Associative learning allows animals to establish links between stimuli based on their concomitance. In the case of Pavlovian conditioning, a single stimulus A (the conditional stimulus, CS) is reinforced unambiguously with an unconditional stimulus (US) eliciting an innate response. This conditioning constitutes an ‘elemental’ association enabling to elicit a learnt response from A+ without US presentation after learning. However, associative learning may involve a ‘complex’ CS composed of several components. In that case, the compound may predict a different outcome than the components taken separately, leading to an ambiguity and requiring the animal to perform a so-called ‘non-elemental’ discrimination. Here we focus on such a non-elemental task, the negative patterning (NP) problem, and provide the first evidence of NP solving in Drosophila. We show that Drosophila learn to discriminate a simple component (A or B) associated to electric shocks (+) from an odour mixture composed either partly (called ‘feature-negative discrimination’ A+ vs. AB-) or entirely (called ‘NP’ A+B+ vs. AB-) of the shock associated components. Furthermore, we show that conditioning repetition results in a transition from an elemental to a configural representation of the mixture required to solve the NP task, highlighting the cognitive flexibility of Drosophila.

Fly rearing: The wild-type line used in this study was a Canton-Special (Canton-S) strain. Flies were raised on standard medium at 25°c, ~60% humidity and a 12h/12h light/dark cycle. The flies were kept in 36x82mm plastic tubes containing ~20mL of medium.

Olfactory conditioning: Odours were diluted in a bottle of mineral oil. Odours used were 3-Octanol (termed ‘A’ for the sake of simplicity, 2.27mM) and 4-Methylcyclohexanol (‘B’, 2.62mM) (figure 1 and figure 2). Benzaldehyde (‘C’, 1.89 mM) was used as a novel odour in some conditions (figure 2). IsoamylAcetate (‘D’) and EthylButyrate (‘E’) were used as alternative odours (electronic supplementary material, figure S2). Odours are prepared at the same concentration whether as components (A, B, D or E) or as mixtures (AB or DE), which is essential for studying non-elementary learning. In the case of mixtures, odours were diluted together in the same bottle of mineral oil. All odours along with the solvent were from Sigma Aldrich (France). The US consisted of twelve 1.5s pulses of 60V electric shock every 5s delivered through a metallic grid. Each experiment included two groups of ~30 flies (2 – 4 days old) and was performed using a semi-automated device based on a previous work. In a T-Maze, two main phenomena drive the preference of flies toward a compartment or another: on the one hand, the learnt information about the stimuli acquired during conditioning and on the other hand, the fact that odours are rarely completely neutral; at the concentrations used in our work, they are in fact repulsive to naïve flies. When two odorants are opposed in the absence of punishment in the T-maze, repulsion balances; yet, if one odorant is opposed to a compound more repulsive, a bias towards the less repulsive stimulus is visible. To disentangle learning from non-learning behavioural components, one of the groups experienced an explicit pairing of CS and US (‘paired group’) while the other group experienced both stimuli unpaired to prevent their association (‘unpaired group’).

Training: Each training trial consisted of 90 s of acclimatisation, after which flies were subjected to their respective conditioning protocol. Each of the two odours (CS) was presented once for 1min with an intertrial interval of 1min. For the paired group (figure 1a), one of the olfactory stimulus (CS⁺) was paired with the US, while the other stimulus (CS^-) remained unpunished. In the unpaired group, flies were exposed to 1min of either shocks or odours, separated by an interval of 1min. This sequence formed one conditioning cycle.

Flies were subjected to one of three training protocols (figure 1b): a differential conditioning (DC) in which they had to learn to discriminate a punished from a non-punished odour (A⁺ vs. B^-), a NF discrimination in which they had to learn to discriminate a punished odour from a non-punished odour compound (A⁺ vs. AB^-) and a NP discrimination in which they had to learn to discriminate two punished odours from a non-punished odour compound (A⁺, B⁺ vs. AB^-). 3-Octanol was always used as the CS+ for DC and NF protocols (with respectively 4-Methylcyclohexanol and 3-Octanol+4-Methylcyclohexanol as CS-). For NP protocol, both 3-Octanol and 4-Methylcyclohexanol were used as CS+ when presented alone and as CS- when presented as a compound. Thus, flies subjected to DC training faced a pure elemental discrimination. On the contrary, flies trained in the NP protocol could only solve the problem if they adopted a non-elemental strategy. Finally, flies subjected to NF training could solve the problem using either an elemental or a non-elemental strategy. For each protocol, training consisted of either one or five cycles, which allowed studying if the amount of experience gathered by flies promoted a particular discriminations strategy.

Test: After training, flies were transferred to a T-maze where they could choose between the CS⁺ and the CS^- in the absence of shock during 1min. Flies from paired and unpaired protocols were sequentially tested. At the end of the test, flies in each arm of the T-maze were counted. If paired flies learned the discrimination, they should be mostly located in the CS- arm, the arm presenting the odour stimulus that was not associated to the shocks during the training. A Performance Index (PI) was calculated as: (Number of flies in the CS^- arm – Number of flies in the CS⁺ arm) / total Number of flies. To control for any experimental bias, each replicate consisted of a ‘paired-group’ PI (reflecting associative learning + bias) from which an ‘unpaired-group’ PI (reflecting bias only) was substracted. 

Please note that no data sheet is available for the electronic supplementary material figure S1 because we used the same data as in Fig1, but we represented all scores (Paired/Unpaired/Corrected index) instead of only the corrected index to illustrate how comparing Paired and Unpaired scores allows for accurate assocative learning quantification. We also only plotted data after 5 conditioning trials in the figure S1.