Complex Spatial Decision-Making: Investigations Into Explore-Exploit Behaviors and CA1 Multi-Field Cell Activation in Megaspace

Abstract

Explore-exploit behavior is a decision-making process involving the exploration of new options without the certainty of consequence or the exploitation of familiar options known to have a predictable outcome. This behavior is critical for organisms to survive and thrive when navigating the real world, but how is such a balance achieved between these decision-making tendencies? What characteristics can be measured/controlled to quantify them? We address these questions using the same changing bandit task in rats and humans. In the changing bandit task, a subject is asked to choose two options with a priori unknown rewards/outcomes. Based on the session type, each option possesses a constant, unknown hazard rate which determines the probability at which the reward amount changes at each option, modulating the volatility/reliability across each trial. We hypothesize that both rats and humans will learn to switch options (explore) after the lowest reward amount is dispensed especially if the hazard rate is high, and will persist if the option yields the highest reward especially if the hazard rate is low. The rats are placed in a maze where they must first go to an unrewarded “home base” before two feeders opposite the home base start blinking. Rats probabilistically receive either zero, one, or three drops of a sucrose solution as rewards. Preliminary results show that both male and female rats are more likely to exploit the highest rewarding feeder than the reliable but less rewarding feeder and seem to display more exploitative behaviors with time. A MATLAB-based GUI was created and used for the human subjects to mimic the changing bandit task using a point-based reward system. Human subjects experience the exact same series of decisions as the rats. Decisions, reaction times and computer mouse tracking data are collected. Initial results from seven participants indicate that humans and rats are qualitatively comparable. Additionally, compared to rats, while humans are more likely to exploit the highest rewards in any case, rats will do so more consistently under certain volatility parameters. More human participants, rat trials, and analysis are required to draw more conclusions about the similarities/differences between the two species.Within the hippocampus, place cells exhibit location-specific firing. Regions of place cells with high firing activity are referred to as place fields. Dorsal place cells have been previously explored in rats in small environments. In small mazes, dorsal place cells form one and rarely two place fields. Recently, place field activity has been studied in megaspace, a large environment about eight to nine times larger than traditional mazes. Dorsal place cells in this larger space commonly feature multiple place fields of varying sizes (one to eight subfields). Why does a single output cell in CA1 fire at different locations in megaspace? We hypothesize that CA1 subfields are ‘bound’ to each other by neural activity in the presynaptic CA3. The binding between subfields can be evidenced by partial synchrony of cells in each field. Thus, to test this hypothesis, we used the NEURON biophysical simulator to create three ‘bound’ CA1 place fields and alter the connectivity matrix to simulate differing conditions. We developed a model consisting of three groups of CA3 neurons. We used this model to test two theories in three environments (small, medium, and megaspace). The first theory relies on neurons shared between fields that propagate activity from one subfield to another, achieving multi-field activity through increasing connectivity. The second theory relies on the projections of the inhibitory neuron onto the shared neurons, multi-field emerge as a function on inhibitory decrease in activity. Our results make specific predictions about the possible mechanisms that yield multi-field properties in megaspace and may provide sufficient support for new animal experiments to further test our hypothesis.