DIALOG: Understanding Disorganisation: A Language-focused Global Initiative in Psychosis

Overview

Disorganized speech, language and communication, also called 'formal thought disorder,' is a key part of severe mental illnesses like psychosis and mood disorders. When someone's communication is disorganized, it makes social interactions difficult, increases stigma and affect educational and employment opportunities. However, we do not know much about why this happens. This project, called DIALOG, aims to understand the brain's role in disorganization by studying everyday language use instead of traditional clinical ratings. The study will look at how our brain creates predictions during interactions and how these processes break down in psychosis. This international project also includes experts with personal experience of mental illness. The study will look at speech, thinking patterns, symptoms, and brain waves. The goal of the study is to see if brain waves are disrupted in psychosis, especially in language-related problems. Speech tasks, like describing pictures, talking about a significant event, and telling a story are administered. These tasks will be audio-recorded for analysis. Non-invasive brain imaging technologies such as Magnetoencephalography (MEG) and Magnetic Resonance Imaging (MRI) are utilized. MRI creates images of the brain's structure, while MEG records magnetic activity from neurons, shown as brain waves. The MRI machine uses a large magnet to create images, and MEG captures small magnetic field changes from brain activity. Participants will also undergo clinical and neurocognitive assessments. The study will combine Large Language Models (LLM) applied to speech recordings with large scale participant data from neuroimaging tools (MRI/MEG). The goal of DIALOG is to pioneer a computationally informed, molecular-to systems-level account of disorganisation, identifying the precise mechanisms that can be targeted with novel treatments. This project aims to gather speech and neuroimaging data from Montreal \[100 healthy volunteers and 50 patients with psychosis\], Groningen \[17 synaptic density PET scans\], Cardiff \[600 participants\] and Marburg \[1600 participants\] with schizophrenia, schizoaffective disorder or mood disorders and user acceptability data at Pavia and Melbourne.

Disorganisation is a multidimensional, cross-diagnostic symptom in SMD, but its quality, persistence, and severity varies markedly across diagnoses. Current knowledge on the pathophysiology of disorganisation highlights context-processing deficits, language network dysconnectivity, and aberrant beta oscillations; but this knowledge is highly fragmented - based solely on unreplicated cross-sectional data and focused predominantly on schizophrenia. DIALOG will approach these interconnected processes of disorganisation within a computational framework of predictive processing using naturalistic language modelled using Large Language Models (LLMs) - the key element for social function. Imprecise predictive processing is a leading neurocomputational theory of disorganisation. Accordingly, the healthy brain implements a generative model of sensory causes (during speech comprehension) and consequences (during production) in the form of probabilistic predictions that propagate down the fronto-temporal cortical hierarchy (messages to words to phonemes). During conversations, our predictions are based on multiple cues (e.g., visual, emotional) and the meaning derived from the full set of preceding words (linguistic context). To be effective, predictions must be precise in their content and timing (i.e., reliable neural representations with low uncertainty). Theoretical models point to the inhibitory modulation of excitatory synapses as the biophysical basis of precision, while temporal dynamics (neural oscillations) appear to modulate this precision. Intact connectivity within the brain's language network (bilateral frontotemporal regions) is key for linguistic context to influence word choice. We propose that the computational failure of predictive language processing arising from synaptic, connectivity and oscillatory dynamics gives raise to disorganisation and social dysfunction. Large Language Models (LLMs), trained on vast text corpora, generate accurate probabilistic predictions from preceding context (i.e., how chatGPT works). In the brain, predictions can be realised in multiple ways and predictive processing operations per se cannot be measured in-vivo, but the probabilistic effects of context at word level and the uncertainty of the predictions provide neurally valid proxies (we pioneered this in healthy individuals). By applying LLMs to natural speech and transforming probabilistic estimates into information theory-based metrics ('contextual uncertainty' or 'perplexity' and lexical surprisal; hereafter, LLM-metrics), predictive operations in daily social life and determine their neurophysiological basis can be tracked. DIALOG will characterise the origins (i.e., the synaptic, connectivity, and temporal dynamics behind LLM-metrics) and consequences (persistent disorganisation and social dysfunction) of imprecise predictive processing and identify interventional opportunities through 5 work packages. Lived-experience experts highlight disorganisation's impact on everyday social function, but the reported correlations are confounded by other symptoms, particularly negative symptoms. Additionally, clinical detection of disorganisation across diagnoses lacks consistency. However, we believe that objective markers of the underlying neurocomputational deficit (imprecise predictive processing) can be trait-like, offering a more consistent causal link to social dysfunction. These markers can be extracted from short segments of natural language use, obtained across time in speech samples. In the predictive processing scheme, effective word selection relies on bidirectional frontotemporal information flow. In disorganised patients, our extensive prior work has uncovered structural/functional deficits in frontotemporal and other multimodal brain regions, but evidence for network dysconnectivity is inconsistent. Functional connectivity does not capture computational processes per se, but biophysical models (e.g., Dynamic Causal Model) can estimate effective connectivity i.e., directionality and the ratio of output-to-input signals for a neural population \[ synaptic gain\]. Resting-state frontotemporal synaptic gain explains cross-sectional LLM-metrics in schizophrenia. This approach can be broadened using anatomical priors based on structural connections (e.g., a larger network including frontotemporal nodes relate specifically to disorganisation in SMD). Impaired comprehension in disorganised patients, a key factor for disrupted therapeutic alliance, clinical exclusion and involuntary treatments, may result from a failure to generate precise predictions. In healthy individuals, generating precise predictions from prior linguistic context during comprehension relies on temporal neural dynamics that can be measured with regional specificity using oscillatory readouts from MEG. In schizophrenia, Singh, Palaniyappan and colleagues have linked resting and motor-related oscillation (especially alpha/beta band power, beta-burst events and connectivity) to clinical disorganisation. Neural oscillations can indicate whether a brain region (target) has engaged with therapeutic neurostimulation. By confirming the role of oscillatory deficits in disorganisation, we can instigate a novel line of clinical neurostimulation trials. DIALOG has 5 work packages (WP) focused on computational \[WP1\], experimental \[WP2-WP3-WP4\], and patient-benefit \[WP5\] aspects. Objectives and Hypotheses Our specific objectives are to answer: 1\. Does imprecise predictive processing influence disorganisation and social dysfunction? 3. How does network-level dysconnectivity affect predictive processing? 3. Do deficits in temporal dynamics of predictive processing relate to impaired comprehension? Hypotheses WP1: (1) LLM-metrics will predict disorganisation severity across SMD, consistently at two timepoints (2) LLM-metrics will explain social/occupational functioning better than predictability-independent language measures (static models) and clinical ratings. This will be tested in \>3000 subjects (pooled sample from all DIALOG sites). WP3: (1) The connectivity of predictive-processing related brain network measured using resting-state functional and structural imaging will predict the LLM-metrics over 1-year. (2) The biophysical substrates of imprecise predictive processing (e.g., functional MRI-based prefrontal synaptic gain index) will differ across diagnoses (schizophrenia, depression, bipolar disorder) (3) The synaptic gain index will correlate with subjective experience of disorganisation based on rating scale scores. This will be tested in an overall sample of n=3000 (pooled from all DIALOG sites). WP4: (1) Imprecise predictions during language production (based on LLM-metrics from naturalistic speech) will relate to aberrant context-related alpha/beta activity during an experimental sentence completion and a story comprehension task in SMD. This will be tested in an overall sample of n=750 (pooled with legacy data from all DIALOG sites).