Using computational modelling to characterise and plan treatments for schizophrenia

Schizophrenia is a mental disorder that involves symptoms such as hallucinations, delusions, and cognitive and motivational impairments. It affects a large number of people - more than 500,000 in the UK alone - and can cause lifelong disability.

We currently only have one family of drugs that can treat psychotic symptoms (delusions and hallucinations). These were discovered in the 1950s, and they block a receptor found in nerve cells in the brain: the dopamine 2 receptor. They help two thirds of people suffering from psychosis, but the remainder have ongoing symptoms. Despite intensive research, we have not found any alternative treatments for psychosis, and no treatment at all for cognitive impairment.

Lots of evidence indicates that dysfunction in another receptor also makes a major contribution to psychosis: the NMDA receptor. The NMDA receptor exists all over the brain, on various kinds of cells - some excitatory (E cells), some inhibitory (I cells). Slightly different subtypes of the NMDA receptor exist on these different cells, and in different parts of the brain.

Different strands of evidence from psychosis research indicate that the primary problem in schizophrenia might be NMDA receptor dysfunction on either E or I cells specifically. There might be different groups of patients with E or I cell dysfunction, for example. Or I cells might try to compensate for NMDA receptor dysfunction on E cells, by reducing their activity.

Pharmaceutical companies have produced many drugs that can act at NMDA receptors, and despite some promising early results, all these drugs have failed large scale clinical trials. One likely reason is that not all patients with the same diagnosis have the same underlying causes of those symptoms. For example, there are many causes of breathlessness (asthma, pneumonia, heart failure, lung fibrosis, etc), and all require different treatments.

There may well be numerous underlying causes of psychosis symptoms: for example, E cell problems in some, I cell problems in others. These groups would need treatments specifically targeted to E or I cells respectively.

The aim of this project is to find ways of identifying these subgroups with 'low E' or 'low I' function, and at what stage they might best be targeted. I will do this using a variety of methods, but all are based around the use of simple tasks (e.g. listening to tones) whilst electrical signals in the brain are recorded using electroencephalography (EEG). I will use computational models to estimate E and I cell function from these recorded EEG signals.

I will study datasets of participants with schizophrenia or psychosis at different stages of their illness and also a very large dataset of young people at risk of developing psychosis. I will also study mice who will undergo the same simple tasks, and who will be given low doses of drugs that cause either 'low E' or 'low I' states. The purpose of all these experiments is to establish i) which cells (E or I) would be the best to target with an existing NMDA receptor-based drug, and ii) when (at what illness stage) that drug should be given.

In Part 2 of the project, I will test these predictions in a sample of 100 people with early schizophrenia. I will use computational models of their brain signals to estimate whether they have 'low E' and/or 'low I' pathology. I will then give them a drug boosting E cell function (for example) for 3 months, and measure its effects on symptoms and cognition. The key question is: can my model estimate of E or I function predict the effects of the drug? If so, this could mean we could use these models to assign treatments for psychosis, and test this approach in a large scale clinical trial.