High-throughput automated methods for classical and operant conditioning of Drosophila larvae
Curation statements for this article:-
Curated by eLife
Evaluation Summary:
A new and interesting operant conditioning paradigm is established for the Drosophila larva. A novel role for serotonergic pathways in the VNC in operant learning points to new circuits and mechanisms for learning and memory. Impressive technology opens doors for new and exciting studies on learned behavior in the small and tractable circuits of the larva.
(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.)
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Neuroscience (eLife)
Abstract
Learning which stimuli (classical conditioning) or which actions (operant conditioning) predict rewards or punishments can improve chances of survival. However, the circuit mechanisms that underlie distinct types of associative learning are still not fully understood. Automated, high-throughput paradigms for studying different types of associative learning, combined with manipulation of specific neurons in freely behaving animals, can help advance this field. The Drosophila melanogaster larva is a tractable model system for studying the circuit basis of behaviour, but many forms of associative learning have not yet been demonstrated in this animal. Here, we developed a high-throughput (i.e. multi-larva) training system that combines real-time behaviour detection of freely moving larvae with targeted opto- and thermogenetic stimulation of tracked animals. Both stimuli are controlled in either open- or closed-loop, and delivered with high temporal and spatial precision. Using this tracker, we show for the first time that Drosophila larvae can perform classical conditioning with no overlap between sensory stimuli (i.e. trace conditioning). We also demonstrate that larvae are capable of operant conditioning by inducing a bend direction preference through optogenetic activation of reward-encoding serotonergic neurons. Our results extend the known associative learning capacities of Drosophila larvae. Our automated training rig will facilitate the study of many different forms of associative learning and the identification of the neural circuits that underpin them.
Article activity feed
-
-
Author Response
Reviewer #2 (Public Review):
Klein et al. have developed a high-throughput tracker to evaluate operant conditioning in Drosophila larvae. Employing this device, they train larvae to prefer bending towards one specific side (left or right), by using as unconditioned stimulus (US) the optogenetic activation of dopaminergic and serotoninergic neurons, demonstrating that larvae are able to perform this behaviour. Furthermore, they show that serotoninergic neurons alone are sufficient to mediate the reward signal, and that specifically serotoninergic neurons in the VNC are required for this behaviour. However, they do not show whether serotoninergic VNC neurons are sufficient. The results are interesting and novel. Operant conditioning had been shown for Drosophila adult. Furthermore, the existence of VNC circuits …
Author Response
Reviewer #2 (Public Review):
Klein et al. have developed a high-throughput tracker to evaluate operant conditioning in Drosophila larvae. Employing this device, they train larvae to prefer bending towards one specific side (left or right), by using as unconditioned stimulus (US) the optogenetic activation of dopaminergic and serotoninergic neurons, demonstrating that larvae are able to perform this behaviour. Furthermore, they show that serotoninergic neurons alone are sufficient to mediate the reward signal, and that specifically serotoninergic neurons in the VNC are required for this behaviour. However, they do not show whether serotoninergic VNC neurons are sufficient. The results are interesting and novel. Operant conditioning had been shown for Drosophila adult. Furthermore, the existence of VNC circuits sufficient for operant conditioning had been shown for other species, as the authors point out in the discussion. Nonetheless, the genetic dissection to identify serotonine expressing neurons as mediators of operant conditioning in the Drosophila larva, and the identification of VNC serotonine cells as necessary are new. Furthermore, given the experimental advantages of the Drosophila larva, including genetic accessibility and a full connectome, the findings open the door to future research into the circuit mechanisms of operant conditioning. I have some comments that I think would be important to address.
The high-throughput tracker is impressive. However, there is no sufficient documentation to ensure that an expert would be able to easily reproduce it. All of the hardware assembly files, the list of materials, as well as the electronic circuit maps and all of the required software needs to be appropriately documented and uploaded onto a public repository. This is a basic requirement when publishing new hardware/software, particularly in an open journal such as eLife.
We have now included all the documentation and CAD files for the high-throughput tracker. The software is publicly available in the following Github repository (https://github.com/ZlaticLab/multi-larva-tracker-scripts-public). The CAD files are available in the Supplementary materials of the paper.
- The differences observed in the results of operant conditioning are very subtle (see for example figure 3c), which means that it is extremely important that statistic analyses are correctly made. The sample number (n) for these experiments is really high (n>100) and for what I understood is not equivalent to the number of animals, because the same animal can generate n >1, eg. n = 2 or n =3 if it collides one or two times, as each time it collides a new identity is given to the larvae. This means that the datapoints collected are not independent, and I think in that case a Wilcoxon rank-sum test is not the appropriate test to take. I recommend the authors and eLife editors to consult with an expert in this type of statistics. Alternatively, the authors could, for each experiment, take into account only the data from larvae that did not collide, and for those that collide only take into account the data before the collision. This can be calculated easily as they just need to exclude from their analysis in each experiment all of the larval IDs where the ID is larger than the initial number of larvae identified by the software.
We apologise if we did not clarify sufficiently that we only took into account (for each time bin) larvae that did not collide. Within the Materials and methods, we describe how objects retained for analysis had to satisfy several criteria. The first criterion is that the object needed to be detected in every frame of the given 60 s bin. In this way, the object identity is stable throughout the bin - a reflection that the object did not collide with another object. In other words, within a single time bin, the same animal only contributes once. Text has been added to the Materials and methods to clarify that this first criterion is selecting for larvae that did not collide.
The reviewer mentions that Wilcoxon rank-sum test is not the appropriate nonparametric test for dependent samples. We agree. In accordance with this, the test used for within-bin comparisons was Wilcoxon signed-rank, which is also nonparametric but is for dependent samples. We believe, then, that there is no need to reconsider the statistical tests used.
-The finding that serotoninergic neurons in the VNC, which with the line they used amount to only 2 neurons per VNC hemisegment, are required for operant conditioning is very interesting. It would be great if they could also test whether they are sufficient. It seems that they would just need to make two split Gal4 lines one for tsh and one for tph, so the experiment does not seem too difficult and would significantly add to their findings.
Generating new intersections is beyond the scope of this already large study which has been significantly impacted by the pandemic. We have therefore added the following sections below explaining that we have identified candidate serotonergic neurons that are required for operant learning and that identifying specific single neuron types that may be sufficient would be an exciting avenue for future follow-up work.
In the Results section entitled, “Serotonergic VNC neurons may play role in operant conditioning of bend direction” we have added:
“The Tph-Gal4 expression pattern contains two neurons per VNC hemisegment (with the exception of a single neuron in each A8 abdominal hemisegment, Huser2012). Future experiments exclusively targeting a single serotonergic neuron per VNC hemisegment could be valuable in determining whether they are sufficient for operant learning.”
In the Discussion section entitled: “Automated operant conditioning of Drosophila larvae”
“Furthermore, developing sparser lines that target single serotonergic and dopaminergic neuron types will enable the identification of the smallest subsets of neurons that are sufficient for providing the operant learning signal. Behavioural experiments with these genetic lines may have the added benefit of mitigating conflicting or non-specific reinforcement signalling.”
-
Evaluation Summary:
A new and interesting operant conditioning paradigm is established for the Drosophila larva. A novel role for serotonergic pathways in the VNC in operant learning points to new circuits and mechanisms for learning and memory. Impressive technology opens doors for new and exciting studies on learned behavior in the small and tractable circuits of the larva.
(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.)
-
Reviewer #1 (Public Review):
Klein and colleagues have developed a new setup to artificially activate genetically targeted neurons in temporal precise correlation with specific behaviors in larva of Drosophila melanogaster. The work explores how the activation of specific sets of reward and punishment coding neurons during the execution of side-specific bending alters the occurrence of this behavior. Indeed, activating serotonergic neurons during specific bending in a training phase, biases bending direction in the test. Since altering behavior as a consequence of its rewarding or punishing outcome is considered operant learning the authors conclude that the targeted neurons mediate operant conditioning. Below I will point out the strength and my criticisms concerning the presented work.
The newly developed closed-loop set-up is …
Reviewer #1 (Public Review):
Klein and colleagues have developed a new setup to artificially activate genetically targeted neurons in temporal precise correlation with specific behaviors in larva of Drosophila melanogaster. The work explores how the activation of specific sets of reward and punishment coding neurons during the execution of side-specific bending alters the occurrence of this behavior. Indeed, activating serotonergic neurons during specific bending in a training phase, biases bending direction in the test. Since altering behavior as a consequence of its rewarding or punishing outcome is considered operant learning the authors conclude that the targeted neurons mediate operant conditioning. Below I will point out the strength and my criticisms concerning the presented work.
The newly developed closed-loop set-up is impressive and will pave the way for many exciting studies on learned behavior and beyond. To validate the set up the authors induce rolling behavior by thermo- or optogenetically activating two sets of previously described neurons in individual larva. Both approaches show convincing induction of the behavior per se. However, it is worth pointing out that there seems to be an interaction of the different tools used (thermo and opto-genetic) and the targeted neurons: the authors observe different dynamics of the behavior across the three stimulation cycles depending on stimulation method and labeled neurons. These findings make it difficult to understand why the authors choose only the optogenetic activation to investigate operant conditioning. The strength of the setup is that individual animals can be targeted. Though the presented data show that behavior can be reliably induced in stimulated animals, it lacks the information about the behavior of non-targeted larva during the stimulation. Thus, it would strengthen the work if the authors could show the behavior of the non-targeted larva during the time when targeted larva receive light or heat.
The authors use their setup to investigate operant conditioning. In operant conditioning an animal learns to associate its action with the consequences. Their new setup allows the authors to artificially induce consequences, the activation of reward or punishment coding neurons, upon side-specific bending behavior. The experiments show that side specific bending in the test is slightly biased towards the side previously paired with the neuronal stimulation. Interestingly, the data suggest that this effect requires the activity of serotonergic neurons outside of the brain (in the VNC) and that it is not mediated by dopamine signaling in the brain. Though, the effects seem to be reproducible with Ddc- and the Tph-GAL4 the reported differences are small, and the origin of the relative difference between left and right bending in the paired group is not entirely clear. Thus, it will be important to strengthen the work by additional experiments and extend the analysis of the presented data. Given the novelty of the method and the differences between the tools in the proof of principle experiments the authors should repeat the key experiments (Figure 3 b and e) with the thermogenetic stimulation. Further it would strengthen the investigation on operant conditioning if the authors would explore the temporal relationship between the CS and US, especially since the effect might be a reduction of the unreinforced behavior (see below). Concerning the analysis, the authors should consider that given the small effect they observe, they want to be sure that it originates from training. Though they show the pretraining results for one of the experiments (Figure 3b, the trained group), the pretraining bending is very relevant for each of the operant learning experiments. In fact, training induced effects should not only be measured by looking at the left vs right bending in the final test but as a change between pre versus post or between a trained and a mock control group. This is done for one group (Ddc-GAL4) in Figure 3b but will be mandatory for all operant learning experiments. It would improve the accessibility of the learning induced change of behavior if the authors could show the pre vs post training results for each run (10-12 larva in a plate). Further, they should plot the numbers of reinforced behaviors in each of the training protocols and relate it to the test performance. The presented data clearly suggests a decrease of the unstimulated bending rather than a change in the reinforced behavior. Though the authors mention it, they do not explain or discuss it. It will be very important for the logic of the manuscript that the authors explain this phenomenon and how it relates to operant conditioning.
Lastly, though the manuscript discusses most of the data carefully, in my view the authors miss an important issue: it remains to be shown if fly larva are capable of operant learning using external reward or punishment. The presented evidence is based on artificial activation of neurons, which arguably is a hint but not a prove that operant conditioning is withing the repertoire of a fly larva, an issue the authors should mention and discuss.
-
Reviewer #2 (Public Review):
Klein et al. have developed a high-throughput tracker to evaluate operant conditioning in Drosophila larvae. Employing this device, they train larvae to prefer bending towards one specific side (left or right), by using as unconditioned stimulus (US) the optogenetic activation of dopaminergic and serotoninergic neurons, demonstrating that larvae are able to perform this behaviour. Furthermore, they show that serotoninergic neurons alone are sufficient to mediate the reward signal, and that specifically serotoninergic neurons in the VNC are required for this behaviour. However, they do not show whether serotoninergic VNC neurons are sufficient. The results are interesting and novel. Operant conditioning had been shown for Drosophila adult. Furthermore, the existence of VNC circuits sufficient for operant …
Reviewer #2 (Public Review):
Klein et al. have developed a high-throughput tracker to evaluate operant conditioning in Drosophila larvae. Employing this device, they train larvae to prefer bending towards one specific side (left or right), by using as unconditioned stimulus (US) the optogenetic activation of dopaminergic and serotoninergic neurons, demonstrating that larvae are able to perform this behaviour. Furthermore, they show that serotoninergic neurons alone are sufficient to mediate the reward signal, and that specifically serotoninergic neurons in the VNC are required for this behaviour. However, they do not show whether serotoninergic VNC neurons are sufficient. The results are interesting and novel. Operant conditioning had been shown for Drosophila adult. Furthermore, the existence of VNC circuits sufficient for operant conditioning had been shown for other species, as the authors point out in the discussion. Nonetheless, the genetic dissection to identify serotonine expressing neurons as mediators of operant conditioning in the Drosophila larva, and the identification of VNC serotonine cells as necessary are new. Furthermore, given the experimental advantages of the Drosophila larva, including genetic accessibility and a full connectome, the findings open the door to future research into the circuit mechanisms of operant conditioning. I have some comments that I think would be important to address.
- The high-throughput tracker is impressive. However, there is no sufficient documentation to ensure that an expert would be able to easily reproduce it. All of the hardware assembly files, the list of materials, as well as the electronic circuit maps and all of the required software needs to be appropriately documented and uploaded onto a public repository. This is a basic requirement when publishing new hardware/software.
- The differences observed in the results of operant conditioning are very subtle (see for example figure 3c), which means that it is extremely important that statistic analyses are correctly made. The sample number (n) for these experiments is really high (n>100) and for what I understood is not equivalent to the number of animals, because the same animal can generate n >1, eg. n = 2 or n =3 if it collides one or two times, as each time it collides a new identity is given to the larvae. This means that the datapoints collected are not independent, and I think in that case a Wilcoxon rank-sum test is not the appropriate test to take. I recommend the authors and editors to consult with an expert in this type of statistics. Alternatively, the authors could, for each experiment, take into account only the data from larvae that did not collide, and for those that collide only take into account the data before the collision. This can be calculated easily as they just need to exclude from their analysis in each experiment all of the larval IDs where the ID is larger than the initial number of larvae identified by the software.
- The findings that serotoninergic neurons in the VNC, which with the line they used amount to only 2 neurons per VNC hemisegment, are required for operant conditioning is very interesting. It would be great if they could also test whether they are sufficient. It seems that they would just need to make two split Gal4 lines one for tsh and one for tph, so the experiment does not seem too difficult and would significantly add to their findings.
-
Reviewer #3 (Public Review):
The manuscript provides evidence that larvae are capable of operant, as opposed to classical, conditioning: optogenetic activation of serotonergic neurons after a larva bends in one direction increases the likelihood that it will bend in that direction again.
Furthermore, the manuscript shows that serotonergic neurons located in the ventral nerve cord are capable of inducing this associative conditioning. While dopaminergic and serotonergic neurons, notably the dopaminergic PAM cluster in the Mushroom Bodies, have previously been implicated in classical conditioning, where different subsets apply positive or negative valence, these data suggest that specific serotonergic neurons may also contribute to learned behaviors. Although the cellular and circuit mechanisms remain unclear, and the consequences of …
Reviewer #3 (Public Review):
The manuscript provides evidence that larvae are capable of operant, as opposed to classical, conditioning: optogenetic activation of serotonergic neurons after a larva bends in one direction increases the likelihood that it will bend in that direction again.
Furthermore, the manuscript shows that serotonergic neurons located in the ventral nerve cord are capable of inducing this associative conditioning. While dopaminergic and serotonergic neurons, notably the dopaminergic PAM cluster in the Mushroom Bodies, have previously been implicated in classical conditioning, where different subsets apply positive or negative valence, these data suggest that specific serotonergic neurons may also contribute to learned behaviors. Although the cellular and circuit mechanisms remain unclear, and the consequences of silencing these neurons in contexts where the larva might more naturally employ operant learning are not tested, this research suggests new areas for exploring the adaptive capacity of a powerful model organism.
-
