Research
Animals learn from experience and flexibly adapt their behavior in a changing world
Our goal is to identify the neural computations that turn past experience into context-appropriate action—and to understand how those processes break down with aging and disease.
Research in the Heys Lab focuses on the computations and population codes in entorhinal–hippocampal circuits that turn past experience into context-appropriate action. We ask how training history and task structure shape these representations and the strategies animals adopt, and how these processes change with aging and under social isolation.
We pair timing and latent-state foraging paradigms with large-scale neural recordings (two-photon calcium imaging, Neuropixels) and projection-specific manipulations, including closed-loop perturbations during behavior. Experiments span precisely controlled virtual reality and naturalistic behavior.
Our approach is theory-driven: recurrent neural network models, latent-state inference and reinforcement learning frameworks, and normative foraging benchmarks generate predictions about neural dynamics (e.g., sequence alignment, remapping, representational geometry) that we test in vivo.
1. Temporal encoding in the brain
From the moment we wake, to the moment we sleep, we are actively keeping track of time. The ability of the nervous system to learn and remember, to infer cause and effect, and to make predictions about future outcomes, all depend critically upon temporal perception.
Despite the incredible ability of the nervous system to keep track of time across twelve orders of magnitude, there is very little mechanistic understanding of how the brain is able to encode it. We are interested in where these "clocks" are located in the brain, how their circuits generate temporal information, and how these signals are used to perform time-dependent behaviors.
2. Principles of synaptic organization
Neurons receive thousands of synaptic inputs that are distributed across their dendritic arbors. During learning, memories are thought to be encoded through persistent changes in the strength of these inputs and their functional and spatial organization.
Although theoretical models and in vitro experiments have proposed many candidate models for mechanisms for synaptic plasticity, synaptic organization, and dendritic integration, it has been difficult to directly test these models in the intact brain of an awake and behaving animal.
Here, we are able to address long standing questions related to the synaptic basis of learning and memory in the behaving animal using 2-photo imaging combined with molecular/genetic approaches. This approach enables monitoring of somatic and synaptic calcium activity at hundreds of individual synaptic spines distributed across the dendritic arbor. Importantly, we are using this approach in stable, chronic imaging preps that allow us to make these measurements over the course of days and weeks as mice learn and recall during memory guided behavior.
3. Dynamics of the recurrent entorhinal-hippocampal circuit
The predominant view of entorhinal-hippocampal circuit function has been feed-forward. In this view, theoretical models demonstrate how integration of synaptic input from medial entorhinal cortex (MEC) grid cells produces tuning properties of downstream hippocampal neurons. However, experiments designed to test these models have provided little evidence to suggest that simple feed-forward models accurately capture the physiological dynamics of the pathway.
In contrast, its architecture suggests that information is transferred in both directions, as synaptic connections provide feedback from the hippocampus into entorhinal cortex -thereby forming a synaptic loop. Therefore, to fully understand the mechanisms underlying memory function in this circuit, we must design experiments and build models that take into account its recurrent synaptic architecture.
4. Neural mechanism dysfunction in Alzheimer’s disease
During the progression of Alzheimer's disease (AD), the earliest detectable behavioral symptoms are often related to difficulties in forming and recalling episodic memories. While there has been some progress in terms of understanding these molecular and behavioral signatures of AD, many question remain open regarding how AD disrupts synaptic, cellular, and circuit level neural mechanisms - and even less is known about how these mechanisms are disrupted in vivo.
Chronic deep brain imaging via microprism.
Imaging methods allow for chronic sub-cellular resolution imaging. Image on top shows sparse labelling technique designed to monitor calcium dynamics across the dendritic arbor and at individual synaptic spines during memory-guided behavior.
Sparse GCaMP expression in MEC. Left, average DF/F image. Right, DF/F time series collected during virtual navigation.
Mice learn to perform head-fixed virtual navigation tasks designed to probe spatial and temporal memory.
Virtual Doorstop timing task: Mice learn to run down and stop at an invisible door for particular wait intervals (eg. 6 or 8 seconds).
Regular sequential neural activation during doorstop timing period. Heatmap (right) shows timing fields spread over the entire wait time period.
Updated imaging methods allow us to do large-scale multi-plane recordings from hundreds up to thousands of simultaneously recorded neurons in deep brain structures like MEC.
Multi-plane imaging allows for large scale population recording from the same or across neighboring cortical layers.