To reconstruct the full chemical synaptic connectivity map of the visual eyes in the Platynereis larva we used ssTEM. Following high-pressure freezing and cryosubstitution, we collected 1690 thin sections (40–50 nm) from a single Epon-embedded 3-day-old larva, encompassing the entire head and part of the first trunk segment. Following contrasting, we imaged the sections at 3.7 nm/pixel resolution, in a 140 µm × 140 µm × 80 µm volume (Figure 1A–C; Video 1).

Figure 1 with 9 supplements see all

Download asset Open asset

Video 1

Download asset

We identified the eyes based on the presence of pigment-filled vesicles in the pigment cells, the presence of a lens formed by the apical extensions of the pigment cells, and the presence of the apical microvillar extensions (rhabdoms) of the photoreceptors (Rhode, 1992; Randel et al., 2013; Figure 1D; Video 2). We then manually traced all neurons lying on synaptic pathways from photoreceptors to locomotor organs (muscles and ciliary bands). We identified 21 photoreceptor cells (PRC), 42 interneurons (IN) and 8 motorneurons (MN) (Figure 1—figure supplement 1–4), forming a ‘minimal eye circuit’ from sensors to effectors. We identified 1106 chemical synapses between these cells and their motor targets. We found further sensory neurons other than the photoreceptors providing direct or indirect input to the minimal eye circuit. These upstream circuits will be described in detail elsewhere. In this paper we focus on the analysis of the 71 neurons constituting the minimal eye circuit.

Video 2

Download asset

First, we describe the anatomy and synaptic contacts of the eyes and the main neuron types identified as part of the eye circuit. To provide an easily accessible representation of the data we generated a comprehensive 3D anatomical atlas of all neurons, synapses, muscles and ciliated cells using Blender (http://www.blender.org/) (Randel et al., 2014; Video 3). Neuron morphologies were reconstructed based on the raw traces (Figure 1—figure supplement 1–4) and represented in the atlas with a simplified morphology (Figure 2). Users can display groups of neurons, query for pre- and post-synaptic partners of neurons, and display synapses.

Video 3

Download asset

Figure 2 with 1 supplement see all

Download asset Open asset

The neurons of the eye circuit only contain axo-axonal synapses. With the exception of the photoreceptors, all neurons lack dendrites. Axons have a median diameter of 160 nm (Figure 1—figure supplement 5), allowing reliable tracing of most processes. The morphology of all neurons is very simple, with one primary branch giving rise to several very short secondary branches in the synaptic regions of the optic neuropils (Figure 1—figure supplement 1–4). These short branches often contain pre- or postsynaptic sites (Video 4). Only two neurons have a branched main axon (MN^r3, MN^l1, see ‘Materials and methods’ for nomenclature).

Video 4

Download asset

Synapses were identified as electron dense accumulations of several vesicles close to the presynaptic membrane (no T-bars as in Drosophila). Prominent postsynaptic structures were not detected, in agreement with previous ultrastructural studies in annelids (Wells et al., 1972). 69% of the synapses could be identified in at least two and up to five consecutive sections (Figure 1—figure supplement 5). To further characterize the ultrastructure of synapses we created high-resolution images (0.2 nm/pixel) of 60 randomly chosen synapses belonging to the different neuron types (Figure 1—figure supplement 6–9). In all cases we observed a cluster of vesicles adjacent to the plasmamembrane, but no pre- or post-synaptic specializations.

We could not identify gap junctions in the eye circuit. Gap junctions exist in annelids, and we could find several innexin genes, encoding gap junction proteins of invertebrates (Kandarian et al., 2012), in the Platynereis transcriptome (data not shown). However, we did not see structures similar to ultrastructurally characterized annelid gap junctions (Muller and Carbonetto, 1979; Shen et al., 2002), even in a stack of high-resolution (1.13 nm/pixel) images of the primary optic neuropil (Randel et al., 2014). If gap junctions are present in the larval stage they may be too small to be distinguished from obliquely cut membranes even at high resolution.

The photoreceptors of the four eyes (3–7 per eye) project to a primary optic neuropil area at the center of the larval brain (Figure 1C,E, Figure 2B), consistent with results obtained by the genetic labeling of the photoreceptors (Backfisch et al., 2013). All photoreceptors from the same eye project along the same axon bundle, forming four separate eye nerves (Figure 2A,B).

The eye nerves from the anterior and posterior eyes from the same side innervate distinct anterior and posterior areas in the primary optic neuropil (Figure 1E,F). Some photoreceptor axons cross the midline and project to the contralateral primary optic neuropil. The primary optic neuropil is surrounded by three giant glia cells. These cells form lamellae that tightly surround the primary and secondary optic neuropils from the dorsal but not the ventral side (Figure 1F, Figure 2C; Video 4). The main targets of the photoreceptors of the four eyes are four primary interneurons (IN¹), with photoreceptors from one eye forming several (up to 30) synaptic contacts to one primary interneuron with a contralateral cell body (e.g., PRC^al to IN^1pr; Video 4). The four eyes and the four primary interneurons show a crosswise arrangement (Figure 2D). Primary interneurons have mutual synaptic contacts (see below) and also form synapses on two other interneuron types (Figure 2E,H; Video 3; Randel et al., 2014). One type of interneuron is intrinsic to the optic neuropil (IN^int) with both ipsilateral and contralateral cell bodies and axons crossing the midline. Another target of the primary interneurons are interneurons that project out of the primary optic neuropil into a secondary optic neuropil area (Figure 1G). The major targets of these trans-optic-neuropil interneurons (IN^ton) are a group of contralateral interneurons, which we named Schnörkel interneurons (IN^sn). Schnörkel interneurons project a curved axon to the ipsilateral secondary optic neuropil (Figure 2H, Figure 2—figure supplement 1). The Schnörkel interneurons are presynaptic to a bilateral group of ventral motorneurons (MN; Figure 2—figure supplement 1). Two groups of dorsal and ventral interneurons (IN^dc, IN^vc) also form synapses on the Schnörkel interneurons and motorneurons (Figure 2H; Video 3).

We identified two distinct motorneuron types. The first type sends a contralateral projection to both the ciliary band and the dorsal longitudinal muscle, and branches to send a descending projection to the trunk (Figure 2G, Figure 2—figure supplement 1). The second type only projects posteriorly along the circumesophageal nerve after crossing the midline. Both motorneuron types form neuromuscular and neurociliary synapses (Figure 1H, Figure 2—figure supplement 1; Video 3, Video 5). Due to the lack of further trunk sections we did not trace descending motorneuron axons along their entire length, therefore we cannot exclude the possibility that motorneurons have other synaptic partners in the trunk. Nevertheless, these data represent, to our knowledge, the most complete connectomic reconstruction of a visual circuit to date, from photoreceptors to effector muscles and ciliated cells.

Video 5

Download asset

To gain insights into the nature of synapses in the eye circuit we measured vesicle diameter in the high-resolution images of synapses (Figure 3—figure supplement 1). The mean diameter of synaptic vesicles in interneurons was significantly larger than in photoreceptors and motorneurons. Photoreceptor synaptic vesicles were not significantly different from neuromuscular synaptic vesicles in motorneurons. However, neurociliary synaptic vesicles in motorneurons were significantly smaller than photoreceptor or neuromuscular synaptic vesicles. These observations indicate the use of different neurotransmitters in the eye circuitry and also suggest that the dual-function muscle- and cilio-motor neurons (MN^r2, MN^l1) may have mixed neurotransmitter content.

To identify possible neurotransmitters in the eye circuit we performed whole-mount RNA in situ hybridizations with marker genes for various neurotransmitters (Randel et al., 2014) and compared the expression domains to cell body positions in the TEM series (Figure 3). In Platynereis larvae the colocalization of distinct gene expression patterns can be determined with near cellular resolution using whole-mount RNA in situ hybridizations and image registration, allowing the molecular profiling of cell types (Tomer et al., 2010; Asadulina et al., 2012). Although expression patterns for several of these genes have already been reported in Platynereis (Denes et al., 2007; Jékely et al., 2008; Tomer et al., 2010), we acquired new whole-body expression data to analyze patterns in the 3-day-old brain in more detail. We found that vesicular glutamate transporter (VgluT), a marker of glutamatergic neurons, colocalized with r-opsin1, a marker of eye photoreceptors (Arendt et al., 2002; Randel et al., 2013; Figure 3—figure supplement 2, Figure 3—figure supplement 3), indicating that the photoreceptors are glutamatergic. In contrast, histidine decarboxylase (hdc), an enzyme catalyzing the synthesis of histamine, the transmitter in arthropod rhabdomeric photoreceptors (Stuart, 1999), does not localize to the eyes, as shown by image registration and double RNA in situ hybridization (Figure 3—figure supplement 3).

Figure 3 with 3 supplements see all

Download asset Open asset

In the cell body regions of the interneurons we found expression of several monoaminergic markers in single cells or small cell clusters. These include tryptophan hydroxylase (TrpH), tyrosine hydoxylase (TyrH), hdc, dopa-beta-hydoxylase (dbh), markers for serotonergic, dopaminergic, histaminergic, and noradrenergic neurons, respectively (Figure 3B, Figure 3—figure supplement 2, Figure 3—figure supplement 3). This indicates that the interneurons of the eye circuit use several distinct monoamine neurotransmitters.

Choline acetyltransferase (ChAT) and vesicular acetylcholine transporter (VAChT), two markers of cholinergic neurons, are localized in the cell body region of the ventral motorneurons (Figure 3C, Figure 3—figure supplement 2). This is consistent with earlier observations showing that the longitudinal trunk muscles, targets of the ventral motorneurons, are influenced by acetylcholine (Denes et al., 2007). A marker for GABAergic cells, glutamate decarboxlase (gad), partly colocalizes with the ventral cholinergic domain, in the area of the motorneuron cell bodies (Figure 3C, Figure 3—figure supplement 2, Figure 3—figure supplement 3). This suggests that some motorneurons may have mixed GABAergic and cholinergic identity. The difference in mean vesicle size at neuromuscular and neurociliary synapses in motorneurons (Figure 3—figure supplement 1) also supports a mixed neurotransmitter identity. Further cellular resolution mapping will be needed to confirm this and to establish the neurotransmitter identities of individual interneurons and motorneurons.

For a detailed analysis of synaptic contacts in the eye circuit, we represented the connectome as a directed graph and as an all-against-all synaptic connectivity matrix (Figure 4A, Figure 4—figure supplement 1, Figure 4—source data 1). In the graph, the nodes correspond to neurons and the directed edges to synaptic contacts, weighted by synapse number. We displayed the full cell complement (79 nodes; Figure 4A) and a trimmed graph (61 nodes) where all cells with <3 pre- and post-synaptic sites were omitted and only edges of three or more synapses were shown (Figure 4—figure supplement 2). Manual or force-field-based clustering was used for graph layout. We also constructed a merged graph where the same neuron types from the left and right body sides were grouped. In this graph the weight of connections is represented as the maximum number of synapses between any individual cell pair of the distinct groups (Figure 4B, Figure 4—source data 2).

Figure 4 with 6 supplements see all

Download asset Open asset

Figure 4—source data 1

All-against-all connectivity matrix of the Platynereis eye circuit.

https://doi.org/10.7554/eLife.02730.025

Download elife-02730-fig4-data1-v2.csv

Figure 4—source data 2

Grouped connectivity matrix of the Platynereis eye circuit.

https://doi.org/10.7554/eLife.02730.026

Download elife-02730-fig4-data2-v2.csv

To analyze information flow we calculated the maximum distance of all nodes to any other node within the directed graph (eccentricity; Figure 4A). The eye circuit is a feed-forward circuit where information flows from the eyes to the muscles and ciliary bands. We also calculated centrality measures to identify strongly connected nodes in the network (Figure 4—figure supplement 2). With three different centrality measures (eigenvector centrality, weighted-degree centrality and authority) the four primary interneurons (IN¹) ranked among the top five nodes (Randel et al., 2014). These four cells receive several synapses from the photoreceptors, with the highest number of each IN¹ received from the crosswise eyes (Figure 4B, Figure 4—figure supplement 2). The four IN¹ cells also have strong mutual connections (see below).

A search for highly interconnected nodes or communities (Blondel et al., 2008) subdivided the eye network into four modules (Figure 4—figure supplement 3). The four IN¹ cells belong to two different modules in the eye circuit (Figure 4—figure supplement 3), which together contain all neurons with projections intrinsic to the primary optic neuropil. Both modules consist of a crosswise pair of eyes and the associated IN¹ cells and trans-optic-neuropil interneurons (IN^ton). The anatomy of the circuitry in the primary optic neuropil thus displays point symmetry rather than bilateral symmetry (Figure 4—figure supplement 3).

The other two modules, the motor modules, contain all neurons intrinsic to the secondary optic neuropil, including Schnörkel interneurons and motorneurons and associated effector organs. A bilateral pair of IN^ton cells links the modules of the primary and secondary optic neuropils, reflecting the anatomy where IN^ton cells receive synapses in the primary and project to the secondary optic neuropil (Figure 4—figure supplement 3). One of the motor modules provides innervation to the prototroch and metatroch ciliary band cells. The other module contains motorneurons that connect to three out of four bundles of longitudinal muscles (we could not find neuromuscular synapses on the left dorsal longitudinal muscle) (Figure 4A,B, Figure 4—figure supplement 2).

The connectome reveals that all four eyes can provide motor input to both body sides. The neuronal pathways from the eyes cross the midline multiple times throughout the circuit. The strongest synaptic connections are formed between photoreceptor—IN¹—IN^ton—Schnörkel interneuron, always to the contralateral side in a feed-forward circuit (Figure 4B). However, most Schnörkel interneurons form several synapses on both ipsilateral and contralateral motorneurons in the secondary optic neuropil, representing a bilateral divergence of the connections. Motorneuron axons receive cell-body-proximal synapses from the ipsilateral Schnörkel interneurons, cross the midline, and receive cell-body-distal synapses on the other side of the body from the contralateral Schnörkel interneurons (Figure 4—figure supplement 6G,H; Video 5).

In order to estimate how stereotypic synaptic connectivity is within the circuit we compared the connectivity matrices of neurons with cell bodies on the left or right side of the body. We used the merged graph (Figure 4B) since not all neurons had a contralateral pair (e.g., we identified three IN^ton cells on the left and two on the right body side). We found a strong correlation between the left and right connectivity matrices (Spearman r = 0.67, p<0.0001; Figure 4—figure supplement 4).

We also analyzed the stereotypy of the spatial arrangement of pre-and postsynaptic sites among neurons of the same type. The four IN¹ cells always receive photoreceptor input in a cluster at the most distal segment of their axon (Figure 4—figure supplement 5). Synapses onto the IN¹ cells from the contralateral photoreceptors are intermingled with presynaptic sites of IN¹ cells to IN^ton cells (Figure 4—figure supplement 6) and are not spatially segregated within the cluster. In contrast, IN¹ cells receive axo-axonal synaptic input from the crosswise IN¹ cell in a cluster positioned more proximal to the cell body (Figure 4—figure supplement 5). The spatial arrangement of synapses is also similar between the left and right sides for other neuron types (Figure 4—figure supplement 6). These observations show that both the number and the spatial arrangement of synapses are stereotypic between the left and right body sides. The reconstruction of further individuals will be needed to assess individual-to-individual variation.

During development of the Platynereis larva, photoreceptors are continuously added to the eyes at the periphery of the pigment cup (Rhode, 1992). Individual photoreceptors in an eye are thus at different stages of differentiation within the same individual. We noted that in our reconstructed specimen the photoreceptor rhabdoms have different sizes, and that rhabdom volumes positively correlate with axon length and synapse number (Figure 5A,B). Using rhabdom volume as a proxy to photoreceptor differentiation we then analyzed how photoreceptor connections change during neuron maturation (Figure 5D).

Figure 5

Download asset Open asset

Photoreceptors with short axons connect weakly if at all to the primary interneurons (IN¹), while photoreceptors with longer axons form more synapses on IN¹ cells. Photoreceptors with the shortest axons have connections to other interneurons (e.g., IN^intl4, IN^intr2) that are not observed in photoreceptors with longer axons (Figure 5D). We defined a photoreceptor connectivity-maturation index (number of crosswise IN¹ connections per all connections) and found that it correlates with rhabdom volume (Figure 5C). These data represent a snapshot during development in one individual, but suggest that during photoreceptor development connections to the IN¹ cells get stronger and initial contacts to other interneurons are eliminated.

The IN^ton cells also show variation in axon length and connectivity, possibly also reflecting developmental progression. The small number of these interneurons precluded similar analyses.

The innervation of the longitudinal trunk muscles and ciliary bands by the eye circuit motorneurons suggested that the eyes could regulate tail bending and ciliary beating during larval swimming. To understand the function of the eye circuit we next analyzed larval photobehavior in detail, in conjunction with eye ablation experiments.

We first analyzed swimming trajectories of freely behaving 3- and 4-day-old larvae exposed to light stimuli of alternating directionality. 3- and 4-day-old larvae swim using their cilia while rotating around their anterior–posterior axis. When we used alternating illumination from the two opposite sides of an assay cuvette we observed directional swimming (phototaxis) of the larvae. Larvae showed mixed behavior, with some swimming towards and others away from the light source (Video 6). Such sign-switch in directional swimming responses is common for marine larvae and can be influenced by various environmental stimuli (Thorson, 1964; Young and Chia, 1982; Marsden, 1990). During reorientations we could observe prominent body bending in 4-day-old larvae and weaker bending in 3-day-old larvae, suggesting that larvae turn by contracting longitudinal muscles while swimming with cilia (Video 7 and data not shown). In agreement with the cholinergic identity of the motorneurons, treatment with mecamylamine, an acetylcholine receptor antagonist, blocked negative phototaxis. Mecamylamine treatment increased swimming speed, probably via influencing cilia (Jékely et al., 2008). The effects could be reversed by washout (Figure 6—figure supplement 1; Randel et al., 2014).

Video 6

Download asset

Video 7

Download asset

Next we illuminated the cuvette constantly from both sides but with different left-right light intensities and measured the efficiency of phototaxis for a population of larvae using a phototaxis index. We found that phototactic efficiency increased with increasing contrast between the light intensity at each side of the cuvette, but was independent of total light intensity (Figure 6—figure supplement 2).

To test the contribution of the eyes and the two eyespots (independent ventral structures that develop in 1-day-old larvae, Figure 1B) to directional swimming, we laser-ablated the eyes or eyespots and subsequently exposed the larvae to directional illumination. Ablation of the two eyespots required for non-visual phototaxis in 1-day-old larvae (Jékely et al., 2008) did not abolish phototaxis responses in 3-day-old larvae. In contrast, ablation of all four eyes led to the loss of directional swimming (Figure 6F), demonstrating that only the eyes mediate this behavior.

Figure 6 with 2 supplements see all

Download asset Open asset

To directly investigate if the eyes are able to mediate body bending we performed selective eye illumination experiments on larvae held between a slide and a coverslip. Unilateral illumination of the eyes triggered reproducible bending of the body in 3-day-old larvae (Figure 6A; Video 8). When we used varying stimulus-light intensities for selective eye illumination we obtained graded bending responses with saturation kinetics (Figure 7D).

Video 8

Download asset

Figure 7

Download asset Open asset

Figure 7—source data 1

Source data of tail bending experiments in Figure 7C.

https://doi.org/10.7554/eLife.02730.041

Download elife-02730-fig7-data1-v2.txt

Figure 7—source data 2

Source data of tail bending experiments in Figure 7D.

https://doi.org/10.7554/eLife.02730.042

Download elife-02730-fig7-data2-v2.txt

Unilateral laser ablation of both eyes (Video 9) followed by uniform illumination also triggered strong and persistent body bending (Figure 6B,E; Video 10). Interestingly, in both experiments, the larvae either bent on or opposite to the illuminated side, with some larvae switching the response during repeated stimulation (Video 8 and data not shown). These results indicate that it is either the contra- or the ipsilateral motor pathway that dominates. Given that the Schnörkel interneurons synapse on both contra- and ipsilateral motorneurons (Figure 4—figure supplement 6), phototactic sign switching may take place at the level of the motorneurons.

Video 9

Download asset

Video 10

Download asset

To dissect the interplay of the four eyes during visual phototaxis, we ablated them in different combinations and tested phototactic ability. Larvae with unilateral ablation of both eyes were not phototactic and swam persistently in circles when illuminated (Figure 6C,D,F). The two anterior and two posterior eyes connect to distinct, though partly overlapping, circuits. To test if input to both the anterior and posterior eyes is necessary for directional swimming, we ablated either the anterior or the posterior eye pair. We also performed crosswise eye ablations. All of these ablated larvae showed phototactic responses similar to non-ablated controls (Figure 6F), hence the presence of at least one eye on both body sides is necessary and sufficient for directional turns.

To investigate if both dorsal and ventral longitudinal muscles are involved in body bending, we performed calcium-imaging experiments using ubiquitously expressed GCaMP6 (Chen et al., 2013). In immobilized larvae we repeatedly stimulated the eyes with 488 nm light on one body side. In agreement with the circuit diagram, upon selective eye stimulation we could observe corresponding calcium signals in both the ventral and dorsal longitudinal muscles on the opposing body side (Videos 11 and 12).

Video 11

Download asset

Video 12

Download asset

Overall, these experiments demonstrate that the eyes mediate visual turns by comparing simultaneous light inputs at each side of the body. This, together with the shading provided by the pigment cups (Video 2), allows the detection of the spatial distribution of light, without body movement. The visual contrast then leads to the graded contraction of the longitudinal muscles during turns. The modulation of ciliary beating, not investigated here, may also contribute to phototactic turns, as is known for younger larvae (Jékely et al., 2008). This sensorimotor structure is similar to visual phototaxis described in the lamprey (Ullén et al., 1997), but is very different from the non-visual phototactic responses in Drosophila larvae (Kane et al., 2013). Visual phototaxis is also fundamentally different from the helical phototaxis mediated by the eyespots during early Platynereis larval stages (1–2 days) (Jékely et al., 2008).

When we analyzed mutual synaptic connections in the eye circuit we found that by far the strongest mutual connections are formed between the crosswise pairs of IN¹ cells (up to 18 synapses one way; Figure 7A,B). These mutual, and thus likely inhibitory, connections are formed at the cell-body proximal segment of the IN¹ cells' axons, spatially segregated from the more distal sites of photoreceptor input (Figure 4—figure supplement 5). This reciprocal wiring may provide mutual inhibition and reinforce small differences in activation between the crosswise eyes, potentially representing a network motif for enhancing contrast-detection. Modularity analysis also revealed that the sub-networks of the crosswise eyes are more strongly connected than those of the pairwise eyes (Figure 4—figure supplement 3). This suggests that two reciprocally connected crosswise eyes are more efficient in visual information processing than a bilateral pair.

To test this, we performed various eye ablations and then subjected the larvae to selective eye illumination. We ablated a bilateral pair of eyes, thus eliminating inputs to one side of both crosswise mutual IN¹ motifs. We expect that these larvae would have a reduced bending response at the same level of contrast. We also performed crosswise eye ablations, leaving input to one of the crosswise IN¹ motifs intact. We then exposed larvae to a 488 nm background illumination and measured their tail bending following unilateral eye illumination with 488 nm stimulus light of varying intensity. Larvae with pairwise ablation of the anterior eyes showed significantly reduced bending relative to non-ablated controls with the average maximum tail displacement reduced to approximately half (Figure 7C,D). In contrast, larvae with crosswise eye ablation showed stronger bending that was not significantly different from non-ablated controls. Although both crosswise and pairwise ablated larvae have only two eyes and are phototactic (Figure 6F), these results show that pairwise ablated larvae have a markedly reduced motor response at the same level of contrast. We conclude that the strong mutual contacts between the crosswise IN¹ cells represent a circuit motif that enhances turning responses during phototaxis.

Platynereis larvae were obtained from an in-house breeding culture following an established protocol (Hauenschild and Fischer, 1969). Larvae were kept at 18°C for development. Behavioral experiments were performed at room temperature.

72 hr post fertilization Platynereis larvae, reared at 18°C, were fixed using a high-pressure freezer (HPM 010; BAL-TEC, Balzers, Liechtenstein) and transferred to liquid nitrogen. Frozen samples were cryosubstituted with 2% osmium tetroxide in acetone and 0.5% uranyl acetate in a cryosubstitution unit (EM AFS-2; Leica Microsystems GmbH, Wetzlar, Germany) over a regime of gradually rising temperatures. Samples were embedded in Epon. 40–50-nm serial sections were cut starting from the anterior end of a larva (Platynereis HT-9-3) on a Reichert Jung Ultracut E microtome. The sections were collected on single-slotted copper grids (NOTSCH-NUM 2 × 1 mm, Science Service, Munich, Germany) with Formvar support film, contrasted with uranyl acetate and Reynold's lead citrate, and carbon coated to stabilize the film. Image acquisition of serial sections was performed at a pixel size of 3.72 nm on a FEI TECNAI Spirit transmission electron microscope (FEI, Hillsboro, Oregon) equipped with an UltraScan 4000 4 × 4k digital camera using the image acquisition software Digital Micrograph (Gatan, Pleasanton, CA). Stitching and alignment were accomplished using TrakEM2 (Cardona et al., 2010). The first 1245 layers were cut at 50 nm and aligned using an elastic alignment algorithm (Saalfeld et al., 2012). We later found that the larva was slightly tilted, and we did not reach the muscles and the ciliary band on the right side of the animal. We therefore sectioned a further 445 sections at 40 nm, and aligned these using the rigid alignment algorithm in TrakEM2. All structures were segmented manually as area-lists or area-trees by an expert tracer (NR). Traces were exported into 3Dviewer, Imaris and Blender. Given the simple anatomy of the neurons and the large average diameter of axons tracing was in many cases unambiguous. Neurons with several short branches or with uncertain continuation were traced again by another expert tracer (LABC).

The minimal eye circuit of the eyes contains only postsynaptic neurons downstream of photoreceptors. Neurons presynaptic to the minimal eye circuit at any level (with possible modulatory functions) are not included here. Five sensory cells, which are both pre- and post-synaptic to the minimal eye circuit were also excluded. 12 postsynaptic targets of the MNs in the ventral nerve cord and five between the secondary optic neuropils as well as 32 postsynaptic targets of interneurons were also excluded because of low connectivity and the occurrence of only individual synapses outside the optic neuropils.

48 neurite fragments could not be traced completely. Most of these fragments are short (<5 µm) with 1–5 synapses. They occur in regions where neurons have small short branches to receive or provide synapses. These fragments likely belong to identified neurons. Nine of the non-traceable fragments were longer (up to 44 µm).

We classified neurons according to their morphology, position, projection pattern and connectivity. Photoreceptors were identified based on the presence of a rhabdom located in the eye pigment cup. Their dominant targets are the four primary interneuron (IN¹) cells. IN¹ cells are named based on their position (anterior-left etc.). IN¹ cells are defined as a separate group based on their unique connectivity patterns, including strong input from photoreceptors and strong reciprocal contacts between the crosswise cells. The second group of interneurons is classified as IN^ints, with projections intrinsic to the primary optic neuropil and weak connectivity to photoreceptors and IN¹ cells. A third group is formed by the IN^ton cells, distinguished by their unique projection pattern from the primary to the secondary optic neuropil. Schnörkel interneurons (IN^sn) form a distinct class with ventral cell bodies and curved axons (‘Schnörkel’ is German for ‘curlicue’) projecting to the ipsilateral secondary optic neuropil and synapsing on the motorneurons. The dorsal interneurons (IN^dc) have dorsally located cell bodies and axons that project to the secondary optic neuropil. The ventral interneurons (IN^vc) have ventrally located cell bodies and axons that project to the secondary optic neuropil. Motorneurons were recognized based on their innervation of dorsal and ventral longitudinal muscles as well as multi-ciliated cells (prototroch or metatroch). Motorneurons form two bilateral clusters. For some motorneurons we could not find motor synapses, however, we could classify these cells as motorneurons based on their cell body position and posteriorly projecting axons. Neurons of all types are labeled based on body side (left ^[l] or right ^[r]) and numbered (e.g., IN^tonl1).

Network analyses were done in Gephi 0.8.2. Modules were detected with an algorithm described in Blondel et al. (2008) with randomization on, using edge weights and a resolution of 1.2. Force-field-based clustering was performed using the Force Atlas 2 Plugin.

Neurons and their connections reconstructed in TrakEM2 were exported as 3D objects in a wavefront (.obj) format and imported into Blender. Cell bodies were approximated with ellipsoids and axons and dendrites with smooth curves. The approximations were performed either manually or automatically using Python scripting embedded in Blender. Muscles were kept in the original form, pigment cups and ciliated cells were approximated with simplified 3D shapes based on TrakEM2 reconstructions. The model was extended with the connectivity and synapse position information obtained from TrakEM2.

The virtual atlas (Randel et al., 2014) can be opened with Blender (http://www.blender.org/). Users new to Blender may consider watching a tutorial (http://www.blendtuts.com/2010/06/blender-25-interface.html). After opening the model file, two panels are visible in the Tool shelf: Hide/Show groups and Highlight subnetworks (Press ‘T’ if Tool shelf did not appear). If the panels do not appear, they can be loaded with the button Reload Trusted in the main menu panel (on the top). If menus still do not appear, the Panel_show_group.py and Panel_synapses.py python scripts need to be run by selecting the Scripting view in the main menu panel instead of Default. In the panel Hide/Show groups the cell types can be selected and hidden or displayed by pressing the Apply button. The panel Highlight subnetworks enables querying pre- and post-synaptic cells by selecting a cell body of interest (right click) and pressing the Postsynaptic or Presynaptic buttons. The complete up- or down-stream network of a cell can also be displayed by selecting a cell body and pressing the Pre.subnetwork or Post.subnetwork buttons. The Show connectors button visualizes the connectors between two cells after selecting the presynaptic cell first and then the postsynaptic cell. The Show out connectors button will reveal all outgoing synapses for a selected cell.

Laser ablation was performed on an Olympus FV1000 confocal microscope equipped with a SIM scanner (Olympus Corporation, Tokyo, Japan). Larvae were immobilized between a slide and a coverslip in NSW containing 100 mM MgCl₂. Larvae were imaged with an UPLSAPO 60X NA:1.20 water immersion objective using a 635-nm laser at 2–5% and transmission imaging with DIC optics. A 351-nm pulsed laser (Teem Photonics™, Grenoble, France) at 8–15% power was used, coupled via air to the SIM scanner for controlled ablation in a region of interest. 12% corresponds to a beam power of 168 µwatts as measured with a microscope slide power sensor (S170C; Thorlabs, Newton, NJ). During eye ablations we also imaged the eye pigment by reflection imaging of the 635-nm light using a PMT. Ablated larvae were placed into NSW for recovery (1–6 hr) before behavioral experiments.

Selective eye illumination was performed on an Olympus FV1000 confocal microscope equipped with a SIM scanner. Larvae were held by trapping them between a slide and a coverslip in natural seawater using several layers of tape as spacer. Larvae were imaged with an UPLSAPO 40X NA:0.90 air objective using a 635 nm laser at 1% and transmission imaging with DIC optics. The scanning speed was 2 μm/msec and we recorded 256 × 256 pixel images with a time increment of 0.43 s. For measuring the stimulus–response curve we used background illumination with the 488 nm laser at 10% power corresponding to 17.8 μwatts in the main scanner. For stimulations we used the 488 nm laser at 1–65% power via the SIM scanner, corresponding to a beam intensities of 0.53–20.9 μwatts. A circular region of interest of 25-pixel diameter (area of 490 pixels) covering the eyes was used for controlled illumination. The pixel dwell time of the beam was 10 μsec and the eyes were stimulated during 150 frames with the SIM scanner corresponding to a total exposure time of 735 msec during a 9.8 s trial. For the bending Video 8 the stimulus laser power was at 10%. The laser power was measured with a microscope slide power sensor (S170C; Thorlabs). To record the spatial and temporal extent of the stimulus the reflected 488-nm light was imaged with a PMT. To prevent the detection of the 488-nm stimulus light by the transmitted light detector, we placed a red long-pass filter in front of the detector.

Whole-mount in situ hybridization and image registration were performed as previously described (Asadulina et al., 2012). For probe generation we used expressed sequence tag clones or PCR-amplified fragments of already published Platynereis genes. The GeneBank accessions are: TyrH—JZ446954, TrpH–JZ446141, VGlut—JZ395359, hdc—JZ396646, VAChT—JZ402823, ChAT—JZ402100, gad—GU169427, dbh—KJ855061.

For registering gene expression patterns of 3-day-old larvae we generated a reference template (Randel et al., 2014) where we removed the bias present in the earlier reference, due to the use of a single larva as a starting template (Asadulina et al., 2012). In brief, we identified a median-size larva and registered it to every single image stack used for template generation (40 stacks), computed the average transformation and applied it to the median-size larva. The corrected larva was used as the starting point for the iteratively template generation (Asadulina et al., 2012). We registered minimum four individual larval whole-body scans for each gene to create an average expression pattern.

Larval phototaxis was assayed in a horizontal 6 × 15 mm rectangular glass cuvette with 3 mm high walls. The cuvette was illuminated uniformly through a light diffuser with white light from a 150-watt halogen cold light source (Schott KL 2500 LCD, Schott AG, Mainz, Germany). Larval behavior was recorded at 21 frames per second on a Zeiss Stemi 2000-CS stereomicroscope equipped with an AxioCam MRm camera (Carl Zeiss AG, Jena, Germany). Phototaxis efficiency was measured using a custom ImageJ macro that calculated larval density at the left and right side of the cuvette before and after the light stimulus (Randel et al., 2014).

Fertilized eggs were injected with capped and polyA-tailed GCaMP6 (Chen et al., 2013) RNA generated from a vector (pUC57-T7-RPP2-GCaMP6) containing the GCaMP6 ORF fused to a 169 base pair 5′ UTR from the Platynereis 60S acidic ribosomal protein P2. 3-day-old larvae were held as described above. Imaging was performed on an Olympus FV1000 confocal microscope equipped with a SIM scanner. Larvae were imaged with an UPLSAPO 60X NA:1.20 water immersion objective using a 635-nm laser at 10% and transmission imaging with DIC optics. The GCaMP6 signal was imaged with a 488-nm laser at 2% intensity. The eyes were stimulated with the 488-nm laser in a region of interest using the SIM scanner. GCaMP6 signals were recorded simultaneously.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank Dan Bumbarger and members of the Jékely group for comments on the manuscript, Ferdinand Kluge for help with tracing, Aurora Panzera for help with microscopy and microinjection, and Martin Gühmann for help with phototaxis scoring. The research leading to these results received funding from the European Research Council under the European Union's Seventh Framework Programme (FP7/2007-2013)/European Research Council Grant Agreement 260821.

Author contributions

GJ designed the research. NR prepared and sectioned the sample. RS imaged the sample. NR traced and assembled the connectome with help from AA and GJ. LABC and CV helped with the tracing and proofreading. NR and GJ performed laser ablations and behavioral experiments. EAW and MC performed gene expression analyses. AA visualized neuronal morphologies. GJ wrote the paper, with contributions from NR and AA.

Animal experimentation: This study only used invertebrate animals.

© 2014, Randel et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.