The Neural Substrates of Probabilistic Reinforcement Learning In Adults with Autism Spectrum Disorders: Relationship to Behavioral Inflexibility

Background:

There has been extensive successful neurocognitive research to explain the unique pattern of strengths and challenges faced by individuals with ASDs. Our program of research seeks to extend the explanatory power of existing approaches through the validation of a new, complementary, mechanistic, and clinically relevant paradigm, which conceptualizes ASDs as learning disorders. Learning deficits are clinically meaningful as they contribute to the academic, social, and adaptive functioning challenges experienced by individuals with ASDs. Here we examine probabilistic reinforcement learning in young adults with ASDs.

Objectives:

Our objectives were to investigate the neural substrates of reinforcement learning deficits revealed in our previous behavioral study (Solomon, Smith, Frank, Ly and Carter, in press) using functional magnetic resonance imaging (fMRI), and to relate findings to symptoms of behavioral inflexibility in affected individuals.

Methods:

Participants were young adults aged 18 – 40 with ASDs (n=8) and age, IQ, and gender-matched controls (n=12). They were scanned while completing a probabilistic reinforcement learning task including three stimulus pairs with 80%, 70%, and 60% valid reinforcement contingencies. Two hypotheses were derived from a computational model of the inter-workings of prefrontal cortex (PFC), orbito-frontal cortex (OFC), and basal ganglia (BG) during the paradigm (Frank et al., 2004). These were (1) that individuals with ASDs would show less activation in fronto-striatal neural circuits required for flexible updating of motivational context, and (2) that repetitive behaviors would be associated with these impairments in fronto-striatal functioning. Hypotheses were tested using fMRI.

 

Results:

Behavioral data indicated that individuals with ASDs learned the task more slowly than neurotypical individuals, replicating our prior study. The probability of learning was assessed using state-space models for the three different trial types for each participant. Individuals’ state-space learning curves were then used as parametric modulators in the GLM for the fMRI analysis. The probability of early learning was associated with activation in the PFC, OFC, and basal ganglia in neurotypical adults, but not in those with ASDs. In individuals with ASDs, rituals/sameness behavior as assessed by a self-report inventory (RBS-R) was negatively associated with activation in the OFC (r = -.67, p = .098) and positively associated with activation in the putamen (r=.64, p =.12). These findings provide preliminary support for the hypotheses generated in the context of the computational model that ASDs involve atypical activation of neural circuits involved in reinforcement learning, and that poor top down modulation of the basal ganglia by the OFC is related to inflexible repetitive behaviors found in the disorders. 

 

Conclusions:

The study of reinforcement learning has clinically significant implications given the centrality of learning problems to ASDs, and the extensive use of learning theory-driven interventions to treat the disorders. Functional neuroimaging studies of reinforcement learning can provide both novel insights into and an important neuroscience "evidence base" for applied behavior analysis and other learning theory-based interventions that constitute the majority empirically supported treatments. Furthermore, reinforcement learning methodologies provide a new way to conceptualize aspects of behavioral inflexibility found in persons with ASDs.