Abstract

We used magnetoencephalography to characterize the spatiotemporal dynamics of cortical activity during top–down control of working memory (WM). fMRI studies have previously implicated both the frontoparietal and cingulo-opercular networks in control over WM, but their respective contributions are unclear. In our task, spatial cues indicating the relevant item in a WM array occurred either before the memory array or during the maintenance period, providing a direct comparison between prospective and retrospective control of WM. We found that in both cases a frontoparietal network activated following the cue, but following retrocues this activation was transient and was succeeded by a cingulo-opercular network activation. We also characterized the time course of top–down modulation of alpha activity in visual/parietal cortex. This modulation was transient following retrocues, occurring in parallel with the frontoparietal network activation. We suggest that the frontoparietal network is responsible for top–down modulation of activity in sensory cortex during both preparatory attention and orienting within memory. In contrast, the cingulo-opercular network plays a more downstream role in cognitive control, perhaps associated with output gating of memory.

INTRODUCTION

Performance in working memory (WM) tasks is strongly modulated by selection cues that allow people to prioritize the most task-relevant item. Sperling (1960) demonstrated that cues picking out the task-relevant items given before or immediately following the presentation of a WM array (when items were still available in the iconic buffer) improved WM performance. This effect has a ready interpretation in terms of gating of encoding to a capacity-limited WM store, reducing memory load to one item. However, more surprisingly, cueing the task-relevant item during the memory retention interval, termed “retrocueing” (Sligte, Scholte, & Lamme, 2008; Nobre et al., 2004; Griffin & Nobre, 2003; Landman, Spekreijse, & Lamme, 2003), improves response accuracy almost as much. This finding challenges the assumption that WM performance is only limited by storage capacity. Retrieval mechanisms or “output gating” (Chatham, Frank, & Badre, 2014) may be just as important a determinant of performance as “input gating” of WM. Although precues are thought to permit input gating of WM, retrocues may facilitate output gating. Our study investigated the different patterns of involvement of control networks in these different forms of top–down control of WM.

One hypothesis holds that a shared top–down attentional mechanism acts both on perceptual input and on WM representations (Gazzaley & Nobre, 2012) to mediate both precue and retrocue benefits. fMRI experiments suggest that whether control is prospective and selects from perceptual input or retrospective, selecting from within WM results in the following: (1) substantially overlapping frontoparietal control regions are recruited (Nee & Jonides, 2009; Nobre et al., 2004) and (2) activity in sensory cortex is modulated retinotopically (Kuo, Stokes, Murray, & Nobre, 2014; Munneke, Belopolsky, & Theeuwes, 2012; Sligte, Scholte, & Lamme, 2009; Mangun, Hopfinger, & Buonocore, 2000). This fits well with a model of WM in which persistent activity in perceptual cortex is responsible for maintaining WM representations (Harrison & Tong, 2009; Pasternak & Greenlee, 2005): Top–down influences bias perceptual activity during WM retention in the same way as it is biased by perceptual attention. However, this model may be too simple. First, recent work suggests that not all items in WM are associated with a persistent activity state in sensory cortex (LaRocque, Lewis-Peacock, & Postle, 2014). Second, retrocues recruit additional sites in frontal cortex besides the frontoparietal network. In particular, Higo, Mars, Boorman, Buch, and Rushworth (2011) and, more recently, Nelissen, Stokes, Nobre, and Rushworth (2013) have suggested that the dorsomedial pFC and the anterior insula or frontal operculum [fO] (hereafter we use the term frontal operculum for clarity) may be the critical sites for top–down control over WM representations—as opposed to frontoparietal sites. These areas are nodes of a second “cingulo-opercular” control network, which is distinct from the frontoparietal network classically associated with top–down control (Petersen & Posner, 2012; Dosenbach et al., 2007). The additional recruitment of the cingulo-opercular network by retrocues, as compared to precues, is an intriguing clue to how its functional role may differ from and complement the frontoparietal network.

In this study, we used a precision/capacity (Zhang & Luck, 2008) visual WM task giving precues (Murray, Nobre, & Stokes, 2011; Sperling, 1960) and retrocues (Murray, Nobre, Clark, Cravo, & Stokes, 2013; Landman et al., 2003) on different trials to compare prospective and retrospective control of WM. We first reviewed prior fMRI work using a meta-analysis technique to confirm the different spatial patterns of cortical recruitment between precues and retrocues and obtain a priori spatial ROIs. We then used magnetoencephalography (MEG), an imaging modality with high temporal resolution, to characterize the pattern of cortical activity in these areas after both precues and retrocues. Our time-resolved data reveal the temporal relationship between activation in these two networks and top–down modulation of activity in sensory cortex. They show that both precues and retrocues similarly recruit the frontoparietal network, but retrocues additionally recruit the cingulo-opercular network at a later time point.

METHODS

Participants and Behavioral Task

Fifty volunteers were recruited (26 women, 24 men; mean age = 24 years, range = 19–34). All participants were healthy, had normal or corrected-to-normal vision, and were right-handed. Ethical approval was obtained from the National Health Service South Central Berkshire ethics committee (11/SC/0053). One participant performed at chance in the behavioral task, and one was unable to complete the MEG session. Forty-eight full data sets were therefore available for behavioral analysis. Further participants were excluded from the MEG analysis. Four made eye movements larger than 2° horizontal displacement during the WM retention intervals (as measured on the basis of the eye-tracker signal), potentially contaminating the MEG signal. Six participants were rejected because their MEG data were of poor quality: For two participants, the spatial coregistration of the MEG forward model with the signal space failed, and for four participants, the signal was heavily contaminated with electronic artifacts. This left us with MEG data from 38 participants. The behavioral task was programmed in Matlab and PsychToolbox (Pelli, 1997). Behavioral and MEG analyses were performed using custom-written Matlab software, SPM8, FSL, Fieldtrip, and the in-house OHBA Software Library.

We used a four-item WM task with predictive cues (precues) and retrodictive cues (retrocues). The behavioral task was run in two separate sessions: The training session contained 6 × 36 trial blocks (216 trials), and the MEG session contained 9 blocks (324 trials).

Figure 1A is a task schematic. Four memory items were presented. Following a delay period of 3310 msec, a single probe item appeared. The probe was identical to one of the items from the memory array but rotated about the circular end body, through 5°, 15°, or 45° clockwise or anticlockwise (equal probability). Participants responded with a right-hand button press; the left button (index finger) indicated an anticlockwise rotation, the right button (middle finger) clockwise. On one-third of trials, a spatial precue 1540 msec before array onset indicated the relevant memory item. On a separate one-third of trials, there was a spatial retrocue 1540 msec into the retention interval. On remaining trials (“neutral-cue trials”), no information was given about which item would be probed. Trial types were randomly interleaved. Cues were always valid. The cue consisted of a white and a black arrow formed by the sides of a square (Figure 1A). For half of the participants, the black arrow indicated the cued quadrant, and for half of participants, the white arrow indicated the cued quadrant, controlling for any effects driven by the physical properties of the cue.

Figure 1. 

Task schematic and behavioral data. (A) Precue/retrocue WM task. The three trial types are randomly interleaved, each making up one-third of trials overall. In this case, the white arrow indicated the cued direction. For half of participants, this was reversed, and the black arrow indicated the cued direction. (B) Accuracy data. The proportion of clockwise responses is plotted for each orientation change of the probe item (x-axis, clockwise positive). Both precues (blue) and retrocues (red) improve performance at all orientation change magnitudes. (C) RT distributions for each trial type, binned into quintiles. Precues and retrocues both speed responses. (D, E) Mixture model analysis. Precues and retrocues substantially reduce the proportion of trials on which participants are guessing (D). Precues substantially increase the precision with which items are represented in memory but retrocues have only a modest effect on precision (E). (F) Misgating analysis. If there were no effect of nontarget items on responses, this plot would show a flat line. Nontarget items influenced responding most strongly in the neutral-cue condition. Error bars represent ±1 SEM.

Figure 1. 

Task schematic and behavioral data. (A) Precue/retrocue WM task. The three trial types are randomly interleaved, each making up one-third of trials overall. In this case, the white arrow indicated the cued direction. For half of participants, this was reversed, and the black arrow indicated the cued direction. (B) Accuracy data. The proportion of clockwise responses is plotted for each orientation change of the probe item (x-axis, clockwise positive). Both precues (blue) and retrocues (red) improve performance at all orientation change magnitudes. (C) RT distributions for each trial type, binned into quintiles. Precues and retrocues both speed responses. (D, E) Mixture model analysis. Precues and retrocues substantially reduce the proportion of trials on which participants are guessing (D). Precues substantially increase the precision with which items are represented in memory but retrocues have only a modest effect on precision (E). (F) Misgating analysis. If there were no effect of nontarget items on responses, this plot would show a flat line. Nontarget items influenced responding most strongly in the neutral-cue condition. Error bars represent ±1 SEM.

Participants were asked to maintain central fixation, and eye movements were monitored using an infrared binocular eyetracker (Eyelink 1000, SR Research, Ottowa Canada). During the training session, stimuli were presented on an LCD screen (Samsung 2233Rz, Samsung, Seoul) at a viewing distance of 80 cm. During the MEG session, stimuli were back-projected (Panasonic PT D7700E, Panasonic, Osaka Japan) onto a screen at a viewing distance of 85 cm. In both sessions, the WM stimuli were presented at an eccentricity 6° visual angle from the fixation cross, and each stimulus subtended 1.2°.

To reduce noise from trials in which the participants were settling into the task, the first 50 trials were discarded from the training block, as well as the first 10 trials from the MEG session (including these trials does not substantively alter any aspect of the results). Any trials in which the RT was shorter than 0.1 sec or longer than 10 sec (the median RT was 1.1 sec) were discarded from the analysis, as they likely represented anticipation responses or a lapse in task engagement, respectively. For most participants (40 of 48), five or fewer trials were excluded on the basis of RT. The maximum number of trials rejected was 20.

A mixture model was fitted for the accuracy data (Zhang & Luck, 2008) in which responses across trials are assumed to come from one of two distributions: a uniform ‘guess’ distribution, representing trials in which participants had no information about the probed item and responded at random, and a Von Mises distribution (circular analogue of a Gaussian) that represents the fidelity of the WM representation on trials when participants were not guessing. We checked for response bias toward “clockwise” or “anticlockwise” responses by taking the difference of net clockwise and anticlockwise responses across all trials for each cue condition. For neutral trials, there was a tendency for participants to respond “anticlockwise” (p(clockwise) − p(anticlockwise) = −0.12, p = .001). This was also significant for precue trials (−0.077, p = .042) and present as a trend for retrocue trials (−0.066, p = .08). This suggested that, when guessing, participants were slightly more likely to press the index-finger than middle-finger button. Clockwise/anticlockwise orientation changes were collapsed together for modeling, factoring out the response bias. The probability that a participant made the correct response on a given trial depends on k, the precision of the Von Mises distribution (higher values indicate a tighter distribution); pGuess, the probability that participants are guessing on any particular trial; and θ, the size of the orientation change. This is captured in Equation 1.
formula
vonmisescdf is the cumulative density function of the Von Mises distribution, which can be evaluated numerically. The model was fitted for each participant and each condition separately using a maximum likelihood approach to get values of pGuess and k (Figure 1D, E).

The mixture model analysis assumes that nontarget items have no effect on responding, but previous work using similar multi-item WM tasks suggests that nontarget items can sometimes influence behavior (Bays, Gorgoraptis, Wee, Marshall, & Husain, 2011; Bays, Catalao, & Husain, 2009). In the current task, the orientation of each memory item was assigned randomly and independently of the orientation of the other items. The effect of nontarget items could therefore be evaluated by running a similar analysis to that for the target item, examining the probability of responding “clockwise” or “anticlockwise,” depending on whether the probe item was rotated clockwise or anticlockwise of each nontarget item. Because the target and nontarget orientations were randomly and independently assigned, the orientation difference between the probe item and a given nontarget item spanned the full circle between −180° and 180°. This range was divided into eight bins: clockwise/anticlockwise 0°–45°, 45°–90°, 90°–135°, and 135°–180°, and for each bin the probability of responding clockwise was calculated. Note that, for any given bin, the probability of the probe item having been clockwise or anticlockwise of the target item was always the same (.5), so if there were no influence of nontarget items, the probability of responding “clockwise” should be equal across all bins.

Meta-analysis of fMRI Data/Derivation of MEG ROIs

We performed meta-analyses of spatial precueing and retrocueing fMRI studies to guide analysis of MEG data using the freely available GingerAle software package (Eickhoff et al., 2009).

Separate meta-analyses were conducted for precue and retrocue experiments. Imaging studies were searched using Pubmed and Google Scholar. Retrocue studies were included if they involved a late cue (>1 sec after the memory array) to focus on an item already in memory, and we used data from contrasts that isolated the brain response to this cue. Precue studies were restricted to studies cueing spatial attention to isolate activations associated with spatial orienting. We included only those studies that could differentiate responses to the cue from subsequent responses to the cued targets. Seventeen studies were included in the retrocue meta-analysis, and eight studies were included in the precue meta-analysis. These studies are listed in Tables 1 and 2. Local maxima were extracted from the resulting meta-analysis activation maps and are listed in Tables 3 and 4.

Table 1. 

Studies Included in the Retrocue Meta-analysis

StudyStudy No.No. of Participants
Rowe, Toni, Josephs, Frackowiak, and Passingham (2000
Rowe and Passingham (2001
Raye, Johnson, Mitchell, Reeder, and Greene (200212 
Johnson, Raye, Mitchell, Greene, and Anderson (2003) 14 
Nobre et al. (200410 
Johnson, Mitchell, Raye, and Greene (2004) 14 
Lepsien, Griffin, Devlin, and Nobre (200510 
Johnson et al. (2005) 14 
Lepsien and Nobre (200614 
Yeh, Kuo, and Liu (200710 10 
Johnson, Mitchell, Raye, D'Esposito, and Johnson (200711 15 
Yi, Turk-Browne, Chun, and Johnson (200812 
Raye, Mitchell, Reeder, Greene, and Johnson (200813 29 
Johnson and Johnson (200914 14 
Nee and Jonides (200915 18 
Roth, Johnson, Raye, and Constable (200916 22 
Higo et al. (201117 21 
StudyStudy No.No. of Participants
Rowe, Toni, Josephs, Frackowiak, and Passingham (2000
Rowe and Passingham (2001
Raye, Johnson, Mitchell, Reeder, and Greene (200212 
Johnson, Raye, Mitchell, Greene, and Anderson (2003) 14 
Nobre et al. (200410 
Johnson, Mitchell, Raye, and Greene (2004) 14 
Lepsien, Griffin, Devlin, and Nobre (200510 
Johnson et al. (2005) 14 
Lepsien and Nobre (200614 
Yeh, Kuo, and Liu (200710 10 
Johnson, Mitchell, Raye, D'Esposito, and Johnson (200711 15 
Yi, Turk-Browne, Chun, and Johnson (200812 
Raye, Mitchell, Reeder, Greene, and Johnson (200813 29 
Johnson and Johnson (200914 14 
Nee and Jonides (200915 18 
Roth, Johnson, Raye, and Constable (200916 22 
Higo et al. (201117 21 
Table 2. 

Studies Included in the Precue Meta-analysis

StudyStudy No.No. of Participants
Mangun et al. (2000
Corbetta, Kincade, Ollinger, McAvoy, and Shulman (200013 
Giesbrecht, Woldorff, Song, and Mangun (200310 
Nobre et al. (200410 
Woldorff et al. (200420 
Wilson, Woldorff, and Mangun (200516 
de Haan, Morgan, and Rorden (200812 
Egner et al. (200814 
StudyStudy No.No. of Participants
Mangun et al. (2000
Corbetta, Kincade, Ollinger, McAvoy, and Shulman (200013 
Giesbrecht, Woldorff, Song, and Mangun (200310 
Nobre et al. (200410 
Woldorff et al. (200420 
Wilson, Woldorff, and Mangun (200516 
de Haan, Morgan, and Rorden (200812 
Egner et al. (200814 
Table 3. 

Activation Clusters from the Retrocue Meta-analysis

LabelCluster No.Local Maxima (MNI Coordinates)Cluster Volume (mm3)
xyz
Right anterior MFG 44 46 24 1696 
34 50 16 
Left anterior MFG −40 36 28 944 
Left precentral / Left anterior MFG −50 −2 40 5040 
−50 22 28 
−54 12 34 
Right anterior insula (fO) 48 12 −4 2056 
Left anterior insula (fO) −40 14 −4 1304 
dACC/pre-SMA 18 48 3272 
−2 30 36 
Right precentral 48 42 440 
Left MTG −58 −38 824 
−66 −42 
Left TPJ −56 −38 32 480 
−48 −38 32 
Left IPS 10 −38 −48 44 488 
Right SPL 11 12 −70 52 1400 
18 −62 54 
Right IPS 12 34 −58 42 1048 
38 −44 44 
30 −64 38 
LabelCluster No.Local Maxima (MNI Coordinates)Cluster Volume (mm3)
xyz
Right anterior MFG 44 46 24 1696 
34 50 16 
Left anterior MFG −40 36 28 944 
Left precentral / Left anterior MFG −50 −2 40 5040 
−50 22 28 
−54 12 34 
Right anterior insula (fO) 48 12 −4 2056 
Left anterior insula (fO) −40 14 −4 1304 
dACC/pre-SMA 18 48 3272 
−2 30 36 
Right precentral 48 42 440 
Left MTG −58 −38 824 
−66 −42 
Left TPJ −56 −38 32 480 
−48 −38 32 
Left IPS 10 −38 −48 44 488 
Right SPL 11 12 −70 52 1400 
18 −62 54 
Right IPS 12 34 −58 42 1048 
38 −44 44 
30 −64 38 
Table 4. 

Activation Clusters from the Precue Meta-analysis

LabelCluster #Local Maxima (MNI Coordinates)Cluster Volume (mm3)
xyz
Left anterior MFG −40 30 24 472 
Right FEF 32 50 1560 
Left precentral/left FEF −36 −4 42 464 
−42 −4 52 
Left FEF −22 −2 50 512 
−22 −8 58 
Left IPS −40 −48 36 544 
Right SPL 26 −56 58 1112 
Left IPS −22 −66 54 3960 
−18 −58 54 
−22 −68 40 
Right SPL 22 −72 50 768 
28 −68 44 
Right IPS0/V7 34 −78 26 584 
34 −74 26 
Left IPS0/V7 10 −28 −78 30 400 
Left occipital 11 −46 −70 −10 800 
Right occipital 12 34 −80 14 800 
34 −84 14 
LabelCluster #Local Maxima (MNI Coordinates)Cluster Volume (mm3)
xyz
Left anterior MFG −40 30 24 472 
Right FEF 32 50 1560 
Left precentral/left FEF −36 −4 42 464 
−42 −4 52 
Left FEF −22 −2 50 512 
−22 −8 58 
Left IPS −40 −48 36 544 
Right SPL 26 −56 58 1112 
Left IPS −22 −66 54 3960 
−18 −58 54 
−22 −68 40 
Right SPL 22 −72 50 768 
28 −68 44 
Right IPS0/V7 34 −78 26 584 
34 −74 26 
Left IPS0/V7 10 −28 −78 30 400 
Left occipital 11 −46 −70 −10 800 
Right occipital 12 34 −80 14 800 
34 −84 14 

To simplify this spatial pattern into a single set of ROIs for MEG analysis (Table 5), with a spatial sampling appropriate to the comparatively coarse spatial resolution of the method (as compared with fMRI), we converted these local maxima into a left/right symmetric spatially sparse set of brain locations by averaging local maxima for each cluster within each meta-analysis map. We then averaged activations common to the precue and retrocue meta-analyses (using a distance threshold of 16 mm). This set of combined coordinates was converted into a symmetric set of ROIs by averaging corresponding left and right hemisphere ROI locations and mirroring the activations that occurred in the left hemisphere only (middle temporal gyrus [MTG], TPJ) in the right hemisphere.

Table 5. 

ROIs Derived from the fMRI Meta-analysis

MNI Coordinates
xyz
IPS0 ±34 −76 26 
Mid-IPS ±30 −68 40 
Anterior IPS ±39 −48 40 
SPL ±12 −68 60 
FEF ±27 −3 52 
Precentral cortex (or iFEF) ±46 43 
Anterior MFG ±40 39 23 
TPJ ±52 −38 32 
MTG ±62 −40 
fO ±44 13 −4 
Pre-SMA 24 42 
MNI Coordinates
xyz
IPS0 ±34 −76 26 
Mid-IPS ±30 −68 40 
Anterior IPS ±39 −48 40 
SPL ±12 −68 60 
FEF ±27 −3 52 
Precentral cortex (or iFEF) ±46 43 
Anterior MFG ±40 39 23 
TPJ ±52 −38 32 
MTG ±62 −40 
fO ±44 13 −4 
Pre-SMA 24 42 

MRI Scan

A structural MRI was acquired for each participant using a Siemens 3T scanner (OCMR, Oxford, UK). A 32-channel head coil was used to obtain a T1-weighted anatomical image with 224 (1-mm) slices. This anatomical image was used to define the single-shell MEG forward model (Nolte, 2003). SPM's spm_eeg_inv_mesh was used to compute the transformation that mapped a set of canonical meshes for the cortical surface, skull, and scalp to each participant's individual anatomical MRI, and this transformation was used to define a MEG forward model tailored to each participant's head shape, computed using Fieldtrip's forward toolbox (Donders Institute, Nijmegen, The Netherlands; shared with SPM). Spatial coregistration between the forward model and MEG space was by aligning the spaces based on anatomical landmarks (nasion and left/right preauricular points) for a first estimate and then refining this fit using points recorded from the scalp surface (see MEG scan), which were matched to the scalp surface mesh using an iterative closest point algorithm as implemented in SPM.

MEG Scan

MEG data were acquired using an Elekta Neuromag 306-channel system (Elekta, Stockholm, Sweden; 204 planar gradiometers, 102 magnetometers). The MEG suite is passively shielded. ECG and vertical/horizontal EOG were recorded. Head position was continuously monitored using emitting coils affixed to the participant's scalp and a Polhemus 3D tracking system (Polhemus EastTrach 3D, Polhemus, Vermont, Unites States). Anatomical landmarks (nasion and left/right preauricular points) were recorded, as well as ∼100 points spread out over the scalp surface. MEG data were recorded in three blocks of ∼15 min each.

MEG Preprocessing

MEG data were initially inspected to remove any channels severely corrupted by noise and then de-noised and corrected for head movements using Elekta's Maxfilter Signal Space Separation algorithm (Taulu, Kajola, & Simola, 2004). Data were epoched and visually inspected again using Fieldtrip's visual artifact rejection tool for standard artifacts: Contaminated trials and channels were tagged on the basis of abnormal variance, kurtosis, and maxima/minima in the time domain data. Eyeblinks were detected from the EOG, and eye-tracker data using a semiautomatic algorithm and data from 200 msec before and until 300 msec after each blink were excluded from all analyses, including estimation of the beamformer weights.

MEG Analysis

Data were cut into three epochs for analysis: precue epoch, array epoch, and retrocue epoch (see Figure 1A). Key experimental contrasts compared (1) activity in trials in which there was a precue or retrocue with neutral-cue trials (“cue effects”) and (2) activity in leftward-cued trials with activity in rightward-cued trials (“cue laterality”—to measure alpha-power lateralization in visual cortex).

Sensor Space Analysis of Alpha Power Lateralization

The time domain sensor space signal was transformed to the frequency domain in 50-msec steps, using a Hanning taper/FFT algorithm with a taper spanning four cycles of the filtered frequency, for frequencies between 3 and 30 Hz in 1-Hz steps. The resulting power spectra were averaged over trials within each cue condition. The power time series in the planar gradiometer pairs were then combined (Cartesian sum), giving a 102-channel combined planar gradiometer map of sensor space power. The cue laterality subtractions [precue left minus precue right] and [retrocue left minus retrocue right] were computed per participant for the precue and retrocue epochs, respectively. Sensor space cluster permutation statistics (Maris & Oostenveld, 2007) were computed for these topographies by permuting cue left/cue right condition labels (using Fieldtrip's ft_freqstatistics). Clusters were formed in space (sensor proximity) and time, averaging over the alpha (8–12 Hz) band.

Sensor Space Classification Analysis of Alpha Lateralization

We used a simple correlation-based pattern classification approach to test whether the pattern of decrease in alpha power to an attentional cue resembled the pattern of event-related alpha desynchronization (ERD) in response to a physical stimulus. We averaged the induced responses to the probe stimulus between 300 and 500 msec (the time window for which within-epoch alpha-band quadrant classification for the probe item was highest), yielding four sensor space patterns for trials in which each of the four visual quadrants was probed. These probe ERD topographies were then correlated against the trial-wise cue-induced topographies from the precue and retrocue epochs. The trials were classified to a quadrant based on whichever probe stimulus quadrant pattern was most correlated with the pattern of activity on that trial. A leave-one-out approach was used to prevent cross-temporal correlations within trials from confounding the analysis: The to-be-classified trial was always excluded from the averages of probe stimulus activity. The classification results were averaged over time–frequency bins in the alpha band (8–12 Hz) and the interval 0.4–0.8 sec after the cue for plotting (Figure 2C, D).

Figure 2. 

Modulation of alpha-band (8–12 Hz) power in visual cortex. (A, B) Sensor space topography of alpha power for the contrast [cue left minus cue right], at 0.6 sec following the precue (200 msec FWHM) (A) and retrocue (B). Sensors belonging to significant clusters (see Results) are circled (black, positive clusters; white, negative clusters). (C, D) Classifying the direction of attention by correlating the cue-induced topography with the topography of the induced response to the probe item. Classification percentages are shown relative to the cued quadrant, averaged over the 8–12 Hz band, from 0.4 to 0.8 sec postcue. E, F show an alpha lateralization index calculated for the IPS0 virtual electrode. Alpha lateralization is persistent following precues, until presentation of the memory array. Lateralization subsides for most of the maintenance interval but ramps up just before presentation of the probe item. Following retrocues, alpha lateralization is transient, returning to near-baseline level by 1 sec postcue.

Figure 2. 

Modulation of alpha-band (8–12 Hz) power in visual cortex. (A, B) Sensor space topography of alpha power for the contrast [cue left minus cue right], at 0.6 sec following the precue (200 msec FWHM) (A) and retrocue (B). Sensors belonging to significant clusters (see Results) are circled (black, positive clusters; white, negative clusters). (C, D) Classifying the direction of attention by correlating the cue-induced topography with the topography of the induced response to the probe item. Classification percentages are shown relative to the cued quadrant, averaged over the 8–12 Hz band, from 0.4 to 0.8 sec postcue. E, F show an alpha lateralization index calculated for the IPS0 virtual electrode. Alpha lateralization is persistent following precues, until presentation of the memory array. Lateralization subsides for most of the maintenance interval but ramps up just before presentation of the probe item. Following retrocues, alpha lateralization is transient, returning to near-baseline level by 1 sec postcue.

Cross-temporal Correlation Analysis

To establish whether the patterns of brain activity we observed were stable or transient, we correlated cue-induced brain states across time, as indexed by the induced response topographies. We randomly subdivided all of the experimental trials into two groups of equal size (discarding trials if more trials had survived preprocessing in one condition than another). Within each of the two halves, data were averaged within conditions, and the cue effects contrast was computed. This yielded two independent estimates of the cue effects topography.

We performed this analysis for the theta (3–7 Hz), alpha (8–12 Hz), and beta (18–30 Hz) bands separately. As in previous analyses (Stokes et al., 2013), one half of the data was designated the training data, and the other half the test data. The topographies were extracted from the training data for each time point in the epoch. This topography was correlated with the topography at every time point in the test data, building up a cross-temporal correlation matrix (King & Dehaene, 2014). If a state is transient, then it gives rise to high correlation values mainly on the diagonal of this matrix. By contrast, temporally stable states will give rise to high correlation values confined to the diagonal of this matrix. The data were split randomly, so each time this analysis is run, a slightly different result will be obtained. We therefore performed this analysis 20 times and averaged the results; this bootstrapping procedure stabilizes the estimate of the correlation structure. We statistically evaluated the strength of the correlations at the group level by forming clusters in the time/time correlation space and then tested these against a permutation distribution of cluster size.

Source Space Analyses

To characterize the time course of activation in each ROI, a virtual electrode was created for each ROI coordinate using a linearly constrained minimum variance beamformer (Woolrich, Hunt, Groves, & Barnes, 2011). The time–frequency representation of the data was then computed at each virtual electrode, between 3 and 30 Hz. The time–frequency data were averaged across task conditions within participants. The condition averages (cue effects and cue laterality) were then subtracted within participants. These contrasts were averaged at the group level to create time–frequency maps of cue-related activity for each ROI. Significance testing was performed by forming clusters in the time/frequency space for each ROI, for positive and negative deviations in power, and then testing against a permutation distribution of cluster size.

To visualize spatial patterns of activation over the whole brain and verify that these were appropriately sampled by the ROIs derived from the meta-analysis, we ran whole brain-induced response analyses for the theta band (3–7 Hz), alpha band (8–12 Hz), and beta band (18–30 Hz). The 4-D spatiotemporal map for each analysis epoch (precue, retrocue) was averaged over successive 300-msec windows. Cluster permutation statistics for the 3-D maps were computed for the informative cue versus neutral cue contrasts with a cluster-forming threshold of 3 (t statistic).

RESULTS

Behavioral Data

Group level accuracy data are shown in Figure 1B: Both precues and retrocues improved response accuracy, increasing the proportion of correct responses across all magnitudes of orientation change. Group level parameters for the mixture model analysis described in Methods are plotted in Figure 1D, E. Precues and retrocues both reduced guess rate, but only precues substantially increased precision. A repeated-measures ANOVA with factor Cue condition (3 levels: neutral, precue, retrocue) found evidence for a main effect of Cue condition upon guess rate (F(1.74, 82.3) = 203.1, p < .0005). Paired sample t tests against the neutral condition confirmed that guess rate was significantly reduced in the precue condition (t(47) = 16.95, p < .0005) and retrocue condition (t(47) = 14.55, p < .0005).

For the precision parameter, there was also a main effect of Cue condition (F(2, 94) = 17.2, p < .0005). Paired-sample t tests indicated that the precision in the precue condition was significantly higher than in the neutral condition (t(47) = 5.63, p < .0005), but the increase in precision in the retrocue condition was only marginally significant (t(47) = 2.04, p = .047). Precision in the precue condition was significantly higher than precision in the retrocue condition (t(47) = 3.56, p = .001).

We also tested whether nontarget items affected behavior. This has previously been termed “misbinding” (Bays et al., 2011) but could also be characterized as “misgating,” if as argued here, the effect is attributable to selecting the wrong item in memory to guide behavior. We performed a repeated-measures ANOVA with factors Orientation bin (8 levels corresponding to the orientation bins described in Methods) and Cue condition (3 levels: neutral, precue and retrocue). There was a main effect of Orientation bin (F(7, 329) = 13.65, p < .0005) confirming that the nontarget items affected responding, and there was an interaction between Orientation bin and Cue condition (F(14, 658) = 4.11, p < .0005) indicating the degree of misgating differed between the cue conditions. These data are shown in Figure 1F.

To quantify the differences between cue conditions contributing to this interaction, we ran repeated-measures ANOVAs comparing pairs of conditions (i.e., as described above, but with cue condition levels (1) precue, retrocue; (2) precue, neutral cue; (3) retrocue, neutral cue). There was no evidence for a Cue condition × Orientation bin interaction for (1) precue versus retrocue (F(7, 329) = 0.598, p = .757), but there was a significant Cue condition × Orientation bin interaction for (2) precue versus neutral cue (F(7, 329) = 5.71, p < .0005) and for (3) retrocue and neutral cue (F(7, 329) = 5.87, p < .0005). Therefore, both precues and retrocues significantly reduced the propensity to respond on the basis of a nontarget item.

Relative to the neutral condition, precues and retrocues also reduced RTs. In Figure 1C, RT distributions are expressed in terms of quintile means: Precues and retrocues were associated with a strikingly similar RT distribution, with a reduction in RT relative to the neutral condition that was present across all quintiles (mean reduction in RT for precues, 422 msec, SEM = 34 msec; retrocues, 432 msec, SEM = 36 msec). We compared median RTs between the cue conditions using a repeated-measures ANOVA with factor Cue condition (3 levels). There was a significant main effect of Cue condition (F(2, 94) = 136.56, p < .0005). We then compared the conditions using paired-sample t tests. There was a significant difference between median RT for neutral and precue trials (p = 5.7 × 10−7) and neutral and retrocue trials (p = 2.9 × 10−7), but no significant difference between precue and retrocue trials (p = .90).

Behavioral analyses performed over the subset of 38 participants for whom we were able to analyze the MEG data were qualitatively the same as for the set of 48 participants who successfully completed the behavioral task, except that the marginally significant effect of retrocues upon precision (kappa) did not reach significance (t(37) = −1.58, p = .12).

Alpha Power Modulation in Perceptual and Parietal Cortex Indexes the Allocation of Attention

Alpha power in perceptual and parietal cortex is robustly modulated by preparatory attention (van Ede, Köster, & Maris, 2012; Haegens, Händel, & Jensen, 2011; Siegel, Donner, Oostenveld, Fries, & Engel, 2008; Worden, Foxe, Wang, & Simpson, 2000). Typically, when attention is directed to one side of space, there is a relative increase in alpha power in the ipsilateral cortex and decrease in alpha power in the contralateral cortex, compared with when attention is directed to the other hemifield. We hypothesized that the pattern of alpha activity in visual and parietal cortex would be similarly modulated by precues and retrocues, as both cue types may recruit a common top–down mechanism. We first performed a sensor space analysis subtracting the pattern of activation in trials in which a quadrant in the right hemifield was cued from the pattern of activation when a quadrant in the left hemifield was cued. The results (for a representative time point 0.6 sec postcue) are plotted in Figure 2A (precues) and B (retrocues). Both precues and retrocues robustly lateralized alpha activity.

Cluster permutation tests revealed a significant cluster of sensors with increased power over the left occipito-parietal sensors for the contrast precue left minus precue right (p = .0005). The cluster of sensors with decreased power over the right hemisphere sensors did not reach significance (p = .10). For the retrocue left minus retrocue right contrast, there was both a significant cluster of sensors with increased power centered over the left hemisphere occipito-parietal sensors (p = .0065) and a significant cluster of sensors with decreased power over the right hemisphere sensors (p = .001).

The source space time course for the laterality contrast was extracted at occipito-parietal ROIs and converted into a lateralization index by flipping the sign of the decrease in alpha power in the right hemisphere and adding it to the increase in alpha power in the left. Lateralization was strongest in the IPS0 ROI, for which precue and retrocue time courses are shown in Figure 2E and F. Precues gave rise to a more sustained alpha lateralization lasting from ∼0.5 sec postcue until the presentation of the memory array, whereas retrocues gave rise to a more transient lateralization of alpha power between 0.5 and 1 sec postcue that returned to baseline before the probe stimulus appeared. Note the very similar time course of decrease in alpha power in IPS for the cue effects contrast (Figure 5A).

To establish whether the alpha-power changes were quadrant-specific and resembled an ERD to a physical stimulus, we classified the cue-induced topographies based on the ERD to the probe item. The results are shown in Figure 2C and D, in which the classification pattern is shown averaged over time–frequency bins in the alpha band (8–12 Hz) between 0.4 and 0.8 sec following precues and retrocues, coded relative to the cued quadrant. We tested the quadrant specificity of this effect statistically by separately comparing up–down classification and left–right classification and testing resulting time–frequency clusters of above-chance classification against a permutation distribution of cluster sizes under the null hypothesis. Quadrant classification was significant for both precues (up/down cluster, p = .029; left/right cluster, p = .0002) and retrocues (up/down cluster, p < .0002, left/right cluster, p < .0002).

We also tested for an induced-response analogue to contralateral delay activity (Vogel & Machizawa, 2004) in which there is a sustained ERP lateralization during the retention interval following lateralized encoding by classifying cued quadrant in the WM retention interval in precue trials. For comparison, Sauseng et al. (2009) reported alpha lateralization in the delay interval of a task similar to that used by Vogel and colleagues. We found a transient effect in the alpha band (8–12 Hz) soon after the memory array, significant only for the left/right decoding (p = .0062) between 0.5 and 0.7 sec following the array onset (compare with the timing of peak lateralization following retrocues). We also found a stronger effect (Figure 2E) that emerged in the run-up (<1 sec before) to the probe stimulus (up/down decoding, p = .0022; left/right decoding, p = .0002). Our results therefore partly replicate those of Sauseng et al. (2009), but the longer retention interval in our experiment (3000 msec as compared to 900 msec in the previous study) revealed that alpha lateralization in the retention interval manifested over short intervals when attention was most likely to be lateralized, consistent with the transient lateralization observed following retrocues.

Cross-temporal Classification Analysis

To picture the overall temporal dynamics of brain activity in response to a precue or retrocue without first spatially selecting the data, we randomly split the trials in each condition into two halves and correlated the induced response topographies of contrasts performed on each half separately, across time (i.e., correlating the topography at time 1 with the topography at time 2—for all combinations of t1 and t2). As illustrated in Figure 3A, on-diagonal correlations reflect the reproducibility of topographies across the independent data sets, and the off-diagonal correlations capture the temporal persistence of these cue-induced brain states (see King & Dehaene, 2014, for an in-depth discussion). The analysis was performed separately for the theta (3–7 Hz), alpha (8–12 Hz), and beta (18–30 Hz) bands (Figure 3B).

Figure 3. 

Temporal dynamics of cue-induced brain states. (A) The expected correlation patterns for unstable and stable brain states. A continually changing brain state will not be correlated when t1 ≠ t2, that is, on the off-diagonal. By contrast a stable state will lead to off-diagonal correlations. (B) The correlation structure for the [precue – neutral] and [retrocue – neutral] topographies, that is, the stability of cue-induced brain states as indexed using sensor space induced responses in the theta (3–7 Hz), alpha (8–12 Hz), and beta (18–30 Hz) bands, over all sensors. Group level t statistics are shown. The correlation structure was tested using a permutation distribution of cluster size and is thresholded at t = 2. Significant correlations are in full color saturation; nonsignificant correlations are unsaturated. The response to precues remains significantly correlated between 0.4 sec and the end of the precue epoch (presentation of the memory array) indicating a persistent, stable state. By contrast, the response to retrocues is less stable, evolving throughout the analysis epoch.

Figure 3. 

Temporal dynamics of cue-induced brain states. (A) The expected correlation patterns for unstable and stable brain states. A continually changing brain state will not be correlated when t1 ≠ t2, that is, on the off-diagonal. By contrast a stable state will lead to off-diagonal correlations. (B) The correlation structure for the [precue – neutral] and [retrocue – neutral] topographies, that is, the stability of cue-induced brain states as indexed using sensor space induced responses in the theta (3–7 Hz), alpha (8–12 Hz), and beta (18–30 Hz) bands, over all sensors. Group level t statistics are shown. The correlation structure was tested using a permutation distribution of cluster size and is thresholded at t = 2. Significant correlations are in full color saturation; nonsignificant correlations are unsaturated. The response to precues remains significantly correlated between 0.4 sec and the end of the precue epoch (presentation of the memory array) indicating a persistent, stable state. By contrast, the response to retrocues is less stable, evolving throughout the analysis epoch.

The cross-temporal correlation matrix shows that, after precues, a stable state emerges from ∼0.6 sec postcue until the presentation of the memory array (Figure 3B, “square” in upper right for precues). Retrocues gave rise to a different pattern. Cue-induced activity following a retrocue continues to evolve throughout the analysis epoch. This overview of the temporal structure of the cue-induced responses can be compared with the source space analysis of induced responses shown in Figure 5.

fMRI Meta-analysis

The results of the meta-analyses are given in Tables 3 and 4, and the ROIs derived from the meta-analysis results are shown in Figure 4. There was overlap between precue- and retrocue-associated activations in the anterior middle frontal gyrus (MFG; or dlPFC) and intraparietal sulcus (IPS). Both cue types also activated regions in the precentral gyrus. Comparing the meta-analysis results, there was a dissociation between cue types, in that precue studies reported activity in the FEF whereas retrocue studies reported activity in a more inferior region we call iFEF (adopting Derrfuss' terminology; Derrfuss, Vogt, Fiebach, von Cramon, & Tittgemeyer, 2012). This is consistent with the previous study by Nee and Jonides (2009), which identified this dissociation. iFEF is in proximity to IFJ, and the retrocue meta-analysis activation cluster encompassed the coordinates of both regions. Kastner and colleagues (2007) have also identified these two dissociable precentral regions.

Figure 4. 

Meta-analysis of prospective and retrospective cueing tasks. MEG ROIs derived from meta-analysis activation results. The ROIs are colored by network membership (following Dosenbach et al., 2007). As ROIs were symmetric across hemispheres, we show the left hemisphere only.

Figure 4. 

Meta-analysis of prospective and retrospective cueing tasks. MEG ROIs derived from meta-analysis activation results. The ROIs are colored by network membership (following Dosenbach et al., 2007). As ROIs were symmetric across hemispheres, we show the left hemisphere only.

Retrocues additionally activated the bilateral fO and the pre-SMA. The left posterior MTG and the left inferior parietal lobule were also activated following retrocues.

The meta-analysis results were consistent with previously described frontoparietal and cingulo-opercular control networks (Petersen & Posner, 2012; Dosenbach et al., 2007). Precues and retrocues both activated the frontoparietal network, and retrocues additionally activated the cingulo-opercular network.

Induced Responses in Control Networks

The ROIs derived from the fMRI meta-analysis were used to extract source space-induced responses from frontal and parietal control regions. The contrast [informative cue minus neutral cue] between 3 and 30 Hz is shown for frontoparietal and cingulo-opercular ROIs in Figure 5A. The remaining parietal ROIs (IPS, SPL) (not shown) showed a similar pattern of responses to the mid-IPS.

Figure 5. 

Cue effects contrasts. (A) Induced responses in frontoparietal and cingulo-opercular ROIs reveals biphasic time course of control network activation. The time–frequency (TF) plots are for the contrast: informative cue minus neutral cue. TF data are averaged over left and right hemisphere ROIs, as most effects were bilateral, with the exception of the two plots marked with asterisks (anterior MFG and mid IPS following precues) for which effects were left-lateralized. Significant activation clusters are shown in full color saturation. Precues give rise to a sustained alpha/beta desynchronization in the left mid-IPS (p < .0002) lasting until the presentation of the memory array, matching the time course of alpha lateralization in occipital cortex, consistent with a role for left IPS as a proximal control region for this attentional effect. Right anterior MFG (p = .0004, early cluster; p = .029, late cluster) and bilateral FEF (p = .0016) are activated in the theta band early following the cue. Cingulo-opercular nodes are not activated with the exception of a short-lived activation in the pre-SMA immediately following the cue (p = .006). Retrocues gave rise to an alpha/beta desynchronization in the mid-IPS (p < .0002), which lasted until ∼1 sec postcue, matching the time course of alpha lateralization. Activations in the theta and alpha/beta band in the anterior MFG (p = .0054, p = .0004) preceded the parieto-occipital effects. Consistent with the pattern in fMRI, retrocues did not activate the FEF but did give rise to a bilateral activation in the more ventral precentral ROI (p = .0036). Retrocues also gave rise to activations in the cingulo-opercular nodes. The pre-SMA was activated immediately following the retrocue in the theta band (p = .0006) and also later in the epoch in the beta-band (∼1.2 sec postcue; p = .006). At this later time point, there was also a bilateral activation in the anterior insula/fO in the theta/alpha band (p < .0002). (B) Whole-brain activation maps for the contrast [informative cue – neutral cue] projected onto the CARET cortical surface, averaged over the 300–600 msec time period following both precues and retrocues. Only activation belonging to a statistically significant activation cluster is shown. (C) Whole-brain activation maps for the contrast [informative cue – neutral cue] projected onto the CARET cortical surface, averaged over the 1200–1500 msec time period following both precues and retrocues. Only activation belonging to a statistically significant activation cluster is shown.

Figure 5. 

Cue effects contrasts. (A) Induced responses in frontoparietal and cingulo-opercular ROIs reveals biphasic time course of control network activation. The time–frequency (TF) plots are for the contrast: informative cue minus neutral cue. TF data are averaged over left and right hemisphere ROIs, as most effects were bilateral, with the exception of the two plots marked with asterisks (anterior MFG and mid IPS following precues) for which effects were left-lateralized. Significant activation clusters are shown in full color saturation. Precues give rise to a sustained alpha/beta desynchronization in the left mid-IPS (p < .0002) lasting until the presentation of the memory array, matching the time course of alpha lateralization in occipital cortex, consistent with a role for left IPS as a proximal control region for this attentional effect. Right anterior MFG (p = .0004, early cluster; p = .029, late cluster) and bilateral FEF (p = .0016) are activated in the theta band early following the cue. Cingulo-opercular nodes are not activated with the exception of a short-lived activation in the pre-SMA immediately following the cue (p = .006). Retrocues gave rise to an alpha/beta desynchronization in the mid-IPS (p < .0002), which lasted until ∼1 sec postcue, matching the time course of alpha lateralization. Activations in the theta and alpha/beta band in the anterior MFG (p = .0054, p = .0004) preceded the parieto-occipital effects. Consistent with the pattern in fMRI, retrocues did not activate the FEF but did give rise to a bilateral activation in the more ventral precentral ROI (p = .0036). Retrocues also gave rise to activations in the cingulo-opercular nodes. The pre-SMA was activated immediately following the retrocue in the theta band (p = .0006) and also later in the epoch in the beta-band (∼1.2 sec postcue; p = .006). At this later time point, there was also a bilateral activation in the anterior insula/fO in the theta/alpha band (p < .0002). (B) Whole-brain activation maps for the contrast [informative cue – neutral cue] projected onto the CARET cortical surface, averaged over the 300–600 msec time period following both precues and retrocues. Only activation belonging to a statistically significant activation cluster is shown. (C) Whole-brain activation maps for the contrast [informative cue – neutral cue] projected onto the CARET cortical surface, averaged over the 1200–1500 msec time period following both precues and retrocues. Only activation belonging to a statistically significant activation cluster is shown.

The spatial pattern of induced responses in the control region ROIs replicated the spatial pattern of BOLD activations captured in the meta-analysis. Both precues and retrocues activated the anterior MFG and mid-IPS, with precues additionally activating FEF and retrocues activating iFEF. In mid-IPS, an increase in theta power was accompanied by a decrease in alpha/beta power that was sustained for precues and transient for retrocues. This decrease was strongest in the left IPS. These frontoparietal network power increases occurred at a similar latency after the cue following both precues and retrocues.

Only retrocues coactivated the fO and pre-SMA (cingulo-opercular network), again replicating the pattern observed in the fMRI meta-analysis. In contrast to the frontoparietal network, these regions coactivated late in the retrocue epoch, after the decrease in power in mid-IPS had returned to baseline ∼1 sec postcue. The fO power increases were in the theta/alpha band, and the pre-SMA in the beta band. The pre-SMA also responded in the theta band immediately following both precues and retrocues. These power increases were not present in the [precue minus neutral cue] analysis for the same retrocue epoch time period and are therefore unlikely to correspond to preparatory activity for the probe stimulus.

We verified that the ROIs appropriately sampled the pattern of activity in source space by computing whole-brain maps of activity over successive 300-msec windows. The whole brain-induced response maps reproduced the patterns expected on the basis of the fMRI meta-analysis. This is illustrated in Figure 5B, which shows the time period 300–600 msec postcue, and in Figure 5C, which shows the time period 1200–1500 msec postcue.

DISCUSSION

Selection mechanisms are effective in improving WM accuracy whether cues are prospective, selecting what “gets into” WM, or retrospective, selecting from within WM. Although much of the previous literature has suggested that both effects may be mediated by a common top–down mechanism modulating activity in sensory cortex, mediated by a frontoparietal network (Gazzaley & Nobre, 2012), other investigators have suggested that a separate cingulo-opercular network is specifically involved in exerting control over the contents of WM (Nelissen et al., 2013; Higo et al., 2011). We addressed this question by recording MEG data while participants processed either a precue or a retrocue. The spatial resolution of the source space MEG recordings was sufficient to replicate the spatial pattern of network activations seen in a meta-analysis of previous fMRI studies, and the high temporal resolution of the method allowed us to dissociate the activation time course of the two networks.

The spatiotemporal pattern of induced responses to precues and retrocues implicated the frontoparietal and cingulo-opercular networks in different aspects of cognitive control. A cross-temporal correlation analysis of the sensor data showed that the brain response to precues had a sustained component, whereas retrocues gave rise to a response that continued to evolve over the analysis epoch. A source space analysis showed that the frontoparietal sites associated with precueing were activated in the early phase of the response following retrocues, but that the cingulo-opercular sites activated later. Comparing these network dynamics with the time course of alpha modulation in perceptual cortex, they are consistent with the suggestion that the frontoparietal network is responsible for top–down control over sensory cortex (as reviewed in Gazzaley & Nobre, 2012). We propose that, consistent with prior hypotheses (Gazzaley & Nobre, 2012; Nobre et al., 2004), a common frontoparietal network is responsible for endogenously modulating activity in sensory cortex, whether this is to bias sensory processing during preparatory attention or to reactivate a sensory representation during memory retrieval. However, we stress that, because we did not use a connectivity analysis, we did not obtain direct evidence that the frontoparietal network was responsible for top–down control.

Following retrocues, the cingulo-opercular power increases occurred later in the trial, once the frontoparietal and sensory power changes had subsided. This disparity in activation timing is inconsistent with the hypothesis that these cingulo-opercular sites directly modulate activity in sensory cortex during top–down control (Nelissen et al., 2013; Higo et al., 2011) but suggests that they are important for a separate operation specifically associated with retrocueing, as discussed below.

The “dual network” account proposed by Dosenbach, Fair, Cohen, Schlaggar, and Petersen (2008) dissociates the frontoparietal and cingulo-opercular networks on the basis of temporal scale of control operations: Although the frontoparietal network is involved in moment-by-moment adjustment of top–down control based on evolving task requirements, the cingulo-opercular network is a parallel system maintaining task set over longer periods. However, the finding that the cingulo-opercular network is transiently recruited following retrocues suggests a more dynamic role in ongoing cognitive control. Our results can be compared with an fMRI study by Ploran and colleagues (2007), in which images were slowly revealed in noise, until participants were able to make a discrimination response. Activity in the frontoparietal network slowly increased as evidence was accumulated, but the cingulo-opercular network was activated only at the moment participants made their response. These data suggested that the cingulo-opercular network has a more “downstream” role, acting on evidence integrated by the frontoparietal network in an interaction with sensory cortex. Analogously, we suggest that in our task the frontoparietal sites act to retrieve perceptual information about the cued item, and the cingulo-opercular network underpins a downstream stage. This is broadly consistent with a “cascade” account of executive function (Banich, 2009), but further work is needed to characterize the nature of this secondary role for the cingulo-opercular network. Plausible functions include prioritization of the retrieved information to drive the response to the probe item or inhibition of interfering information (uncued memoranda). The latter hypothesis might explain why we did not observe a similar power increase in the cingulo-opercular network before the presentation of the probe item in the precue condition: If there is only one item in memory, there is are no other items that might interfere with the cognitive operation performed on the relevant item (in this case, orientation comparison with the probe stimulus).

Our results are also consistent with a conservative view of the role of attention in memory maintenance. Modulations of alpha power in perceptual and parietal cortex are a reliable marker for preparatory attention (van Ede, de Lange, Jensen, & Maris, 2011; Rihs, Michel, & Thut, 2007). We found both the expected quadrant-specific modulation of alpha power following precues, but also a very similar pattern following retrocues, consistent with recent reports by Poch, Campo, and Barnes (2014) and Myers, Walther, Wallis, Stokes, and Nobre (2015). Alpha lateralization was sustained following precues, until the presentation of the memory array, but transient following retrocues, peaking between 0.5 and 1 sec postcue and then returning to baseline. We also observed alpha lateralization in the retention interval as previously reported (Sauseng et al., 2009), but interestingly this was also not sustained, as might have been expected if memory maintenance consisted in sustained top–down activation of representations in parietal or sensory cortex (Kiyonaga & Egner, 2012; Awh, Vogel, & Oh, 2006). Instead, over the relatively long retention interval in our experiment, there were two periods of significant alpha lateralization: one close after the presentation of the memory array (significant between 0.5 and 0.7 sec postcue in the classification analysis) and then another that ramped up toward the presentation of the memory probe (see Figure 2E). Intriguingly, alpha lateralization did not also ramp up in readiness for the probe item following retrocues. We speculate that this may be because, in the 1.5 sec between retrocue and probe item, participants are occupied retrieving and preparing to use the retrocued item, and there is not enough time to make a preparatory switch in attention back toward the probe item.

Taken together, these results are consistent with the idea that alpha lateralization in parietal/perceptual cortex during memory maintenance tracks top–down activation, but that this may be dissociable from maintenance processes (LaRocque et al., 2014; Lewis-Peacock, Drysdale, Oberauer, & Postle, 2012). Rather than interpreting the retinotopic reactivation representing the cued item following a retrocue as biasing of ongoing maintenance activity, an alternative interpretation is that this reactivation reflects a brief “access” event in which the sensory properties of the cued item are reactivated in order that it can be prioritized in memory. Various contemporary models of WM have adopted a multilevel framework (Oberauer & Hein, 2012; Olivers, Peters, Houtkamp, & Roelfsema, 2011), in which, of the set of items in memory, a single item can be elevated to a special prioritized state. Sensory reactivation following a retrocue may be involved in moving items into this prioritized state.

Our behavioral data also support this interpretation, as they were more consistent with retrocues modulating the retrievability of an item in memory (“output gating”) than with retrocues changing whether (or with what fidelity) the item was maintained in memory. Both precues and retrocues reduced guess rate and the rate of responding about uncued items and also caused a leftward shift in the entire RT distribution compared with neutral-cue trials, consistent with faster retrieval of cued items. However, precues had a substantial effect on the precision with which the cued items were represented, whereas precision was only marginally modulated by retrocues. This is not consistent with the proposal that retrocues act primarily by protecting cued items from gradual decay (Pertzov, Bays, Joseph, & Husain, 2012), as were this the case we would expect to have seen a more substantial precision advantage following retrocues. An alternative explanation in terms of maintenance processes is that retrocues protect items from “sudden death” during the retention interval, which might explain the difference in guess rate. However, in Murray et al. (2013) retrocues boosted performance even if compared against a condition in which the probe item was presented early, at the same time as the retrocue, implying that the retrocue benefit could not depend only on protecting items from forgetting during the remainder of the retention interval. Finally, in the current study, retrocues almost completely abolished the effect of nontarget items on behavior, implying that retrocues can prevent retrieval errors in which the wrong item is selected to guide behavior—that is, they mitigate against errors in output gating. These aspects of the behavioral data are all consistent with retrocues facilitating output gating, as opposed to optimizing memory maintenance.

In summary, although selection from WM following a retrocue involves a similar top–down modulation of sensory and parietal cortex and a similar pattern of frontoparietal network activation as does preparatory attention, our data suggest that control over WM is not identical with top–down attention acting to bias memory maintenance activity in sensory and parietal cortex. Sensory reactivation was transient, and there was no evidence for sustained biasing of maintenance activity following cues. Instead we would suggest that the frontoparietal network mediates top–down control over sensory cortex, which can be recruited either to bias perception (attention) or to retrieve perceptual content associated with WM. We found, in line with previous studies (Nelissen et al., 2013; Higo et al., 2011), that the second cingulo-opercular network was specifically recruited by retrocues, but not precues. However, the activation timing suggested it was not directly involved in control over sensory representations, as prior studies have suggested. The precise role of the cingulo-opercular network in cognitive control remains to be elucidated. Broadly, precues facilitate input gating whereas retrocues may facilitate output gating of memory (Hazy, Frank, & O'Reilly, 2007). The frontoparietal network has a role in both input and output gating, but the cingulo-opercular network may be specifically associated with output gating.

Acknowledgments

This work was funded by Wellcome Trust studentships to G. W. and H. C., MRC fellowship MR/J009024/1 to M. S., a Wellcome Trust Equipment Grant to A. C. N. to support OHBA, the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre based at Oxford University Hospitals Trust Oxford University, and an MRC UK MEG Partnership grant (MR/K005464/1). The authors would like to thank Nils Kolling, Nicholas Myers, and Franz Neubert for helpful discussions and Sven Braeutigam, Henry Luckhoo, Diego Viduarre, and Adam Baker for their assistance with the MEG analysis.

Reprint requests should be sent to George Wallis, Oxford Centre for Human Brain Activity, Warneford Hospital, Oxford, OX3 7JX, UK, or via e-mail: wallisgj@gmail.com.

REFERENCES

REFERENCES
Awh
,
E.
,
Vogel
,
E. K.
, &
Oh
,
S. H.
(
2006
).
Interactions between attention and working memory
.
Neuroscience
,
139
,
201
208
.
Banich
,
M. T.
(
2009
).
Executive function: The search for an integrated account
.
Current Directions in Psychological Science
,
18
,
89
94
.
Bays
,
P. M.
,
Catalao
,
R. F. G.
, &
Husain
,
M.
(
2009
).
The precision of visual working memory is set by allocation of a shared resource
.
Journal of Vision
,
9
,
7.1
711
.
Bays
,
P. M.
,
Gorgoraptis
,
N.
,
Wee
,
N.
,
Marshall
,
L.
, &
Husain
,
M.
(
2011
).
Temporal dynamics of encoding, storage, and reallocation of visual working memory
.
Journal of Vision
,
11
,
1
15
.
Chatham
,
C. H.
,
Frank
,
M. J.
, &
Badre
,
D.
(
2014
).
Corticostriatal output gating during selection from working memory
.
Neuron
,
81
,
930
942
.
Corbetta
,
M.
,
Kincade
,
J. M.
,
Ollinger
,
J. M.
,
McAvoy
,
M. P.
, &
Shulman
,
G. L.
(
2000
).
Voluntary orienting is dissociated from target detection in human posterior parietal cortex
.
Nature Neuroscience
,
3
,
292
297
.
de Haan
,
B.
,
Morgan
,
P. S.
, &
Rorden
,
C.
(
2008
).
Covert orienting of attention and overt eye movements activate identical brain regions
.
Brain Research
,
1204
,
102
111
.
Derrfuss
,
J.
,
Vogt
,
V. L.
,
Fiebach
,
C. J.
,
von Cramon
,
D. Y.
, &
Tittgemeyer
,
M.
(
2012
).
Functional organization of the left inferior precentral sulcus: Dissociating the inferior frontal eye field and the inferior frontal junction
.
Neuroimage
,
59
,
3829
3837
.
Dosenbach
,
N. U. F.
,
Fair
,
D. A.
,
Cohen
,
A. L.
,
Schlaggar
,
B. L.
, &
Petersen
,
S. E.
(
2008
).
A dual-networks architecture of top–down control
.
Trends in Cognitive Sciences
,
12
,
99
105
.
Dosenbach
,
N. U. F.
,
Fair
,
D. A.
,
Miezin
,
F. M.
,
Cohen
,
A. L.
,
Wenger
,
K. K.
,
Dosenbach
,
R. A. T.
, et al
(
2007
).
Distinct brain networks for adaptive and stable task control in humans
.
Proceedings of the National Academy of Sciences, U.S.A.
,
104
,
11073
11078
.
Egner
,
T.
,
Monti
,
J. M. P.
,
Trittschuh
,
E. H.
,
Wieneke
,
C. A.
,
Hirsch
,
J.
, &
Mesulam
,
M. M.
(
2008
).
Neural integration of top–down spatial and feature-based information in visual search
.
Journal of Neuroscience
,
28
,
6141
6151
.
Eickhoff
,
S. B.
,
Laird
,
A. R.
,
Grefkes
,
C.
,
Wang
,
L. E.
,
Zilles
,
K.
, &
Fox
,
P. T.
(
2009
).
Coordinate-based activation likelihood estimation meta-analysis of neuroimaging data: A random-effects approach based on empirical estimates of spatial uncertainty
.
Human Brain Mapping
,
30
,
2907
2926
.
Gazzaley
,
A.
, &
Nobre
,
A. C.
(
2012
).
Top–down modulation: Bridging selective attention and working memory
.
Trends in Cognitive Sciences
,
16
,
129
135
.
Giesbrecht
,
B.
,
Woldorff
,
M. G.
,
Song
,
A. W.
, &
Mangun
,
G. R.
(
2003
).
Neural mechanisms of top–down control during spatial and feature attention
.
Neuroimage
,
19
,
496
512
.
Griffin
,
I. C.
, &
Nobre
,
A. C.
(
2003
).
Orienting attention to locations in internal representations
.
Journal of Cognitive Neuroscience
,
15
,
1176
1194
.
Haegens
,
S.
,
Händel
,
B. F.
, &
Jensen
,
O.
(
2011
).
Top–down controlled alpha band activity in somatosensory areas determines behavioral performance in a discrimination task
.
Journal of Neuroscience
,
31
,
5197
5204
.
Harrison
,
S. A.
, &
Tong
,
F.
(
2009
).
Decoding reveals the contents of visual working memory in early visual areas
.
Nature
,
458
,
632
635
.
Hazy
,
T. E.
,
Frank
,
M. J.
, &
O'Reilly
,
R. C.
(
2007
).
Towards an executive without a homunculus: Computational models of the prefrontal cortex/basal ganglia system
.
Philosophical Transactions of the Royal Society of London, Series B, Biological Sciences
,
362
,
1601
1613
.
Higo
,
T.
,
Mars
,
R. B.
,
Boorman
,
E. D.
,
Buch
,
E. R.
, &
Rushworth
,
M. F. S.
(
2011
).
Distributed and causal influence of frontal operculum in task control
.
Proceedings of the National Academy of Sciences, U.S.A.
,
108
,
4230
4235
.
Johnson
,
M. R.
, &
Johnson
,
M. K.
(
2009
).
Top–down enhancement and suppression of activity in category-selective extrastriate cortex from an act of reflective attention
.
Journal of Cognitive Neuroscience
,
21
,
2320
2327
.
Johnson
,
M. R.
,
Mitchell
,
K. J.
,
Raye
,
C. L.
,
D'Esposito
,
M.
, &
Johnson
,
M. K.
(
2007
).
A brief thought can modulate activity in extrastriate visual areas: Top–down effects of refreshing just-seen visual stimuli
.
Neuroimage
,
37
,
290
299
.
Kastner
,
S.
,
DeSimone
,
K.
,
Konen
,
C. S.
,
Szczepanski
,
S. M.
,
Weiner
,
K. S.
, &
Schneider
,
K. A.
(
2007
).
Topographic maps in human frontal cortex revealed in memory-guided saccade and spatial working-memory tasks
.
Journal of Neurophysiology
,
97
,
3494
3507
.
King
,
J.-R.
, &
Dehaene
,
S.
(
2014
).
Characterizing the dynamics of mental representations: The temporal generalization method
.
Trends in Cognitive Sciences
,
18
,
203
210
.
Kiyonaga
,
A.
, &
Egner
,
T.
(
2012
).
Working memory as internal attention: Toward an integrative account of internal and external selection processes
.
Psychonomic Bulletin & Review
,
20
,
228
242
.
Kuo
,
B.-C.
,
Stokes
,
M. G.
,
Murray
,
A. M.
, &
Nobre
,
A. C.
(
2014
).
Attention biases visual activity in visual short-term memory
.
Journal of Cognitive Neuroscience
,
26
,
1377
1389
.
Landman
,
R.
,
Spekreijse
,
H.
, &
Lamme
,
V. A. F.
(
2003
).
Large capacity storage of integrated objects before change blindness
.
Vision Research
,
43
,
149
164
.
LaRocque
,
J. J.
,
Lewis-Peacock
,
J. A.
, &
Postle
,
B. R.
(
2014
).
Multiple neural states of representation in short-term memory? It's a matter of attention
.
Frontiers in Human Neuroscience
,
8
,
5
.
Lepsien
,
J.
,
Griffin
,
I. C.
,
Devlin
,
J. T.
, &
Nobre
,
A. C.
(
2005
).
Directing spatial attention in mental representations: Interactions between attentional orienting and working-memory load
.
Neuroimage
,
26
,
733
743
.
Lepsien
,
J.
, &
Nobre
,
A. C.
(
2006
).
Attentional modulation of object representations in working memory
.
Cerebral Cortex (New York, N.Y.: 1991)
,
17
,
2072
2083
.
Lewis-Peacock
,
J. A.
,
Drysdale
,
A. T.
,
Oberauer
,
K.
, &
Postle
,
B. R.
(
2012
).
Neural evidence for a distinction between short-term memory and the focus of attention
.
Journal of Cognitive Neuroscience
,
24
,
61
79
.
Mangun
,
G. R.
,
Hopfinger
,
J. B.
, &
Buonocore
,
M. H.
(
2000
).
The neural mechanisms of top–down attentional control
.
Nature Neuroscience
,
3
,
284
291
.
Maris
,
E.
, &
Oostenveld
,
R.
(
2007
).
Nonparametric statistical testing of EEG- and MEG-data
.
Journal of Neuroscience Methods
,
164
,
177
190
.
Munneke
,
J.
,
Belopolsky
,
A. V.
, &
Theeuwes
,
J.
(
2012
).
Shifting attention within memory representations involves early visual areas
.
PLoS One
,
7
,
e35528
.
Murray
,
A. M.
,
Nobre
,
A. C.
,
Clark
,
I. A.
,
Cravo
,
A. M.
, &
Stokes
,
M. G.
(
2013
).
Attention restores discrete items to visual short-term memory
.
Psychological Science
,
24
,
550
556
.
Murray
,
A. M.
,
Nobre
,
A. C.
, &
Stokes
,
M. G.
(
2011
).
Markers of preparatory attention predict visual short-term memory performance
.
Neuropsychologia
,
49
,
1458
1465
.
Myers
,
N. E.
,
Walther
,
L.
,
Wallis
,
G.
,
Stokes
,
M. G.
, &
Nobre
,
A. C.
(
2015
).
Temporal dynamics of attention during encoding versus maintenance of working memory: Complementary views from event-related potentials and alpha-band oscillations
.
Journal of Cognitive Neuroscience
,
27
,
492
508
.
Nee
,
D. E.
, &
Jonides
,
J.
(
2009
).
Common and distinct neural correlates of perceptual and memorial selection
.
Neuroimage
,
45
,
963
975
.
Nelissen
,
N.
,
Stokes
,
M.
,
Nobre
,
A. C.
, &
Rushworth
,
M. F. S.
(
2013
).
Frontal and parietal cortical interactions with distributed visual representations during selective attention and action selection
.
Journal of Neuroscience
,
33
,
16443
16458
.
Nobre
,
A. C.
,
Coull
,
J. T.
,
Maquet
,
P.
,
Frith
,
C. D.
,
Vandenberghe
,
R.
, &
Mesulam
,
M. M.
(
2004
).
Orienting attention to locations in perceptual versus mental representations
.
Journal of Cognitive Neuroscience
,
16
,
363
373
.
Nolte
,
G.
(
2003
).
The magnetic lead field theorem in the quasi-static approximation and its use for magnetoencephalography forward calculation in realistic volume conductors
.
Physics in Medicine and Biology
,
48
,
3637
3652
.
Oberauer
,
K.
, &
Hein
,
L.
(
2012
).
Attention to information in working memory
.
Current Directions in Psychological Science
,
21
,
164
169
.
Olivers
,
C. N.
,
Peters
,
J.
,
Houtkamp
,
R.
, &
Roelfsema
,
P. R.
(
2011
).
Different states in visual working memory: When it guides attention and when it does not
.
Trends in Cognitive Sciences
,
15
,
327
334
.
Pasternak
,
T.
, &
Greenlee
,
M. W.
(
2005
).
Working memory in primate sensory systems
.
Nature Reviews Neuroscience
,
6
,
97
107
.
Pelli
,
D. G.
(
1997
).
The VideoToolbox software for visual psychophysics: Transforming numbers into movies
.
Spatial Vision
,
10
,
437
442
.
Pertzov
,
Y.
,
Bays
,
P. M.
,
Joseph
,
S.
, &
Husain
,
M.
(
2012
).
Rapid forgetting prevented by retrospective attention cues
.
Journal of Experimental Psychology: Human Perception and Performance
,
39
,
1224
.
Petersen
,
S. E.
, &
Posner
,
M. I.
(
2012
).
The attention system of the human brain: 20 Years after
.
Annual Review of Neuroscience
,
35
,
73
89
.
Ploran
,
E. J.
,
Nelson
,
S. M.
,
Velanova
,
K.
,
Donaldson
,
D. I.
,
Petersen
,
S. E.
, &
Wheeler
,
M. E.
(
2007
).
Evidence accumulation and the moment of recognition: Dissociating perceptual recognition processes using fMRI
.
Journal of Neuroscience
,
27
,
11912
11924
.
Poch
,
C.
,
Campo
,
P.
, &
Barnes
,
G. R.
(
2014
).
Modulation of alpha and gamma oscillations related to retrospectively orienting attention within working memory
.
European Journal of Neuroscience
,
40
,
2399
2405
.
Raye
,
C. L.
,
Johnson
,
M. K.
,
Mitchell
,
K. J.
,
Reeder
,
J. A.
, &
Greene
,
E. J.
(
2002
).
Neuroimaging a single thought: Dorsolateral PFC activity associated with refreshing just-activated information
.
Neuroimage
,
15
,
447
453
.
Raye
,
C. L.
,
Mitchell
,
K. J.
,
Reeder
,
J. A.
,
Greene
,
E. J.
, &
Johnson
,
M. K.
(
2008
).
Refreshing one of several active representations: Behavioral and functional magnetic resonance imaging differences between young and older adults
.
Journal of Cognitive Neuroscience
,
20
,
852
862
.
Rihs
,
T. A.
,
Michel
,
C. M.
, &
Thut
,
G.
(
2007
).
Mechanisms of selective inhibition in visual spatial attention are indexed by alpha-band EEG synchronization
.
European Journal of Neuroscience
,
25
,
603
610
.
Roth
,
J. K.
,
Johnson
,
M. K.
,
Raye
,
C. L.
, &
Constable
,
R. T.
(
2009
).
Similar and dissociable mechanisms for attention to internal versus external information
.
Neuroimage
,
48
,
601
608
.
Rowe
,
J. B.
, &
Passingham
,
R. E.
(
2001
).
Working memory for location and time: Activity in prefrontal area 46 relates to selection rather than maintenance in memory
.
Neuroimage
,
14
,
77
86
.
Rowe
,
J. B.
,
Toni
,
I.
,
Josephs
,
O.
,
Frackowiak
,
R. S.
, &
Passingham
,
R. E.
(
2000
).
The prefrontal cortex: Response selection or maintenance within working memory?
Science
,
288
,
1656
1660
.
Sauseng
,
P.
,
Klimesch
,
W.
,
Stadler
,
W.
,
Schabus
,
M.
,
Doppelmayr
,
M.
,
Hanslmayr
,
S.
, et al
(
2009
).
A shift of visual spatial attention is selectively associated with human EEG alpha activity
.
European Journal of Neuroscience
,
22
,
2917
2926
.
Siegel
,
M.
,
Donner
,
T. H.
,
Oostenveld
,
R.
,
Fries
,
P.
, &
Engel
,
A. K.
(
2008
).
Neuronal synchronization along the dorsal visual pathway reflects the focus of spatial attention
.
Neuron
,
60
,
709
719
.
Sligte
,
I. G.
,
Scholte
,
H. S.
, &
Lamme
,
V. A. F.
(
2008
).
Are there multiple visual short-term memory stores?
PLoS One
,
3
,
e1699
.
Sligte
,
I. G.
,
Scholte
,
H. S.
, &
Lamme
,
V. A. F.
(
2009
).
V4 activity predicts the strength of visual short-term memory representations
.
Journal of Neuroscience
,
29
,
7432
7438
.
Sperling
,
G.
(
1960
).
The information available in brief visual presentations
.
Psychological Monographs: General and Applied
,
74
,
1
.
Stokes
,
M. G.
,
Kusunoki
,
M.
,
Sigala
,
N.
,
Nili
,
H.
,
Gaffan
,
D.
, &
Duncan
,
J.
(
2013
).
Dynamic coding for cognitive control in prefrontal cortex
.
Neuron
,
78
,
364
375
.
Taulu
,
S.
,
Kajola
,
M.
, &
Simola
,
J.
(
2004
).
Suppression of interference and artifacts by the signal space separation method
.
Brain Topography
,
16
,
269
275
.
van Ede
,
F.
,
de Lange
,
F.
,
Jensen
,
O.
, &
Maris
,
E.
(
2011
).
Orienting attention to an upcoming tactile event involves a spatially and temporally specific modulation of sensorimotor alpha- and beta-band oscillations
.
Journal of Neuroscience
,
31
,
2016
2024
.
van Ede
,
F.
,
Köster
,
M.
, &
Maris
,
E.
(
2012
).
Beyond establishing involvement: Quantifying the contribution of anticipatory α- and β-band suppression to perceptual improvement with attention
.
Journal of Neurophysiology
,
108
,
2352
2362
.
Vogel
,
E. K.
, &
Machizawa
,
M. G.
(
2004
).
Neural activity predicts individual differences in visual working memory capacity
.
Nature
,
428
,
748
751
.
Wilson
,
K. D.
,
Woldorff
,
M. G.
, &
Mangun
,
G. R.
(
2005
).
Control networks and hemispheric asymmetries in parietal cortex during attentional orienting in different spatial reference frames
.
Neuroimage
,
25
,
668
683
.
Woldorff
,
M. G.
,
Hazlett
,
C. J.
,
Fichtenholtz
,
H. M.
,
Weissman
,
D. H.
,
Dale
,
A. M.
, &
Song
,
A. W.
(
2004
).
Functional parcellation of attentional control regions of the brain
.
Journal of Cognitive Neuroscience
,
16
,
149
165
.
Woolrich
,
M.
,
Hunt
,
L.
,
Groves
,
A.
, &
Barnes
,
G.
(
2011
).
MEG beamforming using Bayesian PCA for adaptive data covariance matrix regularization
.
Neuroimage
,
57
,
1466
1479
.
Worden
,
M. S.
,
Foxe
,
J. J.
,
Wang
,
N.
, &
Simpson
,
G. V.
(
2000
).
Anticipatory biasing of visuospatial attention indexed by retinotopically specific-band electroencephalography increases over occipital cortex
.
Journal of Neuroscience
,
20
,
1
6
.
Yeh
,
Y.-Y.
,
Kuo
,
B.-C.
, &
Liu
,
H.-L.
(
2007
).
The neural correlates of attention orienting in visuospatial working memory for detecting feature and conjunction changes
.
Brain Research
,
1130
,
146
157
.
Yi
,
D.-J.
,
Turk-Browne
,
N. B.
,
Chun
,
M. M.
, &
Johnson
,
M. K.
(
2008
).
When a thought equals a look: Refreshing enhances perceptual memory
.
Journal of Cognitive Neuroscience
,
20
,
1371
1380
.
Zhang
,
W.
, &
Luck
,
S. J.
(
2008
).
Discrete fixed-resolution representations in visual working memory
.
Nature
,
453
,
233
235
.