The ability to inhibit prepotent responses is a central facet of cognitive control. However, the role of perceptual factors in response inhibition processes is still poorly understood and an underrepresented field of research. In the current study, we focus on the role of conflicts between perceptual stimulus features (so-called S-S conflicts) for response inhibition. We introduce a novel semantic Stroop Condition task and analyze EEG data using source localization and temporal EEG signal decomposition methods to delineate the neural mechanisms how semantic S-S conflicts modulate response inhibition. We show that semantic conflicts enhance response inhibition performance by modulating neural processes relating to conflict resolution mechanisms in the middle and inferior frontal cortex, as well as the ACC. Opposed to that, Stroop-like (S-S) conflicts compromise response execution by affecting decision processes in inferior parietal cortices. The data suggest that when action control processes and their neurophysiological correlates depend on regions specialized in the processing of semantic conflicts, there is an improvement in response inhibition. The results show that Stroop-like semantic conflicts have opposite effects depending on whether a response has to be executed or inhibited. These opposing effects are then also associated with different functional–neuroanatomical structures. The results of the study show mechanisms by which stimulus-related processes influence mechanisms of response control.
The ability to inhibit prepotent responses is a central facet of cognitive control (Bari & Robbins, 2013; Diamond, 2013). The last years have witnessed a flurry of findings, dealing with that topic using functional imaging and electrophysiological approaches in humans and animals (Bari & Robbins, 2013). Yet, most research still does not consider the possible importance of processes at the stimulus feature level. Nevertheless, accumulating evidence shows that processes at the sensory level play an important role during inhibitory control (Stock, Popescu, Neuhaus, & Beste, 2015; Barry & De Blasio, 2013; Gondan, Götze, & Greenlee, 2010; Huster, Westerhausen, Pantev, & Konrad, 2010; Boehler et al., 2009; Verbruggen, Liefooghe, & Vandierendonck, 2006; Miller, Kühlwein, & Ulrich, 2004; Shedden & Reid, 2001). For example, it has been shown that there are differences between sensory modalities (Bodmer & Beste, 2017) as well as within sensory modalities (Friedrich, Mückschel, & Beste, 2018) to trigger inhibitory control. Also, the integration of stimulus information from different sensory modalities plays an important role (Fiedler, Schröter, & Ulrich, 2011; Gondan et al., 2010; Cavina-Pratesi, Bricolo, Prior, & Marzi, 2001), especially when different sensory modalities convey conflicting information (Chmielewski, Mückschel, Dippel, & Beste, 2016; Chmielewski, Wolff, Mückschel, Roessner, & Beste, 2016). These studies have shown that when different sensory modalities convey contradictory information, this imposes high demands on conflict monitoring processes and cognitive control. The strength of such cross-modal effects differs from those of conflicting sensory information that is presented within a modality. This shows that conflicts at the stimulus level may have a strong effect on inhibitory control processes.
From a cognitive perspective, conflicts can come in different fashions, that is, stimulus–stimulus (S-S) and stimulus-response conflicts (S-R; Hommel, 1997; Kornblum, Hasbroucq, & Osman, 1990; Simon & Berbaum, 1990). The critical distinction between the S-S and S-R conflicts is that the S-S conflict refers to a similarity between stimulus dimensions whereas the S-R conflict refers to how a relevant stimulus feature is mapped on a response (Hasshim & Parris, 2015). S-R conflicts, supposed to drive conflicts in a Simon task (Hommel, 2011; but see Hasbroucq & Guiard, 1992), have recently been shown to modulate response inhibition performance in studies combining a Simon task with a Condition task (Chmielewski, Mückschel, & Beste, 2018; Chmielewski & Beste, 2017). In particular, it has been shown that S-R conflicts (incongruent Simon trials) increase the ability to refrain from responding on no-go trials. Opposed that, when S-R translation processes work rather “automatically” in congruent Simon trials, response inhibition performance is compromised. In sharp contrast to S-R conflicts, the precise impact of S-S conflicts on inhibitory control processes is elusive. A prominent example of an S-S conflict may be the semantic Stroop task (Hommel, 2011; Kornblum et al., 1990). Here, the task-relevant stimulus color (e.g., green) and a task-irrelevant color word identity (e.g., “YELLOW”) are incompatible and hence produce conflicts. Though some evidence suggest that the Stroop conflict is likely not a pure S-S conflict and also includes S-R processes (Hasshim & Parris, 2015; Hommel, 2011; De Houwer, 2003), several lines of evidence show S-S conflicts and independently contributes to the Stroop interference effect (Lansbergen & Kenemans, 2008; Schmidt & Cheesman, 2005; De Houwer, 2003; Zhang & Kornblum, 1998). Interestingly, meta-analytic functional imaging data show that especially S-S conflicts modulate processes in inferior frontal and anterior cingulate areas (Li et al., 2017) and hence areas that are part of a cortical response inhibition network (Aron, Cai, Badre, & Robbins, 2015; Bari & Robbins, 2013). This functional neuroanatomical overlap makes it likely that response inhibition processes are affected by S-S conflicts. The reason is the prefrontal neurons/circuits have been shown to respond to very different stimuli, cognitive operations, and motor responses under different task conditions (Duncan, 2010; Duncan & Owen, 2000). Neural processes, for example, those related to domain-unspecific “inhibitory control,” may act on different aspects during information processing. Resolving semantic S-S conflicts requires the suppression of irrelevant semantic information (Li et al., 2017; Coderre & van Heuven, 2013). Because neural processes subserving inhibitory control are already needed for processing the semantic stimulus-related conflict (Li et al., 2017), they may concomitantly facilitate neural processes related to the inhibition of motor responses that occur in similar or overlapping neural circuits. Importantly, the difficulty to exert inhibitory control depends on the strength of automated response tendencies, that is, the more the response runs automatically, the more faulty/difficult is the inhibition of the response (Chmielewski et al., 2018; Chmielewski & Beste, 2017; Dippel, Chmielewski, Mückschel, & Beste, 2016; Stevenson, Russell, & Helton, 2011; Helton, 2009; McVay & Kane, 2009). Because conflicts make the behavior less dependent on a rather unsupervised or automated execution of responses (Botvinick, Braver, Barch, Carter, & Cohen, 2001), it is likely that S-S conflicts increase the ability to perform motor inhibitory control. We examine this hypothesis using a novel combination of a semantic Stroop task with a Condition task. In particular, we hypothesize that response inhibition is worse in nonconflicting, compared with conflicting, Stroop no-go trials. We set out to investigate this hypothesis considering neurophysiological mechanisms by means of ERP, source localization, and temporal EEG signal decomposition methods. As outlined below, ERP methods may be especially useful in discerning how much S-S and S-R conflicts impact response inhibition.
EEG studies show that, especially semantic, Stroop conflicts are reflected in an increased (i.e., more negative) N450 ERP amplitude (Barbet & Thierry, 2018; Schreiter, Chmielewski, & Beste, 2018a, 2018b; Imbir, Spustek, Duda, Bernatowicz, & Żygierewicz, 2017; Chuderski, Senderecka, Kałamała, Kroczek, & Ociepka, 2016; Larson, Clayson, Kirwan, & Weissman, 2016; Larson, Clayson, & Clawson, 2014; Szűcs & Soltész, 2012; Tillman & Wiens, 2011; Szűcs & Soltész, 2010). This ERP component is associated with ACC (Carter & van Veen, 2007; Botvinick, Cohen, & Carter, 2004). However, the N450 is not the only ERP component that is important to consider. Another relevant ERP correlate is the late-stage centroparietal positivity (CSP) that is generated in subdivisions of the pFC (West, Jakubek, Wymbs, Perry, & Moore, 2005; West, 2003). The CSP is more enhanced (i.e., more positive) following conflicting (incongruent) trials compared with nonconflicting (congruent) trials and has been associated with conflict resolution processes (Larson et al., 2014). Of note, recent data from an experiment systematically varying Simon (S-R) and Stroop (S-S) conflicts (Chmielewski & Beste, 2019) showed that S-R and S-S conflicts are processed at different time points. It has been shown that Stroop processes do not modulate processes in the N2 time window. Yet, this was the case for Simon conflicts that clearly reflect S-R conflict effects (Chmielewski & Beste, 2019). Modulations observed in the N450/CSP ERP components indicate the processing of the S-S conflict part involved during Stroop information processing (Chmielewski & Beste, 2019). This may be the case because Stroop-like S-S conflicts require time-consuming processing of semantic information also involving the suppression of irrelevant semantic information (Li et al., 2017; Coderre & van Heuven, 2013; Lei et al., 2013). A result in which especially later time windows show experimental modulation of neurophysiological data therefore indicates that the S-S conflict part, but not so much the S-R conflict part, of Stroop-like stimuli modulates response inhibition processes. Also, another study (Lansbergen, van Hell, & Kenemans, 2007) provides evidence for an electrophysiological distinction between S-S and S-R conflicts.
Because such long-latency ERP components are prone to high intraindividual variability (Ouyang, Hildebrandt, Sommer, & Zhou, 2017; Ouyang, Sommer, & Zhou, 2015; Ouyang, Schacht, Zhou, & Sommer, 2013; Kutas, McCarthy, & Donchin, 1977), it is important to control for that to detect effects of experimental manipulations with high reliability. This can be achieved using residue iteration decomposition (RIDE; Ouyang et al., 2013). The method decomposes EEG data into three component clusters of distinct functional relevance: the S-cluster, the C-cluster, and the R-cluster. These different component clusters then comprise conventional ERP components with different peaks, topographies, and functional significance (Ouyang, Herzmann, Zhou, & Sommer, 2011), with the advantage that intraindividual variability has been corrected. The S-cluster refers to stimulus-related processes and therefore contains modulations of the P1, N1, and parts of N2 (Mückschel, Chmielewski, Ziemssen, & Beste, 2017; Ouyang et al., 2015). The C-cluster refers to intermediate processes like stimulus evaluation and response selection (Ouyang et al., 2011, 2017). It has been shown that Stroop conflict-related ERP correlates are represented by the C-cluster (Schreiter et al., 2018a). Moreover, it has been shown that ERP correlates of inhibitory control, like the no-go N2 and no-go P3 (Huster, Enriquez-Geppert, Lavallee, Falkenstein, & Herrmann, 2013), are represented in the C-cluster (Ouyang et al., 2013). Therefore, we hypothesize that modulations of neurophysiological activity in late time ranges (similar to the time range of the N450 or CSP) are evident between conflicting and nonconflicting no-go trials in the C-cluster. We assume that the C-cluster is more positive during conflicting than nonconflicting no-go trials in late time windows. In the source localization analysis, we hypothesize that above-mentioned modulations are associated with activation differences in the ACC as well as lateral inferior and middle frontal areas. The reason is that EEG (Lansbergen & Kenemans, 2008) and recent meta-analytic evidence from fMRI data (Li et al., 2017) show that S-S conflicts are processed inside and outside the ACC (in the right inferior frontal cortex) as well as the ACC and that inferior frontal areas play a central role inhibitory control processes (Aron et al., 2015; Bari & Robbins, 2013).
Considering trials in which a response has to be executed, the common finding is that conflicts compromise performance (Botvinick et al., 2004). Therefore, we expect worse performance in conflicting go trials than nonconflicting go trials. The C-cluster has been shown to reveal modulatory effects known from the standard P3b ERP component (Ouyang et al., 2017; Verleger, Grauhan, & Śmigasiewicz, 2016; Verleger, Metzner, Ouyang, Śmigasiewicz, & Zhou, 2014), which may overlap with modulatory processes in the N450 time range. Moreover and similar to the P3 (Geng & Vossel, 2013), sources of C-cluster modulation in the P3 time window have been associated with the TPJ (BA 40; Wolff, Mückschel, & Beste, 2017). The P3b is associated with the TPJ, is smaller during more demanding (conflicting) choice trials (Falkenstein, Hohnsbein, & Hoormann, 1994), is modulated during Stroop-like information processing (Lansbergen & Kenemans, 2008; Lansbergen et al., 2007), and likely reflects the formation of a decision using accumulated stimulus information (Twomey, Murphy, Kelly, & O'Connell, 2015; O'Connell, Dockree, & Kelly, 2012). Therefore, we expect that the C-cluster in the P3 time window is smaller during conflicting go than nonconflicting go trials.
Taken together, we expect that S-S conflicts or the S-S conflict part during Stroop information processing modulates neurophysiological processes at different time points and in different neuroanatomical structures, depending on whether a response has to be executed or to be inhibited. These differential modulations may also then be associated with the probably opposing behavioral effects of S-S conflicts on response execution and inhibition.
The study used a within-subject design. We conservatively assumed a moderate effect size for condition-related effects (i.e., ηp2 = .10). On that basis, the power analysis revealed that power greater than 95% is achieved with n = 22 participants. Note that the actually obtained effects sizes in were much larger (cf. Results section). We included n = 22 young, healthy participants (11 women) between 19 and 30 years (mean age = 25.3 ± 0.7 years) and collected their written informed consent for the participation in this study. All participants reported no psychiatric or neurological disorders and current use of medication and had normal or corrected-to-normal vision. This study was conducted after approval from the institutional review board of the medical faculty of the Technische Universität Dresden.
To examine how stimulus feature conflicts (S-S conflicts) affect response inhibition, we combined a Stroop task with a Condition task. The experimental conditions of the task are shown in Figure 1.
We presented colored stimuli within a white frame box in the middle of a monitor (black background, 58 cm distance to the participant). A white fixation cross was presented within the white box between trials. Trials were separated by an intertrial interval, which was jittered between 1300 and 1700 msec. Trials began with the presentation of a color word (German grün for “green” or German gelb for “yellow”) in a congruent or incongruent color (i.e., green or yellow), thus creating the Stroop component of the task. Importantly, these color words were either presented in roman font (“green” or “yellow” ➔ go trials) or in bold-italic font (“green” or “yellow” ➔ no-go trials) to indicate whether to respond or to withhold the “manual” response. In go trials (roman font), participants were required to report the actual color of the word (green ➔ left hand response on left CTRL button, yellow ➔ right hand response on right CTRL button) within 1000 msec, regardless of the meaning of the color word (“green” or “yellow”). Hence, congruent (“green” in green or “yellow” in yellow) and incongruent (“green” in yellow or “yellow” in green) go trials were employed. These color words were chosen because both words consist of the same number of letters and start with the same letter. These aspects make it difficult to use superficial perceptual aspects as a decision strategy to respond to these stimuli and force the participant to fully evaluate the stimuli. In no-go trials (bold-italic font), participants were required to withhold the response regardless of the meaning (i.e., “green” or “yellow”) and the actual color of the word (i.e., green or yellow). Hence, congruent (“green” in green or “yellow” in yellow) and incongruent (“green” in yellow or “yellow” in green) no-go trials were presented. These conditions are shown in Figure 1. The stimuli used in the experiment were matched in luminance.
If no response was recorded in go or no-go trials, trials ended after 1700 msec. Go trials were coded as hits if a correct response was executed within a time window of 1000 msec after stimulus presentation. If an incorrect response was recorded in this time window, go trials were coded as errors, and if no response was obtained, go trials were coded as misses. No-go trials were coded as correct if no response was recorded. No-go trials were coded as a false alarm if a response was executed within 1000 msec after stimulus presentation. We collected these behavioral measures for congruent and incongruent go and no-go trials separately. In total, 720 trials were presented. These consisted of 252 congruent go (35% of all trials; the same amount of trials for each possible stimulus configuration), 252 (35%) incongruent go, 108 congruent no-go (15%), and 108 incongruent no-go (15%) trials. The 720 trials were divided into six blocks. Trial types were equally distributed in each block and were randomly presented within these blocks. To familiarize participants with the experiment, participants conducted a short exercise of 40 trials.
EEG Recording and Analysis
EEG data were collected with a setup of 60 Ag/AgCl electrodes (reference at electrode Fpz and ground at θ = 58, ф = 78) connected to a BrainAmp amplifier (Brain Products, Inc.). Data were recorded with a sampling rate of 500 Hz, and the electrode impedance was kept below 5 kΩ during recording (online recording filter settings: 0.1–80 Hz). Offline data processing was performed using the BrainVision Analyzer 2 software package (Brain Products, Inc.). We then used a bandpass filter with a slope of 48 dB/oct between 0.5 and 20 Hz. Then, technical artifacts were manually excluded by means of a raw data inspection. Periodically reoccurring artifacts, which were caused by pulse or by vertical/horizontal eye movements, were identified and corrected using an independent component analysis (infomax algorithm). In the next step, we segmented the data for each experimental condition to obtain segments for congruent and incongruent go and no-go trials. Only correct go and correct no-go trials were included in the further data analysis steps. The segmentation began 250 msec before and stopped 1000 msec after stimulus onset (Time Point 0). Subsequently, we applied an automated artifact rejection procedure utilizing a maximal difference of 150 μV in a 200-msec time window and an activity below 0.5 μV in a 100-msec time window as criteria for rejection. In the next step, a current source density (CSD) transformation was conducted to eliminate reference potentials from the data and to help to identify the specific electrodes displaying the strongest effects (Nunez & Pilgreen, 1991). Values obtained after the CSD transformation are reported in μV/m2. Afterward, we conducted a baseline correction, which was applied in a time period from −200 to 0 msec before stimulus presentation, before averaging the segments.
We measured ERP components at the single-subject level by means of mean amplitude in defined time windows. We chose time intervals and particular electrodes based on the visual inspection of the ERPs and scalp topography plots. This visual inspection was based on time intervals reported in previous literature. Importantly, we validated the selection of time intervals and electrodes using the following statistical approach (Mückschel, Stock, & Beste, 2014): We extracted the mean amplitude for all electrodes in the previously selected time intervals. We compared each electrode to an average of the other electrodes using Bonferroni correction for multiple comparisons (critical threshold, p = .0007). This was done across participants and was done for each condition separately, that is, the experimental factors were not directly used to extract the relevant electrode sites. We exclusively quantified amplitudes at electrodes that differed significantly to the other electrodes (negative for negative potentials and positive for positive potentials). Electrodes selected on the basis of this statistical approach were identical to electrodes identified by means of the visual inspection. We consequently used the following electrode sites and time intervals to quantify the ERP components: P1 (93–118 msec) and N1 (151–176 msec) were quantified at electrodes P7 and P8, N2 (258–348 msec) was quantified at electrode Cz, the parietal P3 was quantified at electrodes PO1 and PO2 (325–365 msec), the frontocentral P3 was quantified at electrode Cz (480–520 msec), and the CSP (480–620 msec) was quantified at electrode Pz. The ERP data revealed no N450 ERP component (cf. Figure 2), which was hence not quantified.
Residue Iteration Decomposition
We applied RIDE using the RIDE toolbox (available on cns.hkbu.edu.hk/RIDE.htm) according to established protocols for Condition tasks (Chmielewski et al., 2018; Mückschel, Chmielewski, et al., 2017; Ouyang et al., 2011). To decompose ERP components, RIDE applies L1-norm minimization to obtain median waveforms. Information about the mathematical details of the RIDE method can be found elsewhere (Ouyang et al., 2011, 2013, 2015). Of note, spatial filter properties of the CSD do not violate RIDE assumptions because RIDE decomposes the signal for each electrode separately (Ouyang et al., 2015). RIDE is used to decompose ERP signals into clusters that are correlated to the stimulus onset (S-cluster) or to a central cluster between stimulus and response (C-cluster). In Condition tasks, it is not possible to reliably calculate the R-cluster because of a lack of a motor response (RTs) in no-go trials (Ouyang et al., 2013). The RIDE algorithm has repeatedly been used in Condition tasks (Mückschel, Dippel, & Beste, 2017; Ouyang et al., 2013). The C-cluster was obtained by applying the RIDE self-optimizing, nested iteration scheme for latency estimation, which is based on an estimate of the initial latency of the C-cluster. Subsequently, the S-cluster is removed, and the C-cluster latency is reestimated by means of a template-matching procedure until convergence of the initial latency estimation and the S-cluster and C-cluster is reached. We set the initial time interval for the C-cluster estimation to 200–1000 msec after stimulus onset. The time interval for the S-cluster estimation was set to −200 to 700 msec around stimulus onset. Except for the exact choice of the time windows, we used the same procedure as used in previous studies (Mückschel, Dippel, et al., 2017; Ouyang et al., 2013). These time windows were chosen, as Stroop conflicts are resolved at later stages of response selection (see the Introduction section). To quantify the S-cluster and the C-cluster, we used the same visual inspection and statistical validation procedure as for the ERP component data. S-cluster data were quantified at electrodes P7/P8 for go and no-go trials. For activity in the P1 time window, the mean amplitude in the time interval from 95 to 120 msec after stimulus onset was quantified. For activity in the N1 time window, we used the time interval from 150 to 175 msec after stimulus onset. Regarding the C-cluster data, we analyzed the activity in the P3 time window at electrodes PO1 and PO2 in the time interval 450–520 msec after stimulus onset for go and no-go trials. Additionally, we analyzed the data in the CSP time window at electrodes Cz and CPz in the time interval of 650–740 msec after stimulus onset for go and no-go trials.
We performed source localization with sLORETA (standardized low-resolution brain electromagnetic tomography; Pascual-Marqui, 2002) using the RIDE data. The reason is that only the RIDE data revealed interaction effects explaining the behavioral effects (see the Results section). sLORETA shows no localization bias and gives a single linear solution to the inverse problem (Marco-Pallarés, Grau, & Ruffini, 2005; Sekihara, Sahani, & Nagarajan, 2005; Pascual-Marqui, 2002). The validity of sLORETA source localization results is underlined by several EEG/fMRI and EEG/TMS studies (Dippel & Beste, 2015; Sekihara et al., 2005), and sLORETA has been previously conducted using RIDE data (Chmielewski et al., 2018). sLORETA partitions the intracerebral volume into 6239 voxels with a spatial resolution of 5 mm. sLORETA uses the standardized current density at each voxel, which is calculated in a realistic head model based on the MNI152 template. Because this study focuses on the modulation of RIDE clusters during response inhibition processes by the congruent and incongruent conditions, the voxel-based sLORETA contrasts were calculated between congruent and incongruent trials. The statistical comparisons were conducted using statistical nonparametric mapping, applying the sLORETA built-in voxel-wise randomization tests with 2500 permutations. Voxels, in which significant differences (p < .01, corrected for multiple comparisons) between contrasted conditions occurred, were located in the MNI152 brain.
We analyzed the behavioral data with dependent-samples t tests. We analyzed the neurophysiological data (i.e., ERP components and RIDE S-cluster and C-cluster data) with repeated-measures ANOVAs using the within-subject factors Condition (go vs. no-go) and Congruency (congruent vs. incongruent). The additional within-subject factor Electrode was included, where necessary. We applied Greenhouse–Geisser whenever needed and used Bonferroni correction for all post hoc tests. All variables included in the analyses had a normal distribution as indicated by the Kolmogorov–Smirnov tests (all zs < 0.81; p > .2).
For go trials, paired t tests showed that hit rates were lower in incongruent (87.10 ± 1.68%) than congruent go trials (92.34 ± 1.13%), t(19) = −4.58, p < .001. Moreover, RTs were longer in incongruent (602 ± 19 msec) than congruent go trials (570 ± 16 msec), t(19) = 6.05, p < .001. For no-go trials, paired t tests showed that false alarm rates were lower in incongruent (7.18 ± 1.76%) than congruent no-go trials (8.56 ± 2.08%), t(19) = −1.98, p = .037. Thus, the direction of Stroop conflict effects is opposed in no-go trials, compared with go trials.
Conventional ERP Components
The conventional ERP components are shown in Figure 2.
In the repeated-measures ANOVA for the P1 ERP component (Figure 2A), a main effect of Condition was observed, F(1, 19) = 12.44, p = .002, ηp2 = .396, showing increased P1 amplitudes in no-go (21.22 ± 4.24 μV/m2) in comparison with go trials (18.37 ± 4.09 μV/m2). All other main or interaction effects did not reach significance (all Fs ≤ 3.45, p ≥ .079). In the repeated-measures ANOVA for the N1 ERP component (Figure 2A), the main effect of Condition was observed as well, F(1, 19) = 9.18, p = .007, ηp2 = .326, showing N1 amplitudes to be larger (more negative) in no-go (−19.37 ± 5.69 μV/m2) than go trials (−15.90 ± 5.34 μV/m2). Moreover, a Congruency × Electrode interaction was observed, F(1, 19) = 16.05, p = .001, ηp2 = .458. Post hoc paired t tests showed that this related to increased (in negativity) N1 amplitudes in congruent (−18.12 ± 5.34 μV/m2) compared with incongruent trials (−16.60 ± 5.42 μV/m2) at electrode P7, t(19) = −3.01, p = .007, but decreased amplitudes in congruent (−16.89 ± 7.08 μV/m2) compared with incongruent trials (−18.93 ± 6.88 μV/m2) at electrode P8, t(19) = 2.53, p = .020. All other main or interaction effects did not reach significance (all Fs ≤ 3.34, p ≥ .083).
In the repeated-measures ANOVA for N2 (Figure 2B), only a main effect of Condition was observed, F(1, 19) = 4.63, p = .045, ηp2 = .196, showing larger (more negative) N2 amplitudes in no-go (−16.28 ± 2.49 μV/m2) than go trials (−14.44 ± 2.68 μV/m2). All other main or interaction effects did not reach significance (all Fs ≤ 2.62, p ≥ .122). The P3 at frontocentral sites (i.e., Cz; Figure 2B) only revealed a main effect of Condition, F(1, 19) = 14.44, p < .001, ηp2 = .46, showing that the amplitude were larger (i.e., more positive) in no-go trials (6.44 ± 2.12 μV/m2) than go trials (−8.23 ± 1.55 μV/m2). This reflects a common finding of the frontocentral positivity in no-go trials (Huster et al., 2013).
For the P3 at parietal sites (Figure 2C), the repeated-measures ANOVA only revealed a Congruency × Electrode interaction, F(1, 19) = 12.08, p = .003, ηp2 = .369. When comparing congruent versus incongruent trials at PO1 and PO2 and when comparing PO1 versus PO2 for congruent and incongruent trials by means of post hoc paired t test, this interaction, however, did not survive post hoc comparisons (all ts ≤ 1.55, p ≥ .137). Moreover, the Condition × Congruency interaction did barely miss significance, F(1, 19) = 3.93, p = .062, ηp2 = .171, and all other main and interaction effects did not reach significance (all Fs ≤ 1.49, p ≥ .237).
In the CSP time window (Figure 2C), the repeated-measures ANOVA revealed no significant main or interaction effects (all Fs ≤ 1.66, p ≥ .214). Importantly, reliable effects in the CSP time window were obtained in the C-cluster (i.e., after applying RIDE), which shows that the lack of effects in the standard ERP data is due to a strong intraindividual variability in the data.
The S-cluster, including scalp topography plots, is shown in Figure 3.
The S-cluster was quantified in the P1 and N1 time window. In the repeated-measures ANOVA for the P1 S-cluster amplitude, only a main effect of Condition was observed, F(1, 19) = 12.30, p = .002, ηp2 = .393, showing larger amplitudes in the no-go (21.22 ± 4.23 μV/m2) than go trials (18.38 ± 4.08 μV/m2). There were no additional main effects or interactions (all Fs ≤ 3.43, p ≥ .080). In the repeated-measures ANOVA for the N1 S-cluster amplitude, the main effect of Condition was also observed, F(1, 19) = 6.35, p = .021, ηp2 = .251, showing amplitudes to be increased (in negativity) in the no-go (−18.38 ± 5.59 μV/m2) in comparison with the go condition (−15.80 ± 5.26 μV/m2). Moreover a Congruency × Electrode interaction was observed, F(1, 19) = 19.12, p < .001, ηp2 = .502. Post hoc paired t tests showed that this interaction related to an increased amplitude for congruent (−17.69 ± 5.40 μV/m2) compared with incongruent trials (−15.95 ± 5.48 μV/m2) at the P7, t(19) = −3.22, p = .004, whereas incongruent trial amplitudes (−18.43 ± 6.70 μV/m2) were increased compared with congruent trials (−16.30 ± 6.94 μV/m2) at the P8, t(19) = −2.76; p = .013. There were no additional main effects or interactions (all Fs ≤ 1.87, p ≥ .188).
Because the lack of a Condition × Congruency interaction effect in the P1 and N1 time window in contrast to obtained effects in the C-cluster (see below) is of theoretical relevance, we examined the lack of interaction in more detail using Bayesian analysis. We applied the method by Masson (2011). With this method, the probability of the null hypothesis being true given the obtained data p(H0/D) can be calculated on the basis of the ANOVA results. For the Condition × Congruency interaction effects, this Bayes statistic revealed p(H0/D) = 0.95. According to Raftery (1995), this provides very strong evidence for the null hypothesis.
The C-cluster, including scalp topography plots, is shown in Figure 4.
In the repeated-measures ANOVA for the parietal C-cluster positivity in the P3 time window, a Condition × Congruency interaction was detected, F(1, 19) = 6.51, p = .019, ηp2 = .255. Post hoc paired t tests revealed that this interaction was based upon a significantly increased parietal positivity in the congruent (16.59 ± 1.95 μV/m2) compared with the incongruent go condition (13.46 ± 1.84 μV/m2), t(19) = 3.11, p = .006. In contrast to that, congruent (13.91 ± 1.88 μV/m2) and incongruent no-go trials (14.83 ± 1.82 μV/m2) did not differ from each other, t(19) = −0.91, p = .374. The sLORETA analysis revealed that differences between congruent and incongruent go trials were associated with activation differences in the right inferior parietal cortex (BA 40) encompassing the TPJ (congruent > incongruent). All other main and interaction effects did not reach significance (all Fs ≤ 2.16, p ≥ .292). As can be seen in Figure 4B (top plot), there is a strong frontocentral positivity in the C-cluster, which has previously been suggested to reflect processes similar to the frontocentral (no-go) P3 (Chmielewski et al., 2018), that is, the implementation of motor inhibitory control (Kenemans, 2015; Wessel & Aron, 2015; Huster et al., 2013). However, this frontocentral positivity only revealed a main effect of Condition, F(1, 19) = 25.83, p < .001, ηp2 = .599, with the amplitude being larger in no-go trials (14.81 ± 3.15 μV/m2) than go trials (−3.42 ± 2.17 μV/m2). There were no differential effects of Condition × Congruency (F = 0.35, p > .677) that may explain the behavioral pattern of results. For the Condition × Congruency interaction, this Bayes statistic revealed p(H0/D) = 0.98. According to Raftery (1995), this provides very strong evidence for the null hypothesis.
In the repeated-measures ANOVA for the amplitude modulations in the CSP time window, main effect of Congruency, F(1, 19) = 6.00, p = .024, ηp2 = .240, showed that there were larger amplitudes in incongruent (4.07 ± 1.51 μV/m2) than congruent trials (2.14 ± 1.66 μV/m2). Most importantly, a Condition × Congruency interaction was observed, F(1, 19) = 10.06, p = .005, ηp2 = .346. Post hoc paired t tests showed that this interaction related to increased amplitudes in incongruent (4.76 ± 1.44 μV/m2) compared with congruent no-go trials (1.09 ± 1.49 μV/m2), t(19) = 3.80, p = .001, whereas incongruent (3.37 ± 1.90 μV/m2) and congruent go trials (3.18 ± 2.05 μV/m2) did not differ from each other, t(19) = 0.20, p = .846. The sLORETA analysis revealed that differences between congruent and incongruent no-go trials were associated with activation differences in the ACC (BA 24), as well as the middle and inferior frontal cortices (BA 9, BA 46; congruent < incongruent). All other main and interaction effects did not reach significance (all Fs ≤ 1.66, p ≥ .366).
In the current study, we examined whether and how S-S conflicts modulate motor inhibitory control. We developed a novel Stroop Condition paradigm and examined neurophysiological processes. In short, the data show that S-S conflicts enhance inhibitory control processes.
For the go trials, the behavioral data revealed the usual pattern of performance declines in incongruent as compared with congruent trials. However, most important is the data on the false alarms in no-go trials. Here, it is shown that the rate of false alarms was lower in incongruent than congruent no-go trials. Go and no-go trials thus show an opposing pattern of modulation by S-S conflicts: S-S conflicts compromise performance in response execution but enhance performance in response inhibition. This reversed pattern is discussed further below. The pattern observed in no-go trials is in line with the study's hypothesis and can well be explained with the conflict monitoring framework: This framework supposes that the detection of conflict increases cognitive control mechanisms (Botvinick et al., 2001), making the behavior less dependent on a rather unsupervised or automated execution of responses. Importantly, there is an inverse relationship between the difficulty to exert inhibitory control and the degree of automated response tendencies (Stevenson et al., 2011; Helton, 2009; McVay & Kane, 2009), that is, the more the response runs automatically, the more error prone is the inhibition of the response. The behavioral data thus show that the emergence of an S-S conflict helps to overcome a rather automated mode of responding, thereby increasing the ability to inhibit responses in no-go trials. However, it is crucial to note that this contrasts with findings showing that response inhibition performance declines when conflicting information is presented using different sensory modalities (Chmielewski, Wolff, et al., 2016), for example, the visual and auditory modalities. The likely explanation for this is that the S-S conflict presented in the current study does not involve a shift between sensory modalities or a cross-modal integration of information. This suggests that the effect of an S-S conflict further depends on the complexity to integrate different features of information or sensory modalities.
The neurophysiological data show that distinct subprocesses are associated with these clearcut behavioral effects. As expected, there were no modulations in standard ERP data that could explain the behavioral effects. The likely reason is that neurophysiological processes showing modulations depending on the presence of conflicting or nonconflicting stimulus features usually show longer latencies, known to be prone to strong intraindividual variability (Ouyang et al., 2013, 2015, 2017; Kutas et al., 1977). Supporting this, the data show reliable effects in line with the behavioral data after accounting for intraindividual variability using RIDE. However, also for RIDE, effects were specifically observed in the C-cluster, but not in the S-cluster. The lack of interactive effects in the S-cluster P1 and N1 time windows, which was supported by Bayesian data analysis, suggests that lower level perceptual gating and attentional selection processes (Herrmann & Knight, 2001) are not important to consider to explain the behavioral effects. During go trials, conflicts modulated the C-cluster amplitude in the P3 time window, with the C-cluster being larger during congruent than incongruent trials. This modulation was associated with activation differences in the TPJ (BA 40). The C-cluster in the P3 time window has been shown to reveal modulatory effects known from the standard P3b (Ouyang et al., 2017; Verleger et al., 2014, 2016). The P3b is associated with the TPJ (Geng & Vossel, 2013), is smaller during more demanding (conflicting) choice trials (Falkenstein et al., 1994), and likely reflects the formation of a decision using accumulated stimulus information (Twomey et al., 2015; O'Connell et al., 2012). It is thus likely that conflicts during go trials compromise the decision which response to execute, and this is reflected by the C-cluster modulations. However, during no-go trials, conflicts also modulated C-cluster amplitudes, but in later time windows than in go trials and also in different functional neuroanatomical regions. It is shown that the amplitude of the C-cluster in the CSP time window was more positive in incongruent trials. Such modulations have been suggested to reflect ongoing conflict resolution processes (Larson et al., 2014). The source localization data suggest that differences between incongruent and congruent trials are associated with the ACC (BA 24), as well as middle frontal and inferior frontal cortices (BA 9, BA 46). Interestingly, the strong frontocentral positivity in the C-cluster, which has previously been suggested to reflect processes similar to the no-go P3 (Chmielewski et al., 2018), that is, the implementation of motor inhibitory control (Kenemans, 2015; Wessel & Aron, 2015; Huster et al., 2013), did not show modulatory effects between conflicting and nonconflicting trials. The same was the case for the no-go P3 in the standard ERP data. This was supported by a Bayesian analysis. Thus, effects of Stroop-like (S-S) conflicts do not primarily affect the implementation of motor inhibitory control. Yet, the finding that modulations of the CSP were observed that can well explain the behavioral effects may be interpreted that the CSP reflects a signal (from ACC) to implement more cautious settings after experienced conflict. This can then affect (increase) behavioral performance in response inhibition. Previous results suggest that modulations of longer latency ERP correlates of conflict monitoring (i.e., the N450 or the CSP) are more affected by S-S conflict than S-R conflicts during Stroop information processing (Chmielewski & Beste, 2019). This may be the case because Stroop-like S-S conflicts require time-consuming processing of semantic information involving the suppression of irrelevant semantic information (Li et al., 2017; Coderre & van Heuven, 2013; Lei et al., 2013). The current results on no-go trials suggest that, particularly, S-S conflicts increase performance in inhibitory control. On go trials, however, a response to the Stroop stimulus is required, which means that there may be an S-R conflict as well as an S-S conflict, so the S-R conflict could be causing the effect on go trials. Indeed, recent findings suggest that S-R- and S-S-related conflict monitoring processes can become integrated with each other during response execution (Chmielewski & Beste, 2019). Even if this has to be acknowledged, the results show that Stroop-like conflicts have opposite effects, depending on whether a response has to be executed or inhibited. These opposing effects are then also associated with different functional–neuroanatomical structures.
In order for Stroop-like (S-S) conflicts to effectively modulate inhibitory control processes, it is necessary that cognitive operations involved in conflict resolution have close access to processes mediating motor response inhibition. Several lines of evidence suggest that the TPJ performs “contextual updating” of task-appropriate actions based on accumulating sensory information (Geng & Vossel, 2013). Parietal mechanisms are also important to guide the rapid selection of appropriate motor effectors (response channel; Sulpizio et al., 2017; Bernier, Cieslak, & Grafton, 2012; Cisek & Kalaska, 2002). This may explain the involvement of the TPJ during conflict modulations in go trials, also because these regions have been shown to play a role in conflict monitoring regardless of the type of conflict (Li et al., 2017). Crucially, especially semantic Stroop conflicts need further processes associated with middle and inferior frontal regions, as well as in ACC (Li et al., 2017). Probably, this is because Stroop-like (semantic S-S) conflicts require in-depth, time-consuming processing of semantic information. However, parietal regions, as opposed to above-mentioned frontal regions, have not been related to the processing of semantic (conflicting) information (Friederici, 2011). In this respect, it is likely that the semantic conflict cannot be resolved during go trials, and performance becomes worse. To understand why semantic conflicts increase performance in no-go trials, it is important to consider that the time-consuming processing of semantic information involves the suppression of irrelevant semantic information (Li et al., 2017; Coderre & van Heuven, 2013; Lei et al., 2013). This suggests that inhibitory control processes are already involved to resolve the semantic Stroop (S-S) conflict. However, it should be noted that meta-analytical findings report that left-hemispheric inferior and middle frontal regions are involved in semantic conflict processing, which may be explained by the fact language processes are involved (Li et al., 2017). In our study, activation differences were observed in the right hemisphere. This, however, can be explained by the fact that response inhibition processes require right-lateralized cortical networks (Aron et al., 2015; Bari & Robbins, 2013). Areas involved in the processing of semantic conflicts are also part of a cortical response inhibition network (Bari & Robbins, 2013). Especially inferior frontal regions have been suggested to implement “braking” processes needed to inhibit a motor response (Aron et al., 2015). Recent data also suggest that the C-cluster modulations during no-go trials are associated with these braking processes in inferior frontal structures (Mückschel, Dippel, et al., 2017). Because inhibitory processes are already needed to process the Stroop-like conflict, it is likely that these also inform response inhibition processes. An alternative scenario could be that Stroop conflict delays the go process; therefore, motor inhibitory control processes (frontal P3) have more opportunity to suppress the go process (on successful no-go trials: less false alarms during Stroop conflict), but motor inhibitory control (frontal P3) itself is not affected by Stroop conflict.
Regarding these considerations and the interrelation of conflict processes and functional neuroanatomical regions, it is interesting that there are likely multiple control mechanisms depending on the nature of conflict (i.e., S-S and S-R), which also show distinct functional neuroanatomical foundations (Zmigrod, Zmigrod, & Hommel, 2016; Egner, 2008). Importantly, the current data suggest that the neurophysiological processes and associated functional neuroanatomical structures involved further depend on whether a response has to be executed or inhibited. This aspect has, until now, not been considered in theoretical frameworks on conflict monitoring. The presented paradigm may represent a novel way of measuring S-S conflicts, because the paradigm presented avoids the issue of needing a specific baseline by measuring the effect of S-S conflict rather than the conflict itself. This is a problem with other studies that has more recently been criticized (Hasshim & Parris, 2015).
In summary, we examined how semantic S-S conflicts (Stroop conflicts) modulate response inhibition processes. We show that semantic S-S conflicts enhance response inhibition performance by modulating neural processes relating to conflict resolution mechanisms in the middle and inferior frontal cortex, as well as the ACC. Opposed to that, semantic Stroop (S-S) conflicts compromise response execution by affecting decision processes in inferior parietal cortices. Thus, semantic conflicts modulate different cognitive functions and associated neurophysiological processes in different neuroanatomical structures, depending on whether a response has to be executed/inhibited. The data suggest that whenever action control processes and their neurophysiological correlates depend on regions specialized in the processing of semantic conflicts, there is an improvement in performance.
This work was partly supported by a grant from the Deutsche Forschungsgemeinschaft (DFG) SFB 940 project B8.
Reprint requests should be sent to Christian Beste, Cognitive Neurophysiology, Department of Child and Adolescent Psychiatry, Faculty of Medicine, TU Dresden, Schubertstrasse 42, D-01309, Dresden, Germany, or via e-mail: email@example.com.