Odor discrimination in Drosophila: from neural population codes to behavior

Abstract

Taking advantage of the well-characterized olfactory system of Drosophila, we derive a simple quantitative relationship between patterns of odorant receptor activation, the resulting internal representations of odors, and odor discrimination. Second-order excitatory and inhibitory projection neurons (ePNs and iPNs) convey olfactory information to the lateral horn, a brain region implicated in innate odor-driven behaviors. We show that the distance between ePN activity patterns is the main determinant of a fly's spontaneous discrimination behavior. Manipulations that silence subsets of ePNs have graded behavioral consequences, and effect sizes are predicted by changes in ePN distances. ePN distances predict only innate, not learned, behavior because the latter engages the mushroom body, which enables differentiated responses to even very similar odors. Inhibition from iPNs, which scales with olfactory stimulus strength, enhances innate discrimination of closely related odors, by imposing a high-pass filter on transmitter release from ePN terminals that increases the distance between odor representations.

Figure 1

Population Representations of Odors (A) ORN (top) and ePN (bottom) representations of 1-octanol (left), ethyl acetate (center), and 2-heptanone (right). The odor responses of 24 ORN classes (expressing the odorant receptors indicated on top) were measured (Hallem and Carlson, 2006; Hallem et al., 2004). ePN responses were predicted from these measurements with the experimentally supported equation (Olsen et al., 2010):$R_{ePN}=R_{\max }\frac{{R_{\mathit{ORN}}}^{1.5}}{{R_{\mathit{ORN}}}^{1.5}+{\sigma }^{1.5}+{\left(m/190\sum R_{ORN}\right)}^{1.5}}$Here, $R_{ePN}$ is the firing rate of a particular class of ePN, $R_{ORN}$ is the firing rate of the cognate class of ORN, $R_{\max }$ is the maximal possible ePN firing rate, σ is a constant, and m is an inhibitory scaling factor. The following parameter values were used: $R_{\max }$ = 165 spikes per s, m = 10.63, and σ = 12 spikes per s (Luo et al., 2010; Olsen et al., 2010). The input-output relationship described by the equation is depicted graphically for the three odors. Note that the transformation differs between odors because inhibitory scaling depends on $\sum R_{ORN}$. (B) Responses in ePN projections to the LH were evoked with 5 s pulses of odors and imaged by two-photon microscopy. Flies carried GH146-GAL4:UAS-GCaMP3 transgenes. Examples of individual responses to the six indicated odors (right) are contrasted with a common reference—the response to 1-octanol (left). The activity maps are pseudocolored according to the key on the right. (C) Correlation distances between the experimentally determined response maps are linearly related to calculated Euclidean distances between ePN activity vectors (mean ± SEM, R² = 0.5334, p < 0.0001, n = 13 flies). See also Figure S1.

Figure 2

The Distance-Discrimination Function (A) Movement traces depicting the position of a single Canton-S fly in a behavioral chamber (horizontal dimension) as a function of time (vertical dimension). The same fly was tested between a common reference odor, 1-octanol (orange), and the indicated test odors (blue). The data are arranged by increasing Euclidean and cosine distances between the respective ePN activity vectors. (B) Decision bias scores of 20 Canton-S flies tested against seven odor combinations. Orange symbolizes a preference for 1-octanol, and blue symbolizes a preference for the comparison odor; the intensity of shading represents the magnitude of bias according to the key on the left. (C) Absolute magnitude of the decision bias scores depicted in (B). The intensity of shading represents the magnitude of bias according to the key on the left. (D) Absolute decision bias scores elicited by 51 odor pairs as functions of Euclidean (left) and cosine (right) distances between ePN signals (mean ± SEM, n = 40–80 flies per data point). The distance-discrimination functions (dotted lines) were obtained from least-squares logistic fits to the data; the fits were constrained to include the origin (Euclidean distance: R² = 0.6577, p < 0.0001; cosine distance: R² = 0.6693, p < 0.0001). Shading indicates the area bounded by the distance-discrimination function where decision bias scores are predicted to fall. See also Table S1. (E) Absolute decision bias scores elicited by 36 odors against air (red) as functions of Euclidean (left) and cosine (right) distances between ePN signals (mean ± SEM, n = 40–80 flies per data point). ePN signals in air were calculated from measured spontaneous ORN activity (Hallem and Carlson, 2006; Hallem et al., 2004). The distance-discrimination function and experimental measurements obtained in (D) are reproduced for comparison (gray). See also Table S2 and Figure S2.

Figure 3

Experimental Manipulations of Distance-Based Discrimination (A) Maximum intensity projection of 117 confocal sections (1.5 μm) through the central brain of a fly carrying NP3062-GAL4:UAS-mCD8-GFP transgenes. (B) Single confocal sections from anterior (B₁) to posterior (B₃) of the antennal lobe region indicated in (A). (C) Absolute decision bias scores of flies carrying NP3062-GAL4:UAS-shi^ts1 transgenes as functions of Euclidean or cosine distances between ePN activity vectors (mean ± SEM, n = 30–40 flies per data point). For each odor pair, colored arrows indicate the behavioral change caused by shifting flies from the permissive to the restrictive temperature. The distance-discrimination functions of WT flies, obtained in Figure 2D, are reproduced for reference. The decision bias scores of NP3062-GAL4:UAS-shi^ts1 flies differ significantly between the permissive and restrictive temperatures, as predicted from the reduction in ePN distances (p = 0.0109 and 0.0411 for Euclidean and cosine distance, respectively; F test). (D) Maximum intensity projection of 111 confocal sections (1.5 μm) through the central brain of a fly carrying NP1579-GAL4:UAS-mCD8-GFP transgenes. (E) Single confocal sections from anterior (E₁) to posterior (E₃) of the antennal lobe region labeled in (D). (F) Absolute decision bias scores of flies carrying NP1579-GAL4:UAS-shi^ts1 transgenes as functions of Euclidean or cosine distances between ePN activity vectors (mean ± SEM, n = 40 flies per data point). For each odor pair, colored arrows indicate the behavioral change caused by shifting flies from the permissive to the restrictive temperature. The distance-discrimination functions obtained in Figure 2D are reproduced for reference. The decision bias scores of NP1579-GAL4:UAS-shi^ts1 flies differ significantly between the permissive and restrictive temperatures, as predicted from the reduction in ePN distances (p < 0.0001 for Euclidean and cosine distances; F test). See also Table S3 and Figure S3.

Figure 4

Innate versus Trained Discrimination Absolute decision bias scores of flies carrying mb247-LexA:lexAop-shi^ts1 transgenes (mean ± SEM, n = 40–60 flies per data point) as functions of Euclidean or cosine distances between ePN activity vectors. The distance-discrimination functions obtained in Figure 2D are reproduced for reference. (A) Innate discrimination at the restrictive temperature (32°C) when synaptic output from KCs is blocked. Absolute decision bias scores as functions of Euclidean (A₁) or cosine (A₂) distances between ePN activity vectors (mean ± SEM, n = 30–60 flies per data point). (B) Avoidance of the innately more aversive odor in a pair was reinforced during a 1 min cycle of electric shock training at the permissive temperature (25°C). After a 15 min rest interval, odor discrimination was analyzed at either the permissive or the restrictive temperature when synaptic output from KCs is intact (25°C, blue) or blocked (32°C, red). Absolute decision bias scores as functions of Euclidean (B₁) or cosine (B₂) distances between ePN activity vectors (mean ± SEM, n = 30–60 flies per data point). See also Table S4 and Figure S4.

Figure 5

Excitatory and Inhibitory Projections from the Antennal Lobe to the MB and LH (A) Maximum intensity projection of 119 confocal sections (1.5 μm) through the central brain of a fly carrying Mz699-GAL4:UAS-mCD8-GFP transgenes. The Mz699 enhancer element labels ∼39 ventral iPNs (D) and ∼86 cells in the vlpr (E). (B) Maximum intensity projection of 113 confocal sections (1.5 μm) through the central brain of a fly carrying GH146-GAL4:UAS-GCaMP3 transgenes. The GH146 enhancer element labels ∼90 mostly excitatory dorsal and lateral PNs. (C) Maximum intensity projection of 67 confocal sections (1.5 μm) through the central brain of a fly carrying Mz699-GAL4:UAS-^GFPDSyd-1;UAS-DenMark transgenes. ^GFPDSyd-1 (magenta) labels presynaptic terminals. Fluorescence in the LH originates mainly from iPN axons, whereas signal in the vlpr arises from ipsi- and/or contralateral projections of vlpr neurons. The vlpr cells may also elaborate presynaptic sites in the LH, but these are obscured by the strong iPN signal. DenMark (cyan) labels putative dendritic regions. Although iPN dendrites are found exclusively in the antennal lobes, vlpr neurons receive their main input in the LH. Faint DenMark labeling suggests additional weak dendritic sites of vlpr cells in the vlpr. See Movie S1 for the complete image stack. (D and E) Confocal sections through the central brain of a fly carrying Mz699-GAL4:UAS-mCD8-GFP transgenes, after immunostaining against GABA (magenta, left column) and GFP (cyan, center column); colocalization of both markers results in white structures in the overlay images on the right. In some images, nuclei are counterstained with TOTO-3 (yellow). The approximate positions of the imaged areas are indicated in (A). The images were acquired and are displayed at different photomultiplier gain and contrast settings. (D) An individual confocal section (1 μm) shows GABAergic iPNs. See Movie S2 for the complete confocal image stack. (E) A maximum intensity projection of 70 confocal sections (1 μm) demonstrates the absence of GABA staining in vlpr neurons. See Movie S3 for the complete confocal image stack. (F and G) Two-photon imaging of odor-evoked calcium transients in flies carrying MZ699-GAL4:UAS-GCaMP3 transgenes. (F) Single-trial responses of iPN axons to 5 s pulses of nine different odors (black bar). The traces, which were recorded in the same fly, are aligned to the time of odor onset and color-coded according to the number of glomeruli an odor activates. (G) Integrated fluorescence transients (area under the fluorescence trace during a 5 s odor pulse) in iPN axons as a function of the number of glomeruli an odor activates (mean ± SEM, R² = 0.9278, p < 0.0001, n = 11 flies per data point). See also Figure S5 and Movies S1–S3.

Figure 6

iPN Inhibition Facilitates Odor Discrimination (A) Absolute decision bias scores of flies carrying Mz699-GAL4:UAS-shi^ts1 transgenes as functions of Euclidean or cosine distances between ePN activity vectors at the permissive temperature of 25°C when iPN-mediated inhibition is intact (mean ± SEM, n = 40–60 flies per data point). The distance-discrimination functions obtained in Figure 2D are reproduced for reference. The distance-discrimination functions of flies carrying Mz699-GAL4:UAS-shi^ts1 transgenes are identical to those of WT flies at the permissive temperature (p = 0.9895 and 0.9813 for Euclidean and cosine distance, respectively; F test). (B) Absolute decision bias scores of flies carrying Mz699-GAL4:UAS-shi^ts1 transgenes as functions of Euclidean or cosine distances between ePN activity vectors at the restrictive temperature of 32°C when iPN-mediated inhibition is blocked (mean ± SEM, n = 40–60 flies per data point). The distance-discrimination functions in the absence of inhibition (red lines) were obtained from least-squares logistic fits to the data; the fits were constrained to include the origin (Euclidean distance: R² = 0.6779, p < 0.0001; cosine distance: R² = 0.5538, p < 0.0001). The distance-discrimination functions obtained in Figure 2D (dotted lines) are reproduced for reference. The distance-discrimination functions of flies carrying Mz699-GAL4:UAS-shi^ts1 transgenes differ significantly between the permissive and restrictive temperatures (p = 0.0058 and 0.0097 for Euclidean and cosine distance, respectively; F test). See also Table S5 and Figure S6A. (C) Absolute decision bias scores of flies carrying Mz699-GAL4:UAS-GAD-RNAi;UAS-vGAT-RNAi transgenes as functions of Euclidean or cosine distances between ePN activity vectors (mean ± SEM, n = 40–60 flies per data point). RNA-mediated interference with the expression of GAD and vGAT in iPNs changes the distance-discrimination functions in the same manner as blocking iPN synaptic output (see B; p = 0.3326 and 0.8711 for Euclidean and cosine distance, respectively; F test). The distance-discrimination functions of flies carrying Mz699-GAL4: UAS-vGAT RNAi, UAS-GAD RNAi transgenes differ significantly from those of WT flies (p = 0.0043 and 0.0263 for Euclidean and cosine distance, respectively; F test). See also Table S6 and Figure S6A. (D) Odor preferences against air of Canton-S flies and of flies carrying Mz699-GAL4:UAS-shi^ts1 transgenes at the restrictive temperature are identical (mean ± SEM, n = 40–60 flies per data point) except for pentyl acetate and 2-heptanone (^∗p < 0.05; Bonferroni-corrected t test). See also Table S2 and Figures S2A and S6B.

Figure 7

iPN Inhibition Imposes a High-Pass Filter on ePN Synaptic Output (A) Raw two-photon images of spH fluorescence in ePN projections to the LH at the indicated stimulation frequencies, in the absence (top) or presence (bottom) of 50 μM GABA. ePN axons were stimulated for 5 s by passing 1 ms pulses of current via an extracellular electrode. (B) Average spH fluorescence changes in ePN projections to the LH, evoked by electrical stimulation at the indicated frequencies, in the absence (black) or presence (red) of 50 μM GABA (mean ± SEM, n = 5 flies). (C) Average ratio of integrated spH fluorescence transients (areas under the fluorescence traces during 5 s electrical stimulation) in the presence and absence of 50 μM GABA (mean ± SEM, n = 5 flies). The ratios of ΔF/F at 0 versus 50 μM GABA differ across frequencies (p < 0.0001; one-way repeated measures ANOVA). Asterisks indicate significant differences between the presence and absence of 50 μM GABA at specific frequencies (^∗p < 0.05, ^∗∗p < 0.005; paired t test). (D) Thermally evoked iPN activity has a similar effect on ePN synaptic release as bath application of 50 μM GABA. Flies carried GH146-QF, QUAS-spH, Mz699-GAL4, and UAS-dTRPA1 transgenes. spH fluorescence changes were measured at two electrical stimulation frequencies (40 and 130 Hz) while flies were held at 25°C and 32°C. Columns depict the ratios of the integrated spH fluorescence transients (areas under the fluorescence traces during 5 s electrical stimulation trains) between 32°C and 25°C (mean ± SEM, n = 7–8 flies). A ratio of 1 indicates no effect of thermally evoked iPN activity on ePN synaptic release, whereas a ratio <1 indicates that iPN activity inhibits ePN output. Red brackets denote significant differences (p < 0.05, with Bonferroni-corrected paired t tests to compare the 32°C:25°C ratios at 40 versus 130 Hz within genotypes and one-way ANOVA with a Tukey-Kramer post hoc test to compare the ratios of the 32°C:25°C ratios at 40 versus 130 Hz across genotypes).

Figure 8

iPN Inhibition Increases ePN Distances (A and C) Euclidean (A) and cosine (C) distances between odors were calculated with the empirically derived transmission characteristics of the inhibitory high-pass filter (Figure 7C). The blocking strength of the filter was linearly adjusted according to the number of active glomeruli (Figure 5G). The scatter plot relates the 5,995 possible pairwise Euclidean distances between 110 odors after filtering to their unfiltered counterparts. (B and D) Histograms of the effect sizes of inhibitory high-pass filtering on the Euclidean (B) and cosine (D) distances between 5,995 odor pairs. The filter causes mean increases in Euclidean distance of 5.5 spikes per s (B) or in cosine distance of 0.14 (D). (E) Application of the empirically derived high-pass filter (Figure 7C) to the ePN activity vectors of two odors (point 1) stretches the cosine distance between the odors (point 2). According to the distance-discrimination model, this results in improved odor discrimination (point 3). Thus, the decision bias of WT flies with inhibition intact (point 4) is identical to the decision bias of flies in which iPN output is blocked but the distance-enhancing effect of inhibition is accounted for computationally (point 3). (F) Sequential applications of the inhibitory high-pass filter and the distance-discrimination model predict the empirical distance-discrimination function of WT flies (red line, reproduced from Figure 2D) from the empirical distance-discrimination function of flies carrying Mz699-GAL4:UAS-shi^ts1 transgenes at the restrictive temperature of 32°C (black line, reproduced from Figure 6B).