeds 954 For the measurements of response to discs expanding with different apparent speeds (Extended Data Fig.  955 1f), the expansion speed was varied for UV = 6 discs appearing out of a green background for r/v of {10, 956 15, 30, 60, 120} ms. We also measured responses to black and green square wave gratings, spatial 957 wavelength 30°, temporal frequency of 5 and 10 Hz, for clockwise and anticlockwise yaw rotations, to 958 generate optomotor responses (data not shown). All stimuli were presented 5 times (10 times including 959 from the left or right), and the stimulus presentation order was randomized for each set of trials. 960 window and retreated leftwards to void the window of UV, leaving the screen blank by 250 ms. As for 984 ON-motion, counterclockwise stimuli were presented in the same way, but reflected along a vertical axis 985 so that green edges moved leftwards, and UV edges moved rightwards. In all other respects, the stimuli 986 were organized as for competing ON-motion. 987

988
Competing ON-and OFF-motion over the range UV = 3 -9 989 To measure responses of flies to both ON and OFF-motion (Figs 2g, 3, Extended Data Figs 2, 3), we 990 presented UV intensities over the limited range {3, 4, 5, 6, 7, 8, 9}, so that responses to both competing 991 ON and competing OFF-motion could be measured in the same flies within a protocol that took 21 992 minutes. We also measured responses to black and green square wave gratings, spatial wavelength 30°, 993 temporal frequency of 5 and 10 Hz, for clockwise and anticlockwise yaw rotations, to generate optomotor 994 responses. All stimuli were presented 5 times (10 times including from the left or right), and the stimulus 995 presentation order was randomized for each set of trials. 996 997

Data analysis for behavioral experiments 998
All data were analyzed in MATLAB. The responses to clockwise and counterclockwise stimuli were 999 averaged, with the counterclockwise responses inverted. Likewise, responses to expanding discs centered 1000 on opposing azimuth locations of ±60° were inverted and combined. 1001 For the responses to expanding discs, the response in the 100 ms after the disc had expanded was 1002 used to calculate how the response varied with UV intensity. For the competing edge stimuli, the mean 1003 response over the duration of the stimulus (2 s) was used to calculate how the response varied with UV 1004 intensity. 1005 The isoluminance level was calculated as the first point when the fly's mean response over all 1006 trials was greater than zero, as the UV intensity increased from UV = 0, using linear interpolation between 1007 stimulus intensities, as illustrated in Fig. 2e-f. data saved in the last 0.4 s of every trial. Before the experimental protocol was displayed, expanding discs 1041 expanding up to q = 30° were shown repeatedly to center the imaging field of view to the same region of 1042 visual space across flies and experiments. 1043 green and black ON-and OFF-motion edges, with a spatial wavelength of 30° and a temporal frequency 1051 of 1 Hz, moving up, down, left, or right ( Fig. 4c-d). For the ON-motion stimuli, the screen switched to 1052 blank and then green edges appeared, as for the green component of the competing motion stimuli in 1053 behavioral experiments. We also recorded responses to a uniform green screen, a uniform UV = 15 1054 screen, and an unilluminated, blank screen (data not shown). All stimuli were shown for 5 trials, with the 1055 order of stimuli randomized for every set of trials, taking 18 minutes in total. 1056 For the experiments with expanding UV discs with a UV background (Extended Data Fig. 4), the 1057 stimuli were as for the expanding UV discs with a UV background, except that the background was set to 1058 UV = 5 for T4 cells, and the background was set to UV = 8 for T5 cells. Before and after the stimuli, the 1059 screen was green, and in all other regards, the trials were organized exactly as for the expanding UV discs 1060 with a green background. 1061 1062 Data analysis for calcium imaging 1063 Imaging data was recorded for the first 7.6s of every trial, that lasted 8s, and the frames were saved during 1064 the last 0.4s. The visual display was triggered by the onset of image recording, and the stimulus frames 1065 were temporally synchronized to the image acquisition. 1066 To spatially align frames, we calculated a binary template of the mean calcium fluorescence for 1067 one stimulus (UV = 0 for OFF cells, or UV = 15 for ON cells) and calculated the spatial cross-correlation 1068 between the template and binarized frames, for all frames in the experiment. Recordings with large 1069 movement (>25 pixels) were discarded.
individual flies contributed equally to the population statistics (Figs 4h, 5f). As an additional method to 1109 average out noise in responses, we also calculated the population isoluminance, the UV intensity when the 1110 mean response across flies, first reached zero (using the same ON/OFF consideration as above; used in 1111 Fig. 4f). In this calculation, every fly again contributed equally to the mean response. 1112 To statistically test the differences between the isoluminances of cell types, we calculated the mean 1113 isoluminance across ROIs for individual flies and used the student's t-test, after checking the data was Extended Data Figure 2 silenced (white) and not silenced (colored). We used paired Wilcoxon signed rank test to compare 1227 isoluminance levels within genotypes, and two sample Wilcoxon rank sum tests to compare isoluminance 1228 levels between rescue and control genotypes, with N = 10 for all genotypes. Plotting conventions for 1229 asterisks as in panel (a). Boxplot conventions are as in Fig. 2d. Genotypes for all flies used in behavioral 1230 experiments are in Table 1 Figure 3 tuning. a. Mean calcium activity responses of T4 and T5 ROIs to UV discs expanding out of a UV 1235 background. For the T4 recordings, the background UV intensity was UV = 5, and so when the discs 1236 shared this intensity, they were unambiguously isoluminant. For the T5 recordings, the background UV 1237 intensity was UV =8, and so when the discs shared this intensity, they were unambiguously isoluminant. 1238 We chose these background intensities to test our methods at comparable illuminance levels to the values 1239 determined from the experiments with UV discs expanding out of a green background (Fig. 4f). Mean 1240 ±SEM shown, NT4, flies = 9, NT5, flies = 8, different flies to those in Fig. 4