15 or 1 c/deg), TF (1 or 4 Hz) and mean luminance level (19 or 36 cd/m2) so that the only difference between these two types of gratings were their color contents (Lu and Roe, 2008). For each stimulus condition, a percentage change map (dR/R) was first calculated using the following CT99021 purchase formula: dR/R = (R8–16 − R1–4)/R1–4, in which R1–4 is the average of frames 1–4 and R8–16 is the average of frames 8–16. Each dR/R map was filtered (digital Butterworth
four-order filtering, low- and high-pass cutoff sizes: 34–38 μm per cycle and 1.02–1.52 mm per cycle). We used t-maps (Wang et al., 1996, Xiao et al., 2007) instead of subtraction maps to represent differential response to different stimulus features. For each pixel in a t-map, its t value is based on the pixel’s response OTX015 order to two stimulus conditions. For any pixel i, its paired t value (titi) is calculated as follows: ti=(C1i¯−C2i¯)∗N/Si in which C1i¯ and C2i¯ are the means of pixel i’s optical response (dR/R) to stimulus conditions 1 and 2. SiSi is the SD
of (Ci1−Ci2)(C1i−C2i), and N is the square root of the sample size (number of repeats). Thus, a t-map is similar to the classical subtraction map (i.e., C1i¯−C2i¯), except that each pixel is divided by its trial-to-trial variation value. A t-map takes into account the amount of variance, so pixels having large variance will have smaller t values. Figure S2 shows the comparison of t-maps and conventional subtraction maps calculated from the same data sets. Both maps reveal the same functional domains in V4. Nevertheless, the large amount of noise in blood vessel regions is better suppressed in the t-maps
than that in the subtraction maps. Polar maps (vector summation of eight conditions) were also calculated for direction and orientation selectivity. To obtain a polar map, each dR/R map was first filtered (digital Butterworth four-order filtering, low- and high-pass cutoff sizes: 170–190 μm per cycle and 1.02–1.52 mm per cycle) and compared with the dR/R map from the gray-screen blank condition to obtain a t-map. The t-maps from different directions/orientations were then vector-summed to obtain a polar map (Bosking et al., 1997). Masks for each type of domain already (orientation, color, direction) were obtained from two-tailed P-maps in paired t tests. In the P-maps, regions that consisted of pixels with p < 0.001 and peak p < 0.00001 were considered to be significant responsive regions and were included in the masks. Since, in our preparation, V4 contains around 100,000 pixels, Bonferroni correction is not practical (high risk of type II error; see Perneger 1998; Lazar, 2008). The choice of threshold P value was based on the overall similarity in sizes between the resulting domains and the domains in the original difference maps. Here, we used more stringent P levels as a correction for multiple comparisons. After this thresholding procedure, a few noise pixels (close to blood vessel or as a result of filtering) were then removed manually.