Biology subjects

Costa, V. D.

Publications and source records attributed to Costa, V. D..

13 recordsLinked to original sources

bioRxiv2026

Spatiotemporal Decoding of Explore-Exploit Decisions in the Human Brain

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 choices 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.

neuroscience↗

bioRxiv2026

AI-Discovered Cognitive Models Reveal Novel Insights into Human and Animal Learning

Scientific models are widely used across the natural sciences as an interface between scientific theories and empirical data [1]. Such models play a key role, for example, in the study of human and animal learning, where they express algorithmic hypotheses and relate them to psychology and neuroscience data [2, 3]. These models are traditionally handcrafted by expert researchers based on existing theory or new insights. Such handcrafted models, however, are now known to fall short of capturing the full richness of behavior, even in their narrow domains [4-7]. An alternative data-driven approach has emerged, seeking to discover new insights by fitting and interpreting flexible models [8-11]. However, these tools require substantial human effort to derive insight from data, and it has been unclear how to discover new ideas from data efficiently. Here, we present DataDIVER, a general approach for automatically discovering computational models from data, and demonstrate that these models surface novel mechanistic insights into human and animal learning. Our approach delivers models that take the form of short computer programs, which are optimized both to fit data well and to be simple. These programs explicitly connect with existing theoretical frameworks and are readily understandable by human scientists. They can also be used to make novel predictions, some of which we show are borne out in re-analysis of existing data. General-purpose tools for surfacing new ideas from data, especially in combination with the large datasets that are increasingly available in many fields, stand to dramatically accelerate scientific discovery.

animal behavior and cognition↗

bioRxiv2026

Retrograde transduction of dopaminergic cells in substantia nigra of rhesus monkey

Neuromodulatory systems regulate neural circuits across broad regions of the brain, and disruption of the dopaminergic system contributes to psychiatric and neurodegenerative disorders. Engineered viral vectors have been used to target the neuromodulatory systems of the nonhuman primate brain (Y. Chen et al., 2023; El-Shamayleh et al., 2016; Gray et al., 2010; Lerchner et al., 2014; Perez et al., 2022). However, a conspicuous obstacle to the isolation and modulation of specific pathways is the inability of many retrogradely infecting viruses to transduce dopaminergic (DA) cells efficiently (Tervo et al., 2016; Cushnie et al., 2020; Weiss et al., 2020). We compare the DA neuron retrograde transduction efficacy of four viral vectors after injection into the striatum of nonhuman primates (NHP). Selectivity was assessed by comparing the neuronal co-expression of fluorescent reporter protein and tyrosine-hydroxylase (TH) antibody in substantia nigra pars compacta (SNc). The rabies pseudotyped lentiviral vector, FuG-B2, produced superior retrograde transduction of DA cells to FuG-C or FuG-E. AAV2.retro was the least effective.

neuroscience↗

bioRxiv2025

Risk reshapes amygdala representation of choice

Modifying behavior in response to changing environmental conditions is a crucial adaptive function. This capacity is exemplified when animals curtail pursuit of a valued outcome that risks being punished by aversive consequences, but the mediating brain mechanisms remain poorly understood. Here, using in vivo cellular-resolution calcium (Ca2+) imaging, optogenetics and chemogenetics, we show that risk of punishment dramatically alters animals choice between a large/risky and small- /safe reward and produces novel, causally necessary, patterns of activity in basolateral amygdala (BLA) neurons. We find that experience of a punished outcome generates a BLA representation that is selectively replayed when animals subsequently abort choice of the large/risky reward option. Additionally, we show that risk leads to the incorporation of newly encoding BLA neurons into the pre-choice representation, which predicts shifting away from the large/risky option. These findings reveal how dynamic reshaping of BLA representations underpins behavioral flexibility in the face of risk.

neuroscience↗

bioRxiv2025

Dissociable effects of curiosity and hedonic valence on reinforcement learning

Curiosity and exploration support learning and adaptive decision-making in uncertain environments. While these processes are sensitive to motivational context, it remains unclear how outcome valence shapes exploration across species. Human studies suggest that aversive contexts increase exploration, but these effects often rely on verbal framing and explicit instructions. To gain deeper insight into how exploration strategies emerge from experience alone, this study investigated the influence of hedonic valence on novelty seeking, exploration, and reinforcement learning in rhesus macaques. Using visual tokens as secondary reinforcers, we found that monkeys explored novel, uncertain options more frequently when exploitation would lead to losses rather than gains. However, our analyses clarified that this heightened novelty seeking was primarily a consequence of the monkeys employing an optimistic prior belief about the value of novelty, rather than a categorical, valence-dependent shift in their underlying curiosity or the information bonus associated with exploration. Approach and avoidance motivation did influence other aspects of reinforcement learning. Monkeys demonstrated faster learning from losses than from gains, indicating that they were averse to losing tokens. They also frequently chose an option and then quickly aborted their choice. These choice balks were strategic responses to approach-avoidance conflicts and uncertainty, and represented self-generated bouts of exploratory behavior that led to valence-dependent use of directed and random exploration. These findings suggest that different strategies are used to manage explore-exploit tradeoffs induced by novelty or internal motivational conflicts, revealing dissociable effects of curiosity and hedonic valence on reinforcement learning.

animal behavior and cognition↗

bioRxiv2025

Synthetic Serum Markers Enable Noninvasive Monitoring of Gene Expression in Primate Brains

We demonstrate a noninvasive approach to measure transgene expression in the brain of nonhuman primates using blood tests with engineered reporters called Released Markers of Activity (RMAs). RMAs can exit the brain and enter the bloodstream via reverse-transcytosis across the blood-brain barrier. We demonstrate that these reporters can be used to repeatedly monitor expression of multiple transgenes in cortical and subcortical brain regions simultaneously over a period of three months. RMAs are also sensitive enough to detect circuit-specific Cre-dependent AAV expression. Through this study, the RMA platform provides a cost-efficient, noninvasive tool for neuroscience study of large animals, enabling sensitive, multiplexed, and repeatable measurements of gene expression in the brain with a blood test.

neuroscience↗

bioRxiv2025

Erasable Synthetic Serum Markers

Gene expression in the brain is typically evaluated using invasive biopsy or postmortem histology. Serum markers provide an alternative way to monitor the brain, but relatively few such markers exist. Additionally, the origin of serum markers often cannot be localized to a specific cell population, and monitoring dynamic changes in their gene expression is compromised by the same factor that makes the markers detectable - long serum half-life. Here we propose a paradigm to improve the sensitivity of serum marker measurement by modifying the markers in vivo, called erasable serum markers, or ESM. As a proof of concept, we use a well-controlled system with known half-life and tunable serum levels. This system, released markers of activity, or RMAs enable measurement of transgene expression in the brain through a simple blood test. RMAs are stable in blood, with a half-life of >100 h and can detect expression from as few as 12 neurons in mice. However, their long serum half-life also generates long-lasting background signals when RMA are used to track temporal changes in gene expression. By engineering on-demand erasable RMAs and injecting an intravenous targeted protease, we reduced RMA background signal by more than an order of magnitude without compromising the detection sensitivity. Similarly to previous RMA iteration, our approach showed a 65,000-fold increase in their signal over the baseline when expressed in a single brain region. Furthermore, we demonstrated that this erasable RMA system improves the dynamic range of detection for low-level promoter activity that is driven by physiological levels of c-Fos.

bioengineering↗

bioRxiv2024

Transcriptomic diversity of amygdalar subdivisions across humans and nonhuman primates

The amygdaloid complex mediates learning, memory, and emotions. Understanding the cellular and anatomical features that are specialized in the amygdala of primates versus other vertebrates requires a systematic, anatomically-resolved molecular analysis of constituent cell populations. We analyzed five nuclear subdivisions of the primate amygdala with single-nucleus RNA sequencing in macaques, baboons, and humans to examine gene expression profiles for excitatory and inhibitory neurons and confirmed our results with single-molecule FISH analysis. We identified distinct subtypes of FOXP2+ interneurons in the intercalated cell masses and protein-kinase C-{delta} interneurons in the central nucleus. We also establish that glutamatergic, pyramidal-like neurons are transcriptionally specialized within the basal, lateral, or accessory basal nuclei. Understanding the molecular heterogeneity of anatomically-resolved amygdalar neuron types provides a cellular framework for improving existing models of how amygdalar neural circuits contribute to cognition and mental health in humans by using nonhuman primates as a translational bridge.

neuroscience↗

bioRxiv2024

Contribution of amygdala to dynamic model arbitration under uncertainty

Intrinsic uncertainty in the reward environment requires the brain to run multiple models simultaneously to predict outcomes based on preceding cues or actions, commonly referred to as stimulus- and action-based learning. Ultimately, the brain also must adopt appropriate choice behavior using reliability of these models. Here, we combined multiple experimental and computational approaches to quantify concurrent learning in monkeys performing tasks with different levels of uncertainty about the model of the environment. By comparing behavior in control monkeys and monkeys with bilateral lesions to the amygdala or ventral striatum, we found evidence for dynamic, competitive interaction between stimulus-based and action-based learning, and for a distinct role of the amygdala. Specifically, we demonstrate that the amygdala adjusts the initial balance between the two learning systems, thereby altering the interaction between arbitration and learning that shapes the time course of both learning and choice behaviors. This novel role of the amygdala can account for existing contradictory observations and provides testable predictions for future studies into circuit-level mechanisms of flexible learning and choice under uncertainty.

neuroscience↗

bioRxiv2023

Recurrent activity within microcircuits of macaque dorsolateral prefrontal cortex tracks cognitive flexibility

Human and non-human primate data clearly implicate the dorsolateral prefrontal cortex (dlPFC) as critical for advanced cognitive functions1,2. It is thought that intracortical synaptic architectures within dlPFC are the integral neurobiological substrate that gives rise to these processes, including working memory, inferential reasoning, and decision-making3-7. In the prevailing model, each cortical column makes up one fundamental processing unit composed of dense intrinsic connectivity, conceptualized as the canonical cortical microcircuit3,8. Each cortical microcircuit receives sensory and cognitive information from a variety of sources which are represented by sustained activity within the microcircuit, referred to as persistent or recurrent activity4,9. Via recurrent connections within the microcircuit, activity can propagate for a variable length of time, thereby allowing temporary storage and computations to occur locally before ultimately passing a transformed representation to a downstream output4,5,10. Competing theories regarding how microcircuit activity is coordinated have proven difficult to reconcile in vivo where intercortical and intracortical computations cannot be fully dissociated5,9,11,12. Here, we interrogated the intrinsic features of isolated microcircuit networks using high-density calcium imaging of macaque dlPFC ex vivo. We found that spontaneous activity is intrinsically maintained by microcircuit architecture, persisting at a high rate in the absence of extrinsic connections. Further, using perisulcal stimulation to evoke persistent activity in deep layers, we found that activity propagates through stochastically assembled intracortical networks, creating predictable population-level events from largely non-overlapping ensembles. Microcircuit excitability covaried with individual cognitive performance, thus anchoring heuristic models of abstract cortical functions within quantifiable constraints imposed by the underlying synaptic architecture.

neuroscience↗

bioRxiv2021

The neurocomputational architecture of explore-exploit decision making

Humans and other animals often make the difficult decision to try new options (exploration) and forego immediate rewards (exploitation). Novelty-seeking is an adaptive solution to this explore-exploit dilemma, but our understanding of the neural computations supporting novelty-seeking in humans is limited. Here, we presented the same explore-exploit decision making task to monkeys and humans and found evidence that the computational basis for novelty-seeking is conserved across primate species. Critically, through computational model-based decomposition of event-related functional magnetic resonance imaging (fMRI) in humans, these findings reveal a previously unidentified cortico-subcortical architecture mediating explore-exploit behavior in humans.

neuroscience↗

bioRxiv2021

Novel entropy-based metrics for predicting choice behavior based on local response to reward

For decades, behavioral scientists have used the matching law to quantify how animals distribute their choices between multiple options in response to reinforcement they receive. More recently, many reinforcement learning (RL) models have been developed to explain choice by integrating reward feedback over time. Despite reasonable success of RL models in capturing choice on a trial-by-trial basis, these models cannot capture variability in matching. To address this, we developed novel metrics based on information theory and applied them to choice data from dynamic learning tasks in mice and monkeys. We found that a single entropy-based metric can explain 50% and 41% of variance in matching in mice and monkeys, respectively. We then used limitations of existing RL models in capturing entropy-based metrics to construct a more accurate model of choice. Together, our novel entropy-based metrics provide a powerful, model-free tool to predict adaptive choice behavior and reveal underlying neural mechanisms.

neuroscience↗

bioRxiv2020

A pragmatic reevaluation of the efficacy of nonhuman primate optogenetics

Translational neuroscience is committed to generating discoveries in the laboratory that ultimately can improve human lives. Optogenetics has received considerable attention because of its demonstrated promise in rodent brains to manipulate cells and circuits. In a recent report, Tremblay and colleagues (2020) introduce an open resource detailing optogenetic studies of the nonhuman primate (NHP) brain and make robust claims about the translatability of the technology. We propose that their quantitative (e.g., a 91% success rate) and theoretical claims are questionable because the data were analyzed at a level relevant to the rodent but not NHP brain, injections were clustered within a few monkeys in a few studies in a few brain regions, and their definitions of success was not clearly relevant to human neuropsychiatric disease. A reanalysis of the data with a modified definition of success that included a behavioral and biological effect revealed an 62.5% success rate that was lower when considering only strong outcomes (53.1%). This calls into question the current efficacy of optogenetic techniques in the NHP brain and suggests that we are a long way from being able to leverage them in "the service of patients with neurological or psychiatric conditions" as the Tremblay report claims.

neuroscience↗