Brain networks in epilepsy: Endophenotypes and generative models

Epilepsy is the commonest serious brain disease, affecting 750,000 people in the UK. It causes catastrophic seizures, resulting in at least 1000 deaths per year in the UK. It is the commonest cause of repeated admission to hospital. About 1 in 3 of all people with epilepsy do not respond to any medication, and continue to have uncontrolled seizures. The personal, social and economic costs of epilepsy are very high.

It is not yet known how epileptic seizures start. There are clues from experiments in animals and genetic studies in people which suggest that epilepsy can be caused by a wide range of disturbances in brain cell function. Furthermore, there are different patterns of epileptic seizure, which occur in different combinations in different people, and associated with other specific findings in clinical tests. This has given rise to the view that epilepsy is at least 20 different diseases (or types of epilepsy), and that each type of epilepsy might have a different cause.

However, this does not explain some well-recognised aspects of epilepsy:

Firstly, although there may appear to be many causes, there are only small number of types of epileptic seizure; and an individual with epilepsy may suffer from several different types of seizure.
Secondly, epilepsy may sometimes be inherited within a family; if epilepsy were a set of entirely different diseases, it might be expected that within one family all affected members would have the same type of epilepsy. On the contrary, it is known that within some families there are affected individuals who have completely different types of epilepsy.
Thirdly, drugs for the treatment of epilepsy have broadly similar therapeutic effects across all types of seizure and epilepsy, with very few exceptions.

These observations suggest that there must be important mechanisms in common between different seizure types, and between different types of epilepsy.

It is increasingly recognised that functions of the brain are properties of complex networks of brain cells. We have worked for several years on new methods to identify whether brain networks in people with epilepsy have abnormal properties, and whether these properties allow seizures to occur. Using MRI brain scans, we have shown abnormalities of specific brain regions and their connections. Using electrical recordings of brain waves (EEG) we have shown that brain networks differ between people with epilepsy and people who do not have epilepsy; and that unaffected relatives of people with epilepsy may have similar brain networks to their affected relatives. Using computer models of how EEG activity is produced, we have shown that these inherited networks have an abnormal ability to suddenly switch to seizure activity.

In this Programme we build on this background, to examine four questions:
(1) Are brain networks which have an abnormal ability to generate seizures found across a range of common types of epilepsy?
(2) Are these network abnormalities inherited?
(3) Does successful treatment act by altering the properties of these brain networks?
(4) In people whose epilepsy spontaneously goes away, is this accompanied by a change in the properties of these brain networks?

We will collect MRI brain scans and EEGs from more than 150 people with epilepsy, in addition to their close family relatives, and a "control" group of people who have no personal or family connection with epilepsy. We will develop new methods to identify the relevant brain networks from MRI and EEG, and examine how these networks differ between people with epilepsy, their relatives and controls. We will also examine how these networks change over time (and how quickly they change) in response to starting treatment, and as a result of spontaneous remission of epilepsy. A special focus of this Programme is the development of new computer-based methods to identify the mechanisms by which seizures arise in these highly complex networks of the brain.

We aim to explain the emergence of epileptic seizures from the interplay between dynamic properties of localised brain regions and network structures connecting them, using a combination of methods from connectomics and dynamic modelling.

We will develop MRI methods to describe distributed abnormalities of grey matter, structural connectivity and functional connectivity in the human brain, using T1-weighted MRI, diffusion tensor imaging and resting-state functional MRI. We will develop EEG methods to describe brain networks. From these data, we will identify candidate brain networks which generate seizures.

We will employ our successful phenomenological and physiological models of EEG, which capture the patterns of onset and evolution of human ictal EEG activity. We will adopt a novel formalism to investigate brain networks: generalized dynamic graphs with active network topology. We will combine forward modelling (physiological generative models), and backwards modelling (Bayesian inversion of generative models) to determine which candidate network is most likely to explain imaging and EEG observations.

We introduce the new concept of Brain Network Ictogenicity (BNI) - the propensity of brain networks to generate seizures. We will implement and further develop our novel approach to estimating BNI in real brain networks, based on combining a dynamic model with real network structures inferred from data.

In data from human patients with epilepsy, their relatives, and normal controls, we will investigate networks which generate seizures and the BNI of these networks. In appropriate subject groups, we will examine the familial basis of brain network properties; we will explain how a familial ictogenic network endophenotype may be further modified to allow seizures to be manifest; and we will reveal the network-level effects of successful drug treatment, natural remission and epilepsy surgery.

(In accord with MRC guidance, this section emphasises impact for non-academic beneficiaries)

The new knowledge generated in the short-term by this Programme will bring benefits to a much wider group than only the academic community: a better understanding of brain network dynamics, and approaches to modelling such dynamics based on integrating data with dynamic models through the framework of generalised dynamic graphs with active network topology, will provide a novel approach to a range of brain diseases. This enhanced knowledge and new multidisciplinary methodological approach is likely to be of interest to commercial sector R&D, especially in the pharmaceutical industry and medical devices industry. For example, the ability to characterise the effects of a drug or neuromodulation device through parameter evolution of a generative model extracted from experimental or clinical data will be of benefit to commercial-sector pharmaceutical manufacturers, where they will be able to plan phase 3 trials to include data-derived prognostic biomarkers. This will be achieved using our platform to either develop their own generative models or use ours; such markers may become useful stratification markers or surrogate outcomes. Similarly, it may be possible to develop biomarkers of response to newly developed compounds in early-stage trials, based upon evolution of parameters of a suitable generative model. We have already taken steps to protect intellectual property through a patent application around the concept that changes over time in a phenomenological network model of EEG may predict clinical outcomes (joint patent application submitted by the University of Exeter and King's College London). These impacts could be felt early after the Programme completes.

The development of multidisciplinary researchers through this Programme will have an early impact which may extend for the length of the researchers' careers. The consequent enhancement of UK scientific competitiveness may be felt in academic and commercial sectors.

Science funding policy in the UK and worldwide increasingly emphasises collaboration between biomedicine and maths / engineering / computer science, but the development of such science would benefit from increasing multidisciplinary expertise and awareness amongst reviewers of funding proposals, and amongst funding panels and policy-makers. In the short term, we will seek to interact with funders to enhance the development of such multidisciplinarity.

We aim for a major impact within 5-10 years for patients with epilepsy and professionals providing healthcare in epilepsy. If successful, our framework will ultimately provide the opportunity both for a more definitive diagnosis and also more accurate prognosis for treatment response and remission. This will enable a rapid and definitive choice of treatment for patients, resulting in improvements to the health and quality of life of patients, reducing mortality and morbidity. This will be achieved through the uptake of our platform by experimental and clinical researchers, and the consequent filtering through of research findings into the clinic. For a new approach to be taken up in the clinic will require engagement with healthcare managers and policy-makers, which will be enabled through translational research following completion of the Programme.

In the short to medium term we intend an impact on public understanding of science, through public meetings we will hold throughout the duration of the programme.