The orbitofrontal cortex (OFC) encodes reward predictions (i.e. value) at the level of both single neurons, and the population. This work aimed to understand the role of local field potentials in relation to value encoding. Unfortunately, dissecting the causal contributions of this encoding can be challenging. I developed a closed-loop system capable of directly targeting theta oscillations without affecting underlying neuronal firing rates. We found that OFC neurons preferentially encode value in phase with theta (4-8 Hz) field oscillations. Theta is a low frequency signal related to long-range communication between brain regions while providing conditions favorable for synaptic learning processes. When theta activity is disrupted with closed-loop electrical stimulation, value encoding of single OFC neurons too was disrupted, resulting in our subjects becoming unable to update their beliefs about the value of task-relevant stimuli. When recording from hippocampus, a region that has significant theta-band modulation during behavior, I found that theta activity in hippocampus is a driver of OFC theta, implicating hippocampus in the task. Disrupting hippocampal theta signals with the same closed-loop system produced a similarl disruption of behavioral flexibility. This work was recently published in Neuron (https://doi.org/10.1016/j.neuron.2020.02.003).
Hippocampus is hypothesized to serve as the computational substrate for a ‘cognitive map’ a representation of structural information whether that structure is physical space or otherwise. Structural information can be thought of as a model of the world that embeds current experience (state) into past experience. This model is then exploited by prefrontal cortex (PFC) to flexibly drive behavior. But how is this model instantiated? A canonical example of a world model is the spatial map found in the hippocampal-entorhinal (HC-EC) system as animals navigate their environment. Place cells in HC provide a readout of current, past, and even future locations within an environment, while grid cells within EC represent the space itself. Together, the two structures provide a detailed model of the environment and the animal’s experience within it.
However, our world is filled with structure beyond the physical. For example, knowing how members of a large family are interrelated represents non-spatial, conceptual structure; knowing that Eric's daughter is Penny and Eric and Maggie are siblings, one can infer that Penny is Maggie's niece. Because of its well-established role in mapping physical structure, the HC-EC system is increasingly thought to provide the means to organize and exploit this structure. Unfortunately there has been little direct neurophysiological evidence to support this notion. This work aims to study the role of HC in value learning by investigating single neuron activity in the hippocampus during a learning task designed to have a clear underlying structure unobservable to subjects. The data suggest that HC neurons encode this abstract structure with the same properties observed in the rodent navigating space: though value place cells. Similar to rodent spatial place cells, value place cells fire consistently in value space across time, map all the dimensions of value space, remap in response to contextual changes in the task, and are sensitive to the direction of movement through space. We are currently preparing to submit this work to Cell.