Edited by: Shuhei Yamaguchi, Shimane University, Japan
Reviewed by: Hidenao Fukuyama, Kyoto University, Japan; Douglas Owen Cheyne, Hospital for Sick Children, Canada
*Correspondence: Lawrence M. Ward, Department of Psychology, University of British Columbia, 2136 West Mall, Vancouver, BC, Canada V6T 1Z4. e-mail:
This is an open-access article distributed under the terms of the
On a regional scale the brain is organized into dynamic functional networks. The activity within one of these, the default network, can be dissociated from that in other task-specific networks. All brain networks are connected structurally but evidently are only transiently connected functionally. One hypothesis as to how such transient functional coupling occurs is that network formation and dissolution is mediated by increases and decreases in oscillatory synchronization between constituent brain regions. If so, then we should be able to find transient differences in intra-network synchronization between the default network and a task-specific network. In order to investigate this hypothesis we conducted two experiments in which subjects engaged in a Sustained Attention to Response Task while having brain activity recorded via high-density electroencephalography (EEG). We found that during periods when attention was focused internally (mind wandering) there was significantly more neural phase synchronization between brain regions associated with the default network, whereas during periods when subjects were focused on performing the visual task there was significantly more neural phase synchrony within a task-specific brain network that shared some of the same brain regions. These differences in network synchrony occurred in each of theta, alpha, and gamma frequency bands. A similar pattern of differential oscillatory power changes, indicating modulation of local synchronization by attention state, was also found. These results provide further evidence that the human brain is intrinsically organized into default and task-specific brain networks, and confirm that oscillatory synchronization is a potential mechanism for functional coupling within these networks.
At a regional scale the brain is organized into functionally specific networks (Passingham et al.,
The dominant brain regions comprising the default network – the medial-frontal cortex, the temporal lobe, the hippocampal formation, and the parietal lobe (Buckner et al.,
Such a pattern of differential synchronization between default and task-specific networks should be manifested at several oscillation frequencies, particularly theta (4–8 Hz), alpha (9–12 Hz), and gamma (30–50 Hz), given the links between cognitive processes and oscillations in these frequency bands (von Stein and Sarnthein,
Alpha-band (8–14 Hz) spectral power in the occipital lobe increases when visual processing is suppressed and decreases when visual processing is enhanced, whereas gamma-band power shows the opposite behavior (Kelly et al.,
During most prolonged tasks attention waxes and wanes, alternating periods of “on-task” and “mind wandering” (Smallwood and Schooler,
In Experiment 1, 15 subjects (11 women, mean age ± SD = 21.5± 3.2) completed a SART for $20; in Experiment 2, 10 subjects (seven women, mean age ± SD = 21.5 ± 2.0) completed a similar SART for $20. One subject in Experiment 2 indicated no off-task periods and her data were discarded. All subjects were right handed with no history of neurological conditions and had normal or corrected-to-normal vision. Subjects provided written informed consent to the experimental procedure, which was approved by the UBC Clinical (Experiment 1) or Behavioral (Experiment 2) Ethics Review Board.
The SART involved presenting a serial stream of visual stimuli to subjects (Figure
In Experiment 1 subjects were asked by the experimenter, who was in the room with the subject, to identify verbally their attention state immediately prior to the experience probe as being “on-task” (fully attentive to task performance at block’s end) or “off-task” (inattentive to the task at the block’s end). The block duration randomly varied between 30 and 90 s. In Experiment 2, subjects, alone in the room, were queried by the computer as to their attention state 115 times during the experiment at random intervals (block length) averaging approximately 42 s after the response to the previous experience probe. They responded by indicating their state on a 1 (“completely off-task”) to 7 (“completely on-task”) attention state scale by pressing the appropriate computer key.
In Experiment 1, EEGs were recorded from 64 active electrodes (Bio-Semi Active2 system) distributed evenly over the head, relative to two scalp electrodes located over medial-frontal cortex (CMS/DRL), using a second order high-pass filter of 0.05 Hz, with a gain of 0.5 and digitized on-line at a sampling rate of 256 samples-per-second. The vertical EOG was recorded from an electrode inferior to the right eye, and the horizontal EOG from an electrode on the right outer canthus. In Experiment 2, EEGs were recorded from 60 passive electrodes in a standard electrode cap (Electro-cap, Inc.) at locations based on the International 10–10 System, relative to an electrode over the right mastoid with ground at AFz. Data were sampled at 500 Hz through an analog pass band of 0.01–100 Hz (SA Instrumentation, San Diego, CA, USA). The EOG was recorded from four periocular electrodes. Electrode impedance was below 10 kΩ. Prior to analysis all signals were re-referenced to an average reference, resampled to 250 Hz, and digitally high-pass filtered at 1 Hz.
EEG data were analyzed using EEGLAB software (Delorme and Makeig,
Wavelet coefficients of the sinusoidal oscillations between 5 and 70 Hz were obtained from a Morlet wavelet analysis on each IC time series. Wavelet analysis divides the total broadband IC signal into a set of frequency bands for which amplitude (power) and phase at each time point are computable from the wavelet coefficients.
Because we could not ascertain precisely how long a subject was in a particular attention state before each attention probe, we used a fixed time window of 12 s before the probe. This window has been used successfully in previous similar studies. For Experiment 1, windows preceding subjects’ indication that they were focused on the experiment were labeled “on-task,” whereas for Experiment 2 windows preceding responses five, six, and seven on the 1–7 attention state scale were labeled “on-task.” For Experiment 1, windows preceding subjects’ indication that they were off-task were labeled “off-task,” whereas for Experiment 2 windows preceding attention scale responses 1, 2, and 3 were labeled “off-task.” Windows preceding attention scale response “4” in Experiment 2 were not analyzed. No non-response stimuli occurred in the analyzed 12-s intervals in either experiment. This was by design, as we wished to analyze a stable on-task or off-task state, and a non-response stimulus occurring in this period could have changed the subject’s attention state.
Cluster analysis based on Talairach locations of dipoles associated with all valid ICs was performed to find neural sources that were common across subjects. In Experiment 1 a total of 230 ICs for the 15 subjects were separated into 13 clusters (Experiment 2: 266 ICs for nine subjects into 15 clusters) by applying the
In order to assess neural synchrony
In order to assess the functional connections
where
To obtain statistically significant differences between PLVs in the on- and off-task conditions, a method relying on surrogate distributions was utilized (Maris and Oostenveld,
Phase locking value differences were only considered to be significant if a second criterion was also met. We used the EEGLAB procedure for determining whether a group of PLVs is generally different from zero to filter the PLV values. In this procedure, individual subjects’ PLVs were masked at
We discuss only those significant PLV differences between the on-task and off-task conditions (from the first test) in which a cluster of pixels in the more significant condition also was significantly different from zero within the indicated time–frequency window, meaning that all or most of the subjects had significantly greater than zero PLV for each of those pixels. The one exception was for the default network in Experiment 2, where none of the indicated significant results passed this latter stringent test. There were a number of IC pairs for which PLV was significantly different from zero by the binomial test in both on-task and off-task conditions, including about half of those where the PLVs were significantly different between conditions by the permutation test. This was to be expected because, even when mind wandering and thus off-task, subjects were still performing the SART task at an acceptable albeit slightly reduced level. Moreover, and especially in Experiment 2, there were also a number of pairs for which the off-task PLV was different from zero and the on-task was not but the permutation test was not passed. In order not to over-interpret our data we do not discuss these cases further, although they are predominantly consistent with our conclusions.
Subjects reported being off-task on 57.5% and on-task on 42.5% of the experience probes (SE = 3.5%) in Experiment 1, and 35.5% (SE = 10.9%) and 49.4% (SE = 11.7%) respectively in Experiment 2 (15.1% discarded as neither; “4” on the 7-point response scale). Average false alarms were 6.1 (SD = 4.4) and 11.0 (SD = 4.3) in on-task and off-task epochs, respectively, in Experiment 1, and 14.8 (SD = 13.2) and 25.0 (SD = 22.7) respectively in Experiment 2. Because amount of time spent in on-task and off-task states differed across attention state, subjects, and experiments, however, we normalized false alarm errors with respect to these times. First, we assumed that the proportion of total time spent in each state was the same as the proportion of experience probes that indicated that state. Then we divided the proportion of false alarms in each state by the proportion of experience probes indicating that state, yielding a normalized false alarm measure for each attention state. We then aggregated this measure across experiments. As is common in SART (e.g., Christoff et al.,
Tables
Cluster brain region | No. of subjects involved | Total no. of ICs | BA | Centroid Talairach |
Mean RV% of dipole fit |
---|---|---|---|---|---|
Occ | 13/15 | 33 | 17 | −10, −97, −5 | 6.84 |
L ACC | 12/15 | 21 | 24 | −21, −16, 45 | 5.31 |
R MTG | 9/15 | 20 | 21 | 62, −25, −5 | 10.27 |
OFC | 13/15 | 32 | 11 | 0, 45, −26 | 7.18 |
L MTG | 10/15 | 19 | 21 | −73, −27, 1 | 8.67 |
PPC | 13/15 | 32 | 7 | 3, −62, 30 | 5.23 |
R ACC | 13/15 | 16 | 24 | 11, 8, 32 | 6.96 |
Cluster brain region | No. of subjects involved | Total no. of ICs | BA | Centroid Talairach |
Mean RV% of dipole fit |
---|---|---|---|---|---|
Occ | 9/9 | 19 | 18 | 15, −84, −7 | 4.65 |
L ACC | 5/9 | 11 | 33 | −3, 10, 19 | 8.64 |
R MTG | 9/9 | 21 | 42 | 71, −12, 3 | 4.45 |
OFC | 9/9 | 12 | 11 | 9, 59, −22 | 8.32 |
L MTG | 9/9 | 21 | 42 | −73, −22, 2 | 5.86 |
PPC | 9/9 | 20 | 2 | −41, −25, 41 | 6.4 |
R MFG | 8/9 | 13 | 8 | 55, 9, 38 | 8.64 |
Figure
For this analysis, we added two additional regions from each experiment to the five discussed above: Experiment 1: posterior parietal (Talairach 3, −62, 30) and right anterior cingulate (11, 8, 32); Experiment 2: posterior parietal (−41, −25, 41) and right middle frontal gyrus (55, 9, 38). These were not sufficiently close in Talairach space to be considered analogous across the experiments but were important in the phase locking analysis. The right middle frontal gyrus (Experiment 2) showed on-task off-task ERSP differences indicating that it might be more strongly involved in the off-task network, whereas the other three areas displayed ERSP differences more consistent with a task-related function (Figure
The pattern of network connectivity in theta, alpha, and gamma frequency bands is clearly different when PLV is greater in off-task than in on-task epochs, presumably involving default network processing in addition to task-specific processing, from that when PLV is greater for on-task epochs, presumably involving primarily task-specific processing. Notably, occipital cortex was significantly more synchronized with other brain regions
The data reported here support the idea that synchronization within default and task-specific networks in the cerebral cortex differs in ways similar to previous studies. That is, functional connectivity within these dissociable brain-regional networks varies transiently during off-task and on-task epochs. The fact that synchronization differs within these networks as a function of whether both default and task-specific processing (off-task, mind wandering) or mostly task-specific processing (on-task) is occurring, is consistent with the idea that inter-regional synchronization is a mechanism that modulates functional coupling within these networks.
Also important is our finding that blind source separation (ICA) of EEG data can be used to localize brain activity to areas within default and task-specific networks. Overall, several core locations in the default network were identified. Some discrepancy between the results in this study and previous findings is to be expected, however, since, even in fMRI studies, differences in default network locations are found across experiments, probably reflecting differences in task contexts or normalization procedures. Overall, the convergence in spatial locations between fMRI and high-density EEG with ICA provides new evidence for the validity of the methodologies used in this study. It also provides additional evidence for the reality of the default network as a robust subset of brain areas whose study is not contingent on a particular brain imaging method.
The pattern of results obtained converges well with the characterization of the activities of default network regions posited by Buckner et al. (
In contrast, synchronization results also revealed a distinct on-task network that was clearest when subjects indicated that they were paying attention to the task. During the on-task condition for Experiment 1, greater inter-regional synchronization was found in at least one frequency band in all clusters examined, with the notable exception of the right middle temporal cortex. The synchronization results were slightly weaker for Experiment 2, but indicated extensive synchronization throughout an on-task network. Importantly, in both experiments synchronization was robust between occipital, parietal, and frontal regions during on-task periods. This set of areas has been linked repeatedly to the control of goal-oriented attention.
The SART often involves potential response conflicts. Because the number of target stimuli to be responded to is much higher than the single no-response stimulus, responses often become automated. In the unlikely event that a no-response stimulus does appear, a potential conflict occurs between the subject’s habitual behavior (press the button) and the correct response (withhold the button press). This tendency likely explains the greater on-task synchrony between the anterior cingulate and frontal and parietal systems, given that the anterior cingulate has been robustly linked to conflict detection in the brain (Kerns et al.,
In both experiments there was overlap between areas within the default and on-task networks. In Experiment 1, of the seven common IC clusters found, only the occipital and right middle temporal clusters displayed significantly greater inter-regional synchronization in only one condition. In Experiment 2, only occipital cortex displayed such synchronization. The result that occipital cortex was less synchronized with other areas during off-task epochs was to be expected, given that sensory processing is attenuated during default network activity (c.f., Kam et al.,
In conclusion, our results contribute to the evolving picture of the cognitive brain as intrinsically organized into somewhat overlapping default and task-specific brain networks, and support the idea that oscillatory synchronization is a potential mechanism for functional coupling within these networks.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This research was supported by Discovery Grants from the Natural Sciences and Engineering Research Council (NSERC) of Canada to Lawrence M. Ward and to Todd C. Handy, by a NSERC predoctoral fellowship to Julia Wing Yan Kam, and by a UBC AURA grant to Lawrence M. Ward.