A Bayesian Account of the Sensory-Motor Interactions Underlying Symptoms of Tourette Syndrome

Abstract

Tourette syndrome is a hyperkinetic movement disorder. Characteristic features include tics, recurrent movements that are experienced as compulsive and "unwilled"; uncomfortable premonitory sensations that resolve through tic release; and often, the ability to suppress tics temporarily. We demonstrate how these symptoms and features can be understood in terms of aberrant predictive (Bayesian) processing in hierarchical neural systems, explaining specifically: why tics arise, their "unvoluntary" nature, how premonitory sensations emerge, and why tic suppression works-sometimes. In our model, premonitory sensations and tics are generated through over-precise priors for sensation and action within somatomotor regions of the striatum. Abnormally high precision of priors arises through the dysfunctional synaptic integration of cortical inputs. These priors for sensation and action are projected into primary sensory and motor areas, triggering premonitory sensations and tics, which in turn elicit prediction errors for unexpected feelings and movements. We propose experimental paradigms to validate this Bayesian account of tics. Our model integrates behavioural, neuroimaging, and computational approaches to provide mechanistic insight into the pathophysiological basis of Tourette syndrome.

Keywords: Tourette syndrome; active inference; basal ganglia; insula; motor cortex; tics.

Figure 1

Cortical and subcortical regions associated with the phenomenological features of TS. Tic genesis and expression (green) is associated with premotor cortex, SMA, insula, M1, S1, the putamen, globus pallidus, and thalamus; premonitory sensations (blue) with the SMA, insula, M1, S1, and the putamen; tic suppression (red) with the inferior frontal gyrus, preSMA, and the subthalamic nucleus.

Figure 2

Schematic of predictive coding within a hierarchical neural system (1, 2). Descending (blue) pathways convey (prior) predictions about the causes of sensory inputs, while ascending (red) pathways convey sensory (likelihood) prediction errors, which are combined at each level through generative models encoding prior and likelihood functions, to form (Bayesian) posterior expectations. The relative contributions of prediction and prediction error are weighted according to their (expected) precision, or “reliability” (see box), which is associated with the activity of neuromodulatory systems. Such neuromodulators can typically act to attenuate or amplify the effect of a pre-synaptic neuron on its post-synaptic target, hence attenuating or amplifying descending prior or ascending prediction error signals (“post-synaptic gain”).

Figure 3

Tics and premonitory sensations arise through overly precise priors for action and sensation within the putamen. In our model of TS, increased cortical signalling from the SMA, combined with reduced regulation by inhibitory interneurons in the putamen (see Figure 4), leads to release of signals for movement to M1 with enhanced precision. These precise priors (blue) generate tics, leading to ascending prediction error signals from putamen (bold red arrows) indicating production of an action that was not predicted within high-level regions, and which is “explained away” by the preSMA ascribing an “unvoluntary” feeling to the action—that it was “somewhat intended.” Similarly, increased cortical signalling from the posterior insula, combined with reduced regulation by inhibitory interneurons in the putamen, leads to signals passing to S1 with enhanced precision. These precise priors (blue) generate bodily sensations, leading to ascending prediction error signals from putamen to anterior insula (bold red arrows) for bodily feelings that were not predicted within this higher level region, which are therefore “explained away” by the anterior insula as unexpected or “untoward” bodily feelings that require mitigating action to remove. To do so, anterior insula sends signals to midline motor regions such as the SMA (dotted black arrow): augmenting the excitatory inputs from SMA to motor putamen, further ramping up the process of tic generation. Within-layer interactions (dashed red and blue arrows) represent canonical microcircuits (88) in which superficial pyramidal neurons (red triangles) compare expectations with predictions from deep pyramidal neurons (blue): discrepancies at the highest hierarchical levels are explained away as “unvoluntary” actions (tics), and untoward bodily feelings (premonitory sensations).

Figure 4

Aberrant integration of cortical inputs within the striatum in TS. Typically, glutamatergic inputs to the basal ganglia direct pathway are regulated by inhibitory interneurons within the striatum, and modulated according to dopamine release by the substantia nigra (SN). In our model of TS, increased cortical signalling from the SMA, combined with reduced regulation of the direct pathway by inhibitory interneurons, leads to increased excitation of the medium spiny striatal output neurons, which results in greater thalamic disinhibition and release of signals for movement to M1. Thus, a tic is generated, leading to ascending prediction error signals from putamen (bold red arrows) for a movement that was not predicted within preSMA, and which is “explained away” as an “unvoluntary” action.