Optogenetic induction of appetitive and aversive taste memories in Drosophila

  1. Department of Zoology and Life Sciences Institute, University of British Columbia, Vancouver, Canada

Peer review process

Not revised: This Reviewed Preprint includes the authors’ original preprint (without revision), an eLife assessment, public reviews, and a response from the authors (if available).

Read more about eLife’s peer review process.

Editors

  • Reviewing Editor
    Ilona Grunwald Kadow
    University of Bonn, Bonn, Germany
  • Senior Editor
    Claude Desplan
    New York University, New York, United States of America

Reviewer #1 (Public Review):

The finding that taste memory formation follows the same or highly similar logic and mechanisms as olfactory memory is very interesting. In particular, the new approach to use an operant learning assay developed by the authors to address this outstanding question in the field is very impressive. The shown data are of high quality and very convincing.

While the current version will be of clear interest to fly people dissecting memory formation, it might be less accessible outside this immediate field. Below I list my suggestions, questions and criticisms.

You have developed an operant assay and stress this in the introduction. This is important because it allows you to gain much better inside into how memory is formed and how it is recalled. Nevertheless, I was somewhat disappointed that you did not exploit that aspect more in your study. First, I suggest showing, at least for the initial figures, the traces (e.g. Fig 1D) not only for the test phase but also for the training phase. As you also mention in your discussion, the extent of memory formation will depend critically on the number of pairings during training. And perhaps not only on their number but also on their evolution/change over time. Second, you only show preference indices. I suggest showing the number of actual interactions with the food source in addition. In my opinion and experience, the preference index can be misleading or at least the interpretation might be questioned if the number of actual choices is very low or very high compared to controls or other groups. Third, regarding the same point, you show traces for test phases, but you do not comment or discuss why they might look the way they look. For instance, it appears that in some cases it takes a while to see an actual difference in the preference index while at other times it seems more instantaneously etc.

Along the same lines, I am wondering why you do not observe extinction. Frequently if the CS is re-experienced without the US over several trials, you start to see memory fade. The preference traces as well as the actual interactions might help to explain this.

You use salt as a negative US. I suggest showing at least one experiment with bitter taste (e.g. quinine) to show how general your finding is to negative conditioning. Your optogenetic data suggests it is.

You analyze the role of energy state in memory formation. This is very interesting. In light of the importance of feeding state, it would be very helpful to include starvation/metabolic state information not only in the methods but also in the results section (at least briefly).

Your data convincingly shows that taste memory is formed in the mushroom body. For instance, you show that inhibition of KCs prevents the change in preference. KC inhibition was done during the entire experiment (training and test). Thus, it's important to show how KC inhibition affects (or does not) training vs. test.

Along the same lines, how do you envision this memory formation to happen at the circuit level? KCs and DANs are likely activated by CS and US. It would be important to at least include a paragraph in the discussion to clarify this.

Reviewer #2 (Public Review):

Jelen et al. developed a new taste conditioning paradigm where they pair a tastant (CS) with optogenetic activation of either sensory neurons or dopamine neurons. Activation of different cell types in training led to decreased sugar attraction or decreased salt avoidance. Depending on the activated cell type, the authors could even induce LTM with optogenetic activation. They found that the neural requirement for aversive or appetitive taste learning widely overlaps with the requirement for learning with other modalities (olfaction). They focus also on appetitive taste LTM formation, which requires caloric food intake after training similar to olfactory LTM.

Strengths:

The newly developed operant paradigm has several advantages compared to previous taste learning paradigms. The flies are freely walking and can be monitored throughout training and test. This allowed the authors to describe the temporal dynamics of learning and learned behavior. They could show that a specific type of dopamine neuron enhances salt sipping during training but was not sufficient to induce learning. Furthermore, they could now investigate both, appetitive and aversive learning, which was not possible before in immobilized flies. Optogenetic activation as the US in training allowed the authors to disentangle the need for caloric value in short-term and long-term memory.

Weaknesses:

Artificial activation of neurons seems to be sufficient to induce different memories in the fly. However, as the flies do not receive actual food in the training, those results may not represent the naturally used neural circuits, or only partial circuits underlying the normal taste learning. Also, the new paradigm has operant training, which might change the requirement or recruitment of learning circuits. Thus, the authors find similar neurons involved as in classical conditioning, which is very interesting, but also some differences.

Reviewer #3 (Public Review):

The manuscript by Jelen and colleagues aims at investigating the neuronal circuits underlying aversive and appetitive gustatory Pavlovian conditioning in Drosophila. To this end Jelen and colleagues employ an intuitive novel training device that allows for automatic optogenetic activation of neuron populations of choice upon physical contact of freely walking flies with a food source. Whereas olfactory appetitive or aversive conditioning is well established for Drosophila, learning paradigms using other sensory modalities like acoustic or gustatory stimulation are either not yet well established or cumbersome for the broad community. In this concern the advancement existing optogenetic gustatory learning setups to the automated optogenetic learning device "STROBE" finds a remedy to allow for high throughput experiments on gustatory learning in freely walking and behaving animals.

In the first part of their study Jelen and colleagues employ the experimental setup to induce an aversive memory in the fly to low concentrated sugar in combination with activation of bitter neurons optogenetically recapitulating earlier studies where low concentrated sugar was presented in combination with quinine. However, in contrast to those earlier studies where sugar was presented to the tarsi of the fly and quinine to the proboscis, allowing for differentiated detection of taste modalities in the study by Jelen and colleagues the presentation of both taste modalities are synchronous. Sugar and bitter appear to be sensed simultaneously and bitter neurons are globally activated once the fly gets in contact with the sugar solution. In this scenario it is difficult to understand how the bitter neuron activation does not directly interfere with the sensation of sugar changing the perception of the sugar itself instead of being sensed as a punitive stimulus. The strong aversion of the flies towards the sugar during the training phase may reflect indeed such a change in perception mixed with the learning. Further, during the first 5 - 10 min of the test phase for short-term memory the flies appear to show a stronger preference to the aversively conditioned sugar when compared the control gustatory stimulus. According to the theory during this phase short-term memory should be displayed the strongest and decline over time reaching a nearly complete attenuation after one hour. However, the displayed cumulative average preference indices let assume that for the aversive conditioning the memory recall takes place after about 30 minutes when middle-term memory starts to emerge. In this regard it is further worth noting that after an initial increase of the sugar preference to about 0.25 the preference index of the trained flies remain stable whereas the control flies only reach the same level of attraction to the sugar after 12-15 min and only start increasing their sugar preference after about 20 - 25 min. Compared to the dynamics of the cumulative average preference indices of the appetitive gustatory neuron activation and the artificial activation of dopamine neuron subsets the dynamics of the cumulative average preference after the aversive reinforcement through bitter gustatory neuron activation appears drastically different. This may further indicate competing pathways between sensing and conditioning the bitter taste stimulus as well as a delayed memory recall according to the metabolic need of the animal, as again the strongly delayed dynamics of the cumulative preference index indicates.

In the subsequent part Jelen and colleagues investigate the role of different subsets of dopaminergic neurons in the formation of aversive or appetitive gustatory short- and long-term memory. Similar to olfactory memory, gustatory memory relies on mainly two major sets of dopaminergic neurons that drive aversive and appetitive memory, namely the PPL1 and the PAM cluster that innervate different compartments of the mushroom body where they provide aversive or appetitive input to the conditioned stimuli encoded through the sparse activity of mushroom body Kenyon cells. As a consequence, in the following experiments Jelen and colleagues interfere with the hereinafter layer of memory and silence the mushroom body during conditioning while opto-genetically activating the PAM dopaminergic neurons, conceptionally recapitulating earlier studies that demonstrated the role of the mushroom body in gustatory memory earlier. Analogous to earlier findings for olfactory learning Jelen and colleagues use their intuitive setup to functionally subdivide the dopaminergic neurons in functional subunits with different roles in memory formation. These results strongly demonstrate how conserved the value-giving neuronal circuits are independent from their stimulus modality.

Consequently, extrapolating questions on olfactory memory formation on gustatory learning the authors use the STROBE essay to investigate how different nutrients may affect the formation of a long-term memory. In accordance with the findings on olfactory memory formation Jelen and colleagues find that long-term memory formation depends on readily accessible energy sources.

The study is interesting and rigorously conducted and reveals striking similarities between olfactory and gustatory memory formation. However, it appears that the authors have put large focus on the recapitulation of an already demonstrated mode of action of learning circuits using their new technique and many of the parallels between olfactory and gustatory memory formation appear pertinent as e.g., the need of readily accessible energy sources for long-term memory formation. The need for energy to form a long-term memory should not depend on the stimulus modality you learn but on the cellular mechanisms underlying learning itself. The innovative technique Jelen and colleagues present in their manuscript has such a strong potential that to me as a reader it appears a pity that the study did not exploit the possibilities of their technique to investigate virgin soil instead of walking on beaten tracks.

Author Response

Reviewer #1 (Public Review):

The finding that taste memory formation follows the same or highly similar logic and mechanisms as olfactory memory is very interesting. In particular, the new approach to use an operant learning assay developed by the authors to address this outstanding question in the field is very impressive. The shown data are of high quality and very convincing.

While the current version will be of clear interest to fly people dissecting memory formation, it might be less accessible outside this immediate field. Below I list my suggestions, questions and criticisms.

You have developed an operant assay and stress this in the introduction. This is important because it allows you to gain much better inside into how memory is formed and how it is recalled. Nevertheless, I was somewhat disappointed that you did not exploit that aspect more in your study. First, I suggest showing, at least for the initial figures, the traces (e.g. Fig 1D) not only for the test phase but also for the training phase. As you also mention in your discussion, the extent of memory formation will depend critically on the number of pairings during training. And perhaps not only on their number but also on their evolution/change over time. Second, you only show preference indices. I suggest showing the number of actual interactions with the food source in addition. In my opinion and experience, the preference index can be misleading or at least the interpretation might be questioned if the number of actual choices is very low or very high compared to controls or other groups. Third, regarding the same point, you show traces for test phases, but you do not comment or discuss why they might look the way they look. For instance, it appears that in some cases it takes a while to see an actual difference in the preference index while at other times it seems more instantaneously etc.

We have now added plots showing the preference indices over time during both training and testing for all the experiments in Figures 1 and 2. We also comment in the text on our view of their interpretation. Although we recognize that interesting features of the learning process could be revealed by examining the process over time, we also caution that earlier timepoints are inherently less robust because of smaller sample size to the measurements (flies tend to not take many sips of the food over the first several minutes). Thus, emergence of a preference after a period of time may not reflect an evolution of the preference as much as a firming up of the data as more sips are recorded. As a notable example, our data in Figure 1E,G show close to a zero preference for activation of sweet sensory neurons during the first 10 minutes of training, despite the innately appetitive nature of this manipulation. This is undoubtedly because it takes some time for flies to sample both choices and build up enough interactions to show a clear preference. This is not to say that the curves are never informative, however. For example, it is reassuring to see that activation of PAM neurons does not produce a positive preference at any time during training (Figure 2F).

We have also added the raw sip/interaction numbers for the experiments in Figure 1 in order to provide an example of how these data relate to the preference. Your concern about reliability differing depending on choice number is certainly warranted (as we also discuss above). However, the raw data does not suggest a major difference in the overall number of choices being made between groups.

Along the same lines, I am wondering why you do not observe extinction. Frequently if the CS is re-experienced without the US over several trials, you start to see memory fade. The preference traces as well as the actual interactions might help to explain this.

This is an interesting question, and one that we have certainly wondered about. Our assumption is that the number of exposures to the CS+ during testing is not sufficient to induce extinction. It would be interesting to run a longer testing period to see whether extinction occurs over a longer time course; however, we have not done so at this point.

You use salt as a negative US. I suggest showing at least one experiment with bitter taste (e.g. quinine) to show how general your finding is to negative conditioning. Your optogenetic data suggests it is.

We actually never use natural taste stimuli as the US; we only use salt as the CS+ in our appetitive learning experiments. We have revised the figures and figure legends extensively for clarity and one of the changes is to try to make it clearer what is the CS+ and CS- in each experiment.

You analyze the role of energy state in memory formation. This is very interesting. In light of the importance of feeding state, it would be very helpful to include starvation/metabolic state information not only in the methods but also in the results section (at least briefly).

We have now indicated in all the figure legends and in the text that flies were all food deprived for 24 hours prior to training.

Your data convincingly shows that taste memory is formed in the mushroom body. For instance, you show that inhibition of KCs prevents the change in preference. KC inhibition was done during the entire experiment (training and test). Thus, it's important to show how KC inhibition affects (or does not) training vs. test.

We appreciate the motivation for this suggestion and how extensively this issue has been explored in olfactory classical conditioning. We also agree that it would be interesting to perform this experiment. However, the practical logistics of doing this experiment were not possible with the constraints we were under. We unfortunately don’t currently have the means to operate the STROBE at a temperature high enough to effectively silence neurons using shibire(ts), and silencing with optogenetics is not possible with our current setup either. Thus, we will need to leave this issue unresolved for the time being.

Along the same lines, how do you envision this memory formation to happen at the circuit level? KCs and DANs are likely activated by CS and US. It would be important to at least include a paragraph in the discussion to clarify this.

The bulk of our characterization of this assay (including the demonstration that KCs are required) was done with 75 mM NaCl as the CS+ and optogenetic activation of PAM neurons as the US. Previous studies have shown activation of KCs by tastes (Kirkhart and Scott, 2015), so we believe that KCs are being activated by the CS+ and DANs are being activated by the US (in this case directly through optogenetics). Based on a great deal of beautiful work in olfactory classical conditioning, we believe it is likely that this co-incident activation leads to plasticity as KC-MBON synapses, thereby skewing the behaviour in favor of attraction. We have now tried to clarify this mechanism in the paper.

  1. Howard Hughes Medical Institute
  2. Wellcome Trust
  3. Max-Planck-Gesellschaft
  4. Knut and Alice Wallenberg Foundation