Asymmetry of Anticipatory Activity in Visual Cortex Predicts the Locus of Attention and Perception

Covert attention to a particular location, in the absence of visual stimulation, modulates activity representing all locations. A, Visual cortex contains a spatiotopic map in which nearby portions of cortex represent nearby portions of space. Representations of the target locations in the experimental task, as well as unattended locations in the lower and central fields, are displayed on a flattened representation of early visual cortex. These outlines are based on independent localizer scans. B, C, Endogenous BOLD activity after cues to covertly attend to the top left (B) or top right (C) target location (5° eccentricity) in the absence of visual stimulation. Note the swath of positive modulation corresponding to the attended location for each condition. BOLD activity in early visual cortex corresponding to the rest of the visual field is mostly negative. This creates a sharp peak of activity at the portion of visual cortex representing the attended location. D, The difference in endogenous BOLD activity between trials with leftward and rightward cues. Portions of cortex representing the attended locations show the largest differences. Cortex representing the lower visual field also shows a weak preference for attending to the contralateral hemifield. Portions of cortex representing the center of the visual field, however, show lower BOLD activity for attending to the contralateral versus ipsilateral hemifield, thus showing a preference for ipsilateral attention. White lines are approximate borders between retinotopic visual areas based on a standard atlas (Van Essen, 2002). Black lines are approximate isoeccentricity lines based on an average of six subjects from another study (Jack et al., 2006). Data maps are summed over the last two pretarget MR frames (frames 5 and 6), smoothed with a 5 mm full-width at half-maximum Gaussian kernel, averaged across subjects, and projected onto a flattened representation of posterior occipital cortex using the PALS (population-average, landmark, and surface-based) atlas (Van Essen et al., 2001).

A network of regions modulated with the covert locus of attention. Most regions displayed higher activity for contralateral attention relative to ipsilateral attention. Homologous regions in opposite hemispheres (e.g., left and right FEF) can be thought of as portions of a single priority map with opposite (leftward attention versus rightward attention) spatial preferences. Regions from the left hemisphere of a single subject are displayed above. Time courses are averaged across subjects and display only prestimulus, preparatory BOLD activity.

Activity in portions of cortex representing the left and right target locations is positively correlated across trials. A, Plot of trial-by-trial preparatory activity in the portion of visual cortex representing the right location (left hemisphere) versus preparatory activity in the portion representing the left location (right hemisphere). Each dot represents a single trial. The histograms on the x and y axes are projections of the trial data and display the distributions of preparatory activity for the left and right hemispheres, respectively. Although the signal related to the locus of spatial attention goes in opposite directions for the regions from opposite hemispheres, preparatory activity is nevertheless positively correlated across trials. Accounting for this positive correlation could dramatically improve the ability to decode the locus of spatial attention. This can be seen by noting that although there is a clear separation between the dots corresponding to the leftward and rightward cued trials, the histograms reveal that there is much less separation within either single region. B, Same as A for FEF. C, D, Histograms of preparatory activity derived by subtracting right hemisphere activity from left hemisphere activity for the same data plotted in A and B. The relative activity is clearly better at distinguishing between leftward versus rightward covert attention than activity in either left or right regions alone. All data are from a random half of the trials from a single subject (2).

Comparing activity between portions of cortex representing the two target locations dramatically improves decoding of the locus of attention compared with measuring activity for a single location. This improvement is mostly caused by removal of positive correlation. Bars above display how well the activity in each region discriminates between leftward and rightward cued trials, according to different decoding schemes. White bars indicate sampling only one of the two target locations, gray bars indicate comparing cortical activity for left and right locations on the same trial, and black bars indicate comparing activity when the trial pairings are randomly assigned. The bracket i represents the increase in the ADI caused by the removal of positively correlated activity. The bracket ii shows the increase in the ADI caused by the constructive combining of signals related to the locus of attention that go in opposite directions. In A, the subtractions (black and gray bars) are between homologous regions in opposite hemispheres. B compares subtracting homologous regions (V3A_H) versus subtracting nonhomologous regions in opposite hemispheres (V3A_N), using V3A as an example. V3A_H is identical to V3A in A. The white bar for V3A_N is the same as for V3A_H and represents the average discriminability between individual left hemisphere and right hemisphere V3A. The gray bar for V3A_N represents the average ADI for all nonhomologous opposite hemisphere subtractions involving V3A (L FEF − R V3A, L V3A − R FEF, L IPS − R V3A, etc.) with trial order preserved, whereas the black bar is the same with trials shuffled before the subtraction. Note that the improvement in the ADI caused by the removal of positively correlated noise is higher for homologous versus nonhomologous subtractions (compare H_i and N_i).

The relative activity difference between left and right V3A predicts performance in the upcoming trial. Relatively more right V3A activity predicts correct performance for leftward trials whereas relatively more left V3A activity predicts correct performance for rightward trials. The only exception across the six subjects and two target locations is leftward trials for subject 3. Mean (left V3A minus right V3A) magnitudes are plotted above separately for each subject and condition. The average performance discriminability (correct vs incorrect) based on the full distributions of these magnitudes was significantly greater than chance at 0.54, averaged across subjects and target location. Although modest, this finding was consistent across subjects (as shown above) and replicated in an independent dataset (see Results, Spatial attention signals predict perception).