1. Neuroscience
Download icon

A resource for the Drosophila antennal lobe provided by the connectome of glomerulus VA1v

  1. Jane Anne Horne
  2. Carlie Langille
  3. Sari McLin
  4. Meagan Wiederman
  5. Zhiyuan Lu
  6. C Shan Xu
  7. Stephen M Plaza
  8. Louis K Scheffer
  9. Harald F Hess
  10. Ian A Meinertzhagen  Is a corresponding author
  1. Life Sciences Centre, Dalhousie University, Canada
  2. Janelia Research Campus, Howard Hughes Medical Institute, United States
Tools and Resources
  • Cited 0
  • Views 911
  • Annotations
Cite this article as: eLife 2018;7:e37550 doi: 10.7554/eLife.37550

Abstract

Using FIB-SEM we report the entire synaptic connectome of glomerulus VA1v of the right antennal lobe in Drosophila melanogaster. Within the glomerulus we densely reconstructed all neurons, including hitherto elusive local interneurons. The fruitless-positive, sexually dimorphic VA1v included >11,140 presynaptic sites with ~38,050 postsynaptic dendrites. These connected input olfactory receptor neurons (ORNs, 51 ipsilateral, 56 contralateral), output projection neurons (18 PNs), and local interneurons (56 of >150 previously reported LNs). ORNs are predominantly presynaptic and PNs predominantly postsynaptic; newly reported LN circuits are largely an equal mixture and confer extensive synaptic reciprocity, except the newly reported LN2V with input from ORNs and outputs mostly to monoglomerular PNs, however. PNs were more numerous than previously reported from genetic screens, suggesting that the latter failed to reach saturation. We report a matrix of 192 bodies each having >50 connections; these form 88% of the glomerulus’ pre/postsynaptic sites.

https://doi.org/10.7554/eLife.37550.001

Introduction

A striking convergence in advanced brains has endowed those structures of the brains in insects and mammals that analyse olfactory information with close similarities (Hildebrand and Shepherd, 1997). In both, segregated modules of neuropile called glomeruli each receive input from the axons of olfactory receptor neurons (ORNs), each of which in turn expresses predominantly a single olfactory receptor gene (Gao et al., 2000; Vosshall et al., 2000). ORNs in the adult fruit fly Drosophila melanogaster belong to one of four classes of sensilla that form a regular pattern on the third segment of the fly’s antenna and maxillary palp (Vosshall and Stocker, 2007). Compared with the ~5×106 ORNs (Kawagishi et al., 2014) that innervate the ~1800 glomeruli in the mouse olfactory bulb (Royet et al., 1988), the different regions of the olfactory system in Drosophila signal with fewer cell types using fewer odorant receptor molecules. Thus, only 1300 ORNs that express 62 odorant receptor proteins (Vosshall and Stocker, 2007) innervate a mere ~50 modular glomeruli (Grabe et al., 2015), the first relay of the insect olfactory system (Stocker, 1994; Laissue et al., 1999; Gao et al., 2000; Vosshall et al., 2000; Benton et al., 2009; Grabe et al., 2015).

Each of the ~50 glomeruli has been individually identified (Laissue et al., 1999; Benton et al., 2009; Silbering et al., 2011) and mapped, both in vitro, after dissection (Stocker et al., 1983; Laissue et al., 1999; Couto et al., 2005; Endo et al., 2007); Silbering et al., 2011) and in vivo (Grabe et al., 2015). Output from the glomeruli is made by the antennal lobe projection neurons (PNs), some of which relay olfactory information to higher-order olfactory centres, the mushroom body and lateral horn (Wong et al., 2002; Marin et al., 2002; Yasuyama et al., 2002; Yasuyama et al., 2003) via three main antennal lobe tracts (ALTs), medial, mediolateral and lateral (mALT, mlALT and lALT).

The cellular composition of the antennal lobe has also been extensively reported, both from early back-fill studies (Stocker et al., 1990) and more recent genetic reporter lines (e.g. Tanaka et al., 2012), and the numbers, types and patterns of innervation these receive from ORNs has likewise been identified from such lines (e.g. Couto et al., 2005). Among the PNs, Tanaka et al. (2012) have identified eleven classes, four in the medial, three in the mediolateral, and four in the lateral tracts. Amongst these, mPN1s project out of the antennal lobe along the mALT to the mushroom body and lateral horn (Tanaka et al., 2012). Relative to these, ORNs have been morphologically less well characterized but are recently catalogued for glomerulus DM6 (Tobin et al., 2017).

Little studied hitherto are the local interneurons (LNs), a major focus of our study. Tanaka has identified LNs of six types with cell bodies in different locations, which arborize in either single or multiple glomeruli and in the antennal lobe of either one or both sides (Table 1). Categorized by the glomeruli they innervate, Chou et al. (2010) identify at least ~100 different LNs for each side. Additional cell types identified include a single 5-HT neuron and several transverse (tALT) neurons (Tanaka et al., 2012).

Table 1
Nomenclatures for antennal lobe neurons
https://doi.org/10.7554/eLife.37550.002
*Tanaka et al. (2012)Gal four enhancer
-trap strain
Cell countVa1v cells,
this study
LN1NP12271817
LN2LNP24264024
LN2VNP242731
LN3NP13264.51
LN4NP84231
LN6NP158712
mPN1593
mlPN1NP52889.52
mlPN2233
mlPN332
t3PN12
t4PN15.5
lPN29.33
lPN413
AST11
AST21
AST311
MBDL11
VUMa51
5HT11§
Cells not reconciled with those reported by Tanaka et al. (2012)
LNV3
LN_LVExit3
LN_commissure4
Total10
Ipsilateral ORNs51
contralateral ORNs56
  1. *Column lists all those cells reported by Tanaka et al. (2012) to enter Va1v

    **Tanaka labelled only non-specifically for several cell classes including those in Va1v

  2. ***Tanaka did not find in Va1v at all

    ****§Not included in matrix

Despite detailed knowledge of their cellular composition, the synaptic networks of the fly’s antennal lobe have been incompletely documented at the requisite level from electron microscopy (EM). The reasons for this are obvious: the difficulty of tracing the tiny neurites of Drosophila neurons, the labour required to do so, and the practical problems of targeting specific glomeruli for EM. Fluorescent pre- and postsynaptic markers provide proxies for claimed synaptic contacts (Mosca and Luo, 2014), but lack independent validation from EM. Two recent studies (Rybak et al., 2016; Tobin et al., 2017) report the synaptic networks of identified glomeruli. The first, a study of three glomeruli (DL5, DM2, and VA7: Laissue et al., 1999), has provided a significant map of synaptic connections and their numerical proportions, while the second, on glomerulus DM6 (Tobin et al., 2017), established many features of the PN. Neither study gave a comprehensive account of the varied synaptic inputs from and between LNs, however. Including the synaptic network of all LNs and PNs to generate an actual map of the complete synaptic network, or connectome (Lichtman and Sanes, 2008), for each glomerulus is thus a final step towards adopting functional connectomic approaches (Venken et al., 2011; Meinertzhagen and Lee, 2012) to the analysis of antennal lobe function.

Results

We report the connectome of a single glomerulus VA1v of the right side of a female Drosophila antennal lobe (Figure 1), using dense reconstruction of an isotropic 8 nm voxel image stack (Figure 2), see Materials and methods, obtained by focused ion beam (FIB) milling scanning electron microscopy, FIB-SEM (Xu et al., 2017). VA1v is a sexually dimorphic fruitless-positive glomerulus responsive to fly odour (Sakurai et al., 2013), that signals the sex pheromones cis-vaccenyl acetate and methyl laurate (Kurtovic et al., 2007; Dweck et al., 2013), and is significantly larger in male flies than in females (Kondoh et al., 2003; Stockinger et al., 2005).

(A) The right antennal lobe of a female Drosophila melanogater wholemounted and immunolabelled with nc82 to detect Bruchpilot at synaptic sites, revealing the neuropiles of the brain.

These included the glomeruli of the antennal lobe (AL). Va1v of the right antennal lobe (blue) is enclosed by a box in A, identified in synapse cloud images. Scale bar 100 µm. (B) Image of a single plane with synapses identified as clouds by synapse detection on a corresponding FIB-SEM image stack, enclosing most of the right antennal lobe within which different glomeruli are identified and colour coded. Glomerular borders are visible in this single image from local rarefactions in the density of recognized synapse puncta. (C) A cloud of synaptic puncta in two image planes parallel and corresponding to the one in (B), with different glomeruli – including VA1v (left panel) – identified.

https://doi.org/10.7554/eLife.37550.003
Single grey-scale image of the densely reconstructed glomerulus VA1v with colour-coded profiles of different cells (key).

Confirming the denseness of reconstruction, note that essentially all profiles are labelled, the few remaining dark grey belonging to orphan elements. Most profiles fall in the range of 0-5-1.5µm in diameter, with larger diameter axons giving rise to tiny dendritic neurites. Scale bar: 5 µm.

https://doi.org/10.7554/eLife.37550.004

Synapses

Each synapse comprised a T-bar ribbon at the presynaptic site opposite membrane densities at the postsynaptic dendrites (Figure 3A). Most commonly these numbered three per synapse, with an average of 3.4, most synapses thus forming triads or tetrads (Figure 2—source data 1), like those of the three glomeruli reported by Rybak et al. (2016). The range was from 1:1 to 1:9. The numbers of postsynaptic dendrites differed slightly for each presynaptic cell, ipsilateral ORN cells having 3.63 dendrites per synapse, LN1 and LN2L 3.17 and 3.06 respectively, and mPN1 had 3.54 postsynaptic dendrites (Figure 2—source data 1).

Representative synaptic profiles seen in FIB-SEM at 8 nm resolution.

(A) Five electron-dense profiles of presynaptic sites, four in ORNs and one (open arrowhead) in an LN, reveal a range of shapes, from a clear T-shape, in canonical cross section (1), to cruciform (2), in an en face view of the pedestal, with a range of other profiles that cut the organelle in different planes. Unifying their common identity, all have the same electron density that is clearly visible after FIB-SEM imaging. Postsynaptic densities are not well resolved. The profiles exhibit a wide range of shapes because the neurites contributing them are not aligned, as are the columns of the medulla (Takemura et al., 2015) and mushroom body output lobes (Takemura et al., 2017), and other neuropiles analysed with this imaging method. (B–D) Single neurite profile with ~84 nm diameter dense-core vesicles (dcv) viewed in three orthogonal planes, revealing each dcv as approximately circular in all three planes, and thus a sphere. (E) Same neurite reconstructed to show a varicosity with a single presynaptic site (yellow). Scale bars: 500 nm.

https://doi.org/10.7554/eLife.37550.006

The size and structural features of each synapse resembled those found between neurons in other Drosophila neuropiles, such as the optic lobe lamina and medulla (Meinertzhagen and O'Neil, 1991; Takemura et al., 2008). At each presynaptic site a T-bar was found with a clear pedestal; the electron density of the ribbon platform surmounting it was weak after the preparation methods adopted for FIB-SEM. Thus unlike transmission EM, TEM (Rybak et al., 2016), ribbons often lacked canonical T-bar profiles when seen in FIB-SEM images. This chiefly resulted because in our FIB-SEM images the grey level of the synapses was about twice that of the membranes. We therefore adjusted the electron density of images to increase membrane contrast, because this proved advantageous to enhance membrane continuity more reliably during later proof-reading steps (see Materials and methods), but rendered the platform of the T-bar ribbon often less distinct in our FIB-SEM images than when seen in TEM.

Occasional synapses were bidirectional, with a T-bar ribbon on either side of the synaptic cleft. Synapse sizes also varied more widely than for tetrad synapses in the lamina (Fröhlich, 1985; Meinertzhagen and O'Neil, 1991). Occasional autapses were seen amongst PN axons, at which the neuron was presynaptic to itself. These typically numbered about 1% of all the PN axon’s synapses, about the same relative number as reported for medulla neurons (Takemura et al., 2015). Synaptic vesicles were not well resolved after FIB-SEM imaging at 8 nm. In addition to synapses with what appear to be small, clear vesicles, dark vesicles having a dense core of between 50 and 100 nm in diameter appeared in the profiles of most LNs. An additional reconstruction which closely resembled the cell called 5-HT1 (Tanaka et al., 2012), and was neither an ORN, PN nor LN, was full of large dark vesicles (84 ± 12 nm in diameter) and had few synapses with what appear to be small clear vesicles (Figure 3B–D) not obviously visible after FIB-SEM imaging. Its axon exited via the mediolateral tract (mlT). Spinules (Gruber et al., 2018), interesting membrane invaginations visible in most cell profiles, fall in the vicinity of synapses, but were neither invariably close nor obviously related to synapses; they may represent regions of temporary membrane disequilibrium at sites of vesicle exo- or endocytosis.

Identified neurons from dense reconstruction in VA1v

We reconstructed 192 neuron bodies (Figure 4), each of which had >50 contacts in glomerulus VA1v. These 192 ( Figure 4—source data 1, Figure 4—source data 2Figure 4—source data 3, Figure 4—source data 4) constituted most of the neurons of that glomerulus, which was thus densely reconstructed (Figure 2). Occasional orphan elements (Figure 2; Figure 4—source data 4) may have constituted an additional cell but were disconnected and had membranes too obscure to trace with greater certainty. Otherwise dense reconstruction provided reassurance that no cell could hide undetected within the neuropile (Figure 2). A total of 87.9% of VA1v synaptic connections have been assigned to these neurons, the remainder being those of orphan neurites or neurons with fewer than 50 contacts. Overall, glomerulus VA1v had a neuropile volume of 4,858 µm3 and the 192 neuron bodies constituted 87% of the volume. Neurons were additionally traced sparsely beyond the borders of the glomerulus where possible, in order to connect to cell bodies, and project via the three ALT tracts, antennal nerve, and commissural tract. The total neurite length traced was 15.8 cm.

Reconstructions of the three main types of antennal lobe neurons in glomerulus VA1v, for comparison with published single-cell reporter expression lines (Tanaka et al., 2012).

Dorsal (A–F) and frontal (G–I) views. (A, B) Composite of 51 ipsilateral (A) and 56 contralateral (B) ORNs. Individual cells shown in Figure 4—source data 4 (see also library of cell types in Tobin et al., 2017, their Figure 1—figure supplement 1). Arrow in (A) represents neuron path to antennal nerve. (C) A single anomalous LN (possibly LN2V in Figure 2B in Tanaka et al., 2012). Asterisk shows path to cell body beyond region of segmentation. (D) The three mPN1 (dark purple) are monoglomerular in Va1v with axons that exit via the medial tract mALT (cf Figure 3A in Tanaka et al., 2012). Soma locations marked with an asterisk. Two monoglomerular mlPN1 cells (intermediate purple) with axons that exit via the mediolateral tract mlALT (cf Figure 4A in Tanaka et al., 2012). Three lPN4 (light purple) that arborize in Va1v and its sister glomerulus Va1d, have axons that exit via the lateral tract lALT (cf Figure 6D2 in Tanaka et al., 2012). (E) Additional multiglomerular PN types. For all PNs see Figure 4—source data 2. (G) 17 LN1 cells with somata in the dorsolateral cortex (cf Figure 2A in Tanaka et al., 2012). (H) 24 LN2L cells (cf Figure 2B in Tanaka et al., 2012). (F, I) Composite images of other cells identified as LNs of types LN3-6 and six multiglomerular local interneurons (LNs) differ in the position of their somata and the incoming tract of their axon; these are not illustrated by Tanaka et al., 2012. For a library of densely reconstructed cell types that arborize in VA1v, see , Figure 4—source data 1, Figure 4—source data 2Figure 4—source data 3Figure 4—source data 4 ); most are incompletely traced to other glomeruli.

https://doi.org/10.7554/eLife.37550.007

We reconstructed the neurons and compared their reconstructed shapes within the glomerulus, and their wider axon trajectories within the antennal lobe, with those revealed by comprehensive mapping using Gal4 enhancer-trap strains (Tanaka et al., 2012). This enabled us to identify them cell-by-cell as one of the three main classes of neurons: input ORNs, output PNs, and local interneuron LNs (Figure 4). The provisional identification of cells relative to published trajectories of cells in reporter lines is listed in Table 1.

A total of 51 ORNs originated in the ipsilateral antennal nerve with a further 56 that entered in the commissure from the contralateral lobe.

We have identified an unanticipated large number of 18 PNs that lay within the region containing glomerulus VA1v. They fall into the three groups that exit the three main tracts of the antennal lobe (tract nomenclature from Ito et al., 2014): (a) Within the group that exits the medial tract we found three tightly interwoven monoglomerular PNs, the mPN1s (Figure 4; Figure 4—source data 2). (In a previous study Yu et al., 2010 identified five mPN cells in VA1lm, which we interpret to reveal the PNs of two glomeruli, one with 3 mPN1 and its neighbour with 2) We also identified three other mPN1s with a few projections within the ROI but these were in two other glomeruli. These are labeled mPN1 external (see Figure 4—source data 2i–n). b) In the group that exits the mediolateral tract there was a total of eight neurons: (i) two mlPN1s which are monoglomerular; (ii) three mlPN2s which are multiglomerular but only extending neurites to select glomeruli (one having a less widespread projection pattern in VA1v) and one other, mlPN2, which is mainly external to Va1v; and (iii) two mlPN3s which are multiglomerular and have a much sparser projection. For all of these, see Figure 4—source data 2o–v. c) In the group that exits the lateral tract we identified eight neurons: (i) three lPN4s that project mainly to VA1v and its sister glomerulus VA1d; (ii) four lPN2s which are multiglomerular and bilateral; and (iii) one cell called lPN2_Comm which resembles lPNS but does not exit in the lateral tract, instead entering from the commissure (see Figure 4—source data 2b–h).

One additional cell, AL-AST3 reported by Tanaka et al., 2012, with few pre- and postsynaptic contacts, had an axon that exits at the posterior of the antennal lobe and a cell body close to the antennal nerve.

We also identified no fewer than 56 LNs that innervated our glomerulus, of the >150 previously reported for the entire antennal lobe (Chou et al., 2010). As previously reported (Chou et al., 2010; Das et al., 2011; Liu and Wilson, 2013), these are morphologically diverse, possibly reflecting the three developmental modes of their origin, as residual larval LNs, as adult-specific LNs emerging before associated sensory and projection neurons, and as LNs that emerge after synaptic connections are established (Liou et al., 2018). As a whole, we find they have a wide range in the numbers of their synaptic contacts. LNs were traced comprehensively in VA1v and also via their axon to their soma; thus although most are panglomerular (Chou et al., 2010), our reconstructed LNs are therefore partial. Their morphologies are illustrated fully in Figure 4—source data 3. Within the different LN groups, some have many synaptic contacts within VA1v. We could assign most to the six morphological classes identified by Tanaka et al., 2012, as follows: (a) The most noteworthy LN is a single largely uniglomerular cell that we identified as LN2V. Although superficially resembling a PN, it does not project to any of the three tracts and has a small neurite which extends towards the ventral cortex of cell bodies but which extends beyond the field of view. This is remarkable in three ways that distinguish it from multiglomerular LNs: in arborizing predominantly within a single glomerulus, in being highly directed (with predominant output to PNs), and in the numbers of its synaptic contacts in glomerulus VA1v. Multiglomerular LNs (Tanaka et al., 2012), by contrast, all have fewer synaptic contacts and lack clear directionality.From their morphology and connectivity, we could recognise at least five other types of LNs: (b) Two major groups previously reported are LN1 and LN2L, which share the same soma locations and enter by the same tract. We distinguish them by whether they made contact with ORNs (LN2L: having 24 cells) or not (LN1: having 17 cells) (Figure 5—figure supplement 1). LN3 (one cell) and LN4 (one cell) both have somata ventrolateral to the antennal lobe and project to the commissure, LN4 arborizing in only a few glomeruli, and LN6 (two cells each) have few connections in VA1v. In addition a further four bilateral cells send neurites into VA1v via the commissure; their somata are located on the contralateral side. These types are hard to differentiate further without knowing their contralateral arbor. (c) We found in addition LNV (three cells) and LN_LV (three cells), were not easily reconciled with the report of Tanaka et al., 2012. The cells called LNV enter the antennal lobe close to the lateral tract and branch profusely; we could not identify these further because we could not find their cell bodies. Three cells we call LN_LV have axons that exit close to the lateral tract and originate from cell bodies close to the antennal nerve; one has an identified axon to the commissure (Figure 4—source data 3) and could therefore be LN3. We could not find the tract to the commissure to know whether this cell might be bilateral. d) Four cells with axons that arrive or exit in the commissure are LN commissure cells. These could also be PNs.

Most cells formed both pre- and postsynaptic contacts (Figure 6). However, ORNs were predominantly presynaptic, ipsilateral ORNs carrying more synapses than contralateral ORNs, and most of the reconstructed neurites from VA1v PNs were predominantly postsynaptic, as also recently reported from ssEM for other glomeruli (Rybak et al., 2016; Tobin et al., 2017)(Figures 5 and 6). LNs had about equal numbers of both pre- and postsynaptic contacts and, despite some details (Rybak et al., 2016), these have not previously been reported comprehensively (Figure 2—source data 1). A log/log plot of post- and presynaptic contacts for each cell shows that the predominantly presynaptic ORNs cluster beneath the corresponding values for LNs; having more outputs than inputs (Figure 6).

Figure 5 with 1 supplement see all
Connectivity matrix of VA1v cell types.

Register of cells with presynaptic sites (x axis, ordinate) plotted against the same cells having postsynaptic sites and colour-coded intercepts denoting the number of synaptic contacts between each pair of cell types (key), and thus the anatomical strength of their connection. Cells are arranged from the top left origin as, first, outputs (PNs), then interneurons (LNs), and finally inputs (ORNs) and further organized within those groups by the particular cell. Among the total of 192 cells, dense pathways occupy few intercepts, mostly concentrated in ORN to PN, and PN to LN intercepts. Only cells with more than 50 pre- or postsynaptic contacts are included. For the complete matrix, as a spreadsheet, see Figure 5—source data 1.

https://doi.org/10.7554/eLife.37550.012
The numbers of pre- and postsynaptic sites for each cell, log-log plot.

Only cells with >50 contacts are shown. Cells are colour-coded for their type (ORN, LN, PN: key); ORNs are either ipsi or contralateral, PNs are labeled as well as anomalous LNs, which do not cluster with other LNs. Cells can be identified from their pre:post synapse ratio. Almost all LNs have equal numbers of pre and postsynaptic contacts; ORNs have more presynaptic than postsynaptic contacts, with a ratio of 3.9 ± 0.6 for ipsilateral and 5.3 ± 1.5 for contralateral (Figure 2—source data 1). PNs are predominately postsynaptic but have more variable pre/post ratios as well as number of contacts than the other two cell types; mPN1s have most presynaptic sites with an average ratio of 0.4 ± 0.03 and mlPN2s are almost completely postsynaptic with a ratio of 0.06 ± 0.11; some PNs have >1000 contacts. Unknown cells, some orphans (not traced to identified neurons), cluster with local neurons, their suspected source. A few cells are anomalous, some (especially LNs) falling outside their cluster are possibly mis-assigned.

https://doi.org/10.7554/eLife.37550.015

ORN, PN and LN cells exhibit a wide range of synaptic loads, most obvious for the PNs which cluster less tightly, and three of which, all mlPNs, lack all presynaptic sites (Figure 6). The three tightly clustered mPN1s by contrast have both pre- and postsynaptic sites, as also reported from ssEM for other glomeruli (Rybak et al., 2016; Tobin et al., 2017). Means for the numbers of synapses are given for each cell and cell type in Figure 2—source data 1.

We can set these features in synaptic organisation in the context of known responses of glomeruli to odours and odour mixtures: Relative to ORN and PN cells, LNs exhibit far greater synaptic diversity (Seki et al., 2010). The synaptic summary we present indicates that LNs are anatomically qualified to relay information within and between glomeruli, and given that each glomerulus processes information about only a single olfactant (e.g. Silbering and Galizia, 2007), are therefore qualified as a substrate to make comparisons between different olfactants. Comparison between ipsi- and contralateral inputs to PNs may provide a pathway for Drosophila to signal the direction of olfactants while PN responses to olfactant mixtures provide evidence for interglomerular inhibition (Silbering and Galizia, 2007).

The synaptic matrix of VA1v

In summary, we first present the numerical features of cells contributing the VA1v connectome. A total of 51 ORNs originated in the ipsilateral antennal nerve with a further 56 that entered in the commissure from the contralateral lobe. Together these innervated 18 PNs, and some of the 56 LNs we found. The latter have mostly not previously been reported at EM level.

We next report the synaptic matrix for 192 neurons having at least 50 connections to other synaptic partners in VA1v. Details of the synapses themselves are reported above and illustrated in Figure 3. After manually identifying all synapses and dense proof-reading we found 11,144 presynaptic T-bars opposite which sat 37,843 postsynaptic dendrites, most as triad synapses, with roughly 2.29 synapses per µm3. Of these, 93% were resolved with both pre- and postsynaptic partners identified, including some connecting in neighbouring glomeruli.

The synaptic organization has not previously been reported for specific PN types, while the anatomical connections of LNs have not so far been reported at all. Synaptic engagements were specific for each cell type. The complete inventory of chemical synaptic contacts is given as a matrix of their pre- and postsynaptic connections (Figure 5—figure supplement 1). Each intercept represents the number of contacts between a single pre- and postsynaptic neuron class, even though individual synapses incorporate several (most frequently three) postsynaptic elements, so that multiple such intercepts are in fact coordinately linked. Indicating the specificity of connections, most intercepts are blank. The matrix in Figure 5 reports VA1v neurons as classes, and is expanded in spreadsheets to reveal the number of contacts between individual pre- and postsynaptic neurons (Figure 5—figure supplement 1). Even though the matrix reveals only the numerical strength of synaptic partnerships and not their physiological strength, it is clear that particular pathways preponderate. These include ORN input to PNs, and selected types of PNs onto some LNs. LN2V, a particular LN considered above, stands out from other LNs in also being highly synaptic, receiving heavy input from ORNs and output to PNs. The matrix we present will serve to interpret future electro- or optophysiological recordings from genetically identified neurons of the glomerulus in terms of both ORN inputs and LN outputs, and especially the connections between LN partners.

The relative frequency of such pathway strengths shows that synaptic load is distributed somewhat unevenly, and that most connections have few synapses, in a distribution that is heavy tailed (Klebanov, 2003) (Figure 5—figure supplement 1). About 37% of connected cell pairs have three or more connections (see Materials and methods). Some assurance for the accuracy of each intercept in the matrix was also provided by the large number of some partnerships, especially between ORNs and PNs and LN2V, and others, 778 with >10 synapses, relative to all others. The matrix exhibited remarkable selectivity. Thus a total of 36,864 contacts was possible within the matrix of 192 cells, but of these only 8398 pairs (23%) actually made contact, and only 3073 (8%) did so with >2 contacts.

Based on the matrix numbers, the ratio of pre- to post- connections for ORN’s was 3.94 for ipsi and 5.29 for contra, while for all PN’s it was 1.3, although this includes one outlier lPN2 which had a ratio of 19.5; excluding the latter gave a ratio for PNs of 0.23, while LN1 and LN2Ls had nearly equal numbers of each, with a ratio of 1.09. Ratios for individual cells and averages for groups are given in Figure 2—source data 1.

Connectivity features of the network

If we summarize the numerical features of the network (Figures 5 and 7) we see that the output from ORNs numerically dominates the connectome, with 59.8% of all pre- connections, of which 39.5% go to the PN’s, 14.5% to LNs, and 5.9% to other ORNs (Figure 7A). Much output from the local interneurons is thought to be inhibitory (Wilson and Laurent, 2005) and anatomically this output accounts for a further 27.8% of synapses, distributed evenly amongst the three classes of target cells: 10.4% to PNs, 7.8% to ORNs, and 9.6% to other LNs. PN output is rare within the glomerulus, with 6.4% to LNs, 4.1% to PNs and only 1.8% to ORNs, but of course most output is predicted to lie among the target circuits in the mushroom body calyx (Yasuyama et al., 2002); Leiss et al., 2009; Butcher et al., 2012) and higher olfactory centres. Many cells are multiglomerular, and generate a complex interglomerular network inferred mostly from light microscopy and thus not at synapse level. Thus inhibitory LN output to other LNs is predicted to generate disinhibition, and is further expected to constitute part of a widespread weak interglomerular inhibitory network. Glomeruli innervated by OSNs expressing a narrowly tuned olfactory receptor may have more PNs and fewer LNs, suggesting that these glomeruli perform less lateral processing.

Network diagram for all cells and cell types within glomerulus VA1v having >50 pre/post contacts and reported in the overall matrix (Figure 5).

Edges show connections by classes of cells, not those of their individual constituent neurons. The width of each pathway arrow shows the number of anatomical synaptic contacts (key). (A) Aggregate network for three major classes of VA1v cells. The direction of pre- to postsynaptic contact is shown by the arrow and colour of the line connection between synaptic partner cells. ORN inputs to PNs are numerically predominant. Most LNs are predicted to be inhibitory, and provide pathways that integrate both ORN and PN activity; LNs also form their own network. Numbers indicate cells per class. Recurrent loops are feedback between different neurons of the same class. Large arrowhead indicates PN output to higher olfactory centres in the mushroom body and lateral horn. (B) Network diagram, as in A, plotted for subgroups of their type (see Figure 4 and Figure 4—Source data 4). Node width represents cell number per group, colour-coded as in Figure 6. Vertical: pre-/post- axis, note that this is interrupted at 2.0; horizontal: number of connections per cell plotted on a log scale.

https://doi.org/10.7554/eLife.37550.016

Looking at the network’s features more specifically, we see that the largest input extends from ORNs, those from ipsilateral ORNs constituting 70% of the total ORN input (Figure 7B). Some ORN synapses are made between an ORN and its neighbouring olfactory receptor neurons. Most of the ORN inputs are made to the three mPN1 neurons (25%), however, although input also falls upon other PNs. A total of 76% of ipsilateral ORN input to PNs goes to the three classes of monoglomerular PNs, mPN1 and mlPN1 and lPN4. The latter projects mainly to VA1v but also its sister VA1d glomerulus (Tanaka et al., 2012). Of the 76%, 43% goes to mPN1. The only other PN class to receive major input is the multiglomerular mlPN2 which receives 18.5% of the total ipsilateral ORN input. Thus, very little of the ipsilateral ORN input goes to the remaining PNs. (ii) Cell LN2V is anomalous. Its pattern of connections resembles that of a PN more than that of the other LNs, receiving its input from ipsilateral ORNs and providing most of its output to PNs. This pattern does not resemble in all details any of the cells previously reported by Tanaka et al., 2012. It appears to be predominantly monoglomerular in VA1v but has sparse projections to other glomeruli. (iii)Unusual among LNs, LN6 is almost entirely presynaptic, with weak output to other LNs and a single strong output to LNv not identified by Tanaka et al., 2012. Possibly these two neurons are not LNs at all, but the arbors of receptor neurons from the maxillary palps. Their axons enter the glomerulus from the posterior antennal lobe, compatible with an entry from the antenno-suboesophageal tract, which was not clearly identifiable in our reconstructions; against this possibility, however, ORNs from the maxillary palps are not reported to enter VA1v (Couto et al., 2005). (iv) Neurons that project to many glomeruli, as identified from reporters (Couto et al., 2005; Tanaka et al., 2012), form few connections in VA1v. (v) Six orphans that lack a soma or identified projections outside the glomerulus are included in the matrix. They have very fine neurites we were unable to trace completely, for which reason we have recorded them as orphans, and they are most likely to be the disconnected parts of recorded LNs.

Synaptic reciprocity

Some connections between neurons are made reciprocally, and many LN cell pairs in particular are so connected, but weakly so (Figure 8). The numerical ratios of synapses formed between different cell types differ. Of 1540 possible pairs of neurons having >2 contacts, only 442 (29%) were reciprocal, 380 (25%) of which met a criterion for strong reciprocity in which there were at least four times as many synapses in one direction as the other (a pre:post ratio of either 4 or 0.25: Figure 8).

Matrix of the 192 cells in Figure 5 highlighting those pairs that are strongly reciprocal.

Pairs have connections of at least three synapses in the pre:post and post:pre directions so having at least six pre- and postsynaptic contacts in total. Each intercept shows the ratio for the pair of partner neurons of its pre- to post- contacts. Ratios of between 0.25 and 4 are highlighted in a graded fashion (see scale). These represent those partner neurons at which the numerical strength of pre and postsynaptic pathways is at least 4:1, where four represents a balance in favour of the presynaptic pathway, 0.25 represents a balance in favour of the postsynaptic pathway, and a value of 1.0 represents equal numerical balance in both directions. Numerical values are reflected about the diagonal of the matrix, changed inversely between the two sectors, 0.25 in one direction corresponding to four in the other.

https://doi.org/10.7554/eLife.37550.017

We next consider reciprocity between particular cell pairs (Figure 8). There is considerable reciprocity with the two main LN classes (LN1 and LN2L): a) Between types LN1 and mPN1, 74% of pairs are reciprocal with 93% of pairs having a direction of mPN1 (pre) to LN1 (post) and an average ratio of 2.2; and b) between LN2L and mPN1, for which 84% of pairs are reciprocal, with 87% of those pairs being mPN1 (pre) to LN2L (post) and having an average ratio of 1.5. The mPN1 are all highly reciprocal on each other with an average connection strength of 9.9.

LN2L forms many reciprocal connections to other partners, too. Pairs of cells comprising LNs as one partner constitute 85% of the total of all reciprocal cell pairs in the matrix. LN reciprocity is widely observed between many LN pairs but mostly undirected, and with relatively few synapses in either direction. These are likely to be inhibitory (Wilson and Laurent, 2005) and feedback from LNs may therefore serve to suppress, or curtail signalling in both LN partners, and thus be disinhibitory, rather than to shorten transmission in the forward direction. Such reciprocal partnerships form connections on average of 6.7 ± 4.8 synapses (8.1 ± 5.2 in the stronger direction and 4.7 ± 3.1 in the weaker). In comparison, the average connection strength of ipsilateral ORNs to mPN1 is 21.6 ± 7.7 and for contralateral ORNs to mPN1 is 9.2 ± 4.6 synapses, compared with a recent report (Tobin et al., 2017) which found an average of 23 synapses from ipsilateral ORNs.

Discussion

Our study reports the complete synaptic connectome for all three cell types, ORN, LN and PN, of a single large glomerulus of the Drosophila antennal lobe, providing a proof of principle for the remaining 50 glomeruli. It complements two contemporary EM studies for different glomeruli (Rybak et al., 2016; Tobin et al., 2017), augmenting these by including important synaptic circuits of the LNs, which had not previously been known in any detail at synaptic level yet which constitute nearly half the cells of the glomerulus. Our data are notable for comprehensively documenting synaptic partnerships, and for having overcome the challenge of reconstructing the dense multidirectional arbors of antennal lobe local neurons, aided by the superior z-axis resolution of FIB-SEM compared with ssEM. Unlike PNs and their ORN inputs, and except LN2V, we find that LNs have three main characteristics: they have fewer synapses on average than PNs and ORNs; these are heterogeneous, and they form undirected, reciprocal synaptic networks which are inhibitory.

The connectome we report is essential to promote functional analyses using genetic dissection methods (Venken et al., 2011). In particular it will enable the interpretation of future analyses of network function made possible: first, by imaging methods, either in vitro, especially using genetically encoded calcium (e.g. Tian et al., 2012) or voltage (e.g. Antic et al., 2016) indicators; or second, during intact behaviour in vivo (e.g. Grover et al., 2016), in response to single odours or their combinations (e.g. Yasuyama et al., 2002). Our data also support computational approaches to insect olfaction.

The volumetric packing density of synapses, one site per 2.29 µm3 of neuropile, corresponds approximately to the differently computed values reported from TEM by Rybak et al. (2016) for glomerulus VA7, but is less than the density of synapses in the first and second optic neuropiles, the lamina and medulla (Meinertzhagen, 2014). We may speculate that the more densely packed visual synapses also employ higher rates of transmitter release and are consequently more energetically demanding, but deeper comparisons are hard to make.

We provide the complete connectome from a sensory neuropile in Drosophila, augmenting previous reports on other brain regions, the optic medulla (Takemura et al., 2015) and the mushroom body, both the calyx to which the PN axons project (Butcher et al., 2012) and their Kenyon cell output neurons in the alpha lobe (Takemura et al., 2017). Both the present report and the previous medulla connectome reveal only part of their neuropile as a whole however, for glomerulus VA1v, a single glomerulus in the antennal lobe, and for seven columns only of the optic medulla (Takemura et al., 2015). Like the latter, for all it is comprehensive the connectome of this single fruitless-positive glomerulus lacks most of the circuits of any of the approximately 50 other glomeruli, even of its neighbours, to which most of the local interneurons, those that are multiglomerular, also extend.

Dense reconstruction captures known genetic cell types

Along with genetic labelling from single-cell reporter lines, serial-section EM provides one of two canonical means to identify the morphological signature of a neuron (Meinertzhagen et al., 2009; Meinertzhagen, 2018). The mostly complete agreement we find between the two methods, reported here for FIB-EM and previously from Gal4 enhancer-trap labels viewed by light microscopy (Tanaka et al., 2012), provides assurance that both methods reliably succeed in identifying cell types of glomerulus VA1v, each yielding information that complements the other. From FIB-SEM, in particular, we can be sure that all cells are detected and none can hide in the forest of others, and that apart from orphan fragments the portions of each cell within VA1v are completely identified. This has been important in particular to plot the more elusive connections of the numerous LNs, which branch with greater complexity and are harder to identify than the 18 PNs, and which previous accounts lack at a synaptic level. On the other hand, most LNs are multiglomerular (Chou et al., 2010; Tanaka et al., 2012) so that their arbors in glomeruli other than VA1v still await completion, and we were unable to trace some neurites to a soma, making cell identification problematic in a few cases.

The synaptic connections we report confirm many that have been conjectured from light microscopy (Tanaka et al., 2012), even if the latter lacks the resolution to assert the presence of synaptic contacts. Most obvious is the pathway provided by ORN input to the three mPN1s, which project via the medial antennal lobe tract and provide input in turn to the mushroom body calyx and lateral horn (Stocker et al., 1990; Tanaka et al., 2012). Reporter Gal4 screens may not have identified all neurons per identified class, however. In particular, from Gal4 lines Tobin et al. (2017) report three PNs on the left side glomerulus DM6 and two on the right side, whereas a total of only 59 types for all 50 glomeruli have been labeled by Gal4 drivers in another study (Tanaka et al., 2012), an average of about one PN per glomerulus. In a more complete inventory revealed by photoactivating GH146-Gal4 positive PNs, most glomeruli are reported to contain an average of 2 ± 1 uniglomerular PNs, while six remaining glomeruli, which include VA1v, are innervated by 6 ± 2 PNs (Grabe et al., 2016).

Cellular organization of VA1v

The VA1v glomerulus is innervated by sensilla containing olfactory receptors involved in courtship and mating (Grabe et al., 2016) and additionally exhibits sexual dimorphism (Kondoh et al., 2003; Stockinger et al., 2005), its output delivered by the three uniglomerular mPN1s that project to the mushroom body and lateral horn (Tanaka et al., 2012). These possibly utilize a strategy to ensure that highly essential odor cues are transferred reliably and quickly from the antennal lobe to higher brain centers. Coincidentally they offer the eventual prospect to examine the network basis for sexual dimorphism in the antennal lobe, a prospect made possible by the connectome we report here.

The large number of VA1v PNs is a feature predicted to improve the detection accuracy and latency of odor stimuli (Jeanne and Wilson, 2015) and may also encode diverse stimulus features (Grabe et al., 2016). The large number leads us to anticipate that the number of LN dendrites is, by contrast, small relative to other glomeruli (Grabe et al., 2016), but we lack data for the latter. The number of PNs previously seen in genetic screens is significantly fewer than the 18 we find from FIB-SEM, but distinction between mono- and multiglomerular PNs must be made for valid numerical comparisons. Even so, the reasons for such wide differences are not entirely clear but suggest that the genetic methods so far used underestimate the reported PN numbers, and thus that GH146 Gal4 screens are far from saturated. This is a major conclusion to be drawn from our connectome.

In addition to their number, or possibly as their correlate, the PNs of VA1v as a whole exhibit considerable morphological diversity between classes. This is probably not typical of most antennal lobe glomeruli that have fewer PNs, as for example is clearly the case for glomerulus DM6 for which the monoglomerular PN synaptic connections have also been recently reported (Tobin et al., 2017). Supporting the uniformity of DM6 PNs as a class, pairs of PNs simultaneously recorded in the same glomerulus have well correlated levels of neuronal activity (Kazama and Wilson, 2009). Morphological diversity cannot of course even be anticipated for the many other glomeruli that contain only a single PN (Grabe et al., 2016). The situation remains open for yet other glomeruli and VA1v is probably not unique in containing so many PN types. For example, DA1 with at least 9 PNs (Grabe et al., 2016) contains PNs with different origins, from both the lateral (Lai et al., 2008) and anterodorsal (Lin et al., 2010) lineages. The latter generates exclusively uniglomerular PNs that project through the inner antennocerebral tract, while the lateral lineage generates various types of neurons, including uniglomeurular PNs (Lin et al., 2010).

Synaptic identification and composition

We have annotated contacts with anatomical features of chemical synapses, in particular with T-bar ribbons and presynaptic vesicles. As for other connectomes, in the antennal lobe (Rybak et al., 2016; Tobin et al., 2017) and optic neuropiles (Takemura et al., 2013; Takemura et al., 2015), we were therefore unable to annotate appositions that we could reliably interpret as putative gap junctions (Bennett and Goodenough, 1978) between neurons. The tortuosity of the tightly woven neurites prevented us from visualising linear stretches of membrane with close appositions, having densities on the membranes of both sides, that might have provided evidence for candidate gap junctions, like those reported from more favourable sites (e.g. Shaw and Stowe, 1982).

Our findings clearly establish that each of the three major classes of neurons is both pre- and postsynaptic to the other two, that in general most neurites have an approximate ratio of one presynaptic site for three postsynaptic contacts, most synapses thus being triads or tetrads.

Neurotransmitters

Detailed evidence for the neurotransmitters used by identified antennal lobe neurons is far from complete. As single classes, the ORNs express a cholinergic phenotype (Wilson, 2013), as do the projection neurons, at least those that project to the mushroom body calyx (Yasuyama et al., 2002) and lateral horn (Yasuyama et al., 2003), while LNs are likely to be either GABAergic (Wilson and Laurent, 2005Okada et al., 2009 ; Seki et al., 2010), and thus inhibitory, or glutamatergic and thus probably also inhibitory (Liu and Wilson, 2013).

Synaptic reciprocity

One opportunity enabled by analysing an entire connectome is to identify comprehensively the extent of synaptic reciprocity between neuron partners. Widespread synaptic reciprocity has been identified among synaptic circuits in many different brains and is hardly new. Various forms have in fact been identified and named, mostly for their inferred or demonstrated functions (Shepherd, 1998). Indeed, the anatomical demonstration of reciprocal synapses was reported in very early EM studies, both on the vertebrate olfactory bulb (Rall et al., 1966), and among the circuits of the inner retina, at feedback synapses of amacrine cells (Dowling and Boycott, 1966). But its wider documentation has mostly lacked specific attention or quantification and the full extent of synaptic reciprocity has never been fully revealed, in a way now possible from recently reported entire connectomes. Not in fact since such data have become available has it been possible to examine this question further, as is now possible in Drosophila, in the lamina (Zhao et al., 2015), medulla (Takemura et al., 2015), and mushroom body (Takemura et al., 2017). The optic neuropiles in Drosophila reveal the widespread incidence of synaptic reciprocity among visual circuits, while in the tadpole larva of Ciona the total proportion of reciprocal synaptic connections between neuron pairs is 0.39 (Ryan et al., 2016). By comparison we find a ratio of 0.25 for strongly reciprocal pairs in the antennal lobe. Takemura et al., 2015 discuss the possibility that the reciprocity of synaptic connections may be used to offset variation in the number of synapses in either direction, the operation of a synaptic circuits then depending not on the strength of a particular connection but on its ratio with other connections. Implicit in the reciprocal arrangement of synaptic partners is the fact that transmission in forward and backward directions should have opposite polarities.

In contrast to the ORNs and PNs, LN input to other cells although widespread, exhibits a wide range of synaptic connections that are numerically weak, overshadowed by the numerical strength of ORN input to PNs. Some LNs share reciprocity with a multiglomerular LN partner, as part of a weak lateral interglomerular network that is presumed to be mostly inhibitory. Our findings highlight in particular the extent of reciprocity between the synaptic elements of VA1v and the undirected nature of LN networks, the function of which may be estimated by those that express a GABA phenotype (Wilson and Laurent, 2005) or GABA receptors (Okada et al., 2009) and are thus predicted to mediate inhibition. The expression of other neurotransmitters offers ground for claims that are less certain.

An exception to these features is shown by LN2V, which is strongly connected to both ORN and PN neurons. The numbers of synapses are otherwise in most cases small, generally not more than 6, compared for example with the ipsilateral ORN to LN2V (about 25 synapses per ORN) and to individual PN pathways (about 100 synapses). Perhaps LN2V is a more direct feedback neuron than other LNs. For further predictions, it will be important to demonstrate that the anatomical ratios for synaptic feedback match the corresponding synaptic currents, which will require identification of postsynaptic receptors and their corresponding membrane conductances. Further comparisons between the connectivity of VA1v and that in other olfactory glomeruli will of course be instructive.

Further predictions for LN function based on our circuit information are still hard to make. It is particularly telling that silencing the populous LN1 and LN2 cells in the entire antennal lobe fails to modulate either glomerular input or output activity, suggesting their lack of influence on odor identity coding as a whole, and their possible role instead in more local interactions, possibly on a global or longer time scale (Strube-Bloss et al., 2017), predictions that are currently hard to evaluate anatomically.

Materials and methods

Animals

The dissected brain of a 6 day female fruit fly, Drosophila melanogaster, a cross between homozygous w1118 and CS wild type, was prepared by high-pressure freezing and freeze substitution as previously reported (Takemura et al., 2013). For this, the fly was stabilized in a collar (Heisenberg and Böhl, 1979), and a 240 µm slice cut from the head in a frontal plane using a vibrating microtome (Leica VT1000), and prefixed for 20 min in 2.5% each of paraformaldehyde and glutaraldehyde in 0.1M cacodylate buffer, then high-pressure frozen and freeze-substituted in 1% OsO4, 0.2% uranyl acetate, 3% water in acetone, and then embedded in Durcupan, as previously reported (Zhao et al., 2015; Xu et al., 2017).

FIB-SEM imaging

The right side was trimmed for FIB milling after it had been screened using X-ray imaging (Xradia Versa XRM-510) and the strongly X-ray-positive images used to select the antennal lobe. This X-ray imaging procedure was required to provide the precise depth to enable precise trimming at which FIB imaging should start.

An image stack of ~8,900 FIB-SEM images was then acquired from the block face (Knott et al., 2008; Xu et al., 2017), collecting images from the brain’s right side at a resolution of 8 nm/pixel, in a frontal plane from anterior to posterior, using a Zeiss NVision (Xu et al., 2017). Imaging extended for most of the entire antennal lobe’s depth and included a portion of the antennal nerve. In between images a focused gallium ion beam was used to remove 2 nm from the block face, and secondary electrons emitted from the block face collected in consecutive images. Minor shifts in these were aligned using affine transforms and consecutive sets of four images summed to generate 8 nm voxels that after final alignment yielded an isotropic stack with 8 × 8 × 8 nm resolution.

Glomerulus VA1v

We chose VA1v because it lay close to the lateral border of the lobe, near the entry point of the antennal nerve. The borders between many glomeruli were usually not well distinguished in single images from material prepared after rapid freezing preparation for FIB-SEM imaging (Figure 2) which, unlike conventional fixation for ssEM, left glial boundaries expanded and pale. To demarcate the specific glomerulus we first used an algorithm that identified the presynaptic organelles, T-bar ribbons (Fröhlich, 1985; Hamanaka and Meinertzhagen, 2010), to enable large-scale, automated imaging of synaptic contacts (Huang and Plaza, 2014; Kreshuk et al., 2011). Applied to the entire antennal lobe’s FIB image stack, these generated a synapse point cloud of over 500,000 synaptic puncta (Figure 1A,B), as previously reported (Zhao et al., 2015). The local variations in the density of synapses, elevated in regions (Figure 1A) corresponding to individual glomeruli (Zhao et al., 2015), mirrored reconstructions from light microscopy of synapses made visible by immunolabelling the presynaptic protein Bruchpilot (Laissue et al., 1999). Detailed comparisons between the light and FIB-SEM datasets enabled us to identify VA1v (Couto et al., 2005); Table 1). Using these features, a specific region of interest (ROI) was determined using the software tool Neutu. Different Gal4 lines combine the previously reported VA1l and VA1m into VA1v (Couto et al., 2005; Endo et al., 2007), indicating that not all anatomical territories coincide with genetic boundaries. 

Synapse annotation

For synapse annotation two trained proofreaders who had attained a recall proficiency in excess of 85% and a precision of >94.7% were used to annotate presynaptic T-bars in the EM volume. They marked pre- and postsynaptic sites manually in software, Raveler (https://openwiki.janelia.org/wiki/display/flyem/Raveler) or the later NeuTu (https://github.com/janelia-flyem/NeuTu; Zhao et al., 2018) in combination with the Distributed, Versioned, Image-Oriented Dataservice DVID (https://github.com/janelia-flyem/dvid Katz and Plaza, 2018). Unlike neuropiles in other species, in Drosophila the presynaptic sites are far more easily recognized than the corresponding postsynaptic densities (PSDs). T-bar annotation was undertaken first followed by PSD annotation, so that each T-bar was reviewed again. A small minority of T-bars (<1%) that lacked clear PSDs were removed from the dataset. PSDs were further refined by checking all autapses. A total of 3822 pathways with one synaptic contact were identified. These were all checked by a second proof-reader and approximately 2881 were verified, to confirm both the contact and the neurons that this connects. Of these, only 58 were then denied.

Reconstructions

Proofreading was performed so that bodies with pre- and postsynaptic partners were connected to progressively more extensive neurites, in both Raveler and NeuTu-EM and in combination with DVID software, according to previously published methods (see Chklovskii et al., 2010; Plaza et al., 2014; Takemura et al., 2015; Takemura et al., 2017) by previously trained proof-readers. The image stack was automatically segmented and putative neurite profiles generated (Chklovskii et al., 2010; Plaza et al., 2014). Reconstructions were made initially by marking pre- and postsynaptic sites manually. Segmentation was undertaken using a context-aware two-stage agglomeration framework (Parag et al., 2015). We undertook dense reconstruction of all elements within the glomerulus using focused proof-reading (Plaza et al., 2012) and sparse reconstruction beyond its borders to determine the location of somata and axon tracts. The density of connectome reconstruction helped us eliminate errors of omission, instances of failures to detect synapses. Neuron reconstructions were closely examined for proof-reading errors. Consensus-based proofreading using up to four proofreaders arbitrated disagreements in the shapes or connectivities of reconstructed neurons. Proof-reading required us to split reconstructed bodies frequently. Cell types of reconstructed neurons were identified by comparing the shapes of their arbors and by the locations of somata and axon tracts, with those previously reported from light microscopy (Tanaka et al., 2012). Connectivity matrices were generated by combining the results of neuron reconstructions with those for the connection strength from the number of pre-/post synapses. Synapses had a multiple-contact composition, incorporating several postsynaptic neurites linked coordinately to receive transmitter released at a single presynaptic site. A matrix was constructed for all bodies with partnerships exceeding 50 connections between a single pre- and postsynaptic contact. In fact we found three or more connections for about 37% of connected cell pairs, and took this criterion as a lower limit to acknowledge a pathway, disregarding connections with yet fewer examples as possible errors of human provenance caused by incorrect proof-reading. Networks were plotted in Cytoscape, v. 3.5 (Shannon et al., 2003).

Our analysis took 60 person months of proofreading time and 20 person months for curation.

References

  1. 1
  2. 2
    Gap junctions, electrotonic coupling, and intercellular communication
    1. MV Bennett
    2. DA Goodenough
    (1978)
    Neurosciences Research Program Bulletin 16:1–486.
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
    DVID
    1. WT Katz
    2. S Plaza
    (2018)
    Github.
  26. 26
  27. 27
  28. 28
    Heavy Tailed Distributions
    1. LB Klebanov
    (2003)
    Prague: Charles University.
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
    The anatomical organization of the compound eye visual system
    1. IA Meinertzhagen
    (2014)
    In: J Dubnau, editors. Handbook of Behavior Genetics of Drosophila melanogaster: Behavioral Phenotyping Protocols, 1. Cambridge University Press. pp. 1–19.
  46. 46
    Of what use is connectomics? (Commentary)
    1. IA Meinertzhagen
    (2018)
    The Journal of Experimental Biology jeb164954.
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
    Freeze-fracture evidence for gap junctions connecting the axon terminals of dipteran photoreceptors
    1. SR Shaw
    2. S Stowe
    (1982)
    Journal of Cell Science 53:115–141.
  60. 60
    The Synaptic Organization of the Brain (4th Edn)
    1. GM Shepherd
    (1998)
    New York: Oxford University Press.
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
  83. 83
  84. 84
  85. 85
  86. 86

Decision letter

  1. Liqun Luo
    Reviewing Editor; Howard Hughes Medical Institute, Stanford University, United States
  2. Eve Marder
    Senior Editor; Brandeis University, United States
  3. Moritz Helmstaedter
    Reviewer; Max Planck Institute for Brain Research, Germany

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "The complete connectome of a Drosophila antennal lobe glomerulus" for consideration by eLife. Your article has been reviewed by three peer reviewers, including Liqun Luo as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Eve Marder as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Moritz Helmstaedter (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Essential revisions:

– Please submit as a Tools and Resource paper (given reviewers' reservations about new biological insight)

– Please provide synaptic calibration data (report precision/recall numbers on what they call a synapse) and other relevant metrics

– Please ensure full data availability, which will be checked by reviewers/editors before final acceptance

– Please re-write the manuscript with the general readership of eLife in mind, integrating as much as possible previous data on the subject and providing biological significance of the finding.

– Please also address other comments from the reviewers; see below for the full reviews.

Reviewer #1:

Horne et al. reported the reconstruction of the connectome for a VA1v glomerulus of the Drosophila antennal lobe using FIB-SIM, a state-of-the-art EM reconstruction method that can achieve 8-nm isotropic resolution. The major advance compared to previous efforts in EM-based reconstruction of antennal lobe connectivity (Tobin et al., 2017; Rybak et al., 2016) is the inclusion of local interneurons (LNs), which are in large numbers and are heterogeneous based on light microscopic single-cell tracing data. Indeed, the authors identified 56 LNs in the VA1v glomerulus, and reported synaptic connections between 56 LNs, >100 olfactory receptor neurons (ORNs), and 18 second-order projection neurons (PNs) in this glomerulus. These quantitative data (summarized in Figures 5 and 7) will be highly valuable in understanding how the Drosophila antennal lobe processes olfactory information. However, the paper contains a number of problems that need to be extensively revised before it can reach the high standard of an eLife publication.

1) My major critique is that the current writing style makes it very difficult for readers besides a small group of specialists to be interested in reading the paper. The description of the data is accompanied with little biological meanings or insights. While EM reconstruction must describe dry data, this one is an extreme; for comparison, Tobin et al., 2016 is much more interesting to read even for a specialist. Major efforts need to be put to streamline the writing and interpret the data with biological context, with the general readership of eLife in mind.

2) There are a few significant errors or vague statements that further impede reading. The most important one occurs in paragraph three of subsection “Identified neurons and their synaptic networks”. I believe "medial lateral tract" should be "medial tract", or mALT. The medial lateral tract contains entirely different type of neurons (belonging to group b). Furthermore, in the same paragraph, it promises three groups, but I did not see c) after a) and b).

3) Assuming my interpretation above is correct, there are only 3 uniglomerular PNs that belong to the mALT group. This number is different from the reported 5 based on lineage tracing using dual-color MARCM (Yu et al., PLoS Biology 2010). The authors should comment on this. In general, it will be useful for the authors to systematically specify which neurons in their reconstruction have been traced to the cell bodies, and what is the location of cell bodies with respect to the antennal lobe neuropil; such information can be presented an additional column or row in Figure 5. These will provide cross-validation as the lineages of antennal lobe neurons have been extensively documented in the literature.

4) Speaking of Figure 5, I need to amplify to about 600% on my large computer screen in order to see the data. I expect that it will be quite difficult to fit into a regular eLife page with sufficient resolution. Figure 8 suffers from the same problem. The authors should think of creative ways of presenting these data that are more readily seen by readers.

5) The manuscript contains a number of citation errors. Two examples: Introduction paragraph four, they should cite Marin et al., 2002, which was published back-to-back with Wong et al., and which contains several unique single cell tracings that are relevant to this study. In paragraph seven of subsection “Identified neurons and their synaptic networks”, I think they mean Tanaka 2012 (rather than 2010).

6) The acronym FIB-SIM should be spelled out in Abstract at the first mention. The authors should also consider adding "nearly" before "complete" to their title given that there are still orphan profiles (despite the tremendous amount of work!).

Reviewer #2:

The manuscript "The complete connectome of a Drosophila antennal lobe glomerulus" by Horne et al. reports the electron-microscopic imaging and dense connectomic reconstruction of the synaptic circuits in a glomerulus of the antennal lobe in Drosophila melanogaster.

This is a comprehensive piece of work containing a lot of potentially useful connectomic information.

In my understanding, compared to previous connectomic reconstructions of glomeruli in Drosophila, which the authors cite (Rybak et al., Tobin et al.), this study includes the connectivity of local interneurons which were not available in the previous publications.

Since I am neither an expert in the olfactory system nor Drosophila neurobiology I will focus my comments on the EM and connectomic reconstruction aspects of this work.

This is a comprehensive piece of data and I commend the authors for this work, but I have a few concerns that in my view are substantial and should be addressed before this manuscript can be considered for publication at eLife.

1) While in principle, FIB-SEM is able to provide extremely well-aligned high-resolution data of neuropil, the data presented in the figures of this manuscript seemed to be of lower resolution and/or staining contrast than one would expect from nominally 8nm resolution imaging. Namely the few EM images in Figure 3 a) to d) have surprisingly little intracellular detail. For example, the mitochondria's internal structure is not visible. It may of course be that these are processed images that appear to be at lower resolution but this gives me a certain amount of concern about the ability to detect synapses in this data.

2) Synapse detection: The authors write a section on this topic, but it describes primarily how difficult synapse detection was -without clearly convincing that authors were able to detect synapses eventually (especially when the T-bar was missing). I did not understand how the authors calibrated the synapse detection in this data. Were high-resolution small image volumes obtained for direct comparison? Or were synaptic profiles of certain previously studied neurons compared, for instance to the Rybak or Tobin study?

3) Especially in the context of these concerns about the raw data it is in my view absolute standard by now that the raw image data and the reconstruction should be made freely available and browsable to reviewers and readers. As far as I can see the authors did not specify the dataset availability beyond the paper figures- also the transparent reporting sheet is actually largely empty. I am a bit surprised by this since obviously the Janelia team has a lot of resources to make EM data available to the community and this constitutes in my view an important aspect of such a publication. Especially also in light of the following points, I think it is even mandatory that the 3D image data and reconstructions are made available already at the review stage.

4) A few important quantifications are missing which are very relevant for putting this study in comparison to other connectomic studies:

a) What was the total annotation time invested to obtain these reconstructions (including curation and proof-reading time)?b) What is the total neuronal wiring length that is contained in the provided reconstruction?c) Density of reconstruction: The authors report a few qualitative statements ("no cell could hide"). What would be very helpful here is the volume fraction of neuropil that was accounted for by the ">250" (please give precise number) or 192 neurons. How does the number 93% , in the final paragraph of subsection “The synaptic dataset of glomerulus VA1v”, relate to the point about dense reconstruction discussed in the following paragraph?d) I find the description of synapse detection, second paragraph of subsection “Synapses” following, rather confusing.e) Is the number 87% , in subsection “The synaptic matrix of VA1v”, the volume fraction a accounted-for neuropil? This should be ideally reported in the section called "dense reconstruction" above.f) Numbers on error rates in the obtained Reconstructions are missing. How often did proof reading have to happen etc.?g) The section on reconstruction methods, is rather short and not very informative quantitatively.

5) I do get the impression that the manuscript does not report a large number of clear novel biological results. This is surprising since the authors state that connectivity of local neurons (LN) was not studied before. Even the Abstract, if I understand correctly, does not contain a biological finding beyond the reported reconstruction. I would therefore suggest considering this manuscript as a resource publication rather than a results manuscript. Alternatively, example analyses that would show the impact of this connectivity data would – at least for me as an outsider to this particular field – be very helpful in order to appreciate the depth and extent of this data.

As a final minor comment, I am surprised that the authors are citing the first but afterwards revised reconstruction of the fly medulla column, Takamura et al., 2013, instead of Takamura et al., 2017. In my understanding it is the 2017 eLife paper that provided the correct connectivity in that piece of fly neuropil.

Reviewer #3:

Horne and colleagues apply recent advances in focused ion beam milling scanning electron microscopy (FIB-SEM) to re-construct a largely complete connectome for the VA1v glomerulus in Drosophila. This is sexually dimorphic glomerulus is responsible for detecting female fly odors and is involved in signaling involving sex pheromones. In the last 3 years, considerable work has focused on reconstructing complete connectomes of olfactory regions by EM in the larval brain and indeed, in multiple olfactory glomeruli. However, these were accomplished largely using serial-section EM (ssEM); though ssEM can provide vast information, it can be difficult to clearly delineate small calibre neurites, thus confounding efforts at complete reconstruction. In using FIB-SEM to solve this problem, Horne and colleagues offer a more comprehensive dense reconstruction of a glomerulus, assigning most synapses to a parent neuron, and conducting a more thorough assessment of projection neurons and local interneurons.

The authors are to be commended on an outstanding contribution to Drosophila neurobiology; it is archival work that will be of great use to olfactory biologists. However, the deeper question is whether this is a significant advance to merit eLife publication. The FIB-SEM method has been established by a similar group of previous authors, so the methodology is not new. And though they were done with ssEM, this isn't even the first glomerulus to be fully reconstructed by EM. The new method does provide more information and more completion, but this is really an incremental increase. Because of these reasons, the novelty of the connectome is limited. Further, I don't have any intellectual issues with the analyses or the reconstructions, but I think this analysis would be more suited to a specialty journal.

The chief advances of this connectome rest in a much more comprehensive map of projection neuron and local interneuron projections. These are useful, but the scope is very narrow. Further, in the presentation of each type of neuron, the manuscript is written in such a way that non-fly-aficionados will have a tough time determining what's important. What are the new concepts that can be gleaned from this reconstruction? What do we learn about LN projections? What about PN projections? These areas are minimized (if not completely omitted), making it difficult to interpret why the new connectome is of value to neuroscience at large. These issues would greatly enhance the readability of the manuscript, though I still feel the core advance represented by the paper is better suited for another journal.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "A resource for the Drosophila antennal lobe provided by the connectome of glomerulus VA1v" for further consideration at eLife. Your revised article has been favorably evaluated by Eve Marder (Senior Editor), a Reviewing Editor, and 3 reviewers. The reviewers have discussed their opinions about the revised manuscript, and are in agreement.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below.

Most importantly, the raw data needs to be available for inspection by reviewers before we can accept the paper. Please send a revised manuscript that addresses the rest of the issues (mostly minor) below, after the data are uploaded to a website that the reviewers can inspect. Thank you.

Reviewer #1:

The revised manuscript by Horne et al. is much improved in its presentation and accessibility to a general readership. Also the new format of "Tools and Resources" fits better with the dataset. So I am enthusiastic in support its publication, assuming that the EM expert reviewer is satisfied with the technical aspect of the data.

Below are a few (minor) suggestions for further improvement.

1) Since the authors emphasize in the Abstract that VA1v is a sexually dimorphic glomerulus, they should tell the readers more explicitly in the main text that the sample is from a female (right now the information is in the Materials and methods and figure legend), and what is the nature of sexual dimorphism (male VA1v has larger volume than female VA1v).

2) Subsection “Identified neurons from dense reconstruction in VA1v2”, paragraph three: most cells (should be changed to LNs) being panglomerular. On that note, none of the LNs in reconstructed LNs in Figure 4—figure supplement 3 appeared panglomerular, presumably because they were not fully reconstructed; the authors should state this explicitly, and how they categorize the LNs even though they were not fully reconstructed.

3) Figure 7A: it would be useful to add an output arrow from PNs, the destination being "higher olfactory centers." Even though it is not part of the reconstruction, it will be useful to remind the readers where the information goes from the antennal lobe.

Reviewer #2:

The revision of Horne et al. is improved, the formatting as a resource paper serves the data much better. The text has improved and contains more quantifications. Synapse detection is properly described and convincing.

As to the availability of the raw image data at the review stage, I maintain the view that by eLife standards (and similarly for many other journals) the raw data should be made available to reviewers for inspection, not just promised to be available later. This is especially true for a resource manuscript, but applies more broadly. I would hope that with the infrastructural capabilities of Janelia, data from the FlyEM team could be rather spearheading such modern data transparency standards than circumventing them.

As to the reporting of quantifications, I am surprised that the authors choose NOT to report certain quantitative measurements. The authors state that the reconstruction comprised a total of "15.8 cm of neurite". Since the authors are free to put this number into context, as in the reply to my comment, I would encourage this number to be reported, not withheld from the reader.

Reviewer #3:

The revised version of Horne et al. is a much-improved manuscript. Overall, I am satisfied with the comments from the authors and the considerable revisions included therein. I think the adjustment to a "Tools and Resource" and the reframing of many of the conceptual advances from this manuscript are absolutely appropriate. I thank the authors for their consideration of most of the comments provided.

I will defer judgment to many of the mathematical and technical issues raised by the second reviewer, however, as these fall out of my purview.

One confusing section, however, is the third paragraph of subsection “Identified neurons from dense reconstruction in VA1v”. It seems like the goal is to highlight aspects of ORNs, but some of it seems to be discussing LNs. The LN cells identified by Chou et al., are panglomerular, but the wording makes it sound like the ORNs are pan-glomerular. Is this accurate? Could the paragraph be adjusted to improve clarity?

https://doi.org/10.7554/eLife.37550.023

Author response

Essential revisions:

- Please submit as a Tools and Resource paper (given reviewers' reservations about new biological insight)

We have revised our article to conform to the Instructions for a ‘Tools and Resource’ paper and, reflecting this change in emphasis, now revise the title to:

A resource for the Drosophila antennal lobe provided by the connectome of glomerulus VA1v

We take notice that Tools and Resources articles “do not have to report major new biological insights or mechanisms, but it must be clear that they will enable such advances to take place. Specifically, submissions will be assessed in terms of their potential to facilitate experiments that address problems that to date have been challenging or even intractable.” We therefore address the widely evoked role for connectomic data in enabling functional studies on identified neuron circuits, in the following original text:

“Including the synaptic network of all LNs and PNs to generate an actual map of the complete synaptic network, or connectome (Lichtman and Sanes, 2008), for each glomerulus is thus a final step towards adopting functional connectomic approaches (Venken et al., 2011; Meinertzhagen and Lee, 2012) to the analysis of antennal lobe function.” see also the following text, now added in the Discussion: “The connectome we report is essential to promote functional analyses using genetic dissection methods (Venken et al., 2011). In particular it will enable the interpretation of future analyses of network function made possible […]”

We summarize our findings with the general short statement: “Unlike PNs and their ORN inputs, and except LN2V, we find that LNs have three main characteristics: they have fewer synapses on average than PNs and ORNs; these are heterogeneous, and they form undirected, reciprocal synaptic networks which are inhibitory.”

“The connectome we report is essential to promote functional analyses using genetic dissection methods (Venken et al., 2011). In particular it will enable the interpretation of future analyses of network function made possible: first, by imaging methods, either in vitro, especially using genetically encoded calcium (e.g. Tian et al., 2012) or voltage (e.g. Antic et al., 2016) indicators; or second, during intact behaviour in vivo (e.g. Grover et al., 2016), in response to single odours or odour combinations (e.g. Silbering and Galizia, 2007). Our data also support computational approaches to insect olfaction.”

We have already emphasized that our EM connectomic data reveal that previous genetic screens of antennal lobe cells were clearly not saturated. This is an important finding for evaluating past reports in the field, which have relied heavily on genetic screens to reveal the antennal lobe’s component cells, but which our report now shows to be incomplete, and thus not wholly reliable. We do not wish to attack previous reports, which have contributed much, but in three places we refer quite clearly to the lack of saturation in previous screens (see Abstract, Discussion, or search “saturat” in the text), and regard this as a major scientific finding of our study.

- Please provide synaptic calibration data (report precision/recall numbers on what they call a synapse) and other relevant metrics

We now report these metrices in several places, as follows:

A newly combined section headed Synapse annotationin the Materials and methods where we report: “[…] trained proofreaders who had attained a recall proficiency in excess of 85% and a precision of >94.7% were used to annotate presynaptic T-bars in the EM volume.” Further information continues on in this paragraph, some parts of which were moved from the Results.

- Please ensure full data availability, which will be checked by reviewers/editors before final acceptance

We have now included the complete matrix of all single-cell synaptic partnerships as a data file, Figure 5—figure supplement 1. This is the burden of our complete dataset, suitable for computational studies, as well as the basis to interpret future functional analyses using methods based on genetic reporters.

- Please re-write the manuscript with the general readership of eLife in mind, integrating as much as possible previous data on the subject and providing biological significance of the finding.

We have endeavoured to do this by highlighting more clearly what we see as the novel, especially functional value of our anatomical analysis, but also take notice that “Tools and Resources articles do not have to report major new biological insights or mechanisms”. We do also endeavour to make it “clear that they will enable such advances to take place. Specifically, submissions will be assessed in terms of their potential to facilitate experiments that address problems that to date have been challenging or even intractable.” Our data fall, we feel, into this category, as we now indicate in several places in the text, as follows:

1) “The connectome we report is essential to promote functional analyses using genetic dissection methods (Venken et al., 2011). In particular it will enable the interpretation of future analyses of network function made possible: first, by imaging methods, either in vitro, especially using for example genetically encoded calcium (Tian et al., 2012) or voltage (Antic et al., 2016) indicators, or second, during intact behaviour in vivo (e.g. Grover et al., 2016), in response to odours or odour combinations (e.g. Silbering and Galizia, 2007). Our data also support computational approaches to insect olfaction.” (Discussion, paragraph 2)

2) “A prospect made possible by the connectome we report here”

3) Finally, we indicate that the genetic screens so widely used to date, seem to underestimate PN numbers considerably, and are thus not saturated, leading us to conclude that “This is a major conclusion to be drawn from our connectome.”

- Please also address other comments from the reviewers; see below for the full reviews.

Please see our responses to each comment in the rather long list given below. We indicate where we find that the reviewers’ suggestions are best met by the altered emphasis in our submission provided by its revision to a Tools and Resource paper, for which the original text was, in any case, already better suited.

We also shortened the text of the Discussion by removing a rather inconclusive terminal paragraph comparing the vertebrate olfactory bulb and fly antennal lobe, and cut:

“Limited comparisons with the circuits of the mammalian olfactory bulb are possible (Hildebrand and Shepherd, 1997). Two features are immediately obvious, the diversity of cell types and the complexity of their circuits. PN diversity in Drosophila is reminiscent of the projection neurons cohabiting individual glomeruli in the olfactory bulb. These are of two types, mitral cells and tufted cells, each with different morphological types and each projecting to different regions in the higher brain (Nagayama et al., 2014), much as PN cells do. The granule cell interneurons of the olfactory bulb share circuit similarities with Drosophila LN cells in being connected reciprocally to mitral and tufted output neurons, inhibiting these and receiving excitation from them (Shepherd and Greer, 1998). In Drosophila many LNs are correspondingly GABAergic (Wilson and Laurent, 2005; Okada et al., 2009) and thus likely to mediate inhibition.”

Reviewer #1:

[…] 1) My major critique is that the current writing style makes it very difficult for readers besides a small group of specialists to be interested in reading the paper. The description of the data is accompanied with little biological meanings or insights. While EM reconstruction must describe dry data, this one is an extreme; for comparison, Tobin et al., 2016 is much more interesting to read even for a specialist. Major efforts need to be put to streamline the writing and interpret the data with biological context, with the general readership of eLife in mind.

The senior author takes responsibility for the previous version of the text. We have endeavoured to address all points and related comments from the other reviewers by revising much of the text so as to rearrange, improve and promote the flow of ideas, as indicated in red in the attached manuscript, and in particular as follows: 1) we have compiled all numerical summaries of our findings, which hitherto were scattered under different headings, into a single section in the Results. We present this summary in the section “The synaptic matrix of VA1v” in which we report the complete connectome, and make only passing reference to it in two other places in the Results, citing the connectome as a cross-reference.

1) We further emphasize the completeness of our synaptic analysis, especially of the numerous LNs, which have not previously been documented at synaptic level.

2) Given that we now submit our study as a Tools and Resource paper, we now expand the synaptic database originally presented as the entire matrix in Figure 5 by adding an additional figure supplement as a new look-up spreadsheet (Figure 5—figure supplement 1), which provides the numerical intercepts of the matrix in readable form. See previous response to this reviewer.

2) There are a few significant errors or vague statements that further impede reading. The most important one occurs in paragraph three of subsection “Identified neurons and their synaptic networks”. I believe "medial lateral tract" should be "medial tract", or mALT. The medial lateral tract contains entirely different type of neurons (belonging to group b). Furthermore, in the same paragraph, it promises three groups, but I did not see c) after a) and b).

We have amended medial lateral tract to medial tract, mALT, and reassigned the cell groups to become a)-c) with subtypes i)-iii). Our apologies for careless writing.

3) Assuming my interpretation above is correct, there are only 3 uniglomerular PNs that belong to the mALT group. This number is different from the reported 5 based on lineage tracing using dual-color MARCM (Yu et al., PLoS Biology 2010). The authors should comment on this. In general, it will be useful for the authors to systematically specify which neurons in their reconstruction have been traced to the cell bodies, and what is the location of cell bodies with respect to the antennal lobe neuropil; such information can be presented an additional column or row in Figure 5. These will provide cross-validation as the lineages of antennal lobe neurons have been extensively documented in the literature.

We were mortified to see our omission of the report by Yu et al. (2010), which we now include in our revision.

On the other hand, we believe that the reason for the difference in the number of PNs (we report 3 in VA1v and they report 5) is because the border between VA1v and the neighbouring glomerulus is not correct in the report by Yu et al. We therefore believe that the 5 reported by these authors are in fact the 3 that we find do enter VA1v plus 2 that enter a neighbouring glomerulus. In short, we think the discrepancy is a border dispute between close-neighbouring glomeruli. Moreover we cannot imagine how two entire PNs from our dataset could have been overlooked: there are nowhere sufficient orphan profiles to constitute two complete large neurons. One possible reconciliation might of course be if different flies have glomeruli of different cellular compositions, but this would not have invalidated conclusions from our dataset. We therefore think the correct attribution of borders between neighbouring glomeruli is the problem, and now state this explicitly in the text.

We have now specified those neurons that have been traced to their somata, and where possible indicated the locations of the latter. Cell bodies are now included as a column in Figure 2-source data 1. We were unable to trace a number of cells back to the cell body of origin, if this lay outside the grey scale ROI. This was in any case a lot of work.

4) Speaking of Figure 5, I need to amplify to about 600% on my large computer screen in order to see the data. I expect that it will be quite difficult to fit into a regular eLife page with sufficient resolution. Figure 8 suffers from the same problem. The authors should think of creative ways of presenting these data that are more readily seen by readers.

Yes, we agree with the reviewer concerning this problem, which we share with other contemporary accounts not least with one by Chou et al., 2010, for which the senior author is Liquin, who present a large matrix, the same size as our Figure 5, which similarly can be read only at 600%, but even then we think not as clearly as ours! The problem is numerical and not of our making, but the reviewer seeks resolution notwithstanding. We considered and reconsidered this problem, trying for example to address the issue by collapsing individual cells, in columns and rows, into cell types, as opposed to single cells. In fact this is not very satisfactory either, because it assigns equal space to columns that contain a great many cells (ORNs especially) as to columns that report only a single cell. To provide the complete matrix for single cells, we therefore wish still to give the complete data for these as a single-cell matrix, as well as in a more readable spreadsheet in supplementary data (see Figure 5—figure supplement 1), enabling interested readers to download our complete dataset for further analyses offline (see comment above). This is the most satisfactory solution we could reach, and can think of no alternative, nor apparently could the reviewer.

5) The manuscript contains a number of citation errors. Two examples: Introduction paragraph four, they should cite Marin et al., 2002, which was published back-to-back with Wong et al., and which contains several unique single cell tracings that are relevant to this study. In paragraph seven of subsection “Identified neurons and their synaptic networks”, I think they mean Tanaka 2012 (rather than 2010).

We now also cite Marin et al., 2002, with apologies for having omitted it, and have amended Tanaka et al., 2010 to 2012. We also additionally cite Silbering and Galizia, 2007, citation of whose work on LNs was previously omitted

6) The acronym FIB-SIM should be spelled out in Abstract at the first mention. The authors should also consider adding "nearly" before "complete" to their title given that there are still orphan profiles (despite the tremendous amount of work!).

We prefer not to spell out FIB-SEM in the Abstract, which would have increased the word total beyond the permissible limit, and reference to FIB-SEM then appears in the Results, where it is indeed spelled out. For the title, and partly for stylistic reasons, we really prefer not to add ‘nearly’ but rather to remove ‘complete’ so as to eliminate all reference to completeness. There are in fact several reasons why our connectome, like all others, is not complete, chiefly in having elements we were unable to identify, but the level of our completeness compares very favourably with contemporary reports in other connectomes. Given that we have redirected our submission to a Tools and Resource paper, we indicate the new emphasis anyway by a change in the title: A resource for the Drosophila antennal lobe provided by the connectome of glomerulus VA1v

Reviewer #2:

The manuscript "The complete connectome of a Drosophila antennal lobe glomerulus" by Horne et al. reports the electron-microscopic imaging and dense connectomic reconstruction of the synaptic circuits in a glomerulus of the antennal lobe in Drosophila melanogaster.

This is a comprehensive piece of work containing a lot of potentially useful connectomic information.

In my understanding, compared to previous connectomic reconstructions of glomeruli in Drosophila which the authors cite (Rybak et al., Tobin et al.), this study includes the connectivity of local interneurons which were not available in the previous publications.

Yes, this is by far the most notable feature of our study.

Since I am neither an expert in the olfactory system nor Drosophila neurobiology I will focus my comments on the EM and connectomic reconstruction aspects of this work.

This is a comprehensive piece of data and I commend the authors for this work, but I have a few concerns that in my view are substantial and should be addressed before this manuscript can be considered for publication at eLife.

1) While in principle, FIB-SEM is able to provide extremely well-aligned high-resolution data of neuropil, the data presented in the figures of this manuscript seemed to be of lower resolution and/or staining contrast than one would expect from nominally 8nm resolution imaging. Namely the few EM images in Figure 3 a) to d) have surprisingly little intracellular detail. For example, the mitochondria's internal structure is not visible. It may of course be that these are processed images that appear to be at lower resolution but this gives me a certain amount of concern about the ability to detect synapses in this data.

There are several things in this comment to which we should respond. As the reviewer indicates, the resolution is not that of TEM, but FIB-SEM. It is true that sampling at 8nm per pixel should be sufficient to reveal the cristae of mitochondria in appropriately aligned images. In fact it is possible to see striations in the mitochondrion at 9 o’clock in Figure 3A, although we agree these are not clear, certainly not as they would be in a single TEM image.

There are two major issues. 1) First is the imaging method. In our images the grey level of the synapse has been saturated so as to see membrane details more clearly, but in reality the synapses are ~2x darker than the membranes, which makes their electron density easier to distinguish as well. For that reason we sacrificed resolution to gain membrane contrast, and increase ease of later segmentation and proofreading steps, and to this extent the images are processed as the reviewer suggests. 2) In addition, the fixation method we have used, light aldehyde fixation followed by high-pressure freezing was in fact selected empirically over many trials to highlight membranes and synapses clearly. We now modify the text to indicate the fact that synapses stand out very clearly by the density of their staining, so that we do not need to rely so much on resolving the substructural details of the synaptic organelles as in TEM images of Drosophila or possibly in mammalian tissue. We hope the reviewer will agree that synapses in Figure 3 are simply unmistakable as dark bodies, and that what is missing in most cases is the clear platform that characterizes fly synapses, but not those of other insect species.

2) Synapse detection: The authors write a section on this topic, but it describes primarily how difficult synapse detection was -without clearly convincing that authors were able to detect synapses eventually (especially when the T-bar was missing). I did not understand how the authors calibrated the synapse detection in this data. Were high-resolution small image volumes obtained for direct comparison? Or were synaptic profiles of certain previously studied neurons compared, for instance to the Rybak or Tobin study?

What is important about our methods is that the synapses (presynaptic T-bars) are electron dense and clearly visible as dark structures, and even though they may lack a clear platform they have a very distinct pedestal. The Rybak and Tobin studies both used TEM images in which the platform is clear, but are otherwise similar to those we see with FIB. However, both studies analysed different neurons than those we have identified. Rybak reported circuits from three different glomeruli (DM2, DL5, and VA7) while Tobin reported DM6 from several brains, but neither previous study reported cells from vA1V.

Why no platform on the T-bar? We explain this further in the following sentence as follows: “This chiefly resulted because in our FIB-SEM images the grey level of the synapses was about twice that of the membranes. We therefore adjusted the electron density of images to increase membrane contrast, because this proved advantageous to enhance membrane continuity more reliably during later proof-reading steps (see Materials and methods), but rendered the platform of the T-bar ribbon often less distinctly in our FIB-SEM images than when seen in TEM.”

Are we accurate? We calibrated synapse detection in a number of ways:

1) First, the annotators were specifically trained to detect these profiles, and their numbers are comparable to those seen by Tobin, as the number of presynaptic T-bars per ORN. We cite the comparison:

“[…] of ipsilateral ORNs to mPN1 21.6 ± 7.7 and for contralateral ORNs to mPN1 is 9.2 ± 4.6 synapses, compared with a recent report (Tobin et al., 2017) which found an average of 23 synapses from ipsilateral ORNs” We could not make more exact comparisons with these two previous studies, because we examined a different glomerulus and cells than did either of these two.

2) Exact comparisons are hard to make because the synaptic data are differently reported, but we can compare our density of synaptic sites with that reported by Rybak using ssEM and we find similar numbers overall. Thus we found “roughly 2.29 synapses per µm3” which “corresponds approximately to the differently computed values reported from TEM by Rybak et al., 2016 for glomerulus VA7”.

3) Especially in the context of these concerns about the raw data it is in my view absolute standard by now that the raw image data and the reconstruction should be made freely available and browsable to reviewers and readers. As far as I can see the authors did not specify the dataset availability beyond the paper figures- also the transparent reporting sheet is actually largely empty. I am a bit surprised by this since obviously the Janelia team has a lot of resources to make EM data available to the community and this constitutes in my view an important aspect of such a publication. Especially also in light of the following points, I think it is even mandatory that the 3D image data and reconstructions are made available already at the review stage.

We will of course provide raw data and 3D image data, from the Janelia link (emdata.janelia.org/VA1v) to the greyscale and segmentation data. We have also added the following files into the transparent reporting sheet, as requested by the reviewer: Figure 5—figure supplement 1, and an Excel spreadsheet for Figure 2-source data 1, replacing Table 2.

4) A few important quantifications are missing which are very relevant for putting this study in comparison to other connectomic studies:

a) What was the total annotation time invested to obtain these reconstructions (including curation and proof-reading time)?

We have now added these times: “Our analysis took 60 person months of proofreading time and 20 person months for curation.” In the Materials and methods under the section on Reconstruction.

b) What is the total neuronal wiring length that is contained in the provided reconstruction?

The 192 bodies densely reconstructed in the glomerulus and sparsely traced outside it comprise 15.8 cm of neurite. We think this measure is a poor representation of neuron complexity and reconstruction effort for our connectome, however, given 1) the extreme size differences between different species; 2) that most of the complexity of the reconstruction resides in small processes by synapses; and 3) that the goal is finding connections not achieving wirelength. The total number of synapses reconstructed was >11,140 presynaptic sites with ~38,050 postsynaptic dendrites, as reported. Given our view, we elect not to include wirelength as a relevant measure but provide it to the reviewer for reference.

c) Density of reconstruction: The authors report a few qualitative statements ("no cell could hide"). What would be very helpful here is the volume fraction of neuropil that was accounted for by the ">250" (please give precise number) or 192 neurons. How does the number 93% , in the final paragraph of subsection “The synaptic dataset of glomerulus VA1v”, relate to the point about dense reconstruction discussed in the following paragraph?

We do report that 87% of the volume of glomerulus VA1v was listed in our connectome of 192 neurons. We also indicate that 93% of connections were resolved with both pre- and postsynaptic partners identified, including some connecting in neighbouring glomeruli. 87.9% of the synaptic connections have been assigned to the neurons in the connectome, leaving 5% of synapses connecting neurons not included in the connectome because they had <50 synapses.

d) I find the description of synapse detection, second paragraph of subsection “Synapses” following, rather confusing.

We have addressed this important point in our responses to comment 1 from this same reviewer. See para beginning “Why no platform on the T-bar?”

e) Is the number 87% , in subsection “The synaptic matrix of VA1v”, the volume fraction a accounted-for neuropil? This should be ideally reported in the section called "dense reconstruction" above.

We removed text under the heading dense reconstruction and coupled it with other metrical data under the section: The synaptic matrix of VA1v.

f) Numbers on error rates in the obtained Reconstructions are missing. How often did proof reading have to happen etc.?

Proof reading was undertaken continuously by trained proof-readers, who undertook dense reconstruction of an entire volume of neuropile, not redundant sparse tracing of that neuropile. We compared most of our LN reconstructions with those of the Wilson lab (Harvard) done by Asa Barth-Maron using sparse reconstruction of the same glomerulus and image series.

g) The section on reconstruction methods, is rather short and not very informative quantitatively.

These methods are based on those used elsewhere at Janelia and are well documented and cited in other reports (e.g. Takemura et al., 2017). We provide additional detail in this section and six references.

5) I do get the impression that the manuscript does not report a large number of clear novel biological results. This is surprising since the authors state that connectivity of local neurons (LN) was not studied before. Even the Abstract, if I understand correctly, does not contain a biological finding beyond the reported reconstruction. I would therefore suggest considering this manuscript as a resource publication rather than a results manuscript. Alternatively, example analyses that would show the impact of this connectivity data would – at least for me as an outsider to this particular field – be very helpful in order to appreciate the depth and extent of this data.

We now provide additional biological conclusions in the Abstract in which we hypothesize the mutual reciprocal inhibition by LNs onto PNs, and also between LNs in different glomeruli. Connections are sparse and variation in the strengths of connections. LNs have four distinguishing characteristics: unlike PNs and their ORN inputs, except LN2V, LNs have three main characteristics: they have fewer synapses on average than PNs and ORNs, have heterogeneous synaptic connections, and are undirected and reciprocal in their synaptic networks. Given the suggestion of the reviewing editor, we have now indeed revised our resubmission as a Tools and Resource paper.

As a final minor comment, I am surprised that the authors are citing the first but afterwards revised reconstruction of the fly medulla column, Takamura et al., 2013 Nature, instead of Takamura et al., 2017. In my understanding it is the 2017 eLife paper that provided the correct connectivity in that piece of fly neuropil.

We appreciate this point but should make it clear that the 2013 report was not in fact incorrect, merely incomplete, because it contained only a single medulla column and because some inputs arise in neighbouring columns that as a result could not be traced. The correction is not merely semantic, because the 2013 paper was a landmark advance that for the first time identified pathways from lamina inputs in the distal medulla to outputs onto T4 in the proximal medulla, laying the groundwork for solving the biological implementation of the H-R EMD. Looking through the text, there are many references to the 2015 paper, but in any case we did not find a citation to the 2013 paper that we think is misplaced and should be changed, so we are otherwise at a loss to know how to respond.

Reviewer #3:

Horne and colleagues apply recent advances in focused ion beam milling scanning electron microscopy (FIB-SEM) to re-construct a largely complete connectome for the VA1v glomerulus in Drosophila. This is sexually dimorphic glomerulus is responsible for detecting female fly odors and is involved in signaling involving sex pheromones. In the last 3 years, considerable work has focused on reconstructing complete connectomes of olfactory regions by EM in the larval brain and indeed, in multiple olfactory glomeruli. However, these were accomplished largely using serial-section EM (ssEM); though ssEM can provide vast information, it can be difficult to clearly delineate small calibre neurites, thus confounding efforts at complete reconstruction. In using FIB-SEM to solve this problem, Horne and colleagues offer a more comprehensive dense reconstruction of a glomerulus, assigning most synapses to a parent neuron, and conducting a more thorough assessment of projection neurons and local interneurons.

We do not agree with the reviewer’s viewpoint. There has been no prior comprehensive dense reconstruction of a glomerulus, only of the ORN and PN circuits. The more numerous local neurons of the glomerulus have never been comprehensively reconstructed, nor their circuits identified. This was clearly stated in the first paragraph of the Discussion, now retained in the revised text.

The authors are to be commended on an outstanding contribution to Drosophila neurobiology; it is archival work that will be of great use to olfactory biologists. However, the deeper question is whether this is a significant advance to merit eLife publication. The FIB-SEM method has been established by a similar group of previous authors, so the methodology is not new. And though they were done with ssEM, this isn't even the first glomerulus to be fully reconstructed by EM.

Again, we do not quite agree on this point. This is the first glomerulus to be fully reconstructed by EM. Previous reports made important contributions by covering specific PN and ORN contributions to a glomerulus, or three glomeruli, but the numerous, widespread synaptic circuits of LNs have not previously been systematically reported. As we state in the first sentence of the Discussion “Our study reports the complete synaptic connectome for all three cell types, ORN, LN and PN, […] providing a proof of principle for the remaining 50 glomeruli. It complements two contemporary EM studies … (Rybak et al., 2016; Tobin et al., 2017), augmenting these by including important synaptic circuits of the LNs, […] which constitute nearly 30% of the cells of the glomerulus, and much of its synaptic diversity.”

The new method does provide more information and more completion, but this is really an incremental increase.

We of course are the authors, but we do feel this comment of the reviewer is as incorrect as it is damaging. Together with the PNs and ORNs for glomerulus vA1v we report the connections of 56 local neurons. Previous reports from TEM, while themselves important additions, have been able to document very few if any connections of the LNs, which we provide based on improved FIB-SEM methods, and therefore claim that the increment to which the reviewer refers is in fact very large (56 additional entire cells, all of them inaccessible to TEM methods, and most of them synaptically diverse).

Because of these reasons, the novelty of the connectome is limited.

This may well be the reviewer’s perspective, but we suggest it is not based on the size of the connectome we have now generated.

Further, I don't have any intellectual issues with the analyses or the reconstructions, but I think this analysis would be more suited to a specialty journal.

This may likewise be true, but our report is, we suggest, at least equal in importance to the excellent recent eLife report by Tobin et al.

The chief advances of this connectome rest in a much more comprehensive map of projection neuron and local interneuron projections. These are useful, but the scope is very narrow. Further, in the presentation of each type of neuron, the manuscript is written in such a way that non-fly-aficionados will have a tough time determining what's important. What are the new concepts that can be gleaned from this reconstruction?

We believe that the comprehensive quantitative analysis of the extent of reciprocity between partner neurons is both novel and important, and has not been previously reported comprehensively for any other neuropile.

What do we learn about LN projections? What about PN projections? These areas are minimized (if not completely omitted), making it difficult to interpret why the new connectome is of value to neuroscience at large. These issues would greatly enhance the readability of the manuscript, though I still feel the core advance represented by the paper is better suited for another journal.

We learn that the LN projections are primarily reciprocal and formed with other LNs, which was not previously known, and we now emphasize this conclusion more strongly.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below.

Most importantly, the raw data needs to be available for inspection by reviewers before we can accept the paper. Please send a revised manuscript that addresses the rest of the issues (mostly minor) below, after the data are uploaded to a website that the reviewers can inspect. Thank you.

We agree unreservedly and Dr Steve Plaza, a co-author, has now provided the following information:

“FlyEM has and will continue to provide unprecedented and timely access to our datasets as evidenced by the public datasets on emdata.janelia.org. In this case, we needed some additional time as we are refactoring our infrastructure to provide ability to access the data in the most optimal ways. There are (only a) few good Big Data tools to access connectomic data, which is why FlyEM spends resources developing such cutting edge technology. While most publications believe a directory of images and dump of a large spreadsheet is sufficient, we feel this to be a grotesque way to release such large and beautiful data. We have not finished this refactoring of our infrastructure but have provided a relatively functional but not fully featured release at http://emdata.janelia.org/AL-VA1v. For more information go to the Readme file. For those desiring to see the data through a web viewer please visit:

http://neuroglancer-demo.appspot.com/#!%7B%22layers%22:%7B%22grayscalejpeg%22:%7B%22source%22:%22dvid://http://35.199.29.12:8600/ab6e610d4fe140aba0e030645a1d7229/grayscalejpeg%22%2C%22type%22:%22image%22%7D%2C%22segmentation%22:%7B%22source%22:%22dvid://http://35.199.29.12:8000/d925633ed0974da78e2bb5cf38d01f4d/segmentation%22%2C%22type%22:%22segmentation%22%7D%7D%2C%22navigation%22:%7B%22pose%22:%7B%22position%22:%7B%22voxelSize%22:%5B8%2C8%2C8%5D%2C%22voxelCoordinates%22:%5B3164.243896484375%2C7989.96728515625%2C3391%5D%7D%7D%2C%22zoomFactor%22:76.76400940694718%7D%2C%22perspectiveOrientation%22:%5B0.1018502488732338%2C-0.1967611163854599%2C0.00035121332621201873%2C0.9751468896865845%5D%2C%22perspectiveZoom%22:121.51041751873501%2C%22layout%22:%224panel%22%7D

and

http://neuroglancer-demo.appspot.com/#!%7B%22layers%22:%7B%22grayscalejpeg%22:%7B%22source%22:%22dvid://http://35.199.29.12:8600/ab6e610d4fe140aba0e030645a1d7229/grayscalejpeg%22%2C%22type%22:%22image%22%7D%2C%22segmentation%22:%7B%22source%22:%22dvid://http://35.199.29.12/d925633ed0974da78e2bb5cf38d01f4d/segmentation%22%2C%22type%22:%22segmentation%22%2C%22segments%22:%5B%22205788%22%2C%223%22%2C%22729439%22%5D%7D%7D%2C%22navigation%22:%7B%22pose%22:%7B%22position%22:%7B%22voxelSize%22:%5B8%2C8%2C8%5D%2C%22voxelCoordinates%22:%5B3752.35009765625%2C8052.685546875%2C3647.19970703125%5D%7D%2C%22orientation%22:%5B-0.02181488275527954%2C0%2C0%2C0.9997619986534119%5D%7D%2C%22zoomFactor%22:29.520611734760028%7D%2C%22perspectiveOrientation%22:%5B0.13834506273269653%2C0.5922386050224304%2C0.7721632719039917%2C0.18405957520008087%5D%2C%22perspectiveZoom%22:864.801499283442%2C%22layout%22:%224panel%22%7D

The first link shows a view similar to Figure 2, and the second reconstructed ORN, mPN1 and LN2L cells.

To access the data programmatically (more documentation forthcoming) consult documentation for github.com/janelia-flyem/dvid. In the future, we expect much better strategies for accessing the data and distributing it through cloud services, which will facilitate better scientific reproducibility, something that has been generally absent in most connectomic works to date.

We corrected an editing error in the legend to Figure 5.

We now also cite a recent report that has appeared since our initial submission, on the developmental origins of LNs, adding the following text: “possibly reflecting the three developmental modes of their origin, as residual larval LNs, as adult-specific LNs emerging before associated sensory and projection neurons, and as LNs that emerge after synaptic connections are established (Liou et al., 2018).”

In addition, we have added a new reference and citation to support this revision:

Liou et al., 2018.

Reviewer #1:

[…] 1) Since the authors emphasize in the Abstract that VA1v is a sexually dimorphic glomerulus, they should tell the readers more explicitly in the main text that the sample is from a female (right now the information is in the Materials and methods and figure legend), and what is the nature of sexual dimorphism (male VA1v has larger volume than female VA1v).

We indicated at the start of the Results section that we used a female Drosophila, and moved a short paragraph from the Materials and methods to the first paragraph of the Results, as follows: “a fruitless-positive glomerulus responsive to fly odour (Sakurai et al., 2013), that signals the sex pheromones cis-vaccenyl acetate and methyl laurate (Kurtovic et al., 2007; Dweck et al., 2013, and is significantly larger in male flies than in females (Kondoh et al., 2003; Stockinger et al., 2005).”

2) Subsection “Identified neurons from dense reconstruction in VA1v2”, paragraph three: most cells (should be changed to LNs) being panglomerular. On that note, none of the LNs in reconstructed LNs in Figure 4—figure supplement 3 appeared panglomerular, presumably because they were not fully reconstructed; the authors should state this explicitly, and how they categorize the LNs even though they were not fully reconstructed.

We now explicitly state that our reconstructions of LNs report only those portions in VA1v and additional regions sufficient to identify the soma and its axon, and as a result are therefore still partial.

3) Figure 7A: it would be useful to add an output arrow from PNs, the destination being "higher olfactory centers." Even though it is not part of the reconstruction, it will be useful to remind the readers where the information goes from the antennal lobe.

We have added an arrow to the figure and indicated its significance in the legend to this figure.

Reviewer #2:

The revision of Horne et al. is improved, the formatting as a resource paper serves the data much better. The text has improved and contains more quantifications. Synapse detection is properly described and convincing.

As to the availability of the raw image data at the review stage, I maintain the view that by eLife standards (and similarly for many other journals) the raw data should be made available to reviewers for inspection, not just promised to be available later. This is especially true for a resource manuscript, but applies more broadly. I would hope that with the infrastructural capabilities of Janelia, data from the FlyEM team could be rather spearheading such modern data transparency standards than circumventing them.

See our response above.

As to the reporting of quantifications, I am surprised that the authors choose NOT to report certain quantitative measurements. The authors state that the reconstruction comprised a total of "15.8 cm of neurite". Since the authors are free to put this number into context, as in the reply to my comment, I would encourage this number to be reported, not withheld from the reader.

We have now added this metric in subsection “Identified neurons from dense reconstruction in VA1v”.

Reviewer #3:

The revised version of Horne et al. is a much-improved manuscript. Overall, I am satisfied with the comments from the authors and the considerable revisions included therein. I think the adjustment to a "Tools and Resource" and the reframing of many of the conceptual advances from this manuscript are absolutely appropriate. I thank the authors for their consideration of most of the comments provided.

I will defer judgment to many of the mathematical and technical issues raised by the second reviewer, however, as these fall out of my purview.

One confusing section, however, is the third paragraph of subsection “Identified neurons from dense reconstruction in VA1v”. It seems like the goal is to highlight aspects of ORNs, but some of it seems to be discussing LNs. The LN cells identified by Chou et al., are panglomerular, but the wording makes it sound like the ORNs are pan-glomerular. Is this accurate? Could the paragraph be adjusted to improve clarity?

We agree: this was a word-processing error and the text now reads: A total of 51 ORNs originated in the ipsilateral antennal nerve with 56 that entered in the commissure from the contralateral lobe.

https://doi.org/10.7554/eLife.37550.024

Article and author information

Author details

  1. Jane Anne Horne

    Department of Psychology and Neuroscience, Life Sciences Centre, Dalhousie University, Halifax, Canada
    Contribution
    Data curation, Software, Formal analysis, Supervision, Methodology, Writing—original draft, Project administration, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9673-2692
  2. Carlie Langille

    Department of Psychology and Neuroscience, Life Sciences Centre, Dalhousie University, Halifax, Canada
    Contribution
    Data curation, Validation
    Competing interests
    No competing interests declared
  3. Sari McLin

    Department of Psychology and Neuroscience, Life Sciences Centre, Dalhousie University, Halifax, Canada
    Contribution
    Data curation, Validation
    Competing interests
    No competing interests declared
  4. Meagan Wiederman

    Department of Psychology and Neuroscience, Life Sciences Centre, Dalhousie University, Halifax, Canada
    Contribution
    Data curation, Validation
    Competing interests
    No competing interests declared
  5. Zhiyuan Lu

    1. Department of Psychology and Neuroscience, Life Sciences Centre, Dalhousie University, Halifax, Canada
    2. Janelia Research Campus, Howard Hughes Medical Institute, Virginia, United States
    Contribution
    Validation, Methodology, Ultramicrotomy
    Competing interests
    No competing interests declared
  6. C Shan Xu

    Janelia Research Campus, Howard Hughes Medical Institute, Virginia, United States
    Contribution
    Visualization, Methodology
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8564-7836
  7. Stephen M Plaza

    Janelia Research Campus, Howard Hughes Medical Institute, Virginia, United States
    Contribution
    Resources, Software, Methodology
    Competing interests
    No competing interests declared
  8. Louis K Scheffer

    Janelia Research Campus, Howard Hughes Medical Institute, Virginia, United States
    Contribution
    Software, Methodology
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3289-6564
  9. Harald F Hess

    Janelia Research Campus, Howard Hughes Medical Institute, Virginia, United States
    Contribution
    Visualization, Methodology
    Competing interests
    No competing interests declared
  10. Ian A Meinertzhagen

    1. Department of Psychology and Neuroscience, Life Sciences Centre, Dalhousie University, Halifax, Canada
    2. Janelia Research Campus, Howard Hughes Medical Institute, Virginia, United States
    Contribution
    Writing—original draft, Writing—review and editing
    For correspondence
    I.A.Meinertzhagen@Dal.Ca
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6578-4526

Funding

Howard Hughes Medical Institute (Janelia FlyEM)

  • Jane Anne Horne
  • Carlie Langille
  • Sari McLin
  • Meagan Wiederman
  • Zhiyuan Lu
  • C Shan Xu
  • Stephen M Plaza
  • Louis K Scheffer
  • Harald F Hess
  • Ian A Meinertzhagen

The funder (HHMI) provided technical support for study design, and data collection.

Acknowledgements

This work has been supported by FlyEM at Janelia Research Campus of Howard Hughes Medical Institute. We acknowledge the generous support and encouragement of Gerald Rubin and the following people also at Janelia; in particular Gary Huang, William Katz and Toufiq Parag, Lowell Umayam and Ting Zhao for IT support; and Pat Rivlin and Shinya Takemura for help with glomerulus identification. We acknowledge Nobuaki Tanaka (Hokkaido University, Sapporo) for help identifying cell types; and Mss Aya Shinomiya, Dorota Tarnogorska and Jola Borycz (Dalhousie) for additional proof-reading, and Asa Barth-Maron (Harvard University, Boston) for help with data analysis.

Senior Editor

  1. Eve Marder, Brandeis University, United States

Reviewing Editor

  1. Liqun Luo, Howard Hughes Medical Institute, Stanford University, United States

Reviewer

  1. Moritz Helmstaedter, Max Planck Institute for Brain Research, Germany

Publication history

  1. Received: April 14, 2018
  2. Accepted: October 31, 2018
  3. Accepted Manuscript published: November 1, 2018 (version 1)
  4. Version of Record published: November 13, 2018 (version 2)

Copyright

© 2018, Horne 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.

Metrics

  • 911
    Page views
  • 216
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

  1. Further reading

Further reading

    1. Neuroscience
    Anton Miroschnikow et al.
    Research Article
    1. Neuroscience
    Masahiro Yamashita et al.
    Research Article