Differences in the structure of the cerebral cortex are thought to give rise to many traits that distinguish us from other species, as well as each other. Hence, understanding how these differences arise is an important challenge. Progress will give us insight into our humanity and individuality, as well as what can go wrong to cause developmental disorders. This thesis examines the development of the outer par t of the mammalian brain, the cerebral cortex. This is a complex structure that develops predominantly in utero and during early postnatal life by self-organisation, under the influence of both genes and sensor y experience. In this thesis, I present three studies that use computational modelling to probe our understanding of genetic and activity-dependent cortical development by testing hypothesised rules of self-organisation. Based on the output of these models, I suggest specific experiments that could improve our understanding of these two aspects of cortical development.
The first and second studies examine the coordinated expression of several genes in precise spatial patterns in the telencephalon during embryogenesis in mouse. Manipulating the expression of these genes during development alters the positions of functionally specialised cortical areas in the adult. Data from loss-of-function and gain-of-function mice suggest that these genes form a regulator y network with many feedback loops. However, our understanding of this network is currently qualitative and incomplete. In this thesis, I formalise the regulatory interactions inferred from mice with genetic manipulations into computational models in order to identify constraints on the structure of the network.
Initially, I treat expression levels as Boolean, reflecting the qualitative nature of the expression data currently available. I simulate gene expression patterns created by 1.68e7 different networks potentially consistent with experimental data and show that only 0.1% of these networks reliably reproduce the wild type expression patterns. These networks all lack certain interactions and combinations of interactions, including auto-regulation and inductive loops. While there is remarkable diversity in the structure of the networks that perform reliably, an analysis of the frequency of each interaction in these networks indicates which interactions are most likely to be present in the true gene network regulating cortical area development.
While the quick simulation time of the Boolean models allows me to examine a large number of possible networks, the binary gene expression levels cannot represent the anterior and posterior shifts of expression gradients that occur in loss-of-function and gain-of-function mice. Hence, in the second study, I re-examine the networks that reliably reproduced the wild type expression patterns in the Boolean model in continuous, differential equation models that are capable of representing the shifted gradient experiments. Even with the additional constraints provided by the gene misexpression experiments, similar results are produced by a diversity of networks. This demonstrates that existing data cannot uniquely specify the network. Based on my analysis using the Boolean and differential equation models, I suggest experiments necessary to constrain the models and help identify the true structure of this regulatory network.
The third study examines the development of patterns of receptive fields in the primary visual cortex, a process critically dependent on visual experience. In many mammals, the primary visual cortex contains maps of preferred features of visual input, including the eye through which the stimulus is viewed and the orientation of the stimulus. These maps are characterised by particular geometric relationships to each other. Manipulations of visual experience during development can be used to probe the rules governing map formation by assessing the effect of manipulated experience on these relationships. In the final part of this thesis, I use an established computational model of map formation, based on dimension-reduction principles, to predict the effect on map relationships of over-representing a single orientation in one eye and the orthogonal orientation in the other eye. Since orientation preference and ocular dominance are tightly coupled under this rearing regime, one might expect orientation and ocular dominance contours to lose their normally orthogonal relationship and instead run parallel to each other. However, surprisingly, the model predicts that orthogonal intersection is sometimes preserved in this case. The model also predicts changes to the geometric relationships between maps that have been preserved in other abnormal rearing regimes. These predictions provide a way to further test the adequacy of dimension-reduction principles for explaining map structure under perturbed as well as normal rearing conditions, and thus allow us to deepen our understanding of the effect of visual experience on visual cortical development.
Overall, this thesis demonstrates the utility of computational modelling for testing existing hypotheses about cortical development and generating new ones. It shows that, as for other self-organising systems, simple rules can generate behaviours that are not apparent using qualitative reasoning. I propose that a complementary program of modelling and experiments is the most efficient route to understanding cortical development, and intend this thesis to form part of that effort.