Article. Classification of Object Size in Retinotectal Microcircuits

Transcription

1 Current Biology 24, , October 20, 2014 ª2014 Elsevier Ltd All rights reserved Classification of Object Size in Retinotectal Microcircuits Article Stephanie J. Preuss, 1,3 Chintan A. Trivedi, 1,3 Colette M. vom Berg-Maurer, 2 Soojin Ryu, 2 and Johann H. Bollmann 1, * 1 Neural Circuits and Behavior Research Group, Department of Biomedical Optics 2 Developmental Genetics of the Nervous System Max Planck Institute for Medical Research, Heidelberg, Germany Summary Background: A principal task of the visual system is to detect and classify moving objects in the visual environment. Information about the size of an object is critical for selecting appropriate behavioral responses. Object size is encoded in retinal ganglion cell (RGC) activity. Little is known, however, about how inputs from the multitude of RGC subtypes are distributed to higher visual centers and how information is combined from these feature-selective inputs. Results: Here we show that in the zebrafish optic tectum, prey- or predator-like moving targets evoke activity in distinct groups of RGC fibers dependent on target size, demonstrating a retinal origin of tectal size classification. Small-size-selective retinal inputs are relatively more frequent in the most superficial layer of the tectal neuropil, whereas large-size-selective inputs predominate in deeper layers. Monostratified superficial interneurons (SINs) process large-size- and small-sizeselective signals dependent on their dendritic target layer, consistent with the retinal input organization. Further downstream, small- and large-sized objects are encoded in population activity of separate sets of tectal neurons. Conclusions: Ethologically relevant size classes are preferentially processed in different layers of the tectal neuropil. The tectum categorizes visual targets on the basis of retinally computed size information, suggesting a critical role in visually guided response selection. Introduction Classifying the size of moving objects is a fundamental computational task for the visual system. In general, local object motion serves as a strong trigger for bottom-up attention [1]. Concurrent classification of object size obviously serves as a useful basis for rapid perceptual decision making: a large object is more likely to turn out to be a threat, e.g., a predator. Conversely, a small object could be prey and should be targeted with little delay. Thus, size discrimination in the visual system is a central classification process for rapid response selection. Classically, the two response categories are orienting, goal-directed movements ( attack ) and aversive, defense-like movements ( flight ). A visual center known to play a key role in response selection is the superior colliculus (SC), or its nonmammalian homolog, the optic tectum [2]. Visually controlled orienting and 3 Co-first author *Correspondence: johann.bollmann@mpimf-heidelberg.mpg.de pursuit movements are impaired by lesions of this structure, while local stimulation or pharmacological disinhibition of the optic tectum can induce both orienting and defensive behavior [2 5]. It is well known that the retina contains distinct ganglion cell types that have different sensitivities to object size [6 9]. Thus, classes of retinal ganglion cells (RGCs) with different size selectivity could form parallel channels toward the brain, but how these channels are distributed and combined in retinorecipient centers is not clear [10]. Specifically, does the size classification process in the optic tectum have access to size-selective signals generated in the retina? Alternative models suggest that size-selective responses may result from surround suppression generated by local or feedback connections, as described in visual cortex [11, 12]. Also, earlier work in the tectum has suggested that size selectivity is an emergent property of the intratectal circuitry [13, 14]. The zebrafish has emerged as an important model system for identifying neural correlates of visuomotor processing, for example by imaging neural activity at the population level [15, 16]. In addition, detailed information on the structure-function relationship of individual, genetically labeled neurons in a neural circuit can be obtained using Ca 2+ imaging and patchclamp recordings from single cells [17, 18]. Together, these provide a rich basis for building models useful also for mammalian visuomotor research [19, 20]. To identify the source of size selectivity in the tectum, we measured the distribution of size-selective Ca 2+ signals in RGC afferents in response to stimuli previously shown to evoke different swim patterns [21, 22]. Furthermore, we recorded from horizontal tectal neurons and found that they receive size-tuned presynaptic input in retinorecipient layers. Surprisingly, these superficial interneurons (SINs) exhibited distinct size-tuning properties, depending on whether their dendrites arborized beyond a critical depth within the tectal neuropil. Also, object size was encoded in different postsynaptic populations in the periventricular layer, which identifies the optic tectum as a critical stage for object classification and response selection. Results Size Discrimination in the Retinotectal Pathway Zebrafish larvae respond to small visual targets moving in the periphery with short orienting swims directed toward the stimulus (Figure 1A). Larger targets evoke avoidance swims in the opposite direction (Figure 1B). To examine size sensitivity in the retinotectal pathway, we projected light rectangular targets moving on a gray background with stationary objects (Figures 1A 1D). This stimulus configuration closely resembles the way paramecia, the natural prey of zebrafish, appear when illuminated from above and imaged from the side, i.e., from the perspective of the larva (Figure 1C). Furthermore, this configuration was shown previously to be ethologically relevant in that it can evoke prey capture sequences in a closed-loop virtual environment [22]. The zebrafish optic tectum is critical for target-directed swims in the presence of prey-like objects. Recent analysis has shown that morphologically distinct RGCs target single layers in the tectal neuropil [23]. Little is known, however, about how functionally distinct RGC classes with different

2 Size Classification in Zebrafish Optic Tectum 2377 Figure 1. Size Discrimination in the Retinotectal Pathway using Ethological Stimuli (A and B) Appetitive versus aversive swim. A zebrafish larva performs swims directed toward a small moving target (A) or away from a large target (B). The stimulus paradigm is shown in the bottom panels; a white rectangle moves on a gray background with stationary objects. (C) Configuration for imaging paramecia as seen from the perspective of hunting zebrafish. Inset: examples of paramecia. Note that paramecia appear as bright objects on a dark background. Scale bar, 0.5 mm. (D) Experimental set up for multiphoton imaging and patch-clamp recordings, in combination with behaviorally relevant visual stimulation. (E) The retina contains functionally distinct RGC subtypes (schematically highlighted by color). The tectum receives RGC input mainly in the superficial layers (stratum opticum and stratum fibrosum et griseum superficiale), but how functionally distinct RGC subtypes map onto these layers is unclear. size-tuning properties are organized in the tectal input stage (Figure 1E). To determine the functional organization of size selectivity in this important visual pathway, we used multiphoton Ca 2+ imaging and targeted patch-clamp recordings to determine size-tuning curves in response to behaviorally relevant stimuli. The Optic Tectum Receives Small- and Large-Size- Selective RGC Input We crossed the atoh7:gal4 transgenic line [24] to a UAS: GCaMP6m transgenic line to target expression of the genetically encoded Ca 2+ indicator (GECI) GCaMP6m [25] to RGCs (Figure 2A). In some experiments, we used a UAS:GCaMP3 line with similar results. Using multiphoton imaging at different optical planes, we observed a dispersed pattern of size-selective Ca 2+ signals in the retinorecipient neuropil (Figure 2B). Notably, many regions of interest (ROIs) throughout the neuropil exhibited fluorescence transients selectively evoked by small targets. Moreover, some regions showed a transient response to the OFF edge of a whole-field flash. To quantify the sensitivity of different presynaptic compartments to stimuli of different sizes, we measured the signal variance during target presentation, from which we obtained size-tuning curves (Figure 2C). We calculated a size index, which was positive for large-size preference and negative for small-size preference. Next, we generated color maps to highlight size-selective regions and compared their location with the laminar structure of the tectal neuropil (Figure 2D). There was no strict laminar organization for size-selective pixels, suggesting that differently tuned RGC types can innervate the same tectal layer. To quantify the distribution of size-tuned afferents, we divided the retinorecipient neuropil into a superficial layer, a central layer, and a deep layer at dorsal and ventral imaging planes in the tectum (Figure 2D, top and bottom). Frequency histograms of size indices grouped according to the assigned layers showed that small-size-selective regions were relatively frequent in superficial layers, whereas large-size-selective regions predominated in deeper layers. This trend was observed both in the dorsal and in the ventral imaging planes (Figures 2E and 2F, top and bottom; n = 11 fish). The percentage of small-size-selective pixels with a size index %20.2 was 16% in the superficial layer, but only 2.5% in the deep layer, in dorsal imaging planes. In ventral planes, the corresponding percentages were 9% versus 3%. Overall, the size index distributions were significantly different between the superficial, central, and deep neuropil layers (Kolmogorov-Smirnov test, p < ; Figure 2G). We also observed small-size- and large-size-selective Ca 2+ signals in larvae, in which GCaMP6m was expressed stochastically in small subsets of RGCs (Figure S1). Together, these observations suggest that size classification already begins in RGC fibers innervating the tectal neuropil, with a relatively strong representation of small-sizeselective inputs in the most superficial layer. SINs Arborize in Distinct Retinorecipient Layers of the Tectal Neuropil Do tectal cells selectively pick up size-tuned signals from retinal inputs? To address this question, we focused our analysis on SINs because their anatomical position suggests that they receive direct synaptic input from RGC terminals. Here we examined the size-tuning properties of SINs to determine whether their activity is sensitive to object size within the ethologically relevant range. Because information on the laminar distribution of SINs is lacking, we performed an analysis of their dendritic depth. In an unbiased sampling approach, randomly selected SINs were patched using the shadow patch technique and were filled with a red fluorescent dye in the pou4f3:gfp line, in which GFP expression reproducibly labels RGC fibers in the most superficial layer (stratum opticum, SO) and the deepest portion of a central layer (stratum fibrosum et griseum superficiale, SFGS) [26], which we use as an anatomical yardstick to determine relative dendritic depth [18]. A comparison of the RGC expression patterns in double transgenic pou4f3:gfp/atoh7:gap43- RFP fish revealed that the RGC axons projecting into the SO in the pou4f3:gfp line represent a subset of the most superficial RGC axons labeled in the atoh7 line (Figure S2).

3 Current Biology Vol 24 No Figure 2. RGC Inputs to Tectal Neuropil Carry Small- and Large-Size-Selective Signals (A) Two optical sections from tectal neuropil with GCaMP6m expressed in RGC fibers. Imaging depth is 18 mm and 51 mm from the dorsal crest. Dashed lines indicate superficial, central, and deep layers. Colored boxes indicate eight representative ROIs. Arrowheads mark skin fluorescence. M, medial direction; C, caudal direction. (B) Normalized fluorescence signals from ROIs in response to moving targets and full-field flash. Object size is indicated above the stimulus trace. Numbers refer to ROIs indicated in (A). Note the whole-field flash OFF response in traces 6 and 8. (C) Size-tuning curves calculated from signal variance in ROIs indicated in (A). Variance was measured in intervals indicated by the blue (small-stimulus size) and green (large-stimulus size) vertical boxes in (B). Size index (S.I.) was calculated as S.I. = (Var large 2 Var small ) / (Var large + Var small ). (D) Size selectivity maps for optical sections shown in (A) with corresponding color scale. (E) Normalized frequency distributions of size indices determined from single-pixel fluorescence signals, sorted for the superficial, center, and deep layers of the neuropil and displayed separately for the dorsal and ventral imaging plane. Data are from (D). (F) Average frequency distributions of size indices from 11 fish, for dorsal and ventral imaging planes. (G) Cumulative histograms of size indices. The same data as in (F) are shown. Scale bars, 50 mm (A and D). See also Figure S1. The majority of patched SINs had a monostratified dendritic tree branching in a single retinorecipient layer (Figures 3A 3F). They covered a large extent of the neuropil in both the rostrocaudal and the mediolateral directions (Figures 3A and 3D). The target layer, however, varied widely for different SINs (Figures 3B, 3C, 3E, and 3F). We used intensity profile analysis along the radial direction of the neuropil (Figures 3G and 3H) to quantify the distribution of SIN target layers. Notably, about 24% of SINs exhibited a very superficial tree in the SO (relative depth class > 90%), whose branches appeared to be interwoven with the thin SO layer labeled in the pou4f3:gfp line, whereas no SINs were found to arborize in the deepest SFGS layers (relative depth class < 20%; Figure 3I). A comparison between the relative depth of SIN somata with dendritic depth showed no significant correlation (Figure 3J). In conclusion, the broad distribution of SIN target layers suggests that these cells may serve distinct roles in visual input processing. Size Tuning of Excitatory Inputs to SINs Is Layer Dependent Next, using whole-cell recordings, we determined the synaptic input patterns of SINs and compared their input tuning properties with dendritic depth. In response to the moving targets of variable size, we observed different patterns of excitatory postsynaptic currents. A frequent example were SINs that

4 Size Classification in Zebrafish Optic Tectum 2379 Figure 3. SINs Stratify in Different Sublayers of Retinorecipient Neuropil (A) Maximum-intensity projection showing a SIN filled with sulforhodamine (red) and retinal afferent fibers (cyan). A dorsal view is shown. (B) Transverse optical section through same cell as in (A). Dendrites of labeled SIN colocalize with the superficial SO band (arrowhead). (C) Maximum-intensity projection along same orientation as in (B). Dendrites remain in the SO layer. Note that the apparent broadening of the SIN dendrites (lower left) is caused by the tectal curvature. Compare also Movie S1. (D F) Representation of a patched SIN in a different fish. The same orientation as in (A) (C) is used. Note that dendritic tree stratifies in a narrow region of the deeper SFGS D,F. (G) Line profile analysis from boxed region in (C). The blue curve indicates GFP-expressing layers in SO and SFGS D,F. The red curve indicates the extent of the SIN dendritic tree. The thin red line is a Gaussian fit. (H) Line profile analysis for the cell in (F). Note: oblique sections and maximum-intensity projections in (B), (C), (E), and (F) were gamma corrected for visualization of the thin SO band, but line profiles were measured in uncorrected maximum-intensity projections. (I) Histogram of dendrite depth of individual SINs measured in the pou4f3:gfp background (n = 41). (J) Scatter plot of soma depth versus dendrite depth for individual SINs exhibits no significant correlation (p = 0.84, Spearman rank correlation). Scale bars, 40 mm (C and F). See also Figure S2 and Movie S1. received tonic excitatory inputs that persisted during presentation of small moving targets but were suppressed during large target presentation (Figures 4A and 4B). Conspicuously, these cells often showed a strong phasic increase in excitatory input toward the end of a large-size stimulus interval. These SIN types typically arborized in different sublayers of the SFGS. By contrast, other SINs exhibited relatively low tonic input currents during stimulus pauses but received excitatory inputs selectively evoked by small targets (Figures 4C and 4D). These SINs predominantly stratified in the superficial SO layer. Because small-size-selective SINs were relatively rarely encountered during random sampling, we also patched cells in a new transgenic line (Oh:GCaMP6s; Supplemental Experimental Procedures) in which individual SINs could be screened for small-size selectivity using multiphoton Ca 2+ imaging prior to patch-clamp analysis. These SINs also exhibited small-size-selective inputs and displayed a dendritic tree in the most superficial neuropil (Figures 4D and 4K). We used the variance of the current recording as a generalized measure for whether a stimulus caused a change in the input activity in a given stimulus interval. This way, size-tuning curves were determined for each cell, and a size index was calculated from the averaged variances for large and small targets, respectively (Figures 4E 4H). Importantly, most large-size-selective SINs with a positive size index had dendritic trees that arborized in deeper layers (relative depth < 90%), while the majority of cells with negative size index featured dendritic trees in a thin superficial layer (relative depth > 90%; Figure 4I). When grouped according to dendrite depth and assigned to the SFGS and SO, respectively, the size indices were significantly different between the two groups (Figure 4J). To further test whether synaptic inputs to SINs may originate directly from RGC terminals in the tectum, we also measured their sensitivity to whole-field flashes (Figure S3). This corroborated the notion that tonic and phasic ON- and OFF-type RGCs may contribute to SIN input activity in a layer-dependent fashion. Notably, SINs with dendrites in the SO were virtually insensitive to whole-field flashes, consistent with their tuning to small objects. Together, these findings suggest that SINs receive distinct size-selective input and that a subpopulation arborizing in the superficial SO is preferentially activated by small moving targets resembling preylike objects. Functionally Distinct Classes of SINs in the Tectal Neuropil Next, we also determined the size-tuning curves of somatic Ca 2+ signals across the SIN population. In the novel Oh: GCaMP6s line, the GECI was expressed in a broader pattern that included most, if not all, SINs. In this line, SINs exhibited heterogeneous size-tuning properties, which could not be inferred from their somatic location (Figure 5A). We confirmed in separate recordings that SINs can generate action potentials (Figure S4), which are therefore likely to underlie somatic Ca 2+ transients. The time course of Ca 2+ transients evoked by moving targets revealed cells that

5 Current Biology Vol 24 No Figure 4. SINs Receive Size-Selective Excitatory Input (A) Synaptic current in a SIN with dendrite in upper SFGS. Note the strong background current, which is suppressed during the presentation of larger targets, and subsequent rebound from input suppression. (B) Recording from SIN with dendrites in deep SFGS with size-tuned synaptic input. (C) Recording from SIN with dendrites in the superficial SO. Note the increase in synaptic current during small target presentation. (D) Recording from a SIN in the Oh:GCaMP6s line with dendrites in superficial SO, preselected because of small-size-driven Ca 2+ signals. (E H) Current variance of cells shown in (A) (D) as a function of target size. Horizontal lines indicate the mean variance for small and large targets, respectively. (I) Size index of SINs versus dendritic depth. Gray symbols indicate SINs randomly sampled in the pou4f3:gfp line (n = 41). Red, crossed circles indicate SINs preselected in the Oh:GCaMP6s line (nominal dendrite depth 100%, n = 7). The vertical line indicates 90% dendrite depth. (J) Size index of SINs with dendrites arborizing within 20% 90% of retinorecipient neuropil (SFGS) and R90% (SO) were significantly different (p = , Wilcoxon rank-sum test). The red line indicates the median, the box indicates the 25% 75%-quartile range, and error bars indicate the full range of data. (K) Example of small-size-selective SIN in the Oh:GCaMP6s line. The same cell as in (D) is shown. The maximum-intensity projection is oriented as in Figure 3C. GCaMP6s expression shown in cyan. Scale bar, 40 mm. See also Figure S3. were selective for either large or small stimuli, respectively (Figure 5B). k-means clustering of the normalized fluorescence traces [27] from the population of imaged SINs (n = 133 cells from 14 fish) yielded different response classes (Figures 5C and 5D; see also Figure S4). Two classes were similar in that they contained neurons selective for largesize objects, but they differed in whether they also exhibited a strong OFF response after a whole-field flash (Figure 5D, left and middle columns). The third class, which comprised 19% of responsive SINs, showed an opposite size-tuning profile (Figures 5C and 5D, right columns), with fluorescence transients evoked primarily by small-size stimuli. Additionally, the response to whole-field flashes in this subset was negligible. Size indices were significantly different for the third class compared to the first two classes (Figure 5E), while the large-size-selective groups were significantly different from each other with respect to their OFF response (Figure 5F). Thus, we conclude that SINs fall into functionally distinct subgroups that are differentially recruited depending on object size. Size Classification in Separate Populations of Periventricular Neurons If the tectum participates in classification of large and small moving objects, we would expect that object size is represented further downstream in the tectal circuitry by activity in different subpopulations of periventricular neurons (PVNs). Yet, evidence for size classification in the PVN population is scarce. Therefore, we measured the size tuning of PVNs in a elavl3:gcamp5g line, which expresses GCaMP5G panneuronally (Figure 6A). Somatic fluorescence transients in PVNs showed a great functional diversity in response to moving targets and/or whole-field flashes (Figure 6B), as observed earlier [28]. Size selectivity maps were generated on a pixel-by-pixel basis (Figure 6A, bottom), and a size index was calculated for responsive PVNs (Figure 6C). This revealed populations of small-size- and large-size-selective cells, distributed without apparent spatial order (Figure 6A, bottom). A broad distribution of size indices was observed in individual fish (Figure 6D). The histogram of size selectivity pooled over five fish (n = 523 neurons) showed that 40% of cells were small-size selective (size index < 20.2), while 43% of PVNs were tuned to large objects (size index > 0.2) (Figure 6E). This indicates that a substantial fraction of tectal PVNs is active during the presentation of behaviorally relevant large objects. This corroborates the notion that at the PVN level, the tectum contributes to classifying both small and large targets and may mediate the suitable response, appetitive versus avoidance swims, depending on classification outcome.

6 Size Classification in Zebrafish Optic Tectum 2381 Figure 5. Size Tuning in Functionally Distinct Classes of SINs (A) Optical section representative of tectal expression in Oh:GCaMP6s fish. ROIs mark three SINs with fluorescence signals in response to moving targets. Scale bar, 40 mm. (B) Examples of fluorescence time course in ROIs shown in (A). The left and middle traces show the responses of large-size-selective SINs. Note difference in response to whole-field flash. On the right, the response of a small-size-selective SIN is shown. (C) Size-tuning curves of SINs shown in (A) and (B) (same color code) using variance analysis (Supplemental Experimental Procedures). The inset shows the proportion of each response type from all SINs (n = 133). (D) Raster plot of normalized fluorescence changes in response to stimuli of different sizes for the entire population, sorted using k-means clustering. Each row in the raster plot represents the fluorescence time course from an individual SIN soma. (E) Box-and-whisker plot of size indices for the three different response types of SINs shown in (D). Group III is statistically different (p = , Kruskal-Wallis-test/Bonferroni correction for multiple comparisons). Red lines indicate the median, boxes indicate the 25% 75%-quartile range, and error bars indicate the full range of data. (F) Box-and-whisker plot of ON/OFF indices calculated from responses to whole-field stimuli for each of the three response types analyzed in (D). Group I is statistically different (p = , Kruskal-Wallis-test with Bonferroni correction). See also Figure S4. Discussion What is the combination of input channels a given visual center receives from the multitude of RGC subtypes, and how does it integrate information across these highly specialized inputs? The main result of this study is that in the zebrafish optic tectum, large and small stimuli known to evoke different behaviors are already classified by selective activation of presynaptic RGC input channels. Unexpectedly, postsynaptic SINs fall into different subpopulations with distinct size-tuning properties, dependent on their target layer (Figure 7). Size Tuning in the Retinotectal Pathway By combining the merits of newly developed GECIs [25], multiphoton microscopy in the visual system [29], and behaviorally relevant visual stimulation, we could measure the tuning properties of retinal inputs to the tectum at high spatial resolution. This approach directly demonstrated the presence of retinal inputs selectively responding to small and large sizes in the retinorecipient layers. While small-size-driven signals tended to be relatively more frequent in the most superficial SO, large-size-driven signals were more frequent in SFGS sublayers, although we did not observe a strict laminar organization (Figures 7A and 7B). Therefore, small-size-selective retinal signals may also be transmitted in the SFGS. By contrast, some classes of direction-selective RGCs specifically target distinct sublayers in the superficial neuropil [18, 30]. What types of RGCs may contribute to size-dependent inputs in the optic tectum? In the case of small-size-selective inputs, a strong inhibitory surround mechanism in the retina most likely suppresses the firing rate when an object with wide leading and trailing edges moves across its receptive field and, by the same token, during whole-field flash stimulation. Versions of this functional RGC type are found in the retinae of most vertebrates, originally described as bug detectors in the frog by Lettvin et al. [31], and subsequently as local-edge detectors (LEDs) in the rabbit [6]. Further examples from this class include theta cells in the cat [32] and W3 RGCs in the mouse retina [33], which also exhibit projections to the SC. These cell types bear resemblance in that they have small dendritic fields arborizing in the central layers of the inner plexiform layer (IPL). In the larval zebrafish, we may derive an estimate of the dendritic field size of smallsize-selective RGCs by geometrical comparison. The inner diameter of the suppressive surround receptive field of LED-type RGCs is approximately one to two times that of its dendritic field diameter [33 35]. Suppression of smallsize-selective RGC signals was observed when or larger stimuli were used. This size corresponds to an area of approx. 30 mm 3 15 mm on the IPL, which is in good

7 Current Biology Vol 24 No Figure 6. Large- and Small-Size-Selective Responses in Periventricular Neurons (A) Top: optical section through periventricular layer in elavl3:gcamp5g fish. Bottom: size selectivity map for the same optical section, calculated from fluorescence signals in single pixels. Unresponsive pixels appear black, and responsive pixels insensitive to size appear gray (size index near 0). Examples of nine responsive PVNs are marked by circular ROIs. Scale bar, 20 mm. (B) Fluorescence time course in ROIs as indicated in (A) in response to moving targets and full-field flash. (C) Size-tuning curves calculated from signal variance in somatic ROIs indicated in (A). Size indices are indicated above the tuning curves. (D) Frequency histogram of size indices from cell bodies in a single fish from five optical sections. (E) Frequency histogram of size indices from cell body signals accumulated in five fish. correspondence with that of some RGC dendritic field sizes in the larval retina [23, 36]. We conclude that a defined set of RGCs with narrow dendritic fields selectively encodes small objects moving on a stationary background, which contributes critically to size classification in the retinotectal pathway. Indeed, the SO was recently shown to receive projections mainly from narrowfield bistratified and diffuse RGCs, which may therefore represent good candidates for transmitting small-size-selective input to the most superficial neuropil [23]. The origin of large-size-selective tectal inputs may lie in RGCs with larger dendritic fields and, if at all, weaker surround suppression. Notably, whole-field-responsive RGCs with wide dendritic fields in the mammalian retina, such as alpha/y RGCs, have been found to project to deep layers in the SC [37,38]. Furthermore, we expect large-sizeselective RGC dendrites to arborize at least partially in the OFF layers of the IPL, shown to be present already at early larval stages [39]. This is because (1) large-size-selective RGC terminals often exhibited a pronounced OFF response during whole-field flashes and (2) SINs selectively activated by large stimuli often received significant tonic OFF inputs (Figure S3). Structure and Function of Superficial Interneurons Direct patch clamp recordings from SINs revealed a remarkable diversity in their morphological and functional properties. What are the functional roles of SIN subtypes stratifying in distinct retinorecipient layers of the tectum? Because of their horizontal morphology and inhibitory phenotype [14], a comparison with retinal amacrine cells may be instructive. SINs probably carry a widespread global signal when excitation exceeds a certain threshold, similar to some spiking widefield amacrine cells. This may suggest that SINs provide lateral inhibition to adjacent tectal regions that receive input from other zones of the visual field. Particularly in the superficial SO layer, this may help to sharpen the positional information about a small target or inhibit simultaneously incoming information in a winner-take-all fashion, generated by competing small targets in multiple, spatially segregated hot spots of activity. Alternatively, a global spike mechanism in SINs may provide wide-field reciprocal inhibition between two hypothesized subnetworks that could process small objects and large objects, respectively (Figure 7C), which would argue for a role of SINs in pattern separation and object classification. Finally, it is possible that individual SIN dendrites may support local transmitter release at subthreshold excitation levels, which could shape the tuning properties of preand postsynaptic compartments, depending on the local connectivity, similar to circuit motifs in the retina or the olfactory system [29, 40]. The role of these cell types remains speculative, however, until the postsynaptic targets of SINs are identified, which may become possible using volume electron microscopy and targeted reconstructions [41]. How do the different types of SINs in the SO and the SFGS respond when prey-like objects move across the visual field? In this case, most SO SINs receive phasic excitation, which may suggest that in this layer, the positions of small-sized objects with prey-like value are preferentially processed. At the same time, deeper SINs in SFGS receive tonic excitatory input,

8 Size Classification in Zebrafish Optic Tectum 2383 complex spatiotemporal response both in the SFGS network and in a substantial fraction of PVNs (Figure 6). This sensitivity to large stimuli suggests that the tectum not only plays a role in detecting small, prey-like objects [42], but may also be critical for the detection of large objects, which usually evoke aversive motor responses. A tectal hemisphere sends efferents to premotor areas in the reticulospinal system via both ipsi- and contralateral projections [43], providing a putative anatomical substrate for target-directed and aversive turns. This is in accord with the notion that the optic tectum participates in the generation of orienting and avoidance behaviors in a wide range of species [2, 3, 44, 45]. Thus, the tectum may represent a bifurcation point at which stimulus classification biases the decision between mutually exclusive responses ( move toward or away ), contributing to adequate response selection (Figure 7C). Experimental Procedures Fish Care Fish larvae were raised at 27 C in embryo medium in a 14 hr/10 hr light/dark cycle. Experiments followed protocols in accordance with the guidelines of the German animal welfare law and approved by the local authorities. Figure 7. Proposed Connectivity of Size-Selective Circuits in the Retinotectal Pathway (A) Small-size-selective RGCs project to retinorecipient layers with a preference for the most superficial sublayers (SO). A subset of SINs costratifies and connects preferentially to this RGC input type. (B) Large objects and global motion are represented in the activity of a different set of RGCs, which predominate in deeper tectal layers. SINs with dendrites in these layers process activity in response to large objects and/or whole-field motion. PVNs are tuned to small or large objects, respectively. Their size tuning could result from specific connections in superficial or deeper retinorecipient layers. (C) Proposed tectal circuit for object classification and size-selective response selection. A tectal hemisphere contains two parallel output stages, one activated by large objects (large-size processing) the other by small objects (small-size processing). The small-size-processing network may receive small-size-selective retinal input predominantly in the superficial SO, whereas large-size-selective RGCs activate a separate tectal subnetwork, predominantly in the SFGS. These tectal subnetworks may inhibit each other reciprocally (dashed lines). The tuning curves in the insets schematically show activity (A) versus size (S). which could result in sustained inhibition of a large-size sensitive tectal subnetwork. Together, this may put the tectal network into a state facilitating the detection and classification of prey-like objects. However, when large objects move across the field, we expect a spatial wave of suppression and rebound excitation of RGCs and subsequent inhibition from deep SINs to sweep through the SFGS. Thus, large-size stimuli evoke a Calcium Imaging and Visual Stimulation Calcium imaging was performed using a custom-built multiphoton microscope and a 203, 1.0 numerical aperture water-immersion objective (Zeiss). For detection of presynaptic Ca 2+ signals in RGC afferents, GCaMP-positive offspring from a cross of atoh7:gal4 fish with UAS:GCaMP6m fish or UAS: GCaMP3 fish were used [18, 24]. The retinorecipient neuropil was imaged in a dorsal plane (between 5 and 30 mm from dorsal crest) and a ventral plane (between 30 and 60 mm). For Ca 2+ imaging in SIN cell bodies, Oh:GCaMP6s fish were used. For Ca 2+ imaging in PVN cell bodies, the established elavl3:gcamp5g line was used [46]. Larvae (5 8 days postfertilization) were anaesthetized and paralyzed using 0.02% MS-222 (tricaine methanesulfonate; Sigma Aldrich) and alpha-bungarotoxin (1 mg/ml; Tocris) in embryo medium, respectively, and mounted in a custom-made acrylic glass chamber. With a microprojector, stimuli were presented on a diffusive screen. So that interference of visual stimulation with fluorescence detection could be avoided, the stimulus light was synchronized with the flyback interval of the fast x-scan galvanometric mirror. This generated a virtually flicker-free stimulus covering the entire visible spectrum. Stimuli consisted of white rectangles (aspect ratio 2:1) moving on a gray structured background in a horizontal range of 120 at a speed of 40 /s. Distortion in size when the target moved to the periphery of the curved screen was compensated for by a nonlinear transformation. Electrophysiology Whole-cell patch-clamp recordings were performed using a Multiclamp 700B amplifier (Molecular Devices). For voltage-clamp recordings, the intracellular solution contained cesium-gluconate to minimize potassium leak currents. Sulforhodamine-B ( mm) was added to visualize the morphology of patched cells. The holding potential was 260 mv. SINs were patched in the established pou4f3:gfp line, in which SIN cell bodies were identified as shadows in the superficial neuropil, selected randomly for patch-clamp and morphological analysis. Alternatively, SINs were patched in the Oh:GCaMP6s line, in which Ca 2+ imaging could be used to preselect SINs with small-size-selective Ca 2+ signals. Data Analysis Calcium Signals Fluorescence signals in the neuropil were extracted pixel by pixel from image time series, averaged for two repetitions of stimulus presentation, and normalized. The distribution of each pixel s fluorescence values over time was tested for skewness. An empirically determined threshold for skewness (jskew.j > 0.4) was used to distinguish pixels that exhibited transient, stimulus-evoked fluorescence changes from those that contained random noise. Fluorescence traces from SINs and PVNs were extracted from circular ROIs manually drawn around the soma. A size selectivity index (SI) was calculated according to S.I. = (Var large 2 Var small ) / (Var large + Var small ), where Var large was the averaged variance during presentation of

9 Current Biology Vol 24 No large stimuli (16,32, and 64 ) and Var small was the averaged variance during presentation of small stimuli (2,4, and 8 ), resulting in indices from 21 (small-size selective) to +1 (large-size selective). For analysis of SIN size tuning in the Oh:GCaMP6s line, the fluorescence time course from SIN somata was classified into three response categories using k-means clustering in MATLAB. Electrophysiology Because different RGC types can increase or decrease their spike rate in response to a specific stimulus, we took the variance of SIN input current during stimulus presentation as a generalized measure of sensitivity to a given stimulus size. Variance values of SIN input currents were averaged for small (2,4, and 8 ) and large (16,32, and 64 ) stimuli, respectively, and a size index was calculated as above. Fluorescence traces and electrophysiological recordings shown in Figures 2, 4, 5, and 6 are averages from two to three stimulus repetitions. All values are given as mean 6 SEM, if not indicated otherwise. A more detailed description of methods is provided in the Supplemental Experimental Procedures. Supplemental Information Supplemental Information includes Supplemental Experimental Procedures, four figures, and one movie and can be found with this article online at Author Contributions S.J.P., C.A.T., and J.H.B. conceived the project and designed experiments. S.J.P. and C.A.T. performed experiments. C.M.v.B.-M. and S.R. generated transgenic lines. S.J.P., C.A.T., and J.H.B. analyzed data and wrote the manuscript. Acknowledgments We thank W. Denk for support and helpful discussions; J. Tritthardt and C. Kieser for help with electronic design; U. Herget and J. Nassal for confocal imaging and injections; G. Giese and A. Scherbarth for help with the confocal microscope; J. Gabriel for help with initial experiments; M. Kaiser, A. Scherbarth, and E. Stier for expert technical help; W. Denk and T. Gollisch for comments on the manuscript; and L. Poggi, J. Wittbrodt, E. Yaksi, and A. Schier for sharing transgenic lines. This work was supported by the Max Planck Society and the Deutsche Forschungsgemeinschaft (BO3746). J.H.B. and S.R. are members of the Interdisciplinary Centre for Neurosciences (IZN) and the Excellence Cluster CellNetworks at Heidelberg University. Received: August 22, 2014 Revised: August 31, 2014 Accepted: September 1, 2014 Published: September 18, 2014 References 1. Itti, L., and Koch, C. (2001). Computational modelling of visual attention. Nat. Rev. Neurosci. 2, Dean, P., Redgrave, P., and Westby, G.W. (1989). Event or emergency? Two response systems in the mammalian superior colliculus. Trends Neurosci. 12, DesJardin, J.T., Holmes, A.L., Forcelli, P.A., Cole, C.E., Gale, J.T., Wellman, L.L., Gale, K., and Malkova, L. (2013). Defense-like behaviors evoked by pharmacological disinhibition of the superior colliculus in the primate. J. Neurosci. 33, Gahtan, E., Tanger, P., and Baier, H. (2005). Visual prey capture in larval zebrafish is controlled by identified reticulospinal neurons downstream of the tectum. J. Neurosci. 25, Fajardo, O., Zhu, P., and Friedrich, R.W. (2013). Control of a specific motor program by a small brain area in zebrafish. Front. Neural Circuits 7, Levick, W.R. (1967). Receptive fields and trigger features of ganglion cells in the visual streak of the rabbits retina. J. Physiol. 188, Roska, B., and Werblin, F. (2001). Vertical interactions across ten parallel, stacked representations in the mammalian retina. Nature 410, Olveczky, B.P., Baccus, S.A., and Meister, M. (2003). Segregation of object and background motion in the retina. Nature 423, Bölinger, D., and Gollisch, T. (2012). Closed-loop measurements of isoresponse stimuli reveal dynamic nonlinear stimulus integration in the retina. Neuron 73, Dhande, O.S., and Huberman, A.D. (2014). Retinal ganglion cell maps in the brain: implications for visual processing. Curr. Opin. Neurobiol. 24, Adesnik, H., Bruns, W., Taniguchi, H., Huang, Z.J., and Scanziani, M. (2012). A neural circuit for spatial summation in visual cortex. Nature 490, Sachdev, R.N., Krause, M.R., and Mazer, J.A. (2012). Surround suppression and sparse coding in visual and barrel cortices. Front. Neural Circuits 6, Sajovic, P., and Levinthal, C. (1982). Visual response properties of zebrafish tectal cells. Neuroscience 7, Del Bene, F., Wyart, C., Robles, E., Tran, A., Looger, L., Scott, E.K., Isacoff, E.Y., and Baier, H. (2010). Filtering of visual information in the tectum by an identified neural circuit. Science 330, Portugues, R., Feierstein, C.E., Engert, F., and Orger, M.B. (2014). Whole-brain activity maps reveal stereotyped, distributed networks for visuomotor behavior. Neuron 81, Kubo, F., Hablitzel, B., Dal Maschio, M., Driever, W., Baier, H., and Arrenberg, A.B. (2014). Functional architecture of an optic flow-responsive area that drives horizontal eye movements in zebrafish. Neuron 81, McLean, D.L., Fan, J., Higashijima, S., Hale, M.E., and Fetcho, J.R. (2007). A topographic map of recruitment in spinal cord. Nature 446, Gabriel, J.P., Trivedi, C.A., Maurer, C.M., Ryu, S., and Bollmann, J.H. (2012). Layer-specific targeting of direction-selective neurons in the zebrafish optic tectum. Neuron 76, Miri, A., Daie, K., Arrenberg, A.B., Baier, H., Aksay, E., and Tank, D.W. (2011). Spatial gradients and multidimensional dynamics in a neural integrator circuit. Nat. Neurosci. 14, Joshua, M., and Lisberger, S.G. (2014). A tale of two species: neural integration in zebrafish and monkeys. Neuroscience. Published online May 2, Bianco, I.H., Kampff, A.R., and Engert, F. (2011). Prey capture behavior evoked by simple visual stimuli in larval zebrafish. Front. Syst. Neurosci. 5, Trivedi, C.A., and Bollmann, J.H. (2013). Visually driven chaining of elementary swim patterns into a goal-directed motor sequence: a virtual reality study of zebrafish prey capture. Front. Neural Circuits 7, Robles, E., Laurell, E., and Baier, H. (2014). The retinal projectome reveals brain-area-specific visual representations generated by ganglion cell diversity. Curr. Biol. Published online August 21, doi.org/ /j.cub Zolessi, F.R., Poggi, L., Wilkinson, C.J., Chien, C.B., and Harris, W.A. (2006). Polarization and orientation of retinal ganglion cells in vivo. Neural Dev. 1, Chen, T.W., Wardill, T.J., Sun, Y., Pulver, S.R., Renninger, S.L., Baohan, A., Schreiter, E.R., Kerr, R.A., Orger, M.B., Jayaraman, V., et al. (2013). Ultrasensitive fluorescent proteins for imaging neuronal activity. Nature 499, Xiao, T., Roeser, T., Staub, W., and Baier, H. (2005). A GFP-based genetic screen reveals mutations that disrupt the architecture of the zebrafish retinotectal projection. Development 132, Jetti, S.K., Vendrell-Llopis, N., and Yaksi, E. (2014). Spontaneous activity governs olfactory representations in spatially organized habenular microcircuits. Curr. Biol. 24, Niell, C.M., and Smith, S.J. (2005). Functional imaging reveals rapid development of visual response properties in the zebrafish tectum. Neuron 45, Euler, T., Detwiler, P.B., and Denk, W. (2002). Directionally selective calcium signals in dendrites of starburst amacrine cells. Nature 418, Nikolaou, N., Lowe, A.S., Walker, A.S., Abbas, F., Hunter, P.R., Thompson, I.D., and Meyer, M.P. (2012). Parametric functional maps of visual inputs to the tectum. Neuron 76, Lettvin, J.Y., Maturana, H.R., McCulloch, W.S., and Pitts, W.H. (1959). What the frog s eye tells the frog s brain. Proc. Inst. Radio Eng. 47,

10 Size Classification in Zebrafish Optic Tectum Isayama, T., Berson, D.M., and Pu, M. (2000). Theta ganglion cell type of cat retina. J. Comp. Neurol. 417, Zhang, Y., Kim, I.J., Sanes, J.R., and Meister, M. (2012). The most numerous ganglion cell type of the mouse retina is a selective feature detector. Proc. Natl. Acad. Sci. USA 109, E2391 E van Wyk, M., Taylor, W.R., and Vaney, D.I. (2006). Local edge detectors: a substrate for fine spatial vision at low temporal frequencies in rabbit retina. J. Neurosci. 26, Russell, T.L., and Werblin, F.S. (2010). Retinal synaptic pathways underlying the response of the rabbit local edge detector. J. Neurophysiol. 103, Mumm, J.S., Williams, P.R., Godinho, L., Koerber, A., Pittman, A.J., Roeser, T., Chien, C.B., Baier, H., and Wong, R.O. (2006). In vivo imaging reveals dendritic targeting of laminated afferents by zebrafish retinal ganglion cells. Neuron 52, Tamamaki, N., Uhlrich, D.J., and Sherman, S.M. (1995). Morphology of physiologically identified retinal X and Y axons in the cat s thalamus and midbrain as revealed by intraaxonal injection of biocytin. J. Comp. Neurol. 354, Huberman, A.D., Manu, M., Koch, S.M., Susman, M.W., Lutz, A.B., Ullian, E.M., Baccus, S.A., and Barres, B.A. (2008). Architecture and activity-mediated refinement of axonal projections from a mosaic of genetically identified retinal ganglion cells. Neuron 59, Nikolaev, A., Leung, K.M., Odermatt, B., and Lagnado, L. (2013). Synaptic mechanisms of adaptation and sensitization in the retina. Nat. Neurosci. 16, Urban, N.N., and Arevian, A.C. (2009). Computing with dendrodendritic synapses in the olfactory bulb. Ann. N Y Acad. Sci. 1170, Briggman, K.L., Helmstaedter, M., and Denk, W. (2011). Wiring specificity in the direction-selectivity circuit of the retina. Nature 471, Muto, A., Ohkura, M., Abe, G., Nakai, J., and Kawakami, K. (2013). Realtime visualization of neuronal activity during perception. Curr. Biol. 23, Sato, T., Hamaoka, T., Aizawa, H., Hosoya, T., and Okamoto, H. (2007). Genetic single-cell mosaic analysis implicates ephrinb2 reverse signaling in projections from the posterior tectum to the hindbrain in zebrafish. J. Neurosci. 27, Gandhi, N.J., and Katnani, H.A. (2011). Motor functions of the superior colliculus. Annu. Rev. Neurosci. 34, Billington, J., Wilkie, R.M., Field, D.T., and Wann, J.P. (2011). Neural processing of imminent collision in humans. Proc. Biol. Sci. 278, Ahrens, M.B., Orger, M.B., Robson, D.N., Li, J.M., and Keller, P.J. (2013). Whole-brain functional imaging at cellular resolution using light-sheet microscopy. Nat. Methods 10,

11 Current Biology, Volume 24 Supplemental Information Classification of Object Size in Retinotectal Microcircuits Stephanie J. Preuss, Chintan A. Trivedi, Colette M. vom Berg-Maurer, Soojin Ryu, Johann H. Bollmann

12 Fig. S1 (related to Fig. 2): Distinct small-size and large-size-selective signals observed in sparsely labeled RGC fibers. (A) Atoh7:Gal4 embryos were injected with a UAS:GCaMP6m construct, resulting in sparse, random expression of GCaMP6m in RGC fibers. Left panels: Optical sections of the tectal neuropil of two fish. Imaging depth 26 µm (top left) and 50 µm (bottomm left) from dorsal crest. Right panels: Corresponding size selectivity maps. Examples of size-selective RGC compartments are highlighted by orange boxes and arrowheads. Scale bars 10 µm. (B) Normalized fluorescence signalss from small tectal ROIs (indicated in (A)) in response to horizontally moving targets and full-field flash. Each trace is an average from two stimuluss presentations. Object size indicated above stimulus trace. (C) Size tuning curves calculated from signal variance of traces in (B). Variance was measured and averaged for the 3 presentationn intervals for small target size (blue boxes in (B)) and large target size (green boxes in (B)), respectively. Size index indicated above each tuning curve.

13 Fig. S2 (related to Fig. 3): Visualization of specific retinorecipient sublayers within the tectal neuropil (A) Overview of tectal neuropil in a double transgenic atoh7:gap43-rfp/pou4f3:gfp larva ( 6 dpf). A vertical and two horizontal optical sections from a confocal stack containing a complete tectal hemisphere. (B) Optical sections as in (A). RFP-fluorescence between the most superficial and deeper neuropil in the shown in red, GFP-fluore escence shown in cyan. Note the gap atoh7:gap43-rfp line (arrowheads). Note thin SO band, a wider gap and the SFGS D,F in the pou4f3:gfp line, as indicated. Scale bar: 40 m. This imaging approach revealed that the RGC axons projecting into SO in the pou4f3:gfp line represent a subset of the most superficial RGC axons labeled in the atoh7 line. Likewise, the deep sublayer SFGSS F in pou4f3:gfp colocalized with the deep SFGS boundary in the atoh7: :gap43-rfp layer. A thin gap that is typically visible in the atoh7 expression pattern (B, white arrowheads) colocalizes with the superficial part of a prominent gap typically observed between the SO and the SFGS D,F in the pou4f3-gfp expression pattern.

14 Fig. S3 (related to Fig. 4): ON/OFF input currents depend on dendrite depth (A) Postsynaptic currents in a SIN (same cell as in Fig. 4A) during whole-field flashes (2.5 s duration) ). Note increased input activity during background (I BKG G), and suppression of input during whole-field illumination (I On,tonic ). This cell had a pronounced phasic OFF response (I Off,p ) and a small phasic ON responsee (I On,p ). Measurement intervals indicated below traces.

15 (B) Example of a SIN with dendrites in SO (same cell as in Fig. 4D). Note absence of synaptic input currents in response to whole-field flashes. (C) Scatter plot of background current versus dendrite depth of SINs in the pou4f3:gfp line (gray symbols, n = 41) and preselected SINs in the Oh:GCaMP6s line (red symbols with crosses at nominal depth of 100%, n = 7). Right: Data grouped into SFGS and SO cells based on dendrite depth (depth criterion: 90% for SO ). Groups were statistically different (p = 3.6 x 10-6, Wilcoxon rank-sum test). (D) Scatter plot of phasic ON currents versus dendrite depth. Right: Grouped data. Groups were statistically different (p = 4.8 x 10-4, Wilcoxon rank-sum test). (E) Scatter plot of phasic OFF currents versus dendrite depth. Right: Grouped data. Groups were statistically different (p = 2.1 x 10-7, Wilcoxon rank-sum test). (F) Phasic ON and OFF responses on expanded time scale. (G) Response latencies of synaptic currents after ON edge and OFF edge of wholefield flash. Phasic OFF currents exhibited longer delays (p = 3.0 x 10-6, Wilcoxon rank-sum test). Classically, RGCs can be functionally grouped into phasically or tonically active neurons in response to the ON or OFF edge of a whole-field flash. SINs exhibited differences in their synaptic input patterns with respect to tonic and phasic current components in a layer-dependent fashion. Often, a high level of background current was observed that was suppressed during whole-field illumination. In neurons receiving tonic background activity, we often observed a pronounced phasic ON or OFF response, or both (Fig. S3A). By contrast, in other SINs we detected only small background currents, and the phasic ON or OFF responses were smaller or absent, making these SINs virtually insensitive to whole-field flashes (Fig. S3B). Similar to the distribution of size indices in our data set (Fig. 4I), the phasic and tonic components of these synaptic inputs were clearly depth-dependent (Fig. S3C-E). Indeed, the small-size-selective SINs in the SO exhibited small phasic and tonic input currents, while SINs with dendrites in SFGS often showed substantial background activity and phasic responses to the ON or OFF edge of whole-field flashes. Furthermore, we analyzed the response latencies in cells that exhibited an appreciable phasic ON or OFF current in response to whole-field flashes (Fig. S3F). Phasic responses following the ON-edge of the stimulus had a median latency of 49.8 ms, consistent with response latencies of ON-type RGCs at this stage [S1]. Responses following the OFF-edge had a longer median latency of 129 ms (p = 3.0

16 x 10-6, Wilcoxon rank-sum test). The finding that large-size-selective RGC fibers often exhibited brief fluorescence transients in response to the OFF edge of a wholefield flash (Fig. 2) is consistent with these RGCs being the source of phasic OFF currents in SINs. Taken together, these data lend further support to the notion that RGCs directly drive different SIN subtypes and contribute to their distinct tuning curves in a layer-dependent fashion.

17 Fig. S4 (related to Fig. 5): SINs fire action potentials. (A) Voltage clamp recording (I post, bottom black trace), subsequent current clamp recording (V m, bottom gray trace) and derivative of current clamp recording (dv/dt, upper gray trace) during presentation of a 2 target. Stimulus is indicated on top (Stim.). Action potentials are detectedd when derivative of the voltage trace crosses a lower threshold ( magenta line). Action potentials marked as ticks in an event channel (top black). (B) Same cell as in (A) during presentation of a 64 target. Note strong excitatory inputs in voltage clamp recording (I post, bottom trace) and larger depolarizations in current clamp recording with similar time course relative to the position of the large moving target.