Neural circuitry for stimulus selection in the zebrafish visual system

•Zebrafish respond to competing stimuli using winner-take-all and averaging strategies•Retinotectal isthmotectal circuits enable context-dependent stimulus selection•Perturbing nucleus isthmi function disrupts behavioral action selection•Isthmotectal circuit layout is consistent with local enhancement global suppression When navigating the environment, animals need prioritize responses most relevant stimuli. Although a theoretical framework for selective visual attention exists, its implementation has remained obscure. Here we investigated how larval zebrafish select between simultaneously presented We found that mix of (WTA) strategies best simulates responses. identified two whose activity patterns predict relative saliencies objects. Stimuli only one eye are selected by WTA computation in inner retina. Binocularly stimuli, on other hand, processed reciprocal, bilateral connections (NI) tectum. This interhemispheric leads or Optogenetic stimulation laser ablation NI neurons disrupt selection selection. Thus, depending locations combination retinotectal enables attention. confronted crowded scene, often choose single object response from multiple Examples such include escaping imminent among several approaching threats focusing individual prey item within herd shoal. Elementary forms spatial exist many species, including flies (Sareen et al., 2011Sareen P. Wolf R. Heisenberg M. Attracting fly.Proc. Natl. Acad. Sci. USA. 2011; 108: 7230-7235Crossref PubMed Scopus (61) Google Scholar), fish (Ben-Tov 2015Ben-Tov Donchin O. Ben-Shahar Segev Pop-out search moving targets archer fish.Nat. Commun. 2015; 6: 6476Crossref (38) mice (Wang Krauzlis, 2018Wang L. Krauzlis R.J. Visual Selective Attention Mice.Curr. Biol. 2018; 28: 676-685.e4Abstract Full Text PDF (28) Scholar). Winner-take-all computations, which an animal responds target while disregarding others, considered be crucial during bottom-up, stimulus-driven (Itti Koch, 2000Itti Koch C. A saliency-based mechanism overt covert shifts attention.Vision Res. 2000; 40: 1489-1506Crossref (2238) In addition mechanisms, evidence suggests presentation primates can also lead gaze toward their mean (Lisberger Ferrera, 1997Lisberger S.G. Ferrera V.P. Vector smooth pursuit movements initiated monkeys.J. Neurosci. 1997; 17: 7490-7502Crossref Scholar; Nummela 2011Nummela S.U. Superior colliculus inactivation alters weighted integration stimuli.J. 31: 8059-8066Crossref (19) Ottes 1984Ottes F.P. Van Gisbergen J.A. Eggermont J.J. Metrics saccade double stimuli: different modes.Vision 1984; 24: 1169-1179Crossref (200) contrast strategy, this “averaging” mode existence neurocognitive process integrates across sensory rather than selecting response. It been postulated that, decision-making, potential actions represented brain these parallel representations compete against each (e.g., Cisek, 2007Cisek Cortical mechanisms selection: affordance competition hypothesis.Philos. Trans. Soc. Lond. B 2007; 362: 1585-1599Crossref (596) According view, importance field through focally enhance salient globally suppress neuronal elsewhere (Ewert, 1997Ewert J.-P. Neural correlates key releasing mechanism: case study concepts.Trends 8: 332-339Google Ullman, 1985Koch Ullman S. Shifts attention: towards underlying neural circuitry.Hum. Neurobiol. 1985; 4: 219-227PubMed Lee 1999Lee D.K. Itti Braun J. activates filters.Nat. 1999; 2: 375-381Crossref (359) Such “saliency map” appears implemented tectum/superior vertebrates as distantly related lampreys, goldfish, birds, Gruberg 2006Gruberg E. Dudkin Wang Y. Marín G. Salas Sentis Letelier Mpodozis Malpeli Cui H. al.Influencing interpreting input: role feedback system.J. 2006; 26: 10368-10371Crossref (42) Kardamakis 2015Kardamakis A.A. Saitoh K. Grillner Tectal microcircuit generating commands gaze-controlling neurons.Proc. 112: E1956-E1965Crossref (32) Knudsen, 2018Knudsen E.I. Circuits That Mediate Attention: Comparative Perspective.Trends 41: 789-805Abstract 2018Krauzlis Bogadhi A.R. Herman J.P. Bollimunta A. without neocortex.Cortex. 102: 161-175Crossref (33) Schellart 1979Schellart N.A.M. Riemslag F.C.C. Sperkreijse Center-surround organisation interactions receptive fields goldfish tectal units.Vision 1979; 19: 459-467Crossref (27) Zhaoping, 2016Zhaoping From optic tectum primary cortex: migration evolution saliency map exogenous attentional guidance.Curr. Opin. 2016; 94-102Crossref Studies barn owl suggest could supported reciprocal loops (NI), satellite nucleus, homologous parabigeminal mammals located tegmentum (Knudsen, Evidence model come mainly single-cell electrophysiology classical tract-tracing studies. decisions larvae when they faced threatening escape direction determined strategies, varying degrees fish. Brain-wide two-photon functional imaging, targeted cell ablations, optogenetic populations revealed impinging contribute location-dependent manner. both same eye, already detectable retinal ganglion terminals eyes, pathway, forming recurrent tectum, weighs strengths required hemispheres. These results pinpoint circuitries intra- vertebrate system. To identify adopted established paradigm larvae. tracked swimming freely arena computer vision. Based location, were projected below at defined positions animal’s orientation (Figure 1A). The consisted looming disks expansion rate contrast. Dark expanding mimicking predator collision course, strongly aversive vigorous movement (Bhattacharyya 2017Bhattacharyya McLean D.L. MacIver M.A. Threat Assessment Reticulospinal Encoding Calibrated Responses Larval Zebrafish.Curr. 2017; 27: 2751-2762.e6Abstract (35) Dunn 2016Dunn T.W. Gebhardt Naumann E.A. Riegler Ahrens M.B. Engert F. Del Bene Underlying Visually Evoked Escapes Zebrafish.Neuron. 89: 613-628Abstract (149) Temizer 2015Temizer I. Donovan J.C. Baier Semmelhack J.L. Pathway Looming-Evoked Escape 25: 1823-1834Abstract (120) Single side highly effective driving opposite (Figures 1D–1F S1). Depending location strength stimulus, adjusted magnitude Increasing resulted higher probability greater vigor S1A–S1E, S1I, S1J). observations confirm earlier studies adapt behavior Next wanted find out locations. displayed non-overlapping parts S2A), first alone then combination. disk positioned anterior triggered sideways (82.5° ± 75, standard deviation [SD]) respect heading direction; Figures S2B, S2E, S2L), whereas posteriorly forward (47° 41 SD; S2C, S2H, S2L). For identical expected either stochastic choice trajectory corresponding average If differ, more should dominate case, may still strengths. Both together yielded distribution angles included S2D S2G). As faster dominated so angle was similar position S2F, S2I, S2J–S2L). objects sides fish, observed bimodal trajectories. expanded rates, largely symmetrical around direction. escaped away one, apparently randomly chosen 1B S1G). Modulating biased escapes 1C S1H). indicate stimulus. WTA-mediated comprise (approximately 80%), notable fraction 20%) had explained strategy. estimate contributions fit data mixed predictions 1G). Models where outweighed provided better equal 1H, 1J, 1K, S1K, S1L). unequal responded disk. asymmetry modeled adding bias term, 0.5 balanced values reflect asymmetry, component keeping mixture 1I, 1L, 1M, S1M, S1N). Observing switch S1P). conclude called up exact conditions, experience, internal states over vice versa not known. next brain-wide calcium imaging. First regions reliably Monocular activated (RGC) axons, pretectum, thalamic area near arborization 4 (AF4) (Heap 2018Heap L.A.L. Vanwalleghem Thompson A.W. Favre-Bulle I.A. Scott E.K. Luminance Changes Drive Directional Startle Thalamic Pathway.Neuron. 99: 293-301.e4Abstract (24) Figure S3A). addition, our recordings responsive midbrain-hindbrain boundary, coincides (Gruberg Northmore, 1991Northmore D.P. teleost (Lepomis macrochirus).Vision 1991; 525-535Crossref (31) Northmore Gallagher, 2003Northmore D.P.M. Gallagher S.P. Functional relationship teleosts: synchrony but no topography.Vis. 2003; 20: 335-348Crossref (21) least types present: (1) scales (2) suppressed competitor signatures, kept constant (S1) systematically (S2) 2A–2C). used transgenic expressing cytosolic genetically encoded indicator GCaMP6s RGCs nucleus-localized all neurons. approach enabled simultaneous recording unambiguous separation RGC axon neuropil layers body layer. expected, elicited monocular S1 S2 organized retinotopically, leading foci posterior cells neuropil, respectively 2D). Presenting variety Some scaled increasing 2F 2H, green traces). On subset S1-responsive stronger enhanced weaker magenta switch-like tuning inhibition (Mysore 2012Mysore Knudsen Reciprocal inhibition: motif flexible categorization selection.Neuron. 2012; 73: 193-205Abstract (45) imaging presence competitor, model. visible level axonal projections 2E 2G). transition population shifted 2I S3B–S3G). noticed predicted value compared (difference correlations: Fisher’s z-transformation p = 0.039; 2I). manifests reduced axons sharpened further test whether modulation due (Henley 1986Henley J.M. Lindstrom Oswald R.E. Acetylcholine receptor synthesis retina transport goldfish.Science. 1986; 232: 1627-1629Crossref (65) imaged GCaMP6s-labeled following chemogenetic nitroreductase-expressing 2J, 2K, S4J). Treatment cell-death-inducing pro-toxin metronidazole led severe impairments S4D–S4G). Switch-like RGCs, however, intact 2L 2M; 0.1797, two-way Mann-Whitney test). shapes does require retrograde cells. Synthetic prey-like have shown previously evoke hunting (Bianco 2011Bianco I.H. Kampff Prey capture evoked simple zebrafish.Front. Syst. 5: 101Crossref (150) 2014Semmelhack Thiele T.R. Kuehn Laurell dedicated pathway detection zebrafish.eLife. 2014; 3: 3Crossref (98) tested dynamics extend small, motile dots simulate prey. showed driven S4A–S4C). might serve efficient background distractors. finding indicates system mechanism, restricted guides diverse natural behaviors. previous work prime candidate shaping binocular 3A). Indeed, concurrent hemispheres 3B 3C). left right unbalanced; hemisphere high activity, low mirroring 3C–3E). Similar competition, inhibited stimulus-selective intensity 3D–3G). matched signature thus identifies correlate characterize cellular composition NI, transmitter-specific Gal4 lines, RNA situ hybridization, immunohistochemistry label glutamatergic (vglut2a lhx9), cholinergic (choline acetyltransferase, ChAT), GABAergic 3H, 3I, S5). Co-registration lines markers (Kunst 2019Kunst Mokayes N. Kramer Kubo Fernandes A.M. Förster D. Dal Maschio Cellular-Resolution Atlas Zebrafish Brain.Neuron. 2019; 103 (21–38.e5)Abstract (54) Scholar) form spatially segregated clusters close boundary 3I). small lhx9-positive, co-express ChAT S5F). express known marker genes isthmic region, e.g., Reelin S5; Volkmann 2010Volkmann Chen Y.-Y. Harris M.P. Wullimann M.F. Köster R.W. cerebellar upper rhombic lip generates tegmental hindbrain nuclei long-distance evolutionary conserved manner.J. Comp. Neurol. 2010; 518: 2794-2817PubMed project panel). vast majority labeled here stimuli; dimming S6A S6D). S6B S6E). However, significantly distractor S6G–S6K). show compatible map. end, developed restrained preparation allowed us selectively activate inhibit presenting eyes 4A). absence stimulation, activation lhx9-positive ChR2 twice likely induce ChR2? controls (ChR2+, 0.44 0.34, n= 21 fish; control, 0.19 0.17, n 18; 0.0217; 4B). Inhibition GtACR2 neither increased nor light (response 0.21 0.14, 9, 0.189 controls). Notably, excitation affected responding probabilities 0.24 0.10, controls; 0.26 0.12, ChR2+; 0.27 0.11, GtACR2+) 4C). demonstrate excitatory sufficient drive sought what kinds swim bouts performed stimulation. dataset, light-evoked, looming-evoked, spontaneous occurred outside period (n 881 total) characterized bout based three kinematic parameters (bout integral, maximum tail amplitude, beat frequency). hierarchical clustering space 4D; STAR Methods). C-starts large amplitude integral (Burgess Granato, 2007Burgess H.A. Granato Modulation locomotor adaptation.J. Exp. 210: 2526-2539Crossref (288) 4D, blue cluster). Burst swims exhibited frequency symmetric midline (Budick O’Malley, 2000Budick S.A. O’Malley D.M. Locomotor repertoire zebrafish: swimming, turning capture.J. 203: 2565-2579Crossref red Slow relatively slow appeared encompass routine turns 1 2 described (Marques 2018Marques Lackner Félix Orger Structure Repertoire Revealed Unsupervised Behavioral Clustering.Curr. 181-195.e5Abstract (79) predominantly drove ipsilaterally site 4E). Interestingly, able elicit significant number burst 0.13 0.16 per trial; 0.03 0.09; 0.0026; 4F). release fast well Finally, control any effector, (C-starts) (burst swims) dynamics, free-swimming 1). preparation, (0.44 0.34 loom, 0.12 C-start SD). Surprisingly, even though isolation swims, unable additional looms upon (0.50 0.33 result ceiling effect, dominating paradigm, being insufficient override innate (Klapoetke 2017Klapoetke N.C. Nern Peek M.Y. Rogers E.M. Breads Rubin G.M. Reiser Card Ultra-selective radial motion opponency.Nature. 551: 237-241Crossref (52) decreased 80% (0.02 0.06 trial, 0.020 controls) 4H). Neither 4G), suggesting isolation. plays instructive generation contributing strategies. manipulation affects dynamics. combined multiplane cell-resolution perturbations computer-generated holographic (2P-CGH) photostimulation (Dal 2017Dal Helmbrecht T.O. Linking Neurons Network Function Behavior Two-Photon Holographic Optogenetics Volumetric Imaging.Neuron. 94: 774-789.e5Abstract (72) 5A). ChR2-expressing induced resembled (one active, opto-activation index less ?0.25 0.25) (both tecta almost numbers active cells,