The topographies of these PCs show only a rough correspondence with the outlines

of the FFA and PPA. For example, the first PC, whose tuning profile showed positive responses only for human faces, Selleckchem Stem Cell Compound Library has positive weights only in small subregions of the FFA. The fifth PC, whose tuning profile showed positive responses to both human and nonhuman animal faces, has positive weights in most, but not all, of the FFA, including the same subregions that had positive weights for the first PC, as well as in more posterior VT regions outside of the FFA. The second PC, which was associated with stronger responses to objects—especially houses—than faces, has only negative weights in the FFA and only positive weights in the PPA, but the topography of positive responses extends into a much larger region of medial VT cortex. By contrast, the third PC,

which also was associated with stronger responses to objects than faces but with a preference for small objects over houses, has a mixture Selleckchem PR-171 of positive and negative weights in both the FFA and PPA, with stronger positive weights in cortex between these regions and in the inferior temporal gyrus. Overall, these results show that the PCA-defined dimensions capture a functional topography in VT cortex that has more complexity and a finer spatial scale than that defined by large category-selective regions such as the FFA and PPA. The topographies for the PCs in the common model that best capture the variance in responses to the movie, a complex natural stimulus, did not correspond well with the category-selective Liothyronine Sodium regions, the FFA and PPA, that are identified based on responses to still images of a limited variety of stimuli. We next asked whether the category selectivity that defines these regions is preserved in the 35-dimensional

representational space of our model. First, we defined a dimension in the model space based on a linear discriminant that contrasts the mean response vector to faces and the mean response vector to houses and objects. The mean response vectors were based on group data in the face and object perception experiment. We then plotted the voxel weights for this dimension in the native anatomical spaces for individual subjects (Figure 6A; Figure S1F). Unlike the topographies for principal components, the voxel weights for this faces-versus-objects dimension have a topography that corresponds well with the boundaries of individually defined FFAs. Thus, when the response-tuning profiles are modeled with this single dimension, the face selectivity of FFA voxels is evident, but this dimension does not capture the fine-scale topography in the FFA that is the basis for decoding finer distinctions among faces or among nonface objects. By contrast, the dimensions in the common model do capture these distinctions.