Higher-order unimodal olfactory sensory preconditioning in Drosophila

Learning and memory storage is a complex process that has proven challenging to tackle. It is likely that, in nature, the instructive value of reinforcing experiences is acquired rather than innate. The association between seemingly neutral stimuli increases the gamut of possibilities to create meaningful associations and the predictive power of moment-by-moment experiences. Here, we report physiological and behavioral evidence of olfactory unimodal sensory preconditioning in fruit flies. We show that the presentation of a pair of odors (S1 and S2) before one of them (S1) is associated with electric shocks elicits a conditional response not only to the trained odor (S1) but to the odor previously paired with it (S2). This occurs even if the S2 odor was never presented in contiguity with the aversive stimulus. In addition, we show that inhibition of the small G protein Rac1, a known forgetting regulator, facilitates the association between S1/S2 odors. These results indicate that flies can infer value to olfactory stimuli based on the previous associative structure between odors, and that inhibition of Rac1 lengthens the time window of the olfactory ‘sensory buffer’, allowing the establishment of associations between odors presented in sequence.


Introduction
property to elicit conditional responses even though they have never been in contiguity with a reinforcer. Sensory preconditioning is one example of higher-order conditioning (Giurfa, 2013, (Heisenberg et al., 1985;Dubnau and Tully, 2001;McGuire et al., 2001). During associative memory acquisition, positive or negative values are assigned to odors by associated reward and 105 punishment respectively. This reinforcement is achieved by the coincident activation of a sparse of the environment (Jones et al., 2012). It was also reported that expressing Rac1 N17 in KC significantly enhances trace conditioning, in which an odor is associated with an electric shock sensory preconditioning in the MBON-γ1pedc>α/β of control animals. That is, the MBON-γ1pedc>α/β does not display a decreased response to the S2 odor ( Figure 1C), but does however, when we shorten the sensory preconditioning interval to 1s, the MBON-γ1pedc>α/β 160 neurons in control flies exhibit a decreased calcium response to the S2 odor, and to the S1 odor 161 that is expected ( Figure 1C). Throughout these experiments, MBON-γ1pedc>α/β responses to 162 naïve odors remained unchanged, indicating the decrease in calcium response to the S2 odor is 163 specific to the preconditioning phase, and not the result of a generalized depression to all odors 164 ( Figure 1C). This suggests that shortening the ISI window during sensory preconditioning allows 165 control flies to form some association between the S1 and S2 odors, thus resulting in an inferred 166 predictive value of the S2 odor after animals were aversively conditioned to the S1 odor. These 167 results demonstrate that flies show, at least at the physiological level, sensory preconditioning.

169
In contrast to control animals, flies expressing dominant negative Rac1 N17 in the KC had 170 previously been found to exhibit a decreased calcium response to the S2 odor when using an 171 ISI of 30 s ( Figure 1D, Figure 1-figure supplement 1B). This was also true when the sensory 172 preconditioning ISI was decreased to 1s ( Figure 1D). The data from our control flies led us to 173 speculate that the timing of the S1/S2 ISI dictates whether an association between odors S1 174 and S2 (MCH and OCT) occurs during the sensory preconditioning stage. Secondly, from prior 175 data on Rac1's role in trace conditioning and our sensory preconditioning data, we suspected 176 that Rac1 inhibition in the KC extends the time period that allows for the S1/S2 association to 177 occur. If this was true, we predicted that extending the ISI should eliminate sensory 178 preconditioning in Rac1DN flies. When the ISI was increased to 5 min, no sensory 179 preconditioning was observed in the Rac1DN animals ( Figure 1D). Again, naïve odor 180 responses in these flies were unchanged. This widening of the ISI window to allow for an S1/S2 association is dependent on the inhibition of Rac1, since flies with the same genotype but raised We tested whether this suspected sensory preconditioning-induced calcium depression might 187 be the result of increased habituation of the non-associated S2 odor. We recorded olfactory 188 responses in flies exposed to the same odor protocol except that the unconditioned stimulus 189 was excluded. Results showed no depression to either S1 or S2 odor presented (

211
To further challenge the assertion that the reduced calcium response in MBON-γ1pedc>α/β 212 neurons to the S2 odor is a result of sensory preconditioning, we trained animals with forward or 213 backwards conditioning but excluded the pre-training odor presentation of S1 and S2. When we 214 eliminated the preconditioning association between the S1 and S2 odors, the results showed 215 significant depression to the classically conditioned S1 odor that was associated with the shock, 216 but no depression to any other odors ( Figure 3). These results demonstrate that the depression 217 to the non-trained S2 odor was not due to a general depression to odors or generalization of 218 learned aversion. Instead, depression of the non-trained S2 odor is dependent on the animal 219 learning the association between S1 and S2 during the preconditioning stage. These results 220 further support the hypothesis that we are observing physiological sensory preconditioning.

222
We then tested if sensory preconditioning could be observed at the behavioral level. For this, we 223 trained the flies using an Arduino microcontroller for precise odor delivery by controlling 224 solenoids automatically. Using this Arduino system, we trained animals as follows (Figure 4-225 figure supplement 1A). Control flies were presented with a single pairing of the odors (5 s pulse 226 each) with a 1 s ISI. Using the standard aversive olfactory conditioning paradigm, flies were 227 then conditioned by the presentation of 1 min of S1 odor along with 12, 90 V shocks. Memory 228 was tested right after training in a T-maze by presenting animals with either the S1 (shock-229 paired odor) and a novel odor (NO), or the S2 (non-shocked, pre-paired odor) and a NO. As in 230 canonical aversive olfactory conditioning experiments, each behavioral experiment was 231 conducted using the reciprocal odor as S1. The final performance index (PI) was calculated by 232 averaging the PI for each odor used as S1. Results were compared to flies trained with backwards conditioning. Despite observing evidence of sensory preconditioning by functional imaging using a similar protocol, we could not observe a behavioral memory to the non-shocked 235 pre-paired odor (S2) (Figure 4-figure supplement 1B). At a behavioral level, sensory 236 preconditioning has previously been observed in animals after repeated presentation of a pair of 237 sensory cues. Thus, we preconditioned flies with ten repeated presentations of odor pairings 238 (S1/S2) before training ( Figure 4A). Using an ISI of 1 s, this resulted in a significant behavioral  These results were We next tested flies expressing Rac1 N17 in KC for evidence of behavioral sensory our physiological data, while a 30 s ISI resulted in no aversive memory to S2 in control flies, a 260 significant aversive memory was observed in flies expressing Rac1 N17 ( Figure 4B). This is 261 consistent with our calcium imaging results in Figure 1B

272
Taken together, these results indicate that flies can undergo unimodal sensory preconditioning 273 and that the small G protein Rac1, previously implicated in memory forgetting, gates the 274 windows for this S1-S2 association.

277
Our results demonstrate that flies can infer value to non-reinforced odors based on the previous 278 associative structure between odors. Memories are not discrete bits of events and associations 279 that are added and accumulated in our brains. Rather, memories are dynamic; they undergo 280 modifications and updates and are integrated into a coherent story that continuously 281 incorporates new information to predict upcoming events better and to inform decision making.
fundamental. Here we present evidence that even the simple brain of a fly can form these associations. reinforcement? If this is true, one could suggest a role for dopamine in this reinforcement. We 289 know that dopaminergic neurons are essential for many types of reinforcement and even for

297
Another interesting finding is that while a single odor pairing is sufficient to induce sensory 298 preconditioned-related plasticity in the MBON-γ1pedc>α/β compartment, it is not enough for the 299 S2 odor to drive the learned behavior. As mentioned above, we suggest that the presentation of

319
A challenging question that remains for future study is on the nature of the circuit 320 connections that link S1 and S2 during the sensory preconditioning phase. Once this 321 S1-S2 link is formed, the key process by which we experimentally detect that link is as 322 an aversive response to S2 during the testing phase. Does this aversive response to S2 323 result during the associative learning phase, in which the S1 presentation summons an 324 S2-representation during CS/US pairing, and therefore, S2 forms a direct association 325 with the punishment? Conversely, is the aversive response to S2 dependent on the 326 memory retrieval phase, during which exposing flies to the S2 odor somehow recalls the 327 S1-punishment association to influence the aversive response? This second possibility 328 would imply that S2 is not directly associated with punishment. Deciphering the neural 329 circuit rules that support unimodal sensory preconditioning in the relatively simple fly 330 brain will help us to unravel how other animals, including humans, use our nervous 331 systems to build models of the world. 332

Material and Methods
Flies were cultured on standard medium at room temperature. Crosses, unless otherwise    into a metal pipette to immobilize the head using proboscis aspiration. Once the head is immobilized, using a micromanipulator, the fly was inserted in a narrow slot the width of their and the proboscis fixed with myristic acid to avoid brain movement during proboscis extension.

364
Using a syringe needle, a small, square section of dorsal cuticle was removed from the head to 365 allow optical access to the brain. Fresh saline (103 mM NaCl, 3 mM KCl, 5 mM HEPES, 1.5 mM 366 CaCl 2 , MgCl 2 , 26 mM NaHCO 3 , 1 mM NaH 2 PO 4 , 10 mM trehalose, 7 mM sucrose, and 10 mM 367 glucose [pH 7.2]) was perfused immediately across the brain to prevent desiccation and ensure 368 the health of the fly. Then the fat bodies and trachea above the brain was removed. Using a 20X                          Back Forw S1 S2 S1 S2 S1 S2 S1 S2

669
The following source data is available for figure 3:                were preconditioned by a single presentation of a S1/S2 odor pair with one second ISI. Later flies were aversively trained to MCH (S1) by pairing a one minute odor presentation with 12 90V, 1.25 s shocks. Right after training memory was tested in a T-maze by presenting either S1   of PA and EL (S1 and S2) with one second ISI; later flies were aversively trained to PA (S1) and