Anatomy and function of retinorecipient arborization fields in zebrafish

In 1994, Burrill and Easter described the retinal projections in embryonic and larval zebrafish, introducing the term “arborization fields” (AFs) for the retinorecipient areas. AFs were numbered from 1 to 10 according to their positions along the optic tract. With the exception of AF10 (neuropil of the optic tectum), annotations of AFs remained tentative. Here we offer an update on the likely identities and functions of zebrafish AFs after successfully matching classical neuroanatomy to the digital Max Planck Zebrafish Brain Atlas. In our system, individual AFs are neuropil areas associated with the following nuclei: AF1 with the suprachiasmatic nucleus; AF2 with the posterior parvocellular preoptic nucleus; AF3 and AF4 with the ventrolateral thalamic nucleus; AF4 with the anterior and intermediate thalamic nuclei; AF5 with the dorsal accessory optic nucleus; AF7 with the parvocellular superficial pretectal nucleus; AF8 with the central pretectal nucleus; and AF9d and AF9v with the dorsal and ventral periventricular pretectal nuclei. AF6 is probably part of the accessory optic system. Imaging, ablation, and activation experiments showed contributions of AF5 and potentially AF6 to optokinetic and optomotor reflexes, AF4 to phototaxis, and AF7 to prey detection. AF6, AF8 and AF9v respond to dimming, and AF4 and AF9d to brightening. While few annotations remain tentative, it is apparent that the larval zebrafish visual system is anatomically and functionally continuous with its adult successor and fits the general cyprinid pattern. This study illustrates the synergy created by merging classical neuroanatomy with a cellular‐resolution digital brain atlas resource and functional imaging in larval zebrafish.


| BACKGROUND AND BEGINNINGS
In the first half of the 1990s, there was growing interest in the zebrafish Danio rerio (Cyprinidae; formerly Brachydanio rerio) as a model animal in the study of the developing teleostean brain. But information on the anatomy of the larval visual system was limited. Burrill and Easter (1994) used the lipophilic carbocyanine dye DiI (1,1 0 -dioctadecyl-3,3,3 0 ,3 0 -tetramethylindo-carbocyanine perchlorate) in fixed whole brains-a new tract tracing method at the time (Honig & Hume, 1989)-to report retinal projections in the early zebrafish. This led to the identification of 10 retinal terminal arborization fields (AFs) by observation of fine and beaded arborizations of retinal ganglion cell (RGC) axons in the zebrafish brain ( Figure 1). Retinorecipient AFs are defined as neuropil areas that contain spatial clusters of RGC terminals. The AFs gradually emerge between 60 and 72 hours postfertilization (hpf) (Figure 1a-c). Interestingly, during this early period, no pruning of more extensive (transient) retinal projections has been observed. Once established at 3 days postfertilization (dpf), the pattern is stable and does not change qualitatively into late larval and juvenile stages despite substantial tissue growth and addition of newborn RGCs (compare Figure 1c,d). AFs develop into neuropil areas associated with retinorecipient nuclei (RNs). Below we will discuss each of the RNs as described in the classical literature.
Twenty years after Burrill and Easter's landmark study, Robles et al. (2014) generated 3D reconstructions of the retinofugal projections from confocal imaging of transgenic fluorescent reporter lines, largely confirming the original description. Axon tracings were combined with labeling of synaptic terminals with a Synaptophysin-GFP fusion protein driven by an RGC-specific promoter/enhancer ( Figure 2). These authors also traced close to 500 individual, stochastically labeled RGC axons to reveal a cellular-resolution "projectome" of the retina. More recently, Kölsch et al. (2021) catalogued the F I G U R E 1 Burrill and Easter's (1994) original descriptions of retinofugal projections in zebrafish at embryonic and larval stages, based on wholemount specimens. The lipophilic tracer DiI was injected into the eye of a fixed embryo or larva and subsequently photoconverted to yield a brownish precipitate suitable for light microscopy. The developmental series shows that AFs form on the third day of development and then remain stable for days. The AFs served as anatomical landmarks already in the first digital atlas of the larval zebrafish brain (z-brain; Randlett et al., 2015). Since then, a new digital atlas resource, mapzebrain (see Box 1), with added functionalities, has been developed, which combines transgene expression patterns and single-neuron tracings. This atlas shows the coordinates of AFs in a standardized, 3D reference space ( Figure 3). In this review, we used the covisualization features of the atlas (see Box 1) to investigate the regional identity, cellular composition, and connectivity of each AF. Figure 4 illustrates how information from this digital resource can be tied to classical neuroanatomy. Three consecutive transverse sections from the mapzebrain atlas are shown in which the position of the AFs is indicated against a background of markers commonly used for neuroanatomical orientation: Synaptophysin as a neuropil stain and HuC (Elavl3) as a cell body stain for differentiated neurons (Figure 4a-c). Furthermore, the right panels show colocalization of AFs with three informative markers, which serve to compartmentalize the larval diencephalon: neurogenin (ngn), a marker of glutamatergic progenitors, tyrosine hydroxylase (th), a marker for catecholaminergic (here: dopaminergic) neurons, and glutamic acid decarboxylase isoform 1b ( gad1b/gad67), a marker of GABAergic neurons (Figure 4a'-c'). These three sections encompass all 10 AFs and will be discussed further in Section 3.
The RNs or regions in the adult diencephalon and mesencephalon are well known in teleosts, including in cyprinids closely related to zebrafish, such as goldfish and carp (reviewed in Vanegas & Ito, 1983;Medina et al., 1993;Rupp et al., 1996;Wullimann, 1998). The positions and identities of RNs in adult zebrafish are depicted in Figure 5 (panels modified from Wullimann et al., 1996). The nomenclature follows Northcutt and Wullimann (1988), as discussed below. We have anglicized most of the anatomical descriptors; Latin adjectives or genitive forms (preopticus, dorsalis, accessorius, thalami, etc.) become English adjectives (preoptic, dorsal, accessory, thalamic, etc.). The RNs of adult zebrafish are listed in Table 1, together with their synonyms. Burrill and Easter (1994) already attempted to draw correspondences between AFs and adult structures. However, these correlations turned out to be difficult to establish because, at early developmental periods, retinal terminal fields are within the neuropil located peripherally to the (largely not yet migrated) periventricular cell masses (see green neuropil areas in Figure 4). Thus, early AFs are not as closely associated anatomically with distinct visual diencephalic nuclei (i.e., neuronal cell body aggregates receiving retinal input; see red lettered structures in Figure 5) as is the situation in the adult brain.
Thus, while it is almost trivial even in the larval zebrafish brain to identify the optic tectum (TeO) or the suprachiasmatic nucleus (SC) as recipients of terminal arborization fields AF10 or AF1, respectively, great uncertainties existed in establishing definitive one-to-one relationships between the remaining AFs 2 through 9 with any particular known adult RN. Since 1994, great advances have been made, both with respect to function, location, and connectivity of retinorecipient brain structures in larval zebrafish. We summarize the current status in this review.
In the next section (Section 2), we will first describe adult RNs in teleosts-in particular in cyprinids (including the zebrafish)-known from tract-tracing studies and also briefly summarize some relevant classical electrophysiological evidence. In Section 3, we will then proceed to the recent implementation of larval zebrafish AFs into the mapzebrain atlas using molecular markers contained in this atlas (Kunst et al., 2019) and give a first overall interpretation of the relationships of AFs to RNs. Section 4 relates modern neurobiological data on zebrafish visual subsystems to data from classical studies.

| RETINORECIPIENT NUCLEI IN THE ADULT TELEOST BRAIN
Although there is some variability in retinal targets among teleosts, the picture within cypriniforms is quite uniform with a series of adult RNs and targets common to all species studied (reviewed in . We summarize briefly data on retinal projections in cyprinids gained with modern tract-tracing methods. Starting in the late 1970s, axonal tract-tracing methods included intraocular injections or optic nerve stump applications of [ 3 H] L-proline (with subsequent radioautography), Phaseolus vulgaris-leucoagglutinin (PHA-L), horseradish-peroxidase (HRP), biocytin, dextran amines and other substances in live brains as well as using lipophilic dyes (such as DiI, see above) in aldehyde-fixed brains (review on tract-tracing methods: Lanciego & Wouterlood, 2020).
In the 1970s, the standard neuroanatomical terminology used for cyprinids was that of  established in the goldfish (and beyond cyprinid species, for example, in the cyprinodont killifish; . The goldfish was then the main neurobiological laboratory species for teleost brain research. However, Northcutt and Wullimann (1988) suggested a terminological revision of RNs based on Braford and Northcutt's (1983) new comparative concept of the cyprinid diencephalon. This diencephalic neuroanatomical terminology,

BOX 1 mapzebrain-Max Planck Zebrafish Brain Atlas
This atlas resource combines diverse, spatially resolvable data from the 6 dpf larval zebrafish brain and offers a range of online visualization and analysis tools. In its current version, the web portal at https://fishatlas.neuro.mpg.de exhibits three data modalities. The Regions subportal (a) shows 3D views of 112 anatomical regions, which segment the larval brain in a mutually exclusive, (nearly) collectively exhaustive fashion. Some of the masks are tentative in their current form, and the segmentation is continually improving as new data are considered. The Neurons subportal (b) allows the user to inspect over 4000 stochastically labeled neuronal morphologies, all collected from different, age-matched specimens and coregistered in a standard brain. Laterality of the cells (position in left or right brain hemisphere) is preserved in the atlas. The Markers subportal (c) lets the user interrogate over 400 transgenic reporter lines and a rapidly growing number (>100) of gene expression and antibody staining patterns. Regions, cells, and markers can be covisualized in any combination. Renditions can be rotated, clipped, and magnified to give the most informative views and colored in any desired way. The mapzebrain resource further enables the user to visualize multiple, selected structures across all data modalities, to create publicationready images or animations, and to download the data for further analysis.
Since the literature on teleostean retinal projections covers both the time period before and after this major terminological change, we shortly discuss the reasons for the change and introduce some main synonyms. Braford and Northcutt (1983) addressed critical misconceptions in the comparative understanding of the teleostean diencephalon by providing an improved neuroanatomy with a combined analysis of retinal projections in the goldfish. For example, two prominent nuclei in the dorsal thalamus of reptiles and birds, the nucleus geniculatus lateralis (NGL) and nucleus rotundus (NR), are relay nuclei of a retinothalamo-pallial (NGL) and retino-tecto-thalamo-pallial (NR) pathway, respectively. These terms for sauropsid brains were used by  for the teleostean (retinorecipient) parvocellular and (tectorecipient) magnocellular superficial pretectal nuclei (PSp/PSm; see Figure 5). However, both teleostean nuclei have no ascending projections to the pallium as do the respective bird/reptilian thalamic NGL and NR (reviews: Braford & Northcutt, 1983;Wullimann, 1998). Furthermore, the teleostean PSp and PSm are most likely developmentally peripherally migrated from the periventricular pretectal histogenetic unit (see also Mueller & Wullimann, 2016) and not from the dorsal thalamic one. Thus, Braford and Northcutt (1983) proposed a general revision of the pretectum, which recognized a superficial (PSp/PSm, see above), a central pretectum (CPN; Peter and Gill's NC; probably meaning nucleus pretectalis centralis, abbreviation not explained) and a periventricular pretectum (PPd/v; see below). Both CPN and PPd/v are retinorecipient, as is the PSp (Braford & Northcutt, 1983). Furthermore, in the goldfish, a retinorecipient dorsal accessory optic nucleus (DAO; ; basal optic nucleus of Braford & Northcutt, 1983; nucleus pretectalis of ) and a ventral accessory optic nucleus (VAO; accessory optic nucleus of Braford & Northcutt, 1983; nucleus ventralis lateralis of  were described. Of note, while the lateral geniculate nucleus is considered an anmniote novelty, the teleostean dorsal posterior thalamic nucleus may be homologous to the sauropsid nucleus rotundus (Striedter & Northcutt, 2020). All cyprinid RNs discussed are depicted for the adult zebrafish in Figure 5.
A related change applied to the teleostean thalamus (see Mueller, 2012), a region we refer to as "thalamus proper" in the mapzebrain atlas (https://fishatlas.neuro.mpg.de), to distinguish it from F I G U R E 3 Retinorecipient AFs in the mapzebrain atlas. Snapshots of a 3D rotatable view generated by visualization tools at https://fishatlas. neuro.mpg.de. The atlas contains a fully segmented 6 dpf zebrafish brain in a standard reference space. The data modalities available in the atlas are annotated anatomical regions, including the 10 AFs (shown here), as well as ca. 4000 single neurons and 450 markers and transgenic lines, which can be covisualized in any desired combination. The atlas can be utilized for mapping functional imaging data or for classical neuroanatomy (see the basal and alar-most parts (habenula, epiphysis) of the dorsal thalamus prosomere. In order to improve the comparability within anamniotes, three main nuclei were described in the goldfish bearing great similarity to their counterparts in the amphibian thalamus: an anterior RN (A), and two (nonretinorecipient) more posterior nuclei, the dorsal posterior (DP) and central posterior (CP) thalamic nuclei (Braford & Northcutt, 1983; see Figure 5). Similarly, the (retinorecipient) ventral thalamus was revised to contain the intermediate (I), ventrolateral (VL), and ventromedial (VM) thalamic nuclei. In  goldfish brain atlas, the three dorsal thalamic nuclei are contained in nucleus dorsolateralis and dorsomedialis thalami (NDL, NDM). Somewhat confusingly,  NDL/NDM also include the most anterior part of the ventral thalamus (while the posterior part of the ventral thalamus is their nucleus ventromedialis, NVM) and, furthermore, the periventricular pretectal nucleus of the revised terminology (PPd/PPv; Figure 5d). Additionally, an area pretectalis is recognized in Peter and Gill's atlas (1975) in the relatively cell-sparse zone between VL and CPN which is not assigned to a nucleus in the revised terminology (see Figure 5). In the goldfish hypothalamic preoptic region, the commonly termed (retinorecipient) nucleus suprachiasmaticus (nucleus anterioris periventricularis-NAPv of  and three additional preoptic subnuclei were recognized by Braford and Northcutt (1983), including the retinorecipient posterior parvocellular preoptic nucleus (PPp; Figure 5) (nucleus preopticus of . We will now describe zebrafish adult retinofugal nuclei, using this new terminology, and then shortly discuss further adult cyprinid retinofugal studies giving some additional important synonyms. Retinal projections in the adult zebrafish (Yáñez et al., 2009(Yáñez et al., , 2018 Figure 5d) is also reported to receive retinofugal terminals (Yáñez et al., 2018). The remainder of retinal fibers continues to various layers of the optic tectum (TeO; not further discussed here, see Robles et al., 2013, andRobles et al., 2014). Retinofugal projections to the superficial pretectum (i.e., PSp), central pretectal nucleus (CPN) and the closely associated dorsal accessory optic nucleus (DAO) (Figure 5b,c) were later reported in the adult zebrafish by Yáñez et al. (2018). The neuropil directly rostrally adjacent to DAO receives input from the contralateral PSp (asterisk in Figure 5a; after Yáñez et al., 2018). The ventral accessory optic nucleus (VAO; Figure 5c) is located somewhat more caudally and was not reported as a retinal target in the adult zebrafish, but a retinorecipient VAO is reported in various other cyprinid species (see below).
Early, excellently detailed retinal projection studies in cyprinids (common roach Rutilus rutilus and carp,  common rudd Scardinius erythrophthalmus, , Peyrichoux et al., 1977 reported in essence the same retinorecipient targets listed above for the zebrafish (the neuroanatomical terms for the rudd are given in parentheses below).  and  described as retinorecipient the PSp (then identified as nucleus geniculatus lateralis) and various additional pretectal nuclei corresponding to the PPd/PCN (centrum opticum commissurae posterioris), the CPN (centrum opticum pretectale dorsale), DAO (centrum opticum pretectale mediale + ventrale), VAO (centrum opticum accessorium), as well as the ventral and dorsal thalamus (centrum opticum dorsolaterale thalami) and the suprachiasmatic and posterior parvocellular preoptic nucleus (centrum opticum hypothalamicum). Interestingly, a VAO and an additional RN corticalis were only described in some predominantly visually guided cypriniforms in which the PSp was furthermore rather large (i.e., folded). These features are present in basal (e.g., salmon or pike) as well as derived (e.g., perches) teleosts (Butler et al., 1991) which supports the hypothesis that parts of the visual system are reduced in certain cypriniforms lacking them (e.g., zebrafish).
A few classical neurophysiological findings coming from single cell recordings related to retinorecipient teleost nuclei may be noted here.
Dorsal and ventral thalamic neurons have large receptive fields, respond to stationary cues, are not directional, and do not habituate.
Only ventral thalamic neurons may show bimodal (i.e., somatosensory-visual) responses (Friedlander, 1983). CPN neurons have small receptive fields, show no habituation, are directional in the horizontal plane, and respond best to slowly moving objects (Friedlander, 1983). Furthermore, pretectal neurons, clearly located in the goldfish CPN, are visually directional sensitive in both horizontal and vertical axes (Debowy & Baker, 2011;Masseck et al., 2010; F I G U R E 4 Three consecutive transverse levels from anterior to posterior, taken from the mapzebrain larval zebrafish brain atlas (https:// fishatlas.neuro.mpg.de). Panels (a)-(c) depict AFs at their greatest extent with a HuC/D and Synapsin background, visualizing neuronal soma density and neuronal fibers, respectively. The registration process of HuC cells might underestimate the most peripheral cells and account for some peripherally located single cells shown in Figures 6-8 to be within the white matter. However, some single cells at the periphery of the gray matter might alternatively really be migrated into the white matter, as is typically the case in the preoptic region (see Figure 6a). Panels (a')-(c') show the same sections with molecular markers critical for compartmentalization of the diencephalon. See text for details. Cer, cerebellum; Di, diencephalon; DT, dorsal thalamus (thalamus); Hy, hypothalamus; lfb, lateral forebrain bundle; M1, early pretectal migration area; MO, medulla oblongata; ot, optic tract; poc, postoptic commissure; pc, posterior commissure; Pr, pretectum; PTv, posterior tuberculum, ventral part; Tel, telencephalon; TeO, optic tectum; TLo, torus longitudinalis; VT, ventral thalamus (prethalamus); ZLI, zona limitans intrathalamica Masseck & Hoffmann, 2009a, 2009b and implicated in the optokinetic reflex (see also Fite, 1985). Interestingly, a band of cells interconnects the CPN with the DAO (see Figure 5b,c) and retinal projections in the common roach cover both CPN and DAO, including this interconnecting band of cells, indicating that CPN and DAO may be functionally connected as an accessory optic system (AOS; see new data in zebrafish discussed in Section 3). This is supported by the fact that CPN/DAO and the interconnecting band of cells furthermore project to the corpus cerebelli in goldfish  and zebrafish (Yáñez et al., 2018; interpreted as intercalated pretectal nucleus Pi in their Figure 8d,e-a prerequisite for the optokinetic reflex-, while thalamic and hypothalamic nuclei do not have cerebellar projections. Interestingly, also VAO also projects to the cerebellum in goldfish and other teleosts (see  indicating that it is a further functionally related nucleus. Based on connectivity and location, the DAO has been compared to the nucleus of the basal optic root in nonmammalian tetrapods and the mammalian medial terminal nucleus, whereas the CPN resembles the avian nucleus lentiformis and the mammalian nucleus of the optic tract (discussed in . However, the differential roles of adult CPN, DAO and VAO neurons in direction sensitivity and the optokinetic reflex may not be conserved and need further investigation.

| RETINORECIPIENT AFs IN THE MAPZEBRAIN ATLAS
The 10 larval AFs of Burrill and Easter (1994) were integrated into the mapzebrain atlas ( Upon crossing brain sides, retinal fibers contribute massively to the large diencephalic neuropil located peripherally to the diencephalic cell masses ( Figure 4; note that the optic chiasma is not shown, but lies between y-levels 297 and 321). This neuropil shows all fiber and neuropil masses (visualized in green with an anti-Synapsin marker). Thus, apart from retinal fibers, major additional fiber tracts are present in this neuropil, such as the medial and lateral forebrain F I G U R E 5 Consecutive transverse Cresyl-stained (cell somata) and Bodian Silver-impregnated (axonal fibers) sections, from anterior to posterior, of an adult zebrafish brain with all known retinofugal target nuclei (RNs) indicated in red. Modified after Wullimann, Rupp and Reichert: Neuroanatomy of the zebrafish brain. Birkhäuser, Basel . Asterisk denotes a neuropil receiving contralateral input from PSp (Yáñez et al., 2018 Mueller & Wullimann, 2005& Wullimann, , 2016. M1 is the area of cell bodies wedged in between AF9 and AF7 and extending somewhat toward AF5 in Figure 4. Note that AF8 lies directly on neurons that form part of M1. However, the formation of distinct nuclei in the zebrafish larval brain is far from completed, which hampers the recognition of future adult RNs as described in Section 2. We shall discuss now the extratectal retinal terminal fields (AFs Level B (y = 297) shows AFs 1 through 5 and 7 through 9. AF4 is at this level associated with the ventral thalamus only. This fits well with the fact that only the adult anterior (A), but not the posterior dorsal thalamic adult nuclei DP/CP; see Section 2) are retinorecipient. In contrast, the various pretectal AFs are most extensive at this level.
While AF9 lies still immediately lateral to the periventricular pretectum, AF7 is most peripherally within the migrated pretectal M1 region and AF8 is directly on M1 cell bodies in-between AF9 and AF7. AF5 has its greatest extent at this level, still located basally to the M1 region and lateral to AF4. AF3 is basally adjacent. Also present at this level are AF1 which is directly adjacent to the suprachiasmatic part of the preoptic area and AF2 somewhat more dorsolateral in the preoptic region.
Level C (y = 321) finally shows AF6, but also still AF9. Thus, the latter has a very long anteroposterior extent, while the former only occurs at this caudal level essentially after AF5 has ended. Whereas AF9 is still directly lateral to the periventricular pretectum, AF6 lies again basal to the migrated M1 pretectal cell masses.
Generally, in vertebrates, incoming retinal fibers lie lateral to their retinorecipient diencephalic cell masses which extend dendrites into these terminal fields . In larval zebrafish, AF5, AF6, AF7 and, to a lesser degree, AF8 are located at the extreme lateral diencephalic periphery, at the edge of, or inside, M1, a pretectal migrated area. Thus, a first comparison of AFs and their association with the above-described anatomical diencephalic divisions using molecular markers and the adult neuroanatomy of RNs leads to the following interpretations: • AF1 is the terminal field associated with the suprachiasmatic nucleus (adult SC).
• AF2 is the terminal field of the posterior parvocellular preoptic nucleus (adult PPp).
• AF3 appears like a basal extension of AF4 and is possibly associated with adult VL.
• AF4 is the terminal field likely associated both with dorsal thalamus (adult A) and ventral thalamus (adult VL).
• AF5/6 are in a position to qualify as likely terminal fields of adult DAO and VAO (see below).
• AF7 qualifies as terminal field of the adult PSp.
• AF8 is the terminal field directly on neurons qualifying as adult CPN.
• AF9 is the terminal field of the periventricular pretectum and likely also the paracommissural nucleus (adult PPd/PPv, PCN).
• AF10 is without doubt the neuropil of the optic tectum (TeO).

| FUNCTION OF THE RETINAL AFs
In the past decade, tremendous progress has been achieved in the functional neuroanatomy of the zebrafish larval visual system. Refined behavioral analysis, in combination with functional calcium imaging, lesion studies and optogenetic manipulations, have revealed the contribution of individual AFs to the processing of visual stimulus features and to behavioral responses (Table 2). In the following, we summarize the state of our knowledge, AFbyAF:

| AF1 and AF2
While the functions of SC (nucleus associated with AF1) and PPp (nucleus associated with AF2) have not been directly tested in zebrafish, they likely contribute to the photo-entrainment of circadian behavior, as they do in other vertebrates, including mammals (Güler et al., 2007;Rollag et al., 2003). One or both of these nuclei may also While a homologous system has not yet been discovered in teleosts, VBA is disrupted in zebrafish that lack RGCs, for example, in lakritz mutants, in which complete loss of atoh7 function prevents RGC neurogenesis (Kay et al., 2001), and in blumenkohl (vglut2a) mutants, in which synaptic transmission from RGCs is reduced (Smear et al., 2007)

| AF3
Very little is known about the function of the associated nucleus, which is hypothesized here to be part of the ventral thalamus (see Section 5). A number of functional imaging experiments Kubo et al., 2014;Naumann et al., 2016;Wang et al., 2019;Wu et al., 2020) agree that AF3-projecting RGCs are not motion-sensitive or directionselective.
Response assumed, but anatomical annotation uncertain.
No response detected.   Zhang et al. (2017) reported that AF4 is contacted by ventral thalamic and/or thalamic eminence neurons, based on their observation that neurons extending dendrites into AF4 project to the left habenula (Turner et al., 2016). Thalamohabenular projections have been observed in goldfish and trout (Villani et al., 1996;Yañez & Anad on, 1996). Cells with dendrites in AF4 receive largely ON ganglion cell input, as shown by calcium imaging, project asymmetrically to the left habenula, and are required for light-seeking behavior (Zhang et al., 2017). Arriving at a similar conclusion, Cheng et al. (2017) and Lin and Jesuthasan (2017) demonstrated that blue, but not red, light activates a thalamic area, which projects to left habenula and includes AF4. Another candidate for these thalamic cells is the rostrolateral thalamic nucleus which is present in the adult zebrafish brain  and also projects to the habenula (Turner et al., 2016). However, this nucleus has neither been reported to receive retinal input in the adul zebrafish brain (see Section 2), nor do we see cells qualifying for it in our single cell larval brain analyses (see Section 5). In

| AF5
Calcium imaging showed that RGC terminals in AF5 respond strongly to visual motion in all directions (Kramer et

| AF6
AF6-projecting RGCs respond robustly to dark, looming stimuli, as well as to the gradual dimming of the entire visual field (Temizer et al., 2015), suggesting they are OFF-responsive. This is consistent with the stimulus preferences of tectal layers to which AF6-targeting RGC axons project, namely SFGS6 and SGC (Robles et al., 2014;Temizer et al., 2015). Other authors have recently argued that AF6 receives direction-selective inputs from the retina and is involved in the OMR (Naumann et al., 2016). This is inconsistent with the projection patterns of individual RGCs (Robles et al., 2014). RGCs that project to the direction-selective superficial layer SFGS1 of the tectum have not been observed to form collaterals in AF6, but rather in the adjacent AF5. Moreover, the zebrafish radar s327 (gdf6a) mutant (Gosse and Baier, 2009), in which dorsal RGCs do not develop and which lacks AF6, but retains AF5, is well able to perform responses to directional motion such as OKR and OMR (Muto et al., 2005). Highresolution calcium imaging showed that direction-selective inputs are APN cells may also contribute dendrites to AF7 neuropil. Laser ablation of AF7 disrupted neither optomotor responses (Roeser & Baier, 2003;Semmelhack et al., 2014) nor phototaxis (Burgess et al., 2010).
In adult teleosts, no specific RN has been implicated in food perception or prey capture. Interestingly, a specific subpopulation of RGCs project to both AF7 and SO, the most superficial retinorecipient layer of the tectum (Robles et al., 2014). The same collateralization pattern was reported for RGCs in adult goldfish that project to the PSp (Springer & Mednick, 1985).
In the retina, AF7-plus-SO-projecting RGCs exhibit either a narrow-diffuse or a bistratified dendrite morphology (Robles et al., 2014). Kölsch et al. (2021) showed that a subset of the RGC subclass that express the transcription factor mafaa exhibit these two morphologies. Some of them project to AF7 plus SO; another mafaa-positive subpopulation project to SFGS2. The tuning properties and function(s) of the five, or so, mafaaexpressing RGC types has so far not been investigated. We predict that a subset of them respond to prey features and are required for visual detection of prey.

Robles et al. (2014) demonstrated that AF7 receives inputs from
RGCs located mainly in the temporal half of the retina. These projections are retinotopically organized. Thus, PSp/AF7 contains a map of the temporal visual field, which contains the area centralis, the zone that represents the prey image immediately before the capture strike . RGCs in the area centralis are enriched for tuning to the ultraviolet (UV) range of the spectrum (Yoshimatsu et al., 2020).
UV contrast likely aids in the detection of prey items (Flamarique & Hárosi, 2000). Interestingly, at least one mafaa-positive RGC subpopulation is enriched in the area centralis (Kölsch et al., 2021).
The neighboring PSm (see Figure 4b), which is not retinorecipient, projects to the hypothalamus. It is strongly innervated by an AF7-associated cell group and also receives projections from the optic tectum (in larval zebrafish: Muto et al., 2017; in adult carp: Yoshimoto & Ito, 1993

| AF9
Functional imaging studies have revealed that AF9, the neuropil of the periventricular pretectum, is further subdivided into an ON-responsive dorsal part, AF9d, and an OFF-responsive ventral part, AF9v (Robles et al., 2014). In some cypriniforms, including adult goldfish, a dorsal and a ventral nucleus can be anatomically distinguished. Both are innervated by RGC axons. We therefore propose that AF9d corresponds to PPd and AF9v to PPv. Yáñez et al. (2018) describe another retinorecipient pretectal nucleus, the paracommissural pretectal nucleus (PCN), in the periventricular pretectum (see Figure 5d). This nucleus was not considered by Burrill and Easter (1994). PCN either develops later or is at this stage fused with PPd and PPv. Kölsch et al. (2021) reported that eomesa-expressing RGCs, which constitute at least five different cell types, project to AF9. A similar diversity of morphologies and collateralization patterns was described for AF9-projecting RGCs by Robles et al. (2014). This is consistent with AF9 being a neuropil conglomerate of different pretectal nuclei.
Laser ablation of cells around AF9 was reported to reduce optomotor responses to grating motion (Semmelhack et al., 2014), and optogenetic activation of a similar region resulted in optokineticlike eye movements (Kubo et al., 2014) and optomotor swimming in the absence of any visual stimulation (F. Kubo & H. Baier, unpublished observation). However, new results Wu et al., 2020) suggest that these earlier perturbations primarily affected cell populations that are immediately ventral to AF9 and may be associated with DAO/AF5.

| AF10
AF10 is the retinorecipient neuropil of the optic tectum (TeO). The tectum's broad function is well known; it is involved in sensorimotor tasks that require the localization and identification of visual objects, such as the approach of prey or the avoidance of obstacles. In zebrafish larvae, this was shown by functional imaging from RGC terminals and tectal neurons (Bianco & Engert, 2015;Förster et al., 2020;Temizer et al., 2015) and in ablation studies by Roeser and Baier (2003), Gahtan et al. (2005) and Temizer et al. (2015). The tectum converts a retinotopic map of visual inputs into a map of directed motor outputs (shown in larval zebrafish by Helmbrecht et al., 2018).
Unilateral lesions of one tectum result in a curious reversal of light-seeking behavior, in that the fish with one tectum swim toward the dark (Burgess et al., 2010). This seemingly paradoxical result was elegantly explained by a contribution of OFF RGC inputs, which normally steer the fish in a direction away from the darker hemifield of the visual scenery (Burgess et al., 2010).
The tectum receives synaptic inputs from 97% of the RGCs (Robles et al., 2014), half of which are shared via collaterals with other AFs. Still, an intact AF10 is dispensable for responses to optic flow stimuli, such as optomotor or optokinetic reflexes, although it is involved in pacing of saccades during the OKR (Roeser & Baier, 2003). These behaviors, which compensate for self-motion, are probably driven by connections of direction-selective RGCs in the DAO of the pretectum without major contribution of the tectum.

| TRACING OF NEURONS THAT TARGET RETINAL AFs
This section explores the mapzebrain atlas for single cells (e.g., from Kunst et al., 2019, plus  Retrieved single cells that are more remote from an AF (e.g., tectal or rhombencephalic cells, but also contralateral diencephalic cells) more likely also extend axons rather than dendrites into an AF. Results are summarized in Table 3.

T A B L E 3 Single cells extracted from mapzebrain atlas with a neurite into one particular AF
Notes: Top line shows AFs and total single cell number. Left column shows brain regions. Abbreviations after regions are used in figures to identify single cells, not the region itself. Highest single cell number for each AF in bold. Brackets show two areas for which not all cells were singly assigned, cell number is in predominant area of the two.

| AF1
The basic search for cell somata with a neurite into AF1 yields no results at the time this review was composed. Nonetheless, there is no doubt that this AF receives RGC input that is picked up by dendrites of suprachiasmatic (preoptic) neurons (see Sections 2 and 3).
Likely, the absence of retrieved cells is a false negative result and can be explained by a regional bias in how single cells were entered into the atlas. This also explains that RGCs are only retrieved by basic searches in a fraction of other AFs (see below). Indeed, in their survey of almost 500 RGC projection patterns, Robles et al. (2014) described several RGCs with terminals in AF1, suggesting that RGCs are relatively undersampled in the mapzebrain atlas.

| AF2
The basic search for cell somata with a neurite into AF2 yields six neurons (Figure 6a), all of which are located in the preoptic region immediately anterior to AF2 (compare with Figure 4b). The transverse level (Figure 6a2) shows the exact histological position of one neuron on the left side (cyan) and one neuron on the right side (yellow). The remaining four bilaterally located neurons are slightly anterior to this transverse level as can be seen on the sagittal view (Figure 6a1).
This sagittal level (Figure 6a1) is chosen as to show the main brain parts and anterior (ac) and postoptic (poc) commissures with the preoptic region in between and all cell somata were singly checked to confirm their exact histological position within the preoptic region using atlas coordinates in all three planes. Two neurons cross with a neurite (presumably an axon) to the other brain side. No retinal neurons are retrieved (see AF1 for explanation).

| AF3
The basic search for cell somata with a neurite into AF3 yields 41 neurons (Figure 6b), the majority of which (25)

| AF6
The basic search for somata with a neurite into AF6 results in 72 neurons ( Figure 8a1). Their distribution in the brain is qualitatively similar to those neurons found for AF5. The bulk of neurons is again in the alar diencephalon (39 mostly migrated pretectal M1 cells, see Figure 8a2, and 13 in thalamus, mostly in ventral thalamus, VT, one dorsal thalamic cell indicated with D in Figure 8a1) with a few

| AF7
The basic search for somata with a neurite into AF7 yields 30 single cells The RGCs use the optic chiasma to do so.

| AF8
The basic search for somata which terminate with a neurite in AF8

| AF9
The basic search for cell somata with a neurite into AF9 yields 217 single cells total (Figure 8d1), with a majority of 129 pretectal cells (many in periventricular and migrated pretectal area M1, see Figure 8d2

| General observations on single-cell projection patterns
These single cell (basic) searches in the mapzebrain atlas were highly consistent with the analysis of larval and adult retinal targets and nuclei provided in Sections 2 through 4. The basic searches reveal in great detail where neurons with a neurite into particular AFs lie. As mentioned above, these neurites may theoretically be axons or dendrites, but we assume that neurons in larval diencephalic areas that are well known in adult teleosts, including the zebrafish, to be retinorecipient brain nuclei (described in Sections 2 and 3) have dendrites into an AF while others (such as RGCs, tectal, or rhombencephalic cells) project an axon into one or more AFs.
In all cases (i.e., AF2 through AF9; for AF1, see below), all (AF2) or the majority of single cells retrieved in each search are in that neuroanatomical diencephalic division as would be expected by the analysis presented in Sections 2 and 3. That is the preoptic region for AF2, the dorsal and ventral thalamus for AF4, the migrated pretectal area M1 for AF5, AF6, AF7, and AF8, and the periventricular pretectum for AF9. The cells retrieved for AF3 were predominantly in the ventral thalamus (see below for interpretation). These matches can be made because the majority of retrieved cells with dendrites into a particular AF always lie in one of these larval diencephalic areas. However, it seems that minor dendritic contacts to AFs from additional than the main diencephalic areas do occur. Whether these are real or false positive remains to be determined.
Long distance (i.e., axonal) inputs to AFs from retina, optic tectum and rhombencephalon are consistent with classical adult brain connectivity studies (see Section 3). One particularly nice example is the input of nucleus isthmi cells to AF7 (Figure 8b1). Similarly, outputs of retrieved cells to cerebellum and medulla oblongata fit the adult connectivity picture, for example, only retrieved cells associated with AF6, AF8 and AF9 (i.e., adult DAO, CPN, PPd/PCN) project to the cerebellum (Figure 8a2,c1,d1), but not cells associated with AF2, AF3, AF4, or AF7 (i.e., adult PPp, VT, DT, PSp).  Figure 5).

| Specific observations on single cells in periventricular and superficial pretectum
Both basic searches for single cells contacting AF7 and AF9 retrieved a majority of cells in the pretectum, those for AF7 in the migrated pretectal area M1, those for AF9 mainly in the periventricular pretectum. This is consistent with their identification as adult parvocellular superficial pretectal nucleus (PSp) and dorsal periventricular pretectal nucleus (PPd) plus likely the paracommissural nucleus (PCN), respectively (see Figure 5).

| Specific observations on single cells in AOS
As suggested in Sections 2 and 3, the dorsal accessory optic nucleus (DAO) and central pretectal nucleus (CPN) form one retinorecipient structure only seemingly interrupted on some transverse levels by the magnocellular superficial pretectal nucleus (PSm, see Figure 5). Our basic searches demonstrate that neurons with neurites into AF5/AF6 and AF8 lie predominantly in the larval migrated pretectum M1 as AF remains open at the moment. As discussed in Section 3, it has not been demonstrated in adult zebrafish and remains elusive in other (but not all) teleosts as well. It is furthermore highly interesting that no DS neurons are found associated with AF1 through AF4 (i.e., preoptic region and thalamus) and AF7 (superficial pretectum). This is expected from their known functional visual context different from direction selectivity (see Section 4).

| CONCLUSIONS
Much progress has been made in assigning functions to the retinofugal pathways in larval zebrafish. The retinal inputs to individual AFs have been traced, and their physiological responses to various features of visual stimuli have been recorded. As a result, most AFs can now be annotated with certainty in the framework of the wellestablished teleost neuroanatomy. Moreover, the knowledge of functional contributions of individual AF channels to visual processing is rapidly growing. Because AF4 and AF9 may each harbor two or three neuropil areas, respectively, it now emerges that 12-14 RNs are embedded in the original array of 10 AFs. This matches the number of RNs that was described in adult goldfish (Springer & Gaffney, 1981), a related cyprinid species, and in other teleosts (e.g., . It is still unclear which diencephalic area is associated with AF6, although it appears likely that it is part of the larval AOS. Moreover, a paracommissural nucleus has not been identified in the larva. If it exists at 6 dpf its neuropil may be embedded in AF9. Future work is sure to shed light on these remaining questions. Much progress is expected to come from mapping molecularly defined RGC populations to individual AFs, from identifying diencephalic cell types and their connectivity, and from targeted genetic manipulations of inputs and outputs in the visual network of the larval zebrafish.

ACKNOWLEDGMENTS
We thank the members of our group for discussions and, in particular, Dun- Open access funding enabled and organized by Projekt DEAL.

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1002/cne.25204.