Mark Gluck - US grants

Cognitive Neuroscience of Category Learning

Abstract

With National Science Foundation support, Drs. Gluck and Poldrack and colleagues will conduct a three-year investigation to test two hypotheses about the roles of the basal ganglia and medial temporal lobe, and their interaction, in human category learning. They will test these hypotheses using a combination of functional magnetic resonance imaging (fMRI) and studies of patients with damage to either the basal ganglia (due to Parkinson's disease) or to the medial temporal lobe (due to anoxia or other causes). The first hypothesis to be tested is that the basal ganglia are particularly important for learning based on trial-by-trial feedback, whereas the medial temporal lobe is more important for observational learning in the absence of feedback. The second hypothesis is that the engagement of these two regions during category learning is modulated by the structure of the category that is being learned. For example, some categories are largely determined by single features; for example, most animals with a beak are classified as birds. Other kinds of categories require integration of information across multiple features. We will test the hypothesis that the medial temporal lobe is crucial for learning categories based on combinations of features, whereas the basal ganglia are important for learning categories based on single features.
The topic studied in this project is categorization, oneof the most important acts of human cognition. Categorization is the recognition that various individual objects can be classified into larger groups that resemble each other in some way. Research in cognitive psychology has provided a set of sophisticated models for how humans learn new categories. However, little is currently known about how these operations are achieved in the brain. Most of our current knowledge comes from studies of patients with brain disorders. In particular, patients with Parkinson's and Huntington's diseases have trouble learning some kinds of new categories, though they are able to learn other kinds of categories. These diseases affect a set of deep brain structures known as the basal ganglia, and their impairment on some category learning tasks suggests that the basal ganglia may be critical for category learning. However, the exact role of the basal ganglia is unknown. Whereas patients with Parkinson's and Huntington's diseases are severely impaired at learning some new categories, patients with amnesia following damage to the medial temporal lobe (including the hippocampus) are only subtly impaired at category learning. Initial neuroimaging studies have shown that this region is deactivated when normal individuals are learning new categories, suggesting that it is not involved. Furthermore, imaging studies have shown that that activity in the medial temporal lobe and basal ganglia during category learning is negatively related: Individuals with more activity in one region tend to have less activity in the other. These findings suggest that these two regions may interact during learning, but the nature of this interaction is unclear at present.
These studies have important implications, both for the basic understanding of category learning and for the understanding of the brain systems involved in such common disorders as Parkinson's and Alzheimer's diseases. Most importantly, the results of these studies will provide stronger constraints on theories of human category learning that are currently possible using behavioral measures alone.

The association of rewards with specific actions is a common learning mechanism observed in day to day life that lead to predictions about potential rewards and shaping of behaviors. Over the years, research has suggested that one particular brain structure, the striatum, is a central region involved in how humans learn what decisions are advantageous (e.g., lead to reward). The goal of this proposal is to explore the neural substrates of human reward learning and decision-making processes drawing on two distinct methodologies in cognitive neuroscience: functional neuroimaging and experimental neuropsychology. Functional neuroimaging in healthy individuals can show whether a brain region is normally involved in a cognitive function; neuropsychological studies of individuals with damage localized to the same brain region can show whether or not that region is also necessary for that function. With support from the National Science Foundation, Drs. Mark Gluck, Mauricio Delgado and colleagues at Rutgers University propose to integrate both methodologies to investigate the role of the human striatum during (1) learning when reinforcement is directly contingent on actions as opposed to learning from passive observation, (2) learning when there is a delay between action and reinforcement. Parallel studies will be conducted in patients with striatal dysfunction due to lack of dopaminergic input that occurs in Parkinson's disease (PD) and neuroimaging studies of striatal function in healthy subjects.

Results from this research will contribute to a deeper understanding of the role of the striatum in decision making and in reward prediction, and of how the striatum, and related brain regions, utilize prediction-error feedback. This research builds on prior studies of the role of the basal ganglia in feedback learning, but also expands into new and more complex cognitive domains, including the creation of chains of linked predictions about sequences of actions. Learning to predict future rewards and other positive outcomes is fundamental to our survival in a complex, dynamic and uncertain world; it provides essential knowledge for making decisions that critically affect our lives, and the lives of those around us. The findings could influence interdisciplinary research in economics and neuroscience concerned about the mechanisms of how humans make decisions. Further, the research will give us a deeper understanding of the non-motoric, more motivational deficits associated with PD, eliciting new ideas for future research directions. Finally, the proposal aims to increase training opportunities for undergraduate, graduate and post-doctoral trainees in both neuropsychology and functional neuroimaging.

The ability to encode emotionally laden memories and exploit them efficiently in novel situations is among the most important learning capabilities of an individual. From correct recognition of facial expressions in emotionally-charged social situations, to discrimination between threatening and safe signals in fight-or-flight scenarios, the way we build on our past emotional experiences to decipher unfamiliar conditions strongly determines our ability to survive and function adaptively. Contemporary research on sleep and memory indicates that sleep plays a key role in the generalization of past experiences to novel situations, a process involving core brain regions such as the hippocampus, amygdala and medial prefrontal cortex. However, significant inconsistencies exist between findings regarding the effect of sleep on emotional versus non-emotional memories. These inconsistencies pose a significant obstacle for reaching a comprehensive understanding of how sleep affects memory and learning in general, and how it differentially affect fear memories in particular. Understanding how sleep assists processing of emotional memories can bridge those gaps, as well as contribute broader knowledge to inform any future developments in prevention and treatment of anxiety disorders that involve both sleep and memory generalization disruptions, specifically post-traumatic stress disorder for which sleep irregularities have been emphasized as a key factor.

Current studies suggest that generalizing acquired associations of non-emotional stimuli to novel situations is influenced by a specific sleep stage termed Slow-Wave-Sleep (SWS). Processing of emotional stimuli, on the other hand, was shown to be influenced mostly by another sleep stage, Rapid-Eye-Movement (REM) sleep. These potential inconsistencies raise the question of how sleep influences processes requiring a combination of the two, namely, generalization of emotional memories. These issues, however, were only sporadically addressed in human studies, and were mainly concentrated on a narrow definition of memory generalization. The current project will attempt to elucidate the mechanism involved in the way sleep affects fear generalization and threat detection in humans, addressing the empirical and theoretical gaps by: (1) Employing long-term monitoring of sleep stages prior to, and following, fear learning, that may be more sensitive in detecting relations between cognitive mechanisms and sleep characteristics compared to studies using only single-night monitoring (2) Using novel neurocognitive assessments of fear generalization processes, focusing on discriminative learning of threat and safe stimuli, accompanied by functional neuroimaging (fMRI); (3) Studying computational network models of the key brain regions known to be involved in fear learning to simulate and interpret the experimental findings. These three approaches will allow establishing how fear generalization is affected by individual baselines of sleep patterns, whether it is differently affected by the type of stimuli generalization (discriminative versus similarity-based) required by the behavioral task, and what the level of involvement of the key brain regions is in each type of generalization task and in relation to the individual sleep patterns.