Cholinergic circuits integrate neighboring visual signals in a Drosophila motion detection pathway

Abstract

Detecting motion is a feature of all advanced visual systems [1], nowhere more so than in flying animals, such as insects [2,3]. In flies, an influential auto-correlation model for motion detection, the elementary motion detector circuit (EMD; [4,5]), compares visual signals from neighboring photoreceptors to derive information on motion direction and velocity. This information is fed by two types of interneuron, L1 and L2, in the first optic neuropile, or lamina, to downstream local motion detectors in columns of the second neuropile, the medulla. Despite receiving carefully matched photoreceptor inputs, L1 and L2 drive distinct, separable pathways responding preferentially to moving ‘on’- and ‘off’-edges, respectively [6,7]. Our serial EM identifies two types of transmedulla (Tm) target neurons, Tm1 and Tm2, that receive apparently matched synaptic inputs from L2. Tm2 neurons also receive inputs from two retinotopically posterior neighboring columns via L4, a third type of lamina neuron. Light microscopy reveals that the connections in these L2/L4/Tm2 circuits are highly determinate. Single-cell transcript profiling suggests that nicotinic acetylcholine receptors mediate transmission within the L2/L4/Tm2 circuits while L1 is apparently glutamatergic. We propose that Tm2 integrates sign-conserving inputs from neighboring columns to mediate the detection of front-to-back motion generated during forward motion.

Results and Discussion

Pairs of L1 and L2 neurons, one for each module or cartridge of lamina neuropile, extend axons down the cartridge axis to terminate in distal strata of the medulla ([8]; Figures 1A,B,I). In the lamina they receive input from the outer photoreceptor neurons R1-R6 of the ommatidia [9]. These are rod-like [10] and in Drosophila their terminals each bear ~50 tetrad synapses [11]. The R1-R6 input to L1 and L2 is closely matched because these two cells are invariable postsynaptic partners at the tetrads [12,13]. In vivo calcium imaging reveals that L1 and L2 both respond positively to light decrements and negatively to light increments [7]. However, behavioral and electrophysiological studies reveal that they nevertheless mediate two separable pathways responding preferentially to moving light- and dark-edges, respectively [6,7]. In addition, the L2 pathway is fine-tuned for front-to-back motion detection at low contrasts [14] and differentially modulates translational and rotational walking behaviors [15]. The circuits downstream of L1 and L2, likely sources of these differential output functions, are still unclear however.

Figure 1. Strata and terminals of the distal medulla.

(A-E) Expression of specific markers designates six outer strata M1-M6. Scale bar (in A): 10 μm.

(A) Relative to photoreceptor-specific anti-Chaoptin (MAb24B10, red), L1’s bistratified terminals (green, arrows) in M1 and M5 are revealed by L1-Gal4 driven GFP, and L2’s terminals in M2 by HA-tagged ORT (cyan, arrowhead) expression in the 21D enhancer trap pattern.

(B) Relative to the same L1 and L2 markers in A, anti-Connectin (α-Connectin, red) immunolabels stratum M3 immediately below the terminals of L2, leaving a space, presumably stratum M4, beneath M3 and the deeper terminals of L1 in M5.

(C) Medulla strata revealed by GFP expressed in the ort pattern (green) and anti-Discs Large (Dlg, cyan). L3 (double arrowhead) and the proximal L1 (arrow) terminals are discernable using GFP and anti-Dlg immunostaining.

(D, E) Single-cell flp-out clones of L4 (GFP, green). (D) A single L4 axon and terminals in the medulla M2 (arrow) and the presumptive M4 (double arrow) strata, between strong anti-Dlg staining at M3 and M5 strata. (E) Relative to L4 terminals, anti-Chaoptin immunolabels photoreceptor R7 and R8 axons.

(F-H) L4 axon and associated collaterals (green) in the proximal lamina. Scale bar (in F): 5 μm. (F) wire transformation of (G) with axon marked in cyan. Lamina cartridges are revealed by anti-Chaoptin staining (MAb24B10, red). (H) View of (F) as seen in a direction looking outwards, from a proximal location towards a distal one. The L4 axon (cyan) is located at the posterior side of its cognate lamina cartridge and extends collaterals (green) to its cognate, posteroventral and posterodorsal cartridge neighbors.

(I) Diagram of cell types, and the respective tiers of their medulla terminals and dendrites for lamina (R7, orange; R8, red; L1,L4, green; L2, cyan) and medulla (Tm1,Tm2) cells.

L1 and L2 provide independent pathways to the medulla

The axons of L1 and L2 traverse the external chiasma, their paired terminals innervating specific strata of each medulla column (Figures 1A,B; [16]), L1 in strata M1 and M5, and L2 with a single expansion in M2 [16,17]. Each horizontal sheet of axons twists en route to the distal medulla so as to invert the retinotopic map by which lamina cartridges project upon the corresponding array of medulla columns [18]. L1 and L2’s terminals align across the array of medulla columns (Figures 1A,B) emphasizing the respective strata, and even though the alignment between neighboring columns may not be perfect [16] the strata can be differentiated by immunolabeling with antibodies against synaptic proteins, such as Bruchpilot (Brp) and Discs large (Dlg; Figures 1C,D). Some strata are differentially labeled by antibodies against various adhesive receptors, such as Capricious, FasciclinIII and Connectin (Figure 1B; [19]). Thus, the definition of stratum M4 originally revealed from Golgi impregnation [17], although not clear in the single columns from our EM series [16] is suggested by the absence of immunolabel using anti-Connectin C1.427 (DSHB) (Figure 1B).

Immunohistochemical evidence reveals a glutamate phenotype in both L1 and L2 [20,21]. However, genetic reporter studies suggest that these two cells -- which define the two motion-sensing channels – might, despite their closely matched R1-R6 input in the lamina [11,13], actually employ different neurotransmitters [19]. To examine their neurotransmitter and receptor phenotypes, we developed techniques to profile relevant transcripts with single-cell resolution. We manually dissociated laminas and isolated single GFP-labeled L1 and L2 cell somata, and designed primers for RT-PCR reactions (see Experimental Procedures and Figure S1 for details). We found that L1 expressed vesicular glutamate transporter (VGlut) and L2 choline acetyltransferase (Cha), but not vice versa, implying that L1 and L2 are glutamatergic and cholinergic, respectively. We quantified VGlut and Cha transcript levels using real-time PCR analyses, and found that L1 neurons express 5980 ± 830 copies of VGlut transcript per cell while L2 neurons express 5730 ± 710 copies of Cha transcript (see Experimental Procedures, Tables S1,S2 and Figure S2 for details), indicating that both are highly abundant transcripts comparable to those for Rp49, a ribosomal protein (L1: 8210 ± 290 copies/cell; L2: 8529 ± 196 copies/cell). Our finding that L2 lacks VGlut transcript fails to confirm its expression of both glutamate [20,21] and a glutamate transporter [22]. Previous observations on the housefly Musca have revealed that L2 and L4 form a network of reciprocal connections in the lamina mediated by L4’s collaterals that invade two posterior cartridges [23,24]. This pattern also occurs in Drosophila (Figures 1F-H; [11,13]) and a critical role for L4 in motion detection is suggested by a recent behavioral study [25]. L1 and L4 lack direct synaptic connections [13,26]. To determine the nature of transmission at L2/L4 connections, we extended our profile analyses to all known cholinoceptor transcripts and to L4. We found that, like L2, L4 also expresses Cha but not VGlut. L2 and L4 share expression of the Dα7 and Dβ1 subunits of nicotinic cholinoceptors (nAcR) but each expresses a unique α-subunit (Dα3 in L2 and Dα4/5 in L4; Table S1). In contrast, L1 lacks detectable acetylcholine receptors. Insofar as L2 and L4 express nicotinic but not muscarinic cholinoceptors, we conclude that L2 and L4 probably provide fast reciprocal excitatory inputs to each other.

Recent studies have begun to probe the internal structure of the EMD [7,27,28]. Two very different computational models, including a weighted four-quadrant detector and a two-quadrant detector with an additional DC component, have been proposed [7,27]. Despite differences, both computational models, like the original Reichardt EMD, require communication between neighboring visual signals [4,5]. The topology of the L2/L4 pathway communicates between anteroposterior rows of columns, and so could provide the substrate for this communication. Missing however are the identities of L1 and L2’s target neurons in the medulla. To address these downstream circuits, we chose the circuits of L2 because of their importance for front-to-back motion sensing [14] in a flying insect, their cholinergic neurotransmitter phenotype, and because they engage another lamina cell type, L4.

L2’s synapses incorporate Tm1 and Tm2 medulla cell targets

L2 has a single subdivided terminal (Figure 2A), and three such completely reconstructed terminals each had between 88 and 98 presynaptic sites in stratum M2 [16]. Like R1-R6 tetrads in the lamina and the synapses of R7 and R8 in the medulla [16], each site was marked by a presynaptic ribbon, T-shaped in cross section (Figure 2B). All synapses were of the multiple-contact type, with at least three dendrites visiting each presynaptic site. Occasional presynaptic sites and T-bar ribbons seen en face, in the plane of the plasma membrane, had a four-fold symmetry suggesting that they had four postsynaptic elements, but because these were tiny we could usually trace at most only three. Although, because of their small size, most dendrites were difficult to trace back to a parent neuron, many made redundant contacts from the same neuron, providing assurance that we had identified individual dendrites accurately. In addition to its presynaptic sites, the L2 terminal in column 2 was also postsynaptic at 26 contacts from other medulla cells, mostly C2 and C3 [16].

Figure 2. L2’s downstream medulla pathways to Tm1 and Tm2.

(A,B) Electron micrographs of L2 terminal in stratum M2. (A) Traced profiles (blue) within column 2 reveal the subdivided composition of L2’s terminal, with the approximate border of a neighboring posterior column (dashed line). A: anterior; D: dorsal. (B) L2 synapses comprising a T-bar ribbon (arrowheads) each abutted by Tm1 and Tm2 postsynaptic dendrites, with at least one other postsynaptic element.

(C,D) Reconstructions of L2 and its two main target neurons, Tm1 and Tm2, shown from anterior (C) and dorsal (D) views. Single (Tm2) and double (Tm1) arrows show sites of input from L2 in M2; arrowheads show Tm2’s walking leg dendrites. Scale bars: 5 μm (for A,C); 0.5 μm (for B).

(E-G) L2 terminals overlap Tm2 dendrites in every column. (E) Shown relative to photoreceptor specific 24B10 immunolabeling of R7 and R8 (red), L2-Gal4 driven expression of HA-tagged ORT, HA immunolabeling (blue) overlaps Tm2-Gal4 driven expression of mCD8∷GFP (green) in stratum M2 of the medulla (Me). Terminals of Tm2 innervate stratum 2 (Lo2) of the lobula, Lo [17]. (F) Enlarged view of medulla from E, showing the layer of L2 terminals in M2 and the layer of Tm2 dendrite tips in M4/M5, relative to the deeper terminals of R7 in stratum M6. (G) Detail of Tm2 dendrites in M4 (arrowhead) and overlap between L2 terminal and Tm2 dendrites in M2 (arrow).

(H-J) Dendritic arbor, axon trajectory and terminal of Tm2. (H) Tm2-Gal4 driven expression of GFP in a single cell flp clone relative to CD2 immunolabeling in all Tm2 cells and 24B10 immunolabeling in R7 and R8. (I) Frontal section plane of the optic lobe showing single cell flp clones of Tm2 with axons in the chiasma between medulla (Me) and lobula (Lo) neuropiles and a single terminal in Lo2 (arrow). (J) Enlarged view of medulla arborization of Tm2 showing distal arbors in M2 (arrow) and two proximal walking leg dendrites (arrowheads) stretching down into M4/M5 (cf corresponding sites in C,D). Scale bars: 50 μm (in E, also for H); 10 μm (in F, also for G,J); 40 μm (J).

Through strata M1-M6 in column 2 we reconstructed two clear columnar neurons with particular morphological features. The axons of both were located in the mid-posterior position of the column cross-section, and ran alongside each other down the length of the distal column. Their identities were confirmed by less complete reconstructions from the two other columns. The first neuron closely resembled transmedulla cell Tm2 previously reported from Golgi impregnation [17], profiles of which have already been the subject of detailed light and electron microscopy comparisons [29]. Although we lacked information on its terminal in the next neuropile, the lobula, we could identify this neuron based especially on the defining feature of two or three branched descending dendrites that projected in a proximal direction like walking legs, from their origin at the axon in stratum M2, to a deeper level (Figures 2C,D,J). The second neuron was identified as Tm1, based on the presence of a densely branched arbor in stratum M2 that extended up into M1, and a coarser system of dendrites in stratum M3 (Figures 2C,D). Although its morphological identity was less clear, its partnership with Tm2 was suggested by the fact that both axons fasciculated together, while the frequent impregnation of Tm1 by the Golgi method [17] suggests that this cell is present in all columns. Tm1 is also one of several previously identified cell types with quantitatively analyzed patterns of co-arborization in different medulla strata that belongs to Pathway 2, tentatively identified as the L2 pathway [30].

From the 98 presynaptic sites of L2’s terminal in column 2, Tm1 and Tm2 were both identified postsynaptic elements at nearly half; at 12 they partnered each other at the same sites. The low probability (12 out of 98, or 0.12) of tracing two postsynaptic elements to the same synapse resulted partly from the combined low probabilities of tracing either Tm1 or Tm2 alone (~40 out of ~98, or about 0.4). This in turn is because the dendrites are generally very fine, and thus hard to follow through consecutive sections. We therefore propose that many more of L2’s synapses incorporate both Tm1 and Tm2. Given this low probability, we suggest that Tm1 and Tm2 are the chief targets of L2 at possibly most of its synapses, and that both are projection neurons that establish a dichotomous pathway to the lobula. L2 terminals overlapped Tm2 dendrites in every column (Figures 2E-G), supporting the primacy of this pathway. The terminals of Tm2 expanded in the two outermost strata of the lobula, Lo1-Lo2, where they formed a continuous array across the face of that neuropile (Figures 2H,I). Tm1, by contrast, occupies only the superficial stratum Lo1 [17]. To confirm that Tm1 is also present in all columns we must await a Gal4 driver specific for this cell type.

The L2 pathway incorporates binary subdivisions

Visual systems process features in a series of parallel pathways in distinct neuropile strata. This requires that light-evoked signals be split to provide input to the circuits of different parallel pathways. Light-evoked signals in R1-R6 are split at the first synapse into two different channels, L1 and L2. As we now see, L2 in turn splits these signals in the medulla into two further channels, Tm1 and Tm2. Other cells also receive the same signal at either synapse. Some, like amacrine cells of the lamina, may provide feedback to the R1-R6 input [13], while others may establish an independent pathway, as L3 does in the medulla [16,19]. The similarity between the matched inputs of L1 and L2 at the first synapse and, at the second, the likely matched inputs of Tm1 and Tm2 in the L2 pathway of the medulla is rather striking, and a candidate mechanism for establishing motion opponent pathways [4]. Further anatomical and functional studies would be needed to determine whether the Tm1 and Tm2 pathways indeed converge antagonistically at downstream targets. While previous reports (review: [31]) have identified a single type of medulla Tm neuron in the L2 pathway, none has clearly identified two, and thus none has recognized the binary split that occurs in this pathway in the medulla.

Tm2 also receives input from L4

L4’s axon moved to one side of its column soon after entering stratum M1 and deeper formed three small terminals that arose from short collaterals and partially overlapped the proximal terminals of L1 and L5 [16], in strata M4/5 (Figures 1D,E). The walking leg dendrites of Tm2 descended into the medulla in either a dorsal, ventral, or anterior direction in the column, with possibly some variation. In strata M4/M5 these contacted the terminals of L4 (Figures 1E, 3A,B) from which they received synaptic inputs, from the parent column and its two posterior neighbors (Figure 3C). Although synaptic contacts were observed in M4/5 the cells also overlapped eath other in M2 (Figures 3D-F). Thus Tm2 received identified synaptic input from two types of lamina neurons, but in different strata, L2 in stratum M2 and three L4 terminals in strata M4/M5; light microscopy also suggested L4 input in M2 as well. The L4 input helped us to identify a combined L2/L4 pathway in the medulla which resembles that seen in the lamina, where L4 provides input directly to L2 (Figures 1 F-H)[13,24,26]. In the medulla, such input is received at a common target neuron, Tm2, however. Given the inversion of the chiasma between lamina and medulla, L4’s synapses in both neuropiles provide input to the columns that neighbor their own in a retinotopically anteroposterior direction (Figure 4). The direction of L4’s spread in both neuropiles thus corresponds to an ommatidial sequence from the retina’s front to its back, and thus to the direction of the fly’s forward locomotion.

Figure 3. L4 provides input to Tm2.

(A,B) Corresponding EM reconstructions of L2 terminal (blue), L4 (red) and Tm2 (green) seen from anterior (A) and ventral (B) views. Three sets of L4 terminals (arrow [posteroventral], arrowhead [cognate] and double arrows [posterodorsal]) contact descending walking leg dendrites from Tm2. Scale bar: 5 μm.

(C) L4 synapses (arrowhead) onto a dendrite of Tm2. L4 terminal contains vesicle profiles with a mean diameter of 37.85 ±2.21 (SD) nm, predicted to contain acetylcholine. Other postsynaptic elements at this synapse are unclear. Scale bar: 0.5 μm.

(D-F) Tm2 dendrites (pseudo-colored in green) form apparent contacts (yellow, arrows) with L4 axons (pseudo-colored in red) from cognate (D,G), posteroventral,(E,H), and posterodorsal (F,I) columns at two neuropile levels, in M2 and M4/M5. For clarity, the neighboring L4s and Tm2s are not shown (see Figure S3 for details). Scale bar: 10 μm.

(G-I) Wire transformations of the corresponding panels D-F (above) shown relative to an array of 24B10-immunolabeled R7,R8 pairs at the center of each column at stratum M3.

(J-L) A Tm2 neuron (green) and two L4 axons from the posterior and posteroventral columns. For clarity, the neighboring processes have been blocked from view. (J) The blue channel has been removed from (K). The Tm2 dendrites form apparent contacts (arrows) with L4 from the posteroventral column but not with L4 from the posterior column (double arrow). (K) Wire representation of (J), viewed from the same direction. (L) View of (K) as seen in a direction looking outwards, from a proximal location towards a distal one.

(M) A schematic representation of Tm2 (green) and four neighboring L4s (red) from the cognate (C), posterior (P), posterodorsal (PD), and posteroventral (PV) columns. Photoreceptor axons (blue) mark the column array. Probabilities of overlap between the L4 terminals of four surrounding columns and the dendrites of Tm2 in the central column are shown as percentage values, first in M2 then in M4, of the total numbers of such pairs examined, shown in parentheses. Despite the proximity between Tm2 and the L4 terminals of all four columns, L4 in the posterior column almost never (only 1 in 29) overlapped Tm2, whereas the three others did so in at least 91% of pairs. All errors occurred in stratum M4.

Figure 4. Network of connections between L2 and L4 and their common medulla target Tm2.

(A) Plan view. Top: Connections between L2 and L4 in the lamina. L4 reciprocally contacts its own L2 and through collaterals the L2 of the two posterior cartridges [13,24,26], shown for a patch of seven cartridges with orientation (A,P: anterior, posterior; D: dorsal; eq: equator) as shown. The chiasma between lamina and medulla inverts this array. Bottom: In the medulla, Tm2 receives input from its own L2 and L4, and from the L4 of two posterior columns. See text for details.

(B) The same connections as in A, in side view. Note that the chiasma has been inverted, to emphasize that connections between L4 and L2, in the lamina, or between L4 and L2’s target in the medulla, Tm2, are both directed in a retinotopically posterior direction. Note the binary split from R1-R6 to L1,L2 in the lamina, and between L2 and Tm1,Tm2 in the medulla.

(C) Neurotransmitter and receptor profiles for L1, L2, L4, and Tm2. The L2/L4/Tm2 network is mediated by cholinergic connections (ACh) and each neuron expresses a shared pair of nicotinic receptor subunits (nAcRα7/β1) as well as type-specific nAcR receptor subunits. L1 is genotypically glutamatergic (Glu) and expresses no detectable acetylcholine receptors. In addition, L1 and L2 are connected via gap junction (resistor symbol).

To determine how reliably the L4-Tm2 connections observed in EM repeat in different columns, we examined approximately 150 pairs of L4/Tm2 neurons by light microscopy. To resolve Tm2’s dendritic arbor, which frequently overlaps neighboring Tm2 neurons, we labeled Tm2 neurons stochastically with a membrane-tethered RFP marker, using a pair of split LexA drivers [32] that express only in approximately 40% of Tm2 neurons (Figure S3A). We also randomly labeled L4 axons with a GFP marker using a combined apterous-Gal4 driver and lamina-specific flipase (Dac-Flip), with a UAS>stop>GFP reporter (Table S3). This double-stochastic labeling method allowed us to determine how often Tm2’s dendrites overlapped the four potential L4s, one in the cognate and three in neighboring columns (posterior, posterodorsal and posteroventral). Supporting our EM images in stratum M2, Tm2’s dendrites formed apparent contacts with all three of the L4 axons but, despite their being in physical proximity, not with L4 from the posterior column (100%, n= 34, 55, 40, for cognate, posterodorsal and posteroventral columns, respectively; 0%, n=29 for posterior columns). Similarly, in strata M4/M5, Tm2’s dendrites also formed apparent contacts with the same appropriate L4 neurons. However, these connections were not as robust as those in M2 and failures to overlap were observed. Such failures were not distributed equally for all partners, however, with 3-4% failures for dorsoventrally aligned pairs, and 9% for the cognate pair (Figure 3M). In addition, we observed a rare posteroanterior L4/Tm2 pair (3.4%; Figure 3M).

Given that both L2 and L4 express Cha and are thus genotypically qualified to synthesize acetylcholine and provide cholinergic input to Tm2, we next profiled the expression of acetylcholine receptors in Tm2. This proved more complex than for L2 and L4. In addition to Dα7 and Dβ1 nAcR shared with L2 and L4, Tm2 also expressed Dα1/2 and Dβ2 nAcR (Table S1; Figure 4C). The exclusive expression of nicotinic rather than muscarinic receptors (nAcR not mAcR) in Tm2 suggests that both L2 and L4 provide fast excitatory inputs to Tm2. We also found that Tm2 expressed Cha but not VGlut, indicating that, like L2 and L4, Tm2 is also genotypically cholinergic. In summary, these data predict that both synaptic connections in the L2/L4/Tm2 network are mediated by excitatory acetylcholine systems, and therefore sign-conserving.

Asymmetrical connections in the motion detection pathway

While either the L1 or L2 channel alone can mediate rudimentary motion detection, each also responds differentially in walking flies [15], and in head-yaw assays the L2 pathway is preferentially tuned to front-to-back motion [14]. Although the connections between L4 and L2 along the anteroposterior direction might account for this front-to-back preference [24], these connections are reciprocal [13,24,26] so that information also flows from posterior to anterior, while L2’s activity fails to reveal asymmetrical responses [7,33]. Between L2’s two targets, only Tm2 receives two additional L4 inputs from neighboring posterior columns; Tm1 does not. These L2/L4/Tm2 connections are highly determinate, underscoring a critical role in connecting neighboring L2 channels along the AP direction, in what is arguably the most important motion direction for flies since it occurs during forward flight. Interestingly, other flies have a Tm neuron closely resembling Drosophila’s Tm2 morphologically, for example Tm1 in the calliphorid Phaenicia [34]. This is proposed to receive L2 inputs, suggesting that an L2/L4/Tm2 network might be conserved in higher Diptera.

Tm2 could conceivably serve as half of the EMD’s multiplier stage, comparing the temporally delayed input from collateral L4s with the cognate signal from L2. However, electrophysiological investigations on calliphorid ‘Tm1’ neurons, which resemble morphologically Drosophila’s Tm2 [17], have yet to provide strong evidence for this role [35,36]. An alternative interpretation is that the L2/L4/Tm2 network serves instead as a prefilter in the preprocessing stage while Tm2’s output feeds into the multiplier stage [37]. The topology and sign-conserving nature of L4/Tm2 connections suggest the spatial summation of neighboring visual signals, which could increase light sensitivity at the expense of spatial acuity. It has been suggested that under low luminance conditions, neighboring visual signals are pooled prior to their interaction at the multiplier stage, while at higher luminance levels nearest-neighbor interactions dominate motion detection [38-40]. Alternatively, the L4/Tm2 connections could convert visual signals sampled from the hexagonal ommatidial array into an orthogonal coordinate upon which motion signals can be derived [41,42]. Differentiating between these possibilities must await future investigations that combine genetic and electrophysiological approaches.

Relating synaptic connections to circuit functions

Our study begins the difficult task of using three-dimensional reconstructions to identify the many different types of medulla cell, especially transmedulla cells with subtly different arbors and, in Drosophila, very fine caliber neurites [17]. We adopted a sparse, cell-by-cell reconstruction strategy to identify relevant circuit elements in the motion pathway. Our results have reinforced the hope that even in densely packed neuropiles, such as the medulla, it is technically feasible, though laborious, to identify connections and assign them to identified neurons, a process also open to validation against genetic reporters [29]. Annotating synaptic contacts must proceed in the face of axon terminals that are not exclusively presynaptic and dendrites that are not exclusively postsynaptic [16]. Indeed, in addition to the L2 inputs they receive in stratum M2, Tm1 and Tm2 dendrites were both also presynaptic, having T-bar ribbons in strata M2 and M3 (Figure S4) that provide inputs to neurons yet to be identified. Axon terminals, such as those of L2, can also be postsynaptic [16]. The power of EM reconstructions is that these alone reveal such important details.

Of course, a complete map of synaptic circuits does not easily relate synaptic connections to neuronal functions. EM reconstruction is moreover low-throughput, making it difficult to determine whether an identified connection occurs reliably in different regions and/or different animals. Here we confront that difficulty by combining a light-microscopic approach to assess the robustness of connections identified from EM. For example, despite their paucity, L4 to Tm2 contacts are highly determinate, suggesting that synapse number need not accurately predict functional significance in any simple synaptic democracy. Although connections revealed by EM provide information on the direction of transmission they fail to reveal its polarity or dynamics, which requires knowledge of receptor expression, as we now provide for the cholinergic genotype of the L2/L4/Tm2 pathways. Unlike promoter constructs, which even when available may suffer from positional effects, single-cell transcript profiling demonstrated in this study directly probes the expression of neurotransmitter receptor genes. Determining not only the anatomical connections, but also their robustness and neurotransmitter components will all be crucial both to understand how information is transformed within identified synaptic circuits, and to attain a major goal of functional connectomics.

Experimental Procedures

Fly stocks

Fly stocks were maintained on standard fruit fly medium at 23-25 °C. Fly stocks used in this study are listed in the Supplemental Table S3 and described in the Supplemental Experimental Procedures.

Single-cell transcript profiling

Single L1, L2, L4, and Tm2 neurons labeled with GFP were dissociated from adult lamina or medulla and collected using a custom-made capillary aspiration system. Total RNA from lysed single cells was reverse-transcribed to cDNA. PCR analyses were carried out to determine the presence of specific transcripts for VGlut, Cha, nicotinic and muscarinic acetylcholine receptors. Real-time PCR was carried out to quantify the transcript levels of VGlut and Cha. Rp49, which encodes a ribosomal protein, was used as an internal reference. Detailed procedures for cell dissociation, single-cell isolation, PCR primer design, single-cell PCR assay, and real-time PCR assay are provided in the Supplemental Experimental Procedures.

Immunohistochemistry

Immunohistochemistry, confocal imaging, image deconvolution, and 3D image rendering were performed as described previously [43]. Neuronal processes were traced using the FilamentTracer module in Imaris (Bitplane). Images shown in figures are maximal projection of multiple optical sections (0.2 μm). The following concentrations of primary antibodies were used: 24B10 (DSHB), 1:100 dilution; Connectin C1.427 (DSHB), 1:100 dilution; mouse anti-GFPMab (IgG2a, Invitrogen), 1:200 dilution; rat anti-CD2 (Serotec), rabbit anti-GFP, 1: 500 dilution (Torrey Pines Biolabs). The secondary antibodies including goat anti-rabbit, rat or mouse IgG coupled to Alexa 488, Alexa 568, or Alexa 647 (Invitrogen) were used at 1:400.

Electron microscopy

Specimens were prepared for electron microscopy, and reconstructions made from the same series of 60 nm sections, both as previously reported [16]. The procedures are summarized in the Supplemental Experimental Procedures.

Highlights.

• An L2/L4/Tm2 circuit integrates neighboring signals in a Drosophila visual motion detection pathway

• Serial-section EM reveals the synaptic circuits L2, L4 and Tm2 in the Drosophila visual system

• Light microscopy evaluates the reliability of the connections identified in EM

• Single-cell transcript profiling reveals the neurotransmitter and neuronal receptor genotypes of identified neurons