Networks of neural dynamics: Knowledge-discovery for experimental neuroscience

What is happening in your brain when you think and act? Cells are firing tiny electrical pulses, little spikes of activity, all across the brain. Some groups of cells emit these spikes at the same time, all of them responding to sudden noise, or to the swinging of your arm. In other groups, the spikes occur in a fixed sequence across the cells, remembering the path you just took from the front door to the bus-stop. Fundamentally, the brain works by co-ordinating activity between its cells.

So when cells stop being precisely co-ordinated, the brain stops working properly. In an epileptic fit, the cells across the cortex all become synchronised and waves of activity drown out the fine control of the muscles. In dementia, the loss of synchronisation between cells prevents reliable recall of past events.

The goal of my research is to enable us to find and analyse the co-ordinated activity of brain cells. Neuroscientists are now able to record the spikes from hundreds of separate cells, for hours at a time, from all across the brain. Yet the resulting data mountain is growing without the ability to analyse those recordings. We have many methods for comparing the activity of two cells, but few for comparing the activity of hundreds. We have even fewer methods for finding when in each recording the co-ordination happens, or for finding which cells are taking part, or for finding if the co-ordination is made up of simultaneous spikes, a sequence of spikes, or something more complex. Without these methods, these recordings cannot reveal what co-ordinated activity of individual cells tells us about how the brain functions and dysfunctions.

I will develop analysis methods that are able to take the recordings and automatically solve all these problems: finding when the cells are active together, which groups they belong to, and what form that co-ordinated activity takes. I will apply these methods to three areas of neuroscience research that seek to study the brain in health and disease by recording many cells at the same time. First, with Dr Constance Hammond's lab in Marseille, we will analyse their recordings of the developing rat striatum, a large forebrain system that is central to both the control and learning of actions. We will use my methods to understand how the co-ordinated activity in the healthy striatum develops over pregnancy and infancy, and then understand how genetic and environmental factors disrupt this correct development, leading to disorders of the striatum that appear in youth, like Tourette's syndrome.

Second, with Dr Sid Wiener's lab in Paris, we will analyse their recordings from the forebrains of rats learning to solve spatial navigation tasks in mazes. We will use my methods to understand how co-ordinated activity across the forebrain develops during learning. Particularly we will analyse how the sudden onset of widespread co-ordination that precedes correct decisions on the task depends on dopamine, and how replays of co-ordinated activity during sleep lead to improved performance. From the first we can gain a better understanding of how abnormal dopamine in the forebrain, as in schizophrenics, disrupts working memory and decision-making; from the second we can gain a better understanding of how poor quality sleep can affect learning.

Third, with Dr Rasmus Petersen's lab in Manchester, we will analyse their recordings from cells in the centre of the rat's brain that fire in response to movements of their whiskers. Dr Petersen's lab study these cells to understand the basic "neural code", the information that is carried by each spike. They have already found that some cells emit spikes in response to single features of movement, such as the whisker's position or velocity, whereas other cells emit spikes only to a complex mix of these features. We will use my methods to understand how these single cell codes combine when co-ordinated, forming the "population code" for sensory information.

The delicate balance of co-ordinated neuron activity within and between brain regions is essential for normal brain function. Disrupting this co-ordination leads to dysfunction, like the excessive synchrony underlying epileptic seizures, or the loss of synchrony underlying loss of recall in dementia. Consequently, our best understanding of both neural function and dysfunction will be achieved when we can record, control, and analyse system-wide neural activity resolved to the level of co-ordination between each neuron.

Large-scale recording technology and system-wide control via optogenetics are becoming routine - but we lack system-wide analyses. I will develop data-mining methods that are able to solve the problems of finding when cells are co-active, which groups they belong to, and what form that co-ordinated activity takes, whether it be simultaneous, in sequences, or something more complex. I will develop a methodological framework based on network theory, modelling the interactions between neurons as links in a network. We will create network-dividing algorithms that automatically determine the size and number of groups of interacting neurons. These algorithms will be capable of data-mining the direction and type of interaction through analysing their delay, sign, and causality. Establishing this systems approach to neurophysiology will enable knowledge-discovery of novel hypotheses for neural coding, neural computation, and their disruption.

These algorithms will be used for collaborative analysis of large-scale neural activity data, to discover: the key factors in the healthy development of the striatum (with C. Hammond, INMED, Marseille); how dynamic re-organisation of neuron populations across the prefrontal cortex-hippocampal system underpins learning and decision-making (with S. Wiener, College de France, Paris); and how sensory information is coded by neuron populations in the whisker-related thalamus (with R. Petersen, University of Manchester)

Our work aims to bring systems biology to neurophysiology, by establishing the data-mining tools necessary to enable knowledge discovery of how neurons co-ordinate their activity from large-scale data-sets of recordings. A critical pathway to the public health and economic impact of this work will be through the intermediate step of maximising take-up of the developed methodology across experimental neuroscience. We anticipate that this impact will be on a relatively short time-scale, realisable within the course of the Fellowship.

The potential beneficiaries are all groups that use large-scale recording to understand neural dynamics in both the normal function and dysfunction of neural systems. This encompasses groups using tetrodes, silicon probes, calcium-imaging, and voltage dyes to study, amongst many possible examples, neural coding in the retina and somatosensory systems; motor commands from motor cortex; dynamical systems in the hippocampal formation; behavioural sequencing and learning in the basal ganglia; and working memory in prefrontal cortex. Such groups would directly benefit from our development of a well-grounded methodological approach to data-mining hypotheses for neural computation and coding from their data, and an easily usable software package to pursue that data-mining.

Collaborations within the project will seek to demonstrate how the take-up of our methodology will, in practice, advance knowledge of the dynamics of neural systems. Moreover, we will seek to demonstrate the potential for impact on translational neuroscience through engaging clinical neurologists. In order to maximise our impact, we will focus on our collaboration with the lab of C. Hammond (Marseille) on the development of the striatum and midbrain dopaminergic systems, as this has currently the clearest pathway to potential impact on public health. This work will elucidate the physical and functional maturation of the striatal and dopaminergic systems from embryonic stages to adulthood, and the factors on which healthy maturation depends - in particular, the influence of dopamine over the maturation of the striatum. Clinical neurologists could benefit from our new models for neurodevelopmental disorders of the striatum and midbrain dopaminergic systems, which link developmental insults to abnormal functional development and suggest new targets for intervention and prevention.

We believe the work has two potential impacts on the UK economy. First, in their joint report "Systems Biology: a vision for engineering and medicine" the Academy of Medical Sciences and The Royal Academy of Engineering recommended that establishing systems biology capacity in the UK was critical to its competitiveness at the forefront of science, public health initiatives, and economic potential. A major benefit of this project is thus that it will establish UK capacity in systems biology within neurophysiology, offering a unique opportunity to establish a world-leading research group in a nascent field on the verge of explosion. Within the lifetime of the project, the immediate beneficiaries will be the research staff employed on the project, who will receive extensive cross-disciplinary experience, and the potential to establish their own groups in this area.

Second, we will develop GPU-enabled versions of our data-mining algorithms, which will allow both real-time and exhaustive, high-throughput analyses of neural activity data. As the development and optimisation of GPU-enabled versions is specialised, technically-demanding work, this has the potential for wealth creation via a spin-out company, which will handle the programming, distribution and support of a commercial software package. This package would benefit any experimentalist wishing to achieve either or both of high-throughput and real-time analyses of their data.