Conflict monitoring processes are central to cope with fluctuating environmental demands. However, the efficacy of these processes depends on previous trial history/experience, which is reflected in the “congruency sequence effect” (CSE). Several theoretical accounts have been put forward to explain this effect. Some accounts stress the role of perceptual processes in the emergence of the CSE. As yet, it is elusive how these perceptual processes are implemented on a neural level. We examined this question using a newly developed moving dots flanker task. We combine decomposition methods of EEG data and source localization. We show that perceptual processes modulate the CSE and can be isolated in neurophysiological signals, especially in the N2 ERP time window. However, mechanisms relating perception to action are also coded and modulated in this time window. We show that middle frontal regions (BA 6) are associated with processes dealing with purely perceptual processes. Inferior frontal regions (BA 45) are associated with processes dealing with stimulus–response transition processes. Likely, the neurophysiological modulations reflect unbinding processes at the perceptual level, and stimulus–response translation level needed to respond correctly on the presented (changed) stimulus–response relationships. The data establish a direct relationship between psychological concepts focusing on perceptual processes during conflict monitoring and neurophysiological processes using signal decomposition.
Response selection and control processes are central to cope with the fluctuating environmental demands. This is particularly the case when we carry out responses that go against natural response tendencies (Keye, Wilhelm, Oberauer, & Stürmer, 2013) and create response conflicts. Experimentally, these can be examined using Flanker or Simon tasks. There, an irrelevant (perceptual) feature may foster the selection of the correct response when the same response is activated by the relevant stimulus feature. This is the case in congruent trials. In incongruent trials, however, an irrelevant perceptual feature activates a response tendency opposite to the correct response (Dignath, Johannsen, Hommel, & Kiesel, 2019; Keye et al., 2013). This produces a “congruency effect,” which is associated with slower and more error-prone responses in incongruent stimulus combinations. The magnitude of that effect depends on previous trial history/experience. This is evidenced by the “congruency sequence effect” (CSE) or “Gratton effect” (Duthoo, Abrahamse, Braem, Boehler, & Notebaert, 2014; Keye et al., 2013; Schmidt, 2013; Gratton, Coles, & Donchin, 1992). The CSE describes a higher congruency/interference effect following a trial with a congruent stimulus–response mapping than following a trial with noncongruent stimulus–response mapping. Several theoretical accounts have been put forward to explain this effect (Egner, 2008, 2014; Schmidt, 2013; Verguts & Notebaert, 2009; Hommel, Proctor, & Vu, 2004; Mayr, Awh, & Laurey, 2003; Botvinick, Braver, Barch, Carter, & Cohen, 2001; Gratton et al., 1992).
The so-called bottom–up associative accounts state the CSE to stem from effects based recurrent stimulus and response features over consecutive trials (Egner, 2014; Hommel et al., 2004; Mayr et al., 2003). The importance of these processes has been discussed and corroborated by various studies (for reviews, Egner, 2014; Kim & Cho, 2014; Schmidt, 2013; Hommel et al., 2004; Mayr et al., 2003). These accounts state that important mechanisms operate at the level of specific stimulus features and associated responses. The important implication of this theoretical account is that the perceptual level (i.e., level of stimulus features) and modulations at that level have an impact on the CSE. However, the distinction of bottom–up from other (top–down) accounts has been criticized (Hommel & Wiers, 2017), and attempts have been made to integrate these accounts (Dignath et al., 2019; Egner, 2008, 2014). Yet, these accounts integrating bottom–up and top–down accounts of the CSE also stresses the important role of external sensory input. Similarly, Verguts and Notebaert (2009) suggest that it is not necessary to distinguish between top–down and bottom–up processes. Yet, they also stress the importance of perceptual processes (Verguts & Notebaert, 2009), because this forms the basis for learning processes that may underlie congruency effects (Verguts & Notebaert, 2009).
However, it is unclear how perceptual codings and processes during the CSE are implemented on the neurophysiological level and the functional neuroanatomical level. The current study fills this gap. It combines EEG signal decomposition and source localization methods. To examine the effect of perceptual factors on the CSE, it is important to circumvent strong feature overlaps between different congruency sequences (Egner, 2014). Therefore, we used visual motion stimuli. Because the CSE is strong in flanker tasks, we created a modified moving dot flanker task, which has originally been introduced by Lange-Malecki and Treue (2012).
On the EEG level, effects of conflicts are reflected by the N2 ERP component, which is associated with ACC and adjacent medial frontal areas (Beste et al., 2012; Willemssen, Falkenstein, Schwarz, Müller, & Beste, 2011; Beste, Domschke, Falkenstein, & Konrad, 2010; Folstein & Van Petten, 2008; van Veen & Carter, 2002). The N2 is more negative in incongruent/conflicting trials than in congruent/nonconflicting trials (Gohil, Bluschke, Roessner, Stock, & Beste, 2017; Chmielewski, Mückschel, Dippel, & Beste, 2016; Mückschel, Stock, Dippel, Chmielewski, & Beste, 2016; Stock, Friedrich, & Beste, 2016; Wolff, Roessner, & Beste, 2016; Larson, Clayson, & Clawson, 2014; Kanske & Kotz, 2011; Danielmeier, Wessel, Steinhauser, & Ullsperger, 2009; Folstein & Van Petten, 2008). The N2 in trial n is also modulated by the congruency in the n − 1 trial and hence affected by the CSE (e.g., Chmielewski, Roessner, & Beste, 2015; Chmielewski, Mückschel, Roessner, & Beste, 2014; Clayson & Larson, 2013; Spapé, Band, & Hommel, 2011; Winkel et al., 2009). However, generally, ERP components reflect activity from different sources and reflect a mixture of different functional processes (Stock, Gohil, Huster, & Beste, 2017; Huster, Plis, & Calhoun, 2015; Nunez et al., 1997). This is particularly the case for the N2 (Folstein & Van Petten, 2008). EEG signal decomposition has shown that the N2 is composed of stimulus-related and response-related processes (Chmielewski, Mückschel, Ziemssen, & Beste, 2017; Mückschel, Chmielewski, Ziemssen, & Beste, 2017). This is important to consider when being interested in examining the role of perceptual processes in neurophysiological mechanisms underlying conflict monitoring and conflict-related behavioral adjustments. Because previous data indicate that only some fraction of the conflict-related neurophysiological signal (i.e., the N2) is due to perceptual processes (Chmielewski et al., 2017; Mückschel, Chmielewski, et al., 2017; Folstein & Van Petten, 2008), it is likely that experimental manipulations of the CSE by perceptual factors cannot be observed in the EEG signal when it has not been decomposed into different coding levels (Chmielewski, Mückschel, & Beste, 2018). Such a decomposition can be achieved applying residue iteration decomposition (RIDE; Ouyang, Sommer, & Zhou, 2015b; Ouyang, Herzmann, Zhou, & Sommer, 2011). Even though the primary goal of this method is to reduce intraindividual variability in EEG data (Ouyang, Sommer, & Zhou, 2015a), the algorithm can be used to dissociate intermingled coding levels in EEG (Chmielewski et al., 2018; Mückschel, Chmielewski, et al., 2017). This is because the ERP signal is decomposed into three functionally different clusters (Ouyang et al., 2015a): The S-cluster reflects stimulus-related perceptional and attentional processes, the R-cluster reflects motor preparation and execution processes, and the C-cluster contains intermediate processes between stimulus evaluation and responding (Ouyang, Hildebrandt, Sommer, & Zhou, 2017; Ouyang et al., 2011). Here, we use RIDE to isolate different coding level in the neurophysiological signal to establish a direct relationship between psychological concepts stressing the role of perceptual processes during the CSE and neurophysiological processes using EEG signal decomposition and source estimation methods. If bottom–up perceptual processes reflect a central mechanism constituting the CSE, it is hypothesized that especially the S-cluster should be modulated by trial history effects reflected by the CSE. However, as mentioned, the C-cluster likely reflects intermediate processes between stimulus encoding and responding (Bluschke, Chmielewski, Mückschel, Roessner, & Beste, 2017; Mückschel, Chmielewski, et al., 2017; Ouyang et al., 2017; Verleger, Siller, Ouyang, & Śmigasiewicz, 2017; Wolff, Mückschel, & Beste, 2017; Verleger, Grauhan, & Śmigasiewicz, 2016; Verleger, Metzner, Ouyang, Śmigasiewicz, & Zhou, 2014). Therefore, stimulation-related information must be visible in the C-cluster signal and should also reflect modulations in amplitude by CSE.
From a neuroanatomical point of view, the CSE modulates the ACC and the middle frontal gyrus/supplemental motor region (SMA; Herz et al., 2014; Clayson & Larson, 2013; Stock, Wascher, & Beste, 2013; Mars et al., 2009; Winkel et al., 2009; Nachev, Kennard, & Husain, 2008; Rushworth, Walton, Kennerley, & Bannerman, 2004). We therefore assume that these regions are also involved in perceptual processes during the CSE. However, this would require that perceptual signals are evident in the human pFC. Likely, this is possible especially because middle frontal regions show direct structural connections to visual association areas (Hagmann et al., 2008). Indeed, it has been shown that, aside from higher-level response selection processes, perceptual signals are also evident in prefrontal brain structures (Rahnev, Nee, Riddle, Larson, & D'Esposito, 2016). Therefore, such prefrontal regions are involved in perception decision-making mechanisms (Ruff, Marrett, Heekeren, Bandettini, & Ungerleider, 2010; Philiastides & Sajda, 2007; Heekeren, Marrett, Ruff, Bandettini, & Ungerleider, 2006; Heekeren, Marrett, Bandettini, & Ungerleider, 2004). Furthermore, these areas are affected and involved in perceptual modulations of cognitive control (Adelhöfer, Chmielewski, & Beste, 2019; Labrenz, Themann, Wascher, Beste, & Pfleiderer, 2012; Westerhausen et al., 2010). Importantly, using RIDE it has also been shown that different coding levels pertaining to stimulus-related processes and motor response-related processes are associated with middle frontal/SMA regions (Mückschel, Chmielewski, et al., 2017). Therefore, we assume that middle frontal/SMA regions are associated with modulations of the S-cluster by perceptual factors during the CSE when performing source localization analyses. The same may be the case for the C-cluster; however, the C-cluster has also been associated with inferior frontal structures (Mückschel, Dippel, & Beste, 2017).
Participants and Power Estimations
None of the participants reported a history of neurological or psychiatric illness. All participants reported to be right-handed. All participants provided written informed consent. The study was approved by TU Dresden's ethics committee.
First, we performed a sensitivity analysis using G*Power (Faul, Erdfelder, Lang, & Buchner, 2007). According to that analysis, a sample size of n ∼ 20 participants is sufficient to detect an effect with an effect size f = 0.33 (which equals a partial eta squared ηp2 = .10) and a power of 95%. This is because the study uses a within-participant design. Previous studies by our group using flanker tasks and identical within-participant study designs in n ∼ 20 participants revealed that the ηp2 of CSE effects in standard flanker tasks was ∼.4 (Chmielewski et al., 2014, 2015). Therefore, n = 21 participants (nine women, aged 24.7 ± 2.4 years) with normal vision took part. Of note, the obtained effect sizes were larger than an effect size of ηp2 = .10, for which the power analysis shows that it can reliably be detected with the recruited sample.
The stimuli were shown on a 24-in. TFT (120 Hz refresh rate) at a viewing distance of 56.5 cm. The diameter of the moving random dot patterns (RDPs) was 1° visual angle. The distance between the target and each flanker was 1.25° from center to center of the RDPs. Dot density was 80/deg2, dot size was 0.05°, and dot speed was 3°/sec . The target RDP was presented centrally on the screen. The screen background was light gray (64.5 cd/m2). Participants sat in a dimly lit room with a chin-rest. During a trial, flanker RDPs preceded the target RDP by 200 msec (i.e., SOA = 200 msec). The target was presented for 150 msec. An SOA to present flanker and target stimuli has also been used by Lange-Malecki and Treue (2012) introducing this paradigm. An SOA between flanker and target in the range of 100–200 msec maximizes response conflict (Sidarus & Haggard, 2016; Wendt, Kiesel, Geringswald, Purmann, & Fischer, 2014; Kopp, Rist, & Mattler, 1996) and is therefore suitable for study questions examining conflict-related behavioral adjustments. The target presentation was followed by a variable intertrial interval of 1200–1500 msec. In the intertrial interval, only a central fixation cross was evident. Participants were asked to press one of two response buttons with their index finger corresponding to the direction of the target. In congruent trials, flanker and target pointed in the same, in incongruent trials in opposite horizontal directions. For a reliable investigation of neural mechanisms underlying the CSE, it is important to ensure that the various possible transitions between consecutive trials occur with a reasonable frequency. There are four possible trial transitions: congruent–congruent (cC), congruent–incongruent (cI), incongruent–congruent (iC), and incongruent–incongruent (iI) trial transitions. To ensure a reasonable frequency in all trial transitions and particularly of iC and cI trials, the following established procedure was run (e.g., Chmielewski et al., 2014, 2015): A sequence of 36 trials was compiled and optimized in MATLAB (The MathWorks) to avoid the recurrent appearance of any further short sequences within the 36 trials. This sequence contained 24 compatible and 12 incompatible trials, in which target motion direction was equally distributed to the left and right was compiled in MATLAB in a pseudorandomized order. The sequence did not contain three or more trials in the same succession (i.e., congruent and incongruent) and no more than five trials in terms of target RDP direction. Doing so, 87 cC, 30 iI, 30 cI, and 30 iC transitions were established in each experimental block, summing up to 348 cC, 120 cI, 120 iC, and 120 iI trials in the whole experiment. The experiment took 30 min to conduct.
EEG Recordings and Analyses
The EEG was recorded from equidistant position (60 Ag–AgCl electrodes, 500 Hz sampling rate; reference electrode coordinates: theta = 90, phi = 90; ground electrode coordinates: theta = 58, phi = 78; recording electrode impedances < 5 kΩ). The signals were amplified with QuickAmp amplifier (Brain Products, Inc.) and recorded using the BrainVision Recorder software. Offline preprocessing was done using the BrainVision Analyzer software (Brain Products, Inc.): First, data were resampled to 256 Hz and bandpass filtered between 0.5 and 20 Hz, including a notch filter at 50 Hz (48 dB/oct). All 60 channels were then re-referenced to an average of all channels. In a manual artifact rejection step, gross technical artifacts (e.g., offsets in the EEG) and muscle artifacts were discarded. In the next step, we performed an independent component analysis to identify components displaying physiological artifacts, that is, eye blinks, saccades, and ECG artifacts. These independent components were removed before data back-projection. Segments were formed from 200 msec prestimulus to 1000 msec poststimulus onset and categorized according to the combination of experimental conditions (i.e., left/right movement of the target dots, congruency of the previous trial, congruency of the current trial). Only correct trials were included. These segments underwent a further automatic artifact rejection procedure with the inclusion criteria that voltage steps are below 50 μV/msec and maximum voltage differences in each 200 msec interval stay below 200 μV. To improve the spatial resolution of the data, we performed a spline Laplacian (current source density [CSD]) transformation (Kayser & Tenke, 2015). The baseline of the segments was set according to the interval [−200 msec, 0 msec] relative to stimulus onset. These single trial segments, together with time markers indicating button press onset, were used as an input to derive decomposed clusters using RIDE (Ouyang et al., 2011). For RIDE, we used the corresponding toolbox available at cns.hkbu.edu.hk/RIDE.htm. As described in the Introduction section, RIDE is a temporal EEG decomposition method. The mathematical rationale behind this method is described in the study of Ouyang et al. (2015b). As mentioned, RIDE decomposes the ERP signal in three clusters; the S-cluster, the C-cluster, and the R-cluster. RIDE uses a time window function to extract each RIDE component. The time window has to cover the range within each component likely occurs (Ouyang et al., 2015b). The time windows were set to −100 to 600 msec relative to stimulus onset (S-cluster), 100–700 msec relative to stimulus onset (C-cluster), and −300 to 300 msec relative to button press onset (R-cluster). The S-cluster and the R-cluster are time-locked to the stimulus onset or response onset, respectively. The C-cluster reflects processes that are neither fully stimulus-locked nor response-locked. The latency estimation of the C-cluster is improved by means of a self-optimized iteration scheme. There, the initial latency of the C-cluster is estimated using a time window function. Then the S-cluster is removed, and the latency of the C-cluster is reestimated. This procedure is repeated until the latency variability in the C-cluster is minimized. For each cluster and condition combination, we averaged the data of all trials and participants to identify relevant channels and time windows visually. These choices were validated with a statistical procedure used by Mückschel, Stock, and Beste (2014): Within each time window, the peak voltage value for each channel was compared with the average peak value of all other channels. Only electrodes with a significantly larger absolute peak value (corrected for multiple comparisons) were included in the analysis. This procedure yielded the same electrodes as were identified visually beforehand. The results are shown in Table 1.
|Cluster .||Time Window .||Electrode .||Quantified Range (msec) .|
|255–380 (all other conditions)|
|R||CP1, CP2||635–870 (cC, iC)|
|CP1, CP2||635–960 (all other conditions)|
|Cluster .||Time Window .||Electrode .||Quantified Range (msec) .|
|255–380 (all other conditions)|
|R||CP1, CP2||635–870 (cC, iC)|
|CP1, CP2||635–960 (all other conditions)|
Source Localization Analyses
We used sLORETA for source localization (Pascual-Marqui, 2002; www.unizh.ch/keyinst/NewLORETA/sLORETA/sLORETA.htm). The algorithm was validated by combined fMRI/EEG and TMS/EEG studies (Dippel & Beste, 2015; Sekihara, Sahani, & Nagarajan, 2005) and has also mathematically been provided to yield reliable source estimations (Pascual-Marqui, 2002). The algorithm requires standard electrode coordinates and uses a three-shell spherical head model (Pascual-Marqui, 2002). The intracerebral volume is partitioned into 6239 voxels (5 mm voxel size), and the standardized current density is calculated for each voxel using a MNI152 head model template. As input data to the algorithm, we used the RIDE cluster data (Chmielewski et al., 2018), which revealed a CSE in line with the behavioral data (see Results section for details). Statistics in the source space were calculated using voxel-wise randomization tests as implemented in the statistical nonparametric mapping method modulate implemented in sLORETA. We used with 2500 permutations. The positions of voxels that were significantly modulated by the CSE (p < .01) are shown in the MNI brain. The activations shown in the brain represent critical t values that have been corrected for several comparisons. The correction for multiple comparisons is performed because the sources localization analysis is calculated over all voxels in the used MNI brain. The correction does only refer to contrasted conditions used in the sLORETA analysis for source localization. It should be noted that the more recent eLORETA software package provides only some advantages over sLORETA in terms of less blurred images and a better suppression of less significant details in source localizations (Jatoi, Kamel, Malik, & Faye, 2014).
Statistical analysis was carried out using IBM SPSS Statistics 20. For the descriptive statistics, the mean and standard error of the mean are provided. For the analysis of the behavioral data and the EEG data, three experimental factors were included in the statistical analyses (repeated-measures ANOVA): (i) Direction, denoting the left or right movement of the dots in the central RDP; (ii) Previous Trial Congruency, denoting the match of direction between the flanker and target RDP of the previous trial; and (iii) Current Trial Congruency, denoting the match of the current trial. In the following, the previous (n − 1) trial congruency will be abbreviated with lower case letters “I” (incongruent previous trial) and “c” (congruent previous trial), whereas upper case letters will be used for current congruency (e.g., “iC” will refer to an incongruent trial preceding a congruent trial). If multiple electrode sites were identified in the EEG data, these were included as an additional within-participant factor. Greenhouse–Geisser correction was applied and post hoc tests were Bonferroni-corrected (p < .05), if these post hoc tests were significant. Importantly, a lack of interaction of Previous Trial Congruency × Current Trial Congruency in the EEG data is also of theoretical importance. Therefore, lack of effects in these interactions was further examined using Bayesian statistics. We used the method by Masson (2011). Using this method, the probability of the null hypothesis being true given the obtained data p(H0/D) can be calculated.
Participants responded more accurately if the previous trial was incongruent (77.8 ± 2.6%) compared with a congruent previous trial (74.6 ± 2.9%), F(1, 20) = 12.42, p = .002, ηp2 = .383. It is noteworthy that we observed the same interactive effects as were shown in RTs. We found an interaction of Direction × Previous Trial Congruency, F(1, 20) = 9.68, p = .006, ηp2 = .326. With right-moving target dots, participants responded significantly more accurate if the previous trial was incongruent (81.1 ± 2.6%) compared with congruent (74.9 ± 3.1%), t(20) = 3.89; p = .001. In addition, there was an effect of Direction after previous incongruent trials, t(20) = −2.09; p = .050: In this case, accuracy decreased to left-moving dots (74.4 ± 3.5%) compared with right-moving dots (81.1 ± 2.6%), all other t(20) ≤ 0.18, p ≥ .859. Most important, a CSE was evident (Previous Trial Congruency × Current Trial Congruency: F(1, 20) = 36.13, p < .001, ηp2 = .644) and is shown in Figure 2.
Post hoc tests revealed two significant directions of this interaction: When previous trials were congruent, accuracy was significantly lower when the current trial was incongruent (69.6 ± 3.6%), compared with when the current trial was congruent (79.6 ± 3.1%), t(20) = −2.94, p = .008, d = 0.64. When current trial was incongruent, participants responded less accurately after a previous congruent (69.6 ± 3.6%) than a previous incongruent trial (76.4 ± 3.0%), t(20) = −5.33, p < .001, d = 1.16. All other comparisons did not turn out to be significant (all other t(20) ≤ 0.93, p ≥ .362). No other main or interaction effects were found (all F ≤ 4.12, p ≥ .056).
Concerning RTs, we found a main effect of Current Trial Congruency, F(1, 20) = 15.41, p = .001, ηp2 = .435. On average, responses were faster in congruent (693 ± 22 msec) compared with incongruent trials (737 ± 24 msec). In addition, we observed two interactive effects: The first was and interaction of Direction × Previous Trial Congruency, F(1, 20) = 9.11, p = .007, ηp2 = .313. We calculated the congruency effect of previous trials (i.e., RTs of congruent trials minus incongruent trials). Positive values of this measure denote longer reactions to congruent compared with incongruent trials, and vice versa for negative values. Post hoc tests showed that this congruency effect significantly differs between left-moving (−11 ± 5 msec) and right-moving target dots (13 ± 6 msec), t(20) = −3.02, p = .007. Importantly, there was also was a CSE, that is, the interaction of Previous Trial Congruency × Current Trial Congruency was significant, F(1, 20) = 10.82, p = .004, ηp2 = .351, albeit much smaller than for the accuracy data (cf. effect sizes). For post hoc tests, the congruency effect of previous trials differs significantly between current congruent (−10 ± 4 msec) and current incongruent trials (12 ± 6 msec), t(20) = 3.28, p = .004. No other main or interaction effects were found (all other F(1, 20) ≤ 1.82, p ≥ .192). It has to be noted that RTs were quite long and longer than in the original paradigm developed by Lange-Malecki and Treue (2012). However, in the original paradigm, the SOA was shorter (i.e., 100 msec) than in our study, which used an SOA of 200 msec. This may explain the generally long RTs produced by the participants.
Standard EEG Data
Figures of the standard ERP data can be found in Figure 3.
No significant main or interaction effects were found in the P1 (all F(1, 20) ≤ 3.83, p ≥ .065) and the N2 time window (all F ≤ 3.19, p ≥ .089). The latter is a to-be-expected finding, because the N2 reflects a mixture of different processes (Folstein & Van Petten, 2008), and only a small fraction of these processes is likely to be modulated by perceptual processes focused and manipulated in the paradigm used in the current study. This is clearly demonstrated by the statistical data analysis of the RIDE-decomposed data (see below). In the N1 time window, there was a main effect of Current Trial Congruency, F(1, 20) = 8.70, p = .008, ηp2 = .303. Amplitudes were larger after congruent trials (−7.68 ± 1.20 μV/m2) than incongruent trials (−6.13 ± 1.03 μV/m2). Also, we found an interaction effect between all factors (i.e., Electrode × Direction × Previous Congruency × Current Congruency, F(1, 20) = 6.20, p = .022, ηp2 = .237). Further investigation revealed that this interaction approached significance if target dots moved to the left, F(1, 20) = 4.19, p = .054, ηp2 = .173, but not to the right, F(1, 20) = 0.02, p = .882, ηp2 = .001. This tendency was still evident when we investigated the congruency effect (i.e., congruent − incongruent trials) of the previous trial, F(1, 20) = 4.19, p = .054, ηp2 = .173. However, no statistically significant effects were found for this measure (all t ≤ 1.88, p ≥ .075). To summarize, the standard EEG data do not reflect the pattern of effects observed at the behavioral level. This, however, is an expected finding (cf. Introduction). The effects detailed above are very likely to be inconsistent with the behavioral findings, because the experimental manipulations were designed to modulate perceptual processes; however, the ERP reflects a mixture of processes. This is corroborated by the RIDE data analyses shown below.
RIDE Decomposed Data: S-cluster
Figure 4 shows the S-cluster data.
In the S cluster P1 time window, no significant effects were found (all F ≤ 3.90, p ≥ .062). However, in the N1 time window, we identified a main effect of Current Congruency, F(1, 20) = 8.85, p = .007, ηp2 = .307, with higher mean amplitudes during congruent (−7.98 ± 1.78 μV/m2) compared with incongruent trials (−6.37 ± 1.13 μV/m2). One interaction was found between all experimental factors (i.e., Electrode × Direction × Previous Congruency × Current Congruency), F(1, 20) = 6.27, p = .021, ηp2 = .239. Investigating this effect further, we found this interaction only at left motions, F(1, 20) = 5.91, p = .025, ηp2 = .228, but not for right motions, F(1, 20) = 0.09, p = .772, ηp2 = .004. Furthermore, this interaction was not found for left motions and current congruent trials, F(1, 20) = 1.63, p = .217, ηp2 = .075, but only for left motions and current incongruent trials, F(1, 20) = 4.96, p = .038, ηp2 = .199. In this case, no comparison between factors indicated statistical significance (all t(20) ≤ 1.56, p ≥ .134). However, the congruency effect (i.e., congruent − incongruent) of previous trials significantly differed depending on electrode location, t(20) = 2.23, p = .038 (TP7: 3.13 ± 2.04 μV/m2, TP8: −4.55 ± 2.91 μV/m2). No effects reached statistical significance in this time window (all F ≤ 3.04, p ≥ .096).
In the N2 time window, there was an interaction of Previous Congruency × Current Congruency, F(1, 20) = 5.20, p = .034, ηp2 = .206. Post hoc tests revealed that if the current trial was incongruent, the previous trial congruency modulated the S-cluster N2, t(20) = −2.14, p = .045: In this case, amplitudes were larger after previous congruent (−8.72 ± 2.52 μV/m2) than incongruent (−2.42 ± 1.75 μV/m2) trials (all other t(20) ≤ 1.66, p ≥ .112). Because this interaction corroborates the behavioral data, a sLORETA analysis was conducted and contrasted/compared amplitudes in current incongruent trials if the previous (n − 1) trial was congruent with current incongruent trials that were preceded by an incongruent trial (see Figure 4). This analysis shows that the superior frontal gyrus (BA 6) encompassing the SMA is associated with larger amplitudes in incongruent trials if the previous (n − 1) trial was congruent, compared with when it was incongruent. No further significant effects were observed (all F ≤ 3.16, p ≥ .091).
RIDE Decomposed Data: C-cluster
The C-cluster data are shown in Figure 5.
We analyzed the C-cluster data in the N2 and P3 time window. For the N2 time window, there was a main effect of current trial congruency, F(1, 20) = 4.60, p = .044, ηp2 = .187. Amplitudes were larger during incongruent (−3.35 ± 1.59 μV/m2) compared with congruent trials (−2.19 ± 0.72 μV/m2). There was an interaction of Previous Trial Congruency × Current Trial Congruency, F(1, 20) = 5.98, p = .024, ηp2 = .230. Similarly to the S-cluster N2 data, post hoc tests revealed that previous trial congruency modulated amplitudes differently if the current trial was incongruent, t(20) = −2.47, p = .023, but not when it was congruent, t(20) < 2.01, p > .058. More precisely, amplitudes were larger after previous congruent (−6.31 ± 2.21 μV/m2) than incongruent trials (−0.40 ± 1.73 μV/m2), when the current trial was incongruent. Again, we applied sLORETA to identify the sources. We contrasted/compared amplitudes in current incongruent trials if the previous (n − 1) trial was congruent with current incongruent trials that were preceded by an incongruent trial (see Figure 5). This revealed regions in the right inferior frontal gyrus (rIFG; BA 45). Again, activity in this area was higher in incongruent trials if the previous (n − 1) trial was congruent, compared with when it was incongruent. No effects turned out statistically significant in the N2 time window (all F ≤ 1.61, p ≥ .219).
RIDE Decomposed Data: R-cluster
The R-cluster data are shown in Figure 6.
Only one main effect was found in the response cluster data, namely Current Trial Congruency, F(1, 20) = 5.83, p = .025, ηp2 = .226. Amplitudes were larger during congruent (4.57 ± 0.68 μV/m2) than in incongruent trials (3.09 ± 0.84 μV/m2). The experimental factors showed no interactive effects or other main effects in this cluster (all F ≤ 3.51, p ≥ .076). The lack of interaction of Previous Trial Congruency × Current Trial Congruency was examined in more detail by means of Bayesian analysis using the method of Masson (2011). According to that analysis, the probability for the null hypothesis given the data p(H0/D) is 85.9%. According to Raftery (1995), this provides positive evidence for the null hypothesis. Regarding power and sensitivity considerations (cf. Participants and Power Estimations section), the lack of findings, substantiated by Bayesian analysis, is reliable to interpret. Therefore, this analysis suggests that the CSE is not evident in the response cluster data. This is of theoretical relevance.
We examined the role of perceptual processes for the CSE as an important indicator of postconflict behavioral adjustments and cognitive control. The study was motivated by theoretical considerations put forward in the so-called bottom–up associative account (Egner, 2014; Hommel et al., 2004; Mayr et al., 2003) and related newer conceptual developments (Egner, 2008, 2014; Verguts & Notebaert, 2009), stressing also that perceptual processes play an important role during postconflict behavioral adjustments. Particularly, we were interested whether perceptual coding levels can be isolated in neurophysiological signals during conflict-related behavioral adjustments.
The behavioral data show that reliable CSEs were evoked by the employed experimental paradigm. Especially the accuracy data revealed a clear interaction of previous trial congruency and current trial congruency indicative for a CSE that was stronger as compared with the RT data (ηp2 = .644 for the accuracy data and ηp2 = .351 for the RT data). For the accuracy data, it was shown that whenever the current trial was incongruent, participants responded less accurately after a previous congruent than a previous incongruent trial. This reflects an instance of a CSE (Duthoo et al., 2014; Keye et al., 2013; Schmidt, 2013; Gratton et al., 1992).
Importantly, no consistent effects in line with the behavioral data were observed in standard (nondecomposed) ERP data. This is an expected finding because ERP components and the N2, in particular, reflect activity from different sources and a mixture of different functional processes (Stock et al., 2017; Huster et al., 2015; Folstein & Van Petten, 2008; Nunez et al., 1997) relating to perceptual processes, motor processes, and processing mediating between stimulus processing and responding (Chmielewski et al., 2017; Mückschel, Chmielewski, et al., 2017). Consequently, after EEG decomposition had been applied to dissociated different coding levels in the EEG signal (i.e., RIDE), clear EEG effects in line with the behavioral data and in line with the hypotheses were evident. The RIDE S-cluster reflects stimulus-related perceptional and attentional processes, the R-cluster reflects motor preparation and execution processes, and the C-cluster contains intermediate processes between stimulus evaluation and responding (Ouyang et al., 2011, 2017). As predicted in the hypotheses, only the S-cluster and the C-cluster revealed interactive effects of trial congruency in the n − 1 and the nth trial, which was in line with the behavioral data. No such interactive effects were observed for the R-cluster. This was substantiated by Bayesian statistics. These Bayesian analyses, together with the fact that the study was sufficiently powered, make it possible to interpret this theoretically important dissociation. The dissociation in effects between RIDE clusters is theoretically important, because the CSE has also been explained by mechanisms related to motor conflicts in medial/middle frontal brain areas (Erb, McBride, & Marcovitch, 2019; Erb & Marcovitch, 2018; Folstein & Van Petten, 2008; Botvinick, Cohen, & Carter, 2004; van Veen & Carter, 2002). Because the R-cluster, reflecting motor activation patterns (Ouyang et al., 2015b), did not reveal CSE-like modulations, the data suggest that the CSE can emerge without a pure motor contribution (Verbruggen, Notebaert, Liefooghe, & Vandierendonck, 2006). Rather, the current data suggest that stimulus-related processes (S-cluster) and mechanisms in-between stimulus processing and motor responding (C-cluster) are sufficient for the CSE to occur. In the S-cluster N2 time window, amplitudes in incongruent trials were larger if the previous (n − 1) trial was congruent, compared with when it was incongruent. It has been suggested that cC and iI trial sequences consist of almost complete (perceptual) feature repetitions or alternations that lead to a fast response (Egner, 2014). The likely reason behind this is that feature-binding mechanisms facilitate processing when current trial features “match” the previous event (Egner, 2014; Hommel, 2004). This is not the case when cI (but also iC) trial sequences are presented. Such partial feature repetitions have repeatedly been shown to impose an “unbinding process” to respond correctly on the presented (changed) stimulus that is associated with cost in performance (Egner, 2014; Colzato, Warrens, & Hommel, 2006; Hommel, 2004), as seen in the CSE. The larger amplitude in the in the S-cluster when the n trial was incongruent and the n − 1 trial congruent may reflect intensified stimulus-related unbinding processes. Indeed, unbinding processes are reflected in the N2 time window (Chmielewski & Beste, 2019; Petruo, Stock, Münchau, & Beste, 2016). The source localization data suggest that regions in the middle frontal gyrus (BA 6) encompassing the SMA are associated with the dynamic reflected by the S-cluster. Though these regions are involved in conflict monitoring processes (Herz et al., 2014; Clayson & Larson, 2013; Stock et al., 2013; Mars et al., 2009; Winkel et al., 2009; Nachev et al., 2008; Rushworth et al., 2004), the finding that especially stimulus-related codes seem to be processed during the CSE in this area is novel. Yet, this finding fits to data that these areas process perceptual information during perceptual decision-making (Ruff et al., 2010; Philiastides & Sajda, 2007; Heekeren et al., 2004, 2006). Moreover, these areas are affected by perceptual modulations of cognitive control (Adelhöfer et al., 2019; Labrenz et al., 2012; Westerhausen et al., 2010). Because middle frontal regions show direct structural connections to visual association areas (Hagmann et al., 2008), it is reasonable that the SMA seems to process stimulus codes during conflict-related behavioral adaptation. Intriguingly, these areas are also involved in “unbinding processes” (Elsner et al., 2002), necessary whenever cI trial sequences are presented (Egner, 2014; Hommel, 2004). The higher activity in BA 6 may reflect unbinding processes of perceptual features during the CSE. Interestingly, the C-cluster activity in the N2 time window was associated with the rIFG (BA 45). Activity in this area was lower in incongruent trials if the previous (n − 1) trial was congruent, compared with when it was incongruent. The finding that perceptual processes during conflict monitoring modulate prefrontal brain regions is, therefore, reasonable from a functional neuroanatomical perspective and a cognitive theoretical point of view. It is, therefore, possible to establish a direct relationship between psychological theories about cognitive control and EEG signal decomposition methods.
However, effects were also evident in the C-cluster, but the pattern of modulation was reversed; that is, the C-cluster amplitude was smaller in the N2 time window when the current trial was incongruent and the n − 1 trial congruent. Although this may be regarded to be at odds with the notion that perceptual processes were modulated/manipulated and should exclusively reflected by the S-cluster, it has to be kept in mind that the C-cluster likely reflects stimulus–response transition processes; that is, intermediate processes between stimulus encoding and responding (Bluschke et al., 2017; Mückschel, Chmielewski, et al., 2017; Ouyang et al., 2017; Verleger et al., 2014, 2016, 2017; Wolff et al., 2017). This means that there is still some stimulus-related aspect in the processes reflected by the C-cluster. This is of course also the case for motor processes. Here, however, the data from the R-cluster clearly show that modulations of these processes were present because of the experimental variations. In this respect, the modulations in the C-cluster can rather be explained by a residual influence of perceptual processes on stimulus–response translation processes reflected by the C-cluster. It seems that the process reflected by the C-cluster cannot fully unfold when the current trial was incongruent and the n − 1 trial congruent. It may be speculated that this is due to intensified processes of stimulus unbinding reflected by the S-cluster. The source localization data show that right inferior frontal regions (rIFG, BA 45) are associated with these processes. Similar sources of the C-cluster in the N2 time window have been shown previously (Mückschel, Dippel, et al., 2017), albeit using an inhibitory control task (i.e., a go/no-go task). Yet, inhibitory control processes are important to control unwanted stimulus–response representations or “bindings” (Klein, Petitjean, Olivier, & Duque, 2014; Ocklenburg, Güntürkün, & Beste, 2011; Verleger, Kuniecki, Möller, Fritzmannova, & Siebner, 2009; Stürmer, Siggelkow, Dengler, & Leuthold, 2000). From that perspective, the source localization data reveal activity differences in brain structures that resemble parts of a generic inhibitory control network (Bari & Robbins, 2013). It is possible that inhibitory control mechanisms are less strong in above-mentioned unbinding processes important during sequence effects (Egner, 2014), when there is a concomitantly intensified process related to stimulus unbinding (as reflected by the S-cluster).
It should be noted that the number of RIDE clusters is heuristic and that especially the C-cluster may be further subdivided (Ouyang et al., 2017). Nevertheless, previous findings have also shown that RIDE can be used in a theoretically meaningfully way to dissociate different coding levels in EEG data (Chmielewski et al., 2018; Mückschel, Chmielewski, et al., 2017; Mückschel, Dippel, et al., 2017). Most important, however, is the high theoretical plausibility of findings particularly regarding the dissociation between nondecomposed data and decomposed data. In the standard ERP data, the behavioral effect was not reflected. This was only the case after decomposing the data. Thus, only after applying EEG signal decomposition, a meaningful pattern was evident being able to explain the behavioral effects. The agreement of the effect pattern between behavioral data and decomposed data underlines the reliability of the findings. Regarding the reliability of findings, it has to be noted that the behavioral data suggested that also the direction of moving dots stimulus movement modulated the effect of previous trial congruency. However, because no corresponding effect was found at the EEG level that withstood post hoc testings, we consider this to reflect a spurious effect.
In summary, the study is the first delineating the neural mechanisms underlying modulations of the CSE and conflict-related behavioral adaptations by perceptual factors. The study establishes a direct relationship between psychological concepts focusing on the role of perceptual processes during conflict-related behavioral adaptation and neurophysiological processes using EEG signal decomposition methods. Perceptual processes proposed in cognitive accounts on the CSE can be isolated in EEG signals. The middle frontal and inferior frontal brain regions are important for neurophysiological mechanisms underlying perceptual processes in conflict-related behavioral adjustments. Middle frontal regions are associated with processes dealing with purely perceptual processes (S-cluster); inferior frontal regions are associated with processes dealing with stimulus–response transition processes (C-cluster). Likely, the neurophysiological data reflect unbinding processes at the perceptual level and stimulus–response translation level.
This study was supported by a grant from the Deutsche Forschungsgemeinschaft (DFG BE4045/37-1) and partly by FOR 2698.
Reprint requests should be sent to Christian Beste, Cognitive Neurophysiology, Faculty of Medicine Carl Gustav Carus, Department of Child and Adolescent Psychiatry, TU Dresden, Fetscherstrasse 74, 01307 Dresden, Germany, or via e-mail: firstname.lastname@example.org.