Topography of multisensory convergence throughout the mouse cortex

Information about an animal's surroundings reaches the brain through multiple sensory modalities. Multimodal information is integrated at multiple stages, including subcortically, and in primary-sensory and higher-order cortical areas. A basic and functionally important feature of multisensory convergence is its topography. For example, in the superior colliculus, visual, auditory, and tactile information converge in register, producing a two-dimensional map that drives saccades. Many studies have demonstrated multimodal convergence elsewhere in the brain, but beyond the superior colliculus, the topography of this convergence is less clear. In this study, we use a model of 100um-voxel-wise connection strengths in the mouse, which is based on 428 tracer injections, to estimate the dimensions of multisensory maps in all isocortex areas. The first step in this process is to propagate 2D cortical coordinates of the voxels of six primary sensory areas through the connection model, resulting in a 12D coordinate for each voxel in the isocortex. We then use singular value decomposition to calculate the explained variance of dimension-reduced coordinates, and report for each area the dimension that accounts for 90% of the variance. Across the isocortex, these dimensions range nearly from one to six. Higher-dimensional multisensory convergence tends to occur in areas that are higher in the cortical hierarchy. These dimensions do not simply reflect the number of converging modalities, because any number of modalities can converge into a lower-dimensional map if some dimensions are either correlated with others or not topographically organized. These results have implications for modelling neural systems. In particular, they suggest a need for higher-dimensional topography in multisensory models.