Automated Analysis of Calcium Signals in Bee Brains

Signals from a total of 14 bees were recorded and analysed. Data analysis was fully automated, based on custom MATLAB (R2019b, MathWorks) scripts. The fluorescence time series, containing an entire experimental sequence, were separated into periods of 3 s pre-stimulus, 3 s during the stimulus and 3 s post-stimulus for each trial. For each frame, we computed the relative fluorescence change $$\mathrm{\Delta }F/F(t)=-\frac{F\left(t\right)-F_b}{F_b},$$ via normalizing the raw fluorescence signal $F\left(t\right)$ by the average signal during the pre-stimulus period F_b. This signal is proportional to the relative change in calcium concentration and thus the neuronal firing rate^{64 (link)}. ΔF/F was averaged over the 10 trials for each odour. Next, the 2D activity maps were segmented for individual glomerular responses. The glomerular boundaries were obtained by recursively comparing three sources of information, the anatomical features from an additionally recorded 3D image stack (Supplementary Fig. 2a), the functional response maps (Supplementary Movie 2) and a regional homogeneity analysis that tests the correlation between the signals of each pixel and those from neighbouring pixels. This measure is high within individual glomeruli and falls off at their borders (Supplementary Fig. 2b). After coherently responding structures were segmented, the glomerular identity was determined using the digital 3D antennal lobe atlas 23. The analysis was limited to the 19 most frequently identified glomeruli. If the identity of individual glomeruli could not be determined with certainty, they were discarded, so that the total number of analysed glomeruli fluctuated across bees (Fig. 4c).
A statistical analysis of the subjects’ mean responses to single odours was performed via paired t-tests with FDR correction.