Research

Topics of interest

Information content of neuromodulatory signals

The finding that dopamine signals reflects differences between expected and experienced rewards to drive value-based learning is one of the most replicated and influential in all of neuroscience. However, it's become clear that dopamine neurons take into account many other forms of information in order to craft a teaching signal, and that this can vary across genetically-defined subpopulations. We are interested in uncovering what types of behavior-relevant variables shape dopamine signals, what are the cellular mechanisms that control this, and how these signals interact with post-synaptic targets and other neurotransmitters to influence behavior.

Mechanisms of model-based learning

Animals, including rats and humans, don't just learn what is explicitly relevant for their survival. Rather, they build abstract representations of their environment that takes into account many different facets of their experience. We aim to uncover how these so called mental models or cognitive maps are built and deployed, what are the qualitative and quantitative parameters that define them, and what are their cellular and circuit substrates.

Understanding brain function across different levels

Neuroscience is concerned not only with the material building blocks of a system (the brain) but also with the abstract information that this system processes. One of the major difficulties that this imposes is that any single cognitive process involves computing information at several different scales, including specific proteins, single cells, local circuits, long range networks, and ultimately behavior. Additionally, the brain is a dynamic system, continuously changing how it processes information with each experience. In the K.M. Costa lab, we integrate different levels of analysis to understand how molecular, cellular, and circuit processes influence each other and behavior, and how these components adapt to different life experiences and physiological challenges.

Methods

We apply a number of tools to test hypotheses on the relationship between brain and behavior at different levels of resolution. Below we highlight some of our approaches.

Behavioral tasks

The lab employs theory-inspired behavioral tasks of different modalities to isolate key variables related to learning and decision-making. We work primarily with freely-moving rats and our tasks include different sets of Pavlovian and instrumental paradigms, including classical tasks like sensory preconditioning and reinforcer devaluation, and complex odor-guided tasks where rats learn to make appropriate decisions across several trials. Many of our tasks focus on latent learning and inference testing, as these processes can reveal a lot about what types of information rats pick up when exploring the world.

Recording neural correlates

By recording different forms of neural activity as rats execute our tasks, we can infer how distinct neural mechanisms guide behavior. The lab employs many different techniques, including multi-site fiber photometry, endoscopic imaging, and electrophysiology, to measure signals like neurotransmitter release, calcium transients, and neuronal firing in freely-moving rats.

Manipulating neural activity and cellular properties

We can test hypotheses on how the brain controls behavior by disrupting or manipulating neural activity and observing the effects of these interventions on behavior. The lab is set up to use several tools for neural manipulation, from more recent approaches like chemogenetics and optogenetics, to classical methods like pharmacology and lesions. In addition, the lab is establishing methods for cell-type specific gene editing in adult rats using CRISPR, which will allow us to understand how specific genes shape behavior.

Computational modelling

A strong theoretical framework is generally important both for formulating hypotheses and interpreting experimental results. This is especially true in behavioral neuroscience, where the studied constructs are often hard to define and quantify. We apply computational approaches, both in house and in collaboration with specialized theoreticians, to guide our experimental work and analyze empirical data.