Fig 1.
Three experiments were conducted and combined to investigate the neural basis for object color and related integrative object knowledge. (A) Examples of stimuli in the grayscale object viewing experiment (Exp 2): 6 types of color-diagnostic grayscale objects from 3 color categories (Red: strawberry, watermelon; Green: cabbage, kiwi; Yellow: banana, corn). (B) Examples of stimuli in the chromatic and achromatic grating viewing experiment (Exp 1): 3 equiluminant chromatic gratings (i.e., red, green, and yellow) and 3 achromatic gratings with 25%, 50%, and 75% luminance contrast that had equal mean luminance with chromatic gratings. (C) Examples of stimuli used in the true- and false-colored objects viewing experiment (Exp 3). Only red and green color categories (Red: strawberry, watermelon; Green: cabbage, kiwi) and their color-swapped false items were included in Exp 3. Three fMRI experiments were performed in the order of Exp 2, Exp 1, and Exp 3 to avoid interference between experiments caused by learning. fMRI, functional magnetic resonance imaging.
Fig 2.
Representations of object color memory in the color patches.
(A) Color patches (chromatic versus achromatic grating) from Exp 1 are shown on lateral views of the template inflated surface. Note that relatively higher thresholds (p < 0.01 for M1, p < 0.001 for M2, and p < 10^-6 for M3) were applied to depict the locations of color patches without confluent activation. Red solid lines indicate that color patches (p < 0.05 uncorrected) could not be presented at the thresholds set for A. For M1 and M2, the achromatic grating was adjusted to the next lower level (i.e., from 50% to 25%) to localize the color patches with weak color bias in the right hemisphere [i.e., TEpd_c (CLc), TEav_c (AVc), and TEa_c (AFc) in M1, and TEad_c (ALc) in M2] at a threshold of p < 0.05 (uncorrected). These color patches are marked with solid white lines. (B, C) The illustration and results of classification of grayscale objects with red and green memory colors: training the classifier to distinguish a half set of the red and green color-diagnostic grayscale objects and testing on the other half. Successful memory color decoding was found in V4d_c, TEO_c, and TEad_c. (D, E) The illustration and results of memory color decoding based on chromatic gratings training: training the classifier to distinguish among 3 chromatic gratings on Exp 1 and then testing on 3 categories of grayscale objects in Exp 2. Successful memory color decoding was found in V4d_c, V4v_c, TEO_c, TEpd_c, TEad_c, and TEa_c. (F, G) The illustration and results of true-false color decoding: training on true- and false-colored objects in N-1 runs and testing on the left-out run. Significant decoding of true-false color was done only in TEO_c and TEpd_c. Bars display mean values +/− SEM. Black asterisks indicate a significant difference from the chance level (0.333 in E, 0.5 in C, and G, indicated by the dashed lines); *q < 0.05, **q < 0.01, ***q < 0.001. The data underlying this figure are available in S1 Data and https://zenodo.org/records/13739051.
Fig 3.
Representations of object color memory in TP.
(A–C) The true-color versus false-color in all sessions (A), the first part of sessions (B), and the second part of sessions (C) in TP are shown in the coronal slices for each of the 3 subjects (M1 to M3) at p < 0.05, respectively. Each slice’s anterior/posterior position is indicated on the top left corner (mm relative to the interaural canal). (D) True-false color decoding accuracy when combining all the sessions in TP. (E) Classification of grayscale objects with red and green memory colors in TP. (F) Memory color decoding accuracy based on chromatic gratings training in TP. Bars display mean values +/− SEM. The dashed lines indicate chance levels (0.5 in D and E, 0.333 in F). The data underlying this figure are available in S1 Data and https://zenodo.org/records/13739051. TP, temporal pole.
Fig 4.
A schematic view of the distributed object knowledge representation in the macaque brain.
Red nodes: visual color patches that could encode the typical colors of fruits and vegetables presented as grayscale pictures. Red nodes with black characters: color patches (V4d_c, TEO_c, and TEad_c), where brain activity patterns could be transferrable among grayscale objects within the same diagnostic color category. The brain regions highlighted in bold in the table indicate areas where significant or marginally significant results were consistently observed across all 3 tested monkeys. Additionally, TEO_c could encode true-false color as well, showing stable object color memory representation across 3 analyses. Blue nodes, TP, might store integrative object shape-color information that is abstracted away from specific sensory properties, which showed stronger responses to true- than false-colored objects and could not decode memory colors of grayscale objects. Only TP but not the above-mentioned color patches showed a short-term learning effect on false-colored objects. TP, temporal pole.