Dimensionality reduction simplifies synaptic partner matching in an olfactory circuit

Abstract

The distribution of postsynaptic partners in three-dimensional (3D) space presents complex choices for a navigating axon. Here, we discovered a dimensionality reduction principle in establishing the 3D glomerular map in the fly antennal lobe. Olfactory receptor neuron (ORN) axons first contact partner projection neuron (PN) dendrites at the 2D spherical surface of the antennal lobe during development, regardless of whether the adult glomeruli are at the surface or interior of the antennal lobe. Along the antennal lobe surface, axons of each ORN type take a specific 1D arc-shaped trajectory that precisely intersects with their partner PN dendrites. Altering axon trajectories compromises synaptic partner matching. Thus, a 3D search problem is reduced to 1D, which simplifies synaptic partner matching and may generalize to the wiring process of more complex brains.

Fig. 1.. During development, PN dendrites are exposed to the antennal lobe surface regardless of their position in adults.

(A) Drosophila brain and antennal lobe schematics, at 30 hours after puparium formation (h APF) (left) and adults (right). Antennal lobes are highlighted in dark grey surrounded by dash squares and magnified to the right. At 30h APF, PN dendrites (magenta) innervate similar positions as in adults and ORN axons (green) navigate along the surface of the ipsilateral antennal lobe from entry point at the bottom right towards midline at the top left. In adults, ORNs and PNs establish one-to-one connections in individual glomeruli that form a 3D glomerular map. (B) Adult antennal lobe schematic with ~50 glomeruli circled. Cyan: glomeruli located at the surface of the antennal lobe; orange: glomeruli located in the interior of the antennal lobe. (C) Optical sections showing dendrites of specific adult-surface PN types (magenta, labeled by a membrane-targeted GFP driven by separate genetic drivers specific to PN types labeled above) viewed from the horizontal plane (top row) and the vertical plane (bottom row) of the antennal lobe in adults. White dash lines outline the antennal lobe neuropil stained by the N-cadherin (NCad) antibody (blue). Yellow dotted lines indicate the intersections with the vertical planes shown below. Vertical planes are reconstructed from 3D image volumes where optical sections were taken horizontally. The top and bottom rows show the same brains. Scale bar = 10 μm in this and all other panels. (D) Same as (C), but with data from 30h APF. (E and F) Same as (C) and (D), but for adult-interior PN types. In (F), arrowheads indicate PN dendrites extending to the antennal lobe surface. *, PN cell body. (G) Probability distribution of VA1d-PN dendritic pixels projected onto the antennal lobe surface during development (middle) and in adults (right). The 2D antennal lobe surface is flattened and decomposed into two axes: the x-axis indicates the angle θ of each vertical plane and the y-axis indicates the position L along the long axis of the antennal lobe. Schematic definition of θ and L on the left. (H) Probability distribution of dendritic pixels from twelve PN types projected onto the antennal lobe surface. Left and middle, each ellipse corresponds to one PN type, with ellipse centers matching PN-dendrite centroids, and ellipse boundaries matching the standard deviations of PN dendrites along the x- and y-axes, respectively. Right, arrows represent the shift of centroid of the same-type ellipses from 30h APF to adults. See Fig. S1 for the n of each group. (I) Probability distribution of the shortest distance in 3D space from PN dendritic pixels to the antennal lobe surface during development (left) and in adults (right). Each line represents data from an individual PN type, population mean ± s.e.m. (J) Schematics of two individual PN types during development (left) and in adults (right), viewed from +45˚ anterior and from a single vertical plane. Note that PN dendrites extend to the antennal lobe surface during development regardless of their surface-or-interior positions in adults.

Fig. 2.. During development, ORN axons take cell-type specific trajectories and contact cognate PN dendrites first on the antennal lobe surface.

(A) Adult antennal lobe schematic highlighting six glomeruli, corresponding to the six ORN types shown in (B–H). (B) Single ORN types (green, labeled by a membrane-targeted GFP driven by separate genetic drivers) at 30h APF (top and middle rows, same brains) and in adults (bottom rows). Top, single optical section from vertical plane with dash lines outlining the antennal lobe neuropil stained by the N-cadherin antibody (blue). Reconstructed from 3D image volumes where optical sections were taken horizontally. Arrows point from the antennal lobe center to the average positions of ORN axons. Trajectory angle θ is defined in the DL4 panel. Middle and bottom, maximum projection of horizontal optical sections of antennal lobes at 30h APF and adults, respectively. The yellow dotted line indicates the intersection with the vertical plane shown above. Scale bar = 20 μm. (C) Probability distribution of the axon’s angular position from single-type ORNs. Population mean ± s.e.m. For all genotypes, n ≥ 9. (D) Same as (C), but with each data curve aligned to its peak to minimize data variance between brains and more accurately reflect the width of the probability distribution. (E) Single optical section showing VA1d-ORNs (green, labeled by membrane-targeted GFP driven by a split-GAL4) and VA1d-PNs (magenta, labeled by membrane-targeted RFP driven by a split-LexA (Ting et al., 2011)). From left to right, anterior, middle, and posterior sections from the same brain. From top to bottom, data from 30h APF, 48h APF, and adults. Filled arrowheads indicate examples where ORN axons and PN dendrites overlap. Open arrowheads indicate examples where PN dendrites do not overlap with ORN axons. Scale bar = 10 μm in this and all other panels. (F) Probability distribution of the angular position of VA1d-ORNs and VA1d-PNs. Same definition of the angle θ as in (C). Only vertical planes with PN dendrites were analyzed. Population mean ± s.e.m.; n = 9. (G) Fraction of VA1d-PNs overlapping with VA1d-ORNs, as a function of the distance from PN pixels to antennal lobe surface. For a given distance on the x-axis, y value of 1 means that all the VA1d-PN dendrites within that distance bin match with VA1d-ORN axons. Population mean ± s.e.m. For all time points, n ≥ 8. (H–J) Same as (E–G), but with data from DC3-ORNs and DC3-PNs. For all groups, n = 12. (K) Schematics of the same ORN-PN pair during development (left) and in adults (right). Note that DC3-ORN axons and DC3-PN dendrites are present at the antennal lobe surface during development but not in adults.

Fig. 3.. Dendrites of adult-interior PNs remain at the antennal lobe surface in adults when most axons of cognate ORNs are rerouted during development.

(A) Schematics of the same ORN-PN pair during development (left) and in adult (right). Same as Fig. 2K. (B) Same as (A), but with ORN axons largely rerouted and missing cognate PNs during development (left). This could lead to adult-interior PNs remain at the surface in adult (right, indicated by a question mark). (C) Adult antennal lobe schematic labeling three glomeruli, corresponding to the three ORN-PN pairs shown in (D–J). Some glomeruli were omitted for visualization clarity. (D) Single optical section of DC4-ORNs (green, labeled by membrane-targeted GFP driven by a split-GAL4) and DC4-PNs (magenta, labeled by membrane-targeted RFP driven by a split-LexA) in a wild-type brain. Dashed lines outline the boundary of PN dendrites. Scale bar = 20 μm in this and all other panels. (E) Same as (D), but with the trajectory of DC4-ORN axons changed through genetic manipulations (Toll-7 overexpression; see Fig. S6). The arrowhead indicates DC4-PNs innervating the antennal lobe surface. (F) Probability distribution of the distance from DC4-PN dendritic pixels to the antennal lobe surface in 3D space. Mean ± s.e.m. For all genotypes, n ≥ 6. (G and H) Same as (D) and (E), but with DC3-ORNs and DC3-PNs shown in a vertical plane and a different genetic manipulation (Toll-6 and Toll-7 RNAi; see Fig. S6). (I and J) Same as (G) and (H), but with the trajectory of VA1d ORNs changed through genetic manipulations (Sema-2b RNAi and Toll-7 RNAi; see Fig. S6). (K) Same as (F), but with data from DC3-PNs. For all genotypes, n ≥ 11.

Fig. 4.. Summary of dimensionality reduction in the ORN-PN synaptic partner matching process.

(A) The distribution of glomeruli appears 3D in adults. (B) During development, both PN dendrites and ORN axons search for partners at the 2D antennal lobe surface. (C) The search space for an individual ORN type is further reduced to 1D because their axons follow a specific trajectory on the 2D antennal lobe surface. See text for detail.

Fig. 5.. The accuracy of ORN-PN synaptic partner matching correlates with the accuracy of ORN trajectories.

(A) Single optical sections showing VA1d-ORN axons from a vertical view during development (top) and horizontal view in adult (middle and bottom). Top, dash lines outline the antennal lobe neuropil. Arrows point from the antennal lobe center to the average positions of ORN axons. Images in the bottom row is a zoom-in from the dashed squares in the middle row. Bottom, dashed lines outline the boundary of PN dendrites. Arrowheads indicate ORN axons mismatching with cognate PN dendrites. Left column represents wild-type condition, other columns represent the three manipulation conditions: (1) Sema-2b RNAi; (2) Toll-7 RNAi; (3) Sema-2b RNAi and Toll-7 RNAi. See Table S1 for detailed genotypes. Scale bar = 20 μm (top and middle), 10 μm (bottom), in this and other panels. (B) Probability distribution of the angular position of VA1d-ORN axons in each condition at 30h APF. Population mean ± s.e.m. (C) Average angular position of VA1d-ORN axons in each condition. Same data as in (B). Circles indicate the averages of individual antennal lobes; bars indicate the population mean ± s.e.m. (D) Percentage of VA1d-ORN axons overlapping with VA1d-PN dendrites in adults. Circles indicate the average of individual antennal lobe; bars indicate the population mean ± s.e.m. (E to H) Same as (A to D), but for DA4l-ORNs and DA4l-PNs. The three manipulation conditions are: (1) Sema-2b RNAi; (2) Toll-6 RNAi, Toll-7 RNAi, and Sema-2b overexpression; (3) Toll-6 RNAi and Toll-7 RNAi. Note that due to limitations on the reagents, the ORN signals from the top row result from a combination of three ORN types: DA4l, DA4m, and DC1, all of which take a similar trajectory (Fig. S3). (I–L) Same as (A–D), but for DL4-ORNs and DL4-PNs. The two manipulation conditions are: (1) Sema-2b overexpression; (2) Toll-6 RNAi, Toll-7 RNAi, and Sema-2b overexpression. (M–P) Same as (A–D), but for DC3-ORNs and DC3-PNs. The two manipulation conditions are: (1) Toll-7 RNAi; (2) Toll-7 RNAi and Sema-2b RNAi. (Q) Percentage of ORN-PN mismatch in adults as a function of the absolute angular changes in ORN axon trajectory at 30h APF. The black dot indicates wild type in each ORN type, which is the origin (x = 0, y = 0) in the plot by definition. The dash line indicates the linear fit. Pearson correlation coefficient = 0.88; p = 3.6 × 10⁻⁴.