Cognitive flexibility is an important aspect relevant to daily life situations, and there is an increasing public interest to optimize these functions, for example, using (brief) meditation practices. However, the underlying neurophysiological mechanisms remain poorly understood. On the basis of theoretical considerations, both improvements and deteriorations of cognitive flexibility are possible through focused attention meditation (FAM). We investigated the effect of a brief smartphone app–based FAM on task switching using EEG methods, temporal signal decomposition, and source localization techniques (standardized low-resolution electromagnetic brain tomography). The study was conducted using a crossover study design. We show that even 15 min of FAM practicing modulates memory-based task switching, on a behavioral level and a neurophysiological level. More specifically, FAM hampers response selection and conflict resolution processes and seem to reduce cognitive resources, which are necessary to rapidly adapt to changing conditions. These effects are represented in the N2 and P3 time windows and associated with ACC. It seems that FAM increases the attention to one specific aspect, which may help to focus but carries also the risk that behavior becomes too rigid. FAM thus seems to modulate both the stimulus- and response-related aspects of conflict monitoring in ACC. Motor-related processes were not affected. The results can be explained using a cognitive control dilemma framework, suggesting that particularly alterations in background monitoring may be important to consider when explaining the effects of FAM during task switching.
Cognitive flexibility is an important aspect of executive functioning and an essential prerequisite for successful goal-directed behavior (Diamond, 2013). A standard approach to examine cognitive flexibility is task switching (Kiesel et al., 2010; Braver, Reynolds, & Donaldson, 2003; Wylie & Allport, 2000; Rogers & Monsell, 1995), and a central behavioral parameter to assess the efficacy of task switch is the so-called switch costs (SCs; Kiesel et al., 2010). They reflect increases in processing times when responses are to be switched, compared to situations in which no such switch is necessary (Monsell, 2003). These costs may reflect active reconfiguration processes (Monsell, 2003), the interference from the previous trial, retrieving goals and rules from working memory or activation, and the inhibition of irrelevant task sets (Gajewski, Hengstler, Golka, Falkenstein, & Beste, 2011; Kiesel et al., 2010; Monsell, 2003). However, because cognitive flexibility (or task switching) is central for success at school and working environments (Diamond, 2013) and is also dysfunctional in many neuropsychiatric disorders (Dajani & Uddin, 2015), there is an increasing interest in cognitive enhancement approaches aiming at proving cognitive flexibility. Regarding this, meditation-based approaches have become surprisingly popular in recent years (Davidson & Dahl, 2018; Van Dam et al., 2018). However, the neurophysiological mechanisms underlying possible beneficial effects of such brief meditation-based interventions on cognitive control and task switching in particular, remain poorly understood. In the current study, we set out to investigate the effects of a brief smartphone app–based mindfulness meditation training on neurophysiological mechanisms underlying task switching.
When considering the neurophysiological mechanisms and cognitive processes that are likely to be altered by mindfulness-based training, it is important to consider differences between practiced meditation forms (Basso, McHale, Ende, Oberlin, & Suzuki, 2019). Broadly speaking, mindfulness training describes a process where the practitioner intentionally engages the mind by bringing increased awareness to thoughts and feelings (Davidson & Dahl, 2018; Van Dam et al., 2018). Yet, a meta-analysis (Fox et al., 2016) analyzing 78 functional neuroimaging studies categorized at least three distinct types of mindfulness meditation: focused attention meditation (FAM), open-monitoring meditation (OMM), and loving-kindness/compassion meditation. During FAM, the practitioner is requested that, whenever the mind starts wandering, attention should be directed back to the specifically focused object or the body. Thus, FAM has been supposed to involve “the direction of attention to one specific object while monitoring and disengaging from extraneous thoughts or stimuli” (Fox et al., 2016; Kabat-Zinn, 2005). Studies show that experienced meditators showed better inhibitory control of (conflicting) information after extended training (Andreu, Cosmelli, Slagter, & Franken, 2018; Zanesco, King, MacLean, & Saron, 2013; Moore, Gruber, Derose, & Malinowski, 2012; Sahdra et al., 2011; Moore & Malinowski, 2009). Similar effects have been reported after a brief mindfulness intervention in novices (Colzato, van der Wel, Sellaro, & Hommel, 2016; Colzato, Sellaro, Samara, & Hommel, 2015). Because inhibitory control processes are important during task switching and cognitive flexibility (Giller, Zhang, Roessner, & Beste, 2019; Gohil, Dippel, & Beste, 2016; Zhang, Stock, Fischer, & Beste, 2016) to suppress interferences from previous tasks (Allport & Wylie, 1999), it may be hypothesized that FAM reduces SCs and thus increases performance in task switching. However, because FAM induces a focused mental state in which cognitive processes become “shielded” from new and potentially relevant information, it is also possible that FAM increases SCs and compromises task switching. The reason behind that refers to the so-called shielding–shifting dilemma (Goschke & Bolte, 2014). According to the control dilemma framework (Goschke & Bolte, 2014), goal-directed action requires that goals are maintained and shielded from incoming stimuli because these bear the risk to distract from pursuing a goal. However, individuals must also be able to disengage (switch) from a currently active task set and flexibly reconfigure responses to cope with changing requirements and rules (Goschke & Bolte, 2014). Because the latter is explicitly counteracted during FAM practices, it is also reasonable to hypothesize that FAM increases SCs and compromises task-switching performance. The effect of FAM training thus depends on which of the abovementioned processes is primarily modulated by FAM. In this respect, the current study allows examining which cognitive mechanisms are strongly modulated by FAM by testing two opposing hypotheses.
To examine the associated neurophysiological processes, we use an ERP approach in combination with source localization methods. There is a wealth of findings that task-switching processes modulate the N2 and P3 ERP components. During task switching, the N2 likely reflects response selection mechanisms related to the resolution of conflict and stimulus–response mapping processes (Zhang, Stock, & Beste, 2016; Zhang, Stock, Fischer, et al., 2016; Gajewski, Stoerig, & Falkenstein, 2008; Karayanidis, Coltheart, Michie, & Murphy, 2003). The P3 likely reflects processes related to the implementation of a switching task set (Petruo, Stock, Münchau, & Beste, 2016; Gajewski, Kleinsorge, & Falkenstein, 2010; Jamadar, Hughes, Fulham, Michie, & Karayanidis, 2010; Jost, Mayr, & Rösler, 2008; Kieffaber & Hetrick, 2005; Poulsen, Luu, Davey, & Tucker, 2005; Gehring, Bryck, Jonides, Albin, & Badre, 2003; Karayanidis et al., 2003; Barceló, Muñoz-Céspedes, Pozo, & Rubia, 2000; Lorist et al., 2000). The existing literature on the effects of FAM stresses that FAM modulates processes related to response selection. From that perspective, it is reasonable to hypothesize that particularly processes reflected by the N2 ERP component are modulated and may explain modulations of SCs at the behavioral level. Because the N2 is well known to be generated in medial frontal cortical areas (Folstein & Van Petten, 2008; van Veen & Carter, 2002), we hypothesize that FAM-related modulations in the N2 time window are associated with activity modulations in the medial frontal cortex when applying source localization methods. However, the abovementioned neurophysiological processes are known to show strong intraindividual variability, especially when demands on task switching are high, that is, when switches have to be initiated from working memory and not by using external sensory input (Wolff, Mückschel, & Beste, 2017; Gajewski et al., 2011). It is, however, important to examine FAM effects in conditions imposing such high demands on task-switching processes to avoid ceiling effects, if modulations induced by short-period FAM turn out to be moderate. To account for problems related to intraindividual variability in EEG data, we employ residue iteration decomposition (RIDE; Ouyang, Hildebrandt, Sommer, & Zhou, 2017; Ouyang, Sommer, & Zhou, 2015b). This was also done in previous studies on working-memory-based task switching (Wolff, Mückschel, Ziemssen, & Beste, 2018; Wolff et al., 2017). Crucially, the RIDE method can also be used to decompose EEG data into several component clusters with dissociable functional relevance (Ouyang et al., 2015b). This is important because the N2 ERP component is known to reflect a mixture of dissociable cognitive processes related to stimulus evaluation and responding or response selection. Both processes can coexist during extended periods during goal-directed behavior (Chmielewski, Mückschel, & Beste, 2018; Mückschel, Chmielewski, Ziemssen, & Beste, 2017; Folstein & Van Petten, 2008). The EEG data decomposition using RIDE thus allows a fine-grained analysis of neurophysiological processes modulated by FAM. The RIDE analysis results in three specific components: the S-cluster, which refers to processes of attention and perception related to the stimulus; the R-cluster, which refers to response-related processes like motor preparation and execution; and the C-cluster, which refers to stimulus evaluation and response selection and intermediates thus between the S-cluster and the R-cluster (Ouyang, Sommer, & Zhou, 2015a). The decomposition of the ERPs into these three specific RIDE components enables to further specify which specific neurophysiological process underlies the behavioral pattern of modulation as an effect of short-term smartphone app–based FAM training. It can be hypothesized that FAM may affect either processes of stimulus-related attention and perception, resulting in effects in the S-cluster (as mentioned above, especially in the N2 component), or processes between stimulus evaluation and response selection, which results in effects in the C-cluster. Considering preceding studies on this paradigm, we do not expect to observe effects in the R-cluster (Wolff et al., 2017, 2018).
Nineteen healthy individuals between 18 and 35 (26.21 ± 4.90) years old (seven men) took part in the experiment, which used a crossover study design (see next paragraph). All participants had normal or corrected-to-normal vision, were right-handed, and reported no psychiatric or neurological disorders. Each participant received an expense allowance of €70. Written informed consent was obtained from all participants, and the local institutional review board of the Faculty of Medicine of the TU Dresden approved the study. Participants were recruited via electronic notifications and completed a telephone screening before getting included in the study in which they were asked to report about their previous meditation experience to ensure that participants were novices or had only a few irregular meditation sessions before participating in the study. Questions were whether they are experienced in meditation and, if yes, for how long they practice meditation. In addition, we asked about the frequency of meditation sessions. Of the n = 19 participants, n = 10 had no experience of meditation at all. The other nine participants reported experiences with meditation in irregular application. Yet, including this factor in the statistical analysis did not change the pattern of results, and no interaction with this factor was obtained (all ps > .4). The results are, therefore, unbiased regarding prior experience with meditation training.
The study was conducted using a within-participant (crossover) design. Each participant was tested at two separate appointments (appointment with meditation [A] before the task and appointment without meditation [B]). To control for possible learning effects, effects of expectations or fatigue (Petruo, Mückschel, & Beste, 2018) appointments were counterbalanced across participants, and participants were randomly assigned to two groups in which the order of meditation administration was changed (10 participants started with meditation at the first appointment, and nine participants started without meditation). When including the factor “meditation order” in the data analysis (both ERP and behavioral data analyses), this factor did not change the pattern of results (all ps > .6), suggesting that neither the time point on which participants meditate (Appointment A or B) nor minimal differences in length (15 min, depending on whether participants meditate or not) between appointments influence the pattern of results. The sequence during testing always followed the same order: Participants arrived at the laboratory, were welcomed, and seated in front of the monitor in the EEG cabin. After providing their consent to participate in the study, the EEG cab was prepared at the participant's head. If participants had the appointment with meditation, the app was started and participants meditated with the help of the app for 15 min directly after preparing the EEG cap. Afterward, the task-switching paradigm was explained, exercised, and executed. If participants had the appointment without meditation, the task-switching paradigm was explained, exercised, and executed immediately after preparing the EEG cab. The mean time interval between the appointments was 7.37 (±4.2) days. Within the meditation appointment, participants listened to a 15-min guided short-term FAM of the German version of the Mindfulness App [Achtsamkeits App], which was presented via a smartphone (Samsung Galaxy J3) through headphones previous to the switching task. An evaluation of the used app can be found in Mani, Kavanagh, Hides, and Stoyanov (2015). The app can be retrieved from the Google Play Store (www.play.google.com/store/apps/details?id=se.lichtenstein.mind.en&hl=de).
The meditation starts with the instruction to take a comfortable sitting position. First, the attention should be directed to the inside of the body and to the current moment, then to the perception of the body without evaluating it, and, after that, to the breath.
During testing, participants were seated in a separate room in front of a monitor and performed a task-switching paradigm (Wolff, Roessner, & Beste, 2016; Gajewski et al., 2011). In contrast to earlier research on that paradigm with a cue- and memory-based switching variant (Wolff et al., 2016; Gajewski et al., 2011), the paradigm used in this study consists exclusively of the memory-based block (see Figure 1).
This was done to avoid ceiling effects. It has been shown that demands during memory-based switching are much higher than in cue-based switching (Wolff et al., 2017; Gajewski et al., 2011). In the memory-based version of the task-switching paradigm, external cues did not exist. Rather, participants had to retrieve, switch, and repeat rules out of memory. Therefore, the order of rules was shown at the beginning of the paradigm and had to be remembered. The paradigm contains an exercise block with nine trials per condition. During testing, six consecutive blocks with a total of 198 trials per condition were administered. Each of these blocks contains the same number of digits (stimuli) and responses. The participants were able to take a self-determined period of rest between blocks. On a black computer screen, digits between 1 and 9 were displayed centrally, without the number 5. Participants were instructed to respond as accurately and quickly as possible according to three alternating rules (Wolff et al., 2016; Gajewski et al., 2011). There was a numeric rule (participants were supposed to assess whether the presented digit was smaller or larger than 5), an even or odd rule (participants had to decide whether the digit was even or odd), and a font-size rule (participants had to determine whether the digit was shown in a large or small font size). There was a fixed sequence of rules to ensure that participants can remember, retrieve, and allocate the rule during testing. Participants were to start with the numeric rule and to continue the categorization with this rule three trials in succession. Afterward, the even or odd rule had to be applied, which should also be applied three times in succession. Finally, digits should be categorized by the font-size rule (also applied three times in succession). Then, participants had to restart from the beginning and continue the digit categorization always in the same order out of memory. Before the digits appeared, participants saw a dummy cue for 1300 msec. In case a participant did not apply the correct rule in three consecutive trials, the following three trials showed the real order with the help of short cues (e.g., “NUM,” for the numeric rule; “GER,” for the odd or even rule; or “SG,” for the font-size rule). The switching frequency was 33.3%, with an evenly distributed share of each rule of 33.3%. A RT of 2500 msec after the presentation of the stimuli was accepted as being a valid response. If this was not the case, the response was considered as missing. A feedback of 500 msec was displayed 500 msec after the response. After the feedback, the next stimuli appeared. The response cue interval included the feedback cue delay (which jittered between 400 and 600 msec [500 msec on average]), the response feedback delay (500 msec), and the duration of the feedback stimulus (500 msec) and was set to ∼1500 msec.
In past years, we have collected data in several studies using this task. Therefore, we were able to calculate split-half reliability measures in n = 156 participants in the same age range as included in the current study. This was done, for (a) RTs and (b) response accuracy. The split-half reliability after Spearman–Brown correction for the split-half reliability is .86 for accuracy data and .96 for RT data.
EEG Recording Analysis
The EEG was recorded from 60 Ag/AgCl electrodes (BrainAmp, Brain Products Inc.) at equidistant positions (approximating the position of the 10/20 system) with a sampling rate of 500 Hz (ground electrode at θ = 58, ϕ = 78; reference electrode at θ = 90, ϕ = 90). Sixty recording electrodes were mounted in an elastic cap (Easy Cap Inc.). LECTERON III-10 gel (Easy Cap Inc.) was used. Electrode impedances were kept below 5 kΩ. After recording, the data were down-sampled to 256 Hz during offline data processing, and an infinite impulse response band-pass filter from 0.5 to 20 Hz with a slope of 48 dB/oct was applied using the Brain Vision Analyzer 2 software package (Brain Products Inc.). To reject technical artifacts from the EEG, the raw data were inspected manually. In the first step, rest periods between the blocks were cut out. After that, single muscular and technical artifacts were removed. To remove recurring artifacts, an independent component analysis (infomax algorithm) was conducted on the data sets to identify and remove recurrent artifacts like horizontal and vertical eye movements, blinks, and pulse artifacts. In a final raw data inspection step, any remaining residual artifacts were removed manually. Afterward, EEG data were segmented into stimulus-locked epochs for repeat and switch trials. The segments ranged from −200 msec before until 1000 msec after stimulus onset. Only correct trials were included. After that step, an automated artifact rejection procedure was applied, which had the following rejection criteria: Maximum voltage step exceeds 200 μV in an interval of 200 msec, maximum voltage exceeds 150 μV in a 250-msec interval, and there was activity below 0.5 μV in a 100-msec interval. After these procedures, 124 ± 7 trials were left for repetition conditions with meditation, 123 ± 6 trials were left for repetition conditions without meditation, 60 ± 5 trials for were left switching conditions with meditation, and 61 ± 4 trials were left for switching conditions without meditation. This allows a reliable quantification of ERPs. Moreover, the comparable trial numbers in the meditation condition exclude that possible effects observed for the conditions are because of differences in the SNR. To rereference the data, a current source density transformation was applied using potential differences between one electrode and the potential total of all surrounding electrodes (Perrin, Pernier, Bertrand, & Echallier, 1989). The rereferencing procedure serves as a spatial filter to identify electrodes that can be analyzed for different ERP components (Tenke, Kayser, Abraham, Alvarenga, & Bruder, 2015). Furthermore, this method eliminates the reference potential from the data (Nunez & Pilgreen, 1991). Then, a prestimulus baseline correction (−200 to zero) was implemented. Segments were separately averaged for each condition at the single-participant level. On the basis of the stimulus-locking procedure, the P1, N1, N2, and P3 ERPs were quantified. Electrodes were chosen on the basis of visual inspection of the scalp topography. According to this, the P1 (interval: 95–110 msec poststimulus) and N1 (interval: 145–160 msec poststimulus) ERPs were quantified at electrodes P7 and P8. The N2 was quantified at the electrode FCz within the interval of 260–315 msec. P3 quantification was based on mean amplitudes at electrode PO2 within the interval from 410 to 465 msec. The abovementioned choice of electrodes and search intervals was validated with statistical methods (Mückschel, Stock, & Beste, 2014): The average amplitude was obtained for all 60 electrodes in each of the mentioned search intervals. To compare each electrode to the average of all other electrodes within a given time interval, Bonferroni correction was applied for multiple comparisons (critical threshold: p = .0007). Only electrodes that had significantly larger mean amplitudes than the other electrodes (i.e., negative for N potentials and positive for P potentials) were selected and kept for analyses. It is important to note that this procedure yielded the same electrodes that we identified during the visual inspection of the data.
We performed the RIDE analysis according to established procedures (Mückschel, Dippel, & Beste, 2017; Verleger, Metzner, Ouyang, Śmigasiewicz, & Zhou, 2014) using the RIDE toolbox (www.cns.hkbu.edu.hk/RIDE.htm). With this procedure, we account for intraindividual variability in the data (Ouyang, Herzmann, Zhou, & Sommer, 2011) and are able to distinguish between different coding levels that are not expressed in the ERP components. Ouyang et al. (2011) show mathematical details on the RIDE method.
The RIDE analysis decomposes ERPs into C-, R-, and S-clusters. While, for the C-cluster, time markers are estimated and repeatedly improved, the decomposition of the S- and R-clusters is based on stimulus onset and RT. This results in nonmarker- and marker-locked components (Ouyang et al., 2011). The C-cluster is based on the assumption of a variable latency time over single trials. These single trials are defined as completely locked, with regard neither to the stimulus onset nor to RTs. The latency time of the C-cluster can be repeatedly estimated in every single trial and thus can reflect a global waveform (Verleger et al., 2014). RIDE uses a self-optimized repetition scheme based on the information given by the estimated latency of the C-cluster, by the timing of the stimulus, and by RTs. This scheme improves the estimation of the C-cluster's latency. To extract the waveform for each RIDE component, a time window function is used. It is assumed that each time window covers the area within which each component should occur. Furthermore, the specific values should be adapted to the application (Ouyang et al., 2015b). For the current study, time windows were chosen as follows: from −200 to 500 msec for the stimulus marker within the S-cluster, from 200 to 1000 msec for the response marker within the C-cluster, and −500 to 500 msec for the response marker within the R-cluster. Relevant electrode sites and decomposed data peaks were first identified by visual inspections. For verification of chosen electrodes, a validation procedure according to Mückschel et al. (2014) was applied. Only electrodes that best reflect neuronal activity were used. Electrodes were chosen on the basis of visual inspection of the scalp topography. According to this, within the S-cluster, P1, N1, and N2 were quantified. The ERPs P1 (interval: 98–108 msec poststimulus) and N1 (interval: 148–158 msec poststimulus) were quantified at electrodes P7 and P8. The N2 was quantified at the electrode FCz within the interval from 278 to 310 msec. Within the C-cluster, N2 and P3 were quantified. N2 was quantified at electrode FCz (interval: 270–320 msec), and P3 was quantified at electrode PO2 (420–520 msec). Finally, in the R-cluster, we quantified the ERP N2 at electrode FCz in the time range between 270 and 320 msec. The choice of electrodes and time windows was validated using the same method as described for the standard ERPs (see above).
For source localization, the sLORETA (standardized low-resolution electromagnetic brain tomography; Pascual-Marqui, 2002) algorithm was used. The intracerebral volume for sLORETA is divided into 6239 voxels with a 5-mm spatial resolution. A calculation of the standardized current density using the MNI52 template (Mazziotta et al., 2001) is performed on each voxel in the realistic head model (Fuchs, Wagner, & Kastner, 2001). According to Sekihara, Sahani, and Nagarajan (2005), sLORETA provides reliable results based on mathematical proof. In addition, sources from EEG/fMRI and EEG/TMS studies prove the validity of the sLORETA method (Dippel & Beste, 2015; Sekihara et al., 2005). A comparison with the built-in voxel-wise randomization tests of sLORETA with 2000 permutations and voxel-based sLORETA images under different conditions was performed in this study. These are based on statistical nonparametric mapping. In the current study, voxels with significant differences (p < .01, corrected for multiple comparisons) between contrasted conditions were localized in the Montreal Neurological Institute brain. The sLORETA analysis was carried out using the RIDE data because only the RIDE data revealed differential effects of FAM depending on the experimental condition of the task-switching paradigm. Previous studies show that sLORETA can reliably be used for RIDE-decomposed data (Chmielewski et al., 2018).
For the behavioral data, we analyzed RTs as well as accuracy in percentages of hits for switching and repetition trials, as well as SCs (SCs = switching trials − repetition trials) by mixed-effects ANOVAs. The models contained the within-participant factors Condition (switch, repetition) and Meditation (with meditation, without meditation). For the neurophysiological data, separate mixed-effects ANOVAs for each ERP component were performed using the within-participant factors Condition (switch, repetition) and Meditation (with meditation, without meditation) and, for the analysis of P1 and N1, the additional factor Electrode (P7, P8). Greenhouse–Geisser correction was applied appropriately, and all post hoc tests were Bonferroni corrected. Included variables in the analysis were normally distributed as indicated by Kolmogorov–Smirnov tests (all zs < 0.99, p > .28).
A repeated-measures ANOVA on the response accuracy data (percentages of hits) showed no significant main and interaction effects (all ps > .14, all Fs < 2.30). A more detailed analysis revealed neither a main effect of Condition, F(1, 18) = 2.30, p = .146, ηp2 = .113, suggesting that accuracy during trials of switching (92 ± 1.04%) versus repetitions (94 ± 0.90%) is similar, nor a main effect of Meditation, F(1, 18) = 0.098, p = .758, ηp2 = .005, showing that accuracy during appointments with meditation (93 ± 1.09%) versus appointments without meditation (94 ± 0.97%) is comparable as well. In addition, the interaction of Meditation × Condition was also not significant, F(1, 18) = 0.001, p = .974, ηp2 < .001.
A repeated-measures ANOVA on RTs revealed a main effect of Condition, F(1, 18) = 29.12, p < .001, ηp2 = .631, with faster RTs during repetition (588 ± 23 msec) than switching trials (644 ± 26 msec). In addition, a significant interaction of Meditation × Condition was observed, F(1, 18) = 5.26, p = .035, ηp2 = .236, which indicates increased RTs during switching versus repetition conditions in the condition with meditation (657 ± 30 vs. 583 ± 23 msec) compared to without meditation (631 ± 30 vs. 593 ± 28 msec). However, the difference between switching and repetitions seemed to be even stronger if participants meditate before the task (see Figure 2). Moreover, in the absence of effects of accuracy, a speed–accuracy trade-off possibly affecting the results can be ruled out.
This difference between the switch and repetition conditions was denoted by SCs. A repeated-measures ANOVA on SCs revealed a significant effect of SCs, F(1, 18) = 5.63, p = .029, ηp2 = .23, indicating increased SCs during the “meditation” condition (72 ± 102 msec) as compared to the “no-meditation” condition (37 ± 64 msec). Thus, FAM induces higher SCs compared to a control condition. However, with respect to clinical research, a significant change can be defined as “the extent to which therapy moves someone outside the range of the dysfunctional population or within the range of the functional population” (Jacobson & Truax, 1991). This can be determined by the reliable change index (RCI). Thus, for describing the intervention (FAM) as being meaningful, the mean difference of SCs between the appointment with meditation versus the appointment without meditation has to be larger than the calculated RCI (refer to Hiller & Schindler, 2011, for the RCI formula). With respect to the present data, the calculated RCI is 20.22, suggesting that, in 11 of 19 participants, FAM leads to reduced cognitive flexibility (increased SCs, after the meditation appointment: values above 20.22).
The repeated-measures ANOVA on P1 amplitudes revealed a significant main effect of Electrode, F(1, 18) = 15.33, p ≤ .001, ηp2 = .460, indicating increased P1 amplitudes over the right (34.82 ± 4.93 μV/m2) as compared to the left (18.34 ± 2.30 μV/m2) hemisphere. No further main effect or interaction was observed (all ps > .44, all Fs < 5.35).
The repeated-measures ANOVA on N1 amplitudes revealed no significant main and interaction effect (all ps > .19, all Fs < 1.84). Neither did the mixed-effects ANOVA on the N2 revealed any main and interaction effect (all Fs < 4.21, all ps > .055). However, we observed a trend for the interaction of Meditation × Condition, F(1, 18) = 4.21, p = .055, ηp2 = .190. Finally, the mixed-effects ANOVA on P3 revealed a significant main effect of Meditation, F(1, 18) = 13.65, p < .002, ηp2 = .43, indicating increased P3 amplitudes in without-meditation (18.28 ± 3.29 μV/m2) as compared to meditation (13.32 ± 2.78 μV/m2) conditions. No further main or interaction effect was observed for P3 (all ps > .24, all Fs < 1.48; see Figure 3).
The repeated-measures ANOVA of P1 amplitudes in the S-cluster revealed a significant main effect of Electrode, F(1, 18) = 14.73, p ≤ .001, ηp2 = .450, indicating increased P1 amplitudes at P8 (35.35 ± 5.03 μV/m2) in comparison to P7 (19.08 ± 2.36 μV/m2). No further main effect or interaction was observed (all ps > .24, all Fs < 1.46). The repeated-measures ANOVA of S-cluster N1 amplitudes revealed no significant main and interaction effect (all ps > .21, all Fs < 1.68). The repeated-measures ANOVA of amplitudes in the S-cluster N2 time window revealed no main effect (all ps > .10, all Fs < .3.00) but a significant interaction of Meditation × Condition, F(1, 18) = 4.54, p < .05, ηp2 = .20. Post hoc t tests revealed significantly increased (more negative) N2 amplitudes during switches after short-term FAM (−10.34 ± 1.85 μV/m2), as compared to switches without prior short-term FAM (−5.81 ± 1.85 μV/m2), t(18) = 2.19, p = .041. There was no FAM effect during repetition trials, t(18) = 0.29, p = .78. To examine the cortical sources of the FAM effect observed in switch trials, we applied sLORETA. Results were corrected for multiple comparisons by means of the sLORETA built-in voxel-wise randomization tests with 2000 permutations (based on statistical nonparametric mapping) and revealed significant activation differences in ACC (BA 32). A repeated-measures ANOVA on SC revealed a significant effect of SCs, F(1, 18) = 4.55, p = .047, ηp2 = .20, indicating increased SCs in the meditation condition (3.05 ± 1.49 μV/m2) as compared to the no-meditation condition (−1.10 ± 1.10 μV/m2). Thus, the effects of N2 in the S-cluster reflect the behavioral data, which also show increased SCs after the meditation as compared to the no-meditation condition. In reference to the RCI, which was calculated for the behavioral data, we calculated the RCI for the neurophysiological data as well. However, it is not possible to analyze RIDE data on a single-trial level, because the RIDE toolbox provides latency corrected data solely for average data in each participant. The existence of data on a single-trial level is, however, necessary to obtain the individual standard deviation/standard error of the mean of each participant, to calculate the RCI. Nevertheless, we calculated a score of the neurophysiological SC difference (conditions with meditation − conditions without meditation) and compared both conditions separately for those 11 individuals who showed significant behavioral changes at the individual level and those eight showing no significant difference. The SC difference is larger in the group, who scored behaviorally above the RCI (4.85 vs. 3.23 μV/m2); however, the difference between both groups is only descriptive and statistically not significant (p = .69).
The C-cluster is shown in Figure 5 for the N2 and P3 time windows including the scalp topography plots (bottom). The scalp topography shows a clear fronto-central negativity around electrode FCz during N2 time range as well as a parietal positivity around electrode PO2 in the P3 time range. An analysis of the N2 at electrode FCz revealed no significant main effect or interaction (all ps < .055, all Fs > 4.22). The repeated-measures ANOVA of P3 amplitudes revealed two significant main effects: Condition, F(1, 18) = 5.76, p = .027, ηp2 = .243, indicating lower amplitudes during switching versus repetitions (8.91 ± 1.90 vs. 11.46 ± 2.08 μV/m2), and Meditation, F(1, 18) = 4.42, p = .049, ηp2 = .197, showing decreased amplitudes after the meditation versus no-meditation condition (8.64 ± 1.90 vs. 11.45 ± 2.08 μV/m2). These effects were further specified by the significant interaction of Condition × Meditation, F(1, 18) = 4.89, p = .040, ηp2 = .214.
Finally, the R-cluster is shown in Figure 5 for the N2 including the scalp topography plots (top). An analysis of the N2 revealed no significant main effect or interaction (all ps > .289, all Fs < 1.19). Therefore, with respect to the N2, neither the C-cluster (see Figure 5) nor the R-cluster did reflect differential modulations of switches and repetitions between the conditions of meditation versus no meditation. This missing interaction was additionally analyzed using Bayes statistics referring to the method of Masson (2011). According to Bayes, the probability that the null hypothesis can be presumed, given the obtained data p (Ho/D), as well as the probability that the alternative hypothesis can be presumed, given the obtained data p (H1/D), can be calculated. The analysis for the interaction of Meditation × Condition of the N2 in the C-cluster revealed p (H0/D) = .79 and p (H1/D) = .22; moreover, the analysis for the same interaction of the N2 in the R-cluster revealed p (H0/D) = .81 and p (H1/D) = .19. Accordingly, both analyses for the assumed N2 interaction in the C-cluster as well as in the R-cluster indicate that the null hypotheses seems to be more likely and that the alternative hypotheses seem to be rejected.
The current study examined which neurophysiological mechanisms underlie effects of a brief meditation-based intervention (FAM) on cognitive control and on working-memory-related cognitive flexibility, in particular. This was done by analyzing both standard ERP and decomposed data as well as by applying source localization techniques (sLORETA).
As stated in the Introduction, two opposing hypotheses are equally possible regarding the effects of FAM on task switching. According to one hypotheses, which referred to the “shielding–shifting dilemma” (Goschke & Bolte, 2014), FAM hampers the flexible reconfiguration of processes, which is necessary to rapidly adapt to changing environmental conditions (i.e., cognitive flexibility). This may happen because FAM trains the meditator to focus attention on one specific aspect. That is, FAM intends to reduce the background monitoring. However, according to the cognitive control dilemma framework (Goschke & Bolte, 2014), some degree of background monitoring is necessary to evaluate the significance of stimuli outside the current focus of state, such that these stimuli may eventually capture attention and trigger a switch in behavior (Goschke & Bolte, 2014). Thus, a too rigid focused state carries the risk that goal-directed behavior becomes too rigid and not adjustable to changed requirements (Goschke & Bolte, 2014). Indeed, the behavioral data are in line with that suggestion, because we observed increased SCs, both on a behavioral level and on a neurophysiological level after FAM as compared to the control condition. Thus, the assumption that FAM may induce a focused mental state in which other (possibly also relevant) cognitive processes become shielded from new information seems to be confirmed. FAM seems to become maladaptive when there is a need to react flexibly to changing environmental conditions. FAM seems to support behavioral persistence as well as cognitive stability, resulting in impaired adaptation to changing contexts or demands. The data suggest that even a brief, 15-min FAM intervention is sufficient to modulate cognitive flexibility, suggesting that FAM is a potent modulator of goal-directed behavior. However, the EEG data provide detailed insights to the mechanism that is modulated by FAM during task switching: With respect to the ERP data, it is shown that processes related to early perception as well as attention-related processes (Herrmann & Knight, 2001) (i.e., P1 and N1 ERP components) are not modulated and can hence not explain the behavioral results. Neither are the N2 nor P3 ERP component data able to explain increased SCs after FAM. This, however, is an expected finding likely attributable to high intraindividual variability in EEG signals that emerges when cognitive demands during task switching are high, that is, when switches have to be initiated from working memory and not by using external sensory input (Wolff et al., 2017, 2018). Consequently, when accounting for this intraindividual variability in EEG data using a temporal EEG decomposition method (i.e., RIDE), interactive effects between Condition and Meditation were observed. Methodologically, this can be explained because ERP data are decomposed in temporally static as well as variable components, which share similar time-locking properties (Ouyang et al., 2015a, 2015b). More important, however, the decomposition method allows to isolate distinct coding levels that are otherwise intermixed in ERPs (Chmielewski et al., 2018; Mückschel, Chmielewski, et al., 2017; Folstein & Van Petten, 2008). This is particularly an issue in both the N2 and P3 time windows, known to reveal strong modulations during task switching. The observed interactive effect between Condition and Meditation in the S-cluster is in line with the behavioral data. Thus, the findings in the N2 time window are of particular importance. In particular, it is shown that the S-cluster amplitude in the N2 time window was larger (more negative) during switches after meditation as compared to switching conditions without preceding meditation. The N2 is usually described to represent an index of the resolution of task-set conflict (Folstein & Van Petten, 2008; Gehring et al., 2003; Karayanidis et al., 2003). The sLORETA analysis further shows that the observed amplitude differences result from activity in ACC (BA 32). ACC interfaces processes between monitoring and control and is one of the central importance whenever decisions have to be taken and whenever cognitive flexibility is needed (Shenhav, Cohen, & Botvinick, 2016). It thus seems that FAM increases conflicts during task switching and reduces the ability to select an appropriate response whenever the task has to be switched. This is reasonable because FAM induces a focused mental state in which other (possibly also relevant) cognitive processes become shielded from new information. By doing so, FAM increases the conflict between a previous but no longer relevant task set and the actually required activation of the task set in ACC. Theoretical accounts on ACC suggest that it allocates cognitive control based on a cost–benefit analysis and on a resulting estimation of how much control needs to be invested (Shenhav et al., 2016). It is possible that FAM impedes the allocation of processing capacities to the new and relevant task set.
Interestingly, the observed modulations of the N2 found in the S-cluster (rather than the C- or R-cluster) suggest that FAM effects are because of stimulus-related aspects during task switching. This may be counterintuitive because a memory-based task-switching paradigm was used in the current study. However, during memory-based task switching, participants have to switch out of memory whenever an uninformative dummy cue was presented. This means that the uninformative cue is still important to signal the participants when to retrieve the correct task rule from memory. Moreover, the task required some form of counting using visual stimuli belonging to the trials to judge, when to switch to another task set. This may be the reason why the stimulus-related processes in the N2 time window (as revealed by the S-cluster) are important during FAM-modulated memory-based task switching. Interestingly, no modulatory effects of FAM in the N2 time range, which are in line with the behavioral data, were evident in the C-cluster and the R-cluster. This is despite motor activity (reflected by the R-cluster) that is evident in the N2 time window (Mückschel, Chmielewski, et al., 2017; Folstein & Van Petten, 2008) and motor processes that are well known to modulate SCs (Vandierendonck, Liefooghe, & Verbruggen, 2010; Steinhauser & Hübner, 2006, 2008; Philipp, Jolicoeur, Falkenstein, & Koch, 2007; Koch & Philipp, 2005; Verbruggen, Liefooghe, Szmalec, & Vandierendonck, 2005; Schuch & Koch, 2003). With respect to the C-cluster and the P3 time range, we observed modulatory effects of FAM as well. Already previous studies suggested that working-memory-triggered task switching modulates processes in the C-cluster P3 time window (Wolff et al., 2017) and also processes mediating between stimulus evaluation and responding (Ouyang et al., 2017; Verleger et al., 2014). Effects of the condition without meditation are thus in line with earlier data and replicate these effects (Wolff et al., 2017). However, for the meditation appointment, amplitudes did not differ between switch and repetition trials. This may be interpreted that FAM reduces the ability to differentiate between conditions requiring higher and lower cognitive resources, which is needed to perform switching processes (Lorist et al., 2000). Together with the findings from the S-cluster data, the results suggest that the applied FAM procedure has very specific processes during task switching, namely, stimulus-related processes (cf. S-cluster) and processes linking such stimulus information to responses (C-cluster). The likely reason for this is the FAM protocol used. The FAM protocol used has been supposed to modulate “the direction of attention to one specific object while monitoring and disengaging from extraneous thoughts or stimuli” (Fox et al., 2016; Kabat-Zinn, 2005). Thus, the FAM protocol aims to modulate how stimulus-related aspects are being processed and also how responses are prepared. Therefore, it seems reasonable that particularly neurophysiological processes reflected by the S- and C-clusters were modulated. The observed modulatory effects fit to the above interpretation based on the cognitive control dilemma framework (Goschke & Bolte, 2014) that particularly alterations in background monitoring may be important to consider when explaining FAM effects during task switching.
It is possible that other meditation practices, which may focus less on how external stimuli are being processed or offer a broader focus on the perception of several incoming information (e.g., OMM), do also show a different neurophysiological profile and may also target other cognitive mechanisms important during goal-directed, flexible behavior. In addition, it appears to be relevant on how meditations like FAM may help patients who have difficulties to focus (i.e., shield) on one specific object and who are easily distractible, like patients with ADHD, may profit from such an intervention. This shall be subject for future research. Moreover, the lack of a group performing some form of a sham-mindfulness training is a limitation for the study. Further studies should add this condition to their experimental setup. Finally, we did not account for expectancy or placebo effects, which may also be mentioned as a limiting factor. Expectancy effects have been shown to increase motivation and thus to influence or, more specifically, improve performance (Boot, Simons, Stothart, & Stutts, 2013). However, this study shows reduced performance after FAM, while participants reported neutral to positive attitudes toward meditation. This suggests that possible expectations associated with the effects of FAM did not lead to improved performance. Still, future studies should control for possible expectations and/or placebo effects.
In summary, we showed that even 15 min of an app-based meditation is able to modulate cognitive functions like memory-based cognitive flexibility, on both behavioral and neurophysiological levels. FAM thus increases conflicts in ACC during task switching and reduces the ability to select an appropriate response whenever the task has to be switched. Interestingly, even for conflict monitoring processes, it is shown that FAM specifically modulated stimulus-related processes, which can be explained by the nature of the mindfulness training examined in this study.
This work was supported by a grant from the Else-Kröner Fresenius Stiftung (2017_A101) to N. W. and by grants from the Deutsche Forschungsgemeinschaft (DFG) FOR 2698 and SFB 940 project B8 to C. B. We thank Martin Wikfalk and Magnus Fridh from www.mindapps.se for cooperating with us in such an uncomplicated and courteously way. Finally, we thank Nathalie Henke and Jenny Tippmann for help during recruitment and data collection.
Reprint requests should be sent to Nicole Wolff, Cognitive Neurophysiology, Faculty of Medicine, Department of Child and Adolescent Psychiatry, Dresden University Hospital, TU Dresden Fetscherstraße 74, 01307 Dresden, Germany, or via e-mail: firstname.lastname@example.org.