Figure 4. Odor encoding in descending neuron (DN) populations.

(a) Neural residuals—obtained by subtracting convolved behavior regressors from raw neural data—can predict the presence of an odor significantly better than behavior regressors convolved with a calcium response function (‘Behavior’). Two-sided Mann–Whitney U-test. The classification score was obtained using a linear discriminant classifier with cross-validation. Shown are five trials for five animals (color-coded). (b) Matrix showing the cross-validated ridge regression unique explained variance (UEV) of a model that contains behavior and odor regressors for one animal. The first row (‘Behavior’) shows the composite $R^2$ for all behavior regressors with odor regressors shuffled. The second and third rows show the UEVs for regressors of the odors methyl salicylate (MSC) or apple cider vinegar (ACV), respectively. Colored asterisks indicate neurons illustrated in panel (a). (c, d) Example DNs best encoding (c) MSC or (d) ACV, respectively. Overlaid are traces of neural activity (gray), row one in the matrix (blue), and row one with odor data shuffled (red). (e, f) Locations of odor encoding neurons in (e) one individual and (f) across all five animals. Scale bar is 10 μm.

Figure 4—figure supplement 1. Odor-modulated behaviors and encoding in descending neuron (DN) populations across individuals.

(a) The probabilities that classified behaviors occur during periods of stimulation with humidified air, apple cider vinegar (ACV), or methyl salicylate (MSC) odors. P-values indicate the significance level for a two-sided Mann–Whitney U-test. (b–d) Swarm plots showing a subset of 250 randomly sampled points indicating (b) walking speed, (c) turning angular velocity, and (d) absolute value of turning angular velocity during periods of stimulation with humidified air, ACV, or MSC odors. (e) Matrices showing the cross-validated ridge regression unique explained variance (UEV) of models that contain behavior and odor regressors for five individual animals. The first row (‘Behavior’) shows the composite $R^2$ for all behavior regressors and shuffled odor regressors. The second and third rows show UEVs for the odor regressors MSC and ACV, respectively. (f) Locations of odor encoding neurons (UEV > 5%) across five individual animals. Scale bar is 10 μm and applies to all images. Color indicates odor. Radii scale with the amount of variance explained.

Figure 4—figure supplement 2. Largely identical descending neuron (DN) populations are recruited during walking and head grooming irrespective of odor context.

(a, b) Amount of (a) walking or (b) head grooming variance explained by each DN using only frames during presentation of humidified air, apple cider vinegar (ACV), or methyl salicylate (MSC). Data during humidified air presentation were split into groups to match the amount of data available during ACV and MSC presentation. Indicated are cases where a two-sided Mann–Whitney U-test comparing the cross-validation folds between humidified air and each of the odors yielded significant differences after Bonferroni correction for multiple comparisons (*${\displaystyle \mathrm{p}<0.05}$; ***${\displaystyle \mathrm{p}<0.001}$). (c) Individual region of interest (ROI) data points resulting in significant differences across conditions by a two-sided Mann–Whitney U-test (shown from left to right: fly 1, ROI 82; fly 1, ROI 87; fly 2, ROI 74; fly 4, ROI 28). Each point is the result of a single cross-validation fold from a single trial. p-Values shown are after Bonferroni correction for multiple comparisons.