Frequency-specific mechanism links human brain networks for spatial attention

Abstract

Selective attention allows us to filter out irrelevant information in the environment and focus neural resources on information relevant to our current goals. Functional brain-imaging studies have identified networks of broadly distributed brain regions that are recruited during different attention processes; however, the dynamics by which these networks enable selection are not well understood. Here, we first used functional MRI to localize dorsal and ventral attention networks in human epileptic subjects undergoing seizure monitoring. We subsequently recorded cortical physiology using subdural electrocorticography during a spatial-attention task to study network dynamics. Attention networks become selectively phase-modulated at low frequencies (δ, θ) during the same task epochs in which they are recruited in functional MRI. This mechanism may alter the excitability of task-relevant regions or their effective connectivity. Furthermore, different attention processes (holding vs. shifting attention) are associated with synchrony at different frequencies, which may minimize unnecessary cross-talk between separate neuronal processes.

Fig. 1.

Posner spatial cueing task.

Fig. 2.

Phase resetting in task-relevant regions following endogenous cue. (A) ITC plots around cue onset (dotted line), over electrodes in the DAN, VAN, SMN, and DMN, in an exemplar patient (thresholded at Rayleigh test, P < 0.001, uncorrected; nonsignificant values are greyed out). (B) ITC averaged over electrodes in each functional network (across patients) in two frequency bands; δ (1–3 Hz), and θ (4–7 Hz). Shaded regions represent SEMs. (C) Spatial distribution of ITC in the δ-band during the cue and delay period across all subjects (Upper Left brain plot), and spatial distribution of ITC in the θ-band during the cue period (Upper Right brain plot), along with the four functional connectivity networks of interest in this study (Fisher z-transformed correlation values), sampled at the electrode coordinates across all subjects (Lower; see General Methods for details on generating functional connectivity network maps). Electrodes were all projected onto a single hemisphere.

Fig. 3.

Spatial specificity of phase-resetting response. (A) ITC plots at electrodes over the DAN in two patients following a contralateral cue (relative to electrode positions), ipsilateral cue, and the difference between the two conditions (thresholded at permutation test P < 0.05, uncorrected). (B) The δ-band ITC time-courses over electrodes in the DAN, visual, and DMN, averaged across all electrodes in all patients. Visual electrodes were defined as those in the occipital cortex that exibited a transient γ-power increase after the presentation of the cue or target stimuli. At the group level, the DAN and visual networks exhibited a significant effect of cue direction on δ ITC that peaked around the time of target onset (paired t test during delay, DAN: P = 0.0115, Visual: P = 0.0016). Shaded regions represent SEMs. (C) Locations of electrodes exhibiting significantly greater δ (2 Hz) ITC at target onset following contralateral versus ipsilateral cues (permutation test, P < 0.01, uncorrected).

Fig. 4.

Phase-resetting in task-relevant regions during stimulus-driven shifts of attention. (A) ITC plots around target onset (dotted line), for valid and invalid targets, over electrodes in the DAN and VAN in an exemplar patient, as well as the difference in ITC between invalid and valid trials (thresholded at permutation test P < 0.01, uncorrected). (B) ITC around target onset, averaged over electrodes in each functional network (across patients) in the θ-band, for valid trials, invalid trials, and the difference between the two conditions. Starred epochs mark time points when electrodes in a non-DMN network exhibit a significantly greater reorienting response than the DMN. Only DAN and VAN electrodes from the right hemisphere were included in this analysis, whereas both right and left hemisphere electrodes were included for the SMN and DMN. Shaded regions represent SEMs. (C) Spatial distribution of the difference in θ-band ITC following invalid vs. valid targets (averaged within the 500 ms following target onset) for both the right and left hemispheres.

Fig. 5.

δ (2 Hz) phase-locking within and between networks across task. (A) Percent change in PLV between different epochs of the task and the ITI, averaged across electrode pairs either within or between networks. Red lines represent increases in PLV from the ITI and blue lines represent decreases from the ITI. The thickness of each line is proportional to the percent PLV change. (B) Maps of PLV between a seed electrode over FEF (colored white) and all other electrodes in a single patient, across different task epochs. IPS, intraparietal sulcus; MT, middle temporal region.

Fig. 6.

Phase-clustering within and between task-relevant and -irrelevant networks. (A) Faded lines represent average phases across trials of individual electrodes, weighted by each electrode's ITC at that frequency and time point, from electrodes over the DAN, SMN, and DMN in a single patient at 2 Hz, at a time point during the delay period. Dark lines represent the interelectrode average. The angle of the resultant vector is the average phase across the group of electrodes and the magnitude relates to the degree of clustering between the electrodes. Note that the 2-Hz phases at DAN and SMN electrodes are much more tightly clustered than the phases at DMN electrodes. (B) Differences between the average DAN and average SMN phases, for the same subject and time point considered in A. (C) Phase difference between average DAN and average SMN phases, over the time course of a trial.