Expression patterns of 10 Gal4 lines used in this study. Green, Gal4 driven mCD8-GFP; Red, Gal4 driven nuclear LacZ or nuclear RFP; Blue, neuropil staining by nc82. Scale bars, 20 μm.

(a) Binary glomerular innervation patterns of 1532 singly labeled LNs organized by hierarchical clustering. Rows represent innervation patterns of individual cells; columns represent 54 glomeruli. 1489 are ipsilateral projecting LNs. Of the 43 bilateral projecting LNs, only the ipsilateral patterns are included in the clustering analysis. Yellow, glomeruli innervated; blue, glomeruli not innervated.
(b) Antennal lobe model. The 54 glomeruli we scored for this study are outlined in three sections of the antennal lobe from anterior to posterior. This model is modified after a number of sources3, 43 and derived from tracing nc82 stained brains.
(c-e) Three representative images from three selective regions of the LN cluster diagram indicated on the right side of Fig. 3a. mCD8-GFP staining in green, nc82 or N-cadherin staining in blue. Dashed lines in (C) mark pheromonal glomeruli, DA1, VA1d and VA1l/m. Scale bar, 20μM.
(f) The LN innervation probability of a glomerulus is positively correlated with the mean odor-evoked firing rate of the ORNs presynaptic to that glomerulus (r=0.63, p<0.005, n=23 glomeruli). ORN data from ref25. Firing rates averaged across 110 odors. Filled symbols represent trichoid glomeruli.
(g) Principal component analysis of LN glomerular innervation patterns. In the second dimension, the dumbbell subclass of LNs (bracket) distinctly separates from all other cell types. LNs labeled by different Gal4 lines are marked with different colors. See Supplementary Fig. 8 for color code of Gal4 lines and histograms of cell distributions in PC1 and PC2.

(a) (Top) A normal adult fly head with 2 antennae (AT, arrows) and 2 maxillary palps (MPs, arrowheads). (Bottom) Occasionally, eyFLP-induced smo clones eliminate both MPs (open arrowheads). Scale bars, 100μm.
(b-c) Brains from normal (top) and 0 MP (bottom) flies carrying Line5-Gal4 (b) or GH146-Gal4 (c) were labeled by nc82 and Gal4 driven mCD8GFP as indicated. The MP ORN target VA7l glomerulus (dotted circle) is not innervated by Line5 processes in 0MP flies, but is still innervated by PN processes. Arrows designate glomeruli that are innervated by GH146-negative PNs. Scale bars, 20 μm.
(d) Quantification of glomerular innervation by Line5-Gal4 LN and GH146-Gal4 PN processes in the presence or absence of ORN innervation of MP target glomeruli VA7l (top) and VC2 (bottom).
(e) Representative single sections of Line6 LN single cell clones after adult removal of MPs, or MPs and ATs. Control samples (uncut) show clear Line6 LN innervation of VA7l (top) and VC2 (bottom). After adult removal of MPs or both MPs and AT, Line6 LNs still innervate VA7l and VC2. Blue, nc82; green, mCD8-GFP; red, Syt-HA.
(f) Quantification of glomerular volume, LN process length and the number of Syt-HA puncta in control (WT), MP cut (-M) or MP and AT cut (-A-M) brains. The VA7l volume is significant decreased when ORN processes are removed, but neither process length nor the number of Syt-HA puncta in VA7l and VC2 is significantly reduced. Error bars represent SEM.

(a-e) Representative arborization patterns of single cell MARCM clones of ipsilateral projecting LNs. Each row represents examples from categories that innervate: (a) all glomeruli, (b) all but a few glomeruli, (c) regional, (d) patchy, (e) a few glomeruli.
(f) Representative arborization patterns of bilateral projecting LNs.
Green, mCD8-GFP that labels LN processes; red, synaptotagmin-HA, a marker for presynaptic terminals; blue, nc82 or N-cadherin staining that highlight glomerular structures. Images in c (left, forth and fifth from left) and f (third, fifth and sixth from left) are not stained for Syt-HA. Arrowheads indicate cell bodies. Dashed lines mark the outline of the antennal lobe. Scale bar, 20 μm. Information about corresponding Gal4 lines in this and subsequent figures can be found in Supplementary Table 4.

(a) Rasters show the similar spiking responses of two Line7 LNs. These LNs are typical of Line7 in having high spontaneous firing rates, weak odor-evoked excitation, and strong odor-evoked inhibition. The nominal odor stimulus period is denoted by the gray box; there is a delay of about 100msec before odor reaches the fly.
(b) Dissimilar responses of two Line9 LNs. The first of these innervated almost all glomeruli (52 of 54) and was mainly inhibited by all odors, whereas the second innervated a smaller subset of glomeruli and was excited by all odors. Overall, Line9 LNs were diverse.
(c) Functional properties of Lines5-9 (mean±SEM). All these properties were significantly dependent on the Gal4 line of the recorded LN (one-way ANOVA, p<0.0001). For each property, post hoc Tukey tests yielded significant differences (p<0.05) between some but not all of the 10 pair-wise comparisons between Gal4 lines. Odor-evoked firing rates are expressed as a change from the spontaneous firing rate.

(a) Odor responses of three pan-glomerular LNs.
(b) Pan-glomerular LNs (n=26) have significantly higher spontaneous activity (p<0.01) and weaker mean and maximum odor responses (p<0.05, t-tests) as compared to all other LNs that were successfully reconstructed (n=67).
(c) Odor responses of three LNs that avoid glomerulus VA1d ± DL3 but innervate all other glomeruli (“pheromone-avoiding”).
(d) Left: pheromone-avoiding LNs (n=9) fire a significantly higher percentage of their spikes during the first 100ms of the odor response as compared to all other LNs (n=84) (p<0.01, t-test; spikes counted during the period 100-200msec after nominal odor onset, divided by total spikes during the 1-sec period shown in rasters). Right: odor response time course is more transient for these LNs than for other LNs (mean peri-stimulus time histogram, ±SEM across cells).

(a) Hierarchical clustering of innervation patterns as in Fig. 3a but only for Line6 LNs (n=131). Cells in (b-e) are indicated. Glomeruli innervated by trichoid ORNs (pheromonal glomeruli) are highlighted in orange. K-means clustering verified that Line6 binary innervation patterns form a single cluster. Some cells were from animals in which antennae, maxillary palps or both had been removed for 10 days prior to fixation (Fig. 8e-f), but these treatments did not affect the number or the variability of glomerular innervation (Supplementary Fig. 11).
(b-e) Odor response of four Line6 LNs. Shaded regions of plots denote odor stimulus period (500 ms).
(f) Comparison of experimental and theoretical frequency of innervation (center black line) if glomeruli were randomly innervated. The envelope of ±2% is the standard error assuming a binomial distribution of innervation frequencies. The experimental frequency (blue dots) reveals that many glomeruli are almost always innervated, significantly above the theoretical distribution; other glomeruli are innervated significantly less frequently than the theoretical distribution. The glomerular identities to the right of DA1 are DM5, VA1m, VA1l, DL3 and VA1d; all except DM5 are pheromonal glomeruli.
(g) Quantification of innervation density of DM1, DA1 and VA1d, from randomly chosen 10 LNs of each class that innervate all three glomeruli. Innervation density=total dendritic length in glomerulus/glomerulus volume. Innervation of DA1 and VA1d by Line 6 LNs is significantly reduced compared to control LNs (pan-glomerular randomly selected from Lines 1, 3, 5), Tukey test, p<0.05.

(a) Clustering of 161 patchy cell innervation patterns as in Fig. 3a. No two cells have identical innervation patterns.
(b) Schematic of MARCM-FLPout that allows two sister cells to be labeled by different colors. In this genetic method, UAS-FRT-CD2-FRT-mCD8GFP40 serves as a reporter of Gal4. After FLP-mediated mitotic recombination causes the loss of Gal80 in the ganglion mother cell (GMC), CD2 should be expressed in both daughter cells derived from this GMC. However, if an additional FLPout event occurs in one of the two daughter cells, this cell will express mCD8-GFP instead of CD2.
(c-h) Examples of two sister patchy cells (c-e) and two sister control cells (f-h) labeled by MARCM-FLPout shown with N-cadherin (blue), GFP (green) and CD2 (red) staining. (c, f) Projection of the entire antennal lobe. (d, e, g, h) High magnification of 5 μm projections from anterior (d, g) and middle (e, h) antennal lobe sections showing non-overlapping processes from 2 sister cells. Dashed lines mark the boundary of glomeruli VA1l/m (d), DL1 and DC2 (e). Scale bars, 20 μm.
(i) Sister cell overlap index as a function of edge length of the dilation kernel (see Methods). Dashed lines indicate the mean for patchy sister cells (right-shifted) and non-patchy sister cells (left-shifted). Colored lines indicate individual pairs of sister cells. A quantitative distinction between patchy and non-patchy classes was verified by K-means clustering into two clusters.