027, Fisher’s exact test), suggesting that it may not be responding solely to contours. It has previously been suggested that the PPA responds to high spatial frequencies (Rajimehr et al., 2011). We find no evidence for this. In both LPP and MPP, we found an inverse correlation between spatial this website frequency and average response magnitude that became insignificant

once we included stimulus category in the regression (LPP: p = 0.10, ANOVA, MPP: p = 0.30, ANOVA; Figures S5F and S5G). Because Rajimehr et al. (2011) based their conclusions on the PPA’s differential response to low-pass filtered images, in which sharp contours are blurred, and high-pass filtered images, in which sharp contours are accentuated, rather than by measuring the correlation between high spatial frequency content and PPA response to natural images, GSK1210151A our results do not necessarily indicate a dissociation between LPP/MPP and the PPA. Further research

will be necessary to determine whether the response of the PPA is better explained by spatial frequency or by the presence of long, straight contours. So far, we have demonstrated that cells in LPP and MPP respond selectively to scenes but are driven to some degree by long, straight contours. The role of these contours in defining spatial boundaries and the comparable fMRI response of macaque LPP to rooms with and without objects (Figures 1 and S1) raise the possibility that cells in these regions might be coding topographical layout in a pure sense: i.e., they would respond the same to all scenes with the

same spatial boundaries, regardless of other visual features. Alternatively, units might jointly encode scene content and scene boundaries. We thus sought to determine the sensitivity of unit responses in these areas to changes in boundary and content. We constructed a stimulus set comprising images with 26 different spatial layouts. For each spatial layout, we constructed a line drawing that contained only the spatial boundaries of the scene (Figure 7A). To determine whether LPP cells encoded spatial layout information invariant to scene content, we recorded from 30 units while presenting both sets of stimuli. For each cell, we computed the correlation between the mean response to each of the original layouts and the line drawings representing Metalloexopeptidase those layouts. Correlation coefficients were significantly greater than a control distribution generated by permuting layout labels (p < 10−9, t test; Figure 7B), indicating that LPP units carry some information about the spatial layout present in the stimulus independent of the content of the scene. Classification analysis confirmed this conclusion. We trained naive Bayes classifiers using the responses to four presentations of each of the 28 scene photographs and tested these classifiers on one presentation of each of the scene photographs that was not used to train the classifier along with one presentation of each of the line drawings.