Predictive Processes in Autistic and Neuro-typical Individuals. A behavioral, neural and developmental investigation


Environmental uncertainty can present challenges for individuals with autism. This may have consequences for navigating daily activities that may be noisy or unpredictable and rely on rapid updating of changing contingencies. Social interaction and communication, in particular, require adapting to rapidly changing perceptual, social, and linguistic demands. Recent theoretical and empirical work suggest that individuals with autism may show differences in predictive skills, but empirical findings are mixed across a variety of paradigms and samples of people with autism. These prior studies each examine one paradigm assuming that the paradigm reflects a unitary process of prediction. The present set of interlocking studies at MIT and Harvard adopt an approach that is more integrative in three ways: (1) predictive skills are assessed across wide range of perceptual, cognitive, and linguistic domains in a single group of people with or without autism. This will allow us to discern if there are or are not shared underlying mechanisms and whether any domain-specific or domain-general differences generalize across people with autism or instead are associated with a specific dimension that varies across autism; (2) all studies include behavioral and brain measures (EEG, fMRI); (3) parallel studies are conducted with children and adults to allow for a developmental analysis of findings. We believe that this is the most comprehensive, integrated, and rigorous study of predictive skills in autism, and is made possible by the new partnerships across these laboratories which have each studied separate aspects of predictive skills in autism but never in such a synergistic fashion.

Component Projects:

Behavioral & Electrophysiological Investigations of Sensorimotor Prediction during Metronomic and Probabilistic Auditory Sequences (Pawan Sinha)

Recent theoretical accounts have proposed that several phenotypic features of autism may be accounted for by underlying differences in predictive abilities (Gomot & Wicker, 2012; Pellicano & Burr, 2012; Sinha et al., 2014). Our goal is to empirically examine whether processes of temporal prediction differ in adults with autism spectrum disorder (ASD) relative to neurotypical (NT) individuals. While recording behavioral and electrophysiological measures from participants, we will employ a series of auditory entrainment and passive listening paradigms as a testbed to characterize sensorimotor prediction with probabilistic and metronomic stimuli. Entrainment involves following along (with finger tapping, for example) to an on-going sequence of sensory stimuli such as tones (e.g. Repp, 2005). NT individuals perform this task proficiently; after a few presentations, they precisely time-synchronize their taps with entrainment stimuli (Repp and Su, 2013). Entrainment requires extracting the regularity of repetition, estimating the time-interval between stimuli to predict their onsets, and planning motor movements to coincide with the stimuli. During this process, after a brief learning phase, individuals adopt a predictive stance rather than a reactive one. By systematically varying auditory stimulus sequences and recording behavioral and electrophysiological responses across multiple domains, we will determine whether there are systematic differences between NT and ASD participants. Importantly, little is known about the levels at which predictive skills may be affected in autism and how predictive processing may be associated with higher-order autism symptoms. Heterogeneity in the ASD population and assessment procedures have been a barrier to discerning consistent findings in the literature related to entrainment, temporal perception, mismatched negativity, and motor control in autism. Project I, as part of a collaborative effort, will focus on characterizing the ability to predict the temporal evolution of simple sensory sequences devoid of high-level semantics (trains of auditory tones). This is a crucial step in determining whether any observed temporal processing difficulties associated with autism are domain specific (e.g., evident only with social stimuli), or are more fundamental and discernible even with simple inputs.

The overarching aim of Project I is to explore how ASD affects temporal prediction during presentation of low-level sensory-perceptual (auditory) sequences by measuring behavioral (button press), physiological (Galvanic Skin Response, Electro-myography) and neural (Electroencephalography) responses.

Aim 1. To compare evolving behavioral and electrophysiological response to metronomic stimuli in NT and ASD adults.
Aim 2. To compare behavioral and electrophysiological response to parametric deviations from metronomic predictability in NT and ASD adults.
Aim 3. To determine the extent to which individuals with ASD, relative to NT peers, extract relationships between events to guide responses to incoming predictable and probabilistic information.

Investigations of adaptation to social and non-social stimuli (John Gabrieli)

The goal of the proposed research is test the hypothesis that individuals with autism spectrum disorder (ASD) show reduced neural adaptation across a range of social and non-social domains, to probe the mechanism underlying reduced adaptation, and to assess whether neural adaptation is associated with ASD traits. ASD is characterized by a heterogeneous grouping of symptoms across motor, cognitive, and sensory domains. Recent evidence suggests that across domains, individuals with ASD may have a reduced ability to predict events and use predictions to optimize behavior (Gomot and Wicker, 2012; Pellicano and Burr, 2012; Lawson et al., 2014; Sinha et al., 2014; Van de Cruys et al., 2014). Adaptation may be a neural mechanism that supports prediction by distinguishing repeated from novel events. Studies in children, adults, and mouse models of ASD find that adaptation and habituation are consistently reduced, and that reduced neural adaptation is associated with increased ASD symptoms (Webb et al., 2010; Schipul and Just, 2016). However, it is not known whether the hypothesized reductions in neural adaptation are broad or specific to ASD relevant domains. Further it is not known whether reduced adaptation arises from differences in bottom-up processing of patterned, repetitive information, or from differences in top-down expectation-driven mechanisms. Here, in combination with Drs. Pawan Sinha (MIT) and Jesse Snedeker (Harvard University), we will examine neural adaptation to a range of simple and complex stimuli to fully characterize prediction differences in individuals with ASD.

Aim 1: Test the hypothesis that individuals with ASD exhibit a broad reduction of neural adaptation for both social and non-social material.
Aim 2: Test the hypothesis that the mechanism of reduced neural adaptation includes top-down processes of expectancy or prediction.
Aim 3: Test the hypothesis that reduced neural adaptation is related to the dimensionality of ASD traits across people with or without an ASD diagnosis.
Aim 4: Integrate these findings with studies of the same ASD and TD adults with Pawan Sinha’s Project and ASD and TD children with Jesse Snedeker’s Project.

Developmental Studies of Prediction in Autism: Sequence Learning, Neural Adaptation and Lexical Prediction (Jesse Snedeker)

The proposed research explores three kinds of predictive processes in children with autism. Specifically, we test whether children with autism show reduced tendeny to predict simple sensory sequences (the Sinha task), reduced neural adaptation to social and non-social stimuli (the Gabrieli task), and reduced reliance on discourse context to predict upcoming words (the Snedeker Storytime Task, see below). We recognize that these three predictive processes are radically different from one another. Neural adaptation, when it is predictive, is the expectation of repetition based on prior experience. The sequence prediction task measures expectations about the temporal relations between stimuli. In both cases, the predictions are about events within seconds of each other that are learned on the time scale of minutes. The Storytime task explores how linguistic context is used to predict upcoming words when hearing a story or conversation. In contrast to the other two tasks, these linguistic predictions require the rapid construction complex higher-order representations on the basis of knowledge that has been acquired over many years. By examining all three processes in the same individuals we may gain a greater understanding of differences in prediction in persons with autism and how these processes are (or are not) linked. By examining, these three processes in children (6-10) as well as adults, we will gain a better understanding of how any developmental differences change between middle childhood and adulthood.

Aim 1: Test the hypothesis that that children with autism have a reduced tendency to make predictions about simple sensory sequences.
Aim 2: Test the hypothesis that children with autism exhibit a reduction of neural adaptation for both social stimuli (faces) and non-social stimuli (objects).
Aim 3: Test the hypothesis that that children with autism are less apt to use context to make lexical predictions during spoken language comprehension.
Aim 4: Explore the relationship between these three predictive abilities across children.
Aim 5: Integrate these findings with studies of ASD and TD adults performing parallel tasks in Project 1 and Project 2.