Approach

Our approach integrates behavior, large-scale neural recordings, targeted circuit manipulations, and computational modeling to uncover how distributed neural circuits generate adaptive behavior.

Behavior

Understanding how neural circuits generate adaptive behavior requires experimental paradigms that are both ethologically grounded and precisely controlled. We develop carefully structured tasks in rodents to isolate specific cognitive operations and relate them directly to neural population dynamics.

We embrace both an "outside-in" and an "inside-out" perspective, using each to generate and test mechanistic explanations that link neural dynamics to cognition (see the dual-perspective framework articulated by György Buzsáki, The Brain from Inside Out, 2019, and David Poeppel & Federico Adolfi, 2020).

Outside-in (behavior → brain): We leverage the flexibility and structure of animals’ behavior to infer the neural computations that generate it (e.g. Drieu et al., 2025).
Inside-out (brain → behavior): We identify candidate neural activity patterns and design tasks that directly test how these dynamics contribute to specific cognitive functions (e.g. Drieu et al., 2018).

Neither lens is sufficient on its own - behavior does not uniquely specify underlying mechanisms, and neural dynamics alone do not determine function. By integrating both approaches, we generate stronger, falsifiable hypotheses about how distributed neural circuits implement learning, memory, and adaptive cognition.

In freely moving rats, we combine rich navigation paradigms with precise experimental manipulations to study hippocampal function and related circuits supporting associative learning and episodic-like memory. We use multiple mazes, including a T-maze with return arms and a novel “wheel maze”, to probe memory-guided decision-making and flexible learning.

For experiments requiring two-photon resolution imaging, we use head-fixed mice performing well-controlled cue-guided instrumental tasks. By manipulating task contingencies, we can uncover latent knowledge and track true task acquisition, and examine how internal task representations evolve over learning.

T-maze

Wheel maze

Head-fixed auditory

go/no-go task

Neural recordings

To understand how distributed circuits implement cognitive computations, we combine precisely controlled behavioral paradigms with large-scale recordings of neural population activity across spatial and temporal scales.

We perform multi-site, high-density extracellular recordings to simultaneously monitor large neuronal populations across multiple brain regions. Using silicon probes with hundreds of recording sites, we capture single-unit and population activity with millisecond temporal resolution during behavior and sleep. This approach enables us to: (1) quantify coordinated dynamics across hippocampus and upstream/downstream targets, (2) track cell assembly activity and sequential structure across brain states, (3) characterize inter-regional communication and large-scale population coupling, and (4) relate moment-to-moment neural dynamics to task variables, behavioral performance, and learning stage. By recording simultaneously across circuits, we aim to identify distributed computations that support learning, memory, and behavioral flexibility.

To achieve cellular-resolution access to defined circuit elements, we use two-photon calcium imaging in behaving mice. We perform longitudinal imaging to track the same neuronal populations across days to weeks, allowing us to directly measure how identified networks reorganize during learning. We employ multi-plane imaging to record large ensembles of genetically defined neurons, and two-color imaging to simultaneously monitor specific long-range axonal projections alongside local circuit activity, or to measure distinct neuronal cell types within the same field of view. Together, these approaches allow us to connect distributed circuit dynamics and identified cell types to the computations that guide learning and behavior.

Hippocampal recordings during sleep

Cell tracking over learning

Neural manipulations

To establish causal links between neural activity and behavior, we perform temporally precise and cell-type–specific circuit perturbations.

We use closed-loop optogenetic approaches to manipulate neural activity patterns in real time. By detecting defined population events or specific behavioral responses, we can trigger temporally precise activation or inhibition of specific cell types or projections during behavior to test the functional relevance of identified neuronal activity patterns to learning and memory.

We employ closed-loop electrical stimulation to manipulate endogenous network events with high temporal precision. For example, we stimulate the ventral hippocampal commissure time-locked to sharp-wave ripple events to disrupt those events and directly test their role in memory consolidation and cross-regional communication.

To modulate circuit activity over longer timescales, we use chemogenetic tools (DREADDs) and pharmacological inactivation. These approaches allow reversible and cell-type-specific modulation of neural excitability across extended behavioral sessions.

1P stimulation of PV-ChR2+ neurons combined with 2P calcium imaging of excitatory auditory cortical neurons

Trial-specific, lick-based, closed-loop optogenetic inactivation over learning

Description of neuronal dynamic changes at multiple timescales revealed by low-rank tensor decomposition

Projection of tensor decomposition output onto principal component subspace

Data analysis

We develop and apply quantitative frameworks to extract structure from high-dimensional neuronal population activity and relate it to behavior, learning, and brain state.

We identify coordinated activity patterns and cell assemblies during wake and sleep, and characterize their sequential structure and reactivation across brain states. We examine how oscillatory activity shapes spike timing and population structure, and how rhythmic coordination across brain regions supports inter-regional communication and large-scale computations.

To capture population structure at scale, we use dimensionality reduction approaches, including state-space methods and low-rank tensor decompositions, to reveal latent dynamics underlying neural activity across neurons and brain areas, over time and across learning and performance.

We further implement encoding and decoding models, as well as statistical classifiers, to quantify how task variables, internal states, and learned representations are reflected in neural population activity. By tracking these representations at trial resolution across learning, we characterize how distributed circuits encode internal models and guide adaptive behavior.

Computational modelling

We have a strong interest in collaborating with computational neuroscientists and theorists to develop new tools for analyzing brain-wide data and to build models that link multi-region neuronal activity with behavior. These models can generate testable predictions about how distributed circuits implement learning, memory, and adaptive behavior.

Google Sites

Report abuse