The following HTML text is provided to enhance online
readability. Many aspects of typography translate only awkwardly to HTML.
Please use the page image
as the authoritative form to ensure accuracy.
Colloquium on Neuroimaging of Human Brain Function
compared with nonfaces, and also a small, but a significant, sustained response over memory delay intervals. Finally, three distinct prefrontal regions were identified that all showed greater levels of sustained activity over memory delays. Moreover, the relative contributions of perception-related activity and memory-related activity differed significantly for these three regions, suggesting they play different, functionally specialized roles in working memory. These results are a direct demonstration of memory-related sustained activity in human prefrontal cortex (see also ref. 72).
FIG. 5. Design and results of an fMRI study of working memory for faces (61). (Upper) Design of the task. For each series of fMRI scans, subjects performed 3 1/2 baseline-activation task cycles, each consisting of two sensorimotor control trials followed by two working memory trials. During the memory task, subjects saw a picture of a face, a delay, and then another picture of a face. Subjects were asked to hold an image of the first face in mind during the delay and to respond with a left or right button press to indicate whether the second face matched the first. During the control task, subjects simply looked at the scrambled pictures and then pressed both buttons when the second scrambled picture appeared. Three time series are shown that represent the different cognitive components of the task: a transient, nonselective response to visual stimuli; a transient, selective response to faces; and sustained activity during memory delays. These time series (smoothed and delayed by convolution with a model of the hemodynamic response) were used as regressors in a multiple regression analysis of the time course of activation in each area. (Lower) Results from a single subject overlaid onto that subject’s anatomical images. Activations are color-coded according to the relative sizes of the three regression coefficients described above. Areas that responded transiently and nonselectively to any visual stimulus, such as posterior occipital cortex (a), are shown in green. Areas that responded transiently and showed a selective response to faces over scrambled faces, such as fusiform gyrus (b), are shown in blue. Areas that showed sustained activation during the memory delay after the stimulus was removed from view, such as inferior frontal cortex (c), are shown in red. Areas that showed a combination of these types of responses are shown in a blend of colors. (From ref. 87.)
Our fMRI results have demonstrated the existence of areas in human cortex with response properties remarkably similar to those described in single-cell recordings in monkeys: (i) early visual areas in occipital cortex with relatively nonselective responses to complex or meaningful stimuli (5); (ii) later visual areas in temporal cortex with selective responses to meaningful stimuli, such as faces (73, 74); and (iii) areas in both temporal (75, 76) and prefrontal (25, 26, 37) cortices with sustained activity during memory delays. The degree of sustained activity in monkey prefrontal cortex during the delay period is typically greater than that observed in the inferior temporal cortex, and, unlike the activity in the temporal cortex, the prefrontal activity is not disrupted when the monkey processes other visual inputs during the delay period (76–79). These results suggest that prefrontal cells may be the major originator of the delay activity and may activate perceptual representations in posterior visual areas during the delay via feedback projections to those areas. The idea that sustained activity in posterior visual areas reflect top-down influences from prefrontal cortex is supported by the results of deactivation studies. Fuster et al. (80) have found that delay activity for object information in the inferior temporal cortex, though not eliminated, becomes markedly less selective during reversible deactivation of prefrontal cortex by cooling. Similarly, Goldman-Rakic and Chafee (81) have found that delay activity for spatial information in the posterior parietal cortex is greatly diminished during prefrontal deactivation. Although these kinds of invasive studies are not possible in the human brain, it is possible in neuroimaging experiments to reveal modulatory influences from feedback projections by mathematical modeling of the data (82, 83). Mathematical modeling clearly adds a new dimension to the inferences one can make about functional interactions among cortical areas subserving specific cognitive operations, such as those engaged in working memory tasks (e.g., see ref. 84).
We thank Robert Desimone for critical comments on the manuscript and Christine Rey-Hipolito for help in manuscript preparation.