Data analysis

All data analysis except the pre-processing of imaging data was done in R⁴⁶. All scripts were custom written. M17 data was analyzed by extracting the number of spikes during the 5 s baseline and during the 4 s odor stimulus period. The M17 response frequency was calculated for each odor stimulus and was normalized with the corresponding baseline frequency.

Imaging data were pre-processed with the ImageBee plugin for KNIME⁴⁷. Movement correction was performed for each bee first between images (i.e. frames) and then between videos (i.e. stimuli). Signals were calculated as the ratio of fluorescence at 340 and 380 nm: . The F_340/380 was then normalized to baseline levels by subtracting the average F_340/380 of the first 40 frames (i.e. before odor onset). For glomeruli detection, videos were processed as follows: A Z-score normalization was performed, images were smoothed with a Gaussian filter, a principal component analysis was run and a convex cone algorithm was used as described elsewhere⁴⁷. The map of glomeruli obtained by this procedure was than overlaid with the F_340/380 calculations. The response of each glomerulus over time was calculated by averaging all pixels in the identified area. On average, 15 glomeruli could be analyzed per bee (Table 1). Bees which showed strong movement during one of the stimuli were excluded from the equivalent part of the analysis (i.e. test or extinction).

We calculated the Euclidean distance from the glomerular responses⁴⁸ for each individual bee. We determined the glomeruli responding to each stimulus as described before³². All glomeruli exceeding 3× SD of the period before odor onset were counted as responsive. We determined the two most active glomeruli (dominant glomeruli) during the peak response for the CS+, new odor and first extinction trial for each individual bee. We pooled the response of those two glomeruli and calculated the mean and SEM across bees. We assessed the two strongest instead of all responding glomeruli as this method avoids introducing a bias caused by the reduced number of active glomeruli in the new odor after RG108 treatment (Fig. 3). Additionally, as each individual bee was trained with either 1-hexanol or 1-nonanol, the identity of the CS+ and new odor was different across bees. These two odors differ in which and how many glomeruli are activated^2,49,50. Baseline response levels were not different between treatments or training groups (Supplemental Fig. 1).

We tested the data for normal distribution using Shapiro-Wilk tests and for equal variance using F-tests. Statistical significance was tested using a t-test, if data was normally distributed and had equal variance. Otherwise, data was tested using a non-parametric test: Mann-Whitney U for unpaired and Wilcoxon signed rank test for paired data. Two-tailed tests were used in all cases, except if a prior hypothesis about the directionality of an effect existed, in which case it is stated in the respective result section.

As bees were reared in their natural environment, inter-individual variation in olfactory responses is present and expected due to prior experiences of individual bees. Such variation allows us to identify biologically relevant effects of Dnmt activity in the bee brain. Importantly, this type of variation rather masks, than over-emphasizes effects on the group level.

The effect size (Cohen’s D) was calculated for all effects reaching the 0.05 significance level. As a guideline effects with sizes below 0.2 were defined as negligible, between 0.2–0.5 as small, between 0.5–0.8 as medium and above 0.8 as large⁵¹. The effect size can be used as an estimate of the real difference between the tested groups.