UC Irvine

Learning unfolds through a complex interplay of behavioral actions, cellular associations, and molecular events. At the behavioral level, learning is often facilitated by associations that develop over time in response to observed outcomes. However, understanding learning at the cellular level presents more challenges. We are aware of molecular outcomes, including changes in synaptic weights of cells that accompany behavioral associations. These changes lead to the development of ensembles, formed through principles such as attractor dynamics, which represent the learned information. Additionally, it is understood that these ensembles can be accessed and categorized as similar or different from past experiences via pattern completion and pattern separation. Yet, research in this area is highly individualized when focusing on neuronal trajectory traces or ensemble activity related to learned stimuli or behaviors. This subjective approach is logical, considering that each individual encodes an abstract stimulus uniquely, due to the vast number of individual differences such as neuronal count, connectivity, and allocation. Nonetheless, if a consistent objective exists and a shared optimal behavioral strategy is employed across individuals, then a communal computational approach should theoretically also be utilized. To explore this hypothesis, I develop novel decision-making paradigms in Chapter 1. I explore the cellular mechanisms of learning by using a behavioral paradigm in which freely-moving mice must decide between two choices based on visual stimuli without dependence on localized cues, providing a controlled experimental setting. In Chapter 2, I analyze the neural activity of mice during this behavior by first testing the decoding ability of neurons to accurately predict task relevant information before and after learning. After confirming the presence of encoded learning effects, I cluster neurons based on average trial activity across many mice. By then specifically observing how the same cells behave under two competing choice conditions with learning, we find that they deviate or spread in opposing directions by a factor that is shared across mice. This result is indicative of generalizable neuronal temporal dynamics for decision-making and supports the plausibility of a universal decoder for choice detection across individuals. This investigation reveals that even as mice learn to make decisions, the underlying comparative processes bear similarities, suggesting a shared element in pattern differentiation despite the known uniqueness of each compared ensemble. Despite knowing that these ensembles form with learning and have shared features for decision-making between individuals, we still do not know how dynamic or stable these ensembles are. Whereas other studies have attempted to explore ensemble stability, most are done within same day sessions. More recent studies have analyzed ensemble stability across days, but in most cases the ensembles were determined by methods such as functional connectivity. Therefore, it could only be theorized that any of the observed stable ensembles may hold prior information, but it could not be tested. These studies were also done using spontaneous or visually evoked cellular activity, but not in the scope of a thorough learning paradigm. Therefore, I shift focus in Chapter 3 to the dynamics of neuronal populations during learning. Notably, we observe significant turnover in active cells, with the greatest changes occurring as mice progress from naive to expert levels. This relative turnover rate diminishes between expert-to-expert sessions, hinting at the emergence of a stable ensemble through learning. By testing the decoding ability of learned information across various groups, we found that using only the neurons present in both the previous and current sessions yielded significantly better results in expert-to-expert sessions compared to strictly using the newly active cells that were not active in the previous session. This provides further evidence for the stabilization of neuronal ensembles with learning.