Molecular and functional dissection using CaMPARI-seq reveals the neuronal organization for dissociating optic flow-dependent behaviors.

Molecular and functional dissection using CaMPARI-seq reveals the neuronal organization for dissociating optic flow-dependent behaviors - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Nat Commun . 2026 Apr 17;17:3411. doi: 10.1038/s41467-026-71371-6 Search in PMC Search in PubMed View in NLM Catalog Add to search Molecular and functional dissection using CaMPARI-seq reveals the neuronal organization for dissociating optic flow-dependent behaviors Koji Matsuda Koji Matsuda 1 RIKEN Center for Brain Science, Wako, Saitama, Japan 2 National Institute of Genetics, Systems Neuroscience Laboratory, Mishima, Shizuoka, Japan Find articles by Koji Matsuda 1, 2 , Chung-Han Wang Chung-Han Wang 1 RIKEN Center for Brain Science, Wako, Saitama, Japan 2 National Institute of Genetics, Systems Neuroscience Laboratory, Mishima, Shizuoka, Japan 3 Genetics Program, Graduate University for Advanced Studies (SOKENDAI), Mishima, Shizuoka, Japan Find articles by Chung-Han Wang 1, 2, 3 , Hisaya Kakinuma Hisaya Kakinuma 1 RIKEN Center for Brain Science, Wako, Saitama, Japan Find articles by Hisaya Kakinuma 1 , Atsushi Toyoda Atsushi Toyoda 4 National Institute of Genetics, Comparative Genomics Laboratory, Mishima, Shizuoka, Japan Find articles by Atsushi Toyoda 4 , Tomoya Shiraki Tomoya Shiraki 5 National Institute of Genetics, Laboratory of Molecular and Developmental Biology, Mishima, Shizuoka, Japan Find articles by Tomoya Shiraki 5 , Koichi Kawakami Koichi Kawakami 5 National Institute of Genetics, Laboratory of Molecular and Developmental Biology, Mishima, Shizuoka, Japan Find articles by Koichi Kawakami 5 , Fumi Kubo Fumi Kubo 1 RIKEN Center for Brain Science, Wako, Saitama, Japan 2 National Institute of Genetics, Systems Neuroscience Laboratory, Mishima, Shizuoka, Japan Find articles by Fumi Kubo 1, 2, ✉ Author information Article notes Copyright and License information 1 RIKEN Center for Brain Science, Wako, Saitama, Japan 2 National Institute of Genetics, Systems Neuroscience Laboratory, Mishima, Shizuoka, Japan 3 Genetics Program, Graduate University for Advanced Studies (SOKENDAI), Mishima, Shizuoka, Japan 4 National Institute of Genetics, Comparative Genomics Laboratory, Mishima, Shizuoka, Japan 5 National Institute of Genetics, Laboratory of Molecular and Developmental Biology, Mishima, Shizuoka, Japan ✉ Corresponding author. Received 2025 May 18; Accepted 2026 Mar 23; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13090360  PMID: 41997902 Abstract Optic flow processing is critical for the visual control of body and eye movements in many animals. Rotational and translational binocular optic flow patterns need to be clearly distinguished to induce different behavior outputs. However, the specific neuron types and their connectivity involved in this computation remain unclear. Here, we developed a method to link the functional labeling using a photoconvertible calcium indicator called CaMPARI2 and single-cell RNA sequencing (CaMPARI-seq) to investigate the transcriptional profile of the pretectum, a center for processing optic flow in larval zebrafish. Using this technique, we identified a pretectal cluster expressing tcf7l2 , which can be further classified into molecularly distinct subclusters. In vivo calcium imaging and cell ablation revealed that nkx1.2lb -positive pretectal neurons are commissural inhibitory neurons required for the optomotor response but not for the optokinetic response. Our genetic and functional dissection using CaMPARI-seq uncovered the neuronal organization essential for dissociating different optic flow-dependent behaviors. Subject terms: Visual system, Sensorimotor processing In this study, the authors develop CaMPARI-seq to link neural activity with molecular profiles in larval zebrafish. They identify inhibitory pretectal neurons required for optomotor response, revealing how the brain distinguishes optic flow to guide behavior. Introduction Optic flow, the visual motion resulting from animals’ movements, plays an important role in estimating self-motion in invertebrates and vertebrates 1 , 2 . Optic flow triggers the stereotyped compensatory responses of the eye (the optokinetic response; OKR) and body (the optomotor response; OMR) by which animals adjust and minimize self-motion-induced displacement 3 – 5 . Previous studies conducted in mammals and teleosts have demonstrated that the activation of the pretectum (a part of the accessory optic system in mammals) is sufficient to evoke OKR, whereas lesions or experimental inactivation suppress this behavior 6 – 8 . The afferent and efferent connections of the pretectum in mammals and teleosts suggest that the pretectum receives direct inputs from the retinal ganglion cells (RGCs) in the contralateral retina and transmits information to the brainstem and other structures in support of visual-oculomotor control 9 – 11 . In the zebrafish, a subset of RGCs that encode the direction of motion, namely the direction-selective (DS-) RGCs, project to a neuropil region of the pretectum 12 – 14 . This retinorecipient neuropil, primarily in the arborization field AF5, is innervated by optic flow-responsive pretectal neurons 13 . A subset of the pretectal neurons projects to downstream premotor/motor areas such as the nucleus of the medial longitudinal fasciculus (nMLF) in the midbrain and reticular formation in the hindbrain, which produce motor output 13 , 15 . Previous systematic and unbiased functional imaging studies in zebrafish have demonstrated that hundreds of DS neurons in the pretectum respond to optic flow 7 , 16 , 17 . Furthermore, these DS pretectal neurons show diverse physiological properties 18 , including preferred directions of motion 19 , receptive field size and location 20 , motion sensitivity defined by higher-order correlations 14 , binocular integration 7 , 16 , 19 , 21 , and temporal integration 22 . Among them, binocular integration of optic flow is of high behavioral relevance. In lateral-eyed animals, including the zebrafish, comparison of the motion information between the left and right eyes is an integral step in estimating optic flow patterns across a wide area of the visual field 7 , 16 , 17 , 23 , 24 . Functional imaging has demonstrated that a large population of zebrafish pretectal neurons integrate binocular optic flow 7 , 16 , 17 . Notably, most of these binocular neurons selectively respond to translational optic flow, and rotation-selective neurons comprise only a minor part 7 , 21 . These observations led to the idea that binocular computation in the pretectum is suited for processing translation-induced optic flow but not rotation-induced optic flow, and this property ensures the appropriate directionality of OMR and OKR behaviors to be unambiguously encoded at the level of the pretectum 21 . However, the specific neuron types and connectivity enabling this separate processing remain unclear. Despite extensive characterization of the physiological properties of the optic flow processing circuit, the causal link between the identified neuron types and behavior remains largely unknown. This is partly due to the unavailability of techniques to manipulate optic flow-responsive neurons in a subtype-specific manner. To overcome this problem, we sought to genetically label and manipulate different functional subtypes of pretectal neurons. Vertebrate pretectum is derived developmentally from an evolutionarily conserved segment of the diencephalon known as prosomere 1 (p1) 25 – 28 . Although some transcription factors that are expressed by the p1 domain have been described, current knowledge about cell-type-specific molecular signatures within the pretectum is limited. A recent study conducted using high-throughput single-cell RNA sequencing (scRNA-seq) revealed the molecular taxonomies of zebrafish visual neurons, including pretectal neurons 29 . However, it is challenging to associate individual transcriptomic types with their function, behaviors, and morphological connectivity. In this study, we employed transcriptional profiling of pretectal neurons linked with visual functions. We strategically combined activity-dependent fluorescent labeling by CaMPARI and the scRNA-seq technique 30 , 31 to molecularly classify optic flow-responsive neurons. We identified a transcriptomic cell type that contained optic flow-responsive cells in the pretectum, which we further classified into seven distinct transcriptomic cell types. Using CRISPR/Cas9 genome editing and intersectional strategies, we established a set of transgenic lines to genetically access several molecularly defined pretectal cell types. We identified two molecularly distinct subtypes of pretectal neurons that were morphologically distinct and composed of different sets of functional response types. We found that a subset of pretectal neurons that express the transcription factor nkx1.2lb are inhibitory commissural neurons that are required for the optomotor response but dispensable for the OKR. Overall, our functionally targeted transcriptional profiling identified the circuit organization and mechanisms required for computing behaviorally relevant information during optic flow processing. Results Molecular profiling of functionally identified pretectal neurons using CaMPARI-seq To obtain genetic access to the functionally diverse types of optic flow-responsive pretectal neurons, we employed single-cell RNA sequencing of functionally identified neurons, which we termed CaMPARI-seq (Fig. 1a ). To specifically isolate optic flow-responsive cells, we fluorescently labeled them using CaMPARI2, a Ca 2+ -dependent photoconvertible protein that changes its fluorescence from green to red by UV illumination, thereby labeling neurons that are activated only during a specific temporal window (Supplementary Fig. 1a–c ) 32 , 33 . We generated a transgenic zebrafish line Tg(elavl3:NLS-CaMPARI2) expressing CaMPARI2 fused with the nuclear localization signal (NLS-CaMPARI2) in most neurons. We reasoned that the conventional cytosolic CaMPARI2 would not be ideal because the cytosolic CaMPARI2 signal distributed in the neurites would be eliminated during the following cell dissociation procedure required for the scRNA-seq, attenuating the CaMPARI2 signals in the dissociated cells. To circumvent this, we fused CaMPARI2 with the nuclear localization signal (NLS) so that the NLS-CaMPARI2 protein localized in the soma would not be lost during cell dissociation. Similar to the previously reported cytosolic CaMPARI in the zebrafish brain 32 , UV illumination at 405 nm induced the robust photoconversion of NLS-CaMPARI2 in response to the proconvulsive compound 4-aminopyridine (4-AP) (Supplementary Fig. 1d ). Fig. 1. Molecular profiling of functionally labeled pretectal neurons using CaMPARI-seq. Open in a new tab a Experimental workflow for identifying marker genes for optic flow-responsive pretectal neurons. b (Top) Moving and stationary conditions used for CaMPARI-seq. Note that a greater number of CaMPARI2 red cells are detected in the “moving” condition compared to the “stationary” condition. Scale bar, 50 μm. (Bottom) Ratio of cells derived from “moving” datasets over those from “stationary” datasets (MS index). c UMAP embedding of all sequenced cells after integrating two “moving” datasets and two “stationary” datasets. d Markers for clusters. Color shade represents the average expression level within a cluster (average expression). Dot size represents the percentage of cells expressing the marker in a cluster (percent expressed). e Relative frequency (y axis) of each cluster (x axis), ordered from highest to lowest. Each data point represents one dataset and their average is shown as bar graph. f Ratio of cell frequency of “moving” datasets over that of “stationary” datasets for each cluster (MS index). Red clusters with MS index higher than 1 were considered as candidates for the pretectal optic flow-responsive clusters (see text for details). Source data are provided on a public repository. To photoconvert NLS-CaMPARI2 during optic flow stimulation, we illuminated Tg(elavl3:NLS-CaMPARI2) larvae with UV LED coupled with the presentation of a visual stimulus that consisted of four binocular optic flow patterns (horizontally moving gratings); namely, clockwise, counter-clockwise, forward, and backward motions, referred to as the “moving” experiment (Fig. 1a and Supplementary Fig. 1a ). As a control, we used stationary gratings, referred to as the “stationary” experiment. During “moving” visual stimulation, neurons in the pretectum and its surrounding brain areas (such as tectum, thalamus, habenula, cerebellum, etc.) of Tg(elavl3:NLS-CaMPARI2) larvae were photoconverted by UV light (Fig. 1b and Supplementary Fig. 1e ). Consistent with the previously reported spatial distribution of optic flow-responsive cells 7 , 16 , 17 , 19 , photoconverted CaMPARI2-red neurons were enriched in the middle part of the pretectum, as well as the lateral pretectum; i.e., the pretectal migrated area M1, in the moving experiments. Significantly fewer neurons were photoconverted in these regions in the stationary experiments, indicating that the neurons that responded to the optic flow were reliably labeled (Fig. 1b and Supplementary Fig. 1e, f ). After the photoconversion during optic flow stimulation, we collected CaMPARI2-red neurons by FACS (Supplementary Fig. 2 ) and transcriptionally profiled them using single-cell RNA-seq. Our transcriptional profiling was not designed to capture activity-dependent transcriptional changes in response to visual stimulation, since the transcriptomic analyses were performed after the visual stimulation and photoconversion (typically minutes to hours later), by which time transiently expressed immediate early genes are already turned off. We obtained two replicate datasets from the “moving” condition and two others from the “stationary” control and integrated all datasets into a single dataset by the SCTransform approach (Supplementary Fig. 3 and Methods). Clustering of the integrated dataset yielded 30 clusters based on the expression of the marker genes (Fig. 1c, d ). To identify transcriptomic clusters in which optic flow-responsive cells were enriched, we focused on cell clusters that contained more abundant cells detected in the datasets obtained from the “moving” condition compared to those from the “stationary” condition (Fig. 1b ). For each cluster, we calculated the fold change in the proportion of cells between the “moving” and “stationary” conditions (the MS index; see the Methods section for details). If the proportion of cells in the “moving” condition is higher than that in the “stationary” condition, the MS index would be greater than 1. Among the 16 clusters that showed an MS index higher than 1, we excluded eight since their identities were unambiguously annotated based on published data (Supplementary Fig. 4a , see the Methods section for details). For example, consistent with their strong responses to optic flow stimuli 34 , the habenular clusters (clusters 11 and 17) exhibited high MS indices (Supplementary Fig. 5 ). However, they were excluded from subsequent analyses since we aimed at identifying pretectal subtypes in this study. The remaining eight clusters were considered candidate clusters that potentially contained optic flow-responsive cells in the pretectum (Fig. 1e, f ). CaMPARI-seq identifies pretectal neuron subtypes labeled by a set of genes including tcf7l2 and gad1b Subsequently, we performed reclustering of the cells using the eight pretectum candidate clusters, resulting in 16 clusters that were identifiable by one or a few specific marker genes (Fig. 2a, b and Supplementary Fig. 4b ). After this round of clustering, two clusters showed a high MS index, suggesting enrichment of the optic flow-responsive cells derived from the “moving” dataset compared with the “stationary” dataset (Fig. 2c, d ). Close inspection of one of the two candidate clusters (cluster 7) suggested that it corresponds to a putative hypothalamus region, based on the expression of sim1a and fezf2 35 . The other cluster (cluster 5) was characterized by the expression of tcf7l2 (transcription factor 7-like 2) , which has been shown to be expressed in the broad areas of the diencephalon (e.g., the thalamus, habenula, etc.) including the pretectum in early zebrafish larvae (2 dpf) 25 . Next, we sought to clarify whether this cluster corresponds to the pretectum but not the other regions of the diencephalon. We used a combination of genes whose expression domains delineate boundaries for the pretectum, thalamus, and prethalamic areas in the larval zebrafish brain 28 . A marker gene for the pretectum, pax7a , was expressed in cluster 5, whereas the thalamus marker gene, lhx9 , was not detectable in cluster 5 and was weakly expressed in cluster 1 (Fig. 2e, f ). Furthermore, we examined known marker genes for the P1 domain in the mouse embryonic diencephalon and prenatal pretectum, such as tal2 and gata3 27 , and found that both tal2 and gata3 were expressed in cluster 5 (Fig. 2b, e, f ). These expression patterns of a combination of genes suggest that cluster 5 corresponds to the pretectum. A similar set of genes (i.e., tcf7l2 , pax7a , tal2 , and gata3 ) were also expressed in cluster 15, which was marked by the expression of an additional gene, tyrosine hydroxylase ( th) , likely corresponding to dopaminergic neurons in the pretectum (Fig. 2b, f ). We excluded this cluster from further analyses because of its low MS index (Fig. 2c, d ). To determine whether tcf7l2 and tal2 are expressed in the pretectum at the late larval stage (5–6 dpf), we generated gene-specific Gal4 driver lines using CRISPR/Cas9-mediated genome editing and crossed them with the Tg(UAS:GFP) reporter line (Fig. 2g–j ). Similar to the expression pattern of tcf7l2 in early zebrafish embryos 25 , Tg(tcf7l2-hs:Gal4FF) labeled cells in the pretectum and other brain regions of diencephalic origin, as well as parts of the tectum and forebrain (Fig. 2g and Supplementary Movie 1 ). In the pretectum, Tg(tcf7l2-hs:Gal4FF) -labeled cells are tightly packed, indicating that the majority of cells expressed tcf7l2 (Fig. 2h ). This was confirmed by immunostaining of Tcf7l2 and HCR-FISH of tcf7l2 mRNA, which showed that the most pretectal cells expressed Tcf7l2 (Fig. 2k–n , Supplementary Fig. 6 ). Tg(tal2-hs:Gal4FF);Tg(UAS:GFP) labeled cells in the pretectum and tectum (Fig. 2i, j and Supplementary Movie 2 ). Compared with the ubiquitous expression of tcf7l2 , Tg(tal2-hs:Gal4FF) -labeled cells were more sparsely distributed in the pretectum. To determine whether cluster 5 neurons are excitatory or inhibitory, we examined the expression of slc17a6b (a.k.a. vglut2a , the marker for glutamatergic excitatory neurons) and gad1b (the marker for GABAergic inhibitory neurons). Interestingly, the majority of cells (80.47%) in this cluster expressed gad1b and only 5.73% of them expressed the glutamatergic marker gene slc17a6b (Fig. 2e, f ). Taken together, our CaMPARI-seq technique identified the transcriptional profile of putative optic flow-responsive GABAergic cells in the pretectum and obtained genetic access to them. Fig. 2. Identification of optic flow-responsive pretectal cluster and marker genes. Open in a new tab a UMAP embedding of pretectal candidate cells. Cells from two “moving” datasets and two “stationary” datasets are integrated and plotted. b Markers for clusters. Color shade represents the average expression level within a cluster (average expression). Dot size represents the percentage of cells expressing the marker in a cluster (percent expressed). c Relative frequency (y axis) of each cluster (x axis), ordered from highest to lowest. Each data point represents one dataset and average is shown as bar graph. d Ratio of cells derived from “moving” datasets over those from “stationary” datasets (MS index). e Gene expression plots of cells embedded in UMAP space. f Expression of tcf7l2 , tal2 , pax7a , lhx9 , gad1b , slc17a6b for each cluster. Color shade represents the average expression level within a cluster (average expression). g – j Maximum z-projection of a substack from Tg(tcf7l2-hs:Gal4FF);Tg(UAS:GFP) larva ( g , h ) and Tg(tal2-hs:Gal4FF);Tg(UAS:GFP) larva ( i , j ). Asterisks depict extraocular muscles. Arrowheads indicate lateral line neuromasts. k – n Immunostaining of Tcf7l2 co-stained with Synapsin in the dorsal ( k , l ) and ventral ( m , n ) pretectum. Scale bars, 50 μm. Source data are provided on a public repository. tcf7l2 + neurons cover most of the optic flow-responsive cells in the pretectum Next, we investigated whether the tcf7l2 -positive pretectum cluster corresponds to the optic flow-responsive neurons. We tested if tcf7l2 + neurons responded to optic flow stimuli by two-photon Ca 2+ imaging using Tg(tcf7l2-hs:Gal4FF);Tg(UAS:GCaMP6s) larvae at 6 dpf. Visual stimulation was presented from a 360-degree arena, consisting of eight different combinations of monocular and binocular horizontal gratings 7 (Fig. 3 ). We found that tcf7l2 -positive neurons, particularly those located near the posterior boundary of the tcf7l2- expressing domain, strongly responded to the optic flow stimuli (Fig. 3a, d ). When stimulated with a series of monocular and binocular motions, pretectal neurons in the zebrafish larvae can be classified into different functional types based on their binary (On or Off) response to each of the eight different motion phases 7 . Of the 256 (2 8 ) theoretically conceivable binary response types, monocular direction-selective and translation-selective response types are frequently represented in the pretectum using the pan-neuronal GCaMP line Tg(elavl3:H2B-GCaMP6s) as previously described 7 (Supplementary Fig. 7 ). The application of the same classification to the response profiles of tcf7l2 -positive neurons revealed that most of these major response types were highly represented (Fig. 3a–c ). Furthermore, the spatial distribution of optic flow-responsive cells in the pretectum of the Tg(tcf7l2-hs:Gal4FF);Tg(UAS:GCaMP6s) larvae revealed that monocular DS cells are located contralateral to the visually stimulated eye in most cases (e.g., neurons responding to motion in the right eye are located in the left brain), whereas translation-selective neurons (especially the more abundant BEL and BER types) are located in both pretectal hemispheres (Fig. 3d ). A similar spatial organization was observed for Tg(elavl3:H2B-GCaMP6s) larvae (Supplementary Fig. 7 ). These results suggest that tcf7l2 labeled most optic flow-responsive neurons in the pretectum. This observation is consistent with the uniform distribution of tcf7l2 expression within the pretectum (Fig. 2g, h, k–n ). Fig. 3. tcf7l2 + neurons in the pretectum respond to optic flow. Open in a new tab a A histogram of the frequent response types (“simple”, blue; “translation-selective”, green) detected in Tg(tcf7l2-hs:Gal4FF);Tg(UAS:GCaMP6s) larvae. The remaining response types were listed as unclassified (gray). Each dot represents an individual fish. Each bar represents the average number of cells per fish. Error bars represent ± SEM ( n = 5 fish). b Nomenclature of the visual stimulus protocol and response types as previously described 7 . c (Top) Raster plot showing the responses to the eight stimulus phases ( n = 546 cells, pooled from 5 fish). Cells are ordered according to their correlation coefficient to the corresponding regressor (within each response type). ΔF/F values across the 3 repetitions of the visual stimulation were averaged. (Bottom) Binary response type barcode of each cell. d Spatial distribution of the response types (as shown in ( a )) in a representative Tg(tcf7l2-hs:Gal4FF);Tg(UAS:GCaMP6s) fish. Cells are color-coded according to their response type. Average image of GCaMP6s signal of the dorsal, medial and ventral representative planes of the pretectum is used as a background. For the middle and ventral planes, cells detected in the same plane, as well as those detected in 10 µm above and 10 µm below the corresponding plane, are plotted. For the dorsal plane, cells detected in the same plane, as well as those detected in 10 µm above the corresponding plane, are plotted. For the data from all individual planes, see Supplementary Figs. 9 , 10 . A anterior, P posterior, L left, R right. Source data are provided on a public repository. Pretectal optic flow neurons can be further clustered into subsets Since tcf7l2 marks the majority of optic flow-responsive neurons in the pretectum, we asked whether the functional subtypes of the pretectum could be identified by further clustering of the tcf7l2 -positive pretectal cluster (i.e., cluster 5 in Fig. 2 ). We reclustered cells contained in the tcf7l2 -positive pretectal cluster, which yielded seven subclusters (Fig. 4a, b, e ). The MS index of each cluster revealed that all clusters had a higher contribution from the “moving” dataset compared to the “stationary” one to varying degrees (Fig. 4c, d ). We examined the expression pattern of the marker genes of different subclusters by hybridization chain reaction (HCR) RNA-FISH. Most genes were sparsely expressed in the pretectum area delineated by the expression of tcf7l2 (Fig. 4f and Supplementary Fig. 8 ). For example, nkx1.2lb (NK1 transcription factor related 2-like, b) 36 was expressed in the dorsal-posterior part of the pretectum, while mafaa (MAF bZIP transcription factor Aa) was expressed in the ventral-lateral part of the pretectum (Fig. 4g–q ). npy ( neuropeptide Y ) was expressed more anteriorly compared with nkx1.2lb , while penkb (proenkephalin b) was expressed more medially compared with mafaa (Supplementary Fig. 8 ). Notably, in most cases, the expression patterns of different marker genes identified from different clusters did not spatially overlap, suggesting that the transcriptional subtypes correspond to anatomically distinct cell types. As expected, for all of the clusters, the cells were largely gad1b + GABAergic, while only a minority of the cells were slc17a6b + glutamatergic (Fig. 4b ). Fig. 4. Pretectal neurons consists of diverse molecular types. Open in a new tab a UMAP embedding of pretectal subtypes. Cells from two “moving” datasets and two “stationary” datasets are integrated and plotted. b Markers for clusters. Color shade represents the average level of marker expression in a cluster (average expression). Dot size represents the percentage of cells expressing the marker in a cluster (percent expressed). c Relative frequency (y axis) of each cluster (x axis), ordered from highest to lowest. Each data point represents one dataset and average is shown as bar graph. d Ratio of cells derived from “moving” datasets over those from “stationary” datasets (MS index). e Gene expression plots of cells embedded in UMAP space. f – n Substack maximum z-projections of double HCR-FISH stains of mafaa and nkx1.2lb in Tg(tcf7l2-hs:Gal4FF);Tg(UAS:GFP) larva in the dorsal ( f – h ), coronal ( i – k ), and sagittal ( l – n ) views. o – q Schematic representation of the expression patterns in the pretectum in the dorsal ( o ), coronal ( p ) and sagittal ( q ) views. D dorsal, A anterior, R right, P posterior. Scale bars, 50 μm. Source data are provided on a public repository. Genetic targeting of mafaa- and nkx1.2lb- positive neurons identified anatomically and functionally distinct pretectal subtypes To determine whether transcriptional subtypes correspond to different anatomical features and/or functional subtypes of the optic flow-responsive pretectal neurons, we focused on the following two subclusters: mafaa + and nkx1.2lb + clusters. First, to label mafaa -positive cells, we generated a Gal4 knock-in transgenic line of mafaa gene Tg(mafaa-hs:Gal4FF) . As expected from the HCR RNA-FISH data (Fig. 4f, g ), neurons labeled in Tg(mafaa-hs:Gal4FF);Tg(UAS:GCaMP6s) were located in the lateral part of the pretectum, the pretectal migrated area M1, partially homologous to the mammalian accessory optic system 29 , 37 (Fig. 5a, b and Supplementary Movie 3 ). The projection of these mafaa -positive neurons remained close to the soma and did not extend outside of this local region, suggesting that they extended local projections within a neuropil region near their cell bodies (Fig. 5c–e ). Furthermore, the neurites of mafaa + neurons overlapped with the AF5 and AF6 of the retinal ganglion cell (RGC) axon terminals, where direction-selective RGCs innervate 13 , 14 (Fig. 5f–h ), suggesting a possibility that they may receive direct inputs from the DS-RGCs. HCR staining of Tg(mafaa-hs:Gal4FF);Tg(UAS:GFP) larvae revealed that the majority of mafaa + neurons in this region are GABAergic and not glutamatergic (Fig. 5i–m ). Ca 2+ imaging using Tg(mafaa-hs:Gal4FF);Tg(UAS:GCaMP6s) larvae revealed that mafaa -positive neurons responded exclusively to temporal motion presented to the contralateral eye (Fig. 5n, o , Supplementary Figs. 9 , 10 ), suggesting that these neurons correspond to a specific functional type, previously termed MoTL ( Mo nocular T emporalward motion presented in the L eft eye) and MoTR ( Mo nocular T emporalward motion presented in the R ight eye) 7 . Fig. 5. Anatomical and functional characteristics of mafaa + pretectal neurons. Open in a new tab a , b Substack maximum z-projections of Tg(mafaa-hs:Gal4FF);Tg(UAS:GCaMP6s);Tg(elavl3:lyn-tagRFP) larva. Asterisks in ( b ) denote extraocular muscles. c – e Substack maximum z-projections of Tg(mafaa-hs:Gal4FF);Tg(UAS:GFP);Tg(UAS:loxp-tdTomatoCAAX-loxp-GCaMP6s) larva (right hemisphere). Note that membrane-targeted tdTomato-labeled neurites are locally extended in the vicinity of the soma expressing GFP. f – h Distribution of mafaa + neurons and the retinal ganglion cell terminals in Tg(mafaa-hs:Gal4FF);Tg(UAS-hs:loxp-tdTomatoCAAX-loxp-GCaMP6s);Tg(isl2b:GFP) larva. i – m Substack maximum z-projections of double HCR-FISH stains of gad1b and slc17a6b in Tg(mafaa-hs:Gal4FF);Tg(UAS:GFP) larva ( i ). Single optical section of the region marked with dotted box in i ( j – m ). n (Top) A histogram of the frequent response types in Tg(mafaa-hs:Gal4FF);Tg(UAS:GCaMP6s) fish (“simple”, blue; “translation-selective”, green). The remaining response types are listed as unclassified (gray). Each dot represents an individual fish. Each bar represents the average number of cells per fish. Error bars represent ±SEM ( n = 7 fish). (Middle) Raster plot showing the average responses to the eight stimulus phases for all cells categorized as one of the frequent response types ( n = 73 cells, pooled from 7 fish). Cells are ordered according to their correlation coefficient to the corresponding regressor (within each response type). (Bottom) Binary response type barcode of each cell. o Spatial distribution of functionally identified cells in a representative Tg(mafaa-hs:Gal4FF);Tg(UAS:GCaMP6s) fish. ROIs distributed across 60 μm in depth (separated by 10 μm) are projected onto an average projection image of the GCaMP6s signals from six planes. In this volume, MoTL and MoTR response types were detected and very few of other response types were detected (for the other response types, see Supplementary Figs. 9 , 10 ). Arrowheads and asterisks denote mafaa + neuron cell bodies and extraocular muscles, respectively. A anterior, P posterior, L left, R right, M medial. Scale bars, 50 μm ( a , b , i ), 20 μm ( c , f ) and 10 μm ( j ). Source data are provided on a public repository. Second, we labeled nkx1.2lb -positive neurons using a Gal4 knock-in transgenic line of nkx1.2lb gene Tg(nkx1.2lb-hs:Gal4FF) . nkx1.2lb + neurons labeled in Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:GCaMP6s) larvae were located in the medial part of the pretectum (Fig. 6a–c and Supplementary Movie 4 ), consistent with its endogenous expression pattern (Fig. 4f, h ). Interestingly, nkx1.2lb -positive neurons extended commissural projections that crossed the midline through the posterior commissure (pc) and reached the contralateral pretectum (Fig. 6c ). Sparse labeling and neurite reconstruction of nkx1.2lb + neurons revealed that the majority of individual neurons (13 out of 14) exhibited commissural projections via the pc (Supplementary Fig. 11 ). To verify that commissural projection through the pc is derived from the pretectal neurons but not from other neurons outside of the pretectum labeled in the Tg(nkx1.2lb-hs:Gal4FF) line, we took advantage of the pretectum-specific expression of tcf7l2 and crossed Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:loxP-tdTomatoCAAX-loxP-GCaMP6s) fish with Tg(tcf7l2-hs:Cre) fish (Fig. 6d ). This intersectional genetic strategy exclusively labels the tcf7l2- positive (pretectal) population of nkx1.2lb + neurons. We observed that both the Gal4 + , Cre − (tdTomato-positive), and Gal4 + , Cre + (GCaMP-positive) populations crossed the midline, indicating that both pretectal nkx1.2lb + neurons and non-pretectal nkx1.2lb + neurons are the source of the commissural projections (Fig. 6e–g ). HCR staining of Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:GFP) larvae revealed that the majority of nkx1.2lb + neurons are GABAergic and not glutamatergic (Fig. 6h–l ). Ca 2+ imaging using Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:GCaMP6s) larvae revealed that nkx1.2lb -positive neurons responded broadly to the optic flow stimulus (Fig. 6m, n ). They covered most of the optic flow-response types that were found in the pan-neuronal (Supplementary Fig. 7 ) and Tg(tcf7l2-hs:Gal4FF) (Fig. 3 ) lines, and no noticeable bias of the labeled response types was found in the nkx1.2lb -positive neurons. Additionally, the spatial distribution of the optic flow-responsive neurons detected from the nkx1.2lb -positive neurons did not differ grossly from elavl3 - or tcf7l2 -labeled neurons (Fig. 6n and Supplementary Figs. 9 , 10 ). Together, our genetic targeting of the different subtypes of optic flow-responsive cells in the pretectum revealed functional and anatomical specializations. Fig. 6. Anatomical and functional characteristics of nkx1.2lb + pretectal neurons. Open in a new tab a – c Substack maximum z-projections of Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:GCaMP6s);Tg(elavl3:lyn-tagRFP) larva ( a ) corresponds to the pretectal region marked by the dotted box in a ( b , c ). Note that GCaMP6s-labeled neurites cross the midline via the posterior commissure. d Schematics for Gal4 and Cre intersectional strategy to refine genetic access to nkx1.2lb + pretectal neurons. In a nkx1.2lb Gal4 driver line, nkx1.2lb + neurons activate expression of tdTomatoCAAX through the UAS-hs:loxP-tdTomatoCAAX-loxP-GCaMP6s reporter. Additional pretectal Tg(tcf7l2-hs:Cre) line results in nkx1.2lb + pretectal neurons switching to GCaMP6s expression, while nkx1.2lb + non-pretectal ( tcf7l2 − ) cells remain to express tdTomatoCAAX . e – g Substack maximum z-projections of a Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS-hs:loxp-tdTomatoCAAX-loxp-GCaMP6s);Tg(tcf7l2-hs:Cre) larva. Note that GCaMP6s-labeled neurites, as well as tdTomatoCAAX-labeled neurites, cross the midline through the posterior commissure. h – l Substack maximum z-projections of double HCR-FISH stains of gad1b and slc17a6b in a Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:GFP) larva ( h ). Single optical section of the region marked with dotted box in h ( i – l ). Asterisks denote nkx1.2lb + cells that express gad1b . m (Top) A histogram of the frequent response types in Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:GCaMP6s) fish (“simple”, blue; “translation-selective”, green). The remaining response types were listed as unclassified (gray). Each dot represents an individual fish. Each bar represents the average number of cells per fish. Error bars represent ± SEM ( n = 7 fish). (Middle) Raster plot showing the average responses to the eight stimulus phases for all cells categorized as one of the frequent response types ( n = 295 cells pooled from 7 fish). Cells are ordered according to their correlation coefficient to the corresponding regressor (within each response type). (Bottom) Binary response type barcode of each cell. n Spatial distribution of monocular direction-selective neurons (left) and translation-selective neurons (right) from a representative Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:GCaMP6s) larva. Cells are color-coded according to their response type. Average image of GCaMP6s signal of the dorsal and middle representative planes of the pretectum is used as a background. For the middle plane, cells detected in the same plane, as well as those detected in 10 µm above and 10 µm below the corresponding plane, are plotted. For the dorsal plane, cells detected in the same plane, as well as those detected in 10 µm above the corresponding plane, are plotted. For the data from all individual planes, see Supplementary Figs. 9 , 10 . A anterior, P posterior, L left, R right. Scale bars, 50 μm ( a , b , e , h ) and 10 μm ( i ). Source data are provided on a public repository. nkx1.2lb -positive neurons are required for processing the translational optic flow Next, we tested whether the molecularly identified subtypes of the pretectal population are required for the optic flow-dependent behavior. Taking advantage of the relatively specific expression pattern of nkx1.2lb compared to mafaa (which is expressed in other neurons and extraocular muscles), we selectively ablated nkx1.2lb + pretectal neurons and tested the fish for two types of behavior, which are the optokinetic response (OKR) and optomotor response (OMR) 38 . The OKR is generally induced by clockwise or counter-clockwise rotational optic flow, while the OMR is induced by forward and backward translational optic flow. To specifically ablate the nkx1.2lb + neurons located in the pretectum, we took the intersectional approach and generated Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:loxP-TagBFP-loxP-epNTR-mScarlet);Tg(tcf7l2-hs:Cre) triple-transgenic larvae (Fig. 7a ). In these larvae, zebrafish-codon optimized enhanced-potency nitroreductase (epNTR) is specifically expressed by nkx1.2lb + neurons located in the pretectum. epNTR converts the prodrug metronidazole (Mtz) into a cytotoxic compound 39 , 40 . The intersectional labeling of nkx1.2lb pretectal neurons was restricted to the anterior portion of the nkx1.2lb + population (Fig. 7b–d and Supplementary Movie 5 ). This anterior portion corresponds to the region where most optic flow responses are detected (Fig. 6n ). The administration of Mtz to these larvae selectively and effectively ablated nkx1.2lb + pretectal neurons (Fig. 7b–d ). First, we tested OKR by presenting moving gratings using the visual stimulus arena that surrounds 360 degrees of the fish’s field of view (Fig. 7e ). We found no significant difference in the OKR index (see Methods for details) between the control and ablated fish (Fig. 7f and Supplementary Fig. 12 ). Second, we tested the OMR by presenting a moving grating stimulus underneath the elongated tank and recording the position of the larvae before and after the presentation of the moving grating stimuli (Fig. 7e ). While groups of control larvae swam in the direction of the moving grating, groups of ablated larvae swam a significantly shorter distance (Fig. 7g,i , p = 0.03788). To better characterize the impaired OMR in ablated fish, we analyzed swim kinematics using a head-fixed OMR assay with binocular forward-moving grating stimuli 41 . OMR bouts in ablated fish were significantly shorter in duration, while no significant difference in bout rate (bout/sec) and mean tail beat frequency (Supplementary Fig. 14 ). Furthermore, whereas control fish exhibited largely left–right symmetric directionality characteristic of forward swimming, ablated fish more frequently displayed left–right asymmetric directionality (Supplementary Fig. 14 ). This finding suggests that the computation of the swim directionality is compromised in ablated fish. These defects are unlikely to result from changes in basal swimming activity in the ablated fish, as ablating nkx1.2lb ⁺ pretectal neurons had no significant effect on spontaneous locomotor activity on either a white background or a stationary grating pattern (Fig. 7h and Supplementary Figs. 12 , 13 ). Furthermore, we investigated swim kinematics in response to looming stimuli using head-fixed preparations 42 . We did not observe any major differences in loom-evoked tail responses between control and ablated fish, either in mean tail beat frequency or in latency to stimulus onset (Supplementary Fig. 15 ). Fig. 7. nkx1.2lb + pretectal neurons are required for processing translational optic flow. Open in a new tab a Schematics for Gal4 and Cre intersectional strategy to refine genetic access to nkx1.2lb + pretectal neurons. In a nkx1.2lb Gal4 driver line, nkx1.2lb + neurons activate expression of TagBFP through a UAS:loxP-TagBFP-loxP-epNTR-mScarlet reporter. Additional pretectal Tg(tcf7l2-hs:Cre) line results in nkx1.2lb + pretectal neurons switching to epNTR-mScarlet expression, while nkx1.2lb + non-pretectal cells continue to express TagBFP. b – d Maximum z-projections of substack of a Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS-hs:loxp-TagBFP-loxp-epNTR-mScarlet);Tg(tcf7l2-hs:Cre) transgenic fish before ( b , c ) and after ( d ) treatment with metronidazole (Mtz). Scale bars, 100 μm ( b ) and 50 μm ( c ). e Optokinetic response and optomotor swimming assays. f Box plot showing the OKR index of control (Ctrl, n = 7) and ablated (Mtz, n = 7) fish. The OKR index was calculated by counting the saccades in the expected direction during 1 min of grating stimulation. All individual data points are shown, with each dot representing one fish. Boxes represent the interquartile range (25th–75th percentile), with the median indicated by the center line; whiskers extend to 1.5× the interquartile range. Statistical analysis was performed using two-sided Mann–Whitney U test (ns, p = 0.4047). g Box plot showing the average distance moved in the direction of motion during OMR for control (Ctrl, n = 7 groups) and ablated (Mtz, n = 7 groups) fish. Each group consisted of 6 fish. Each dot represents one group. Boxes represent the interquartile range (25th–75th percentile), with the median indicated by the center line; whiskers extend to 1.5× the interquartile range. Statistical analysis was performed using Welch’s two-sided t -test (∗ p < 0.05, p = 0.03788). h Box plot showing the locomotor activity of control (Ctrl, n = 12) and ablated (Mtz, n = 12) fish. Total distance moved during 90 seconds is shown. All individual data points are shown, with each dot representing one fish. Boxes represent the interquartile range (25th–75th percentile), with the median indicated by the center line; whiskers extend to 1.5× the interquartile range. Statistical analysis was performed using Welch’s two-sided t -test (ns, p = 0.3038). i Position of fish before (pre-stim) and after (post-stim) the exposure to a moving visual stimulus (30 s). Each small dot represents one of the 6 fish tested in one group and large dot indicates the mean position of the 6 fish in a group. Control larvae are either Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:loxp-TagBFP-loxp-epNTR-mScarlet);Tg(tcf7l2-hs:Cre) transgenic larvae untreated with Mtz or Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:loxp-TagBFP-loxp-epNTR-mScarlet) transgenic larvae treated with Mtz. Source data are provided on a public repository. To test whether the loss of nkx1.2lb + pretectal neurons influences the composition of optic flow response profiles, we performed pan-neuronal calcium imaging using Tg(elavl3:H2B-GCaMP6s);Tg(nkx1.2lb-hs:Gal4FF);Tg(tcf7l2-hs:Cre);Tg(UAS:loxP-TagBFP-loxP-epNTR-mScarlet) transgenic larvae. We found that the number of translation-selective response cells was significantly reduced in the Mtz group compared with the control group (Wilcoxon rank-sum test, FDR-adjusted p = 0.037), whereas no significant difference was observed for the monocular direction-selective category ( p = 0.76) (Supplementary Fig. 16 ). In conclusion, commissural inhibitory nkx1.2lb + neurons are required specifically for OMR but not for OKR, and this requirement is associated with a reduction in translation-selective optic flow-responsive neurons. These results suggest that nkx1.2lb + neurons are required for processing translational optic flow but not rotational optic flow (see the Discussion section). Discussion By strategically combining the functional labeling of neurons using CaMPARI and subsequent scRNA-seq, we developed a technique to analyze the transcriptomic features of functionally active neurons during optic flow stimulation in larval zebrafish. Most pretectal neurons expressed the tcf7l2 gene and encompassed most of the previously characterized optic flow response types. Further clustering revealed several transcriptomic subtypes of pretectal neurons that were distinguishable from one another by anatomical and functional characteristics. Finally, we molecularly identified inhibitory commissural pretectal neurons that were necessary for optic flow-dependent behavior, specifically for the OMR, which depends on the translational optic flow; however, it is dispensable for the OKR, which depends on the rotational optic flow. In this study, we integrated the functional tagging of neurons by CaMPARI2 and subsequent single-cell transcriptomic analyses. BARseq 43 and spatial RNA-seq 44 both link molecular data to spatial context: BARseq for neuronal projections, and spatial RNA-seq for gene expression, although the latter may not achieve perfect single-cell resolution. In contrast, CaMPARI-seq, while lacking spatial information, uniquely enables the selective labeling and transcriptomic profiling of single neurons that are functionally active during defined behaviors or stimuli. This functional specificity is not achievable with BARseq or spatial RNA-seq, making CaMPARI-seq a powerful tool for linking neuronal activity to molecular identity. Although the physiological properties and transcriptomic profiles of single cells have generally been independently studied, a couple of methods have been described to link the physiological and transcriptomic features of individual neurons 45 . One such method, Patch-seq, is based on the electrophysiological recording and subsequent RNA-seq analysis of single neurons 46 – 48 . Although it is a powerful tool to study both aspects of the neuronal diversity of the same cells, the throughput of cell sampling is generally limited. In contrast, the optical tagging of functional neuron types and the subsequent RNA-seq analysis we employed in this study offer the advantage of sampling thousands of neurons simultaneously. Furthermore, since CaMPARI2-based labeling uses light, it provides a precise temporal window for labeling neurons of interest, compared with other techniques based on immediate early gene promoters, which have been primarily used for behavior where the temporal kinetics are relatively slow 49 , 50 . A combination of CaMPARI2 and scRNA-seq applied in the mouse identified previously known transcriptomic types of layer 2/3 neurons in the visual cortex and validated that these transcriptomic types exhibit distinguishable negative and positive prediction error responses 31 . Our current study extends this research by identifying previously unknown molecular markers for the behaviorally relevant functional cell types in the pretectum. A previous scRNA-seq study characterized the transcriptomic diversity of neurons in the diencephalon, including the pretectum, thalamus, and prethalamus, as well as several surrounding brain regions in the larval zebrafish 29 . Some of the pretectal marker genes identified in their dataset, such as npy , penkb , and zic1 , were also detected in our dataset. Most of these clusters are consistently GABAergic in our data and that of Sherman et al 29 . The mafaa and nkx1.2lb genes that we focused on in our study were not described as representative marker genes in the study by Sherman et al. We speculate that our functionally targeted approach successfully dissected out a rare and functionally relevant population of neurons that have been missed in the conventional anatomically based approach. Binocular integration of the optic flow critically depends on the interhemispheric transfer of DS information in the pretectum 7 , 16 . Previous anatomical studies conducted using single-neuron reconstructions and tracer experiments in larval and adult zebrafish pretectum have demonstrated that some pretectal neurons cross the brain side through either of the two commissural fibers: the posterior commissure (pc) or postoptic commissure (poc) 11 – 13 . The pc represents a conserved landmark located in the dorsal part of the pretectum of vertebrates, while the poc is located at the extreme ventral part of the diencephalon 51 . Furthermore, electron microscopy-based reconstruction of functionally identified optic flow-responsive cells in larval zebrafish revealed that the optic flow-responsive cells frequently cross the midline via either the pc or poc and project to the contralateral pretectum 52 . Our study identified nkx1.2lb as a specific marker for pretectal neurons that cross the midline via the pc. Genetic markers for the poc-crossing pretectal neurons are yet to be identified. Contralaterally projecting pretectal neurons identified in the EM reconstruction had dendrites in ipsilateral retinorecipient neuropil areas that receive DS inputs (AF5 and AF6) and/or medial pretectal neuropil (mPN) and projected axons to the contralateral mPN and/or AF6 in the pretectum 52 . The similarity between their anatomical features suggests that nkx1.2lb + commissure neurons may receive direct DS retinal inputs and innervate pretectal neurons located on the contralateral side (Fig. 8 ). Future studies are needed to identify the pre- and postsynaptic neurons of the nkx1.2lb + pretectal neurons. Fig. 8. Model for the optic flow processing circuit for controlling OKR and OMR. Open in a new tab a Summary of previous results 7 . Clockwise (CW) and counter-clockwise (CCW) rotational optic flow mainly induces the OKR. In the pretectum, monocular optic flow-responsive cells are active during rotational optic flow, whereas rotation-selective binocular cells are rarely observed. For processing of the forward (FW) and backward (BW) translational optic flow that induces OMR, monocular optic flow-responsive cells and binocular translation-selective cells are active. b Proposed circuit model. Pretectal output neurons are depicted based on 1) projection neurons that project to cerebellum and reticular formation in the hindbrain 13 and 2) pretectal neurons that extend dendrites into AF5 (which contains the axon terminals of direction-selective RGCs) and project axons to the contralateral abducens nucleus (nVI) 77 . PC posterior commissure, AF5 arborization field 5, AF6 arborization field 6, mPN medial pretectal neuropil. Our previous functional imaging study proposed a parsimonious circuit wiring diagram of the optic flow processing circuit in the pretectum, where the monocular and binocular response types are hierarchically organized to compute the rotational and translational optic flow 7 . The current study provides updates on this model. First, while the previous model predicted that all of the commissural pretectal neurons would be excitatory, we found that at least some of the commissural pretectal connections are inhibitory, since the inhibitory nkx1.2lb + pretectal neurons project contralaterally. The discrepancy can be explained, at least in part, by the direct or indirect connectivity of inhibitory inputs. The previous model proposed that interhemispheric inhibition is mediated indirectly by commissural excitatory relay neurons, which, in turn, activate local inhibitory neurons. We now propose that the inhibition is provided directly by commissural inhibitory neurons. Second, the previous model predicted that the monocular response types consist of both excitatory and inhibitory neurons, while the binocular response types consist exclusively of excitatory neurons. Our data suggest that the majority of pretectal transcriptomic types, including mafaa + and nkx1.2lb + neurons, express the inhibitory neuron marker gad1b . For instance, we found some of the binocular neurons to be inhibitory using the nkx1.2lb transgenic line. Although we do not rule out the possible contribution of excitatory neurons, the majority of optic flow-responsive neurons are inhibitory. The excitatory connections predicted in the previous model could be derived from the DS-RGC inputs, which are glutamatergic and excitatory. Together, our current study provides direct experimental evidence for anatomical projections and the sign of connectivity (i.e., excitatory vs. inhibitory) of the functional circuit model of the pretectum. Our data showed that nkx1.2lb + neurons consist of both monocular and binocular types and the ablation of pretectal nkx1.2lb + neurons impaired the OMR but left the OKR intact. Prior functional imaging studies have shown that the majority of active neuron types during rotational motion in the horizontal plane are monocular simple neurons (namely MoNL, MoNR, MoTL, and MoTR), suggesting that rotational motion and the resulting rotational OKR are largely computed by monocular simple cells in the pretectum (Fig. 8 ) 7 , 19 , 21 . These observations led to the idea that during rotational motion, the optic flow information from both eyes is separately processed and not combined in the pretectum 7 , 19 , 21 . In contrast, during translational motion, binocular translation-selective cells are active in addition to the simple monocular neurons 7 , 19 , 21 . Moreover, the activities of these binocular translation-selective cells are selectively suppressed during rotational motion, meaning that they are, either directly or indirectly, inhibited by the simple cells that respond to the motion sensed by the other eye during rotation (e.g., the FEL is inhibited by MoTR during CW motion). The intact OKR after pretectal nkx1.2lb + neuron ablation suggests that the computation of rotational optic flow does not require monocular cells labeled in the nkx1.2lb transgenic line. This indicates that other monocular cells that were unlabeled in the nkx1.2lb line play essential roles in the OKR. We speculate that one such candidate is mafaa + neurons located in the migrated pretectal region M1 12 , 53 . Neurons in the M1 have been shown to play critical roles in OKR 37 . mafaa + neurons closely resemble those neurons in terms of 1) direction tuning to motion, 2) proximity to RGC arborization fields AF5 and AF6 that harbor DS-RGC axon terminals 13 , and 3) their neurite projections in the local pretectal neuropil 37 . Unfortunately, pretectal mafaa + neuron ablation is technically difficult at present even when using the intersectional genetics with Tg ( mafaa-hs:Gal4FF) and Tg(tcf7l2-hs:Cre) lines, since both transgenic lines label the eye and tail muscles (see for example Figs. 2 g, h and 5b, c ). Given the critical role of M1 neurons in the OKR, we speculate that mafaa + neurons may contribute significantly to the motion processing and/or generation of the OKR. On the other hand, the processing of the translational optic flow during OMR requires nkx1.2lb + pretectal neurons. The requirement of nkx1.2lb⁺ pretectal neurons for OMR was associated with reduction in translation-selective optic flow-responsive neurons. The observed OMR defect can be caused by 1) the ablation of monocular neurons that provide inhibitory inputs to generate binocular neurons, 2) the ablation of binocular neurons that encode translation-selective information, or a combination of both processes. In any case, the essential role of nkx1.2lb + inhibitory neurons for OMR but not for OKR is consistent with the previous model in which processing of the rotational optic flow does not require binocular inhibitory integration while the processing of the translational optic flow and behavior requires inhibitory connectivity between hemispheres. Although optogenetic stimulation of mafaa ⁺ and nkx1.2lb ⁺ pretectal neurons would provide further support for their roles in optic flow-dependent behavior, performing these experiments is currently challenging due to technical limitations. For example, the Tg(mafaa-hs:Gal4FF) line lacks sufficient specificity to selectively activate mafaa ⁺ pretectal neurons, as it also labels extraocular muscles located in close proximity to these neurons. In addition, nkx1.2lb ⁺ neurons comprise a broad range of optic flow response types; thus, simultaneous activation of the entire nkx1.2lb ⁺ population may induce non-physiological network activity, potentially obscuring specific behavioral effects. Future studies will be needed to overcome these limitations by (1) implementing advanced photostimulation approaches, such as a 3D holographic system, and/or (2) employing intersectional optogenetic strategies to restrict opsin expression specifically to pretectal populations. The cell-type-specific transcriptome is generally used to classify neuron diversity 54 , 55 and is believed to give rise to distinct morphology, connectivity, and function 56 – 58 . Consistent with this view, we found that one pretectal transcriptomic type corresponds to one morphological type: mafaa + and nkx1.2lb + pretectal neurons project locally and contralaterally, respectively. However, in terms of function, we found that nkx1.2lb + neurons consisted of multiple optic flow-response types, suggesting that a single transcriptomic type can consist of multiple functional types. This supports the conclusion of a prior work showing that transcriptionally similar optic tectum neurons can be functionally divergent 59 . The functional diversity of nkx1.2lb + neurons may arise from different inputs received by each neuron (e.g., nasally tuned vs. temporally tuned DS inputs). In summary, our functionally targeted transcriptional profiling provided a molecular handle on the optic flow processing circuit and gained new insights into the circuit mechanism for computing behaviorally relevant visual information. The molecular markers we identified may help elucidate how diverse pretectal neuron types are generated and wired together to form the functional optic flow processing circuit during development. Future studies will identify comprehensive catalogs of the transcriptomic, functional, and morphological neuron types underlying the optic flow processing and synaptic connectivity of the identified neurons. Methods Animal care and transgenic zebrafish Adult and larval zebrafish ( Danio rerio ) were housed and handled according to standard procedures. All protocols were approved by the Animal Care and Use committees of the National Institutes of Genetics and the RIKEN Center for Brain Science and animal experiments were performed according to regulations of both institutions. We used the following previously described transgenic lines: Tg(UAS:GFP)nns19, Tg(UAS:GCaMP6s)mpn156, Tg(UAS:mCherry)s1984t, Tg(elavl3:LOXP-DsRed-LOXP-EGFP)nns16Tg, Tg(elavl3:H2B-GCaMP6s)jf5Tg, Tg(elavl3:lyn-tagRFP)mpn404 (a.k.a. HuC:lyn-tagRFP). Transgenic fish were kept in either a TL or TLN (nacre) background and larvae lacking trunk pigmentation (outcrossed to TLN, nacre) were used in the experiment. Zebrafish larvae were raised in E3 solution on a 14/10 h light/dark cycle until 5–7 days post-fertilization (dpf). Plasmids To generate the pTol2-elavl3 :NLS-CaMPARI2 plasmid, the CaMPARI2 coding sequence was amplified from pAAV_hsyn_NES-his-CaMPARI2-WPRE-SV40 plasmid (gift from Dr. Eric Schreiter, Addgene plasmid, Cat# 101060), fused with nuclear localization signal (NLS), and inserted into a pTol2 vector that contains elavl3 pan-neuronal promoter. To generate the pTol2-UAS-hsp:epNTR-mScarlet plasmid, fragments containing heatshock promoter (hsp) sequence from the UAShspzGCaMP6s 60 , epNTR (codon optimized for zebrafish) 40 and mScarlet (codon optimized using CodonZ) 61 were inserted between UAS and SV40 sequences synthesized pTol2-UAS vector using In-Fusion cloning. To generate the pTol2-UAS-hs:loxP-tdTomatoCAAX-loxP-GCaMP6s plasmid, three fragments for loxP-tdTomatoCAAX 62 , SV40-loxP and GCaMP6s 41 were PCR amplified and cloned between a hsp sequence and SV40 sequence of pTol2-UAS-hsp:epNTR-mScarlet plasmid by In-Fusion cloning (Takara). To generate the pTol2-UAS-hsp:loxP-TagBFP-loxP-epNTR-mScarlet plasmid, epNTR-mScarlet were inserted into the UAS-hsp:loxP-TagBFP-loxP vector. To generate the pTol2-UAS:Cre plasmid, the Cre coding sequence was amplified by PCR from the pTol2-HuC:Cre-shSV40-cold-heart-fwd plasmid 63 and cloned between the hsp sequence and SV40 sequence of pTol2-UAS-hsp:epNTR-mScarlet plasmid. To generate the pTol2-HuC:CreV318K-shSV40-cold-heart-fwd plasmid, the coding sequence carrying the CreV318K mutation 64 , which confers reduced recombination efficiency, was generated by combining two PCR fragments amplified from pCR8GW-Cre-FRT-kan-FRT-cold_heart 62 as the template. Fragment 1 was amplified using the primer pair pCR8GW-Cre-InFU1 (GTACAAAAAAGCAGGCTCCgaattcGCCCTTGCCACCATGtCCAAT) and CreV318K_InFL1 (AATATTTTTATTGGTCCAGCCACCAGC). Fragment 2 was amplified using the primer pair CreV318K_InFU1 (ACCAATAAAAATATTGTCATGAACTAT) and SV40pA-pCRGW8-InFL1 (TTGTACAAGAAAGCTGGGTCgaattcGGACAAACCACAACTAGAAT). Subsequently, the two fragments corresponding to CreV318K were cloned between the HuC promoter and the SV40 sequence of the pTol2-HuC:Cre-shSV40-cold-heart-fwd plasmid 63 using In-Fusion cloning (Takara). Transgenic line establishment Tg(elavl3:NLS-CaMPARI2), Tg(UAS-hs:loxP-tdTomatoCAAX-loxP-GCaMP6s) and Tg(UAS-hs:loxP-TagBFP-loxP-epNTR-mScarlet) transgenic lines were generated using the standard Tol2 transposon system. Tg(tcf7l2-hs:Gal4FF) , Tg(tcf7l2-hs:Cre) , Tg(tal2-hs:Gal4FF) , Tg(mafaa-hs:Gal4FF) , and Tg(nkx1.2lb-hs:Gal4FF) lines were generated using the published CRISPR knock-in method 65 with some modifications. Briefly, single guide RNA sites were designed in a region 200–600 bp upstream of the transcription start site of the target gene using CCTop 66 . gRNA sequences for each line are as follows: tcf7l2 (CGGCACAGCTAGAATCATATTGG), tal2 (ATATTTATTTAGTATAGGAAAGG), mafaa (CGACGTGGAAGGTGAACGCGAGG), nkx1.2lb (TAAGCTAGAGCTGAGACGGGCGG). CRISPR-Cas9 RNP complex was prepared according to the manufacturer’s instructions (IDT). In short, gRNA was produced by annealing custom-synthesized crRNA (Alt-R CRISPR-Cas9 crRNA, IDT) with tracrRNA (IDT, Cat# 1072532) in Nuclease-Free Duplex Buffer (IDT, Cat# 11-05-01-12) and subsequently incubated with Cas9 protein (IDT, Cat# 1081060) at 37 °C for 15 min. Another CRISPR-Cas9 RNP complex targeting either G-bait or M-bait sequence of the donor plasmid was prepared separately. Donor plasmid containing Gbait-hsp70:Gal4FF, Mbait-hsp70:Gal4FF or Mbait-hsp70:Cre 65 (gift from Dr. Shin-ichi Higashijima) was added to a final concentration of 25 ng/μl. The freshly prepared mix of two CRISPR-Cas9 RNP complexes and the donor plasmid was injected into the cell of Tg(UAS:mCherry) , Tg(UAS-hs:loxP-tdTomatoCAAX-loxP-GCaMP6s) , Tg(UAS:GCaMP6s) or Tg(elavl3:LOXP-DsRed-LOXP-EGFP) transgenic zygotes. Transient expressors were raised and screened for germline transmission. Gal4- or Cre-positive larvae (F1 fish) were used to confirm correct insertion of the donor plasmid into the target gene locus by PCR using primer pairs designed to amplify fragments that span the region between the genomic locus of the target gene and the donor plasmid. CaMPARI photoconversion during visual stimulation To perform photoconversion of CaMPARI2, we used a custom-built microscope equipped with a 405 nm UV LED (Thorlabs, Cat# M405L4) and a 20x water immersion objective lens (Olympus, UMPLFLN20XW, NA0.5) (Supplementary Fig. 1b ). Continuous, wide-field UV illumination was delivered through the objective of the microscope. The UV intensity measured at the sample was around 2.5 mW and the power density was around 326.1 mW/cm 2 . To validate the photoconversion efficacy of NLS-tagged CaMPARI2, Tg(elavl3:NLS-CaMPARI2) transgenic larvae between 6 and 7 dpf were embedded in 2% agarose in the center of a dish with a diameter of 3 cm and UV light was applied to the center of the pretectal area for 10 min in the presence of 800 μM 4-aminopyridine (4-AP). To photoconvert neurons during visual stimulation, Tg(elavl3:NLS-CaMPARI2) transgenic larvae were similarly embedded in agarose. We determined the target pretectum region for photoconversion based on anatomical landmarks. The anterior–posterior position was determined using the midpoint between the posterior edge of the habenula and the midbrain–hindbrain boundary. The dorsal–ventral position was determined by focusing ventral to the optic tectum through the objective (Supplementary Fig. 1c ). Sinusoidal grating stimuli were generated by a custom-written script using PsychoPy and presented to a custom-built oLED arena that consists of four oLED screens (3.81 inch, Wisecoco) that covered 360 degrees around the fish. Binocular gratings moved horizontally in four phases in the following order: clockwise, counter-clockwise, forward and backward (duration: 10 s each at spatial frequency of 0.033 cycles/degree and temporal frequency of 2 cycles/sec, interspersed with 2 s stationary gratings, Supplementary Fig. 1a ) with 10 times repeats. The total length of the visual stimulus protocol was about 8 min 20 s. In separate experiments using fish embedded with eyes free, we confirmed that fish showed normal optokinetic response under this UV photostimulation condition. This result indicated that the UV light does not interfere with the detection of the visual stimuli. However, we cannot exclude the possibility that cells responsive to UV illumination 67 are photoconverted in our experiments. In the control visual stimulation, the same photostimulation protocol was used except that the speed of grating stimulus was set to zero (stationary gratings). Confocal imaging was performed on a Zeiss LSM900 microscope to visualize CaMPARI2 green and red fluorescent proteins. When comparing photoconversion efficiency among different conditions (Fig. 1 and Supplementary Fig. 1 ), same confocal imaging settings (laser power and gain) were used across conditions. Cell dissociation and fluorescence-activated cell sorting (FACS) Photoconversion of CaMPARI2 from green to red state is irreversible 32 . Consistent with this, we observed that the CaMPARI2 red signal persisted for several hours after photoconversion. After photoconversion, larvae were kept at 27.5 degrees for maximally 4 h until brain dissection. Brain was dissected in ice-cold Neurobasal medium (Thermo Fisher Scientific, Cat# 21103049) supplemented with B-27 (Thermo Fisher Scientific, Cat# 17504044). After the eyes were removed from the head, anterior to the habenula and posterior to the otolith was cut away to increase the proportion of pretectal cells (Fig. 1a ). Cell dissociation was performed as previously described with some modifications 29 . Tissues were pooled from 50 to 60 larvae and digested in 25 U/ml papain, 130 U/ml DNaseI, 2 mM L-cysteine in oxygenated Ames at 37 °C for 36 min. Every 12 min, tissues were triturated by pipetting, followed by a final incubation for 5 min. The cell suspension was then passed through a pre-wet 40 µm filter (Falcon, Cat# 352340). To stop the digestion, the papain solution was replaced by papain inhibitor solution containing 15 mg/ml ovomucoid and 15 mg/ml BSA. Tissue was gently dissociated by trituration using a broad end tip pipette in the papain inhibitor solution. To wash the cell suspension, cells were pelleted at 300 g for 7 min at RT and resuspended in FACS buffer (oxygenated Ames containing 0.25% BSA (Sigma-Aldrich, Cat# A4161), 10 mM HEPES (Sigma-Aldrich, Cat# H0887), 25 U/mL DNase I (Thermo Fisher Scientific, Cat# 90083), 5 mM MgCl 2 ). CaMPARI2 red cells were sorted based on green and red fluorescence using SONY Cell Sorter SH800S equipped with 488 nm & 561 nm lasers (Supplementary Fig. 2 ). In a pilot experiment, Tg(elavl3:NLS-CaMPARI2) larvae without photoconversion were used to determine sorting gates. First, FACS gates were set after 50,000 recorded events to sort live single-cells. Similar gates were used across experiments. Specifically, a scatter gate was initially set using forward scattering (FSC-A) and back scattering (BSC-A) to select the live cell population. Next, Forward Scatter Area (FSC-A) and Forward Scatter Height (FSC-H) were used to eliminate debris as much as possible and isolate single-cells. Then, CaMPARI2 green and CaMPARI2 red fluorescence were used to collect CaMPARI2 red cells labeled by photoconversion of Tg(elavl3:NLS-CaMPARI2) larvae. The gating was set to include the CaMPARI2 red fluorescent cell population, which was not observed in the Tg(elavl3:NLS-CaMPARI2) control larvae that did not undergo photoconversion. Overall, 60–70% of the population consisted of live single cells, and ~8% of those were CaMPARI2 red cells. These cells were sorted using a 100 µm nozzle (20 PSI) and collected to a 2 ml protein LoBind Eppendorf tube filled with 1 ml oxygenated Ames containing 1.0% BSA, 10 mM HEPES. A total of 130,000–150,000 CaMPARI2 red cells were collected from 50 to 60 larvae. After FACS was completed, cell suspension was centrifuged for 5 min at 100 g at RT and the pellet was resuspended in 300 µl of Ames medium/0.5% BSA. This washing step was repeated three times, and finally, the cells were resuspended in 30 μl of Ames/BSA solution. To check cell density and viability, cells were stained with a dead cell stain Trypan Blue (Sigma-Aldrich, Cat# T8154) and both live and dead cells were counted using a DIC microscope. Cell density was typically between 300 and 800 cells/µl with a viability rate of over 90%. Single cell RNA-seq Droplet RNA sequencing was performed according to the manufacturer’s instructions with no modifications (10x Genomics Chromium System, Chromium Next GEM Single Cell 3’ Reagent Kits v3.1). A single-chromium chip channel was loaded with a cell suspension, aiming to capture 5000 cells with a doublet rate of <5%. For the “moving” condition, 2 replicates were collected from 2 experiments. For the “stationary” condition, 2 replicates were collected from 2 experiments. cDNA libraries were sequenced on a NovaSeq 6000 (illumina) or DNBSEQ-G400 (MGI Tech) to a depth of ~60,000 reads per cell. Computational analysis of single cell transcriptomics data Alignment of gene expression reads, preprocessing and batch integration Initial preprocessing was performed using the Cell Ranger software suite (version v6.0.1, 10x Genomics) following standard guidelines. “cellranger count” was used to align the reads to a custom reference genome created by combining the improved zebrafish transcriptome annotation file (GTF for v4.3.2) 68 and the FASTA file (GRCz11) with CaMPARI2, using “cellranger mkref”. Each biological replicate resulted in 3000–5000 estimated number of cells with 3500–5000 median unique molecular identifier (UMI) counts per cell. Output from cellranger was loaded into Seurat R package 69 (version 4.1.1) using the Read10X function, with each condition representing different experiment: stationary1, stationary2, moving1, and moving2. The data were stored as Seurat objects, with a minimum threshold of 200 genes per cell and a minimum of 3 cells per gene for inclusion. For quality control, cells with greater than 10% mitochondrial gene expression were excluded from further analysis. Cells were also filtered based on the number of detected genes, retaining those with between 200 and 5000 genes. This quality control was applied uniformly to all datasets. The numbers of cells before and after quality control for each dataset were as follows: stationary1, 3117 cells before QC and 3093 cells after QC; stationary2, 3884 and 3869; moving1, 3735 and 3678; moving2, 4812 and 4804, respectively. Following filtering, the datasets were integrated using the SCTransform function. The datasets for stationary1, stationary2, moving1, and moving2 were stored in a list and processed through the SCTransform function. For each transformed dataset, the 3000 most variable genes were selected as possible integration features using SelectIntegrationFeatures. We then followed the standard integration pipeline, employing PrepSCTIntegration to prepare the data for integration. Integration was performed using the FindIntegrationAnchors function, where stationary1 and stationary2 datasets were used as the reference. Normalization was done using the SCTransform. The resulting integration anchors were used to combine the datasets into a single Seurat object using IntegrateData. To reduce dimensionality, principal component analysis (PCA) was performed using the RunPCA function. Subsequently, UMAP was generated using the first 20 principal components (PCs) to visualize the data in a two dimensional space (RunUMAP). Clustering was performed using the FindNeighbors and FindClusters functions with the PCA reduction and a resolution of 0.8. Marker genes for each cluster were identified using the FindAllMarkers function, selecting only positive markers with a log fold change threshold of 0.25. To assess transcriptional relatedness, hierarchical clustering was performed with BuildClusterTree on the integrated assay using the first 20 principal components. The resulting dendrogram was used to reorder clusters for downstream visualization (Supplementary Fig. 4a ). We annotated clusters based on the expression data of marker genes using the Zebrafish Information Network website (zfin.org). Identification of optic flow-responsive pretectal cluster To identify clusters in which optic flow responsive cells are enriched, we calculated the composition of different clusters within each dataset and compared them across moving and stationary conditions. For each cluster, we calculated the fold change of the proportion of “moving” datasets (average of the two replicates) with respect to that of “stationary” datasets (average of the two replicates) (MS index) using the following formula MS index = M e a n p r o p o r t i o n o f m o v i n g d a t a s e t s ( % ) M e a n p r o p o r t i o n o f s t a t i o n a r y d a t a s e t s ( % ) Among the 16 clusters that showed MS index higher than 1 (Fig. 1f ), we excluded 8 clusters since their identities were unambiguously annotated based on published data (Supplementary Fig. 4a ). Cluster 7 was classified as telencephalon/prethalamus due to the expression of dlx5a and arxa , cluster 9 as hindbrain based on the expression of slc30a2 and hoxb3a , cluster 11 as dorsal habenula due to the expression of gng8 and GPR151 , cluster 16 as trigeminal sensory neurons based on p2rx3b and p2rx2 , cluster 17 as ventral habenula due to the expression of kiss1 and GPR139 , cluster 22 as cerebellum based on the expression of gsg1l and trpc3 , and cluster 26 as hypothalamus based on the expression of pdyn and oprd1b . Cluster 23, which expressed krt4 and krt91 , was determined not to be neuronal. The remaining 8 clusters, “2”, “3”, “6”, “8”, “12”, “15”, “20”, and “21” were considered as candidate clusters that potentially contain optic flow-responsive cells in the pretectum (Fig. 1f ). The subset was split by the “orig.ident” variable, and 2000 variable genes were identified for each subset using the “vst” selection method. The datasets were integrated using SelectIntegrationFeatures and FindIntegrationAnchors, followed by integration using the IntegrateData function. The standard workflow for clustering and visualization was performed, including normalization, PCA (with 10 PCs), and UMAP. Clustering was performed with a resolution of 0.8. Marker genes for each cluster were identified using the FindAllMarkers function, selecting only positive markers with a log fold change threshold of 0.25. As a result, 16 new clusters were obtained (Fig. 2 ), and the MS index calculation and brain region annotation were performed as described above (Supplementary Fig. 4b ). For the subset analysis of candidate pretectal clusters, hierarchical clustering was performed using BuildClusterTree on the integrated assay with the first 10 principal components. The dendrogram guided cluster reordering for visualization (Supplementary Fig. 4b ). Re-clustering of the pretectum cluster (cluster 5) was performed in a similar manner (Fig. 4 ). However, due to the low number of cells, the k.weight parameter in the IntegrateData function was set to the minimum sample size of 68 (moving1: 165 cells, moving2: 132 cells, stationary1: 68 cells, stationary2: 106 cells). Immunohistochemistry Larvae (6 dpf) were fixed in 4% PFA/PBS at 4 °C overnight and processed for whole-mount immunostaining. Samples were incubated with primary antibodies for 7 days at 4 °C and then with secondary antibodies for 5 days at 4 °C. Primary antibodies used were as follows: chicken anti-GFP (1:1000, Invitrogen, Cat# A10262), rat anti-RFP (1:1000, ChromoTek, Cat# 5F8), mouse anti-TCF4 (also known as TCF7L2) clone 6H5-3 (1:400, Merck Millipore, Cat# 05-511), rabbit anti-Synapsin 1/2 (1:1000, Synaptic Systems, Cat# 106002). Secondary antibodies were as follows: anti-chicken Alexa Fluor 488 (1:250, Thermo Fisher Scientific, Cat# A-11039), anti-rat Alexa Fluor 555 (1:250, Thermo Fisher Scientific, Cat# A-21434), anti-mouse Alexa Fluor 555 (1:250, Thermo Fisher Scientific, Cat# A-21127), and anti-rabbit Alexa Fluor 647 (1:250, Thermo Fisher Scientific, Cat# A-21244). Confocal imaging was performed on a Zeiss LSM900 microscope. Hybridization chain reaction (HCR) fluorescent in situ RNA labeling (HCR-FISH) Larvae (5–6 dpf) were fixed in 4% PFA at 4 °C overnight and processed for HCR fluorescent in situ RNA labeling (HCR-FISH) according to the manufacturer’s instructions (Molecular Instruments). Tg(tcf7l2-hs:Gal4FF);Tg(UAS:GFP), Tg(mafaa-hs:Gal4FF);Tg(UAS:GFP) , Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:GFP) larvae were hybridized with two probe sets, each of which was detected using HCR amplifiers conjugated to Alexa Fluor 546 or Alexa Fluor 647. Imaging was performed on a Zeiss confocal microscope (LSM900). Functional calcium imaging Larvae expressing GCaMP were mounted in agarose (LMP-agarose, 2% w/v in E3 medium). We used a two-photon microscope (Bergamo II, Thorlabs) equipped with a 20× water immersion objective (XLUMPLFLN 20x, NA1.0, Olympus) and a fiber laser with a fixed-wavelength at 920 nm (ALCOR-920, Spark Lasers). The visual stimuli were presented to the fish using a custom-built oLED arena that consists of four flat oLED screens (3.81 inch, Wisecoco) arranged in square and covered 360 degrees around the fish. To ensure monocular stimulation, the central 45° portion of the visual field was covered with black paper. Sinusoidal grating stimuli were generated by a custom-written script using PsychoPy. RGB color space of the oLED screens were set to [1, −1, −1] (red stimulus). Furthermore, magenta gelatin filter (No. 32, Kodak) was applied in front of the oLED screens to block green light emitted from the screens. Visual stimulus protocol was as reported previously 7 . In each experiment session, gratings moved horizontally in eight phases (3 s each at spatial frequency of 0.033 cycles/degree and temporal frequency of 2 cycles/sec, interspersed with 10 s stationary gratings). Four of the eight phases are monocular, four are binocular: 1) left nasalward, 2) left temporalward, 3) right temporalward, 4) right nasalward, 5) clockwise, 6) counter-clockwise, 7) forward, 8) backward. The sequence of eight phases was repeated three times. During visual stimulation, GCaMP fluorescence was imaged at about 3 Hz using a Bergamo II multiphoton microscope (Thorlabs), with image acquisition controlled by ThorImageLS software (version 4.3 or 4.4) (0.501 µm/pixel, laser power ca. 15 mW after the objective, imaging region of about 256 ×256 µm). We typically imaged a volume centered around the pretectal neuropil with the z-step size of 10 μm. Imaging of the habenula was performed across five planes separated by 10 μm. Calcium imaging analysis To eliminate motion artifacts, raw GCaMP image time series was registered to median image of all frames using TurboReg ImageJ plugin. To identify neurons that show correlated activity to visual stimuli, we performed regressor-based correlation analysis as described previously 7 , 70 . Briefly, the visual stimulus data and the registered image time series were loaded using a custom MATLAB script. To detect cells that respond to optic flow stimuli, we generated a motion-sensitive variable that is 1 during motion phases and 0 during motionless phases of the visual stimuli. This variable was then convolved with a kernel of GCaMP6s kinetics (τdecay = 3 s for GCaMP6s and τdecay = 4 s for H2B-GCaMP6s) to generate a motion-sensitive variable, called regressor. Based on the correlation value with the regressor in each pixel of the time series, a heat map (Z-score map) was generated and we selected individual cells (ROIs) for a subsequent correlation analysis. Care was taken to only include individual cell-like structures and avoid neuropil signals. Response type classification was performed as previously described 7 . In short, since there were 8 different stimulus phases in the visual stimulus, a total of 2^8 (256) binary response patterns were conceivable. In order to classify the response of each cell as a particular response type, 256 convolved regressors were generated that corresponded to the 2^8 possible combinations of monocular and binocular stimulations in nasalward and temporalward direction. The regressor that showed the highest correlation with the cell’s response (best regressor) defined the response type of the cell. Average response was calculated by averaging ΔF/F during each of the 8 motion phases across the 3 repetitions. Single-neuron tracing and morphological analysis Single neurons were sparsely labeled by transient microinjection at the single-cell stage, using one of the following combinations of plasmids (1–5 ng/µl) and transgenic embryos. 1) pTol2-14xUAS:GCaMP6s plasmid 41 injected into Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:loxP-tdTomatoCAAX-loxP-GCaMP6s) embryos, 2) pTol2-HuC:Cre-shSV40-cold-heart-fwd plasmid 63 injected into Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:loxP-tdTomatoCAAX-loxP-GCaMP6s) embryos, 3) pTol2-HuC:CreV318K-shSV40-cold-heart-fwd plasmid (generated in this study) injected into Tg(nkx1.2lb-hs:Gal4FF);Tg(UAS:loxP-tdTomatoCAAX-loxP-GCaMP6s) embryos, 4) pTol2-UAS:loxP-tdTomatoCAAX-loxP-GCaMP6s 62 and pTol2-HuC:Cre-shSV40-cold-heart-fwd plasmids co-injected into Tg(nkx1.2lb-hs:Gal4FF) embryos, and 5) pTol2-UAS:loxP-tdTomatoCAAX-loxP-GCaMP6s and pTol2-UAS:Cre (generated in this study) plasmids co-injected into Tg(nkx1.2lb-hs:Gal4FF) embryos. At 6 dpf, injected larvae were fixed in 4% paraformaldehyde at 4 °C overnight and processed for whole-mount immunostaining as described above. Confocal imaging was performed on a Zeiss LSM900 microscope. Individual neurons were semi-automatically traced using the open-source Fiji/ImageJ plugin Simple Neurite Tracer (SNT), saved as SWC files, and registered to Synapsin-staining standard brain of the mapzebrain.org using ANTs 71 . The ANTs registration command used was as follows antsRegistration \-d 3 --float 1 -v 1 \--winsorize-image-intensities [0.01,0.99] \ -o [${output1},${output2}.nrrd] \-- interpolation WelchWindowedSinc \--use-histogram-matching 0 \-r [${template},${input1}.nrrd,1] \-t rigid[0.1] \-m MI[${template}.nrrd,${input1}.nrrd,1,32,Regular,0.25] \-c [200×200×200×0,1e-8,10] \-- shrink-factors 12×8×4×2 \--smoothing-sigmas 4×3×2×1vox \-t Affine[0.1] \-m MI[${template}.nrrd,${input1}.nrrd,1,32,Regular,0.25] \-c [200×200×200×0,1e-8,10] \-- shrink-factors 12×8×4×2 \--smoothing-sigmas 4×3×2×1vox \-t SyN[0.1,6,0] \-m CC[${template}.nrrd,${input1}.nrrd,1,2] \-c [200×200×200×200×10,1e-7,10] \--shrink- factors 12×8×4×2×1 \--smoothing-sigmas 4×3×2×1×0vox \ Subsequently, single-neuron tracings in SWC format were converted to CSV files via a custom Python script and co-registered using the antsApplyTransformsToPoints function. antsApplyTransformsToPoints \-d 3 \-i ${input2.csv} \-o ${output4} \-t ${output1} 1InverseWarp.nii.gz \-t ${output1}0GenericAffine.mat,1 The registered neuron tracings were converted back to SWC format and visualized in the SNT Reconstruction Viewer, overlaid with the mapZebrain reference brain 71 . Cell ablations Larvae expressing the enhanced potency nitroreductase (epNTR)-mScarlet were sorted at 3 dpf. Cell ablation was induced at 5 dpf by incubating larvae with 2 mM metronidazole (Mtz) (Sigma-Aldrich, Cat# M3761) for 24 hours, followed by washout and recovery with E3 medium for more than 20 hours. Behavioral experiments were carried out at 7 dpf. Successful ablation was confirmed by confocal imaging before and after treatment with Mtz. Behavior assays for optokinetic response (OKR), optomotor response (OMR) and spontaneous swimming We adapted previously established methods for OKR assays 72 with some modifications. Sinusoidal grating stimuli were generated by a custom-written script using PsychoPy and presented to a custom-built oLED arena that consists of four oLED screens (3.81 inch, Wisecoco) that covered 360 degrees around the fish. Binocular gratings moved horizontally in clockwise and counter-clockwise directions (duration: 60 s at spatial frequency of 0.003 cycles/degree and temporal frequency of 0.33 cycles/sec, interspersed with 60 s stationary gratings), with 2 times repetitions. Eye movements were recorded at 10 frames per second (fps) using a CMOS camera (Mako U-051B, Allied Vision) equipped with a lens (MLH-3XMP, MORITEX), with image acquisition controlled by StreamPix7 software (NorPix). Angles of left and right eyes were calculated using a custom ImageJ macro based on the “Moment Calculator” plug-in. The “OKR index” was calculated by counting the saccades in the expected direction, with CCW saccades during the CW motion for the right eye and CW saccades during the CCW motion for the left eye, focusing on those with an amplitude greater than 7.5 degrees. OMR was measured using previously established methods 73 with some modifications. A group of 6 larvae were placed in a custom-built acrylic chamber (2 × 2 × 25 cm, RIKEN machine shop). The chambers were placed on a 23-inch monitor, on which visual stimulus was presented. Fish behavior was recorded at 10 fps using a CMOS camera (CS505MU, Thorlabs) equipped with a lens (LM16HC, Kowa), with image acquisition controlled by ThorCam software (Thorlabs). Before the OMR tests, animals were kept in the chamber for at least 2.5 min to allow acclimation to light and temperature conditions of the experiment. Sinusoidal grating stimuli were generated by a custom-written script using PsychoPy. Moving gratings (grating width of 1 cm and temporal frequency of 1 cycle/sec) were in motion for 30 s. Average distance traveled was calculated by the difference between the average positions of 6 fish in one chamber before and after the moving grating stimulus. To measure spontaneous swimming, one larva was placed in a plastic weighing dish (70 × 70 × 22 mm, BIO-BIK). After acclimation for 3 min, swimming was recorded for 1.5 min at 10 fps using the same CMOS camera described above, controlled by ThorCam. Swimming trajectories were tracked using ImageJ Plugin TrackMate 74 . Free-swimming assay using stationary grating stimulus Free-swimming behavior with a stationary grating pattern was recorded using a CMOS camera (Mako U-051B, Allied Vision) equipped with a lens (LM16HC, Kowa) and StreamPix7 software (NorPix). A gray-and-white stripe pattern with a stripe width of 5 mm was printed on tracing paper and placed directly on an infrared (IR) light source (MDBL-CIR70, MORITEX). Gray stripes were rendered as RGB (178, 178, 178) relative to white (RGB 255, 255, 255). A single larva was placed in a watch glass (diameter of 60 mm) positioned on top of the printed tracing paper. After 3 min of acclimation, swimming behavior was recorded for 1.5 min at 10 fps. Swimming trajectories were tracked using the ImageJ plugin TrackMate 74 . Head-fixed behavior assays For head-fixed OMR and looming assays, all agarose posterior to the swim bladder was carefully removed, allowing the tail to move freely. For the OMR assay, sinusoidal grating stimuli were generated using a custom-written script in PsychoPy and presented in the same oLED arena used for the OKR assay. Binocular gratings moved horizontally in the forward direction (spatial frequency: 0.025 cycles/degree; temporal frequency: 2 cycles/sec). Each trial consisted of two 5 s presentations of moving gratings, separated by a 10 s stationary grating period. Three trials were performed per fish, with 30 s of stationary gratings presented between trials. For the looming assay, looming stimuli were generated using a custom-written PsychoPy script and projected using a DLP projector (LightCrafter 4500, Texas Instruments). Larvae were mounted by fixing the head with agarose at the edge of a partially cut 3-cm-diameter dish. The dish was secured to a custom-built acrylic stage using waterproof silicone and placed inside a custom acrylic chamber (6 × 6 × 6 cm), which was filled with E3 medium. A rear-projection translucent film was attached to one side of the chamber, onto which the looming stimulus was projected. The distance between the projection surface and the larva was adjusted to ~1 cm. After 10 min of acclimation, responses to looming stimuli were recorded. The looming stimulus was a dark expanding disk on a gray background, starting from 2° and ending in 60° in diameter. The expansion size-to-speed ratio (l/v) was 60 ms. After reaching the maximum size, the looming stimulus remained stationary for 5 s 75 . The looming disk was presented in the center of the visual field. Three trials were performed per fish, with a uniform gray background presented for 50 s between trials. Tail movements were recorded using a CMOS camera (Mako U-051B, Allied Vision) equipped with a lens (MLH-3XMP, MORITEX) and controlled with StreamPix7 software (NorPix). Recordings were performed under IR illumination at 300 fps for the OMR assay and 400 fps for the looming assay. Tail kinematics were analyzed by tracking 20 points along the tail using the open-source software ZebraZoom (v1.34.89) 76 . From the tail positions tracked by ZebraZoom, we extracted bout timing and kinematic parameters to calculate bout rate (bout/sec), mean tail beat frequency (Hz), bout duration (sec), max tail angle (degrees) and response latency (sec). Quantification and statistical analysis To assess the normality of behavioral data, a Q-Q plot and the Shapiro–Wilk normality test were performed. If normality was confirmed, Welch’s two-sample t -test was used (OMR, spontaneous locomotor activity, free-swimming assay using stationary grating stimulus, looming assay, head-fixed OMR). For data that did not follow a normal distribution, the Mann–Whitney U test was applied (OKR). For calcium imaging response-type analyses, group differences were evaluated using the Wilcoxon rank-sum test. P -values were adjusted for multiple comparisons using the Benjamini–Hochberg procedure. For comparing the directionality of OMR bouts, Mardia–Watson–Wheeler test was used. All statistical analyses were conducted using R (version 4.1.0). Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information (12.9MB, pdf) 41467_2026_71371_MOESM2_ESM.pdf (247.5KB, pdf) Description of Additional Supplementary Files Supplementary Movie 1 (72.5MB, avi) Supplementary Movie 2 (57.5MB, avi) Supplementary Movie 3 (180MB, avi) Supplementary Movie 4 (173.3MB, avi) Supplementary Movie 5 (188.3MB, avi) Reporting Summary (95.4KB, pdf) Transparent Peer Review file (280KB, pdf) Acknowledgements We thank Shin-ichi Higashijima (National Institute for Basic Biology) for providing the donor plasmids for the CRISPR-Cas9 knock-in, Shachar Sherman (Max Planck Institute for Biological Intelligence), Grigorios Oikonomou, and David Prober (Caltech) for sharing cell dissociation protocols prior to publication, Shigehiro Kuraku (National Institute of Genetics) for advice on bioinformatics analysis. We thank Masahiko Hibi, Ayjan Urazbayeva, and Hitoshi Okamoto for their discussion and feedback on our manuscript. We thank the members of the Kubo laboratory for their technical assistance and advice and the Research Resource Division of RIKEN CBS for fish care and technical assistance. Bioinformatics computations (Cell Ranger analysis) were performed on the NIG supercomputer at ROIS National Institute of Genetics. We are grateful to National BioResource Project (NBRP) Zebrafish Japan for providing Tg(UAS:GFP) and Tg(elavl3:LOXP-DsRed-LOXP-EGFP) lines. This study was supported by JSPS Grant-in-Aid for Scientific Research (KAKENHI) (19K23787, 20K15906, 23K14293 to K.M.; 24K02008 to K.K.; 17K20147, 21H02586, 22K21353 to F.K.), “Strategic Research Projects” grant from ROIS (Research Organization of Information and Systems) (to K.M.), and Tomizawa Jun-ichi & Keiko Fund of the Molecular Biology Society of Japan for Young Scientist (to F.K.). Author contributions K.M. and F.K. conceived and designed the study, and wrote the paper with inputs from C.H.W., H.K., A.T., T.S. and K.K. K.M., C.H.W., H.K., A.T. and F.K. conducted experiments. K.M., C.H.W. and F.K. analyzed the data. T.S. and K.K. generated the pTol2 UAS-hsp:epNTR-mScarlet plasmid. Peer review Peer review information Nature Communications thanks Inbal Shainer and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Data availability Single-cell RNA sequencing raw and processed data files are available through the NCBI Gene Expression Omnibus (GEO) under the accession number GSE320054 . Data supporting the findings of this study, including source data, are publicly available on the RIKEN CBS data sharing platform (10.60178/cbs.20260306-001). Code availability Custom codes used in this study are publicly available on the RIKEN CBS data sharing platform (10.60178/cbs.20260306-001). Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-71371-6. References 1. Masseck, O. A. & Hoffmann, K.-P. Comparative neurobiology of the optokinetic reflex. Ann. N. Y. Acad. Sci. 1164 , 430–439 (2009). [ DOI ] [ PubMed ] [ Google Scholar ] 2. Srinivasan, M. V. & Zhang, S. W. Visual navigation in flying insects. Int. Rev. Neurobiol. 44 , 67–92 (2000). [ DOI ] [ PubMed ] [ Google Scholar ] 3. Brockerhoff, S. E. et al. A behavioral screen for isolating zebrafish mutants with visual system defects. PNAS 92 , 10545–10549 (1995). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Muto, A. et al. Forward Genetic Analysis of Visual Behavior in Zebrafish. PLoS Genet. 1 , e66 (2005). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Neuhauss, S. C. F. et al. Genetic Disorders of Vision Revealed by a Behavioral Screen of 400 Essential Loci in Zebrafish. J. Neurosci. 19 , 8603–8615 (1999). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Cazin, L., Precht, W. & Lannou, J. Pathways mediating optokinetic responses of vestibular nucleus neurons in the rat. Pflug. Arch. 384 , 19–29 (1980). [ DOI ] [ PubMed ] [ Google Scholar ] 7. Kubo, F. et al. Functional architecture of an optic flow-responsive area that drives horizontal eye movements in zebrafish. Neuron 81 , 1344–1359 (2014). [ DOI ] [ PubMed ] [ Google Scholar ] 8. Schiff, D., Cohen, B. & Raphan, T. Nystagmus induced by stimulation of the nucleus of the optic tract in the monkey. Exp. Brain Res. 70 , 1–14 (1988). [ DOI ] [ PubMed ] [ Google Scholar ] 9. Giolli, R. A., Blanks, R. H. I. & Lui, F. The accessory optic system: basic organization with an update on connectivity, neurochemistry, and function. In Progress in Brain Research (ed. Büttner-Ennever, J. A.) vol. 151 407–440 (Elsevier, 2006). [ DOI ] [ PubMed ] 10. Vanegas, H. & Ito, H. Morphological aspects of the teleostean visual system: a review. Brain Res. Rev. 6 , 117–137 (1983). [ DOI ] [ PubMed ] [ Google Scholar ] 11. Yáñez, J., Suárez, T., Quelle, A., Folgueira, M. & Anadón, R. Neural connections of the pretectum in zebrafish (Danio rerio). J. Comp. Neurol. 526 , 1017–1040 (2018). [ DOI ] [ PubMed ] [ Google Scholar ] 12. Baier, H. & Wullimann, M. F. Anatomy and function of retinorecipient arborization fields in zebrafish. J. Comp. Neurol. 529 , 3454–3476 (2021). [ DOI ] [ PubMed ] [ Google Scholar ] 13. Kramer, A., Wu, Y., Baier, H. & Kubo, F. Neuronal Architecture of a Visual Center that Processes Optic Flow. Neuron 103 , 118–132.e7 (2019). [ DOI ] [ PubMed ] [ Google Scholar ] 14. Yildizoglu, T., Riegler, C., Fitzgerald, J. E. & Portugues, R. A Neural Representation of Naturalistic Motion-Guided Behavior in the Zebrafish Brain. Curr. Biol. 30 , 2321–2333.e6 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Orger, M. B., Kampff, A. R., Severi, K. E., Bollmann, J. H. & Engert, F. Control of visually guided behavior by distinct populations of spinal projection neurons. Nat. Neurosci. 11 , 327–333 (2008). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Naumann, E. A. et al. From Whole-Brain Data to Functional Circuit Models: The Zebrafish Optomotor Response. Cell 167 , 947–960.e20 (2016). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Portugues, R., Feierstein, C. E., Engert, F. & Orger, M. B. Whole-brain activity maps reveal stereotyped, distributed networks for visuomotor behavior. Neuron 81 , 1328–1343 (2014). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Matsuda, K. & Kubo, F. Circuit Organization Underlying Optic Flow Processing in Zebrafish. Front. Neural Circuits 15 , 709048 (2021). [ DOI ] [ PMC free article ] [ PubMed ] 19. Wang, K., Hinz, J., Haikala, V., Reiff, D. F. & Arrenberg, A. B. Selective processing of all rotational and translational optic flow directions in the zebrafish pretectum and tectum. BMC Biol. 17 , 29 (2019). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Wang, K., Hinz, J., Zhang, Y., Thiele, T. R. & Arrenberg, A. B. Parallel Channels for Motion Feature Extraction in the Pretectum and Tectum of Larval Zebrafish. Cell Rep. 30 , 442–453.e6 (2020). [ DOI ] [ PubMed ] [ Google Scholar ] 21. Zhang, Y., Huang, R., Nörenberg, W. & Arrenberg, A. B. A robust receptive field code for optic flow detection and decomposition during self-motion. Curr. Biol. 32 , 2505–2516.e8 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 22. Dragomir, E. I., Štih, V. & Portugues, R. Evidence accumulation during a sensorimotor decision task revealed by whole-brain imaging. Nat. Neurosci. 23 , 85–93 (2020). [ DOI ] [ PubMed ] [ Google Scholar ] 23. Krapp, H. G., Hengstenberg, R. & Egelhaaf, M. Binocular Contributions to Optic Flow Processing in the Fly Visual System. J. Neurophysiol. 85 , 724–734 (2001). [ DOI ] [ PubMed ] [ Google Scholar ] 24. Wylie, D. R. W., Bischof, W. F. & Frost, B. J. Common reference frame for neural coding of translational and rotational optic flow. Nature 392 , 278–282 (1998). [ DOI ] [ PubMed ] [ Google Scholar ] 25. Brożko, N. et al. Genoarchitecture of the Early Postmitotic Pretectum and the Role of Wnt Signaling in Shaping Pretectal Neurochemical Anatomy in Zebrafish. Front. Neuroanat. 16 , 838567 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Puelles, L. & Rubenstein, J. L. R. Forebrain gene expression domains and the evolving prosomeric model. Trends Neurosci. 26 , 469–476 (2003). [ DOI ] [ PubMed ] [ Google Scholar ] 27. Virolainen, S.-M., Achim, K., Peltopuro, P., Salminen, M. & Partanen, J. Transcriptional regulatory mechanisms underlying the GABAergic neuron fate in different diencephalic prosomeres. Development 139 , 3795–3805 (2012). [ DOI ] [ PubMed ] [ Google Scholar ] 28. Wullimann, M. F., Mokayes, N., Shainer, I., Kuehn, E. & Baier, H. Genoarchitectonics of the larval zebrafish diencephalon. J. Comp. Neurol. 532 , e25549 (2024). [ DOI ] [ PubMed ] [ Google Scholar ] 29. Sherman, S., Arnold-Ammer, I., Schneider, M. W., Kawakami, K. & Baier, H. Retina-derived signals control pace of neurogenesis in visual brain areas but not circuit assembly. Nat. Commun. 14 , 6020 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 30. Carmona, L. M. et al. Corticothalamic neurons in motor cortex have a permissive role in motor execution. Nat. Commun. 16 , 4735 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. O’Toole, S. M., Oyibo, H. K. & Keller, G. B. Molecularly targetable cell types in mouse visual cortex have distinguishable prediction error responses. Neuron 111 , 2918–2928.e8 (2023). [ DOI ] [ PubMed ] [ Google Scholar ] 32. Fosque, B. F. et al. Neural circuits. Labeling of active neural circuits in vivo with designed calcium integrators. Science 347 , 755–760 (2015). [ DOI ] [ PubMed ] [ Google Scholar ] 33. Moeyaert, B. et al. Improved methods for marking active neuron populations. Nat. Commun. 9 , 4440 (2018). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Lavian, H., Prat, O., Petrucco, L., Štih, V. & Portugues, R. Visual motion and landmark position align with heading direction in the zebrafish interpeduncular nucleus. Nat. Commun. 16 , 9924 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Wolf, A. & Ryu, S. Specification of posterior hypothalamic neurons requires coordinated activities of Fezf2, Otp, Sim1a and Foxb1.2. Development 140 , 1762–1773 (2013). [ DOI ] [ PubMed ] [ Google Scholar ] 36. Bae, Y.-K. et al. Expression of sax1/nkx1.2 and sax2/nkx1.1 in zebrafish. Gene Expr. Patterns 4 , 481–486 (2004). [ DOI ] [ PubMed ] [ Google Scholar ] 37. Wu, Y., Dal Maschio, M., Kubo, F. & Baier, H. An Optical Illusion Pinpoints an Essential Circuit Node for Global Motion Processing. Neuron 108 , 722–734.e5 (2020). [ DOI ] [ PubMed ] [ Google Scholar ] 38. Portugues, R. & Engert, F. The neural basis of visual behaviors in the larval zebrafish. Curr. Opin. Neurobiol. 19 , 644–647 (2009). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Curado, S., Stainier, D. Y. R. & Anderson, R. M. Nitroreductase-mediated cell/tissue ablation in zebrafish: a spatially and temporally controlled ablation method with applications in developmental and regeneration studies. Nat. Protoc. 3 , 948–954 (2008). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Tabor, K. M. et al. Direct activation of the Mauthner cell by electric field pulses drives ultrarapid escape responses. J. Neurophysiol. 112 , 834–844 (2014). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Thiele, T. R., Donovan, J. C. & Baier, H. Descending control of swim posture by a midbrain nucleus in zebrafish. Neuron 83 , 679–691 (2014). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Temizer, I., Donovan, J. C., Baier, H. & Semmelhack, J. L. A Visual Pathway for Looming-Evoked Escape in Larval Zebrafish. Curr. Biol. 25 , 1823–1834 (2015). [ DOI ] [ PubMed ] [ Google Scholar ] 43. Chen, X. et al. High-Throughput Mapping of Long-Range Neuronal Projection Using In Situ Sequencing. Cell 179 , 772–786.e19 (2019). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Moses, L. & Pachter, L. Museum of spatial transcriptomics. Nat. Methods 19 , 534–546 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 45. Yagishita, H. & Sasaki, T. Integrating physiological and transcriptomic analyses at the single-neuron level. Neurosci. Res. 10.1016/j.neures.2024.05.003 (2024). [ DOI ] [ PubMed ] 46. Cadwell, C. R. et al. Electrophysiological, transcriptomic and morphologic profiling of single neurons using Patch-seq. Nat. Biotechnol. 34 , 199–203 (2016). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Fuzik, J. et al. Integration of electrophysiological recordings with single-cell RNA-seq data identifies neuronal subtypes. Nat. Biotechnol. 34 , 175–183 (2016). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Scala, F. et al. Phenotypic variation of transcriptomic cell types in mouse motor cortex. Nature 598 , 144–150 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 49. Chen, M. B., Jiang, X., Quake, S. R. & Südhof, T. C. Persistent transcriptional programmes are associated with remote memory. Nature 587 , 437–442 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Sun, W. et al. Spatial transcriptomics reveal neuron–astrocyte synergy in long-term memory. Nature 627 , 374–381 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 51. Wilson, S. W., Ross, L. S., Parrett, T. & Easter, S. The development of a simple scaffold of axon tracts in the brain of the embryonic zebrafish, Brachydanio rerio. Development https://www.semanticscholar.org/paper/The-development-of-a-simple-scaffold-of-axon-tracts-Wilson-Ross/7e0cf3dfb244127e8ab50f162437aa9533aa16f1 (1990). [ DOI ] [ PubMed ] 52. Svara, F. et al. Automated synapse-level reconstruction of neural circuits in the larval zebrafish brain. Nat. Methods 19 , 1357–1366 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 53. Mueller, T. & Wullimann, M. F. BrdU-, neuroD (nrd)- and Hu-studies reveal unusual non-ventricular neurogenesis in the postembryonic zebrafish forebrain. Mech. Dev. 117 , 123–135 (2002). [ DOI ] [ PubMed ] [ Google Scholar ] 54. Zeng, H. What is a cell type and how to define it? Cell 185 , 2739–2755 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 55. Zeng, H. & Sanes, J. R. Neuronal cell-type classification: challenges, opportunities and the path forward. Nat. Rev. Neurosci. 18 , 530–546 (2017). [ DOI ] [ PubMed ] [ Google Scholar ] 56. Goetz, J. et al. Unified classification of mouse retinal ganglion cells using function, morphology, and gene expression. Cell Rep. 40 , 111040 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 57. Huang, W. et al. Linking transcriptomes with morphological and functional phenotypes in mammalian retinal ganglion cells. Cell Rep. 40 , 111322 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 58. Kölsch, Y. et al. Molecular classification of zebrafish retinal ganglion cells links genes to cell types to behavior. Neuron 109 , 645–662.e9 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 59. Shainer, I. et al. Transcriptomic neuron types vary topographically in function and morphology. Nature 638 , 1023–1033 (2025). [ DOI ] [ PMC free article ] [ PubMed ] 60. Muto, A. et al. Activation of the hypothalamic feeding centre upon visual prey detection. Nat. Commun. 8 , 15029 (2017). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 61. Horstick, E. J. et al. Increased functional protein expression using nucleotide sequence features enriched in highly expressed genes in zebrafish. Nucleic Acids Res. 43 , e48 (2015). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 62. Förster, D. et al. Genetic targeting and anatomical registration of neuronal populations in the zebrafish brain with a new set of BAC transgenic tools. Sci. Rep. 7 , 5230 (2017). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 63. Förster, D., Dal Maschio, M., Laurell, E. & Baier, H. An optogenetic toolbox for unbiased discovery of functionally connected cells in neural circuits. Nat. Commun. 8 , 116 (2017). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 64. Hartung, M. & Kisters-Woike, B. Cre Mutants with Altered DNA Binding Properties. J. Biol. Chem. 273 , 22884–22891 (1998). [ DOI ] [ PubMed ] [ Google Scholar ] 65. Kimura, Y., Hisano, Y., Kawahara, A. & Higashijima, S. Efficient generation of knock-in transgenic zebrafish carrying reporter/driver genes by CRISPR/Cas9-mediated genome engineering. Sci. Rep. 4 , 6545 (2014). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 66. Stemmer, M., Thumberger, T., Del Sol Keyer, M., Wittbrodt, J. & Mateo, J. L. CCTop: An Intuitive, Flexible and Reliable CRISPR/Cas9 Target Prediction Tool. PLoS One 10 , e0124633 (2015). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 67. Guggiana Nilo, D. A., Riegler, C., Hübener, M. & Engert, F. Distributed chromatic processing at the interface between retina and brain in the larval zebrafish. Curr. Biol. 31 , 1945–1953.e5 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 68. Lawson, N. D. et al. An improved zebrafish transcriptome annotation for sensitive and comprehensive detection of cell type-specific genes. eLife 9 , e55792 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 69. Satija, R., Farrell, J. A., Gennert, D., Schier, A. F. & Regev, A. Spatial reconstruction of single-cell gene expression data. Nat. Biotechnol. 33 , 495–502 (2015). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 70. Miri, A., Daie, K., Burdine, R. D., Aksay, E. & Tank, D. W. Regression-Based Identification of Behavior-Encoding Neurons During Large-Scale Optical Imaging of Neural Activity at Cellular Resolution. J. Neurophysiol. 105 , 964–980 (2011). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 71. Kunst, M. et al. A Cellular-Resolution Atlas of the Larval Zebrafish Brain. Neuron 103 , 21–38.e5 (2019). [ DOI ] [ PubMed ] [ Google Scholar ] 72. Schoonheim, P. J., Arrenberg, A. B., Del Bene, F. & Baier, H. Optogenetic Localization and Genetic Perturbation of Saccade-Generating Neurons in Zebrafish. J. Neurosci. 30 , 7111–7120 (2010). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 73. Orger, M. B. et al. Behavioral screening assays in zebrafish. Methods Cell Biol. 77 , 53–68 (2004). [ DOI ] [ PubMed ] [ Google Scholar ] 74. Ershov, D. et al. TrackMate 7: integrating state-of-the-art segmentation algorithms into tracking pipelines. Nat. Methods 19 , 829–832 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 75. Zhao, P. et al. The visuomotor transformations underlying target-directed behavior. Proc. Natl. Acad. Sci. USA 122 , e2416215122 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 76. Mirat, O., Sternberg, J. R., Severi, K. E. & Wyart, C. ZebraZoom: an automated program for high-throughput behavioral analysis and categorization. Front. Neural Circuits 7 , 107 (2013). [ DOI ] [ PMC free article ] [ PubMed ] 77. Dowell, C. K., Hawkins, T. & Bianco, I. H. Subsets of extraocular motoneurons produce kinematically distinct saccades during hunting and exploration. Curr. Biol. 35 , 554–573.e6 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Information (12.9MB, pdf) 41467_2026_71371_MOESM2_ESM.pdf (247.5KB, pdf) Description of Additional Supplementary Files Supplementary Movie 1 (72.5MB, avi) Supplementary Movie 2 (57.5MB, avi) Supplementary Movie 3 (180MB, avi) Supplementary Movie 4 (173.3MB, avi) Supplementary Movie 5 (188.3MB, avi) Reporting Summary (95.4KB, pdf) Transparent Peer Review file (280KB, pdf) Data Availability Statement Single-cell RNA sequencing raw and processed data files are available through the NCBI Gene Expression Omnibus (GEO) under the accession number GSE320054 . Data supporting the findings of this study, including source data, are publicly available on the RIKEN CBS data sharing platform (10.60178/cbs.20260306-001). Custom codes used in this study are publicly available on the RIKEN CBS data sharing platform (10.60178/cbs.20260306-001). Articles from Nature Communications are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (5.7 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top