Article activity feed

  1. Evaluation Summary:

    The authors perform a tour-de-force study of classical conditioning in fly larvae. Experiments are original, findings are exciting, and we expect this paper to have a substantial impact. There is potentially an issue in the assay of larvae preference taking an hour of unrewarded presentation of CO2, while the training and extinction happen on much shorter scales, muddying the ability to interpret the results. A mathematical model of the conditioning process is also missing, which also makes interpretation harder.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)

    Was this evaluation helpful?
  2. Reviewer #1 (Public Review):

    The manuscript attempts to analyze the process of Pavlovian conditioning in fly larvae, where the conditioned stimulus (CS) is the presence of CO2 (usually an repulsive signal), and the unconditioned stimulus (US) is an ontogenetic manipulation of reward neurons. In the course of the manuscript, the authors try an astonishing variety of different conditioning protocols, changing the order and the duration of CS and US presentation and the strength of the CS, introducing extinction phases, testing the duration of persistence of the association, and so on.

    Major findings of the manuscript include that:

    1. This is, indeed, a classical Pavlovian system, where the order of the CS/US presentation matters (and not just their co-occurrence).

    2. It is impossible for larvae to be trained to like CO2, and the strongest learning achieved is to become indifferent to it.

    3. The learning and the extinction in these animals is supposedly all-or-nothing - every presentation of CS/US pairing makes a fixed fraction of the animals fully trained, and similarly extinction only changes the fraction of fully trained animals.

    4. Memories persist overnight.

    I find the manuscript illuminating and thought-provoking. I did not expect (2) above, for example. The studies are quantitative, done with high statistical power. and focus on individual animals, rather than ensemble-averaged. Thus I believe the manuscript will be a gold standard for associative learning work in small animals.

    Nonetheless, the manuscript left me wondering about a few serious things.

    1. The choice of CO2 as a CS is both a curse and a blessing. The experimentalists must overcome innate avoidance of the signal, instead of the value of the signal being neutral to a naive animal. The authors speculate that the conditioning here is through inhibition of avoidance, and the picture they try to build (and it would be useful to have this as a simple mathematical model rather than just a picture) is that an unconditioned optogenetic stimulus decreases avoidance of the conditioned stimulus. This is not the standard Pavlovian scheme, where, traditionally, positive reinforcement increases preferences (+ / ++) and negative reinforcement increases avoidance (- /+-) or decreases preference (- /-+). Instead it's an unusual structure where positive reinforcement decreases avoidance (+ / --). This is uncommon -- and results in precisely the same behavior limitations that the authors noted: the most one can do is to decrease avoidance to zero, and then subsequent presentation of CS/US pairs does not lead to emergence of the preference. I think the manuscript would become stronger if the authors tried to speculate what aspects of the animal's ecology would make this uncommon functional organization favored.

    2. Potentially a bigger issue is that the training in these experiments last for a very short time (from 30 s to 15 min or so), while the readout of the behavioral preference takes an hour, during which many unrewarded presentations of CS happen. In the paper, the authors themselves show that unrewarded CS presentations lead to reduction in the behavioral response (fig 3), to the point that overnight memory consolidation is not observed (fig 4). Thus this long scale of the assay compared to the time scale of dynamics of the learning and extinction themselves makes interpretation of the findings very hard, at least for me. For example, is the 50% maximum choice of CO2 due to the animal not being able to establish the preference to it (and only being able to suppress the avoidance), or is it because the animal establishes a strong preference, which then gets partially washed away during the one hour of testing? There are a few ways that this and similar concerns can be addressed. First, a different assay can be established, where the preference is measured as quickly as it gets established and extinguished. Second, one can explore if the preference of animals does not change during the course of the testing phase. This could be done by analyzing the preference over fifteen minute segments, and checking for a drift (one could even combine animals to do so). Third, one can try to establish a mathematical model of conditioning and extinction, which would account for unrewarded CS presentations, and then see whether all of the data can be explained within this model. Or maybe one can do something totally different -- but I believe that some analysis of the effects of the assay on the conditioning state must be performed.

    3. The authors talk about quantized response as compared to gradual learning. This makes it seem that there are only two states that the animals can be in. But this is, in fact, unclear from the data. It's clear that there are two modes: indifferent to CO2 and avoiding it, but the modes are wide. Is there an additional signal there? Where is the width of the modes coming from? Is it simply the counting statistics of making, on average, pN out of N choices? Or are the data hiding something more interesting? This could be addressed by being a bit more careful with statistical analysis, and not treating the data as being fit by two Gaussians with arbitrary widths, but as a mixture of two Bernoulli distributions -- would such model work? If not, then why?

    Was this evaluation helpful?
  3. Reviewer #2 (Public Review):

    The authors have developed a novel apparatus that promises to accelerate mechanistic studies of how individual Drosophila larvae learn odor preferences. Previous studies of olfactory learning in Drosophila larvae have often used mass assays, testing groups of freely moving larvae trained under different conditions (but see, for example, Gerber and Stocker, 2007). Here, the authors trained individual animals via Pavlovian conditioning in a Y-maze, using an innately aversive olfactory cue, carbon dioxide (CO2), as a conditioned stimulus, and optogenetic stimulation of a pair of reward neurons as the unconditioned stimulus. They then assessed changes in the animal's preference for carbon dioxide versus air in the Y-maze after systematically varying the temporal relationship between carbon dioxide presentation and reward stimulation during training.
    The results show that, consistent with the associative nature of the learning, the aversion of larvae to CO2 decreases when CO2 presentation is paired with reward stimulation (with the necessary and sufficient condition being only that CO2 predicts reward stimulation), and that the learned preference increases as a function of the number of training cycles, extinguishes when reward stimulation is removed after paired training, and under the right conditions, can be shown to be protein-synthesis dependent (i.e., a long-term memory).
    Notably, by using their apparatus to repeatedly assess the choices made by individual animals given different amounts of training, they were also able to demonstrate that the learning underlying the change in preference occurs in an all-or-none fashion for each animal, rather than in a graded manner in which the preference gradually changes across training cycles for each animal. This is an observation that could not have been made using the conventional mass assays in which groups of animals trained together.

    Strengths: The authors clearly demonstrate the power of their new apparatus for testing learned odor preferences in Drosophila larvae, in providing researchers with improved experimental control over the presentation of olfactory cues and reward stimulation using optogenetic activation. Using their device, they show systematically the training conditions under which Drosophila larvae show changes in preferences to an innately aversive odor, and that this learning undergoes extincion, exhibits protein-synthesis dependent long-term memory, and occurs in an all-or-none mode. As the authors present their remarkably clean findings with equally clear rationales, their conclusions are well supported by data. As the authors claim, this apparatus promises to shed greater light on the neural circuit mechanisms underlying olfactory learning.

    Weaknesses: While the results shown in the paper are thorough and demonstrate that many of the major findings shown in adults with respect to the neural basis of behavior can be replicated in larvae, the authors appear uncertain about what the main focus of their article should be. Whereas the title suggests the authors' desire to highlight the evidence on the all-or-none nature of learning, other sections, including the introduction and conclusions, seem to suggest their desire to emphasize the utility of the new technique. If the former is the case, readers might be interested in knowing what neural mechanisms underlie the behavioral evidence showing that learning occurs in a switch-like fashion. For example, is a switch-like change also induced at the level of neurons?

    The authors also clearly showed that the "Forward Paired" condition is necessary and sufficient to induce the change in preference to CO2. Why did they not use this condition to carry out all of their experiments, if this shows the crucial learning of interest (not contaminated coincident stimulation)? Have the authors used the "forward paired" condition to examine extinction and overnight memory retention? If so, did they differ from the findings using the "coincidence" condition, or were they the same?

    Was this evaluation helpful?
  4. Reviewer #3 (Public Review):

    Over the past two decades, the Drosophila larva has proven to be an advantageous system to study the neural basis of memory and its effects on orientation behavior. While larvae clearly learn, this behavior has been mostly characterized through en masse assays. To this date, it has been extremely difficult - if not impossible - to characterize learning at the level of single larvae. Gershow and colleagues present a truly ingenious assay to control the frequency and the exact timing of the presentation of the conditioned and unconditioned signals. With their new assay, they demonstrate the switch-like nature of learning in individual larvae - a really exciting finding, which alone justifies the assay development. But, the authors did not stop there. Their work revisits multiple aspects of the theory of associative learning in the Drosophila larva, including the role of repeated training, the emergence of memory extinction and overnight consolidation of memory.

    This manuscript will have a major impact on the field of memory and learning in Drosophila. It provides a groundbreaking tool that will enable a wealth of new experimental work that would have been impossible until now. The study of various parameters influencing memory formation is controlled by the right set of conditions (e.g., the use UAS-Chrimson alone with ATR to preclude leaking expression of the effector; the potential effect of sensory habituation, etc.). The data analysis and statistical models are simple but powerful. I have no major criticism about the methodology and key conclusions of the study.

    Was this evaluation helpful?