Behaviour.

To test the effect of rTMS on memory performance, we conducted a 2 (List 1 versus List 2) × 2 (DLPFC versus vertex) mixed ANOVA. There was a significant positive effect of DLPFC stimulation on memory performance (main effect rTMS, F(1,38) = 5.096, p = 0.03, η²_p = 0.118) and a significant difference between memory for the first and second lists (main effect list, F(1,38) = 17.242, p < 0.001, η²_p = 0.312). We also found a significant rTMS × LIST interaction (F(1,38) = 8.837, p = 0.005, η²_p = 0.189). Post hoc independent samples t tests revealed that the DLPFC group showed better memory performance compared to the vertex group for words presented during rTMS application (List 2, t(38) = 2.820, p = 0.008, Cohen’s d = 0.892; Fig 2D), but not for words presented before rTMS application (List 1, t(38) = 1.399, p = 0.170, Cohen’s d = 0.443; Fig 2B). Hence, the effects were specific to the application of rTMS to the left DLPFC.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Behavioural memory performance in experiment 1.

(A) Serial position curve for List 1 words. Error bars depict standard errors of the mean. (B) Raincloud plots of average memory performance for List 1 words across all blocks with paired boxplots [16]. Coloured area within the box plots indicate the standard error, while the circles depict individual data points. (C) Serial position curve for List 2 words. Error bars depict standard errors of the mean. (D) Memory performance for List 2 words. (E) Difference in average memory performance between the DLPFC and vertex condition for each list (List 2 = Stimulation). The data and scripts used to generate this figure can be found at https://osf.io/dyxjv/. DLPFC, dorsolateral prefrontal cortex; n.s., not significant; TMS, transcranial magnetic stimulation.

https://doi.org/10.1371/journal.pbio.3001363.g002

In an exploratory follow-up ANOVA, we investigated a possible effect of rTMS on serial position to assess whether left DLPFC stimulation affected the likelihood of recalling a word as a function of its list position [17]. Analysis of serial position curves revealed a significant LIST × POSITION × rTMS interaction (F(9,342) = 2.435, p = 0.011, η²_p = 0.06). To unpack this 3-way ANOVA, we calculated two 2-way ANOVAs for each list separately. These ANOVAs showed a significant POSITION × rTMS interaction for List 1 (F(9,342) = 2.703, p = 0.005, η²_p = 0.066), but no significant POSITION × rTMS interaction for List 2 (F(9,342) = 0.893, p = 0.532, η²_p = 0.023; Fig 2C). The significant interaction in List 1 was due to enhanced recall rates for late position words in the DLPFC group compared to the vertex group (see Fig 2A). These results suggest that online rTMS to the left DLPFC equally increased memory performance in List 2 regardless of position, whereas for List 1, only late position words benefitted from stimulation.

Participants.

The data reported here were collected as part of a larger study (reported in [14] experiment 2). A total of 48 healthy human participants were tested, and participants were randomly assigned to one of the 2 stimulation conditions. After artefact rejection and inspection of the EEG data, 40 participants remained in the sample, resulting in 20 participants per group (DLPFC group: mean age = 21.7, range 18 to 26, 8 males; vertex group: mean age = 22.3, range 18 to 27, 6 males). All participants were right handed, had normal or corrected-to-normal vision, reported no history of neurological disease or brain injury, and were screened for contraindications against rTMS [10]. Written informed consent was acquired from each participant prior to the experiment. The study was approved by the ethics committee of the University of Konstanz (Project ID: “How the synchronized brain forms enduring memories”) and conducted in accordance with the principles expressed in the Declaration of Helsinki.

Task and stimulus material.

The stimulus material consisted of 240 nouns derived from the MRC Psycholinguistic Database [57]. The material was translated into German and divided into 24 lists of 10 words. The lists were matched according to word frequency, number of letters, number of syllables, concreteness, and imageability [14]. The presentation of the lists was counterbalanced across participants. Each list was presented equally often across 4 conditions (Forget List 1, Forget List 2, Remember List 1, and Remember List 2). The data were collected as part of a study that focussed on the causal involvement of the left DLPFC in voluntary forgetting (reported in [14]; experiment 2). Participants performed 12 encoding-recall runs. In each run, participants were presented with 2 lists of 10 words. After having studied the first 10 words, a cue was presented for 5 s, prompting participants to either forget the previously studied words or to continue remembering this list. The second list of 10 words was always followed by a remember cue. For this study, only the 6 remember runs, i.e., runs in which the first and second lists had to be remembered, are included in the analysis. The words were presented in a randomised order one at a time for 2.5 s, with a variable interstimulus interval of 1.5 to 2.5 s (during which a fixation cross was shown). After a short distractor task of 2 min (counting backwards in steps of 3 from a random number), participants were asked to freely recall as many words from this run as possible in any order. Participants’ reponses were recorded manually by the experimenter outside of the EEG room.

rTMS.

During encoding of List 2, 45 pulses of 1 Hz rTMS were applied at 90% resting motor threshold. One group of participants received rTMS to the left DLPFC, while the control group received rTMS to the vertex. The vertex was chosen as a control site, as it has been shown to not have any wide-ranging network effects for 1 Hz stimulation [51]. The rTMS pulses and stimulus presentation were not synchronised by the experiment. Due to the nature the ISI being randomly chosen as a multiple of 0.25 s, there appeared to be a weak 4 Hz rhythm present (see S3 Fig). However, this bias did not systematically differ between stimulation conditions and therefore cannot explain the observed behavioural effects. rTMS was delivered using a Magstim Rapid2 stimulator with a figure-of-eight air filmed cooled coil (magstim; www.magstim.com). Prior to the main experiment, individual T1-weighted MRI scans were acquired with a 1.5T Philips scanner. In order to assure that the exact regions of interest were targeted, the stimulation was guided by a neuronavigation system (ANT-Visor; www.ant-neuro.com). Individual MRI scans were coregistered with the position of the rTMS coil, and the precise targeting of the stimulation sites was monitored throughout the experiment. The coil was approximately angled 45° from the midline axis of the participant’s head with the handle pointing backwards and laterally. The MNI coordinates for DLPFC stimulation were x = −45, y = 6, z = 39 [14].

EEG recording and preprocessing.

EEG was recorded throughout the task from 128 electrodes in an equidistant montage (ANT; www.ant-neuro.com). Participants were seated in a shielded room, and data were recorded with a DC amplifier (ANT) at a sampling rate of 2,048 Hz; data were offline re-referenced to average reference. Individual electrode positions were digitised at the beginning of the experiment (Xsensor, ANT). EEG data were preprocessed and analysed using Fieldtrip [58]. Due to excessive artefacts in the EEG during rTMS [59], List 1 (no rTMS) and List 2 (during rTMS) trials were preprocessed separately. Preprocessing of rTMS-EEG data followed the guidelines and procedure outlined by Herring and colleagues [60] and described on the Fieldtrip tutorial website (https://www.fieldtriptoolbox.org/tutorial/tms-eeg/). EEG data were first cut into segments of −0.9 s to 0.9 s around the rTMS pulses. Data were visually inspected, and data around the rTMS artefacts resulting from ringing and recharging of the stimulator were removed from further analysis, as these can impact the performance of the subsequent preprocessing steps. The epoched data were subsequently subjected to an independent component analysis (runICA). This allowed the removal of rTMS-related artefacts, eye blink, eye movement, and other remaining artefacts. Any missing data were interpolated with a cubic interpolation algorithm to avoid artificially induced artefacts in the data. The cleaned data epoched around word onset (−2 s to 4 s) were then downsampled to 500 Hz. A low-pass filter (40 Hz cutoff) was applied, and the data were visually inspected for remaining artefacts. Missing and rejected channels were interpolated (mastoids were removed resulting in 126 channels). For trials without rTMS (List 1), data were epoched −2 s to 4 s around the onset of the word, downsampled to 500 Hz, and low-pass filtered (40 Hz cutoff). After visually inspecting the data for artefacts, an ICA was applied in order to identify and remove ocular and muscle artefacts. The cleaned data were again visually inspected.

Behavioural analysis.

In order to assess the effect of stimulation on recall performance, a mixed ANOVA with the within-subjects factor LIST (List 1 and List 2) and the between-subjects factor rTMS (DLPFC and vertex) was performed. We further tested whether DLPFC stimulation influenced the likelihood of recalling words as a function on a words’ list position. To this end, serial position curves were calculated [17]. For every participant at every list position, we coded whether a word was later recalled (1) or not (0). This was done for all 6 encoding-recall runs and subsequently averaged for every participant over the 6 runs. These data were then subjected to a 2 (DLPFC versus vertex) × 10 (position in list) × 2 (List 1 or List 2) ANOVA.

EEG analysis.

EEG data (−1.5 s to 3 s) were subjected to a time frequency decomposition (2 to 35 Hz in steps of 1 Hz) using Morlet wavelets (width 7) and z-transformed per trial across time for each participant, within each stimulation condition, to enable analysis of post- as well as prestimulus activity [60]. Since we analysed the data in the context of an increased memory performance, which, according to the sync/desync hypothesis, should be characterised by cortical alpha/beta power decreases, only negative clusters were expected [3,61]. Therefore, data from the DLPFC and vertex group were subjected to a 1-tailed cluster-based permutation test, averaged over beta (13 to 30 Hz) and the poststimulus time window of interest (0 to 1 s). Alpha values were set to 0.05. All further analyses were conducted on the electrode sites identified as showing significant differences in beta between the 2 conditions.

To ensure that any observed effects were specific to stimulation trials, an additional analysis was performed comparing the List 1 and List 2 trials for the DLPFC and vertex groups, respectively, in a time window from −1 s to 1.95 s relative to stimulus onset. This time window was split into 6 nonoverlapping time bins. The data were then analysed using a TIME (time bins) × LIST (List 1 versus List 2) × GROUP (DLPFC versus vertex) ANOVA accompanied by post hoc independent samples t tests (see S1 Text for a control analysis regarding potential trial imbalances).

The properties of observed power changes were further investigated using the FOOOF toolbox [22]. This method uses simultaneous fitting of the aperiodic spectrum component as well as spectral peaks. For this, we analysed a 1- to 80-Hz band-pass filtered signal in the time window of interest (resulting from the time frequency analysis) and an identically sized time window before stimulus presentation. This time window was chosen to minimise any effects the filtering process might have on the frequency spectrum. The model was subsequently fit using a frequency range of interest of 1 to 40 Hz to optimise fits for the low-frequency (alpha and beta) bands of interest. These components were then analysed separately. We performed a 2 × 2 mixed repeated measure ANOVA (Pre versus Post (2) word presentation × DLPFC versus vertex (2) stimulation, for each component (the aperiodic exponent, the offset and the periodic peak power). Additionally, a control analysis was performed to ensure that there were no differences in model fits or residuals between the different FOOOF models that could alternatively explain any of the effects presented in this study (see S1 Table).

Participants.

A total of 24 healthy human participants took part in this experiment (mean age = 19.04, range 18 to 28, 5 male). All participants were right handed, had normal or corrected-to-normal vision, reported no history of neurological disease or brain injury, and were screened against contraindications against rTMS [10]. Written informed consent was acquired from each participant prior to the experiment, and participants were fully debriefed at the end. The protocol was approved by the ethics committee of the University of Birmingham (Project ID: ERN_14–0651) and conducted in accordance with the principles expressed in the Declaration of Helsinki.

Task and stimulus material.

In this study, the participants as well as the experimenter interacting with the participants were blind towards the hypotheses.

A total of 240 nouns were derived from the MRC Psycholinguistic Database [57] and divided into 24 lists of 10 words. As in experiment 1, the lists were matched according to word frequency, number of letters, number of syllables, concreteness, and imageability [14]. The presentation of the lists was counterbalanced across participants so that each list was used equally often across 8 conditions (DLPFC–Forget List 1, DLPFC–Forget List 2, DLPFC–Remember List 1, DLPFC–Remember List 2, vertex–Forget List 1, vertex–vertex List 2, vertex Remember List 1, and vertex–Remember List 2). Participants performed 12 encoding-recall runs, split by stimulation condition. Whether the 6 DLPFC runs or the 6 vertex runs were conducted first was counterbalanced across participants. The task was the same as in experiment 1. For this study, only the 3 remember runs per stimulation condition are included in the analysis. Participants’ responses were recorded manually inside the testing room.

rTMS.

The same stimulation parameters were used as in experiment 1. However, in this experiment, participants received both DLPFC and vertex stimulation in a blocked manner. The stimulation was delivered using a Magstim Rapid stimulator with a figure-of-eight coil (magstim; www.magstim.com). Prior to the main experiment, individual T1-weighted MRI scans were acquired using a 3T Philips Achieva MRI scanner. In order to assure precise stimulation, individual MRI scans were coregistered with the position of the rTMS coil, and the stimulation was guided by a neuronavigation system (Brainsight; Rogue Resolutions; https://www.rogue-resolutions.com). The coil was held in place manually, and the precision of the stimulation was monitored throughout the experiment. The same MNI coordinates as in experiment 1 were used.

Temporal clustering.

To investigate whether the observed memory effects could be explained due to contextual effects resulting from the stimulation, we calculated temporal clustering scores per participant for each respective list and stimulation condition (procedure is based on [62]). This procedure can be summarised with the formula: where the observed distance was defined as the absolute difference between the observed recall position and the position during encoding for each subsequently recalled item (R). For example, if a participant recalls an item in the third position and subsequently recalls an item in the fifth position, the observed distance would be 2. The expected difference is the distance value that would be expected during optimal temporal clustering (Expected difference = 1; e.g., one would expect the fourth item to be recalled following the third item yielding a difference of 1).

This yielded a temporal clustering value for each list and condition per participant (see S2 Table for values per condition). The items in List 1 were coded with the numbers 1 to 10, while items belonging to the second list were coded with numbers 11 to 20. These were then directly compared to each other using a 2 × 2 repeated measure ANOVA (List × Stimulation Condition).

Meta-analysis.

In order to combine the effect of stimulation over the 2 studies, a cumulative meta-analysis of the stimulation effect for the List 1 and List 2 items was performed using the R-package metafor [26]. The analysis was performed by computing effect sizes (Hedge’s g) for the individual relevant t tests (independent and dependent for study 1 and 2, respectively), which were then used to run a weighted fixed-effect meta-analysis [26,63].