Neuro-computational Study of Thymic Fluctuations in Mood Disorders

Mood disorders are common diseases that represent the first psychiatric cause of morbidity and mortality, and a major public health and medico-economic issue at the national and international level.

Although effective treatments exist, understanding of the pathophysiology of these diseases remains largely incomplete. The assessment is entirely clinical, without any reliable biological marker which has serious consequences in terms of morbidity.

Neurocomputational approaches, which consist of describing the cognitive mechanisms underlying the symptoms on which the clinic is based, so as to better articulate them with their biological substrates, are proving promising in the field.

The first mood models mathematically describe the intuitive idea that positive events have a positive impact on mood, and vice versa for negative events. More recent work has developed models to characterize the interaction of mood with decision-making. They describe how mood affects the perception of events and influences the resulting decisions: when mood is high, subjects tend to overestimate potential gains and underestimate losses, and vice versa when mood is low.

The investigators recently replicated these results and identified the neural bases of this phenomenon by showing, using a functional MRI (fMRI) study in healthy volunteers, that mood was encoded in the activity of two brain regions, the ventromedial prefrontal cortex (vmPFC) and the anterior insula (aIns). The activity of these regions in turn modulated the participants attitude towards risky choices. The investigators recently replicated this result in an intracerebral stereo-electroencephalogram (sEEG) study.

The neurocomputational approach to mood has therefore already made it possible:

1. to clarify the reciprocal interactions between mood and decision-making and
2. to propose a first characterization of the neural bases of mood.

This double advance constitutes a first step towards a true mechanistic explanation of mood disorders.

However, these models validated in the context of experiments carried out in the laboratory on healthy volunteers suffer from two limitations. They were used (1) on a short time scale (one hour) compared to daily fluctuations which evolve over a few days or weeks (2) to account for minimal, barely perceptible fluctuations, while the fluctuations observed in the context of mood disorders are extremely intense. To demonstrate their clinical relevance, it is necessary to make a double change of scale, in terms of time and amplitude.

With the aim of bridging this explanatory gap, this study aims to make this change in scale, by including patients and using a mobile application to conduct long follow-up (12 months). Indeed, the use of mobile applications has emerged in recent years as a promising approach, used in several studies with encouraging results, to collect data in ecological conditions (longitudinal, multidimensional monitoring, large samples).

To this end, the investigators will recruit 588 subjects, in 4 groups :

• 96 patients with depressive disorder (DD)
• 96 patients with bipolar disorder (BD)
• 96 healthy controls, as a reference group for comparison
• 300 non selected subjects from the general population.

After receiving written and oral information on the research, how it will be carried out and what their participation will entail, participants can be enrolled in the protocol if they have no exclusion criteria.

At inclusion, socio-demographic, clinical and therapeutic data will be collected, and updated at each visit.

An initial visit with a cognitive assessment will be carried out for patients and controls. For those who accept it, and in whom MRI is not contraindicated, this assessment will be performed during a functional MRI of the brain.

Minimal mood fluctuations will be measured using the feedbacks of a reasoning task, and a decision-making task will be used to reveal their effects on decision-making. Subjective mood ratings provided during this (short-term) lab-experiment will be fitted with our computational model of mood. The investigators will thus obtain, for each participant, a computational fingerprint describing how positive and negative events are integrated into a mood signal. Thus, a first step will be to investigate to what extent this computational fingerprint of mood differs between groups of patients, and compared to healthy controls.

All patients and healthy subjects will then be monitored daily for one year, via a dedicated mobile application. All participants will be required to complete a 3-minutes evaluation on a daily basis. The investigators will collect three types of longitudinal data: subjective mood ratings, positive and negative life events (valence and intensity of the most salient event of the day) and economic choices involving sensitivity to reward, loss, risk, delay, and effort (e.g. do the participant prefers 10 euros now or 20 euros in one week?). As in short-term data, the investigators will obtain, for each participant, a computational fingerprint describing how of positive and negative event are integrated into a mood signal. The investigators predict to observe the same group differences than in short term data: a higher reciprocal effect of mood on the perception of life events in patients with bipolar disorder, and a higher asymmetry between positive and negative life events in patients with recurrent depressive disorder.

Though informative in themselves about the effect of syndromic or sub-syndromic mood fluctuations on decision-making, the investigators also aim to use this ambulatory tool to anticipate relapses in these stabilized patients. Given the frequency of mood relapse in these patients and based on our preliminary data, the investigators expect to observe an actual episode in around one third of them during this 12-month follow-up. Importantly, referring psychiatrists will not have access to the collected data during this follow-up (no risk to alter the usual care), allowing us to test to what extent any of our measures is sensitive and specific enough to predict relapses. A trivial prediction is a shift in mood rating will be observed a few days before actual relapse. However, recent results suggested that decision-making tests could be more sensitive than auto or hetero-questionnaires to detect relapse.

Intermediate visits will be carried out every 3 months, with repeated clinical scales. An end-of-study visit will take place 12 months after inclusion.

The combination of these two steps, first a lab-recorded experiment, then a 12-month follow-up, is a key aspect of this study. Critically, the investigators will compare the computational phenotypes obtained in the very same patients but fitting short-term (a few dozens of minutes) or long-term (12 months) subjective mood ratings. The investigators will investigate to what extent the computational phenotype obtained during the short-term evaluation (first step) is informative about subsequent clinical (and subsyndromic) mood fluctuations.

Finally, a group of healthy volunteers will be monitored only online via the mobile application. This group will allow us to show the validity and stability of computational phenotypes of mood and decision-making, and to calibrate population-based measures of these relevant measures.

In addition, the investigators wanted to add two optional ancillary studies. The first uses a mobile application for implicit, passive, and longitudinal mood assessments through emotion tracking. Indeed, it seems relevant to add this type of evaluation alongside explicit assessments to more accurately detect mood fluctuations.

The second study uses a mobile application that allows voice recordings. The analysis of these vocal parameters will help to characterize a specific linguistic and vocal profile within the three groups, as well as to identify specific symptoms of conditions such as depression and bipolar disorder.

These ancillary studies will be offered to both patients and the control group.