Jérémie Naudé

Behavioural experiments often hinge on moments when animals crowd together. In these situations, existing pose-estimation methods usually fail because detected crops include neighbouring animals, and their body parts are easily confused between individuals. We solved it with DeepLabSAM, a pipeline that segments and tracks every animal from a single text prompt using the SAM 3 foundation model, followed by a mask-gated pretrained pose estimation network from DeepLabCut. Across recordings of multiple interacting mice, the proposed method substantially reduces misassigned keypoints under heavy occlusion. Furthermore, we used DeepLabSAM to automatically construct the Multi-Task Mouse Behaviour (MTMB) dataset, comprising 14,300 frames spanning nine behavioural assays, with each instance annotated for bounding box, mask, identity, and pose. SuperAnimal pretrained pose models transferred unpredictably across this dataset. Conversely, MouseLite, our RF-DETR pose-estimation model trained on the MTMB dataset, matched their in-distribution accuracy, generalized better to unseen assays, and achieved real-time (faster-than-acquisition) inference. Combined, DeepLabSAM, MTMB dataset, and MouseLite provide a scalable framework for robust, cross-assay behavioural pose estimation of freely interacting mice.

Learning from outcomes requires weighing each surprise by how much it is likely to mean. Two properties of the environment set that weight in opposite directions: volatility, the rate at which contingencies change, and stochasticity, the noise in outcomes around a stable contingency. Whether animals separate the two is unresolved, in part because varying reward probability in a reversal task makes outcomes noisier and the two options harder to tell apart at once. We trained male and female mice on a probabilistic reversal task for intracranial self-stimulation, scaling reward magnitude with reward probability so that the expected reward of each option and the difference between them stayed constant while outcome variance varied, and crossing this with different reversal rates. Simulations show that adapting the learning rate alone and adapting the decision policy alone are near-equivalent solutions across the environments we tested. Mice adopted the first adjustment: their fitted learning rate rose with volatility and fell with stochasticity, in both directions predicted by reward maximization, while their inverse temperature stayed fixed. This joint dependence was reproduced only by a meta-learning rule that reads stochasticity from the size of prediction errors and volatility from how often recent outcomes favored switching.

Defensive survival behaviors are shaped by both danger attributes and internal physiological states. Well-characterized defensive responses depend not only on whether danger is potential or perceived, but also on its spatial and or temporal proximity. The relationship between danger imminence and defensive behaviors is referred to as the threat imminence continuum, which provides a conceptual framework bridging ecology and behavioral neuroscience. Although this theoretical framework is gaining momentum, another much less explored danger attribute is the extent to which the environment may be controllable. Most paradigms assessing defensive behaviors rely on conditioned stimuli as danger warning-signals in stable environments that hamper the study of environment controllability and its impact on defensive behaviors. To address this question, we repurposed the active place avoidance (APA) paradigm combined with supervised machine learning to characterize multivariate exploratory and defensive behaviors in male and female mice. One important benefit of APA is that it allows the study of active avoidance when the platform is rotating as well as deliberative avoidance when the platform is immobile. As such, APA is very well suited to test the impact of environment controllability on the expression of defensive behaviors. Using this unified preparation, we systematically compared quantitative danger attributes (punishment intensity, behavioral agency) across danger states (absence, presence and omission of punishment) and learning experience (early and late learning stages), while considering sex as a biological variable. Our results demonstrate that defensive behaviors depend on punishment intensity in a sex-dependent manner in APA. Defensive responses can be classified as proactive (risk assessment) or reactive (freezing, escape) in APA. Their response probability depends on behavioral agency, is mutually exclusive and depends on danger spatial proximity. Acute treatment with norepinephrine transporter blocker selectively increases reactive defensive behaviors while sparing proactive defensive behavior. Collectively, these results provide experimental evidence that environmental stability calibrate proactive and reactive defensive behaviors whose neural correlates may be divergent.

Adaptive decision-making in dynamic environments requires flexible adjustment of learning speed to balance stability and flexibility. When outcomes are highly stochastic, learners must avoid over-interpreting noise and update more slowly, whereas in volatile environments where contingencies change frequently, learning should accelerate to rapidly incorporate new evidence. Theories propose that internal estimates of uncertainty tune learning rates through neuromodulation-dependent mechanisms. Here, we investigated how noradrenergic inputs from the locus coeruleus (LC) to the orbitofrontal cortex (OFC) support adaptive learning under uncertainty. We show that, in a probabilistic reversal learning task performed across different levels of stochasticity, rats exhibited behavior best explained by an adaptive reinforcement-learning model in which learning rates dynamically adjusted according to model-estimated stochasticity and volatility, outperforming standard fixed-rate models. Noradrenaline release in the OFC closely tracked trial-by-trial, model-derived volatility estimates around contingency changes. Disrupting LC→OFC noradrenergic inputs reproduced the model-predicted impairment in adaptive learning-rate adjustment associated with model-estimated volatility. Together, these findings identify OFC noradrenergic signaling as a key circuit mechanism supporting learning-rate adjustment in response to internal estimates of volatility during adaptive decision-making under uncertainty.

Adaptive decision-making in dynamic environments requires flexible adjustment of learning speed to balance stability and flexibility. When outcomes are highly stochastic, learners must avoid over interpreting noise and update more slowly, whereas in volatile environments where contingencies change frequently, learning should accelerate to rapidly incorporate new evidence. Theories propose that internal estimates of uncertainty tune learning rates through neuromodulatory-dependent mechanisms. Here, we investigated how noradrenergic inputs from the locus coeruleus (LC) to the orbitofrontal cortex (OFC) support adaptive learning under uncertainty. We show that rats performing a probabilistic reversal learning task exhibited behavior that was best explained by an adaptive reinforcement-learning model in which learning rates dynamically adjust according to estimated environmental volatility and stochasticity, outperforming standard fixed-rate models. Noradrenaline release in the OFC closely tracked trial-by-trial, model-derived volatility estimates around contingency changes. Disrupting LC→OFC noradrenergic inputs reproduced the model-predicted deficit in volatility-dependent adjustments of learning rate. Together, these findings identify OFC noradrenergic signaling as a key circuit mechanism for volatility-dependent modulation of learning rates during adaptive decision-making.

The striatum plays a key role in decision-making, with its effects varying with anatomical location and direct and indirect pathway striatal projecting neuron (d- and iSPN) populations. Using a mouse gambling task with a reinforcement-learning model, we described individual decision-making profiles as a combination of three archetypal strategies: Optimizers, Risk-averse, and Explorers. These strategies reflected stable differences in the parameters generating decisions (sensitivity to the reward magnitude, to risk, or to punishment) derived from a reinforcement-learning model of animal choice. Chemogenetic manipulation showed that dorsomedial striatum (DMS) neurons substantially affect decision-making, while the nucleus accumbens (NAc) and dorsolateral striatum neurons (DLS) have lesser or no effects, respectively. Specifically, DMS dSPNs decrease risk aversion by increasing the perceived value of risky choices, while DMS iSPNs emphasize large gains, affecting decisions depending on decision-making profiles. Hence, we propose that striatal populations from different subregions influence distinct decision-making parameters, leading to profile-dependent choices.

Summary: Social organization and division of labor are fundamental to animal societies, but how do these structures emerge from individual interactions, and what role does neuromodulation play in shaping them? Using behavioral tracking in a semi-natural environment, neural recordings, and computational models integrating reinforcement and social learning, we show that groups of three isogenic mice spontaneously develop specialized roles while solving a foraging task requiring individual decisions under social constraints. Moreover, these roles are shaped by dopaminergic activity in the ventral tegmental area. Strikingly, despite minor sex-differences in behavior when mice were tested alone, male triads formed stable worker-scrounger relationships driven by competition, whereas female triads adopted uniform, cooperative strategies. Model analysis revealed how intra- and inter-sex parameter differences in resource exploitation, combined with contingent and dynamic social interactions, drive behavioral specialization and labor division. Most notably, it highlighted how contingency, amplified by competition, magnifies individual differences and shapes social profiles. The plastic, adaptive nature of social organization within triads was confirmed by manipulating dopaminergic cell activity, which reshaped social roles and altered group structure. Our findings support a feedback loop where social context shapes neural states, which in turn reinforce behavioral specialization and stabilize social structures.

Phasic variations in dopamine levels are interpreted as a teaching signal reinforcing rewarded behaviors. However, behavior also depends on the motivational, neuromodulatory effect of phasic dopamine. In this study, we reveal a neurodynamical principle that unifies these roles in a recurrent network-based decision architecture embodied through an action-perception loop with the task space, the MAGNet model. Dopamine optogenetic conditioning in mice was accounted for by an embodied network model in which attractors encode internal goals. Dopamine-dependent synaptic plasticity created “latent” attractors, to which dynamics converged, but only locally. Attractor basins were widened by dopamine-modulated synaptic excitability, rendering goals accessible globally, i.e. from distal positions. We validated these predictions optogenetically in mice: dopamine neuromodulation suddenly and specifically attracted animals toward rewarded locations, without off-target motor effects. We thus propose that motivational dopamine reveals dopamine-built attractors representing potential goals in a behavioral landscape.

The striatum plays a key role in decision-making, with its effects varying with anatomical location and direct and indirect pathway striatal projecting neuron (d- and i-SPN) populations. Using a mouse gambling task with a reinforcement-learning model, we described of individual decision-making profiles as a combination of three archetypal strategies: Optimizers, Risk-averse, Explorers. Such strategies reflected stable differences in the parameters generating decisions (sensitivity to the reward magnitude, to risk or to punishment) derived from a reinforcement-learning model of animal choice. Chemogenetic manipulation showed that dorsomedial striatum (DMS) neurons substantially impact decision-making, while the nucleus accumbens (NAc) and dorsolateral striatum neurons (DLS) have lesser or no effects, respectively. Specifically, DMS dSPNs decrease risk aversion by increasing the perceived value of risky choices, while DMS iSPNs emphasize large gains, affecting decisions depending on decision-making profiles. Hence, we propose that striatal populations from different subregions influence distinct decision-making parameters, leading to profile-dependent choices.