Our recent studies have focused on a group of noncholinergic neurons in the basal forebrain (BF), which we refer to as BF bursting neurons. The bursting response of these BF neurons leads to faster decision speed (Avila & Lin, 2014), while their neuronal inhibition leads to rapid behavioral stopping (Mayse et al, 2015). Based on these observations and a thorough review of the literature, we propose that BF bursting neurons likely represent a group of corticopetal GABAergic neurons, and their rapid modulation of cortical response likely improves behavioral performance by enhancing the speed of decision making, therefore serving as a key mechanism for top-down attention (Raver & Lin, 2015). In the context of aging research, these studies raise the question of whether the factors that lead to the early degeneration of BF cholinergic neurons in Alzheimers disease also affect the survival and normal functioning of spatially-intermingled noncholinergic BF bursting neurons, and the possibility that GABAergic BF neurons may be a critical yet previously neglected mechanism in BF-dependent executive functions that are vulnerable in aging. Given that older adults often show problems dealing with situations where reward is uncertain and requires constant behavioral adjustments, we have also investigated the role of BF bursting neurons in reward-based associative learning in the current reporting period. While much evidence suggests that animals and humans use internal reward-prediction models to guide decision-making and associative learning, little is known about how such internal models are first established during the early phase of new learning. By tracking the temporal evolution of a reward prediction error (RPE) signal in BF bursting neurons during learning, we show that animals' internal model undergoes stepwise expansion to incorporate new reward predictors. Reward predictors were sequentially incorporated based on their temporal proximity to the reward, and this process was mirrored by the temporal backpropagation of the BF RPE signal from the time of reward to earlier epochs in discrete steps. As a result, a new stimulus that was clearly perceptible and objectively predicted reward was not represented in the internal model during the early phase of new learning to guide reward-seeking behavior, and was only incorporated into the internal model when the BF RPE signal backpropagated to that stimulus. Together, these results reveal an effective strategy for discovering novel associations in complex decision trees without the need for representing all potential events concurrently. These findings provide the foundation to understand how noncholinergic BF neurons may be involved age-related impairments in decision making and top-down attention.
