Toward a Computationally-Informed, Personalized Treatment for Hallucinations

Overview

Auditory hallucinations are among the most distressing aspects of psychotic illness, and between 10 and 30% of people with hallucinations do not respond to antipsychotic medications. The authors have used computational modeling of behavior to link brain activity to development of auditory hallucinations in the hope of guiding new treatment development. The proposed studies take the first step toward individualized treatment approaches to hallucinations by attempting causal, pharmacological manipulation of relevant model parameters underlying these phenomena.

Auditory verbal hallucinations (AVH) are present throughout the course of psychotic illness and are among its most distressing symptoms. The presence of hallucinations alone increases risk of suicide in patients with psychosis. While antipsychotic medications often succeed in ameliorating auditory hallucinations, 10-30% of those with hallucinations exhibit little to no response to these treatments. Understanding how the processes underlying auditory perception might go awry to produce auditory hallucinations is a critical next step in the development of new treatments that are more soundly based upon systems neuroscience and brain pathophysiology. Perceptual systems do not rely entirely upon information coming from sensory organs like the retina and the cochlea. Rather, they blend this input with perceptual beliefs about the sensory environment in order to produce an internal model of that environment. The authors and others have proposed that hallucinations may be seen as an over-weighting of these perceptual beliefs when combined with sensory evidence during perceptual inference. In this work, the authors take advantage of a long history of sensory conditioning research to elicit hallucinatory experiences via traditional learning mechanisms: subjects are exposed to repeated pairings of visual and auditory stimuli and subsequently perceive the presence of the auditory stimulus when only the visual is present. The authors applied this Conditioned Hallucinations paradigm to four groups of subjects who varied orthogonally in having or not having hallucinations and psychosis. The authors found that conditioned hallucinations readily occur in all subjects but with markedly increased frequency in those who hallucinate compared to those who do not. The authors then employed a computational approach that formally models perception as a combination of prior knowledge and sensory input: the Hierarchical Gaussian Filter (HGF). Results indicate that the weight prior knowledge exerts during perception is significantly higher in those with hallucinations, and is related to prior-related functional activity specific brain regions like the anterior insula. This 'prior weighting' alteration may represent a novel, personalized, and computationally-informed target for the treatment of hallucinations. Mathematically, prior weighting is the ratio of the precision of prior knowledge to the precision of incoming sensory evidence exhibited by an individual during perception. Therefore, it may be normalized by either decreasing the precision of prior knowledge or increasing the precision of incoming sensory evidence. The precision of sensory evidence appears to depend critically upon cholinergic signaling: acetylcholine increases auditory discrimination abilities and biases perceptual inference toward sensory data. Antagonism at central cholinergic receptors decreases sensory sensitivity and decreases reliance on incoming sensory evidence during perceptual inference. Consistent with this, scopolamine, a safe and reversible antagonist at the M1 cholinergic receptor used routinely for its anti-emetic effects, can both cause spontaneous hallucinations and enhance conditioned hallucinations. By contrast, increased cholinergic signaling ameliorates psychotic symptoms in schizophrenia and Alzheimer's Disease. Rivastigmine, a reversible, centrally-acting cholinesterase inhibitor, has been used study the cholinergic system and has been found to ameliorate hallucinations in some patients with schizophrenia. The authors plan to characterize the effects of cholinergic agents on the perceptual, computational, physiological, and clinical signatures of hallucinations in healthy participants and individuals with psychosis via the following aims: Aim 1: Characterize the effects of cholinergic antagonism on the behavioral, computational, and neural signatures of conditioned hallucinations in healthy subjects. Hypotheses: 1) Non-hallucinating healthy subjects will show increases in prior weighting and conditioned hallucinations with scopolamine vs. saline. 2) Scopolamine-related changes in prior weighting will be accompanied by increased prior-related activity in anterior insula on functional MRI (fMRI). Aim 2: Determine the effect of cholinergic potentiation on the behavioral, computational, and neural signatures of conditioned hallucinations in subjects with psychosis and hallucinations. Hypotheses: 1) Subjects with hallucinations and high prior weighting will show decreases in prior weighting and conditioned hallucinations with rivastigmine patch vs placebo patch. 2) Rivastigmine-related changes in prior weighting will be accompanied by lower prior-related functional activity in anterior insula. 3) Subjects with hallucinations and lower prior weighting will show none of these physostigmine-related changes. In proposing these aims, the authors apply a formalized, theoretical understanding of perceptual processing to probe the interplay between perceptual, computational, circuit-level, and neurotransmitter-level dysfunction seen in hallucinations. This approach also has the potential for an immediate clinical impact: it is the first attempt to leverage the powerful tools of computational psychiatry to identify distinct patient subgroups likely to respond to emerging cholinergically-mediated treatments for hallucinations.