Abstract
Control over internal representations requires the prioritization of relevant information and suppression of irrelevant information. The frontoparietal network exhibits prominent neural oscillations during these distinct cognitive processes. Yet, the causal role of this network-scale activity is unclear. Here, we targeted theta-frequency frontoparietal coherence and dynamic alpha oscillations in the posterior parietal cortex using online rhythmic transcranial magnetic stimulation (TMS) in women and men while they prioritized or suppressed internally maintained working memory (WM) representations. Using concurrent high-density EEG, we provided evidence that we acutely drove the targeted neural oscillation and TMS improved WM capacity only when the evoked activity corresponded with the desired cognitive process. To suppress an internal representation, we increased the amplitude of lateralized alpha oscillations in the posterior parietal cortex contralateral to the irrelevant visual field. For prioritization, we found that TMS to the prefrontal cortex increased theta-frequency connectivity in the prefrontoparietal network contralateral to the relevant visual field. To understand the spatial specificity of these effects, we administered the WM task to participants with implanted electrodes. We found that theta connectivity during prioritization was directed from the lateral prefrontal to the superior posterior parietal cortex. Together, these findings provide causal evidence in support of a model where a frontoparietal theta network prioritizes internally maintained representations and alpha oscillations in the posterior parietal cortex suppress irrelevant representations.
- alpha dynamics
- EEG
- internal representations
- neural oscillations
- theta connectivity
- transcranial magnetic stimulation
- working memory
Significance Statement
Control over internal representations requires the prioritization of relevant information and suppression of irrelevant information. We targeted theta-frequency frontoparietal coherence and lateralized alpha oscillations in the posterior parietal cortex using online rhythmic transcranial magnetic stimulation (TMS) in human participants while they prioritized or suppressed internally maintained working memory (WM) representations. For suppression, alpha-TMS to the posterior parietal cortex increased the amplitude of lateralized alpha oscillations contralateral to the irrelevant visual field. For prioritization, theta-TMS to the prefrontal cortex increased theta-frequency connectivity in the prefrontoparietal network contralateral to the relevant visual field. In a separate dataset of patients with implanted electrodes, we demonstrate anatomical specificity in that theta connectivity was directed from the lateral prefrontal to the posterior parietal cortex.
Introduction
The manipulation of internal representations is essential to higher-order cognition (Miller and Cohen, 2001; Gazzaley and Nobre, 2012). After the selective admittance of information from the environment (Posner, 1980; Jensen and Mazaheri, 2010), acquired information must be prioritized, integrated, sequenced, and transformed (Chatham et al., 2014; Chatham and Badre, 2015; Wallis et al., 2015). Understanding the mechanisms by which internal representations are manipulated will enable more accurate decoding of cognition (Kwak and Curtis, 2022; Wan et al., 2022) and contribute to the development of techniques using noninvasive brain stimulation to modulate cognitive control (Bergmann et al., 2016; Romei et al., 2016; Riddle and Frohlich, 2021). Experimentally, a retrospective cue (retro-cue) in the delay period of a working memory (WM) task, which indicates a subset of items that are more likely to be tested, engages the prioritization of relevant and the suppression of irrelevant information (Wallis et al., 2015; Souza and Oberauer, 2016). Previous studies using a retro-cue WM task suggest distinct mechanisms for prioritization via frontal and parietal theta oscillations (4–7 Hz; Wallis et al., 2015; Albouy et al., 2017; Riddle et al., 2020) and suppression via the amplitude of posterior parietal alpha oscillations (8–12 Hz; Wallis et al., 2015; Schneider et al., 2016; Poch et al., 2017, 2018; de Vries et al., 2018; Riddle et al., 2020). However, noninvasive brain stimulation with concurrent electrophysiology is required to establish causally tested mechanisms for how internal representations are prioritized and suppressed.
To address this gap in the literature, we delivered rhythmic TMS at the precise time at which prioritization and suppression processes were recruited in a retro-cue WM task that enabled us to establish the frequency-specific contributions of the frontal and parietal cortexes (Romei et al., 2011; Thut et al., 2011; Hanslmayr et al., 2014; Riddle et al., 2019, 2020) with concurrent electroencephalogram (EEG) to understand how frontoparietal theta connectivity and alpha amplitude in the posterior parietal cortex correspond to changes in behavioral performance (Thut et al., 2011; Hanslmayr et al., 2014; Albouy et al., 2017). In addition, we investigated theta frequency directed functional connectivity in the frontoparietal network during the same retro-cue WM task using invasive EEG with greater anatomical resolution in patients with electrodes that were implanted for clinical reasons. Together, these experiments provided causal evidence for frontoparietal theta connectivity in the prioritization of relevant internal representations with information flowing from the lateral prefrontal to the posterior parietal cortex and lateralized alpha amplitude for the suppression of irrelevant internal representations in the regions of the posterior parietal cortex at which they were maintained.
Materials and Methods
Experiment 1—concurrent rhythmic TMS and EEG
Experimental design
The experiment reported here was approved by the Institutional Review Board at the University of North Carolina at Chapel Hill (UNC). Participants were recruited from the Raleigh-Durham-Chapel Hill community and provided written consent before participation. Based on our previous experiment that delivered rhythmic TMS in theta and alpha frequencies during the processing of a retro-cue in the delay of a WM task (Riddle et al., 2020), we performed a power analysis to determine our recruitment goal. The previous experiment (N = 20) found a significant interaction between TMS frequency and TMS site on WM capacity (k) with an effect size of 0.532 (Cohen's d). Using the G*power toolbox, a power estimate was performed as a t test between dependent sample estimates with this previous effect size (two-tailed, p < 0.05), which revealed that 48 participants were required to reach a power of 1-β >95%. Thus, data were collected until at least 48 participants completed the experiment.
The hypotheses for the study were preregistered on the Open Science Framework (https://osf.io/vxdkb), and the study was registered as a randomized clinical trial in the National Clinical Trials registry (NCT03828734). The experiment comprised four sessions. In the initial session, participants performed the retro-cue WM task at three different memory loads to calibrate the difficulty of the task and to exclude participants who did not utilize the retro-cue to improve their task performance. A total of 77 participants were enrolled in the experiment and took part in the initial behavioral screening session. Our behavioral analysis of the first session was run with all 77 participants. Of the 77 participants who completed the first session, 16 participants did not demonstrate a benefit to their accuracy from the retro-cue for any load, defined as an improvement by at least 5% accuracy between retro-cue and neutral-cue trials. Thus, 61 participants were invited to participate in the second session in which high-density EEG was recorded during the performance of the task with a fixed load determined to be above chance with a retro-cue effect on accuracy. Three participants dropped out of the study prior to the second session due to the time commitment. The baseline EEG analysis consisted of 58 participants. From these EEG data, the individualized peak frequency of theta and alpha neural oscillations were localized for each participant and used for stimulation in their subsequent sessions.
In the third and fourth sessions of the experiment, rhythmic TMS was delivered to either the anterior middle frontal gyrus (aMFG) or to the inferior intraparietal sulcus (iIPS) in a randomized and counterbalanced order. Immediately following the presentation of the retro-cue, five pulses of TMS were delivered in one of four patterns that were randomized and intermixed within each block. On every trial, stimulation was delivered in individual theta frequency (ITF), individualized alpha frequency (IAF), arrhythmic pattern matched for the duration and number of pulses of ITF, or an arrhythmic pattern matched for IAF. Of the 58 participants who were invited to participate in the TMS sessions, five participants discontinued participation between sessions due to the time commitment, and four participants did not complete the TMS sessions due to discomfort with TMS. Thus, 49 participants completed the full experiment. After data collection, it was discovered that the sessions of two participants contained technical errors: incorrect EEG sampling rate without the fast-recovery setting in one session and an incorrect stimulation frequency for another session. These data were removed from our analysis such that the final dataset for concurrent TMS-EEG was run with 47 participants (33 female, ages 18–30, 21.2 ± 3.2).
Retro-cue WM task
At every session, participants performed a change detection WM task with retrospective cues during the delay period (Fig. 1A; Riddle et al., 2020). The task was presented using Psychtoolbox-3 (Brainard, 1997; Kleiner et al., 2007) running on MATLAB version 2018A (MathWorks). Participants were seated comfortably with their chin in a height-adjustable chinrest. The LCD screen was set to a refresh rate of 120 Hz and positioned at an eye level of 70 cm from the eye. The task was synchronized with the EEG recording using either a parallel or serial port, and the timing of the digital triggers was verified with a photosensitive diode that was recorded in the EEG amplifier.
Retro-cues increased posterior alpha lateralization and frontoparietal theta connectivity. A, The task comprised three epochs. In the encoding epoch, participants encoded two arrays of colored squares presented in each visual hemifield. After a short delay period, a retro- or neutral-cue was presented. The retro-cue was 100% predictive of the location of the probe, whereas the neutral-cue provided no additional information. After a second delay period, a probe was presented and the participant responded whether the probe matched the encoding array. B, Spectral activity in left frontal electrodes across the task for all conditions revealed increased theta amplitude followed by decreased alpha amplitude for each epoch. The dashed rectangles indicate time–frequency regions for alpha amplitude and theta-connectivity analysis. The cue epoch was analyzed and is outlined for emphasis. C, Alpha amplitude increased contralateral to the visual field from which irrelevant information was encoded: retro-cues to the left versus right visual field. D, Theta-frequency functional connectivity, measured with weighted phase lag index (wPLI), increased between frontal and parietal electrodes for retro-cues relative to uninformative neutral-cues. The rounded rectangle depicts the seed from which scalp functional connectivity was calculated. N = 58. The gray outline is significant clusters at p < 0.05 with cluster correction. The dots are p < 0.05 with cluster correction.
On each trial, two arrays of colored squares were presented for 500 ms in a circle 5° visual angle (dva) from a central fixation cross. We used a fixed stimulus set of nine colors equally spaced around a color wheel that was matched for saturation and brightness. The colored squares could appear at 12 locations equally spaced around the circle such that there were 6 possible locations in each visual hemifield. The two memory arrays were randomly placed contiguously in each visual hemifield and were never contiguous to each other within the encoding display with at least one empty location between them. The left and right arrays included an equal number of stimuli (two, three, or four items per array) on each trial. Following a delay period of 1,000 ms, a retro-cue (50% of trials) or a neutral cue (50% of trials) was presented at fixation for 100 ms. Retro-cues were an arrow presented at fixation with a width of 3 dva that pointed toward either the left or right visual field. Participants were instructed to “remember” the array from the cued visual field and to “forget” the array from the other visual field. The retro-cue was fully predictive of the array that would be tested at the probe. Neutral cues were a double-sided arrow that provided no predictive information about the upcoming memory probe. Participants were instructed that a neutral cue indicated that they could be tested for their memory of either array.
The task specifications were identical to that of our previous experiment (Riddle et al., 2020) with the exception that the second delay period was increased from 1,000 to 2,000 ms to increase the time poststimulation to measure the aftereffects of rhythmic TMS on neural oscillations. After a second delay period of 2,000 ms, participants were probed with a test array on either the left or right side of fixation, in the exact location as the original encoding display. Participants indicated whether the entire test array matched the memory array or whether at least one color had changed in color or position. On nonmatch trials, the test array included a novel color not in either memory array for half of the trials or a color from the memory array not tested. Participants were given 2,000 ms to indicate whether the probe array matched or did not match the encoding array by making a button press with the index finger of their left or right hand. The mapping between left and right index fingers to match or nonmatch was kept consistent throughout the experiment and was randomized for each participant.
The screening session was used to exclude participants who did not use the retro-cue to improve their WM accuracy and to titrate the difficulty of the task to the maximum load (2, 3, or 4) in which participants exhibited a retro-cue benefit: 5% or greater difference in WM accuracy for retro-cue trials versus neutral-cue trials. Then, the WM load was fixed for all subsequent sessions. In addition, WM load was used as a covariate in RM-ANOVAs for the impact of stimulation on behavior. Analysis of behavior investigated neural correlates of WM capacity and the impact of rhythmic TMS on WM capacity. After removing trials where the participant responded faster than 100 ms, treating misses as incorrect, Pashler's WM capacity was calculated for each condition as the WM load multiplied by hit rate minus the false alarm rate normalized by the correct rejection rate, where a hit was the detection of a change from encoding (Rouder et al., 2011).
Rhythmic TMS and neuronavigation
Rhythmic TMS was delivered using a MagVenture X100 with MagOption (MagVenture). During the third session, the resting motor threshold of the participant was calculated using single-pulse TMS with a MagVenture C-B60 butterfly coil using motor-evoked potentials. For rhythmic TMS, stimulation was delivered at 120% of the resting motor threshold using a MagVenture Cool-B65 liquid-cooled butterfly coil. All stimulation was delivered through a low-profile passive MicroCel 128-channel EEG net on the scalp with a NetAmps400 amplifier (Electrical Geodesics).
Five pulses of rhythmic TMS were applied on each of 32 trials for 10 blocks at IAF, ITF, arrhythmic IAF, or arrhythmic ITF. EEG data from the second session was used to personalize the frequency of stimulation for each participant. IAF was selected as the frequency with maximum amplitude between 8 and 12 in 0.25 Hz increments for the contrast of retro-cues to the left minus the right in parietal occipital electrodes (PO3 and surrounding) from 500 to 900 ms after cue onset (10.1 ± 0.9 Hz), and ITF was the frequency with maximum amplitude between 4 and 8 Hz for the contrast of retro-cues minus neutral cues for the frontal-midline (Fz and surrounding) from 100 to 500 ms after cue onset (5.8 ± 0.5 Hz). We used Fz for localization of ITF because bilateral activation of homologous regions often shows peak activity in the midline of sensor space; for example, auditory evoked potentials are strongest in Cz and upon source localization are revealed to originate in the bilateral primary auditory cortex (Debener et al., 2007). Based on our previous fMRI findings showing bilateral prefrontal activation to the retro-cue (Riddle et al., 2020), we used Fz to define ITF. However, we expected to find the strongest impact of TMS on neural activity in the left hemisphere and so F3 was used for analysis of the impact of TMS on the EEG. IAF and ITF were manually inspected to ensure that they were at least 2 Hz apart for each participant and that the peak frequency in this range reflected a cluster centered in the window of interest; otherwise, canonical stimulation frequencies of 6 Hz for theta and 10 Hz for alpha were used. In only 2 out of the 49 participants, the canonical theta and alpha frequency were used after a failure in localization. Arrhythmic stimulation was matched to the duration of the individualized burst such that the first and last pulse occurred at the same time. The three middle pulses were randomly selected until two conditions were met: no two pulses were within 20 ms and at least one interpulse interval needed to be 20 ms different from the individualized interval. For all TMS conditions, the pulse sequence began 100 ms after the onset of the retro-cue.
Using 3D stereotaxic tracking (Localite), we used an automated method in the Localite software for normalizing the scalp of the individual into the Montreal Neurological Institute (MNI) space. In brief, many scalp coordinates are acquired to construct a surface model of the scalp, and a warp from MNI space is estimated. This warp was then applied to canonical spherical regions of interest that were then backnormalized into the participant's native space. The coil was positioned in the posterior-to-anterior direction over aMFG (−40, 39, 23) or iIPS (−34, −76, 26) coordinates derived from a meta-analysis of fMRI studies on retro-cue WM (Wallis et al., 2015). The left hemisphere was targeted based on previous literature demonstrating a moderate left hemisphere dominance for WM (Nee and Jonides, 2008; Wallis et al., 2015). For this reason, we focused on the left hemisphere in our analyses. The position of the coil was recorded for each burst of TMS, and any trials with a position greater than three median absolute deviations were discarded from the final analysis (5.5% of trials on average).
Prior to the start of the task blocks, participants wore EEG-compatible headphones, and white noise was played for the duration of the experiment that involved TMS. The noise level was raised until participants reported that they could not hear a TMS pulse that was delivered into the air near the participant. Confirming that participants could not hear the clicking sound from the delivery of TMS reduced the possibility that auditory evoked potentials were the driving source of rhythmic activity in the brain.
EEG preprocessing
EEG data were preprocessed using the EEGLAB toolbox (Delorme and Makeig, 2004) and custom code with additional steps applied to the concurrent TMS-EEG data. First, the onset of each TMS pulse was identified by a deviation from the average EEG signal and verified to ensure that every pulse was found. Second, the TMS artifact was removed from every channel using a custom algorithm using a hybrid of approaches used by existing pipelines (Rogasch et al., 2017; Bertazzoli et al., 2021): (1) Perform a linear detrend of the signal. Starting with the final pulse of each burst, (2) run a 2-degree polynomial fit on 20–120 ms after pulse onset and subtract the first component that reflects the decay artifact as the amplifier recovers from the TMS pulse. (3) Replace the “artifact zone” around the TMS pulse onset, −10 ms to +20 ms, with Gaussian noise sampled from the mean and standard deviation of the data 20 ms before and after the artifact zone. (4) Add a linear trend to the artifact zone estimated between the mean of the 20 ms before and after the artifact zone. (5) Smooth around each pulse with a 5 ms moving average. Third, in three sessions, we replaced one or two channels with a spherical interpolation as these channels were so noisy that artifact subspace reconstruction in subsequent steps unnecessarily removed too much data. Fourth, the data were filtered from 1 to 50 Hz. Fifth, the data were downsampled to 200 Hz. Sixth, artifact subspace reconstruction was run with increasingly conservative parameters until 10 or fewer channels were flagged and spherically interpolated. In four sessions, this method failed, and a conservative threshold of 0.5 correlation was selected. In these sessions, there were more electrodes that exhibited above-average noise levels, and the removal of these additional electrodes likely was proportional to increased noise in these data. These individual differences reflect typical variation also seen in independent component analysis (ICA) rejection. Seventh, the data were global average rereferenced. Eighth, trials and channels were manually inspected and flagged for removal or interpolation, respectively. Invariably, this method flagged and removed channels directly under the TMS coil as these were physically jostled over the course of each session. Ninth, baseline correction was applied using the intertrial interval, and ICA was run based on the rank of the data. By manual inspection, independent components were removed that corresponded to line noise, muscle activity, movement of the eyes, or focal activity under the TMS coil. Finally, trials were removed that were flagged by neuronavigation, the participant responded quicker than 100 ms, or the participant did not make a response.
Spectral analysis and functional connectivity
Five-cycle Morlet wavelet convolution was used to extract instantaneous amplitude for 150 frequencies from 1 to 50 Hz in a distribution that mimicked the aperiodic background noise of the brain (Donoghue et al., 2020). The time–frequency spectrogram of amplitude values for each condition was normalized by subtracting, then dividing, by the mean of baseline activity from the intertrial interval, 700–400 ms prior to encoding onset. Spectral analyses were run in three primary regions of interest defined by a central electrode averaged with its neighbors: F3 for the left frontal, PO3 for the left parietal, and PO4 for the right parietal. Based on a priori hypotheses, we extracted theta amplitude, 4–7 Hz; soon after the cue, 100–500 ms; and alpha amplitude, 8–12 Hz, at a later window, 500–900 ms. These a priori time-frequency regions of interest were selected based on previous EEG studies: Wallis et al. (2015) found lateralized alpha power as a function of the cued visual field was significantly modulated from approximately 500 to 900 ms following the retro-cue. The same study localized a time-frequency cluster in the theta range in aMFG as a function of retro-cue versus neutral cue ranging from approximately 100 to 500 ms following the retro-cue (Wallis et al., 2015). For topographic analyses, we applied a z-transformation across all scalp electrodes to normalize across participants prior to statistical testing. We hypothesized that alpha lateralization—the difference between the left parietal versus the right parietal cortex for retro-cues left versus right—would be modulated by the retro-cues. When analyzing the impact of rhythmic TMS on neural activity, the time course IAF and ITF were calculated by averaging across estimates from plus to minus 1 Hz in the regions targeted by rhythmic TMS.
Functional connectivity was estimated using the weight phase lag index (wPLI), an estimate that accounts for the confounding effect of volume conduction (Vinck et al., 2011). We extracted the average theta and alpha phase by band-filtering and applied the Hilbert transformation to extract the instantaneous phase. Phase values were concatenated across trials of each condition for the window of 100–900 ms postcue. The wPLI from the left parietal to each other electrode on the scalp was estimated. In order to increase the sensitivity of this metric, we increased the time window for this analysis to encompass the full window of 100–900 ms. This decision was also based on a finding from Wallis et al., who found a change in the pattern of theta topography across the brain was strongest from 100 to 900 ms following the retro-cue, despite significant power modulation being restricted to primarily the first 500 ms following the retro-cue (Wallis et al., 2015).
Experimental design and statistical analysis
The first, behavior-only, session consisted of 10 blocks of 24 trials each with six conditions for analysis: cue (retro or neutral) and WM load (2, 3, or 4). Thus, each condition of interest included 40 trials. A two-way analysis of variance (ANOVA) with within-participant factors, cue and WM load, was run for WM capacity, accuracy, and response time. Additional conditions that were counterbalanced in each block, but were not included in the analysis, were the correct response (match or nonmatch) and the side to be remembered (left or right).
The second session with EEG consisted of 10 blocks of 24 trials each with four conditions for EEG analysis, cue (retro or neutral) and side (left or right), and one condition of interest for behavioral analysis, cue. WM load was fixed based on the first session. Thus, each condition of interest for the EEG analysis included 60 trials and 120 trials for the behavioral analysis. To understand the pattern of evoked activity, we compared activity in the left frontal electrodes to baseline across participants. We corrected for multiple comparisons using permutation-based cluster correction with an alpha threshold of 0.05 for cluster mass. Next, we analyzed the difference between retro-cues to the left versus the right in the left parietal using the multiple-comparisons correction described above. We hypothesized to find posterior alpha lateralization for retro-cues left versus right 500–900 ms after cue onset and analyzed the difference in alpha power across the scalp and corrected for multiple comparisons with a cluster size of at least 5% of the 90 analyzed scalp electrodes. We hypothesized that left frontal theta oscillations would increase for retro-cues versus neutral cues in the early time window of 100–500 ms, so we investigated the difference between retro-cues and neutral cues using permutation-based cluster correction. Next, we investigated evoked theta power from 100–500 ms after cue onset across the scalp as a function of retro- versus neutral-cues with cluster correction as described above. We conducted an individual differences analysis to correlate left parietal alpha power and right parietal alpha power for retro-cues left versus right, and frontal theta amplitude for retro-cues versus neutral-cues to WM capacity controlling for differences in overall WM load. In addition, we investigated theta and alpha functional connectivity across the scalp with a seed in left parietal electrodes for retro-cues versus neutral-cues using a multiple-comparisons cluster-correction threshold as described above.
The third and fourth sessions each consisted of 10 blocks of 32 trials each with 16 conditions of interest: side (left vs right), frequency of TMS (theta vs alpha), rhythmicity (rhythmic vs arrhythmic), and site of TMS (frontal vs parietal). Thus, there were 40 trials per condition of interest. For all analyses, the arrhythmic-matched condition was subtracted from the rhythmic condition to control for non-frequency-specific effects of TMS. We ran a three-way repeated-measures ANOVA (RM-ANOVA) for WM capacity with factors side, TMS frequency, and TMS site and used the WM load of each participant as a linear covariate. We hypothesized an interaction between TMS frequency and TMS site such that frontal theta TMS and parietal alpha TMS were beneficial to performance, but the reverse pairing was disruptive to performance. To investigate the impact of rhythmic TMS on neural activity, we performed a two-way RM-ANOVA on theta amplitude in the site of TMS with factors: TMS site by TMS frequency. We ran another RM-ANOVA for the impact of rhythmic TMS on alpha amplitude. In addition, we quantified the impact of rhythmic TMS on ITF and IAF in the left frontal and left parietal over time and estimated significance using permutation-based cluster correction over time. The impact of rhythmic TMS on alpha lateralization was also estimated and corrected for multiple comparisons with permutation-based cluster correction over time. The time range for which this analysis was run was from negative one to positive six interpulse intervals. Finally, we hypothesized that theta-TMS to the frontal cortex would increase frontoparietal theta connectivity, so we ran a comparison between theta and alpha rhythmic TMS for retro-cues to the right (contralateral to TMS target) and to the left (ipsilateral).
Experiment 2—invasive EEG
Experimental design
This experiment was approved by the institutional review board at the University of North Carolina at Chapel Hill. Participants were recruited from the epilepsy monitoring unit who were undergoing neurosurgery for the treatment of intractable epilepsy that was unrelated to this research. Written informed consent was acquired for 14 participants (8 female, ages 22–63, 34.4 ± 13.5). After fulfilling clinical obligations, participants completed the retro-cue WM task on a laptop computer running MATLAB and Psychtoolbox-3 with bimanual response plungers. One participant was not able to perform the task, and one participant was excluded from the analysis for completing only a single-task block. Invasive EEG data were synchronized to the task using digital triggers, and a photodiode was sent to the clinical monitoring system (Quantum Amplifier, Natus Medical).
Preprocessing
The invasive EEG data were processed using custom scripts and the EEGLAB toolbox in MATLAB. Channels with epileptic activity from clinical observation by the neurologist and those that were observed to display interictal discharge during the task recording were removed from the analysis. Bipolar referencing was applied between neighboring electrodes. Data were epoched by trial and baseline-corrected using the intertrial interval. Finally, the data were band-filtered from 1 to 250 Hz and notch-filtered from 58 to 61 Hz and then downsampled to 512 Hz. Spatially, the preoperative structural MRI and postoperative CT scan were used to anatomically register each electrode using an established protocol (Stolk et al., 2018) that uses the FieldTrip toolbox (Oostenveld et al., 2011). After registration, the anatomical location of each electrode was labeled for time-frequency and effective connectivity analysis (Mercier et al., 2022). Electrodes in white matter, cerebral spinal fluid, or the skull were ignored.
Time–frequency analysis in invasive EEG
The amplitude of theta (4–8 Hz) and alpha (8–12 Hz) oscillations following the retro-cue (100–500 ms and 500–900 ms, respectively) was quantified for each bipolar electrode pair in invasive EEG and baselined corrected relative to −700 to −300 ms relative to the onset of encoding. These a priori time–frequency windows were selected to match the time windows used in the EEG analysis. Statistical comparisons pooled electrodes based on anatomical labels for the lateral prefrontal [MFG and superior precentral sulcus (sPCS)] and posterior parietal cortexes [superior parietal lobule (SPL) and superior intraparietal sulcus (IPS)]. In addition, we quantified high gamma activity (100–200 Hz from 100 to 900 ms) as this signal is considered a proxy for neural activity (Crone et al., 2006). Statistical analysis was performed using paired t tests for the contrast of retro-cue versus neutral cue and retro-cues left versus right. For lateralization analysis, data from electrodes in the right hemisphere were sign-flipped. In our dataset, there was limited coverage of the inferior parietal and occipital cortexes. As an exploratory analysis, we identified seven bipolar electrode pairs that were located within the lateral occipital (LO) cortex and repeated these analyses for descriptive purposes.
Effective connectivity and statistical analysis
Phase slope index (PSI) is a metric for calculating directed functional connectivity between two regions in a predefined frequency band (Nolte et al., 2008). We hypothesized to find increased theta-frequency effective connectivity from the lateral prefrontal cortex, MFG and sPCS, to the posterior parietal cortex, SPL and IPS, during the retro-cue. These regions are members of the frontoparietal executive control network and dorsal attention network. We analyzed PSI in the timeframe from 0.1 to 0.9 s after the cue in the theta band from 4 to 8 Hz using five-cycle Morlet wavelet convolution on 33 frequencies from a modified log distribution based on previous research (Voytek et al., 2015). We investigated the difference in effective connectivity between retro-cues and neutral cues across each electrode pair in our a priori frontoparietal regions using linear mixed-effects models with a random effects term to account for differences between participants (Mercier et al., 2022). In addition, we ran a fixed effects analysis using a paired t test. As a control, we analyzed three additional regions: the primary somatosensory cortex (postcentral gyrus), the postcentral sulcus, and the temporal parietal junction (encompassing the angular gyrus).
Results
Retro-cues increased posterior alpha lateralization and frontoparietal theta connectivity
Participants performed a retro-cue WM task (Fig. 1A), and task difficulty was titrated to the performance of the individual (N = 77; Extended Data Fig. 1-1). Two-way RM-ANOVA for cue (retro- or neutral-cue) and WM load (2, 3, or 4) were run for WM capacity, response time, and accuracy. As expected, we found a main effect of cue such that WM capacity was greater for retro-cues (mean ± SD, 1.56 ± 0.49) relative to neutral cues (1.18 ± 0.62; differences of 0.38 ± 0.63, F(1,76) = 28.03, p < 0.001, ηp2 = 0.27; Extended Data Fig. 1-1A). In addition, there was a significant main effect of WM load (F(1,76) = 6.90, p = 0.0104, ηp2 = 0.08) and interaction between cue and WM load (F(1,76) = 4.76, p = 0.032, ηp2 = 0.06) reflecting reduced WM capacity for neutral-cues as WM load increased. We found a main effect of cue for both accuracy (F(1,76) = 40.77, p < 0.001, ηp2 = 0.35; Extended Data Fig. 1-1B) and response time (F(1,76) = 583.8, p < 0.001, ηp2 = 0.88; Extended Data Fig. 1-1C) such that performance was overall improved by the retro-cue. Of the 77 participants who completed the screening session, 16 participants did not show a retro-cue benefit for any load, 22 participants were allocated to load 2 for subsequent sessions, 32 participants to load 3, and 7 participants to load 4.
Figure 1-1
Working memory performance is improved by retrospective cues. (A) Retro-cues increased WM capacity relative to neutral-cues across WM loads in the screening session (N = 77). WM capacity with a retro-cue was balanced across WM loads; n.s. is not significant. Behavioral performance was improved for (B) accuracy and (C) response time as a function of cue (dark versus light grey). Error bars are within-participant SEM. RM-ANOVA revealed a significant main effect of cue (retro- versus neutral-cue) at p < 0.05. Download Figure 1-1, TIF file.
Figure 1-2
Frequency specificity of contributions of lateralized alpha amplitude to suppression and frontal-parietal theta connectivity to prioritization. (A) In left parietal electrodes, alpha change was concentrated on the alpha band from 9-15 Hz and from 400 to 900 ms following cue onset. (B) Theta amplitude was not modulated for retro- versus neutral-cues. (C) Broadband decrease in amplitude from 2-15 Hz followed a retro-cue versus a neutral-cue from 400-800 ms. The amplitude of theta oscillations in frontal electrodes was not increased for retro- versus neutral-cues (t(57) = -0.28, p = 0.78, d = .04). (D) Similarly, topographic analysis did not reveal any change in theta amplitude as a function of cue. In addition, theta amplitude in left frontal electrodes for retro- versus neutral-cues was not predictive of the retro-cue benefit to WM capacity (partial WM load, Pearson, r(56) = -0.21, p = 0.111). Grey outline is significant clusters at p < 0.05 with cluster correction. Dots are p < 0.05 with cluster correction. Download Figure 1-2, TIF file.
In a separate session, EEG was acquired during task performance (N = 58). Our analysis of evoked spectral amplitude across task conditions revealed an increase in theta oscillations from 100 to 900 ms after the presentation of the cue and decreased alpha oscillations from 500 to 900 ms (Fig. 1B). Alpha amplitude was significantly lateralized based on the direction of the retro-cue (t(57) = 4.53, p = 0.00003, d = 0.60) such that alpha amplitude was increased in left parietal electrodes for retro-cues to the left versus right visual field (t(57) = 4.31, p = 0.00007, d = 0.57) and decreased in right parietal electrodes (t(57) = −3.62, p = 0.00062, d = 0.48). Topographic analysis confirmed the expected pattern that alpha lateralization was concentrated around left and right parietal electrodes (Fig. 1C). Spectral analysis of left parietal electrodes showed that lateralization was specific to the alpha band from 9 to 15 Hz from 400 to 900 ms following cue onset (Extended Data Fig. 1-2A). In an individual differences analysis, we found a significant positive relationship between alpha lateralization and the benefit of the retro-cue to WM capacity (partial correlation for WM load, Pearson’s, r(56) = 0.30, p = 0.020). Post hoc analysis revealed that the effect was consistent in the left (r(56) = 0.27, p = 0.045) and right (r(56) = −0.26, p = 0.049) parietal electrodes. These findings are consistent with the role of alpha amplitude in suppressing irrelevant internal representations in contralateral visual processing regions (Jensen and Mazaheri, 2010).
In addition, we found increased theta-frequency functional connectivity between frontal and parietal electrodes for retro- versus neutral-cues (100–900 ms postcue, t(57) = 3.55, p = 0.001, d = 0.47). Topographic analysis revealed that this effect was spatially specific to the left frontoparietal network (Fig. 1D). By comparison, there was no change in alpha-frequency functional connectivity with cues (t(57) = −0.82, p = 0.414, d = 0.11; Extended Data Fig. 1-2B). Critically, theta amplitude was not modulated by the retro-cue (Extended Data Fig. 1-2C,D). These findings suggest a specific role for theta connectivity within the frontoparietal network during prioritization.
Rhythmic TMS improved WM capacity in a frequency and site-specific manner
In separate sessions, TMS was delivered to left anterior MFG, frontal, or left IIPS, parietal, in a counterbalanced and randomized order (Fig. 2A). Rhythmic TMS began 100 ms after onset of the retro-cue in either ITF, IAF, or an arrhythmic burst that was matched for duration and number of pulses but with a random interpulse interval (Fig. 2B). RM-ANOVA revealed an interaction between TMS frequency and TMS site (N = 47; F(1,44) = 4.36, p = 0.043, ηp2 = 0.09; Table 1) such that frontal theta and parietal alpha-TMS increased WM capacity relative to the reverse pairing (Fig. 2C). Post hoc analysis revealed a numerically greater impact of mismatched TMS to disrupt performance relative to condition-matched arrhythmic TMS (t(46) = −1.59, p = 0.12, d = 0.23) relative to the improvement from matched TMS (t(46) = 0.58, p = 0.59, d = 0.08) although neither effect was significant. In addition, we hypothesized that theta-TMS to the left frontal while prioritizing internal representations encoding the contralateral visual field (retro-cues to the right) and alpha-TMS to the left parietal while suppressing the contralateral visual field (retro-cues to the left) would improve WM capacity. We found a trend-level interaction between TMS frequency and TMS site in support of this hypothesis (F(1,44) = 3.57, p = 0.065, ηp2 = 0.08). Analysis of the control conditions to these predicted relationships, that is, prioritizing or suppressing the ipsilateral visual field (frontal TMS with retro-cues to the left and parietal TMS with retro-cues to the right), found no interaction between TMS frequency and TMS site (F(1,44) = 1.35, p = 0.252, ηp2 = 0.03). Thus, our behavioral findings suggest that when rhythmic TMS matched the predicted endogenous activity related to prioritization or suppression, TMS provided a benefit to WM capacity.
Rhythmic TMS improved WM capacity when matched to endogenous activity. A, TMS was targeted to the aMFG and iIPS. B, Rhythmic TMS was delivered in individual theta frequency (ITF), individual alpha frequency (IAF), and two arrhythmic conditions matched for the duration and number of pulses but with randomized interpulse intervals. C, Rhythmic TMS matched to the preferred oscillation of the targeted region (green) improved WM capacity relative to the mismatched frequency (red). Error bars are within-participant SEM *p < 0.05. D, Raw EEG trace with artifacts from rhythmic TMS before and after the preprocessing pipeline in an example trial.
Analysis of covariance of rhythmic TMS effect on WM capacity
Rhythmic TMS increased the targeted neural oscillation
After removing electrical artifacts caused by rhythmic TMS from the EEG data (Fig. 2D), we investigated the impact of rhythmic TMS on neural oscillations during the second half of the rhythmic TMS burst of five pulses. RM-ANOVA for the impact of the frequency of rhythmic TMS on frontal theta and parietal alpha oscillations revealed a TMS-site by TMS-frequency interaction (F(1,46) = 47.51, p < 0.001, ηp2 = 0.51) such that frontal theta-TMS increased the amplitude of frontal theta oscillations (t(46) = 4.80, p < 0.001, d = 0.70) with peak effects near the fourth pulse (Fig. 3A) and parietal alpha-TMS increased the amplitude of parietal alpha oscillations (t(46) = 2.77, p = 0.008, d = 0.40) with peak effects near the fifth pulse (Fig. 3B). In addition, frontal alpha-TMS decreased frontal theta amplitude (t(46) = −4.58, p < 0.001, d = 0.67) and parietal theta-TMS decreased parietal alpha amplitude (t(46) = −2.47, p = 0.017, d = 0.36). Theta and alpha oscillations displayed an inverse relationship such that rhythmic TMS produced a decrease in the off-target rhythm that temporally aligned with increased activity in the targeted rhythm.
Rhythmic TMS increased the amplitude of the targeted neural oscillation. A, Theta-TMS to the frontal cortex delivered after cue onset (vertical black line) increased theta amplitude relative to arrhythmic TMS matched for duration and number of pulses (left). The targeted region is depicted in the inset. The black rectangle highlights the time and frequency range of delivered rhythmic TMS. Lightning bolts depict the approximate time of each TMS pulse. The amplitude of ITF and IAF during frontal theta-TMS was estimated, and amplitude modulation peaked at the fourth TMS pulse (right). B, Alpha-TMS to the parietal cortex increased alpha amplitude (left) with peak effects around the fifth TMS pulse (right). The vertical gray lines depict each TMS pulse in the burst. The shaded area is a 95% confidence interval relative to arrhythmic TMS. The amplitude estimation for the panels on the right were corrected for the individualized frequency of stimulation. C, The amplitude of ITF oscillations in the stimulated region as a function of the TMS site and TMS frequency revealed a robust increase in ITF amplitude when targeting both the frontal and parietal cortexes. D, IAF amplitude in the targeted region was increased from alpha-TMS in the parietal cortex, but not the frontal cortex. Error bars are SEM. *p < 0.05 after Bonferroni’s correction. The highlighted area depicts the condition hypothesized to show maximal target engagement.
We investigated whether the impact of rhythmic TMS on neural oscillations was consistent for the frontal and parietal cortexes. RM-ANOVA revealed a main effect of TMS frequency (F(1,46) = 46.6, p < 0.001, ηp2 = 0.50) such that theta-TMS to both the frontal and parietal cortexes produced a significant increase in the amplitude of ITF oscillation (Fig. 3C). For alpha-TMS, RM-ANOVA revealed an interaction between the TMS site and TMS frequency (F(1,46) = 8.28, p = 0.005, ηp2 = 0.16) such that alpha-TMS only increased IAF amplitude in the parietal cortex (t(46) = 2.77, p = 0.008, d = 0.40), but not the frontal cortex (t(46) = −2.29, p = 0.027, d = 0.33; Fig. 3D). Thus, theta-TMS increased oscillatory amplitude in the theta band in both regions but alpha-TMS only increased alpha amplitude in the parietal cortex.
Rhythmic TMS increased alpha lateralization and theta connectivity
At baseline, the amplitude of alpha oscillations in parietal electrodes increased when the retro-cue signaled suppression of internal representations of irrelevant information from the contralateral visual field and frontoparietal theta connectivity increased for the prioritization of internal representations. Therefore, we investigated whether alpha-TMS to the parietal cortex increased alpha amplitude as a function of the retro-cue and found that alpha lateralization was significantly increased and peaked around the fourth TMS pulse (p < 0.05, permutation-based cluster correction; Fig. 4A). By contrast, alpha-TMS to the frontal cortex did not modulate posterior alpha lateralization (Fig. 4B). Furthermore, we found increased frontoparietal ITF connectivity for theta- versus alpha-TMS to the frontal cortex (t(46) = 2.09, p = 0.042, d = 0.31; Fig. 4C). There was no change for TMS to the parietal cortex (Fig. 4D). Frontoparietal ITF connectivity was not increased by theta-TMS during retro-cues that prioritized the ipsilateral visual field (t(46) = −0.29, p = 0.772, d = 0.04; Extended Data Fig. 4-1A,B). Together, these findings provide causal evidence that alpha amplitude increases for suppression of internal representation of irrelevant information from the contralateral visual field and frontoparietal theta connectivity increases for prioritization of internal representations of relevant information from the contralateral visual field.
Alpha-TMS increased alpha lateralization and theta-TMS increased frontoparietal theta connectivity. A, IAF lateralization over posterior parietal electrodes was calculated as left minus right parietal cortex for retro-cues left minus right. Alpha-TMS to the parietal cortex increased IAF lateralization around the fourth TMS pulse. The black rectangle with an asterisk depicts a significant difference from arrhythmic TMS at p < 0.05 with permutation-based cluster correction. The shaded area is within-participant SEM. B, Rhythmic TMS to the frontal cortex did not produce a significant change in IAF lateralization. C, ITF connectivity between the frontal and parietal cortices increased for rhythmic TMS versus condition-matched arrhythmic TMS with a retro-cue to the right visual field. Error bars are within SEM. *p < 0.05. D, Frontal theta-TMS, but not parietal theta-TMS, increased ITF connectivity between frontal and parietal regions. The dots are p < 0.05. The circled region depicts the electrodes with significant theta connectivity at baseline. *p < 0.05 for a priori cluster of electrodes. The rounded rectangle occludes electrodes near the parietal seed.
Figure 4-1
Left fontal theta-TMS increased left frontal-parietal theta connectivity when prioritizing information from the contralateral visual field. (C) With a retro-cue to the left visual field, there is no significant modulation in ITF connectivity from frontal theta-TMS. (D) Frontal-parietal connectivity did not increase from frontal theta-TMS when the ipsilateral visual field was prioritized. n.s. is not significant. Download Figure 4-1, TIF file.
Theta connectivity is directed from the lateral prefrontal to the posterior parietal cortex
Invasive EEG provides an opportunity to investigate the neural origination of electrical signals with greater anatomical precision directly in the cerebral cortex and to provide evidence for the direction of information flow between the frontal and parietal cortices during prioritization. In an independent sample, 14 participants with implanted electrodes performed the retro-cue WM task while invasive EEG was recorded. Two participants were excluded from the analysis: one participant did not understand the task, and one participant had a low trial count (n = 13). The participants were able to perform the WM task with accuracy greater than chance (77.8 ± 10.4%; t(11) = 9.31, p < 0.001, d = 2.69) and benefitted from the retro-cue in response time (−244 ± 24 ms; t(11) = −3.48, p = 0.005, d = 1.00) but not accuracy (9.3 ± 11.3%; t(11) = 1.29, p = 0.23, d = 0.37; Fig. 5A). Three participants did not show a benefit from the retro-cue in either response time or accuracy and, thus, were removed from analysis. Of the remaining nine participants, seven participants included electrodes in both the frontal and parietal cortices and were included in the final analysis (Fig. 5B).
Theta connectivity during prioritization is directed from the lateral prefrontal cortex to the posterior parietal cortex. A, Participants performed the retro-cue WM task and benefited from the retro-cue in response time, but not accuracy. The red lines depict participants who did not benefit from the retro-cue relative to the neutral-cue. **p < 0.01. Error bars are SEM. One participant with an average response time of 3.62 s is not pictured and did not show a retro-cue benefit. B, For those that demonstrated a retro-cue benefit in response time or accuracy, electrodes were implanted in both the frontal and parietal cortices in seven participants. Frontal electrodes are indicated in orange and parietal in blue. A, anterior; P, posterior; L, left; R, right. C, Time-frequency analysis revealed a similar pattern of activity for intracranial EEG as with EEG. Theta oscillations were increased in the lateral prefrontal cortex (orange rectangle) and posterior parietal cortex (blue rectangle) for retro-cues versus neutral-cues, but alpha oscillations were not. Nonetheless, there was a quantitative increase in alpha oscillations and a reduction in high gamma activity contralateral to irrelevant information in LO. Ipsi, ipsilateral; contra, contralateral. D, Effective connectivity in the theta-frequency band, PSI, was increased from the left lateral prefrontal to the left posterior parietal cortex for retro-cues relative to neutral-cues (highlighted in blue), but did not increase for other regions in the parietal cortex. PFC, prefrontal cortex; MFG, middle frontal gyrus; sPCS, superior precentral sulcus; SPL, superior parietal lobule; IPS, intraparietal sulcus; S1, primary somatosensory cortex; post-CS, postcentral sulcus; TPJ, temporoparietal junction; LO, lateral occipital. Error bars are a 95% confidence interval. ***p < 0.001; **p < 0.01; *p < 0.05; n.s. = not significant.
For the contrast of retro-cues versus neutral-cues, we found a significant increase in theta amplitude in the lateral prefrontal (t(109) = 2.44, p = 0.016, d = 0.23) and posterior parietal cortices (t(38) = 2.63, p = 0.012, d = 0.42; Fig. 5C). High gamma was increased for retro-cues in the posterior parietal cortex (t(38) = 2.54, p = 0.015, d = 0.41), but not the lateral prefrontal cortex (t(109) = 0.09, p = 0.929, d = 0.01). For the contrast of retro-cues to the ipsilateral visual field versus the contralateral, we did not find a significant increase in alpha activity for the posterior parietal cortex (t(38) = −2.84, p = 0.007, d = 0.45) or lateral prefrontal cortex (t(110) = −1.70, p = 0.092, d = 0.16). The posterior parietal cortex, comprising the SPL and superior IPS, showed the inverse pattern to what was found with noninvasive EEG; in fact, high gamma activity was also increased (t38) = 3.09, p = 0.004, d = 0.49) suggesting increased processing contralateral to relevant information, and not suppressed processing contralateral to irrelevant information. As an exploratory analysis to explain this pattern of findings, we investigated the impact of the direction of the retro-cue on the limited electrodes from the LO cortex (n = 7). We found the expected pattern of increased alpha activity contralateral to irrelevant information (t(6) = 1.03, p = 0.35, d = 0.39) and reduced high gamma activity (t(6) = −1.66, p = 0.15, d = 0.63) although these effects were not significant given the limited recordings from this region.
We investigated whether directed theta-frequency functional connectivity from the left lateral prefrontal cortex to the posterior parietal cortex, composed of the SPL and IPS, was increased by the retro-cue. We found increased theta connectivity for retro-cues relative to neutral-cues [random effects (RFX), F(1,202) = 65.16, p < 0.001; fixed effects (FFX), t(202) = 8.05, p < 0.001, d = 0.56) driven by an increase for retro-cues (RFX, F(1,202) = 31.04, p < 0.001; FFX, t(202) = 5.56, p < 0.001, d = 0.39] and decrease with neutral-cues (RFX, F(1,202) = 12.10, p < 0.001; FFX, t(202) = −3.47, p = 0.001, d = 0.24). In the right hemisphere, there was no significant difference in connectivity for retro-cues versus neutral-cues (RFX, F(1,93) = 1.97, p = 0.16) nor independently for either type of cue (retro-cue, F(1,93) = 1.46, p = 0.23; neutral-cue, F(1,93) = 0.04, p = 0.84). This pattern of findings was specific to the left posterior parietal cortex; there was no significant change in theta connectivity during retro-cues or neutral-cues for the left primary somatosensory cortex, postcentral sulcus, or temporal parietal junction (RFX analysis p > 0.05 for all three regions; Fig. 5D). These results support a model where the lateral prefrontal cortex transmits theta-frequency control signals to the posterior parietal cortex during prioritization.
Discussion
In a series of two experiments, we investigated the neural oscillatory mechanisms for prioritization and suppression of internal representations. Following a retro-cue in the delay period of a WM task, we found that alpha oscillations increased in the parietal cortex contralateral to information being suppressed, and frontoparietal theta connectivity increased during prioritization. Individualized rhythmic TMS delivered during the retro-cue in theta to the frontal cortex and alpha to the parietal cortex improved WM capacity relative to the mismatched frequency. Rhythmic TMS drove an increase in the amplitude of neural oscillations in the targeted frequency band and suppressed amplitude in the off-target band. Alpha-TMS to the left parietal cortex increased the lateralization of posterior alpha amplitude contralateral to information that was suppressed; whereas theta-TMS to the frontal cortex increased frontoparietal theta connectivity contralateral to information that was prioritized. Using invasive EEG, we found that directed theta-frequency connectivity increased from the left lateral prefrontal to the left posterior parietal cortex, but not other regions of the parietal cortex, during prioritization. Together, these findings provide causal evidence that the amplitude of posterior alpha oscillations supports the suppression of internal representations of irrelevant information and top-down frontoparietal theta connectivity supports the prioritization of internal representations with relevant information.
When irrelevant information is suppressed, alpha oscillations emerge in the corresponding region (Fries et al., 2001; Sauseng et al., 2009; Jensen and Mazaheri, 2010). In previous work, alpha-frequency TMS delivered to the parietal cortex contralateral to distractors increased the amplitude of alpha oscillations (Thut et al., 2011) and improved task performance (Sauseng et al., 2009). By using a retro-cue, we isolated the suppression of internal representations from the suppression of encoding information from the environment. Our findings are consistent with increased alpha amplitude during precues (Boettcher et al., 2021), which suggests a conserved mechanism in support of the gating by inhibition theory (Jensen and Mazaheri, 2010). It is possible that inhibitory gating is directed by a secondary control mechanism (Jensen, 2023). Previous studies found that while alpha oscillations increased in cortical regions processing irrelevant information, the amplitude of these oscillations scaled with the amount of relevant information processed in other regions (Noonan et al., 2018; Zhigalov and Jensen, 2020; Jensen, 2023). In our data, we found that theta connectivity after the retro-cue preceded and encompassed the time at which alpha oscillations were engaged. Thus, we speculate that prioritization via frontoparietal theta connectivity might act as a secondary control mechanism that initiates the suppression of irrelevant internal representations.
Theta oscillations are generated in the prefrontal cortex during the maintenance of internal representations in WM (Jensen and Tesche, 2002), and the magnitude of theta activity is further increased when internal representations are manipulated (Albouy et al., 2017; Berger et al., 2019). However, our previous experiment demonstrated that the amplitude of frontal theta oscillations during the delay period of a WM task linearly increased over consecutive task blocks spanning multiple hours, but glucose administration simultaneously improved WM capacity and decreased frontal theta amplitude (McFerren et al., 2021). Thus, theta amplitude might correspond with cognitive effort rather than the prioritization process itself. Alternatively, functional connectivity analyses revealed increased theta-frequency phase coherence (Sauseng et al., 2005; Payne and Kounios, 2009) and increased covariance in BOLD fluctuations (Gazzaley et al., 2004) in frontoparietal networks during the delay period of WM tasks. Researchers used noninvasive brain stimulation to increase the amplitude of frontal theta oscillations and found improved WM capacity (Alekseichuk et al., 2016; Riddle et al., 2020), but stimulation that synchronizes the frontal and parietal cortexes to promote theta connectivity also improved WM capacity (Polanía et al., 2012; Violante et al., 2017; Reinhart and Nguyen, 2019). Here, we investigated the impact of theta-TMS on theta amplitude as well as theta connectivity. While theta-TMS to the frontal and parietal cortices increased theta amplitude, only frontal theta-TMS improved WM capacity. Critically, theta-TMS to the frontal cortex, and not the parietal cortex, increased frontoparietal theta connectivity contralateral to relevant information. Thus, our findings suggest that the theta-oscillatory substrate for prioritization might correspond most directly with the strength of functional connectivity within the frontoparietal network rather than the amplitude of theta oscillations in the prefrontal cortex. Future studies are required that target the frontal and parietal cortices with in-phase or anti-phase theta-TMS in order to provide direct causal evidence for frontoparietal theta connectivity in the prioritization of internal representations.
Parietal theta-TMS did not increase frontoparietal theta connectivity, which may be explained by distinct subregions within the parietal cortex that specialize in prioritization (with high connectivity to PFC) versus suppression (local alpha generators). Previous studies found distinct roles for superior versus inferior subdivisions of the IPS: superior IPS is more resistant to distraction and encodes WM items (Xu and Chun, 2006; Bettencourt and Xu, 2016), whereas the inferior IPS is more sensitive to the presence and predictability of distractors (Gillebert et al., 2012). Sensitivity to distractors in inferior IPS is consistent with a model where alpha oscillations are dynamically engaged to prevent distraction and indeed alpha oscillations often localize to inferior IPS (Wutz et al., 2018). Furthermore, functional connectivity analyses in fMRI revealed that superior IPS displays strong functional connectivity to our targeted region in aMFG (Thomas Yeo et al., 2011). Consistent with this prediction, our analyses using invasive EEG found an increase in theta amplitude and theta connectivity for retro-cues in the superior posterior parietal cortex. We predict that the deleterious effect of theta-TMS to inferior IPS on WM capacity might not be found when targeting superior IPS, and also that theta-TMS to superior IPS may increase connectivity with PFC.
As in any scientific investigation, the current study has limitations. First of all, our preregistered behavioral analysis proposed to find an interaction between the TMS-site and TMS-frequency such that theta-TMS to the frontal cortex with contralateral relevant information and alpha-TMS to the parietal cortex with contralateral irrelevant information would improve WM performance. This analysis was trend-level in the predicted direction (two-tailed, p = 0.065; one-tailed, p = 0.033). The inverse condition pairing was defined as TMS without endogenous activity (theta-TMS to the frontal cortex with ipsilateral relevance and alpha-TMS to the parietal cortex with ipsilateral irrelevance). The overall impact of TMS on behavior was consistent with our previous experiment: in both studies, we found the strongest TMS-site by TMS-frequency interaction for all cognitive conditions, a marginally weaker interaction with endogenous activity, and no interaction without endogenous activity (Riddle et al., 2020). Analysis of neural data provides critical evidence in support of the association between theta connectivity with prioritization and lateralized alpha amplitude with suppression. Namely, frontal theta-TMS increased frontoparietal theta only while prioritizing the contralateral visual field and parietal alpha-TMS increased lateralized alpha amplitude corresponding to increased suppression of irrelevant information.
The use of concurrent EEG and TMS is challenging from a technical perspective and subject to confounding effects through peripheral stimulation (Belardinelli et al., 2019). We delivered white noise using EEG-compatible headphones to prevent auditory entrainment, but we did not control for peripheral nerve stimulation via the scalp. Additionally, participants might hear stimulation due to mechanical vibration of the skull. Nonetheless, our results are robust to peripheral entrainment due to the specificity of our TMS effects. First, TMS to the parietal cortex, but not the frontal cortex, increased the amplitude of alpha oscillations. This effect is consistent with a neural model in which TMS enhances endogenous activity, but not auditory or peripheral entrainment. Second, the impact of theta-TMS on theta connectivity and alpha-TMS on alpha power critically depended on the direction of the retro-cue to the left or right visual field, which is unlikely to be explained by peripheral factors. Third, evoked oscillatory activity gradually increased with each subsequent pulse of the TMS burst, consistent with entrainment. By comparison, electrical artifacts from TMS would be uniformly distributed across the burst. Thus, the preponderance of evidence provided by the impact of rhythmic TMS on cognition, plausible neurophysiology, and their interaction renders alternative accounts via peripheral nerve stimulation or artifact unlikely.
Within the context of the present study, frontoparietal theta connectivity during prioritization either corresponds to a top-down control signal that prevents a representation in the posterior cortex from degrading, that is, sensory recruitment (D'Esposito and Postle, 2015; Scimeca et al., 2018), or corresponds to an abstract internal representation maintained in the prefrontal cortex that is transmitted posterior during prioritization. Recent experiments found that not only can internal representations be successfully decoded from the prefrontal cortex (Parthasarathy et al., 2017; Lorenc et al., 2020) but these representations are actively transformed during prioritization (Wan et al., 2022). Neural activity in visual areas and inferior IPS recodes visual information into an abstract representation, for example, a reduced space of only behaviorally relevant features (Kwak and Curtis, 2022). Thus, both prefrontal and sensory regions exhibit patterned activity from which abstract internal representations can be decoded. As a causal test, a virtual lesion with TMS to superior IPS resulted in a change in the abstract representation of WM items (Iyer et al., 2022). Future research is required to adjudicate between a role for frontoparietal theta connectivity protecting against sensory degradation versus transmitting an abstract representation.
Footnotes
This work was supported in part by the National Institute of Mental Health of the National Institutes of Health under Award Numbers K99MH126161 to J.R. and R01MH101547 and R01MH111889 awarded to F.F. We thank Amber McFerren, Davindra Rammani, Emma Milligan, and Alexander Reardon for their help with data collection and Sankaraleengam Alagapan for his methodological contributions to the data collection with invasive EEG.
↵*J.R. and T.M. contributed equally to this work.
F.F. is the lead inventor of intellectual property filed on the topics of noninvasive brain stimulation by the University of North Carolina at Chapel Hill. F.F. is a consultant for Electromedical Products International and has received honoraria from the following entities in the last 12 months: Academic Press and Insel Spital. J.R., T.M., A.S., H.S., and E.H. have no conflict of interest.
- Correspondence should be addressed to Flavio Frohlich at flavio_frohlich{at}med.unc.edu.
