Seven scientists.One isthmus.

Risk preference is a promising transdiagnostic behaviour marker for many mental disorders. I will present a series of studies to understand the computational and neural mechanisms underlying individual differences in risk preferences. First, I will show a computational work on how bias in positive and negative learning rates can give rise to these individual differences and their association with anxiety and depression. Then, with a high-field functional Magnetic Resonance Imaging (7T-fMRI), we showed that this learning rate bias is associated with habenula and VTA BOLD functional connectivity. Finally, we study the neuromodulator basis of risk preferences. Specifically, leveraging the opportunity of deep brain stimulation (DBS) device implant surgeries in patients with movement disorders, we record sub-second (10 Hz) dopamine (DA) fluctuations in the human brain using fast-scan cyclic voltammetry (FSCV). We show that DA response to reward prediction errors is associated with individual differences in risk preferences.

Adaptive behavior requires flexibly shifting between exploiting familiar rewards and exploring novel opportunities. These explore-exploit decisions are implemented via a distributed brain network—anchored in frontopolar cortex (FPC) and ventromedial prefrontal cortex (vmPFC)—that computes the total value of a given choice by weighting the immediate value of familiar options against the latent future value of exploration. Capturing the precise temporal dynamics of these neural computations across spatially distributed cortical networks is critical to resolving how the brain manages this tradeoff. Here, we combined magnetoencephalography (MEG) with partially observable Markov decision process (POMDP) modeling during a reinforcement learning task. By decoding POMDP-derived choice policies in MEG source space, we mapped the precise spatiotemporal emergence of explore-exploit decisions across the cortex. We demonstrate that explore-exploit policy implementation unfolds via a hierarchical functional dissociation across the rostral prefrontal cortex. During decision formation, the lateral FPC initiates the strategic shift toward exploration hundreds of milliseconds before choice execution. Conversely, vmPFC and orbitofrontal cortex (OFC) do not exhibit early significant divergence from the baseline exploitative trace on exploration trials, displaying only a delayed, transient peak prior to choice execution. Following feedback, the vmPFC and OFC transition to a sustained representation of the choice’s empirical value to update reward expectancies and guide future actions. Mapping reinforcement learning variables onto millisecond-resolved neural data reveals how the human brain resolves the explore-exploit dilemma: through early strategic initiation of exploration in the FPC, a delayed exploratory shift in vmPFC and OFC, and sustained outcome evaluation to optimize future actions.

Two infants with similar brain injuries on clinical imaging can follow strikingly different motor trajectories—one developing typical movement, the other cerebral palsy. This divergence may depend not only on the visible brain injury itself, but on how the developing motor system reorganizes afterward. During the first two postnatal years, the corticospinal tract is refined through activity-dependent competition: bilateral projections present early in life are progressively withdrawn as predominantly contralateral motor control emerges. Perinatal brain injury can disrupt this process, creating a unique window into human motor-circuit development. We are longitudinally tracking corticospinal structure and function in infants with perinatal brain injury from 3–6 months through 24 months. Diffusion MRI measures corticospinal microstructure, while single-pulse transcranial magnetic stimulation (TMS), which our laboratory established as safe and feasible in infants, tests whether stimulation of the motor cortex produces a response in the contralateral limb. Standardized neuromotor and selective motor-control assessments link these circuit-level measures to emerging behavior. Preliminary analyses suggest that these measures capture complementary features of corticospinal organization. Infants with more organized corticospinal microstructure scored higher on concurrent neuromotor examination, and hemisphere-specific microstructure was associated with better selective control of the contralateral arm. Infants in whom TMS elicited a contralateral motor response also had higher neuromotor scores, and this difference persisted through 24 months. Together, these findings indicate that early corticospinal organization is associated with motor function across development. Combining structural and functional measures may therefore reveal how motor circuits reorganize after early brain injury, providing a window into human motor-system plasticity during a uniquely dynamic period of development.

Noradrenergic A1/A2 neurons are key contributors to aversion and negative affect, including during opioid and alcohol withdrawal. A substantial subset of these neurons co-expresses prolactin-releasing peptide (PrRP); however, little is known about the role of this peptide system. We combined genetic, behavioral, and whole-brain light sheet microscopy to test PrRP function and adaptations during ethanol abstinence. Using a conditional knockout of PrRP in tyrosine hydroxylase-expressing A1/A2 neurons, we observed a significant increase in time spent in the open arms of the elevated plus maze, time spent immobile in the forced swim test, and preference for saccharin, predominantly in females. Together, these data suggest a decrease in anxiety-like behavior alongside a shift to a more passive stress coping strategy and altered hedonic valuation. Knockout mice also displayed attenuated conditioned taste avoidance to ethanol-paired saccharin, consistent with altered hedonic/aversive processing. In the Chronic Drinking Forced Abstinence (CDFA) paradigm, PrRP knockout did not alter voluntary ethanol consumption. However, female knockouts failed to exhibit the protracted abstinence-related increase in forced swim immobility seen in female controls. Notably, this stands in contrast to the effects on forced swim immobility in ethanol-naïve animals, suggesting that PrRP plays a divergent, context-dependent role in behavioral responses to a stressor that is shifted by previous voluntary ethanol experience. Using whole-brain clearing and light-sheet microscopy, we observed a female-specific increase in PrRP terminal labeling in the parabrachial nucleus (PB) during protracted ethanol abstinence. Whole-brain cFos mapping across ethanol abstinence showed network reorganization, with decreased telencephalic and increased brainstem engagement. Additionally, functional coordination between the intermediate reticular nucleus (A1) and PB was heightened, paralleling the observed structural plasticity. These data identify PrRP as a sex-sensitive neuromodulator that shapes stress responses and ethanol abstinence-related negative affect possibly through parabrachial adaptations following chronic voluntary ethanol.

Environmental cues can reinforce actions and evoke strong, dynamic motivational states during drug seeking and relapse. In animal models, intermittent access to cocaine promotes sensory gated, state-level control of motivation that results in binge-like seeking thought to approximate human drug use. The basolateral amygdala (BLA) is critical for the associative learning underlying cue-guided behavior by linking affective valence to predictive sensory stimuli. Dopamine signaling is important for cue-driven motivational states and in vivo fiber photometry experiments revealed BLA dopamine signaling reflects flexible disambiguation of relative stimulus importance, independent of value. Further, in vivo optogenetic stimulation of ventral tegmental area (VTA) TH-neurons will prompt dopamine release in the BLA, and this evoked dopamine transmission was significantly larger after cocaine exposure. Fiber photometry was then used to measure in vivo BLA dopamine signaling and BLA D1-neuron activity during an intermittent access self-administration paradigm, where shifts between drug available and unavailable periods are signaled by changes in sensory context. Both BLA dopamine and D1-neuron signals dynamically tracked the transitions between drug availability states and cocaine-seeking actions. However, these signals exhibited different temporal profiles during cued state transitions, suggesting they may underlie unique facets of drug-related behaviors. The functional role of D1-containing BLA neurons in state-level control of drug seeking and reinforcement was then explored using optogenetics. Optogenetic augmentation of BLA D1-neurons potentiated the vigor of binge-like cocaine-seeking bouts, without affecting the amount of cocaine taken. Interestingly, optogenetic activation of the BLA D1-neurons will elicit dopamine release in the nucleus accumbens and is inherently reinforcing, as rats will readily seek self-stimulation. Together, these studies reveal the BLA dopamine system as a key hub for processing dynamic environmental control of addiction-like behaviors, and D1-neurons there encode emotional salience and contribute directly to reinforcement to promote operant learning.

Psychedelics, like psilocybin, are potentially powerful drugs to treat various disorders and diseases. Unlike other psychiatric medications, psilocybin clinical studies carefully optimize the treatment environment as it’s believed to foster a productive therapeutic mindset. While observational human studies support this notion, few animal models studies have investigated the contextually specific, psilocybin-elicited effects. Increased neuroplasticity in 5-HT2A–rich layer V pyramidal cells of the mPFC could underly these benefits, with glucocorticoid signaling also playing a modulatory role in psilocybin’s therapeutic efficacy. To address this, this study utilizes a 2×2 factorial design, where Sprague Dawley rats (n=43) are housed in either standard caging (SC) or environmental enrichment (EE), a larger cage with regularly changed novel objects, at weaning. After a month, rats are administered either psilocybin (1 mg/kg) or saline, and serum is collected measure acute corticosterone levels. Rats are then placed in an open field arena (OFT) for 10 minutes/sessions/day for 3 days to evaluate psilocybin- and EE-induced behavioral changes. Brains are collected for DiI labeling to assess dendritic spines in layer V mPFC pyramidal cells. We found psilocybin acutely increases corticosterone levels (p<0.0001). We also found trending increases in time in center for EE (p=0.056) and psilocybin (p = 0.058) animals, with an enhanced decrement time spent in center arena over the days in the SC group (housing×day: p=0.0003). When coding self-grooming behavior as a measure of arousal in the OFT, EE rats spent more time grooming per grooming bout on average (p<0.0001), but this effect could be influenced by the drug and day of the behavioral test (drug×housing×day: p=0.074). Early analysis shows that psilocybin increases spine density is increased in basal dendrites (p=0.029). Preliminary findings replicate prior effects on psilocybin corticosterone release and EE self-grooming length, with spine and mRNA sequencing data collection underway.

Semantic knowledge supports abilities shared across species, such as object recognition, and uniquely human abilities including language. A major unresolved question concerns the interpretability of individual neural units (cells or populations) in semantic representations. Units might each encode an independently-interpretable feature such as “has eyes”. Alternatively, as observed in deep neural networks, units may encode information interdependently – although individual units do not have clear meanings, the activity pattern across units is similar for similar concepts. Multivariate analysis methods showed initial promise for resolving this debate – yet they generate contradictory results that support opposing theories. First, I will present a new theoretical framework for describing representations. Kinds of representation proposed by different theorists can be classified on one framework, illustrating their similarities and differences and paving the way for research that adjudicates theories. Crucially, the framework highlights assumptions made by different multivariate methods that constrain the kinds of neural code each can detect. Differences in assumptions may explain heterogeneous findings. In the second project, I turned this methodological limitation to my advantage. I applied customised multivariate decoders, each making specific predictions about representational interpretability, to fMRI (7T) data. Models constrained to prefer interdependent codes outperformed models without this constraint, suggesting that semantic representations resemble the interdependent representations in neural networks. My third project, challenged this conclusion. I decoded semantic information from human intracranial electrocorticography (ECoG) data from 19 surgical patients. I tested the generalisation of decoders to time-windows on which they were not trained. Models trained on voltage data exhibited a gradually widening pattern of generalisation, a pattern also observed in neural networks. Conversely, models trained on high gamma power showed broad and constant generalisation, indicating a static, feature-based code. These studies indicate that representational interpretability – and degree of correspondence between brain and machine representation – varies crucially with granularity of measurement.

How neural circuits generate diverse behaviors is a fundamental question in neuroscience. Distinct behavioral outputs may arise from either dedicated motor circuits or shared circuits operating in different functional states. Using the rich vocal repertoire of the singing mouse (Scotinomys teguina), this work identifies two distinct vocal modes—ancestral ultrasonic vocalizations used for short-range communication and loud, rhythmic songs used over long distances. Despite their acoustic and contextual differences, both modes share a sound-production mechanism, phonatory-respiratory coupling, and vocal gating from the midbrain caudolateral periaqueductal gray. Mathematical modeling and circuit perturbations show how parametric tuning of this central circuit node can produce distinct vocal modes, offering a mechanistic basis for rapid behavioral evolution in mammals.

Temporal contiguity between events—how close they occur in time—is not sufficient to explain learning. However, prominent models of reinforcement learning that are frequently fit to human and animal behavior rely on contiguity to compute values of states and actions. My postdoctoral research sought to understand whether dopamine dynamics still resemble prediction errors when temporal contiguity between events in the environment is left unchanged, but the causal relation between events is altered. Using fiber photometry and optogenetics with outcome-selective contingency degradation in rats, I found that dopamine neuron activity in the midbrain and dopamine release in the ventral striatum are best explained by a model in which dopamine triggers retrospective causal inference. These results were consistent with simulations in which phasic dopamine encodes the sum of learned contingencies between meaningful events, but inconsistent with a standard temporal-difference model in which phasic dopamine encodes a reward prediction error. Overall, this work asks us to rethink how dopamine contributes to basic mechanisms of plasticity.

Alcohol Use Disorder is defined by shared diagnostic criteria, yet individuals show strikingly diverse drinking patterns and relapse vulnerabilities. To understand how bed nucleus of the stria terminalis (BNST) function varies across individuals, we used the Structured Tracking of Alcohol Reinforcement task to phenotype mice as high, low, or compulsive ethanol drinkers. BNST activation correlated with ethanol intake, while compulsive drinkers displayed elevated BNST calcium transients during drinking despite similar intake across groups. Forced abstinence uncovered phenotype-specific adaptations: high and low drinkers reduced seeking across abstinence, whereas compulsive drinkers persisted and showed increased BNST transients during protracted abstinence. Abstinence experience itself also potentiated compulsive drinking. Together, these findings reveal that ethanol abstinence precipitates increased seeking and compulsive intake linked to dynamic, phenotype-specific BNST activity, identifying a neural correlate of relapse vulnerability and offering insight into the biological diversity of alcohol use disorder.

Body-to-brain communication influences nearly every aspect of behavior, from immediate feelings of thirst or hunger to lifelong pair bonds and mood states. Yet technical and conceptual barriers have kept the neural mechanisms underlying these effects largely unexplored. This work develops and applies new tools for studying how signals from internal organs influence neural dynamics in behaving animals, combining systems and computational neuroscience, physiology, biochemistry, and bioengineering. Mouth- and gut-to-brain signals enable the brain’s thirst system to predict changes in hydration before they occur and adjust drinking preemptively. Delayed gut-to-brain food-poisoning signals also reactivate the brain’s representation of flavors from a recent meal, providing a mechanism for learning food aversions. Together, these findings offer neural explanations for long-enigmatic elements of everyday experience while revealing general principles for how learning and motivation are instantiated in the brain.

How the brain’s wiring diagram shapes neural computation is a fundamental question in neuroscience. To test the limits of connectome-based functional predictions, I developed a sparse imaging paradigm to characterize the visual selectivity of 91 cell types in the fruit fly optic lobe. Comparing measured responses with connectome-based predictions revealed good performance for some properties, such as orientation tuning, but surprisingly poor performance for others, such as receptive-field size. Strong connections consisting of numerous input synapses exerted a disproportionately large influence on postsynaptic activity: predictions based on a neuron’s single strongest input were as accurate as models considering its full complement of inputs. Strong inputs to a given cell type were also more functionally homogeneous than expected by chance. These results reframe our understanding of structure and function in the brain and define principles for improving future connectome-based functional predictions.

Learning by imitation is the foundation for verbal and musical expression, but its neural basis remains obscure. A juvenile male zebra finch imitates an adult tutor through a song-specialized cortico-basal ganglia circuit, providing a powerful system for locating the synaptic substrates of imitative motor learning. Combining a computational framework for quantifying song learning with synapse-specific optogenetic and chemogenetic manipulations, we identified the cortico-basal ganglia synapses that drive the acquisition and expression of rapid vocal changes and characterized the hours-long timescale over which those changes consolidate. Transiently augmenting postsynaptic activity in the basal ganglia briefly accelerated learning and persistently altered song. These results identify the synaptic locus that enables a juvenile songbird to learn to sing and reveal the circuit logic and behavioral timescales of imitative learning.

Anthropogenic change is creating environments enriched in toxic chemicals, including arsenic-contaminated drinking water. To understand biological resilience, we study Tokorhabditis tufae, a culturable nematode isolated from arsenic-rich Mono Lake. Unlike most nematodes, T. tufae hatches its progeny internally and gives birth to live young. Pharmacology and immunostaining reveal distinct neurobiological control of live birth, with changes that support egg retention and internal hatching. Gestation protects progeny from toxins until they are born in a stress-resistant dauer stage, and developing progeny transition from quiescence to a mobile state that facilitates birth through their own movement. These findings suggest that adaptation to an extreme environment involves coordinated neurobiological, physiological, and behavioral changes. Ongoing international collaborations are establishing T. tufae as a model for studying how neural circuits and behaviors evolve in contaminated environments.

Flexible visuospatial behavior requires transforming sensory input into stable, internally maintained action plans, a process thought to involve posterior parietal cortex. We recorded spiking activity and local field potentials from areas CIP and V3A in rhesus macaques performing memory-guided and visually guided saccade tasks. A brief, task-specific transient of beta-band synchronization emerged after target offset during memory-guided trials but was absent during visually guided behavior. It appeared earlier in CIP than V3A, predicted subsequent saccade accuracy, and preceded a sustained increase in induced beta power that also scaled with performance. Visually guided trials instead showed sustained beta suppression related to accuracy. Across tasks, beta dynamics showed a push–pull organization: suppression accompanied externally driven orienting, while enhancement accompanied endogenous maintenance. We propose that transient beta synchronization marks a hierarchical control signal that reconfigures network states for stable, internally guided action.

The ability to generalize learning across contexts is essential for navigating the world. We used calcium imaging to monitor hippocampal neurons as animals performed a conditioning task across multiple spatial contexts, asking whether cells could preserve task information even as their spatial encoding remapped. Dimensionality-reduction and machine-learning methods revealed that task-related neural representations remained stable while place-cell representations changed, producing similar embedding geometries across contexts. These patterns were consistent not only within individual animals but also across animals, suggesting a shared “neural syntax” in the hippocampus. The findings bridge memory and navigation research by showing how the hippocampus maintains cognitive consistency across different spatial environments, and suggest a common neural framework with broad implications for understanding memory, learning, and related cognitive processes.

Aggression is a fundamental social behavior used to gain and defend resources, territories, and mates. To explore its genetic basis, we leveraged Siamese fighting fish, which have been selectively bred for fighting for more than 700 generations. Controlled visual stimuli, including custom fish animations and 3D-printed models, quantified the markedly increased aggressive displays and biting attacks of fighting betta. Population-genetic analyses identified a strong selection signal centered on the neuroligin 1 gene. Although no coding differences were found between fighting and wild fish, fighting betta express two to four times less nlgn1 mRNA and show distinct spatial expression patterns during development. CRISPR/Cas9-mediated knockout fish will help establish nlgn1’s role in aggression and its mechanism of action. This work provides insight into the molecular, genetic, and neuronal mechanisms of aggressive behavior and the evolution of social behavior.