Understanding factors driving the origins and plasticity of somatosensory representations

Somatosensory cortex contains a topographic representation of the body surface, whereby neighbouring regions on the body are represented by adjacent cortical regions. Furthermore, some regions are magnified, such that their representations are larger than would be expected by their physical size. What determines the setup and plasticity of these representations? We first investigate the possible contributions of the peripheral afferent densities and the statistics of tactile input, of which both tend to be greater in over-magnified regions. We consider that the brain has limited capacity resources---bottlenecks, which constrains representation of input regions. Building on previous work in efficient coding, we use linear second-order models to maximise information. We show that the optimal representation depends crucially on the width of the bottleneck; however, exact patterns of region over and under-representation differs depending on the combination of receptor density and activation. We test the model’s predictions using published empirical measurements of both factors and resulting cortical sizes for a highly touch-specialised organism, the star-nosed mole. We demonstrate the importance of usage statistics in determining allocations in this case. Second, we investigate a popular unsupervised self-organising model, which can produce topographic maps of inputs. We find that this model largely disregards the density of receptors and instead follows the input statistics. Furthermore, this model was unable to reproduce the inverse receptive field relationship, a key feature of cortical sensory maps, where magnified regions tend to have smaller receptive fields. Finally, we consider other contributors to somatosensory representation by modelling empirically measured effects of a short-term plasticity protocol. We found the canonical hand representation could be reproduced by implementing more divergent feedforward connectivity and explicit lateral connections. Furthermore, we suggest a role for homeostatic control of cortical activation, which was able to reproduce the global activation decrease across the hand map under digit anaesthetic. Overall, we demonstrate that somatosensory representations are dependent and differentially affected by the input statistics and density of receptors. Other factors such as network connectivity and resource capacity are also crucial for forming and maintaining representations.