Atypical Generalization of Learning in Adolescents with Autism Spectrum Disorders: An fMRI Study of Transitive Inference
Individuals with autism spectrum disorder (ASD) exhibit impairments in generalizing learning, which profoundly affect their adaptive functioning. However, little is known about the neurobiology of these deficits.
To use a new child-friendly paradigm and task-based functional magnetic resonance (fMRI) to illuminate cognitive and neural mechanisms underlying generalization impairments in ASD.
21 high functioning adolescents with ASD aged 12-18 years, and 23 gender, IQ, and age-matched adolescents with typical development (TYP) were recruited (Mean = 14.9 years). They completed a Transitive Inference (TI) task with training on a stimulus hierarchy in which A>B>C>D>E>F during rapid event-related fMRI. There were 2 Training Sessions including stimulus and feedback phases, followed by a “Big Game” with testing on generalization to novel pairs BD, BE, AF. Whole-brain univariate, region of interest, and functional connectivity analyses were performed. Based on our prior behavioral study of TI (Solomon, Frank, Smith, Ly, & Carter, 2011), we hypothesized that individuals with ASD would use a conjunctive learning approach reliant on the hippocampus; whereas TYP would use an associative learning strategy reliant on the striatum and prefrontal cortex (PFC).
Between group differences in accuracy in Training Sessions 1 and 2 were examined using linear mixed models. Tests of fixed effects showed a main effect of Session (F(1,45) = 42.2, p<.001); a main effect of pair (F(4, 271) = 18.7, p < .001), and a Session x pair interaction (F(4,271) = 18.5, p < .001), suggesting the both groups used associative learning by Session 2. Consistent with hypotheses, during the stimulus phase of Training, whole brain analyses using ANOVA showed that TYP demonstrated greater recruitment of left dorsolateral PFC (BA 9 [-30, 20, 40]) than ASD. Contrary to hypotheses, during the feedback phase of Training, both groups showed comparable recruitment in the caudate bilaterally [-18, -10, 31], [18, 17, 19]. During the Big Game, TYP showed greater recruitment of left posterior cingulate cortex [BA 31; -6, -16, 49]. In the ASD group, functional connectivity between the hippocampus and the caudate, and the caudate and the left dorsolateral PFC during training were significantly associated with Big Game performance [(r = .61, p = .004) and (r = .67, p = .001), respectively]. TYP showed significantly greater functional connectivity between the BA 31 seed and regions of the PFC including right BA 10 [27, 53, 19], right BA 11 [45, 47, -8], and left BA 47 [-33, 32, -5], [-42, 29, -11] and [-45, 29, -2].
While ASD engaged the hippocampus to some extent, they also showed strong relationships between indices of associative learning and task performance. TYP engaged a more extensive network reminiscent of one used for general reasoning and mature math problem solving. While effective, learning in those with ASD may be less flexible due to their failure to reliably engage prefrontal and default mode brain regions involved in attention regulation. Changing this balance offers a potential avenue for treatment.