Pain is so close to pleasure: the same dopamine neurons can mediate approach and avoidance in Drosophila

  1. Daniel Döringer
  2. Christian Rohrsen
  3. Benjamin de Bivort
  4. Wolf Huetteroth
  5. Björn Brembs

  • Curated by eLife

    eLife logo

    eLife Assessment

    This is a valuable study that uses several operant paradigms to test Drosophila for the appetitive or aversive behavioral effects of self-activation of specific dopaminergic neuron (DAN) populations. The work provides solid data, although the range of paradigms and genetic driver lines will require more specific future evidence in case-by-case assays. The results suggest that a more nuanced view of the role of these neurons in ongoing behaviour and learning is necessary. The results will be of broad interest to the neuroscience community.

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Dopaminergic neurons (DANs) have long been implicated in reward, punishment, and value processing across animal phyla, yet whether individual DAN subpopulations convey generalized valence signals or serve more specialized functions remains contested. We addressed this question directly by screening about fifty Drosophila DAN subpopulations for their ability to drive consistent approach or avoidance across different operant behavioral paradigms in which flies controlled contingent optogenetic DAN activation through their own behavior. No single DAN subpopulation drove consistent preferences across all paradigms, and DAN activation produced substantially weaker valence effects than activation of sensory neurons used as positive controls. One subpopulation produced a striking time-dependent valence switch: initial aversion gave way to approach over approximately ten minutes of operant engagement. This switch required both dopamine and active behavioral contingency, demonstrating that agency itself is a necessary condition for such a valence switch. Together, our results suggest that individual DANs function as specialized, context-sensitive reporters for specific goals rather than general-purpose value signals, and that their apparent generality under some circumstances may reflect the consequences of non-contingent DAN stimulation rather than a true unified valence code.

Article activity feed

  1. eLife

    eLife Assessment

    This is a valuable study that uses several operant paradigms to test Drosophila for the appetitive or aversive behavioral effects of self-activation of specific dopaminergic neuron (DAN) populations. The work provides solid data, although the range of paradigms and genetic driver lines will require more specific future evidence in case-by-case assays. The results suggest that a more nuanced view of the role of these neurons in ongoing behaviour and learning is necessary. The results will be of broad interest to the neuroscience community.

  2. eLife

    Reviewer #1 (Public review):

    This study addresses an important question: whether dopaminergic activity in Drosophila functions as a general value signal across behavioral contexts. Its main strength is the breadth of the experimental program, combining multiple DAN driver lines, several operant paradigms, rescreening, and more targeted follow-up experiments. The cross-paradigm design is well suited to test whether self-stimulation effects generalize.

    The main weakness is that several conclusions extend beyond what the experiments directly establish. The genetic tools define populations rather than the same identified neurons at single-cell resolution. In the key time-course experiments, the data more clearly support a reduction of initial avoidance toward neutrality than a consistent aversive-to-appetitive sign reversal. Similarly, the non-contingent comparison uses continuous pre-exposure rather than a stimulation-history-matched open-loop condition; it therefore shows that simple non-contingent exposure is insufficient, but does not isolate agency itself as the causal variable.

    The general claims are also constrained by incomplete overlap among the initial screens and by design changes between the original and follow-up joystick experiments. Overall, the study provides useful evidence that the behavioral effects of DAN activation are weak, context-dependent, and time-dependent, while the claims about neuronal identity, positive valence, agency, and a general dopamine "common currency" require more cautious framing.

  3. eLife

    Reviewer #2 (Public review):

    Summary:

    Broadly, the authors use a wide range of lines targeting dopaminergic neurons in different behavioral paradigms to determine their role in behavior.

    Strengths:

    There is a comprehensive use of lines and assays by a single group, such that internally consistent comparisons can be made.

    Weaknesses:

    It is not clear to what extent these broad assays will provide concrete answers that enable a "unified hypothesis" of how dopamine functions in behavior.

    The manuscript has a feel of being a mix of an original paper, commentary, hypothesis-generating outlook, a focused summary, and a broad review of the field. As someone outside this field, I find that it has its advantages and disadvantages. For example, the final sentence of the Abstract, despite multiple readings, is not obvious to someone outside the field. Plus, due to the large number of references for any one view, it is not clear which work is saying exactly what. For example, which of the fly papers on dopamine clearly say they favor the "common currency" model? Do these references also use this term? Even for a reference that occurs multiple times (Jovanoski and Waddell, 2025), it's not clear what this paper actually proposes.

    Also, what does "a reporter for a specific goal" mean (Line 16 of Abstract, and lines 1986-1987 of Discussion)? They provide a nice example of a tethered fly and what happens when it is suddenly released, vs a free-flying fly that has entered a maze--what is the difference here in terms of dopamine function and the fly's "goal" and how does the fly know the difference? When a fly is free flying in a maze, how does it know that it has just been released from a tether, or whether it has entered a maze as it was freely flying outside it?

    And for their "unified hypothesis", how does a fly know that it is in "controls its environment" versus "lack of control"?

    In sum, I found this paper to be valuable in bringing up the different views of dopamine signaling. However, it was a bit unclear at the end as to what dopamine is actually doing in the fly.

  4. eLife

    Reviewer #3 (Public review):

    Summary:

    Dopaminergic neurons (DANs) are known to play a role in conveying rewarding and punishing signals that influence future behaviour. This study uses optogenetic manipulation of a range of subpopulations of DANs to assess their influence in several operant learning paradigms, such that the animal's own behaviour determines the DAN activation. The results show surprisingly inconsistent effects and do not support the widespread assumption that specific DANs consistently convey either positive or negative valence. In one case, the apparent valence of the DAN activation switches over a time frame of minutes from aversive to attractive, and the relevant circuits are further dissected to show this effect is dopamine-dependent, for a specific DAN subset, and contingent on operant control of behaviour.

    Strengths:

    This study has two major strengths. The first is to address the relatively neglected field of operant learning in Drosophila. The detailed unpicking of learning circuits in the Drosophila brain that has been made possible by neurogenetics and other advances has almost exclusively focussed on classical conditioning paradigms, but these are often not representative of the real behavioural situations that animals face. Cross-comparison of the behavioural results of identified neuron activation in a wider range of paradigms is very welcome. The second is the use of multiple paradigms and a wide-ranging screen to survey the effects across many lines (targeting different DAN populations), including many that have been previously tested in classical conditioning. A further strength is the useful review (in the introduction and discussion) of existing research that already raises problems with the view that a single dimensional positive and negative "value" (as conceived of in standard reinforcement learning theory) can be directly mapped to the activity of two broad clusters of DANs in the brain.

    Weaknesses:

    The main weakness is that it is difficult from the manuscript to understand what the authors believe is the underlying mechanism by which DAN activation should affect behaviour in operant paradigms, and how this should relate to the better-known role of DAN activity in classical conditioning. For example, they refer to Heisenberg and Wolf's 1991 distinction between operant activity and operant learning. That paper proposed a mechanism involving a goal, some degree of random variation in action, comparison of sensory input to a goal, correlation of the efference copy of the action to the output of this comparison, with the strength of this correlation subsequently shaping the action probabilities. It would help to be explicit about what DAN activity is assumed to represent in this scheme - the outcome of comparison to a goal? The actual assumptions behind the work seem more simplistic: if different groups of DANs 'signal' reward or punishment, animals should persist in actions that activate the reward DANs and desist in actions that activate the punishment DANs. This might be justified by the standard way that innately rewarding/punishing stimuli are operationally defined in learning theory. But what brain circuitry is assumed to underlie the variation and choice of actions, and how is this influenced by DANs?

    Perhaps what is most clearly shown is that avoidance behaviour to activation of innately punishing sensory neurons is not directly a consequence of downstream activation of specific DANs, such that DAN activation can be substituted to produce the same behavioural phenomena (at least for the DANs tested here). Indeed, this is how the core results are summarised at the start of the discussion. But should this have been expected? There are a lot of assumptions needed to connect the expected effect of a reinforcement signal in classical conditioning to its effect on operant behaviour that are not made sufficiently explicit.

    Appraisal:

    The authors met their aim to carry out a (fairly) comprehensive test of the effects of optogenetic activation of a variety of DAN subpopulations under several operant paradigms that differ in relevant respects (group vs. individual testing, freely moving vs. tethered). The lack of consistent results across these paradigms, and generally weak effects (with limited statistical analysis) make it difficult to interpret whether the outcome here is mostly 'noise', but perhaps this is an example where the failure to find a clear effect (as predicted) is itself informative. The paper appropriately describes the limitations (such as the mixed range and specificity of DAN subsets manipulated across the paradigms) which were probably unavoidable given the tools available.

    Likely impact on the field:

    This work should have an impact in countering assertions that dopamine neurons in the fly brain can be understood as (collectively) conveying a single dimension of valence, from positive to negative, across all behavioural and learning situations. It also provides strong motivation to more clearly understand the nature of operant learning and how/where it might be implemented in the insect brain.

  5. eLife

    Author response:

    We thank the reviewers and the Reviewing Editor for their careful and constructive assessments. We find the reviews fair and helpful, and we plan to revise the manuscript along the lines they suggest. In brief:

    Scope of the conclusions (Reviewer #1):

    We can see how a reader may get the impression that some of our conclusions go beyond what the experiments directly establish. In the revision we will:

    - Describe the genetic tools as targeting neuronal populations (DAN clusters) rather than identified single neurons, and adjust statements about neuronal identity accordingly.

    - Rephrase the description of the time-course results, to emphasize that the shift from aversive to appetitive only appears across experiments and that some single experiments may only show a shift to neutrality. This should help moderate claims about positive valence.

    - State explicitly that our non-contingent comparison used continuous pre-exposure rather than a yoked control. It therefore shows that simple non-contingent exposure is insufficient but does not isolate agency as the causal variable.

    - Better motivate the design changes between the original screen and follow-up joystick experiments.

    - Frame the “common currency” hypothesis more cautiously throughout.

    Clarity and accessibility (Reviewer #2):

    We appreciate that the manuscript is hard to follow for readers outside the field. We will revise the final sentence of the Abstract and clarify what we mean by a dopaminergic “reporter for a specific goal”. We will also spell out how the contrast between a tethered fly and a free-flying fly in a maze bears on this, and what we mean by “control” versus “lack of control” of the environment. In the Introduction and Discussion we will make clearer which cited studies propose or favor the “common currency” view, including Jovanoski and Waddell (2025), and which ones merely bear on it.

    Conceptual framework (Reviewer #3):

    We will make our assumptions linking the role of DANs in classical conditioning to their expected effects in operant paradigms more explicit. The reference to the Heisenberg and Wolf (1991) scheme as a suitable level of explanation was particularly helpful and we will thrive to develop our hypotheses at the same level. At least for the designers of the original screen (BB and CR), the lack of consistent effects was clearly unexpected, or we would have designed the experiments differently. We will justify these expectations and from this why we consider this negative result to be so informative.

    Code availability (Reviewing Editor): We will deposit the code in a repository with a permanent DOI, such as Zenodo, in addition to GitHub.

  6. Version published to 10.1101/2021.10.04.463010 on bioRxiv