A connectome of the Drosophila central complex reveals network motifs suitable for flexible navigation and context-dependent action selection

Abstract

Flexible behaviors over long timescales are thought to engage recurrent neural networks in deep brain regions, which are experimentally challenging to study. In insects, recurrent circuit dynamics in a brain region called the central complex (CX) enable directed locomotion, sleep, and context- and experience-dependent spatial navigation. We describe the first complete electron microscopy-based connectome of the Drosophila CX, including all its neurons and circuits at synaptic resolution. We identified new CX neuron types, novel sensory and motor pathways, and network motifs that likely enable the CX to extract the fly's head direction, maintain it with attractor dynamics, and combine it with other sensorimotor information to perform vector-based navigational computations. We also identified numerous pathways that may facilitate the selection of CX-driven behavioral patterns by context and internal state. The CX connectome provides a comprehensive blueprint necessary for a detailed understanding of network dynamics underlying sleep, flexible navigation, and state-dependent action selection.

Keywords: D. melanogaster; central complex; connectome; insect; navigation; neuroscience; sensorimotor; sleep.

Figure 1.. The central complex (CX) and accessory brain regions.

(A) The portion of the central brain (aquamarine) that was imaged and reconstructed to generate the hemibrain volume (Scheffer et al., 2020) is superimposed on a frontal view of a grayscale representation of the entire Drosophila melanogaster brain (JRC 2018 unisex template [Bogovic et al., 2020]). The CX is shown in dark blue. The midline is indicated by the dotted black line. The brain areas LO, ME, and SEZ, which lie largely outside the hemibrain, are labeled. (B) A zoomed-in view of the hemibrain volume, highlighting the CX and accessory brain regions. (C) A zoomed-in view of the structures that make up the CX, the ellipsoid body (EB), protocerebral bridge (PB), fan-shaped body (FB), asymmetrical body (AB), and paired noduli (NO). (D) The same structures viewed from the lateral side of the brain. (E) The same structures viewed from the dorsal side of the brain. The table below shows the abbreviations and full names for most of the brain regions discussedin this paper. See (Scheffer, 2020) for details. Anatomical axis labels: d: dorsal; v: ventral; l: lateral; m: medial; p: posterior; a: anterior.

Figure 1—figure supplement 1.. The central complex (CX) and additional accessory brain regions.

(A) Posterior view of the hemibrain volume shown in Figure 1A. (B) A zoomed-in view of the CX and additional accessory brain regions not shown in Figure 1. (C) Lateral view of (B).

Figure 1—figure supplement 2.. Fan-shaped body (FB) neurons tracts.

(A) Top and side (inset) views of the PDM1 to PDM4 cell clusters, corresponding to the DM1 to DM4 hemilineages, also known as w,x,y,z hemilineages in other insect species (Boyan and Williams, 2011; Izergina et al., 2009; Williams, 1975). These cell clusters encompass all FB types except the tangential FB neurons. (B–E) Lateral view of the FB neurons in the PDM clusters (top) and an EM cross-section of the bundle of processes connecting their somata to the FB (bottom). h∆ and v∆ neurons travel with the PFN neurons, but have neurites with much smaller diameters. FC, FS, and FR neurons travel in between the PFN and PFL neurons, and generally have small-diameter processes. Scale bar 5 μm.

Figure 1—figure supplement 3.. Main neurite diameter of central complex (CX) neurons.

Median diameter of the processes between the somata and main branchpoints of all CX neurons, grouped by type. Each point is a neuron, each x-coordinate a type. Note that there is some variability in the detection of the main branchpoint of neurons. (A) EB neurons (B) PB, NO and SA neurons (C) FB neurons (except FB tangentials) (D) FB tangential neurons.

Figure 2.. High-level schematic and an example sensorimotor pathway through the central complex (CX).

(A) The CX integrates information from multiple sensory modalities to track the fly’s internal drives and its orientation in its surroundings, enabling the fly to generate flexible, directed behavior, while also modulating its internal state. This high-level schematic provides an overview of computations that the CX has been associated with, loosely organized by known modules and interactions. (B) A sample neuron type-based pathway going from neuron types that provide information about sensory (here, visual) cues to neuron types within the core CX that generate head direction to self-motion-based modulation of the head direction input and ultimately to action selection through the activation of descending neurons (DNs). The neurons shown here will be fully introduced later in the article. Note that the schematic highlights a small subset of neurons that are connected to each other in a feedforward manner, but the pathway also features dense recurrence and feedback. (C) Ci-iii show three different views (anterior, lateral, dorsal, respectively) of individual,connected neurons of the types schematized in B.

Figure 2—figure supplement 1.. Selected central complex (CX) input, intra, and output neurons.

(A) Three different views (Ai: anterior; Aii: lateral; Aiii, dorsal, respectively) of selected individual neurons that provide input to the CX. (B) Same as (A), but for intra-CX connections. (C) Same as (A), but for CX output pathways.

Figure 3.. Quantitative impact of different levels of proofreading on neuronal connectivity in the ellipsoid body (EB).

(A) Morphological rendering of an example EPG neuron before and after dense tracing in the EB. Inset, zoomed-in view of part of the EPG arbors highlighting changes resulting from dense reconstruction. The neuron segmentation is in pink. One newly added fragment is colored in green and marked with a red star. Synapses to neurons that were initially identified are in orange. Synapses to neurons that were identified after dense tracing are in blue. These new additions often resulted from joining previously unidentified fragments to their parent neurons, which partner with the example EPG neuron. (B) Change in the number of input synapses from known neurons (left panel) and output synapses to known neurons made with selected EB neurons after dense tracing. Each neuron in this subset had at least 200 presynaptic sites in the EB for the left panel, 200 postsynaptic sites in the EB for the right panel, and at least a 10% change in known synapse numbers after dense tracing. The EB neurons are ordered by type and colored by supertype (see Materials and methods). Each colored dot represents a single neuron of the type indicated. Throughout, we analyze input and output connectivity separately. The example neuron shown in (A) is circled in black. (C) Comparison of the input connectivity of the neuron shown in (A) before and after dense tracing. Each point is the relative weight of a connection between that EPG and a single other neuron. Relative weight refers to the fraction of the inputs that comes from the given partner (see Materials and methods). The color denotes the type of the partner neuron. The gray line is a linear fit with 95% confidence intervals (the confidence interval is too small to be seen). The dashed line is the identity line. (D) Slope of the linear fits (similar to the one in C) with 95% confidence intervals for all neurons considered. Many confidence intervals are too small to be seen. The example shown in (A) is circled in black.

Figure 3—figure supplement 1.. Influence of the amount of change from tracing on fit results.

(A) Influence of the percentage change in the number of input synapses (left) and output synapses (right) made with known partners after dense proofreading (the same quantity as plotted in Figure 3B) on the slope of the fit for each neuron considered. (B) Influence of the percentage change in the number of input synapses (left) and output synapses (right) made with known partners after dense proofreading (the same quantity as plotted in Figure 3B) on the quality of the fit as measured with the corrected r² for each neuron considered. (C) Influence of the total number of input synapses (left panel) or output synapses (right panel) to known partners in the densely proofread dataset on the slope of the fit for each neuron considered. (D) Influence of the total number of input synapses (left panel) or output synapses (right panel) to known partners in the densely proofread dataset on the quality of the fit as measured with the corrected r² for each neuron considered.

Figure 4.. Differences in connectivity between compartments at different levels of tracing.

(A) Differences in connectivity between mirror-symmetric protocerebral bridge (PB) glomeruli. We compare glomeruli that are densely proofread (L4/R3) or not (R4/L3). R or L refer to the right or left half of the PB, respectively. Each half of the PB is made up of nine distinct glomeruli, with glomerulus 1 the most medial and glomerulus 9 the most lateral. (Ai) Sample PFNa neurons that each arborize in a single PB glomerulus. Two arborize in L3, and the other two in its mirror symmetric glomerulus, the densely proofread PB glomerulus R3. (Aii) Percentage increase in input connectivity (left) and output connectivity (right) to known partners for neuron types innervating single glomeruli between R4 and L4 or L3 and R3. Types were selected if they had neuron instances that innervate all four of these glomeruli, with each instance having at least an average of 20 synapses per glomerulus and at least 80% of their PB synapses in the given glomerulus. For a given type, circles denote the L3-to-R3 comparison and triangles the R4-to-L4 comparison. Few output comparisons can be made because most columnar neurons mainly receive input in the PB. (Aiii) Comparison of input connectivity for the type shown in (Ai) in R3 and L3. Each point is the relative weight of a connection between that type and another neuron type. The color denotes the supertype of the partner. The gray line is a linear fit with 95% confidence intervals. The dashed line is the identity line. (Aiv) Slope of the linear fit (similar to the one in Aiii) with 95% confidence intervals for all types considered. (B) Differences in connectivity between a densely proofread section of the FB (denoted as ‘column 3’, or C3) and other parts of the FB. (Bi) Sample hΔA neurons. One (in blue) has almost all of its output synapses in C3. The other four avoid C3 altogether. Output synapses are in orange. (Bii) Comparison of the average number of synapses to known partners per type between neuron instances innervating the heavily traced C3 and instances innervating other columns. Types are selected as having instances innervating C3 with at least an average of 200 synapses of a given polarity in the fan-shaped body (FB) and having at least 80% of those synapses in C3. They are compared to neurons of the same type with no synapses in C3 (e.g., the hΔA neurons in gray in Bi, circled in black). Plotted are the percentage increases in input connectivity (left) or output connectivity (right) to known partners for neurons in FB C3 versus other columns, by type. (Biii) Comparison of output connectivity for the type shown in (Bi) between neuron instances innervating C3 and instances avoiding C3. Each point is the average relative weight of a connection between that type and another neuron type. The color denotes the supertype of the partner type. The gray line is a linear fit with 95% confidence intervals, the dashed line is the identity line. (Biv) Slope of the linear fit (similar to the one in Biii) with 95% confidence intervals for the types considered. hΔA neurons are circled in black.

Figure 4—figure supplement 1.. Comparing protocerebral bridge (PB) connectivity in glomeruli with similar levels of tracing.

In the main figure, we compare glomeruli that are densely proofread (L4/R3) or not (R4/L3). R or L refer to the right or left half of the PB, respectively. The same analysis is done in this figure on glomeruli that have simillar of tracing, namely L5/R5 and L6/R6. (A) Similar to Figure 4Aiii, but for PB glomeruli L5-R5. (B) Similar to Figure 4Aiii, but for EPGs in L6-R6. (C) Similar to Figure 4Aii, but for glomeruli L5-R5/R6-L6. (D) Similar to Figure 4Aiv, but for glomeruli L5-R5/R6-L6. (E) Distribution of slopes for the fits for the equally traced (glomeruli 5 and 6) and the densely vs. sparser traced (glomeruli 3 and 4) conditions.

Figure 5.. Overview of input pathways to the central complex (CX).

(A) Schematic of input pathways, that is, pathways from non-CX brain regions, to the CX (see Figure 5—figure supplement 1B). (Ai) Input pathways to the ellipsoid body (EB) (red arrows), noduli (NO) (brown arrows), and protocerebral bridge (PB) (green arrows). (Aii) Input pathways to the fan-shaped body (FB) (blue arrows) and asymmetrical body (AB) (turquoise arrows). The width of the arrow is a qualitative indicator of the relative amount of input. (B) Input pathway classification for the EB, PB, and NO input neurons. Types are counted as inputs if they have at least 20 synapses of a given polarity outside of the CX and are the postsynaptic partner in at least one significant type-to-type connection outside of the CX. See Appendix 1—figure 3 for an explanation of pathway weight. The corresponding data for FB and AB input pathways is presented in the FB section (Figure 36—figure supplement 1C, Figure 40E).

Figure 5—figure supplement 1.. Additional information on input pathways to the central complex (CX).

(A) Input synapses to CX neurons in regions that are outside of the CX (but in the hemibrain volume), with the exception of synapses to fan-shaped body (FB) tangential neurons, which are shown in Figure 28—figure supplement 1B in the FB section. The synapses are color-coded by their supertype. Brain regions are colored as in Figure 1—figure supplement 1. (Ai) Posterior view. (Aii) Lateral view. (B) Total number of input synapses for CX input neurons grouped by input region outside of the CX and primary CX neuropil that is targeted by the input neuron.

Figure 6.. Overview of the anterior visual pathway and organization of the small unit of the anterior optic tubercle (AOTU).

(A) Schematic of the fly brain indicating the neuropils that are part of the anterior visual pathway, which starts at the medulla (ME) and projects via the AOTU and the bulb (BU) to the ellipsoid body (EB). The anterior visual pathway only passes through the smaller subunit of the AOTU (AOTUsu). The light blue shaded region indicates the coverage of the hemibrain dataset. (B) Morphological renderings of a subset of neurons that are part of the anterior visual pathway. (Bi) and (Bii) highlight two of several parallel pathways. (Bi) TuBu01 neurons tile a subregion of the AOTUsu and project to the BU, where they form glomeruli and provide input to ER4m neurons. ER4m neurons project to the EB. All TuBu01 and ER4m neurons from the right hemisphere are shown. (Bii) TuBu03 neurons also arborize in the AOTU, but these neurons target different regions of both the AOTU and BU and form larger arbors in the AOTU than do TuBu01 neurons. TuBu03 also form glomeruli in the BU, where they connect to ER3d_d. Inset shows the TuBu03 arbor in the AOTU as seen from the ventral position. (C) Connectivity graph of the inputs to TuBu neurons in the AOTU (significant inputs were selected using a 0.05 [5%] cutoff for relative weight). AOTU046 neurons are included here as they provide input to TuBu neurons in the BU (see Figures 7 and 8). TuBu are colored from pink to green based on the regions they target in the BU (see Figure 7). The dashed rectangle marks neuron types that also project to the contralateral AOTU. An asterisk marks TuBu types with likely tuning to polarized light based on their morphology and connectivity (see text). (D) Projections of the normalized synapse densities for medulla columnar types (Di) and each TuBu type (Dii) along the dorsal-lateral (left), the dorsal-anterior (center), and the anterior-lateral (right) plane, respectively. The synapse locations of MC61 and MC64 define two subregions of the AOTUsu, which are marked with a dashed line. Projections for the 10 TuBu types were split up in subplots for ease of readability. Types that arborize in similar regions were grouped together. Note the columnar organization of TuBu01 and TuBu06-10 as opposed to the more diffuse projections of TuBu02-05. (E) Projections of individual synapse locations from medulla columnar to TuBu neurons. (Ei). Synapses from MC61 onto TuBu01 neurons. Projections are shown along the same planes as in (D). Synapse locations are color-coded by the identity of the presynaptic neuron (MC61, top) or the postsynaptic neuron (TuBu01, bottom). The large, black-outlined dots indicate the center of mass for synapses from an individual neuron. Note that there are many more MC61 than TuBu01 neurons. (Eii). Same as (Ei), but for synapses from MC64 to TuBu03. ME: medulla, AOTU: anterior optic tubercle, AOTUsu: small unit of the AOTU, BU: bulb, EB: ellipsoid body.

Figure 6—figure supplement 1.. Connectivity motifs between MC and TuBu neurons in the AOTU.

(A) Quantification of the level of convergence from MC to TuBu neurons in the AOTU (see schematic on the right). Each dot represents the number of distinct MC neurons that give input to a given TuBu neuron. The total number of synapses that a TuBu neuron receives from all MC neurons of a given type is encoded in the dot size. Boxplots show interquartile range and medians. A single TuBu neuron receives input from 20 to 50 MC neurons of the primary MC input type. The dashed vertical line indicates 1:1 connections. (B) Quantification of the level of divergence in the connections from MC to TuBu neurons (see schematic on the right). Here a single dot represents the number of distinct TuBu neurons of a given type that a single MC neuron gives inputs to. Dot size represents total number of synapses from a MC neuron to the respective TuBu neuron and the dashed line indicates 1:1 connections.

Figure 7.. The bulb (BU) is more than just a relay station of visual information.

(A) Region arborization plot of cell types that innervate the BU, showing average pre- and postsynaptic counts by region. The following types were excluded upon manual inspection based on their relatively small number of synapses in the BU. ExR7, SMP238, CRE013, LHCENT11, LHPV5l1. The LNO neuron (LCNOp) is an input neuron to the noduli (NO), which will be described in a later section. (B) Morphological rendering of processes from one AOTU046 and one ExR5 neuron, which both arborize widely within the BU, as well as one TuBu01 and one ER4m neuron, which form a glomerulus (dashed circle). Different anatomical zones of the BU are labeled. (C) Projections of the normalized synapse densities for TuBu types (Ci) and ER types (Cii) along the dorsal-lateral (left) and the anterior-lateral (right) planes of the BU, respectively. Borders between different anatomical zones are indicated with dashed lines. For readability, synapse densities of TuBu and ER types that arborize in the BUs (top) versus the BUi or Bua (bottom) are displayed separately. All populations of neurons, except ER6, form glomeruli. (D) Neuron-to-neuron connectivity matrix of connections from TuBu neurons to ER neurons. Neurons were grouped according to type and, within a type, ordered such that most connections lie on a diagonal. The yellow boxes mark connections between neurons (putatively) tuned to polarized light. The blue box marks connections of sleep-related neurons. (E) Morphological rendering of the glomeruli formed by TuBu06 and ER5. (Ei). All TuBu06 and ER5 neurons. (Eii). Same as (Ei) but just TuBU06 neurons. (Eiii) Same as (Ei), but with only one ER5 neuron shown to highlight how a single ER neuron can target multiple glomeruli. Top view shown on the right. BUs: superior bulb, BUi: inferior bulb; BUa: anterior bulb; pBUi: posterior inferior bulb, aBUi: anterior inferior bulb.

Figure 7—figure supplement 1.. Connectivity motifs between TuBu and ER neurons in the BU.

(A) Quantification of the level of divergence in the connections from TuBu to ER neurons. Visualization as in Figure 6—figure supplement 1A. (B) Quantification of the level of convergence and from TuBu to ER neurons in the right BU. Visualization as in Figure 6—figure supplement 1B. (C) Scatter plot of synapse locations for TuBu06 neurons (Ci) and ER5 neurons (Cii) in the right BU. Synapses are color-coded based on body id. Synapse locations were projected onto projected onto the x/y axis (left) and z/y axis (right). For the z/y projection, only a thin slice as indicated by the dashed lines in the x/y plot is considered.

Figure 8.. Source of contralateral visual information.

(A) Morphological renderings of neurons in the anterior visual pathway together with neurons that connect to the contralateral anterior optic tubercle (AOTU) and/or bulb (BU). (Ai) TuBu09, ER2_d, and TuTuB_a. (Aii) TuBu01, ER4m, and AOTU046. (Aiii) TuBu03, ER3d_d, and ExR3. (B) Connectivity graph of TuBu and ER neurons as well as other neurons, ExR and AOTU046, that provide input to TuBu and ER neurons in the right BU. To highlight the organizational principles of connectivity in the BU, the nodes representing ER neurons are placed in an outer ring, those representing TuBu neurons (for brevity named TB here) in a middle ring, and nodes representing ExR and AOTU046 inside a central circle. (C) Projections of the normalized synapse densities of AOTU046 (Ci) and TuTuB (Cii) neurons in the right AOTU. Visualization as in Figure 6D. (D) Projections of the normalized synapse densities of AOTU046 and ExR neurons in the right BU. AOTU046 and ExR1 shown in (Di); ExR2, ExR3, and ExR5 shown in (Dii). Visualization as in Figure 7C. (E) Schematic of the projection pattern of a right AOTU046 neuron, piecing together innervations of the right AOTU046 neuron in the left hemisphere from the innervation of the left AOTU046 neurons in the right hemisphere, assuming mirror symmetric innervation patterns of the left and right neurons. Qualitative indication of input/output ratios per region is given based on region innervation plots shown in Figure 7A. (F) Schematic as in (E), but for the right ExR3 neuron.

Figure 8—figure supplement 1.. Connectivity of AOTU046 and ExR3 with TuBu neurons.

(A) Type-to-type connectivity matrices from AOTU046 to TuBu neurons for the right anterior optic tubercle (AOTU) and bulb (BU). Connectivity shown on per-type level. (B) Type-to-type connectivity matrices as in (A), but for ExR3 input to TuBu (left) and TuBu input to ExR3 (right) in the right BU.

Figure 9.. Mechanosensory input to the ellipsoid body (EB).

(A) Connectivity graph of paths from putative APN2 and APN3 to ER neurons. Only pathways with a minimal total weight of 1E-05 and a maximum length of 5 were considered. APN: AMMC projection neuron; WPN: wedge projection neuron; WLL: wedge-LAL-LAL neuron. (B) Hierarchical pie charts showing the fraction of inputs from various neuron types separated by input region for ER1_a (left), ER1_b (center), and ER3a_b (right) neurons. The fractions represent the average per type (computed only for neurons from the right hemisphere). Arrows highlight inputs from WPN (LHPV6q1) and WL-L (LAL138). (C) Morphological renderings of putative APN2 (SAD003, SAD004), APN3 (SAD077), WPN (LHPV6q1), and WL-L (LAL138) neurons as well as ER1_b and ER3a_b. (Ci). Frontal view. (Cii). Top view. (D) Morphological renderings of ER1_a (Di) and ER1_b (Dii). Only neurons with cell bodies in the right hemisphere are shown. Individual neurons are colored differently. (E) Projections of synapse locations of the neurons shown in (D). Synapses are colored by neuron identity (see legend). Larger, black-outlined dots mark the mean synapse position (center of mass) of each neuron. Synapses of individual ER1_b neurons separate along the dorsal-ventral axis (Eii), whereas synapses of ER1_a neurons are more spatially mixed (Ei).

Figure 9—figure supplement 1.. Organization of inputs to ER1 neurons in the lateral accessory lobe (LAL).

(A) Connectivity graph of paths from putative APN2 and APN3 to ER neurons. Only pathwayswith a minimal total weight of 1E-05 and a maximum length of 5 were considered. APN:AMMC projection neuron, WPN: Wedge projection neuron, WLL: Wedge-LAL-LAL neuron. (B) Hierarchical pie charts showing the fraction of inputs from various neuron types separatedby input region for ER1_a (left), ER1_b (center) and ER3a_b (right) neurons. The fractionsrepresent the average per type (computed only for neurons from the right hemisphere).Arrows highlight inputs from WPN (LHPV6q1) and WL-L (LAL138). (C) Morphological renderings of putative APN2 (SAD003, SAD004), APN3 (SAD077), WPN(LHPV6q1) and WL-L (LAL138) neurons as well as ER1_b and ER3a_b. Ci: Frontal view. Cii:Top view. (D) Morphological renderings of ER1_a (Di) and ER1_b (Dii). Only neurons with cell bodies inthe right hemisphere are shown. Individual neurons are colored differently. (E) Projections of synapse locations of the neurons shown in D. Synapses are colored by neuronidentity (see legend). Larger, black-outlined dots mark the mean synapse position (center of mass) of each neuron. Synapses of individual ER1_b neurons separate along the dorsal-ventral axis (Eii) whereas synapses of ER1_a neurons are more spatially mixed (Ei).

Figure 10.. Overview of the organization of the ellipsoid body (EB).

(A) Region arborization plot of neuron types that innervate the EB, showing average pre- and postsynaptic counts by region. For each neuron type, the number of cells from the right hemisphere is noted in the x-axis label. (B) Two-dimensional histograms of synapse counts of ER4m after projection onto the EB cross-sections along the dorsolateral (Bi), dorsoanterior (Bii), and anterior-radial axes (Biii). Note that for (Biii) anterior-radial cross-sections along the circumference of the EB were collapsed onto a single plane. The dashed line in (Bii) indicates one of the cross-sections that were collapsed in (Biii). The shapes of the anterior-radial cross-sections vary along the circumference of the EB, which is shown in Figure 10—figure supplement 4. (C) Normalized synapse densities of ring neurons onto the EB cross-section along the anterior-radial axes (see dashed outline in Bii, solid outline in Biii). (Ci). The synapse densities are color-coded by ring neuron type. (Cii). The synapse densities are color-coded by input regions. The dashed line indicates the outline of the EPG synapse density as seen in (D), for reference. (D) Same as in (Ci), but for columnar EB neurons. (E) Same as in (Ci), but for extrinsic ring (ExR) neurons. (F) Connectivity graph of neurons innervating the EB. Relative weight as measured on a type-to-type level has been mapped to the edge width. Gray shapes indicate groups of neuron types that likely share similar functional tuning based on existing literature. Only connections with a minimal relative weight of 0.05 (5%) are shown. Connections of a type to itself are omitted for simplicity.

Figure 10—figure supplement 1.. Ring neuron synapse positions.

Two-dimensional histograms of pre- and postsynaptic synapse counts of all ring neurons afterprojection onto the EB cross sections along the dorso-lateral (A), dorso-anterior (B) and anterior-radial axes (C).

Figure 10—figure supplement 2.. Ellipsoid body (EB) columnar neuron synapse positions.

Two-dimensional histograms of pre- and postsynaptic synapse counts of all EB columnarneurons after projection onto the EB cross sections along the dorso-lateral (A), dorso-anterior (B) and anterior-radial axes (C).

Figure 10—figure supplement 3.. Extrinsic ring (ExR) neuron synapse positions.

Two-dimensional histograms of pre- and postsynaptic synapse counts of all ExR neurons afterprojection onto the EB cross sections along the dorso-lateral (A), dorso-anterior (B) and anterior-radial axes (C).

Figure 10—figure supplement 4.. Synapse projections onto the anterior-radial axis along the circumference of the ellipsoid body (EB).

Normalized synapse densities of ring neurons projected onto the EB cross section along theanterior-radial axes for 8 wedge-shaped sections around the EB circumference are shown (see schematic for reference). Illustration of the position of cross sections on the upper right cornerof panel A. Synapse densities are color-coded by neuron type. (C) Ring neuron types. (D) EB columnar neuron types. (E) ExR neuron types.

Figure 10—figure supplement 5.. Morphological renderings of ring neurons.

Morphological renderings of ring neuron types and their primary regions of innervation: ER1_a, ER1_b, ER2_a, ER2_b, ER2_c, ER2_d, ER3a_a, ER3a_b, ERa_c, ER3a_d, ER3d_a, ER3d_b, ER3d_c, ER3d_d, ER3m, ER3p_a, ER3p_b, ER3w, ER4d, ER4m, ER5, ER6. Left column: Rendering of a single ring neuron from righthemisphere population with blue dots marking the location of postsynaptic sites and yellowdots those of presynaptic sites. Middle and right columns: Two views of the full population of ring neurons for each type.

Figure 10—figure supplement 6.. Morphological renderings of ring neurons.

Morphological renderings of all ring neuron types and their primary regions of innervation: ER2_d, ER3a_a, ER3a_b, ERa_c, ER3a_d. Visualization as in Figure 10—figure supplement 5.

Figure 10—figure supplement 7.. Morphological renderings of ring neurons.

Morphological renderings of all ring neuron types and their primary regions of innervation: ER3d_a, ER3d_b, ER3d_c, ER3d_d, ER3m. Visualization as in Figure 10—figure supplement 5.

Figure 10—figure supplement 8.. Morphological renderings of ring neurons.

Morphological renderings of all ring neuron types and their primary regions of innervation: ER3p_a, ER3p_b, ER3w, ER4d, ER4m. Visualization as in Figure 10—figure supplement 5.

Figure 10—figure supplement 9.. Morphological renderings of ring neurons.

Morphological renderings of all ring neuron types and their primary regions of innervation: ER5, ER6. Visualization as in Figure 10—figure supplement 5.

Figure 11.. Ring neuron to columnar connectivity.

(A) Neuron-to-neuron connectivity matrix for connections from ring neurons to EL and EPG neurons in the ellipsoid body (EB) on a single neuron level. The boxes on the right side are colored according to the ring neuron’s input region. (B) Morphological renderings of EL neurons and renderings of innervated regions of interest (ROIs). Note that EL neurons target a small region next to the GA, called the gall surround (GAs). (Bi). Single left hemisphere EL neuron with blue dots marking the location of postsynaptic sites and yellow dots those of presynaptic sites. (Bii). Full population of EL neurons. (C) Schematic illustrating variation in synaptic strength in ring neuron to EPG connections due to neural plasticity. Top: connectivity between ring neurons and EPG neurons. Bottom. Illustration of receptive fields (RFs) of single-ring neurons. (D) Connectivity matrix of ER4m inputs to EPG neurons that have been sorted and averaged according to the EB wedge they innervate.

Figure 11—figure supplement 1.. Wedge-specific modularity of inputs from ring neurons to EPG neurons.

(A) Neuron-to-neuron connectivity matrix for connections from ER4d neurons to EPG neurons in the ellipsoid body (EB), shown for the matrix that preserves (Ai) versus shuffles (Aii) the individual EPG neurons onto which individual ER4d neurons synapse (highlighted boxes). EPG neurons are ordered according to the EB wedge that they innervate. (B) Pairwise Pearson’s correlation measured between individual EPG neurons according to the pattern of their ER4d neuron inputs. Solid red boxes highlight clusters of EPG neurons that innervate the same EB wedge. Highlighted wedges in (Bi) are shown in (Bii). The modularity of the matrix in (Bi) measures whether individual EPG neurons are more correlated with those EPG neurons within the same wedge (solid boxes in Bi and Bii) than would be expected based on their average correlation with neurons across all wedges (dashed boxes in Bii). (C) Modularity of connectivity from different ring neuron types onto EPGs. Histograms show the distribution of modularity values computed for 1000 shuffled versions of each connectivity matrix (one example of which is shown in Aii). Insets show the correlation matrix of the measured (unshuffled) connectivity matrix; the modularity of this matrix is marked by a green line on the histogram. p-values indicate the fraction of shuffles that produced higher modularity than that of the measured connectivity matrix.

Figure 12.. Morphology analysis of ring neuron connectivity to EPG neurons.

(A) Skeleton of a single EPG (id. 1447576662) with the selected root point indicated in yellow. Inset: schematic indicating how the electrotonic distance from a point on the skeleton to the root point is calculated. The Euclidean metric is used to calculate the length of each segment (A–F) and ${\lambda }_{}$ (for = A, B, C, D, E, F) represents the length constants of the edges (see Materials and methods). (B–E) Localization of synaptic inputs to EPGs in the ellipsoid body (EB) along the dendritic tree, split by modality group. (B) The modality groups, the neuron types that fall into these groups, and the colormap that is used for modality groups for the rest of the panels in this figure. (C) Density of synapse locations onto EPGs in the radial vs. depth plane for all EPGs included in the analysis (n = 44). The black outline approximates the EB outline in this plane. Left: synapse locations are shown in gray (included here are synapses from partner types ER, ExR, PEG, PEN, EPG, EPGt). Overlaid contour lines indicate the distribution of the mean of the normalized electrotonic distance from the root. The yellow points indicate where the root points of the EPGs are located in this plane. Right: synapse locations from selected inputs separated and color-coded based on input modality (see B for input assignment to modality). (D) Cumulative density function (CDF) of the distribution of the normalized electrotonic distance to root for synapses separated by input modalities for a single EPG (id. 632544268). (E) Medians of the normalized electrotonic distance distributions grouped by modality. The connecting lines indicate the points corresponding to each individual EPG (n = 44), with the black line corresponding to the EPG whose CDFs are shown in (D).

Figure 12—figure supplement 1.. Additional information on the analysis of electrotonic distances of synapse locations of different ring neuron types onto EPG neurons.

(A) Synapse densities of each modality type separated to show where overlap occurs (most notably, between motor and mechanosensory). (B) Rank ordering of the input modalities determined via the location of their median for each EPG included in the analysis (n = 44). Group A indicates the most common (‘standard’) ordering. Most other groups are only a single permutation from group A (e.g., group B is one permutation, (2,3), from group A). The only exception to this is group F, which consists of a single neuron and is separated by three permutations from group A ((2,3), (1,2), and (4,5)). Schematic on the right shows where the neurons innervate the ellipsoid body (EB) for groups that contain the (2,3) permutation (which shows the largest separation in distributions of all the permutations, see C) from the standard ordering (groups B and F). Each shaded region indicates the arbor locations of one neuron, except for the regions indicated by the arrows, which contains arbors of two neurons. (C) Box plot showing the distance between the median of the modalities with consecutive rank order distributions. Neurons are included in the standard group if they do not show a permutation between the rank orders considered, and in the permuted group otherwise. The groups included in each boxplot are given by letter below the label of standard or permuted (note that these match the group labeling in B). (D) Same as Figure 12C (left), but for physical distance along arbor. (E) Same as Figure 12D, but for physical distance along arbor. (F) Same as Figure 12E, but for physical distance along arbor.

Figure 12—figure supplement 2.. Comparison of EPG synapse locations by ring neuron type.

Medians of the normalized electrotonic distance distributions grouped by neuron type within each modality group. The connecting lines indicate the points corresponding to each individual EPG (n = 44), with the black line corresponding to the EPG whose CDFs are shown in Figure 12D. (A) Motor group. (B) Mechanosensory group. (C) Ipsilateral visual and polarization sensitive group. (D) Contralateral visual and motor group. (E) Sleep group.

Figure 12—figure supplement 3.. Morphology analysis of ring neuron connectivity to EL neurons.

Same as Figure 12C, E and Figure 12—figure supplement 1B, but for synapses onto EL neurons instead of EPG neurons.

Figure 13.. Inter-ring neuron connectivity.

(A) Connectivity matrix for connections between ring neurons in the ellipsoid body (EB) on single neuron level. Connections between neurons of the same type are highlighted with black boxes. (B) Normalized contributions of different ring neuron types to EL and EPG neurons (left) vs. normalized contributions of EL neurons to different ring neuron types (right, EPGs make very few synapses to ring neurons, see Figure 13—figure supplement 1B). (C) Connectivity graph of connections between ring neurons. The graph nodes are arranged along the x-axis to group ring neuron types with putatively similar tuning. Vertices are ordered on the y-axis according to their rank-ordered connectivity strength to EPG neurons. Vertex size is scaled by the ratio of the sum of all outputs divided by the sum of all inputs. Only connections with a relative weight of at least 0.05 (5%) are shown. Furthermore, connections between neurons of the same type are not shown.

Figure 13—figure supplement 1.. Connectivity between ellipsoid body (EB) columnar neurons and ring neurons.

(A) Neuron-to-neuron connectivity matrix for connections from ring neurons to PEG and PEN neurons in the EB. (B) Same as (A), but for connections from all columnar neurons (EL, EPG, PEG, and PEN neurons) to ring neurons.

Figure 14.. Overview of extrinsic ring (ExR) neurons.

(A) Region arborization plot of all ExR types from the right hemisphere, showing average pre- and postsynaptic counts by region. Indicated below the plot is a qualitative categorization into three groups: mostly input to the ellipsoid body (EB) (blue), mostly output from the EB (pink), and mixed (black). (B) Similarity matrices (see Materials and methods) for ExR neurons based on all their inputs (Bi) and outputs (Bii). ExR-type labels are colored according to groups in (A). (C) Type-to-type connectivity matrix of ExR to EB columnar neurons (Ci) and EB columnar to ExR neurons (Cii).

Figure 14—figure supplement 1.. Morphological renderings of all ExR types: ExR1, ExR2, ExR3, and ExR4.

Some of the innervated neuropils are shown. The left column shows a single, right hemisphere ExR neuron for each type with presynaptic sites marked by yellow dots and postsynaptic sites marked by blue dots. The middle and right columns show morphological renderings of the complete population.

Figure 14—figure supplement 2.. Morphological renderings of ExR type. ExR4, ExR5, ExR6, ExR7, ExR8.

See Figure 14—figure supplement 1 for details on presentation.

Figure 14—figure supplement 3.. Comparison of inputs and outputs of extrinsic ring (ExR) neurons.

(A) Similarity matrices (see Materials and methods) as in Figure 14B but excluding inputs in the ellipsoid body (EB) (Ai) or outputs in the EB (Aii). (B) Stacked bar graph illustrating the fraction of inputs from and outputs to extrinsic ring (ExR) partners, grouped into supertypes and separated by brain region. Inputs and outputs are normalized per neuron type and brain region. The connectivity strength for inputs and outputs is measured by relative weight and output contribution, respectively. ExR-type labels are colored according to groups in Figure 14A.

Figure 14—figure supplement 4.. Neuron-to-neuron connectivity matrices for connections between extrinsic ring (ExR) and columnar neurons in the ellipsoid body (EB).

(A) Connections from ExR to columnar EB neurons. (B) Connections from columnar EB neurons to ExR.

Figure 15.. Extrinsic ring (ExR) connectivity motifs.

(A) Schematic explaining the ExR connectivity motif analysis, which compares connectivity within the ellipsoid body (EB) to connectivity outside the EB. The top row shows the three circuit motifs that were considered, and the bottom row their equivalent representation in a compact circular network plot. Here we compare connections from ExR to other EB neurons outside and inside the EB. We only consider out-of-EB pathways for ExR neurons. The out-of-EB pathways can be direct or indirect connections (pink arrows) to other EB neurons (in green). ‘Parallel connections’ occur when the source neurons also contact the pathway target neuron inside the central complex (CX) (in red). The ‘canonical feedback’ motif describes the case where the target of the pathway contacts the source type in the CX (in yellow). ‘Linked targets’ are neurons connected in the CX that are targets of the same neuron outside of the CX (in green). (B) Summary of motif prevalence across different ExR types. The colored circles represent the prevalence of each specific motif, whereas the gray circles represent the total number of all the motifs of the same type that could form given that type’s partners outside of the CX (normalized per type and motif). (C) Bar graph showing the contribution (measured by relative weight) of ExR partners in the EB to the observed connectivity motifs. The sum of the relative weights of each connection for an ExR to its partner is shown, separated by motif and partner type. (D) Morphological rendering of one ExR2_R (Di) and ExR3_R (Dii). Some of the innervated brain regions are shown in gray. Blue dots mark postsynaptic sites, and yellow dots mark presynaptic sites. (E) Graphical representation of connectivity motifs as depicted in (A) for ExR2_R (Ei) and ExR3_R (Eii). (F) Schematic relating groups of connectivity motifs in ExR2 (Fi) and ExR3 (Fii) to the anatomical location of the connections that are involved.

Figure 16.. EPGs connect the ellipsoid body (EB) to the protocerebral bridge (PB).

(A, Ai) A morphological rendering of two EPG neurons. Black dots are presynaptic sites. (Aii) A morphological rendering of the entire population of EPG neurons, color-coded by PB glomerulus. (B) Schematic showing where the EPG processes arborize in the EB and in the PB. The EPG neurons map the different locations around the ring of the EB to the right and the left PB. A fictive bump of activity in the EB will therefore split into both a right and a left bump of activity in the PB. Note that the bumps in the PB are slightly shifted with respect to one another due to the 22.5° offset between the right- and left-projecting wedges in the EB.

Figure 17.. PEN_a neurons connect the protocerebral bridge (PB) back to the ellipsoid body (EB), with a shift, forming feedback loops with the EPG neurons.

(A, top) PEN_a neurons on the left side of the PB send projections to the EB that are counterclockwise shifted with respect to the EB processes of their EPG inputs in the PB (see Figure 16). (Bottom) PEN_a neurons on the right side of the PB send projections to the EB that are clockwise shifted with respect to the EB processes of their EPG inputs in the PB. Black dots are presynaptic sites. (B) Schematic showing where the PEN_a processes arborize in the EB and in the PB. The processes in the right PB project to different locations in the EB than the processes in the matched glomerulus in the left PB. A bump of activity at the same location in the right and left PB will therefore form two shifted bumps of activity in the EB. The EB processes of the PEN_a neurons form eight equiangular tiles, each of which covers two of the EPG wedges. (C) Neuron-to-neuron connectivity matrix for EPG, PEN_a, PEN_b, and PEG neurons in the EB. The neurons are arranged according to their angular position in the EB (Ci) or according to their arrangement in the PB (Cii). Dotted lines are overlaid on the diagonal of the PEN to EPG quadrants to emphasize the offset in connectivity. Though not represented in the axis labels, multiple neurons often cover the same angle or arborize in the same PB glomerulus. (D) Neuron-to-neuron connectivity matrix for EPG, PEN_a, PEN_b, and PEG neurons in the PB. The EPG neurons directly connect to the PEN_a, PEN_b, or PEG neurons in glomeruli where they both have processes (L2–L8 for the PEN neurons and L1–L8 for the PEG neurons). The EPG neurons also occasionally synapse onto partners in neighboring glomeruli. As in (C), multiple neurons often cover the same glomerulus. (E) A force-directed network layout of the EPG and PEN_a connections. Weight refers to the number of synapses between partners.

Figure 17—figure supplement 1.. PEN_a and PEN_b connectivity.

(A) Type-to-type PEN_a and PEN_b input connectivity matrix in the protocerebral bridge (PB). (B) Type-to-type PEN_a and PEN_b input connectivity matrix in the ellipsoid body (EB). (C) Type-to-type PEN_a and PEN_b output connectivity matrix in the EB.

Figure 18.. EPGt neurons extend EPG-like connectivity.

(A) Morphological renderings of all EPGt neurons. The EPGt neurons arborize only in glomeruli L9 and R9 in the protocerebral bridge (PB) and, in the ellipsoid body (EB), their arbors line the canal at the bottom of the torus (Ai). A side view of the EB shows the position of EPGt processes in the EB (Aii). (B) Type-to-type connectivity matrix showing the inputs (Bi) and outputs (Bii) for the EPG and EPGt neurons in the PB. (C) Total number of presynaptic sites for the EPG and EPGt neurons by brain region. (D) Type-to-type connectivity matrix showing the inputs to the EPG and EPGt neurons in the EB. (E) Neuron-to-neuron input connectivity from the PEN_a and PEN_b neurons to the EPG and EPGt neurons in the EB (Ei) and outputs from the EPG and EPGt neurons to the PEN_a and PEN_b neurons in the PB (Eii).

Figure 19.. An overview of the protocerebral bridge.

(A) A diagram of the input (Ai) and output (Aii) pathways for the protocerebral bridge (PB). Connected brain regions include the ellipsoid body (EB), the inferior bridge (IB), the superior posterior slope (SPS), the posterior slope (PS), the crepine (CRE), the lateral accessory lobe (LAL), the fan-shaped body (FB), and the noduli (NO). (B) Morphological rendering of an EPGt neuron, which only arborizes in a single glomerulus in the PB. Yellow dots mark presynaptic site. Blue dots mark postsynaptic sites. (C) Morphological rendering as in (B) of an LPsP neuron, which has arbors throughout the PB. Yellow dots mark presynaptic site. Blue dots mark postsynaptic sites. (D) Region arborization plot for each neuron type that contains arbors in the PB. Neuron types that provide input to the PB are denoted by the dashed vertical boxes. The horizontal boxes at top indicate which neurons arborize in multiple glomeruli (filled gray boxes) and which arborize in single glomeruli (gray outline). (E) The average number of synapses per neuron in each PB glomerulus for each neuron type that contains arbors in the PB.

Figure 20.. E-PG to Δ7 connectivity forms a cosine-like profile.

(A) A morphological rendering of a Δ7 neuron that outputs to glomeruli R8, L1, and L9. (B) Type-to-type connectivity table from EPG and Δ7 neurons to themselves and to all other protocerebral bridge (PB) neurons. (C) Synaptic connectivity matrix between EPG and Δ7 neurons. (D, Di) The EPG to Δ7 synapses were added together within each EPG glomerulus for each Δ7 neuron. The total synapse counts were then averaged across all Δ7 neurons that have the same arborization pattern. (Dii) Each column in the EPG to Δ7 connectivity matrix in (Di) was circularly shifted to align the peaks. (Diii) The mean and standard deviation across aligned Δ7 neurons. A cosine fit to the mean profile is shown with the dotted red line. (E, left) A simulated von Mises bump profile in the ellipsoid body (EB) leads to von Mises profiles in the right and left PB. (Middle) The profile is multiplied by the EPG to Δ7 synaptic connectivity and then by the Δ7 to EPG connectivity to simulate the Δ7 input onto the EPG neurons. (Right) The normalized mean Δ7 to EPG input profile in the right and left PB, averaged across all possible bump positions and assuming a von Mises input. The standard deviation is shown in gray, and a cosine fit to the right or to the left mean is shown with the dotted red curve. (F) A simulated impulse profile to one glomerulus in the PB (Fi) and the resulting simulated activity profile in the Δ7 neurons (Fii). The procedure follows that used in (E). (G) The residual sum of squares error between a cosine and the mean Δ7 input to a given neuron type assuming either a von Mises (black outline) or an impulse (black fill) input from the EPG neurons. The error is averaged across the fits to the right and to the left PB.

Figure 20—figure supplement 1.. EPG and Δ7 neuron-to-neuron connectivity to PEG, PEN, PFGs, PFL, and PFR neurons.

Figure 20—figure supplement 2.. EPG and Δ7 neuron-to-neuron connectivity to PFN neurons.

Figure 20—figure supplement 3.. The Δ7 neurons get input in glomeruli that represent angles ~180° offset from their output glomeruli.

(A) Connectivity table between the EPG neurons and the Δ7 neurons in which the EPG synapses are combined for all EPG neurons that arborize in a given protocerebral bridge (PB) glomerulus. (B) The EPG neurons’ angles in the ellipsoid body (EB) mapped onto the PB glomeruli to which they project. (C) The mean input angle for each Δ7 neuron as a function of their output glomeruli. Each EPG neuron is assigned an angle as in (B), these angles are then weighted by the synapse count, as shown in (A), and the circular mean is then calculated. (D) Connectivity table between the Δ7 neurons and all EPG neurons that arborize in a given PB glomerulus, as in (A). (E) The difference between the mean input angle (from C) and the mean output angle for each Δ7 neuron’s left (L) or right (R) PB outputs. The output angles are calculated similarly to (C); each glomerulus is assigned an angle based on the EPG neurons that arborize there; the angles are weighted by the total synapse count; and the circle mean calculated. The dotted lines indicate 180° ± 11.25°.

Figure 21.. P6-8P9 neuron morphology and connectivity resembles that of the Δ7 neurons that arborize in the outer glomeruli.

(A) A morphological rendering of a P6-8P9 neuron. There are two P6-8P9 neurons on each side of the protocerebral bridge (PB), both of which are presynaptic in glomerulus 9. (B) Both Δ7_L8R1R9 (top) and P6-8P9 (bottom) neurons get input in PB glomeruli 5–9. Both output in PB glomerulus 9 (not shown here). P6-8P9 neurons have the highest number of input synapses in glomerulus 8, while the Δ7_L8R1R9 neurons have the highest number of input synapses in glomeruli 5 and 6. The left PB is not considered as one of the two P6-8P9_L neurons was not able to be fully connected due to a hot knife error. (C) The mean number of output synapses from each Δ7_L8R1R9 neuron (left) or P6-8P9_R neuron (right) in PB glomerulus R9. The color code is identical to that in (C).

Figure 22.. Protocerebral bridge (PB) input and inner neuron connectivity to output neurons.

(A) Schematic depicting the neuropil that bring input to the PB via columnar neurons that target single PB glomeruli. (B) Morphological renderings of single SpsP (Bi) and IbSpsP (Bii) neurons. (C) Type-to-type connectivity matrix from select PB inputs (IbSpsP, PFNv, and SpsP neurons) to PB output neurons. The SpsP neurons also connect to themselves. (D) Region arborization plot for the right IbSpsP neurons. The left IbSpsP neurons were not fully contained in the imaged volume. (E) Type-to-type inputs to the PFNv neurons, separated by neuropil region.

Figure 22—figure supplement 1.. Presynaptic partners of the IbSpsP neurons, outside of the protocerebral bridge (PB).

Neuron-to-neuron connectivity matrix showing the presynaptic partners of the IbSpsP neuronsin all regions outside of the PB.

Figure 23.. Neuromodulatory neurons in the protocerebral bridge (PB) output broadly across types.

(A) A morphological rendering of a putative octapaminergic P1-9 neuron. (B) Type-to-type connectivity matrix for the outputs of the P1-9 neurons and the putative dopaminergic LPsP neurons in the PB.

Figure 24.. The number of neurons per glomerulus varies for each columnar neuron type.

(A) Number of neurons per protocerebral bridge (PB) glomerulus for each of the PB-EB neuron types. (B) As in (A), for the PFGs, PFL, and PFR neurons. (C) As in (A), for the PFN neurons. The irregular PFNp_d neurons have minimal arborizations in the PB.

Figure 24—figure supplement 1.. Neuron types with more instances in a glomerulus have fewer total input or output synapses per region of interest (ROI).

(A) The total number of input and output synapses per ROI for the EPG neurons as a function of the protocerebral bridge (PB) glomerulus in which those neurons arborize. (B) The total number of input and output synapses per ROI for the EPG neurons as a function of the number of EPG neurons per glomerulus. The points were jittered by up to ± 0.2 to either side of their vertical centerline for ease of visualization. (C) For each neuron type, individual neurons were grouped according to how many neurons of that same type arborize in that neuron’s PB glomerulus. The mean total input (output) synapse count was then calculated and normalized by the mean total input (output) synapse count for the neurons with the fewest number of instances per glomerulus. This normalized total synapse count is displayed as a function of the numerosity factor. The numerosity factor is the ratio of the number of instances per glomerulus for the given glomerulus divided by the fewest number of instances across all glomeruli. The dotted line is the function y = 1/x. The points in the fan-shaped body (FB), noduli (NO), and PB input plots and the FB and NO output plots were jittered by up to ±0.05 to either side of their vertical centerline for ease of visualization.

Figure 25.. Overview of the noduli and illustration of separate compartments.

(A) Region arborization plot summarizing all cell types that innervate the noduli (NO), showing average pre- and postsynaptic counts by region. Boxes mark groups of neuron types that will be described in more detail in this section. (B) Connectivity graph of all neuron types in the right NO, highlighting clusters that approximately correspond to anatomically defined subcompartments (see inset). The line thickness corresponds to the relative weight of a given type-to-type connection. Only connections with a relative weight of at least 0.05 (5%) are shown. (C) Schematic of how the NO connect to other brain regions.

Figure 26.. Columnar neurons in the noduli.

(A) Morphological rendering of columnar neurons. (Ai) PFNd neurons. Left: two example neurons from the left and right PFNd population. Right: complete population of PFNd neurons. (Aii) PEN_a neurons. (B) Stacked bar graph illustrating the fraction of inputs and outputs to PFN and PEN partners grouped into supertypes and separated by brain region. Inputs and outputs are normalized per neuron type and brain region. The connectivity strength for inputs and outputs is measured by relative weight and output contribution, respectively. (C) Similarity matrices (see Materials and methods) for columnar noduli (NO) neurons based on their inputs in the NO (top) and protocerebral bridge (PB) (bottom). (D) Neuron-to-neuron connectivity matrix for columnar neurons in the right NO. Connections between neurons of the same type are highlighted with black boxes.

Figure 26—figure supplement 1.. Comparison of PFN inputs and outputs.

(A) Stacked bar graph illustrating the weight of inputs and outputs to partners grouped into supertypes and separated by brain region. (B) Connectivity matrix showing inputs to PEN and PFN types in the noduli (NO). Connectivity is measured on a type-to-type level. (C) Same as (B), but for inputs in the protocerebral bridge (PB).

Figure 27.. Comparison of LNO neurons, which provide input to columnar neurons.

(A) Stacked bar graph illustrating the fraction of inputs and outputs of LNO to partners grouped into supertypes and separated by brain region. Inputs and outputs are normalized per neuron type and brain region. The connectivity strength for inputs and outputs is measured by relative weight and output contribution, respectively. (B) Morphological rendering of LNO neurons. (Bi) GLNO, (Bii) LNO1, and (Biii) LCNOp. Note that LCNOp crosses the midline and arborizes in the contralateral noduli (NO). Additional morphological renderings. LCNOpm, LNO2, LN03, LNa. (C) Illustration of how similarity between LNO neuron types relates to their connectivity to columnar NO neurons. Left: dendrogram depicting the similarity between GLNO, LNO, and LCNO neuron types based on their inputs outside of the NO (i.e., excluding feedback connections from PFN or PEN neurons). The branch height in the dendrogram indicates the normalized distance between types within the similarity space. Right: connectivity from GLNO, LNO, and LCNO neurons onto columnar NO neurons, visualized as in the connectivity graph in Figure 25B. Note that LCNOp and LNO3 neurons project to the ipsilateral NO and therefore target the right-side population of certain PFN types. PFN types that receive inputs from ipsi- and contralateral LNO and LCNO types are highlighted with dashed boxes.

Figure 27—figure supplement 1.. Comparison of LNO inputs and outputs.

(A) Similarity matrices (see Materials and methods) for LNO neurons based on their inputs outside of the noduli (NO) (Ai) and within the NO (Aii). (B) Connectivity matrix showing all inputs (including those in the NO) to GLNO, LNO, and LCNO neurons. The matrix columns and row were rearranged based on clustering GLNO, LNO, and LCNO neurons on the basis of their inputs and the input types based on their outputs to GLNO, LNO, and LCNO neurons.

Figure 27—figure supplement 2.. Connectivity graph of paths from putative directionally tuned wind sensitive neurons (putative WPN neuron and WLL neuron) to any of the LNO neurons.

Only pathways with a minimal total weight of 1E-05 and a maximum length of 3 were considered. Given these criteria, we only found pathways to GLNO and LNOa. The pathway to GLNO goes through the EB via ER1_b.

Figure 28.. Fan-shaped body (FB) overview.

(A) Schematic showing the FB, its main associated input and output neuropil, and the general types of information thought to be conveyed. Here, the FB is divided into nine vertical columns defined by protocerebral bridge-FB (PB-FB) neurons (see Figure 29), which map the nine glomeruli in the left and right PB to columns in the FB, as indicated by the color of each glomerulus/column (see Figure 30). However, unlike in the PB, the number of FB columns is not rigidly set, but depends on cell type. In addition to columns, FB tangential cells divide the structure into nine horizontal layers. The ventral FB (layers ~ 1–6) receives columnar input from the PB while the dorsal FB (layers ~ 7–9) does not. (B) Morphological renderings of individual columnar neurons (shown in black; red circles are presynaptic sites) from each of the four broad columnar neuron classes. PB-FB-*, FX, vΔ, and hΔ (where X and * stand for an additional, neuron type-specific neuropil). Each class contains many distinct neuron types. The population of neurons comprising each neuron type innervates all columns of the FB, but in a layer-restricted manner. As shown next to each anatomical rendering, a neuron type’s morphology can be summarized by illustrating the location of dendritic (rectangle) and axonal (circles) compartments for the nine FB layers and any associated neuropil. Here, each neuron type is colored according to its class (see legend in D). (C) Same as in (B), but for 2 of the 145 types of FB tangential cells. (D) Schematic showing the innervation pattern of every FB columnar neuron type, each illustrated as in (B). Columnar neurons can be roughly grouped into four putative functional groups: those that convey information from outside the FB to specific FB layers (columnar inputs; subset of PB-FB-* neurons), those that convey information between layers of the FB (intra-FB columnar neurons; vΔ and hΔ neurons), those that convey information out of the FB (columnar outputs; PB-FB-* and FX), and those that could perform a mixture of these functions (input/output). Columnar inputs have axons in every FB layer they innervate, intra-FB columnar neurons have processes confined to the FB, and columnar outputs have dendrites (and very few axons) in every FB layer they innervate. Note that while some columnar types are grouped (e.g., PFNm and PFNp), these types can be distinguished by their connectivity both within and outside of the FB (e.g., PFNm and PFNp receive distinct noduli [NO] inputs). In addition, tangential cells innervating the superior protocerebrum (SMP/SIP/SLP), crepine (CRE), and/or lateral accessory lobe (LAL) (and additional structures) provide input to (left panel) and output from (right panel) specific FB layers. Tangential cells in many different layers send processes to the SMP/SIP/SLP, CRE, and/or LAL, but only consistently target these regions in most cell types in the layers that are shown. See Figure 28—figure supplement 1 for average pre- and postsynaptic counts by region and columnar neuron type.

Figure 28—figure supplement 1.. Fan-shaped body (FB) regional connectivity.

FB columnar neuron region arborization plots. Circle size indicates the number of synapses eachFB columnar neuron type (x-axis) makes in a given neuropil (y-axis). Circles are shadedaccording to polarity, with darker circles indicated the presence of mostly presynaptic sites. This data was used to construct schematic in Figure 28D.

Figure 29.. Most PB-FB-* neurons form nine columns in the fan-shaped body (FB).

(A) Morphological renderings of PFNp_a and PFNa populations, colored by column (C1–C9). Left schematic shows location of dendritic (rectangle) and axonal (circles) compartments for the nine FB layers and any associated neuropil. Right panels show zoomed-in views of the FB, revealing a nine-column structure. Notice that PFNp_a columns are clustered while PFNa columns tile the FB more evenly. Note also that neurons innervating the same column often share the same fiber tract. The blue and yellow arrows in the top-right panel mark columns C8 and C9, which are closely spaced but show a clear spatially offset, as shown in panel (C). (B) Morphological rendering of the 18 neurons composing the PFGs population. Schematic on left as described in (A). Left panel shows front view (note that not all cell bodies are visible in this view). Arbor width is variable between cells. In addition, there is more substantial overlap in the dorsal FB arbors. The nine columns defined by this cell type are therefore more distinguishable in the ventral arbors. Right panel grays the nine neurons that project to the right gall, revealing that each column comprises two neurons, one of which projects to the left and the other to the right gall-surround, and that these right- and left-projecting neurons alternate in the FB. This projection pattern breaks the nine columns into ~18 ‘demi-columns,’ one neuron per demi-column, with two exceptions (purple and gray arrows). The purple arrow marks a demi-column which lacks separation from adjacent demi-columns. Similarly, the gray arrow marks a demi-column containing two neurons, whereas all other demi-columns contain one neuron. Whether these are the result of wiring errors requires further investigation. (C) Top-down view showing every neuron’s median location for all individual neurons in the PFNp_a, PFNa, and PFGs populations. Notice that while PFNp_a forms nine clear clusters, PFGs tile space more evenly. The distinct clustering seen in the PFNp_a arbors is reflected by the unique, scalloped morphology of layer 1 of the FB. The arrow in the PFNp_a panel points to a neuron that innervates both C1/C2 (assigned to C2) and C9, which is why its synapse location ended up outside of either cluster. (D) Distribution of neuronal arbor widths for PFGs, PFN, PFL, and PFR neurons. As shown in the inset, the width (red line) of synaptic point clouds (black dots) from individual neurons was measured along a direction locally tangent to a line bisecting the FB layer (green). To account for differences in layer size, the raw width (red line) was normalized by dividing the length of the layer (green line). Each distribution was normalized to have a peak of 1. The vertical dashed line in the graph marks 1/9th of the layer width, the arbor width that would result from nine evenly spaced columns that have minimal overlap and collectively tile the layer. Notice that most neurons take up slightly less than 1/9th of the layer. Importantly, this measure is independent of the neuron’s column assignment. (E) Distribution of inter-column distance, expressed as a fraction of the layer width, as in (D). Inter-column distance was measured by calculating the distance between the mean location of pairs of neurons in adjacent columns (as shown in inset), normalized to the length of the layer.

Figure 29—figure supplement 1.. Columnar structure of PB-FB-* neuron types.

Population morphological renderings (top panels) and median neuron locations (bottompanels) for every PB-FB-* neuron type, with the exception of PFL1-3, which are shown later: PFGs, PFR_a, PFNa, PFNd, PFNm_a, PFNm_b, PFNp_a, PFNp_b, PRNp_c, PFNp_d, PFNp_e, PFNv, PFR_b. Median neuron locations are shown for the FB layer with the most synapses for eachneuron type. Neurons are colored by column (see legend). Note that most neuron types thatinnervate L1 most heavily show evidence for 9 clustered columns. Neuron types that innervatemore ventral layers tend to show less clear clustering. Unlike all other PFN neurons, PFNdneurons form 8 columns. At present, the PFNd neuron names reflect the 9 column scheme, butwill be changed to 8 columns in future database versions. The arrow in the PFNp_a panel pointsto a neuron that innervates both C1/C2 (assigned to C2) and C9, which is why its synapselocation ended up outside of either cluster. The neuPrint link for PFNa, PFNp_a, and PFNp_bdisplays two neurons per PB column.

Figure 30.. PB-FB-* neurons have type-specific phase shifts in fan-shaped body-to-protocerebral bridge (PB-to-FB) projections.

(A) PFGs and PFRa neurons connect PB glomeruli to FB columns with no phase shift. (Ai) Schematic of a PB-to-FB projection pattern with no phase shift. PB glomeruli and FB columns are colored according to anatomical phase. Based on EB-to-PB columnar neuron projection patterns (EPG neurons, see Figure 16), when a bump is centered at L5 in the left PB, a second bump will be centered between R5/R4 in the right PB (both marked in purple). With no phase shift in their projection pattern, neurons innervating R5/L5 both project to C5 in the FB. This pattern, repeated across glomeruli/columns (see Aiii), would bring the two bumps in the PB to approximately the same FB location. (Aii) Morphological renderings of single neurons innervating R5 and L5, from the PFGs (top panel) and PFR_a (bottom panel) populations. Neurons are colored according to their FB column. Notice that the R5/L5 neurons end up at matching locations (C5) in the FB. (Aiii) Graphs showing the projection pattern from PB glomeruli to FB columns for all neurons in the PFGs (top panel) and PFR_a (bottom panel) populations. R5 and L5 projections have been highlighted as in (Ai). Lines connecting PB glomeruli to FB columns are colored according to PB glomerulus (i.e., anatomical phase). Blue dots mark glomeruli R1 and L1, whose neurons project to the opposite hemisphere (GAL for PFGs; ROB for PFR_a) than the other neurons in their half of the PB, the functional significance of which is unknown. (B) PFN types have one-column contralateral phase shifts in their PB-to-FB projection pattern. (Bi) Schematic of a PB-to-FB projection pattern, as in (Ai), but now showing a one-column contralateral phase shift. Notice that R5 projects to C6, and L5 projects to C4. This pattern, repeated across glomeruli/columns (see Biii), would cause PB bumps centered at R5 and L5 to end up at different locations in the FB. PFN neurons do not innervate glomeruli R1 and L1, as indicated by the gray shading. (Bii) Morphological renderings of single neurons innervating R5 and L5, as in (Aii), but now for PFNp_a and PFNa. Notice that the R5 neurons project to C6 and the L5 neurons project to C4. (Biii) Graphs showing the projection pattern from PB glomeruli to FB columns, as in (Aiii), but for PFNp_a and PFNa. Edges beginning at R5 and L5 have been highlighted, as in (Bi). Lines are colored according to PB glomeruli (i.e., anatomical phase).

Figure 31.. Overview of vΔ and hΔ columnar structure.

(A) Vertical columnar interneurons – the vΔ neuron types – have individual neurons with processes centered around one fan-shaped body (FB) column. Schematic on left shows two schematized neurons with arbors centered on C3 and C6. (Ai) Morphological rendering of the vΔA_a population, along with their schematized innervation pattern. Individual neurons are colored by FB column (from C1 to C9). In addition to innervating the FB, vΔA_a neurons (and some vΔA_b) are unique among vΔ neurons in that they innervate an extra-FB area, the asymmetric body (AB). Also notice the high degree of overlap of processes in the dorsal FB and the messy columnar structure of the population. Inset to the left shows a ‘C0’ neuron, which has arbors in both C1 and C9. (Aii) Same as in (Ai), but for the vΔB population. As with all other vΔ types, these neurons have processes restricted to the FB and receive most of their input in ventral layers while sending most of their output to more dorsal layers. (Aiii) Same as in (Ai), but for the vΔH population. (B) Horizontal columnar interneurons – the hΔ types – have individual neurons with processes centered on two distant FB columns, as shown in the illustration for two generic hΔ neurons. In particular, each hΔ neuron has a dendritic compartment that is ~180° away from its axonal compartment (i.e., separated by half the FB’s width). Half of the population has dendrites in right FB columns and project to left FB columns, while the other half of the population does the opposite. Individual hΔ neurons are assigned to columns based on the location of their dendritic compartment. (Bi) Morphological rendering of the hΔK population, along with their schematized innervation pattern. Individual neurons are colored according to FB column, with paired columns given matching colors. To achieve the ~180° phase shift, all hΔ types form an even number of columns. In this case, 12 columns (marked with six colors). In addition to innervating the FB, hΔK neurons are unique among hΔ neurons in that they innervate an extra-FB area, the EB. (Bii) Same as in (Bi) but for the hΔA population, which also forms 12 columns. Like most hΔ neurons, hΔA receives most of its input in ventral FB layers and provides most of its output to more dorsal FB layers. (Biii) Same as in (Bi) but the for the hΔH population, which forms eight columns instead of 12. Note the highly columnar structure of hΔ neuron types compared to the vΔ neuron types from (Ai) to (Aiii).

Figure 31—figure supplement 1.. Columnar structure of vΔ neuron types.

Population morphological renderings (top panels) and median neuron locations (bottom panels) for every vΔ neuron type: vΔA_a, vΔA_b, vΔB, vΔC, vΔD, vΔE, vΔF, vΔG, vΔH, vΔI, vΔJ, vΔK, vΔL, vΔM. Median neuron locations are shown for layer 1, where most vΔ types have primarily dendritic arbors, as well as the dorsal layer containing the most synapses, where vΔ types have axonal arbors. Neurons are colored by column (see legend). Gray neurons indicate those vΔ neurons that project to both C1 and C9, which we refer to as C0. Instead of marking median neuron location in the bottom panels, the gray dots mark the median location of the two arbors. Note that most vΔ neuron types show a highly variable columnar structure.

Figure 31—figure supplement 2.. Columnar structure of hΔ neuron types.

Population morphological renderings (top panels) and median input and output arbor locations (bottom panels) for every hΔ neuron type: hΔA, hΔB, hΔC, hΔD, hΔE, hΔF, hΔG, hΔH, hΔI, hΔJ, hΔK, hΔL, hΔM. Median input (circles) and output (triangles) arbor locations are shown for the FB layer with the most synapses for each neuron type, except for hΔK, which has axons and dendrites in separate layers (so all layers were used). Neurons are colored by column in a way that preserves left- and right-projecting pairs (see legend). Note that hΔ neuron types make a variable number of columns and that some types show a tighter columnar structure than others.

Figure 32.. Overview of FX columnar structure.

(A) FX neurons types all have a vertical morphology, with processes centered around one fan-shaped body (FB) column. Schematic on left shows two schematized neurons with arbors centered on C3 and C6. (Ai) Morphological rendering of the FR1 population, along with their schematized innervation pattern. Individual neurons are colored by FB column (from C1 to C9). In addition to innervating the FB, FR types innervate the ROB. (Aii) Same as in Ai, but for the FS1A population. FS types innervate both the FB and the SMP/SIP/SLP. (Aiii) Same as in Ai, but for the FC1E population. FC types innervate both the FB and the CRE.

Figure 32—figure supplement 1.. Columnar structure of FR and FS neuron types.

Population morphological renderings (top panels) and median neuron locations (bottom panels) for every FR and FS neuron type: FR1, FR2, FS1A, FS2, FS3, FS4A, FS4B, FS4C. Median neuron locations are shown for the FB layer containing the most synapses for each neuron type. Neurons are colored by column (see legend). Note that FR1 and FR2 are each composed of 18 neurons, with 2 neurons per column. Note also that some FS types, such as FS1A, show evidence for 9 clustered columns.

Figure 32—figure supplement 2.. Columnar structure of FC neuron types.

Population morphological renderings (top panels) and median neuron locations (bottom panels) for every FC neuron type: FC1A, FC1B, FC1C, FC1D, FC1E, FC1F, FC2A, FC2B, FC2C, FC3. Median neuron locations are shown for the FB layer containing the most synapses for each neuron type. Neurons are colored by column (see legend).

Figure 33.. Fan-shaped body (FB) columnar type to columnar type connectivity.

(A) The type-to-type connectivity between FB columnar neuron types arranged in a three-layer network diagram. FB inputs are shown at far left while FB outputs are shown at far right. Neuron nodes are color-coded by that neuron’s class. Only connections where most of the presynaptic neurons connect to a postsynaptic neuron of the given type are shown (more than 2/3 of the columns must connect across types). (B) The number of steps between columnar FB inputs and columnar FB outputs through other columnar FB neurons. (Bi) While PFN neurons directly connect to a few of the FB columnar output neurons in the FB (top), the pathways between PFN neurons and columnar outputs are often longer, traveling through one (middle), two (bottom), or more intermediate columnar neurons. (Bii) Direct (top), two-step (middle), and three-step (bottom) connections between PFN and FB columnar output neurons are shown in black. (C) Neuron-to-neuron connectivity matrix for the PFNa, FC1, and PFL1 neurons. Type-to-type connections between these neurons are shown below the dotted horizontal line in (A).

Figure 33—figure supplement 1.. Type-to-type connectivity matrix between fan-shaped body (FB) columnar neurons.

(A) Type-to-type connectivity matrix for the FB columnar neurons. Data is the same as that in Figure 33A. Neuron-type labels are color-coded by that neuron’s class. FB inputs are noted on the y-axis, while FB outputs are noted on the x-axis. Only connections where most of the presynaptic neurons connect to a postsynaptic neuron of the given type are shown (more than 2/3 of the columns must connect across types). (B) The number of type-to-type connections as a function of the percentage of presynaptic or postsynaptic neurons that appear in the neuron-to-neuron connectivity matrix between types. The dotted vertical line denotes the 2/3 (66.7%) threshold used in (A).

Figure 33—figure supplement 2.. Clustering by upstream and downstream partners.

(A) Hierarchical tree showing the similarity of different fan-shaped body (FB) columnar neuron types based on their FB columnar downstream (Ai) or upstream (Aii) partners. The dotted line shows the cutoff of 0.8 that was used to form the clusters shown in (C). (B) Cosine distance similarity matrix for columnar FB neuron downstream (Bi) and upstream (Bii) partners. (C) Each neuron type is linked to its downstream and upstream cluster. The thickness of each edge denotes the number of types within the given connection. The color denotes the supertype. The dotted boxes emphasize neuron types that fall into the same upstream and downstream clusters.

Figure 33—figure supplement 3.. The vΔF, G, H, and I subnetwork.

(A) Morphological renderings of the vΔF, G, H, and I neurons. These neurons connect to common upstream and downstream partners, forming a subnetwork. (B) Type-to-type connectivity matrix. Common input and output partners of the vΔF, G, H, and I neurons are highlighted in gray.

Figure 34.. Protocerebral bridge to fan-shaped body (PB-FB) projection patterns determine FB neuron’s phase shift and directional tuning.

(A) Schematic of a PB-to-FB projection pattern showing the one-column contralateral phase shift employed by PFN types, as in Figure 30B. (B) Graphs showing the projection pattern from PB glomeruli to FB columns for all neurons in the PFNa (top panel) and PFNp_a (bottom panel) populations, as in Figure 30B. (C) Connectivity between PFNa (top panel) or PFNp_a (bottom panels) neurons and two of their downstream partners within the FB. Notice that PFN neurons that arborize in glomeruli R5 or L5 connect with distinct columns in the FB, consistent with their PB-FB phase shifts. (D) Scatter plot showing the estimated directional tuning of FB neurons as a function of their medial-lateral position. For every v∆, h∆, or FX neuron postsynaptic to a PB-FB type, directional tuning was estimated by assigning angles to PB-FB neurons according to the PB glomerulus they innervate and by taking a circular mean across all angles inherited by the postsynaptic FB neuron, weighted by connection strength (Lyu et al., 2020; Figure 34—figure supplement 1, and Materials and methods). Medial-lateral position was normalized from 0 (right border to FB) to 1 (left border of FB) to account for the varying width of the FB layers occupied by each postsynaptic type. (E) Anatomical phase shift for PB-FB neuron types. Each circle is an estimated phase shift from the presynaptic PB-FB type to one of its postsynaptic types (v∆, h∆, or FX). Phase shifts were estimated across all postsynaptic neurons of a type individually and the circular mean was taken as the type average (black line). Note that PFR_b, PFNp_a, and PFNp_d types were excluded from this analysis due to inconsistent downstream connectivity (Figure 35) or because they exclusively target h∆ types on both axonal and dendritic compartments (Figure 37), both of which complicated phase shift estimates (see Materials and methods). (F) Histograms of PFN phase shift magnitude across all postsynaptic FB neurons (v∆, h∆, or FX), colored according to whether the postsynaptic FB neurons sample from presynaptic PFN neurons from two glomeruli (black) or from presynaptic neurons from more than two glomeruli (red). For individual neurons to have a 90° phase shift, they must sample from presynaptic PB-FB neurons that innervate at least two PB glomeruli (see Figure 34—figure supplement 1B).

Figure 34—figure supplement 1.. Estimating protocerebral bridge-fan-shaped body (PB-FB) phase shifts and directional tuning of FB neurons.

(A) Schematic showing angular assignments of PB glomeruli based on the projection pattern of EPG neurons from the EB to the PB. Note that corresponding glomeruli in the left and right PB have a 22.5° difference in their preferred directional tuning (Figure 16), consistent with recent physiological estimates (Lyu et al., 2020). R9 and L9, which do not receive direct EPG input, were assigned angles that preserved the 45° sampling interval in the left and right PB, even though their EPGt inputs suggest a slightly different directional tuning (see Materials and methods). (B) Schematics showing several commons ways in which FB neurons (v∆, h∆, or FX) sample from presynaptic PFN neurons to generate phase shifts that are ~90°. Each neuron’s phase shift is computed as the phase difference between the average angles inherited from the left PB population and the angles it inherits from the right PB population. In taking the average from the left/right populations, the circular mean is weighted by connection strength from each presynaptic neuron. As shown in the left and middle panels, if an FB neuron samples from PB-FB neurons that innervate only two glomeruli (one on the left, one on the right), its phase difference will either be 112.5° or 67.5° since no left-right pair of PB glomeruli are separated by 90°. Instead, if a postsynaptic neuron is to have a 90° phase shift, it must sample from more than two glomeruli, as shown in the right panel. (C) Histogram of PB-FB phase shift magnitude across all FB neurons (v∆, h∆, or FX) postsynaptic to PFNa (let panel) and PFNp_c (right panel). Vertical lines mark phase shifts generated in ways similar to those shown in (B).

Figure 35.. Right and left PB-FB-* populations target the same FB neuron types and neurons.

(A) Schematic showing one potential mechanism – type-specific targeting by left and right PB-FB-* populations – by which activity from the left and right PB could propagate through separate FB channels. This model predicts that PB-FB-* neurons from the left and right PB should target distinct downstream neuron types in the FB. (B) Scatter plot showing the average input from left (x-axis) and right (y-axis) PB-FB-* neurons onto downstream neuron types. Each circle is a downstream neuron type, and circles are colored according to the upstream PB-FB-* type (see legend). If the model from (A) were true, some points should lie along the x and y axes, indicating specific input from the left or right PB populations. Instead, every downstream type receives approximately equal input from left and right PB populations, ruling out the model from (A). (C) Schematic showing a second potential mechanism – demi-column-specific targeting by left and right PB-FB-* populations – by which activity from left and right PB could propagate through separate FB channels. This model predicts that individual neurons in a downstream population should receive input from the left or the right PB population (high ‘lateralization’), but not both. (D) Scatter plot showing the average input from left (x-axis) and right (y-axis) PB-FB-* neurons onto individual neurons in downstream populations. Each circle is a downstream neuron, and circles are colored according to the upstream PB-FB-* type (see legend). If the model from (C) were true, all neurons (circles) in a downstream population would lie along the x- or y-axis, indicating specific input from the left or right PB population. While some points do lie along the axes, most circles lie along the diagonal, suggesting roughly equal input from the left and right PB populations, similar to (B). (E) Scatter plot showing left/right lateralization (y-axis. the proportion of downstream neurons that receive input from the left or the right PB but not both) according to the connection consistency (x-axis. the proportion of neurons in a downstream neuron type targeted by a PB-FB-* neuron type). The size of circles indicates connection strength. Each circle is a downstream neuron type targeted by an upstream PB-FB-* type. The model from (C) predicts that points should lie in the upper-right portion of the plot, indicating a strong connection that is highly lateralized. instead, only weak and inconsistent connections show lateralization. With few exceptions (e.g., PFNd-to-PFNd), strong and consistent connections have downstream neurons that receive input from both left and right PB populations (low lateralization). This rules out the model from (C). Note that a small jitter was introduced so that overlapping points could be resolved. Arrow marks PFNd to PFNd connectivity, a connection that is fairly strong and lateralized.

Figure 36.. Overview of the asymmetric body (AB).

(A) Morphological renderings of the vΔA_a neurons, which arborize in the fan-shaped body (FB) and in the AB. They are columnar in the FB, with columns 1–5 projecting into the right AB and columns 6–9 projecting into the left AB. (Inset at right) The right AB is noticeably larger than the left AB. (B) Region arborization plots for each neuron type that contains arbors in the AB. (C) Renderings of FS4A (Ci) and FS4B (Cii) neural populations. These neuron types are columnar and receive input in both the AB and the FB and output to the superior medial protocerebrum (SMP). (D) Type-to-type connectivity matrix for the right (Di) and left (Dii) AB. The smaller left AB has fewer types that make significant connections. (E) The mean number of downstream (top) and upstream (bottom) synapses in the right (cyan) or left (red) AB by their FB column of origin for the columnar FB-AB neurons. (F) Type-to-type connectivity matrix for the downstream partners of the columnar FB-AB neurons in the FB.

Figure 36—figure supplement 1.. Additional asymmetrical body (AB) connectivity.

(A) Input pathway classification for the AB and non-tangential fan-shaped body (FB) input neurons. Types are counted as inputs if they have at least 20 synapses of a given polarity outside of the central complex (CX) and are the postsynaptic partner in at least one significant type to type connection outside of the CX. See Appendix 1—figure 3 for an explanation of pathway weight. (B) As in Figure 36E for the vΔA_a neurons, but now with individual neurons on the x-axis. (C) As in Figure 36F, but now for the upstream partners of the columnar FB-AB neurons.

Figure 37.. The intra-fan-shaped body (intra-FB) columnar network is built from a small number of circuit motifs.

(A) FB columnar neurons can be divided into vertical and horizontal types. Throughout the figure, vertical types are marked in maroon and horizontal types are marked in dark blue. Note that hΔ neurons are named according to the column containing their dendritic arbor, which impacts the connectivity matrix structure, as shown in the examples in (B). Vertical and horizontal neurons give rise to four connection types. vertical-to-vertical (V to V), vertical-to-horizontal (V to H), horizontal-to-horizontal (H to H), and horizontal-to-vertical (H to V). (B) Three columnar-to-columnar connectivity motifs generated by three circuit motifs. Top panels show circuit motifs that generate the corresponding column-to-column connectivity matrix shown in the bottom panels. The middle panels show how excitatory or inhibitory connections would impact bump phase. In each circuit diagram, all presynaptic columns are marked with hexagons, while postsynaptic columns can be dendritic (squares), axonal (circles), or contain multiple hΔ neurons with either dendritic or axonal arbors (rounded rectangles). See legend in Figure 37—figure supplement 1 for details. Circuit motifs are shown with ellipsis (…) to indicate variable column numbers, while connectivity matrices and bump change diagrams are shown with the nine-column pattern typical of most FB columnar neurons. (Bi) Motifs that generate a diagonal column-to-column connectivity matrix. Excitatory connections could pass the bump to a second layer while maintaining its phase, while inhibitory connections could shift the bump’s position by 180°. (Bii) Motifs that generate a shifted column-to-column connectivity matrix. Excitatory connections would shift the bump by 180° while inhibitory connections would maintain its phase (bottom panel). (Biii) Motifs that could produce a column-to-column connectivity matrix with two diagonal bands. Excitation and inhibition could produce a double-bump pattern, as shown in the bottom panel. Alternatively, if the axonal compartment receives inhibitory input and the dendritic compartment receives excitatory input, a single bump would be preserved (not shown). (C) Scatter plot showing that FB connectivity matrices can be grouped into one of the three motifs. Each circle in the scatter plot marks the location of a single connectivity matrix in principal component space. Briefly, each column-to-column connectivity matrix was coerced into a nine-column scheme, binarized, and transformed into a vector. Principal component analysis (PCA) was performed using a matrix containing all connectivity vectors (n = 903 connectivity matrices), and each vector was projected onto the largest two PCs (PC1 and PC2). Circles are colored according to pre-to-post connection type (see legend). Note that the large majority of connectivity matrices correspond to motifs 1 and 2 (diagonal point clouds), whose orthogonality is preserved in PC space. Points lying off these diagonals largely correspond to motif 3. The three large circles outlined in red correspond to the connectivity matrices in the bottom panels of (B).

Figure 37—figure supplement 1.. Detailed description of intra-fan-shaped body (intra-FB) columnar connectivity motifs.

(A) FB columnar neurons can be divided into vertical and horizontal types. Throughout the figure, vertical types are marked in maroon and horizontal types are marked in dark blue. hΔ neurons are named according to the column containing their dendritic arbor, which impacts the connectivity matrix structure, as shown in (B). (B) Same as in Figure 37A. (C) Same as in Figure 37B, but showing eight circuit motifs that could generate the corresponding connectivity matrices. The additional circuit motifs shown here occur more rarely that those shown in Figure 37B. (D) Same as in Figure 37C, but now showing eight connectivity matrices, each of which corresponds to one of the eight circuit motifs from (C).

Figure 37—figure supplement 2.. Principal component analysis of fan-shaped body (FB) columnar connectivity.

(A) Plot showing the proportion of variance accounted for by the first 10 principal components, ordered from highest to lowest. Note that the first two components account for greater than 50% of the variance and subsequent components much less. (B) Matrices showing the first six principal components. The proportion of variance explained by each is listed above each matrix. Note that the first two principal components do not correspond to motif 1 or motif 2 (Figure 37Bi and Bii). Instead, the first component is composed of two diagonal bands with positive and negative values. The second component is a rectified version of the first principal component. Linear combinations of PC1 and PC2, with appropriate weighting, produce column-to-column connection matrices that correspond to motif 1 and motif 2.

Figure 38.. Δ neurons in the fan-shaped body (FB) preferentially take input in lower FB layers and output to upper FB layers.

(A) All presynaptic sites for all the PFN neuron types. (B) Morphological renderings of vΔF (Ai) and hΔI (Aii) neurons. Presynaptic sites are shown in yellow while postsynaptic sites are shown in blue. Both types have postsynaptic sites throughout their arbors, but their presynaptic sites output in their upper layer FB arborizations. (C) All postsynaptic (left) and presynaptic (right) sites for all the vΔ (top) or hΔ (bottom) neurons. Postsynaptic sites are visible in the lower FB layers while presynaptic sites are restricted to the upper layers of the FB. (D) Region arborization plots for each vΔ and hΔ type.

Figure 39.. PFL neurons, a major fan-shaped body (FB) output, have type-specific phase shifts in protocerebral bridge-to-fan-shaped body (PB-to-FB) projections.

(A) PFL2 neurons have a four-column (~180°) PB-FB phase shift and bilateral LAL projections. (Ai) Schematic of a PB-to-FB projection pattern with a four-column phase shift, as shown for the R5 and L5 glomeruli. PB glomeruli and FB columns are colored according to anatomical phase, which indicates matching bump locations. (Aii) Morphological renderings of PFL2 neurons innervating R5 and L5. Neurons are colored according to their FB column. Notice that R5 and L5 neurons end up at C1 and C9, respectively. R5/L5 have been given asterisks because individual PFL neurons can innervate multiple PB glomeruli (in this case, R4/R5 and L4/R5). (Aiii) Graph showing the projection pattern from PB glomeruli to FB columns for all neurons in the PFL2 population. R5 and L5 projections have been highlighted as in (Ai), and edges are colored according to PB glomerulus. Unlike all other PB-FB-* neurons, the PFL2 population should only inherit one bump in the PB since the neurons sample from an ~360° region of PB space, split between left and right halves (R5–R1 and L1–L5). (Aiv) Functional graph showing the mapping between phases in the PB (top row) and phases in the FB (bottom row). Circles are colored by anatomical phase (legend). With one exception (R1), every PB glomerulus connects to a FB column with an ~180° phase shift. (B) PFL1 neurons have a one-column (~45°) ipsilateral phase shift and project to the contralateral LAL. (Bi) Same as in (Ai), but schematizing the one-column ipsilateral phase shift of PFL1 neurons. (Bii) Same as in (Aii), but for two PFL1 neurons. Notice that the R5 neuron projects to C4, and the L5 neuron projects to C6. (Biii) Similar to (Aiii), but for PFL1. Black and gray arched lines indicate groups of glomeruli that project to the right or left LAL, respectively. R1 and L1 are marked with blue dots because they project to the ipsilateral LAL, unlike the other neurons in the population. (Biv) Similar to (Aiv), but for PFL1. Here, instead of dividing glomeruli by their left vs. right PB innervation (as in Biii), glomeruli are grouped by whether they project to the left LAL (top row, gray outlined circles) or the right LAL (bottom row, black outlined circles), and sorted by anatomical phase. With the exception of R7 and L7, each glomerulus projects to FB columns with a one-column (~45°) ipsilateral phase shift. (C) PFL3 neurons have a two-column (~90°) ipsilateral phase shift and project to the contralateral LAL. (Ci) Same as in (Bi), but schematizing the two-column ipsilateral phase shift of PFL3 neurons. (Cii) Same as in (Bii), but for two PFL3 neurons. Notice that the R5 neuron projects to C3, and the L5 neuron projects to C7. (Ciii) Same as in (Biii), but for PFL3. R1/R2 and L1/L2 are marked with blue dots because these glomeruli contain neurons that either project to the contralateral LAL (like most of the population) or project to the ipsilateral LAL (unlike most of the population). (Civ) Same as in (Biv), but for PFL3. Notice that every neuron that projects to the left LAL (top row) and right LAL (bottom row) samples FB columns with a two-column (~90°) phase shift.

Figure 39—figure supplement 1.. Columnar structure of PFL types.

(A) Population morphological renderings for the PLF2 (Ai), PFL1 (Aii), and PFL3 (Aiii) neuron types. As in previous figure, each neuron is colored according to its column (see legend).

Figure 40.. Fan-shaped body (FB) tangential overview.

(A) Overview of FB tangential neurons. (Ai) FB tangential neurons output in single or multiple layers of the FB (e.g., in layer 4, shown in green) and may have mixed arbors in the noduli (NO), superior protocerebrum SMP/SIP/SLP, lateral accessory lobe (LAL), crepine (CRE), or other brain regions outside of the central complex. (Aii) A morphological rendering of FB4O neurons, which receive input in the SMP and CRE and output to layer 4. (Aiii) A morphological rendering of FB4I neurons, which receive input in the LAL and output to layer 4. (B) While most FB tangential neurons arborize in one FB layer and receive input external to the central complex (CX), there are exceptions. Some FB tangential neurons, for example, arborize in multiple FB layers. (Bi) shows a morphological rendering of one such type, the FB1I neurons. In contrast, some FB tangential neurons arborize exclusively within the FB. (Bii) shows a morphological rendering of one such type, the FB4Z neurons. (C) The number of FB tangential types that receive input from (top) or give output to (bottom) the CRE, SMP/SIP/SLP, or LAL. The FB layer refers to the layer where a given type has the most expansive processes. For this analysis, only the right neurons of the type are considered, and each right neuron of that type must have, on average, at least three synapses in the given region. (D) The number of neurons per FB tangential type. With a few rare exceptions, both the right and left FB contribute an equal number of neurons to each type. (E) Input pathway classifications for the FB tangential input neurons. Types are counted as inputs if they have at least 20 synapses of a given polarity outside of the CX and are the postsynaptic partner in at least one significant type to type connection outside of the CX. See Appendix 1—figure 3 for an explanation of pathway weight.

Figure 40—figure supplement 1.. Fan-shaped body (FB) arborizations by region.

(A) Region arborization plots for each FB tangential neuron type. Only FB tangential neurons from the right side of the brain are shown due to the constraints imposed by the hemibrain volume. (B) The arborizations of some FB tangential neurons are structured within a layer. These neurons tend to selectively target only certain neurons within that layer, and their structure follows the arborizations of those specific targets. (Bi) A morphological rendering of the FB4Z neurons shown in (A), now viewed along the dorsal-ventral axis. The FB4Z neurons do not fill the entire FB layer. (Bii) A rendering of one of the downstream targets of the FB4Z neurons, the hΔA neurons, whose processes also do not fill the entire layer. (Biii) The FB4Z and hΔA arborizations overlap.

Figure 40—figure supplement 2.. Fan-shaped body (FB) tangential synaptic sites that are outside of the central complex (CX).

(A) Presynaptic sites in regions of interest (ROIs) external to the CX for the FB tangential cells. The sites are color-coded by the FB layer where the given neuron has the most synapses. (B) As in (A), now for postsynaptic sites.

Figure 41.. FB2B_a connectivity.

(A) FB2B_a neurons, arborize in the SIP, the crepine (CRE), and fan-shaped body (FB) layer 2. (B) Postsynaptic FB2B_a partners in the FB. Partners include other FB tangential cells, FB neurons that are both pre- and postsynaptic in the FB, and FB outputs. (C) Presynaptic FB2B_a partners in the FB.

Figure 42.. Fan-shaped body (FB) tangential postsynaptic partners in the FB.

Type-to-type connectivity matrix for the FB tangential presynaptic partners in the FB. The connectivity of known dopaminergic neurons is highlighted in gray. Vertical lines adjacent to the y-axis mark groups of FB tangential neurons that primarily arborize in the same layer.

Figure 43.. Fan-shaped body (FB) tangential to FB tangential connections in the FB.

Type-to-type connectivity matrix between FB tangential types in the FB. The connectivity of known dopaminergic neurons is highlighted in gray. Boxes surround connections between FB tangential neurons that have their primary arborizations in the same FB layer.

Figure 44.. Fan-shaped body (FB) tangential postsynaptic partners in the FB.

(A) Type-to-type connectivity matrix for the FB tangential postsynaptic partners in the FB. The connectivity of known dopaminergic neurons is highlighted in gray. Horizontal lines adjacent to the x-axis mark groups of FB tangential neurons that primarily arborize in the same layer. (B) Type-to-type connectivity matrix for the FB tangential presynaptic partners in the FB, as seen in Figure 42A, where the axes are now flipped and only the partners from (A) are plotted.

Figure 45.. Several fan-shaped body (FB) tangential neuron types show all-to-all connections that resemble connectivity patterns within and between ER neuron types.

(A) Neuron-to-neuron connectivity matrix for the FB tangential types. (B) Locations of synapses between individual FB2I neurons (box in blue in A). FB2I neurons synapse onto other neurons of the same type across all columns of the layers that they innervate.

Figure 46.. Direct connections from mushroom body output neurons (MBONs) to central complex (CX) neurons.

(A) Network graph showing direct connections from MBONs to CX neuron types, all of which are fan-shaped body (FB) tangential cell types, with the exception of one connection involving LCNOp, a LAL-NO type. MBON nodes are colored according to their neurotransmitter identity as determined by RNA-seq (filled circles) (Aso et al., 2019) or as predicted by an artificial neural network trained on EM images of presynaptic boutons from MB neurons with known transmitter types (open circles) (Eckstein et al., 2020; Li et al., 2020). Both typical (01–23) and atypical (24–35) MBONs are included. The size of each node is proportional to the total number of outgoing (MBON types) or incoming (CX types) synapses, and the width of each edge is proportional to the connection’s relative weight, averaged over all right hemisphere regions of interest (ROIs) outside of the CX. Graph includes all direct connections with at least 10 synapses (as in Li et al., 2020) and a relative weight greater than 0.01 (see Materials and methods and Figure 46—figure supplement 1 for more details). (B) Morphological renderings illustrating several aspects of MBON-to-CX connectivity. (Bi) shows an example of a strong direct connection, from MBON09 to FB4R. (Bii) shows an example of an MBON (MBON04) that contacts multiple FB tangential neurons that innervate dorsal (FB6P) or ventral (FB1H) FB layers. (Biii) shows an example of two MBONs that release different neurotransmitters (MBON12, acetylcholine; MBON04, glutamate) but provide convergent output to the same target (FB4A neurons). (Biv) highlights the one direct connection that does not involve an FB tangential type, with atypical MBON30 synapsing onto LCNOp. Yellow circles mark synapse locations.

Figure 46—figure supplement 1.. Connection threshold dependence of mushroom body output neuron (MBON) to central complex (CX) connectivity.

(A) Scatter plot of type-to-type weight (i.e., synapse count; x-axis) versus relative weight (y-axis) for all MBON-to-CX connections. Horizontal dashed line marks relative weight threshold of 0.01. Vertical dashed line marks raw weight threshold of 10 synapses. Notice that the two thresholds mostly exclude the same cluster of connections near the origin. (B) Bar graph showing each MBON type’s strongest CX connection, expressed as a relative weight. Only MBONs that made at least three synapses onto at least one CX type were included. Horizontal dotted line marks the 0.01 relative weight threshold. (C) Line graph showing the number of CX neuron types directly downstream of MBONs as a function of the relative weight threshold. Vertical dotted line marks the 0.01 relative weight threshold employed here.

Figure 47.. Indirect mushroom body output neuron (MBON) to central complex (CX) connections.

(A) Plot showing the number of fan-shaped body (FB) tangential types, per layer, indirectly targeted by MBONs through one intermediate neuron. Only indirect pathways where each connection involved more than 10 synapses and a relative weight greater than 0.01 were considered. (B) Same as in (A), but for strong pathways, with greater than 20 synapses and relative weights greater than 0.02. (C) Network graph showing all strong (thresholds as in B), indirect connections from MBONs that receive at least 20% of their input from visual projection neurons (vPNs) to CX neuron types through one intermediate layer. Edges are colored by the intermediate neuron’s supertype, which largely reflects the region of interest (ROI) that contains its arbors. Non-FB tangential targets have gray nodes. (D) Same as in (C), but for MBONs the receive at least 20% of their input from thermosensory or hygrosensory projection neurons. (E) Network graph showing all strong, indirect connections from MBONs to non-FB tangential CX neurons. Notice that, other than ExR2, the non-FB tangential targets belong to a LAL-NO type.

Figure 47—figure supplement 1.. Indirect pathways from mushroom body output neurons (MBONs) to central complex (CX) neurons.

(A) Type-to-type connectivity matrix showing connections from MBON types to intermediate neuron types with projections to CX types. (B) Type-to-type connectivity matrix showing connections from the intermediate types in (A) to downstream neuron types in the CX. The connectivity matrix has been rotated 90° so that presynaptic types are arrange along the x-axis, which facilitates matching neurons in (A) to those in (B).

Figure 48.. Identification of the sleep-promoting dorsal fan-shaped body (dFB) tangential neuron types in the R23E10 GAL4 line.

(A) Front view of a 3D rendering of a confocal stack showing the R23E10 expression pattern (blue) along with immunohistochemical staining against nc82 (gray). The raw confocal stack was warped to a standard reference brain and rendered in 3D using VVDviewer, which facilitates direct comparison of the R23E10 expression pattern to the EM morphologies of candidate neuron cell types (see Materials and methods and Figure 48—figure supplements 2–4). (B) Stochastic labeling of subsets of R23E10 neurons made using the MCFO method (Nern et al., 2015). Note the differences in arbor morphology outside the FB and the different layers of arbors within the FB for the individual neurons in the line. (C) Matrix comparing the similarity in connectivity within the FB for the nine putative R23E10 neuron cell types (31 neurons total, see Materials and methods). (D) Single neuron morphological renderings from the EM dataset for each of the nine neuron types that were identified in the R23E10 line. Magenta circles mark presynaptic sites. Two anatomical features of R23E10 neurons —the lateral location of their soma and a fiber tract that enters the FB slightly medial to the lateral border— unambiguously identified 14 candidate tangential neuron types with processes in layers 6 and 7 whose general morphology matched that of the R23E10 pattern. Comparison of these neuron morphologies with candidate EM neuron types allowed us to exclude 5 of the 14 candidates based on the presence of arbors that lie well outside the R23E10 pattern (Figure 48—figure supplements 2–6).

Figure 48—figure supplement 1.. Region arborization plot of R23E10 and dopaminergic neuron cell types (FB6H and FB7B, see Figure 49).

Average pre- and postsynaptic counts by region are shown. Red asterisk marks synapses from FB6A located in the FB but that are incorrectly assigned to the left BU, whose ROI boundary requires revision.

Figure 48—figure supplement 2.. Summary of sleep-promoting and PPL1 dopaminergic neuron (DAN)-type identification.

Morphological renderings of the 14 dFB neuron types whose general morphology matches that of the R23E10 pattern. For each neuron type, the color of text indicates whether the type is confirmed to be in 23E10 (blue text), is confirmed to be a PPL1 dopaminergic neuron (magenta), or is neither (gray text). Arrows mark anatomical features not present in R23E10, as evaluated by directly comparing R23E10 expression to EM morphologies in the same reference brain. See Figure 48—figure supplements 3 and 4 for more details.

Figure 48—figure supplement 3.. Overlap of individual R23E10 and dopamine neurons with corresponding EM neuron types.

Morphological renderings comparing individual R23E10 cells from the right hemisphere (green), generated using the MCFO stochastic labeling technique (Nern et al., 2015), to single EM neuronal morphologies (magenta). In every case but FB7K, one or more high-quality matches (that is, those with a high degree of overlap between EM and LM processes) was obtained between single 23E10 neurons and their corresponding EM neuron type. See Videos 14 and 15 for a direct comparisons in 3D of layer 6 and layer 7 neurons, respectively.

Figure 48—figure supplement 4.. Cell types not found in R23E10, though they have similar morphology.

These three candidate EM neuron types contain processes, marked by magenta arrows, that lie well outside the R23E10 pattern.

Figure 48—figure supplement 5.. Summary of the morphologies of all layer 8 and layer 7 tangential neurons.

For each neuron type, the color of text indicates whether the type is confirmed to be in R23E10 (blue text), is confirmed to be a PPL1 dopamine type (magenta text), or does not have a general morphology consistent with R23E10 neurons (black text). In layer 7, we were able to identify high-quality matches (that is, those with a high degree of overlap between EM and LM processes) with a subset of FB7A neurons and a moderate-quality match to FB7K. As presently defined, the FB7A neuron type contains three neurons per hemisphere. Two of these neurons send processes to the lateral portion of the superior neuropil — a feature not observed in R23E10— while the remaining neuron had a high-quality match in several R23E10 MCFO samples. Therefore, we include all FB7A neurons while recognizing that future work may further refine this neuron type and its relation to the R23E10 line.

Figure 48—figure supplement 6.. Summary of the morphologies of all layer 6 tangential neurons.

For each neuron type, the color of text indicates whether the type is confirmed to be in R23E10 (blue text), is confirmed to be a PPL1 dopamine type (magenta text), or does not have a general morphology consistent with R23E10 neurons (black text).

Figure 48—figure supplement 7.. Summary of the morphologies of all layer 5 tangential neurons.

For each neuron type, the color of text indicates whether the type is confirmed to be in R23E10 (blue text), is confirmed to be a PPL1 dopamine type (magenta text), or does not have a general morphology consistent with R23E10 neurons (black text).

Figure 49.. Identification of wake-promoting, PPL1 dopaminergic dorsal fan-shaped body (dFB) tangential neuron types.

(A) Confocal micrographs showing a portion of the expression pattern of a split-GAL4 line, SS56699 (green), focused on the cell bodies of the three neurons expressed in each brain hemisphere of this line along with immunohistochemical staining against TH using a polyclonal (red) and monoclonal (blue) antibody. Inset shows a zoomed-in view of the three SS56699 soma in the right hemisphere, marked by green arrows, which are all TH+. This result was consistent across 12 hemispheres from six brains. (B) Expression pattern of the SS56699 line with nc82 reference staining (top) and zoomed-in view of the expression pattern alone (bottom). One of hemidriver parents of this line uses an 11kb genomic segment of the TH tyrosine hydroxylase (TH) gene (see Aso et al., 2014a) to drive its expression. Morphological renderings comparing the three putative dFB dopaminergic neuron types (magenta)—FB7B (top panel), FB6H (middle panel), and FB5H (bottom panel)—to individual neurons from SS56699, generated by MCFO stochastic labeling (green; Nern et al., 2015). (C) Single neuron morphological renderings from each of the three identified PPL1 dopaminergic neuron types: FB7B, FB6H, FB5H. Magenta circles mark presynaptic sites. See Video 16 for 3D comparisons.

Figure 50.. A potential sleep-wake flip-flop switch in the dorsal fan-shaped body (dFB).

(A) Neuron-to-neuron connectivity matrix between R23E10 neurons (marked by blue text) and dopaminergic neurons (marked by magenta text). Note that most layer 6 neurons connect to other layer 6 neurons, and layer 7 neurons to other layer 7 neurons, but there are many fewer connections between layers, consistent with tangential neurons’ layer-specific innervation patterns. (B) Network graph showing the intra-FB connections between the R23E10 and dopaminergic types. Arrow width is proportional to connection strength, and arrow color indicates connection type. Node color indicates whether the neuron type belongs to a putative wake-promoting type (magenta) or a putative sleep-promoting type (blue). Connections within a type have been omitted for clarity, but can be observed in (A).

Figure 51.. Downstream targets of dorsal fan-shaped body (dFB) sleep-wake neurons.

(A) Type-to-type connectivity matrices showing the neuron types targeted by each of the sleep-wake neuron types. The downstream neurons are divided into groups, with non-FB tangential targets shown in the top panel and FB tangential targets shown in the bottom panel. Green text marks neurons involved in ellipsoid body-FB (EB-FB) sleep-wake circuit (see Figure 53). Connections with relative weights below 0.005 were excluded from this analysis. (B) Number of synaptic connections from each sleep-wake neuron type to other FB neuron types (tangential or columnar), neuron types with prominent arbors outside the FB (EB/bulb [BU], superior protocerebrum SMP/SIP/SLP, lateral accessory lobe [LAL], crepine [CRE], and olfactory), or unknown types (i.e., neurons that have not been assigned a type name). (C) Same as in (B) but plotting the number of downstream neuron types reached.

Figure 52.. Inputs to dorsal fan-shaped body (dFB) sleep-wake neurons.

(A) Type-to-type connectivity matrix showing the neuron types that target each of the sleep-wake neuron types. The upstream neurons are divided into groups, with non-FB tangential targets shown in the left panel and FB tangential targets shown in the right panel. Connections with relative weights below 0.005 were excluded from this analysis. (B) Number of synaptic connections to each sleep-wake neuron type from other FB neuron types (tangential or columnar), neuron types with prominent arbors outside the FB (ellipsoid body/bulb [EB/BU], superior protocerebrum SMP/SIP/SLP, lateral accessory lobe [LAL], crepine [CRE], and olfactory), or unknown types (see legend). (C) Same as in (B) but plotting the number of upstream neuron types.

Figure 53.. A direct pathway linking sleep-wake neurons in the dorsal fan-shaped body (dFB) and ellipsoid body (EB).

Network graph showing the connections between ExR1, ExR3, and hΔK, along with some of their major upstream and downstream connections in the dFB (top panel) and EB (bottom panel). dFB types contained in the R23E10 line have "(sleep)" below their name, while the wake-promoting dopaminergic types have "(DAN)". Note that these circuits are embedded in the highly recurrent dFB and RB networks, whose many neuron types and connections have been omitted for clarity.

Figure 53—figure supplement 1.. Ellipsoid body (EB) neuron types in 5HT7-GAL4.

(A) Front view of a 3D rendering of a confocal stack showing the expression pattern of 5HT7-GAL4, which contains EB neuron types that express the serotonin 5HT7 receptor and receive input from ExR3 (Liu et al., 2019). 5HT7-GAL4 labels neurons in the ER3d, ER3p, and ER4d populations. (B) Morphological rendering of ER3d (a–d) EM morphologies overlaid on 5HT7-GAL4 expression pattern. Individual neurons. ER3d_a (1261086734), ER3d_b (1261427885), ER3d_c (1261419142), and ER3d_d (1261423534). (C) Morphological rendering of ER3p (a–b) EM morphologies overlaid on 5HT7-GAL4 expression pattern. Individual neurons: ER3p_a (1229288307) and ER3p_b (1260327246). (D) Morphological rendering of ER4d EM morphologies overlaid on 5HT7-GAL4 expression pattern. Individual neuron: ER4d (1198693217).

Figure 54.. Central complex (CX) neurons with downstream synapses outside the CX.

(A) Neuropil innervation plot of all CX types having downstream connections outside the central complex. Only CX neuron types that have a significant number of presynaptic terminals in other brain regions are shown. The CX is excluded to highlight the connections in non-CX neuropils. The CX innervation of the same neurons can be found in Figure 10 (ellipsoid body [EB] columnar, ER neurons, extrinsic ring [ExR] neurons), Figure 28—figure supplement 1 (FB columnar), and Figure 40—figure supplement 1 (FB tangential neurons). (B) Presynaptic site locations outside of the CX in the right hemibrain for the neuron types shown in (A). Left: frontal view. Right: side view. The locations are overlaid on an anatomical rendering of the relevant neuropils. Dot colors indicate the neuron type. The color code for neuropils is identical in (A) and (B). (C) The total number of synapses across all significant type-to-type connections outside of the CX in the right hemibrain for all neurons of the types shown in (A and B).

Figure 55.. Divergence of output networks.

(A) Diagram of the number of neurons contacted while walking five steps downstream from central complex (CX) neurons that arborize outside the CX. Size of the circles represents the number of new neurons in each layer. The layer a neuron is assigned to corresponds to the length of the shortest path from the CX to that neuron. The thickness of the connecting lines indicates the number of neurons reached in the same layer (loops), in the next layer (top arc) or in previous layers (bottom arcs). Color of the connector indicates the layer of origin. (B) The number of types per layer (black), the total number of targets of the previous layer to any layer (dark gray), and the projected number of types from the mean divergence of the previous layer (light gray). The difference between the total number of targets (dark gray) and the number extrapolated from the divergence (light gray) reveals the level of convergence of the output pathways. The difference between the number of types per layer (black) and the total number of targets (dark gray) corresponds to connections that are not simple feedforward connections and reveals the amount of recurrence of the output pathways. On average, each type connects 12.3 other types (divergence of 12.3) and is contacted by 9.21 other types (convergence of 9.21). Note that the total here exceeds the number of neurons in the database as they include simulated pathways on the side of the brain not present in the volume (see Materials and methods). Of the 34,100 neurons reached, 21,363 (out of 26,190 for the entire dataset) are in the hemibrain dataset, and 12,737 are mirror symmetric neurons from existing neurons inferred from symmetric connections. (C) Relative type composition of the different layers weighted by the pathway weight (see Materials and methods, Appendix 1—figure 3) they receive from the CX. Circles on the top row represent the total pathway weight received in every layer. The total pathway weight decreases as the layer gets farther away from the CX, as is expected with the metric used, which multiplies relative weights across a pathway then sums pathways ending on the same neuron. Reflecting the composition of the database, the majority of neurons reached either belong to poorly studied neuropils (‘terra incognita’) or have no name in the database (‘unidentified’). Note that in the first layer, most identified targets are CX types. (D) Same as (C), but zoomed-in onto known types excluding CX types. Types with a gray background in the legend are those for which most existing neurons of that type are present in the database. The fraction of known targets increases to reach a maximum in the fourth layer. (E) Number of types reached outside of the CX for every CX neuron innervating outside of the CX in different downstream layers. Note that very small numbers are not visible on this scale.

Figure 56.. Central complex (CX) to CX connections in the gall (GA), bulb (BU), round body (ROB) and rubus (RUB).

(A) Downstream synapses of potential CX output neurons, colored by the fraction of their target pathways that contribute to pathways coming back to the CX (see Materials and methods). The GA, BU, ROB, and RUB contribute most of their outputs to the CX and the lateral accessory lobe (LAL) almost none, while the upper neuropiles are mixed. (Ai) Frontal view; (Aii) side view. (B) Pathway weights of all pathways that start in the GA, BU, ROB, and RUB and end on another CX neuron. The weights are normalized for each type of origin. If the normalized pathway weight is 1, it corresponds to a neuron for which all output pathways come back to the CX. Connections are separated by the supertypes that these recurrent pathways reach. EB neurons mostly reach other EB neurons, whereas FB neurons mostly reach other FB neurons. (C) Morphological renderings of seven selected neuron types that innervate the GA, gall surround (GAs), ROB, and RUB. (Ci) Closeup of the four structures. (Cii) Illustration of the full morphology of the seven neuron types, showing the left population only. (D) Connectivity matrix of neurons that arborize in the right GA and GAs region. All connections outside of the CX regions (ellipsoid body [EB], protocerebral bridge [PB], fan-shaped body [FB], noduli [NO]) were considered because the GA region of interest does not capture the GAs. PFGs neurons were included in the analysis, but did not make any significant connections. (E) Illustration of selective connectivity between EPG neurons to PEG neurons from odd and even wedges of the EB in the dorsal and ventral GA, respectively. Left: schematic. Middle: rendering of EPG cells targeting the right GA with those from odd wedges colored in orange and those from even wedges colored in brown. Right: rendering of PEG neurons shown analogously as EPG cells (maroon: even-numbered PB glomeruli; pink: odd-numbered PB glomeruli). (F) Connectivity graphs on the level of neuron types showing any connections with at least 0.05 relative weight. (Fi) Connectivity in the GA and GAs. (Fii) Connectivity in the EB between the same neuron types as in (Fi), including the neurons from the other hemisphere.

Figure 56—figure supplement 1.. Gall (GA) and gall surround (GAs).

(A) Neuron-to-neuron connectivity matrix of all neurons that make connections in the right GA. (B) Neuronal profiles of PFGs, EL, and PEG neurons in the electron microscopy (EM) micrographs taken from the fan-shaped body/ellipsoid body (FB/EB) regions (left panels) and the GA/GAs output terminal regions (right panels). The presynaptic densities that are observed for PFGs and EL neurons (circled in white) are not traditional T-bar-style synapses, while those in PEG neurons have clear T-bars (white arrowheads). Dense-core vesicles (DCVs) in PFGs and EL neurons are larger than those in PEG (yellow arrows). The fills correspond to the neuron segmentation. Scale bars: 500 nm. (C) Zoom in on a non-T-bar synapse (top), and a T-bar synapse next to a dense core vesicle (bottom). Views correspond to the areas in dashed rectangles in (B). Scale bars: 200 nm. (D) Table summarizing the finding of DCVs and synapse types for various output neurons.

Figure 56—figure supplement 2.. Round body (ROB).

(A) Neuron-to-neuron connectivity matrix of PFR-to-PFR connections in the round body. PFR_b neurons contact both PFR_a neurons and themselves in the ROB in a homogeneous manner. (B) Connectivity graph of output pathways from PFR_a and PFR_b. (C) Morphological rendering of LAL002_R, the only non-central complex (non-CX) neuron targeting the ROB in a very targeted manner, and the main relay for PFR_b outputs.

Figure 56—figure supplement 3.. FR connectivity in the rubus (RUB).

(A) Neuron-to-neuron connectivity matrix of FR-to-FR connections in the RUB. FR1 neurons form all-to-all connections between themselves, FR2 is not involved in direct central complex (CX) to CX connections in the RUB. (B) Neuron-to-neuron connectivity matrix of non-CX targets of FR neurons in the RUB. FR1 and FR2 neurons have largely different targets. (C) Connectivity graph of output pathways from FR1 and FR2.

Figure 57.. Central complex (CX) to CX connections in other regions.

(A) Pathway weights (see Appendix 1—figure 3) of pathways that end on another CX neuron for neurons innervating the neuropils not described in Figure 56. Lateral accessory lobe (LAL), wedge (WED), posterior slope (PS), crepine (CRE), and superior medial protocerebrum (SMP). The weights are normalized for each type of origin (see Materials and methods). The connections are separated by the supertypes that these recurrent pathways reach. (B) Type-to-type pathway connectivity matrix of those same neuron, excluding the fan-shaped body (FB) tangential neurons. Connections from the FB to ellipsoid body (EB) and noduli (NO) neurons are highlighted. (C) Pathways from PFL neurons to EB and NO neurons. (Ci) PFL3 neurons connect ipsilaterally to LCNOp and contralaterally, through midline crossing LAL interneurons, to ER6, ExR6, and ExR4 neurons. (Cii) PFL1 neurons, through a multilayered network, reach ER1_a neurons ipsilaterally and ER1_b neurons contralaterally. Note that LAL138 is the WL-L neuron described in the section about mechanosensory inputs. (Ciii) PFL2 neurons reach LNO3 in two steps.

Figure 57—figure supplement 1.. All central complex (CX-to-CX) connections.

(A) Pathway weights (see Appendix 1—figure 3) of all pathways that end on another CX neuron. The weights are normalized for each type of origin (see Materials and methods). If the normalized pathway weight is 1, it corresponds to a neuron for which all output pathways come back to the CX. Connections are separated between axonal, dendritic, self, and self-contralateral. ‘Self’ pathways end on the same neuron type they started with, whereas ‘self-contralateral’ pathways end on the mirror symmetric type. ‘Axonal’ pathways end on a neuron for which more than 75% of the synapses located outside of the CX are output synapses. (B) Connectivity matrix of all CX-to-CX pathways outside of the CX, filtered for pathway weights larger than 0.5%.

Figure 58.. Central complex (CX-to-CX) motifs.

(A) Schematic of the three motifs considered and their equivalent representation in a compact circular network plot. The CX output type of interest is in gray and at the center of the circular diagram. It reaches other CX neurons (green) through pathways that can be constituted of multiple steps (pink). Motifs are formed by the relation between those pathways outside of the CX and the connections formed inside of the CX by the same CX neurons. ‘Canonical feedback’ corresponds to the target of the pathway contacting the source type in the CX (yellow). ‘Parallel connections’ occur when the source neurons also contact the pathway target neuron inside the CX (red). ‘Linked targets’ are neurons connected in the CX that are targets of the same neuron outside of the CX (green). The equivalent circular plot is provided below each motif, and their combination in a single polar plot is on the right. (B–D) Example motifs. Left: circular motif graph showing the motif in the context of all the CX-to-CX motifs that the type of origin is implicated in. Right: frontal and lateral morphological renderings. (B) FB2B_b forms a canonical feedback loop with PFL1 neurons. PFL1 contacts FB2B_b in the LAL while FB2B_b contacts PFL1 in the fan-shaped body (FB). (C) Parallel connections. FB6T contacts FB6E both in the SIP/SMP and FB. (D) Linked targets. FB8F_a contacts four FB6 neurons who are themselves interconnected in the FB. (E) Prevalence of the three motifs for all the potential CX output types. The colored circles represent the prevalence of each specific motif. The gray circles represent the total number of all the motifs of the same type that could form given that type’s partners outside of the CX. Ellipsoid body (EB) columnar, ExR2, and ExR3 neurons form a large proportion of all the possible motifs, reflecting the high level of recurrence in EB circuits.

Figure 59.. Feedforward output networks.

(A) Total pathway weights contributed by the different central complex (CX) output neurons (summed over all neurons in the graph, see Materials and methods and Appendix 1—figure 3). Color indicates if the receiving types are identified or not (unidentified means that they are part of poorly studied neuropils or unknown). Compare to Figure 54C. some types (e.g., FR and FS neurons) reach a large number of neurons but have much lower pathway weights than other prominent types (for example, PFL neurons). This discrepancy arises because these types only make modest contributions to their targets. (B) Histogram of total pathway weight received from the CX for every downstream type. Most neurons receive very weak inputs from the CX. Note the log scale, without which the handful of neuron types receiving strong CX contributions would be invisible. (C) Connectivity matrix from the CX to every type in the downstream network graph receiving more than 0.5% total pathway weight (filtered for individual weights > 0.5%). The CX output types are ordered according to the similarity of their output vectors (see Materials and methods). Most important targets are influenced mainly by a single CX type. PFL3 neurons contact more types than any other CX type. Note that for convenience of display the names of the targets are not displayed on the x-axis. (D) Downstream neuropil innervation of the targets of CX output neurons, starting on the right side of the brain. The innervations are weighted by the pathway weight they receive. The region of interest (ROI) score is the sum of pathway weights for all the target types innervating the ROI times their number of downstream synapses in the same ROI. (Di) Measured innervations (right and central neuropils). (Dii) Simulated innervations (left neuropils) from the known symmetric types. This is necessarily an underestimate of the extent to which the CX pathways reach the contralateral side of the brain. (E) Downstream synapses of targets receiving more than 0.5% of pathway weight from the CX, colored by the CX type contributing the most to their inputs.

Figure 59—figure supplement 1.. Clustering at different depths.

Cosine distances between CX output types were computed for all connections originating from the CX at a given pathway length (labeled at top). "Full graph" means that the pathway weights are used, so that all pathway lengths contribute to it. Modularity is strong at the onset (and for the full graph as the short paths dominate) and remains strong relatively deep in the network. CX outputs were clustered on their full connectivity profile (the pathway weights they contribute to every neuron in the network). For all panels the ordering is the clustering order obtained on the full pathway weight distances.

Figure 59—figure supplement 2.. Modularity of output networks.

(A) The full graph of all strong feedforward targets of the central complex (CX), in a stress minimizing layout (see Materials and methods), which tends to keep strongly connected neurons close to each other. Nodes are colored by their main CX contributor. (B) Same graph, but where the nodes are colored by the results of the label propagation community detection algorithm (see Materials and methods). Neurons that receive their main input from the same CX neuron tend to form communities. A community is a set of nodes that are more connected between themselves than they are with the rest of the network. (C) Average connection strength between two neurons as a function of their main CX contributor (filtered for average connections larger than 0.1%). Connections between neurons that share their main CX input (on the diagonal) tend to be stronger.

Figure 59—figure supplement 3.. Same as Figures 54B and 59E for PFL, FS, and FC neurons alone.

Left: output synapses made by those neuron types outside of the central complex. Right: output synapses made by the strongest downstream partners of those same neuron types.

Figure 59—figure supplement 4.. Same as Figures 54B and 59E for FR, PFR, and ExR neurons alone.

Left: output synapses made by those neuron types outside of the central complex. Right: output synapses made by the strongest downstream partners of those same neuron types.

Figure 59—figure supplement 5.. Neuron-to-neuron output connectivity of the main columnar output neurons.

Relative weight of connections from FB columnar types, ordered by the columns they innervate, to neurons receiving at least 0.5% pathway weight from the CX. No columnar structure is visible at the output stage, which reflects the fact that all columnar neurons of the same type innervate almost perfectly overlapping territories.

Figure 60.. Connections to identified types.

(A) Normalized pathway weights (see Methods Appendix 1—figure 3 and Materials and methods) of pathways that end on various known types. Only central complex (CX) neurons that contribute at least 0.5% of their outputs to feedforward networks are included. ‘Unknowns’ correspond to pathways that never reach a known type. Note that some of the known groups are still very broadly defined. For example, the group labeled ‘LH’ contains a lot of functionally diverse neurons with branches in the lateral horn (one of which is the WPN neuron mentioned in the mechanosensory inputs section). (B) Same as (A), but zoomed in on the known types. (C) Pathway weights from the CX received by the most prominent known targets (getting at least 0.5% of their inputs from CX pathways), colored by the CX types of origin.

Figure 61.. Connections to mushroom body output neurons (MBONs), dopaminergic, and antennal lobe neurons.

(A) Pathway weights (see Appendix 1—figure 3) from the central complex (CX) to MBONs, dopaminergic (DANs), and antennal lobe neurons for which the total pathway weight is greater than 0.05%, colored by the CX type of origin. (B) Network diagram showing the interconnections between those pathways. Grayed areas correspond to the morphological renderings in (C–E). CX neurons have bold labels and circles. (C) FR2 to PPL107 and CRE054, morphological rendering. (D) FR1 to PPL102 and MBON30, morphological rendering. (E) FB8F_a/MBON23/PPL105 loop, mFB8F_aorphological rendering.

Figure 61—figure supplement 1.. Main circuits converging onto oviIN and MBON27.

(A) Network diagram showing the FS and FC connections to oviIN and MBON27. (B) Morphological rendering of example FS and FC neurons, oviIN and MBON27.

Figure 62.. Connections to visual projection neurons (vPNs).

(A) Total pathway weights (see Materials and methods and Appendix 1—figure 3) from the central complex (CX) to vPNs for which the total pathway weight is greater than 0.05%, colored by the CX type of origin. (B, C) PFL3 neurons interact with LC10 neurons through anterior optic tubercle-lateral accessory lobe (AOTU-LAL) neurons. (B) Network diagram showing how PFL3 neurons interact with LC10 neurons both ipsilaterally and contralaterally through AOTU-LAL neurons. The midline is denoted by the vertical dotted line. (C) Morphological rendering showing that PFL3 interacts with LC10 along a dorsoventral axis in the AOTU, corresponding to the anteroposterior axis in the lobula. Connections are stronger on the ventral side of the AOTU, corresponding to LC10 neurons innervating the posterior part of the lobula. (D, E) PFL3 neurons interact with a subpopulation of LC33 neurons. (D) Network diagram. PFL3 neurons synapse onto LC33 in the LAL both directly and through one of its strong targets (LAL141). (E) Morphological rendering showing which subset of LC33 neurons is associated with PFL3 neurons. (F, G) PFL1 neurons interact with LC27 neurons. (F) Network diagram showing that the connection is through two layers constituted by LAL and PLP neurons, respectively. (G) Morphological rendering showing that the PLP neurons downstream of PFL1 specifically target the LC27 glomerulus. (H, I) ExR8 neurons contact ventral centrifugal horizontal (VCH) and dorsal centrifugal horizontal (DCH), centrifugal neurons of the horizontal fiber system. (H) Network diagram showing that ExR8 reaches CH neurons both directly and indirectly through a PS neuron. (I) Morphological rendering. PS047 innervation closely follows ExR8 innervation.

Figure 62—figure supplement 1.. PFR_b-to-visual projection neurons (PNs) connections.

(A) Network diagram. PFR_b reaches visual PNs through the MBON20 and mALD1 neurons. (B) Morphological rendering of MBON20 with its visual target, LT85. (C) Morphological rendering of mALD1 with one of its visual targets, MC62.

Figure 62—figure supplement 2.. PFL3 and LC33 neuron-to-neuron connectivity.

(A) Neuron-to-neuron connectivity of PFL3 to LC33 neurons. Only 4 LC33s out of 15 receive any PFL3 synapses (and only three consistently so). (B) Postsynaptic connectivity of PFL3 and LC33 neurons. For ease of display, this matrix has its axis flipped compared to the convention used in the paper. Note that the downstream connectivity of LC33 neurons is not consistent and only partially overlaps with that of PFL3 neurons. Targets of LC33 neurons that are not contacted by PFL3 are circled in teal. In yellow, targets of PFL3 not contacted by LC33. In green is an example of shared connectivity. Even in that case, usually only one of the LC33 neurons contacts the same neuron as the PFL3 neurons. (C) Schematic representation of the degree of overlap between LC33 and PFL3 targets.

Figure 63.. Connections to descending neurons.

(A) Pathway weight (see Appendix 1—figure 3 and Materials and methods) from the central complex (CX) to descending neurons (DNs) for which the total pathway weight is greater than 0.05%, colored by CX type of origin and separated by their putative ventral nerve cord (VNC) innervation. (B) Network diagram of the main CX to DN connections, restricted to DNs on the right side of the brain. DN connections primarily come from PFL2_R and PFL3 neurons, with smaller contributions made by ExR7 and ExR8 neurons. PFL2 neurons reach DNs through ipsilateral networks while PFL3 neurons also reach them via lateral accessory lobe (LAL) interneurons crossing the midline. Much of the circuit is shared between PFL2 and PFL3 pathways. VNC innervations for each DN type are indicated below, highlighting their diversity. (C) Morphological rendering of PFL3_R neurons with two of its midline crossing targets, AOTU019_R and LAL121_R. (D) Morphological rendering of PS013_R, DNa04_R and LAL018_R. Note how LAL018 innervations follow those from DNa04. (E) Morphological rendering of DNa02_R and LAL010_R.

Figure 63—figure supplement 1.. Other connections to descending neurons (DNs).

(A) Network diagram showing the connection from PFL2,3 neurons in the right lateral accessory lobe (LAL) to the bilateral MDNs. LAL160, a midline crossing LAL neuron linking PFL2 to the MDN, is also targeted by MBON30, a MBON neuron receiving direct central complex (CX) input (from FR1 neurons, see Figure 61). (B) Morphological rendering of LAL160_R, PS010_R, and MDN neurons. (C) Network diagram of the indirect connections between ExR8 neurons and DNp15 and DNp16/17 neurons. (D) Morphological renderings of ExR8, PS235_R, and DNp15_R neurons. (E) Network diagram of the indirect connection between FR2 and DNp32 neurons.

Figure 64.. PFL3 outputs distribution.

(A) Connectivity matrix between PFL3_L and its direct downstream partners outside of the central complex (CX), on the right side of the brain. PFL3 neurons are binned by protocerebral bridge (PB) glomerulus (numbers of neurons per glomerulus are indicated in parenthesis on the y-axis). (B) Sum of the relative weights across glomeruli. Connections are strongest for glomerulus L3.

Figure 65.. Asymmetric connection to the flange (FLA).

(A) Connectivity matrix between FS4A_L and its direct downstream partners outside of the central complex (CX), on the right side of the brain. FS4A neurons are binned by fan-shaped body (FB) column (numbers of neurons per column are indicated in parenthesis on the y-axis). Right: sum of the relative weights across columns. Connections are biased towards columns C7–C9. (B) Morphological rendering of two of the strongest direct targets of FS4A, SMP297, and SMP304 (also circled in green in the connectivity matrix in A). All strong direct targets of FS4A project to the FLA, potentially participating in the control of feeding behaviors. FS4A neurons are columnar and project unilaterally in the superior medial protocerebrum (SMP), raising the possibility that they control feeding behaviors in a directed fashion. Besides, the asymmetry in FB innervation also suggests that this behavior could have a default directionality corresponding to the border columns of the FB.

Figure 66.. Mapping multisensory cues to a flexible head direction representation.

(A) Illustration of different types of visual cues found in a natural setting that can inform the fly about its orientation. The sun represents a prominent bright landmark but also creates a polarization pattern that covers the full sky. In addition, terrestrial features create a visual scene that can be mapped onto the head direction representation (Fisher et al., 2019; Kim et al., 2019). (B) Ring neurons bring sensory information to the central complex (CX), where they provide input to the fly’s head direction system. Sensory pathways have been described for mechanosensory information about wind direction (Okubo et al., 2020), celestial visual cues related to the polarization pattern of the sky or visual features (Seelig and Jayaraman, 2013). (C) Hypothetical competition and transformation that could occur through interactions between ring neuron types conveying distinct directional information. Due to hierarchical competition, one sensory cue, for example, polarization pattern, could dominate at the expense of other, less reliable cues. The transformation from sensory information represented by ring neurons to the head direction estimate allows for complementary directional cues to be combined. (D) Schematic of ring neurons that respond to local features in a visual scene (Di). Plasticity between these ring neurons and EPG neurons (Dii) ensures that the compass reliably tethers to the visual scene.

Figure 67.. Disambiguating directional information from polarized light sensors.

(A) Connectivity matrix of ER4m inputs to EPG neurons in the ellipsoid body (EB). EPG neurons are sorted according to the EB wedge they innervate. See also Figure 11D. (B) Pairwise Pearson’s correlation measured between individual EPG neurons according to the pattern of their ER4m neuron inputs. See Figure 11—figure supplement 1 for details. (C, D) Under most conditions, the two eyes, and thus the left and right polarization-sensitive dorsal rim areas, are expected to receive different input. (C) Schematic of the polarization pattern and the sun position of the sky in relation to the fly’s eyes depending on the fly’s orientation. Receptive fields of the polarization sensitive dorsal rim area for the left (green) and right (orange) eye are overlayed. (D) Receptive fields of the left and right dorsal rim area now shown with an indication of the orientation of the e-vector direction that different parts are sensitive to. (Ci, Di) The sun is located to the left of the fly. (Cii, Dii) The sun is located to the right of the fly.

Figure 68.. Conveying and transforming the head direction representation from the ellipsoid body (EB) to the fan-shaped body (FB).

(A) Schematic showing how a bump of activity gets conveyed from the EB to the left and right PB. EB wedges and PB glomeruli are colored by their anatomical phase (i.e., directional tuning). Based on data from Figure 16. (B) Δ7 neurons in the PB transform any EPG activity profile into a sinusoidal activity profile that gets inherited by FB columnar neurons. PFN neurons receive the EB bump directly from EPG neurons (dashed arrow) as well as through Δ7 neurons (solid arrow) (Bi). Schematic showing how the connectivity of Δ7 neurons might transform any EPG activity profile into a sinusoidal activity profile (illustrated for PFN neurons; yellow curve) (Bii). Based on data from Figure 20. (C) Overview of bump propagation through the central complex (CX) along with the major computations carried out in each region. (D) Phasor representation of a sinusoidal activity profile. Any sinusoidal activity bump can be represented as a vector whose angle encodes bump phase and whose magnitude encodes bump amplitude. Schematic of hypothetical sinusoidal activity bump (purple line), centered at 0°, encoded by a population of neurons that function as a sinusoidal array, and phase representation (purple arrow) of the same activity (Di). Vector addition can easily be implemented by sinusoidal arrays carrying different activity bumps (purple, blue, light blue) (Dii). Two vectors (purple and light blue) can be summed to generate a new vector (blue) (Diii). Based on data from Figure 20.

Figure 69.. Conceptual model showing that PFN phase shifts, when combined with differential noduli (NO) input, could produce ±45° bump shifts between the protocerebral bridge (PB) and fan-shaped body (FB).

(A) Schematic of a PB-to-FB projection pattern with a one-column contralateral phase shift. PB glomeruli and FB columns are colored according to anatomical phase (from –180° to 180°), which indicates matching bump locations. PFN neurons innervating the right PB project to the left NO, where they receive input from LAL-NO neurons carrying self-motion information from the left lateral accessory lobe (LAL). Similarly, PFN neurons innervating the left PB project to the right NO and receive self-motion inputs from LAL-NO neurons innervating the right LAL. As shown in the bottom panel, when these two LAL inputs are equal, the spatially offset bumps from the left and right PB sum to generate a new bump located halfway between the two. (B) Same as in (A), but with differential NO input. In this case, as shown in the bottom panel, the R LAL neurons increase the bump amplitude of the left PB population (pink bump), and the L LAL neurons decrease the bump amplitude of the right PB population (blue bump). The sum of these two bumps will end up closer to –90°, the location of the left PB bump, due to the difference in bump amplitudes. (C) Phasor diagram interpretations of the scenarios from (A) and (B). In the left panel, with equal NO input, the left PFN and right PFN bumps sum to produce the purple vector located at 0. In the right panel, with differential NO input, the left PFN bump becomes bigger than the right PFN bump (as in B), and therefore, their sum is closer to the left PFN bump. This effectively shifts the bump by a phase that depends on the difference in amplitude between the left and right PFN populations. Importantly, the PFN neurons’ one-column ipsilateral phase shift limits such bump shifts to ±45° from the PB bump. Phases shifts outside this area (marked in gray) cannot be produced. (D) Illustration of a head direction to body direction coordinate transformation. (E) Illustration of a forward model of head direction. Figure describes conceptual models based on data from Figures 30 and 34.

Figure 70.. Conceptual model showing how two PFN populations, when combined with differential noduli input, could form a four-vector basis set whose summation could produce any vector.

(A) Schematic of a protocerebral bridge-to-fan-shaped body (PB-to-FB) projection pattern with a one-column contralateral phase shift. (B) Phasor diagram interpretation showing how hΔ and vΔ motifs could be used to transform PFN bumps. For the first PFN population (top panels), an inhibitory hΔ or an excitatory vΔ could pass the PFN1 bump along while maintaining is phase. In contrast, for the second PFN population (bottom panels), an excitatory hΔ or and inhibitory vΔ could shift the PFN2 bump by 180°. For both PFN population 1 and 2, differential noduli (NO) input could shift their summed FB bump location by ±45°, but not outside this region (marked by gray portions of each circle). (C) Phasor diagram of a hypothetical downstream neuron that sums the input from the PFN1 and PFN2 populations. In this case, the downstream neuron’s population activity would be the sum of four independent bumps, located 90° apart from one another and whose amplitudes are modulated by independent noduli input. These four bumps could act as a basis set for computing an arbitrary vector, thus freeing the resulting bump from the ±45° range. (D) Phasor diagram showing how modulating the amplitude of the left and right PFN1 and PFN2 populations could be used to compute an arbitrary vector (shown in black). Figure describes a conceptual model based partially on data from Figure 30, Figure 31, Figure 33, Figure 34, and Figure 37.

Figure 70—figure supplement 1.. Dynamic updating of the four-vector basis set.

Schematic showing how the four-vector basis set gets updated as the fly's head direction changes. Here a fly is shown walking in a loop, with phasor plots showing the state of the FB network's four vector basis set at eight different head directions. The fly's head direction is marked by the large arrow in each phase diagram. As the fly's head direction changes, so too do the bump positions of the PFN populations that form the four-vector basis set, though the relative positions across the four populations are maintained. For example, the Right PFN1 population always has a bump located 45° clockwise from the fly's head direction vector. In this view, at every moment in time, the FB network has access to four vectors pointing in four distinct allocentric directions. Modulating the amplitude of each vector according to self-motion cues may allow for various navigational computations, as described in later figures. Figure describes a conceptual model.

Figure 70—figure supplement 2.. The fan-shaped body (FB) network has the necessary connectivity and depth to form a basis set: bump propagation using simulated activity through actual FB connectivity.

(A) Two PFN types (PFNp_c and PFNd) receive compass input that generates two simulated activity bumps in the left and right protocerebral bridge (PB), centered at R5 and L5. (B) In FB column space, the two simulated activity bumps end up 90° apart owing to the PFN one-column contralateral phase shift. In particular, the activity bump at PB R5 is centered at FB C6 (blue bump), and the activity bump at PB L5 is centered at FB C4 (pink bump). (C) An excitatory vΔK inherits the PFNp_c activity bump and moves it to downstream types while maintaining bump phase. In contrast, an excitatory hΔA receives the PFNd activity and shifts it by 180° before passing it on to downstream types. (D) PFL3 neurons, a major central complex (CX) output neuron type, receive input from both vΔK and hΔA, which could potentially instantiate a four-vector basis set. Note that while this figure keeps the left and right PFN bumps separate as they propagate through the network, each layer would represent the summation of their inputs as a single bump. And while we assume both vΔK and hΔA are excitatory, this currently remains unknown. However, many similar pathways exist and an excitatory vΔ type could function like an inhibitory hΔ type. Figure shows a simple model that uses actual connectivity to simulate bump propagation through the FB network (see Materials & Methods for details).

Figure 71.. A conceptual model that computes an allocentric translational velocity vector using head-centered optic flow sensors during flight.

(A) Illustration of a fly whose head and body direction are pointed north and whose translational velocity vector is 22.5° east of north. (B) Schematic of noduli circuitry, showing that the left and right PFN1 and PFN2 populations receive input from right and left LN1 and LN2 neuron types, respectively. (C) LN1 and LN2 neuron types are those described by Stone et al., 2017. They function as optic flow-based velocity sensors with preferred expansions points spaced at 45° intervals around the fly’s head. (D) Schematic of a four-vector basis set. Importantly, note that each PFN vector points in the same direction as its upstream LN neuron’s preferred optic flow direction. (E) Schematic showing how the four-vector basis set, whose vectors are amplitude-modulated by the LN velocity sensors, can compute the fly’s translational velocity vector. In this case, because the fly is moving just east of north, LN1_L is driven most strongly, which increases the amplitude of right PFN1 population (blue vector). When properly calibrate, summing the amplitude-modulated PFN vectors compute the fly’s translational velocity vector. Figure describes a conceptual model based on previous work (Stone et al., 2017).

Figure 72.. A conceptual model that computes an allocentric translational velocity (TV) vector using body-centered velocity estimates during walking.

(A) Illustration of a walking fly whose head direction, body direction (BD), and TV direction are all different. The fly’s head is pointing north, its body 22.5° east of north, and it’s walking northeast (that is, 45° east of north). The fly’s TV vector can be computed by summing the component of its movement parallel to its body axis (TV_//) with the component of its movement perpendicular to its body axis (TV_⊥). Circuits for computing these quantities are shown in (B) and (C), respectively. (B) Circuit for computing the component of the TV vector parallel to the fly’s body axis. As shown in the bottom panel, this circuit uses a four-vector basis set whose PFN vector amplitudes are modulated by lateral accessory lobe-noduli (LAL-NO) inputs that encode whether the head is left (HA_L) or right (HA_R) of the fly’s body axis as well as either a forward (For.) or reverse (Rev.) velocity signal. The firing rate of each PFN population is noted below each PFN node. Arrow width is proportional to firing rate. Gray arrows indicate neurons that are silent. Note that head angle input alone is insufficient to bring the LN neurons to threshold, but it can boost PFN firing when combined with a velocity input. In this case, LN2_L remains silent despite receiving a head angle input from HA_L, and LN1_L is strongly driven by both the forward velocity signal and HA_L. LN1_R, meanwhile, is moderately driven by the forward velocity signal alone. This conditional effect of the head angle input could be achieved in other ways, but the core conceptual model would remain the same. In all cases, the circuit would require proper calibration for the vector summation to accurately compute the fly’s TV_// vector. (C) Circuit for computing the component of the TV vector perpendicular to the fly’s body axis. The circuit shown in the bottom panel operates like that described in (B), but the forward and reverse velocity signals have been replaced by left (SS_L) and right (SS_R) sideslip velocity signals. As in (B), a head angle input alone is insufficient to bring LN neurons to threshold. Note that these circuits function regardless of which direction the fly’s head is facing and which direction the fly is moving, as detailed for four other examples in Figure 72—figure supplement 1. (D) Phasor diagram showing how summing the output from the circuits in (B) and (C) yields an exact TV vector whose integration would compute the path integration vector. Figure describes a conceptual model.

Figure 72—figure supplement 1.. The circuit for computing TV_⊥ operates independent of the fly’s head-body angle and which direction the sideslip component is towards.

(A) Left panel shows an example of a fly whose head direction is north, body direction is 22.5° east of north, and whose TV_⊥ vector is 22.5° south of east, which is towards the fly’s right (green array). The circuit on the right is the same as in Figure 72B and uses a four-vector basis set and head-angle and sideslip velocity to compute TV_⊥. (B) Same as in (A) but for a fly that is sideslipping towards its left. (C) Example showing a fly whose head direction is north, whose body direction is now 22.5° west of north, and whose translational velocity vector has a component 22.5° north of east, which is towards the fly’s right. (D) Same as in (C), but for a fly that sideslipping towards its left.

Figure 73.. PFL neurons could generate egocentric motor commands by comparing the fly’s allocentric head direction to an allocentric vector stored in the fan-shaped body (FB).

(A) PFL2 neurons could use their 180° protocerebral bridge (PB)-FB phase shift to generate a forward velocity signal that is largest when the fly is oriented towards the ‘goal vector,’ which in our formulation is away from the ‘stored vector’ (see bottom panel and ‘Discussion’). PFL2 neurons sample a single bump in the PB and individual PFL2 neurons project to both the left and right lateral accessory lobe (LAL), consistent with a bilateral velocity signal like forward walking. Top panel shows a schematic of the PFL2 180° phase shift between PB glomeruli (top row) and FB columns (bottom row). In this example, the stored vector points due north. To return to the goal location, PFL neurons compare the fly’s instantaneous head direction to the stored vector. The 180° phase shift ensures that PFL2 output will be largest when the fly is oriented towards the goal direction (and opposite the stored vector). (B) Similar to (A), but for the PFL3 neuron type and its 90° PB-FB phase shift. Unlike PFL2 neurons, PFL3 and PFL1 neurons (C) sample head direction bumps from the left and right PB, and individual neurons project to either the left or right LAL, consistent with motor commands with a left/right asymmetry, such as turning. In the case of PFL3 neurons, the 90° phase shift ensures that the left PFL3 population will be most active when the fly is 90° to the right of the goal direction. Similarly, the right PFL3 population will be most active when the fly is 90° to the left of the goal direction. If we assume that the right PFL3 neurons generate right turns and left PFL3 neurons generate left turns, then the motor command would act to align the fly’s heading with that of the goal direction. (C) Same as in (B), but for PFL1 neurons and their 45° PB-FB phase shift. Note that in all cases, the PB-FB phase shifts are an idealized version of those from Figure 39. The actual PFL phase shifts are not as stereotyped, since the phase shifts are continuous in anatomical space, unlike the discrete mapping schematized here.

Figure 73—figure supplement 1.. Numerosity and systematic asymmetries in synapse counts across columns may set up a potential ‘default goal vector’ through the PFL neurons.

(A) Connectivity between EPG and PFL2 and PFL3 neurons shows systematic columnar variation in synapse counts and across-column spread, beyond those expected from differences in proofreading. (B) Simulated EPG activity in the protocerebral bridge (PB) (Bi) propagated across connectivity matrix in (A) would evoke differential activity in the population of PFL2 (Bii) and PFL3 (Biii) neurons in the PB depending on their columnar identity. Note that this ignores any influence that the Δ7 neurons may have on activity propagation between the EPG and PFL populations. (C) Connectivity matrix of PFL2 and PFL3 neurons to descending neurons (DNs). (D) Resulting summated DN activity based on propagating the activity of PFL neurons across the DNs for different positions of the EPG bump in the PB. The activity propagation shown explicitly excludes any influence on PFL activity from their many inputs in the FB. Under these assumptions, DN activity would peak at different positions (phase-shifted by 45°) for the two DNs, based on whether they were activated by the bilaterally projecting (and likely forward-movement modulating) PFL2 neurons or the unilaterally projecting (and likely turn-modulating) PFL3 neurons. For reasons spelled out in Figure 72, this could, in principle, create a default ‘goal’ that could be moved in the FB. A scheme with some similarities to this, and also relying on somewhat different inhomogeneities in synaptic weights onto PFL3 neurons and modulation of activity in the FB, has been proposed by Rayshubskiy et al., 2020. Importantly, synaptic count inhomogeneities in the PB are not required for the FB-driven framework conceptualized in Figure 72.

Figure 74.. Summary of output networks.

Schematic representation of the contributions of CX output neurons to various subnetworks and their potential functions. Outputs are divided between unilateral (like PFL3) versus bilateral (like PFL2), as those are likely to control different types of behavior (asymmetric vs symmetric), and between columnar and non-columnar, likely distinguishing between orientation-dependent and orientation-independent action selection.

Figure 74—figure supplement 1.. PFL1 subnetworks, rationale behind Figure 74.

Network diagram of all targets of PFL1 receiving at least 0.5% of pathway weight from PFL1. We divide the network into three domains: Projections to the ipsilateral WED and PLP. Projections to the contralateral WED through a single neuron, LAL138. Criss-crossing motifs: a set of LAL neurons that cross the midline back and forth. We analyzed these networks, and the morphology of the neurons that constitute them, to construct Figure 74.

Figure 75.. The central complex (CX) seen as a deep recurrent neural network for navigation.

(A) A layered representation of the connectivity of a selection of neuron types in the CX, with a bias towards those involved in navigation. Layers have been labeled by their putative computational roles in a navigational context. (B) The connectivity of ER4m, ER3a_a,d, ER3m, ER4d, ER2_a,b,d, ER1_a,b, and EPG neurons is densely recurrent. However, different neuron types have specific roles in circuit function. The ER types plotted here are also the types plotted in layer 2 (cue competition/stimulus selection) in (A). (C) If neurons in (B) were unsorted, the structure in their connectivity would be difficult to recognize (left). When properly sorted by types, the structure in the network connectivity becomes clear (right). The neuron names were randomly shuffled to generate the unsorted plot at left.

Figure 75—figure supplement 1.. The structure in the fan-shaped body (FB) connectivity becomes clear when neurons are sorted by type.

As in Figure 75C, but now for the neurons in layers 4-7 (vector computations/coordinate transformations, action selection) of Figure 75A.

Appendix 1—figure 1.. Regular and convergent synapses in the CX EM micrographs from the CX.

Scale bars: 200 nm. Typical polyadic synapses (in FB, arrowheads), and synaptic vesicles (red arrows). Convergent synapses found in EB (double arrowheads).

Appendix 1—figure 2.. An example of connectivity subtypes within a single morphology type.

All these neurons were classified as type ‘FB2F’ but subdivided into three connectivity types. Scale bar: 50 μm.

Appendix 1—figure 3.. Graphical methods for pathway tracing and computation of pathway weights.

(A) By walking n layers (5 in the case of the outputs network) downstream from a starting layer (here, potential CX output neurons), one obtains a complex interconnected graph. (B) Computing pathway weights: Bi The graph obtained in A yields an adjacency matrix Adj of relative weights. T₁ is the connectivity matrix of direct connections from the source neurons. Bii Formulas used to compute the pathway weight. The full pathway weight can be obtained by summing the powers of the adjacency matrix. (C)Toy example for a network with four neurons. Ci Network graph and associated adjacency matrix. The first line of the matrix is the output connectivity vector of neuron A. Cii Multiplying the first line of the matrix by the full adjacency matrix yields the two step connectivity vector from neuron A (this would be T₂ in Bii). In this case only A to D is non zero. The 2-steps weight from A to D is obtained by multiplying weights along paths and summing across path as shown in the schematic formula below. (D) Since the metrics used are between zero and one, the norm of the connectivity matrix of connections of length n converges to zero as n grows. Intuitively, when considering long paths, connectivity gets very weak and diffuse. As a consequence, the pathway weights matrix, which is the sum of T_N converges to a stable value.