Dopaminergic systems create reward seeking despite adverse consequences

Unconstrained reward seeking in humans is commonly associated with substance use disorders. Rodent studies show that electrical or optogenetic self-stimulation of dopaminergic neurons can induce persistent self-administration, even under punishment, paralleling cocaine-infused behaviour. Such approaches model acquisition of unconstrained reward seeking while avoiding confounds of pharmacology. However, mammalian DAN populations are heterogeneous and hard to target precisely. Drosophila, with a reduced dopaminergic system, enables cellular-resolution analyses of reward memory and seeking. In flies and mammals, specific dopaminergic neurons (DANs) convey reward or aversive teaching signals to assign valence to stimuli. In adult flies, reward-encoding DANs are much more numerous than aversive DANs. Functional analyses and connectivity show extensive heterogeneity within reward DANs, supporting parallel coding of diverse rewards (e.g., sweet taste, nutrient value, water, courtship, safety, absence of punishment, relative aversive value). Combinations of aversive and rewarding DANs regulate need-specific expression of reward-seeking memories and innate food-seeking. The authors hypothesized that simultaneous engagement of multiple reward-specific signals could generate a compound reward memory that drives reward seeking despite adverse consequences.

• Prior work in rodents: optogenetic/electrical self-stimulation of ventral tegmental area DANs produces persistent self-administration under punishment, similar to cocaine; inhibition of these DANs can generate aversion.
• Heterogeneity of mammalian midbrain DANs complicates isolating mechanisms of unconstrained seeking.
• In Drosophila, defined DAN subsets in PAM (reward) and PPL1 (aversion) clusters convey positive and negative valence for associative learning; reward DAN population is larger than aversive DANs.
• Specific PAM DAN subtypes encode distinct rewards: sweet taste vs nutrient value of sugar, water, male courtship reward, safety/absence of expected punishment, and relative aversive value.
• Combinatorial activity of aversive and rewarding DANs controls need-dependent expression of learned and innate food-seeking behaviours.
• Ethanol reward in flies can produce shock-resistant reward seeking; ethanol also broadly activates PAM DANs.
• Input connectivity studies reveal additional heterogeneity and parallel input structures to PAM subtypes, suggesting multiplexed reward representations.

• Model organism: Drosophila melanogaster adults.
• Behavioural conditioning and testing:
  • Olfactory appetitive conditioning with either natural sucrose reward or optogenetic DAN activation (red light, CsChrimson/CsChr), versus mock training.
  • T-maze tests of odour preference under varying conditions and durations (15, 30, 60, 120 s).
  • Shock-punishment tests: presenting the CS+ odour concurrently with 90 V electric shocks (1.5 s every 5 s; standard for 60 s aversive training) to assess reward seeking despite punishment.
  • Food competition tests: choice between CS+ odour (paper-lined) and CS− odour with dried sucrose to assess neglect of physiological need (feeding) in starved flies.
  • Consecutive training paradigms: reward implantation via DAN activation followed by aversive conditioning to assess interference with subsequent aversive learning.
  • Naive shock avoidance assays after prior transient optogenetic activation of reward DANs to test effects on aversion processing and time course (immediate, 10 min, 1 h).
• Genetics and circuit targeting:
  • Optogenetic activation: UAS-CsChrimson with GAL4 drivers targeting reward-encoding DANs and other neurons: 0273-GAL4 (Fer2-driven; ~130 PAM DANs among other types), R48B04-GAL4 (PAM-β′2, γ4, γ5n), split-GAL4 lines (MB056B, MB109B, MB312C, VT006202, MB042B, MB316B) for subsets, and R58E02-GAL4 (~90 PAM DANs).
  • Intersectional suppression with GAL80 lines to refine expression: R15A04-GAL80, R48B04-GAL80, R58E02-GAL80, TH-GAL80, tsh-GAL80.
  • Optogenetic inhibition: UAS-GtACR1 (green-light chloride channel) to inhibit 0273 neurons or β′2&γ4 DANs during CS+ to implant aversive memory; MB504B-driven GtACR1 to inhibit PPL1 aversive DANs; TH-GAL4 used to label/inhibit/activate PPL1 DANs.
  • Controls included genetic controls for light-only and driver-only conditions.
• Single-cell transcriptomics:
  • 10x Genomics Chromium scRNA-seq on two biological replicates (32 brains total), yielding 11,502 cells (mean ~5,673 transcripts per cell), aligning to Drosophila reference and CsChr transgene.
  • UMAP mapping of CsChr expression vs Fer2 expression; neurotransmitter marker analysis (Vmat, DAT, cholinergic, GABAergic, glutamatergic markers) to quantify cell-type composition.
• Neuroanatomy and connectomics:
  • Confocal imaging to quantify PAM DAN somata labeled under GAL4/GAL80 intersections.
  • FlyEM adult female hemibrain connectome analysis to identify upstream neurons (USNs) providing dendritic input to β′2, γ4, and γ5 DANs; clustering USNs by morphology and connectivity; visualization and quantification of parallel inputs.
• In vivo functional imaging:
  • Two-photon calcium imaging in R48B04-GAL4 flies co-expressing GCaMP6f and tdTomato; simultaneous recording of β′2, γ4, and γ5n DAN presynaptic arbors.
  • Odour stimuli: MCH and OCT, 10 s presentations, 4 repeats each, 20 s inter-odour intervals in satiated flies.
  • Feeding stimuli: 1 M sucrose, 20 s x4 with 160 s intervals in starved flies; additional 5 s water or sucrose in starved, dehydrated, or satiated flies, with post-water sucrose 10 min later to dissociate water vs sucrose components and state dependence.
• Statistics:
  • ANOVAs (one-way, two-way, repeated measures) with Tukey’s HSD or Šidák corrections where indicated; t-tests for imaging mean difference curves; effect sizes and detailed statistics in Supplementary Information.

• Synthetic activation of specific reward-encoding PAM DANs creates reward seeking despite punishment:
  • After training with optogenetic activation of 0273-GAL4 neurons (PAM-rich), ~50% of flies approached the electrified CS+ regardless of test duration (PI ~0), while genetic controls avoided shock.
  • Sucrose-trained wild-type flies avoided the CS+ when electrified, indicating natural sugar reward did not induce punishment-resistant seeking.
  • Flies trained with 0273 activation maintained conditioned approach even when shocks (90–120 V) were paired during training or when CS+/CS− sequence was reversed.
• Cell-type composition and contributions:
  • scRNA-seq of CsChr-expressing cells showed 16.7% dopaminergic, 63.1% cholinergic, 18.4% GABAergic, 1.8% glutamatergic; CsChr expression overlapped strongly with Fer2 expression.
  • Removing cholinergic or glutamatergic expression (via GAL80) enhanced 0273-mediated appetitive memory; reducing DAN expression with TH-GAL80 impaired memory and shock-resistant seeking, implicating DANs as necessary.
• Specific DAN subsets sufficient for unconstrained seeking:
  • Intersections identified R48B04-labeled PAM DANs as critical; β′2 and γ4 DANs (R15A04-GAL80; R48B04-GAL4) were sufficient to implant state-independent, punishment-resistant reward seeking comparable to 0273 activation, persisting up to 120 s tests and in satiated flies.
  • Sparse split-GAL4 lines labeling only fractions of β′2 or γ4 failed to reproduce robust, shock-resistant memory, suggesting co-activation of sufficient numbers of both β′2 and γ4 DANs is required.
• Functional antagonism with aversive DANs underlies punishment resistance:
  • Optogenetic inhibition of 0273 neurons or β′2&γ4 DANs during CS+ implanted aversive memory, indicating these reward DANs normally oppose aversive valuation.
  • Simultaneous inhibition of PPL1 aversive DANs (TH-GAL4/MB504B) abolished the aversive memory formed by 0273 inhibition, showing reliance on PPL1 output.
  • Prior activation of 0273, R48B04, or β′2&γ4 DANs impaired subsequent aversive learning and naive shock avoidance (transiently lasting at least 10 min, recovering by ~1 h), consistent with indirect suppression of PPL1 function.
• Reward seeking overrides physiological need:
  • Starved flies trained with 0273 or β′2&γ4 activation preferred the reward-predicting CS+ over a sucrose-laden alternative, indicating neglect of food.
  • Prior activation of β′2&γ4 DANs reduced subsequent sucrose approach in starved flies, suggesting a satiety-like demotivational signal; activation of γ5 alone did not produce this effect.
  • Co-activation of γ5n with β′2&γ4 weakened appetitive short-term memory across states, indicating γ5n provides auxiliary modulatory (including satiety-like) signals.
• Parallel, heterogeneous inputs to reward DANs:
  • Connectome analysis identified 1,996 upstream source neurons (USNs) to β′2/γ4/γ5 DANs (86 DANs total), including 1,718 USNs to β′2 and/or γ4, organized into 200 top input clusters spanning multiple brain regions.
  • Connectivity is highly parallel: 40 clusters specific to β′2, 8 to γ4, 33 to γ5, and 24 clusters shared between β′2 and γ4 (14 of which also connect to γ5), supporting multiplexed reward representation.
• State-dependent DAN responses:
  • Two-photon imaging showed β′2 and γ4 respond most to odours, γ5n responds strongly to sucrose with adaptation over repeats; β′2/γ4 show off-responses to sucrose.
  • Starvation enhanced responses to sucrose (but not water) in β′2, γ4, and γ5n; dehydration enhanced water responses in γ5n and sucrose responses in γ4/γ5n but not post-water sucrose, indicating specific gating by hunger/thirst.
• Overall, unconstrained seeking arises from both high incentive value assigned by co-activated β′2&γ4 DANs and concomitant transient suppression of aversive processing (PPL1), alongside satiety-like devaluation of food.

The study identifies precise dopaminergic mechanisms that enable reward seeking despite adverse consequences. Artificial co-activation of β′2 and γ4 PAM DANs assigns a high incentive value to predictive cues, sufficient to drive approach even when pursuit entails electric shock or foregoing food. At the network level, reward DAN activation indirectly suppresses aversive PPL1 DAN function, thereby reducing punishment sensitivity and impairing both aversive learning and naive shock avoidance. This antagonism explains risk-prone behaviour observed after reward DAN activation. Connectomic analyses reveal that β′2 and γ4 DANs receive diverse, parallel inputs, consistent with convergent representation of multiple reward types. Imaging demonstrates that, in normal conditions, these DANs are gated by physiological state: β′2 is modulated by hunger, γ4 by thirst, and γ5n by both, ensuring that learning and seeking are aligned with need. Synthetic co-activation bypasses this gating, producing state-independent, punishment-resistant and need-indifferent seeking. Together, these observations provide a mechanistic account wherein both elevated expected value and reduced aversive processing contribute to unconstrained reward pursuit. Given parallels with mammalian dopaminergic systems, these findings offer insight into neural dysfunctions underlying compulsive-like reward seeking.

This work shows that targeted activation of specific reward-encoding PAM dopaminergic neurons (β′2 and γ4) in Drosophila is sufficient to create reward seeking that persists despite punishment and suppresses feeding, recapitulating key features of unconstrained reward pursuit. The behavioural phenotype arises from (i) heightened incentive value assigned to cues through simultaneous activation of multiple reward-specific pathways and (ii) transient suppression of aversive (PPL1) processing, with additional satiety-like signals that devalue food. Connectomics confirms extensive, parallel inputs to these DANs, and functional imaging reveals state-dependent gating that is bypassed by synthetic activation. Future research should dissect how prolonged or substance-specific DAN dysregulation (e.g., by ethanol or sugars) drives long-term compulsive seeking, determine causal pathways mediating reward–aversion antagonism, and investigate inter-individual variability in susceptibility to unconstrained seeking.

• Driver specificity: 0273-GAL4 and R48B04-GAL4 also label non-dopaminergic and non-PAM neurons; while GAL80 intersections and controls implicate PAM DANs, contributions from other neurons in the brain cannot be fully excluded.
• GAL80 tools: GAL80 transgenes may not precisely mirror promoter-driven GAL4 patterns, potentially leading to incomplete or off-target suppression.
• Sparse split-GAL4 lines labeling only fractions of β′2 or γ4 failed to reproduce shock-resistant memory, limiting precise cell-by-cell sufficiency claims and suggesting that sufficient co-activation is necessary.
• Transience of aversion impairment: Suppression of shock avoidance after reward DAN activation is transient (minutes), and the longer-term dynamics are not fully characterized.
• Species and paradigm constraints: Findings are in Drosophila with acute, brief training sessions; extrapolation to mammalian systems and chronic addiction-like states requires caution.
• Natural vs synthetic rewards: Natural sucrose did not reproduce unconstrained seeking; how different natural rewards (e.g., ethanol) map onto specific DAN subsets over time remains to be established.