Functional parcellation of the inferior frontal and midcingulate cortices in a flanker‐stop‐change paradigm

Abstract

Conflict monitoring and motor inhibition are engaged in the performance of complex tasks. The midcingulate cortex (MCC) has been suggested to detect conflicts, whereas the right inferior frontal cortex (IFC) seems to be of relevance for the inhibition process. The current experiment investigates the neural underpinnings of their interplay via a modified flanker paradigm. Conflict was manipulated by the congruency of flanking stimuli relative to a target (congruent vs. incongruent) and motor inhibition by a within‐trial response change of the initiated response (keep response vs. stop‐change). We used event‐related functional magnetic resonance imaging, decomposition with high model order ICA, and single trial analysis to derive a functional parcellation of the whole‐brain data. Results demonstrate the segmentation of the MCC into anterior and posterior subregions, and of the IFC into the pars opercularis, pars triangularis, and pars orbitalis. The pars opercularis and pars triangularis of the right IFC constituted the foundation of inhibition‐related networks. With high conflict on incongruent trials, activity in the posterior MCC network, as well as in one right IFC network was observed. Stop‐change trials modulated both the MCC as well as networks covering extended parts of the IFC. Whereas conflict processing and inhibition most often are studied separately, this study provides a synopsis of functionally coupled brain regions acting in concert to enable an optimal performance in situations involving interference and inhibition. Hum Brain Mapp, 2013. © 2012 Wiley Periodicals, Inc.

INTRODUCTION

A central question in research on executive functions is how human beings respond and adapt smoothly in an environment full of interfering information. Within such situations various processes are involved that enable persons to achieve their goals with specific actions. Theories of conflict monitoring assume that this function is part of a more general performance monitoring system that processes pre/postresponse conflict, errors, and unfavorable outcomes [Ridderinkhof et al., 2004a]. The midcingulate cortex (MCC), often referred to as the dorsal anterior cingulate cortex, is suggested to serve as an online detector of conflict in information processing [Botvinick et al., 1999, 2001; Carter et al., 2000]. For instance, simultaneously activated response representations lead to competition, which is in turn detected and reflected in activation of the MCC [Braver et al., 2001]. A considerable body of evidence illustrates the crucial role of the posterior MCC (pMCC) for conflict detection [e.g., Hester et al., 2004; Taylor et al., 2006]. After conflict detection, the MCC is thought to exert top‐down influence on other brain structures to adjust future performance [e.g., Botvinick et al., 2001, 2007; Danielmeier et al., 2011]. For the actual resolution of a conflict, an active suppression of incorrect response representations is suggested [Eriksen and Schultz, 1979; Ridderinkhof et al., 1999; Verbruggen et al., 2006].

The second function of interest, motor inhibition, is thought to be an active mechanism that serves to suppress physical motor responses [Aron et al., 2004; van den Wildenberg et al., 2009]. This mechanism is assumed to be implemented via a fronto‐subcortical circuit with the right inferior frontal cortex (rIFC) as the initiating structure for inhibitory mechanisms, involving an “indirect” or “hyperdirect” pathway through the basal‐ganglia and the primary motor cortex as downstream target [Aron, 2008, Aron et al., 2006; Band and Boxtel, 1999; Mink, 1996; Nambu et al., 2002]. Electrical stimulation of corresponding regions in the monkey's brain led to cancellation of a manual response in go trials of a go‐nogo task [Sasaki et al., 1989]. In humans, activations in the rIFC due to inhibitory processes have been observed with go‐nogo and stop‐signal tasks [e.g., Aron et al., 2006; Konishi et al., 1999; Rubia et al., 2003]. In addition, Aron et al. [ 2003] demonstrated disrupted inhibition with damage to the rIFC. This finding was further supported by Chambers et al. [ 2006] who showed impaired inhibition performance with “temporal disruptions” of the rIFC induced via transcranial magnetic stimulations.

Stroop, flanker, and go/nogo tasks are often utilized to assess disrupted inhibitory capabilities in clinical populations. However, most often the processing of conflicts is involved in such paradigms as well, thereby hampering definite conclusions. Recently, Verbruggen et al. [ 2004, 2006] demonstrated behavioral interactions of inhibitory and conflict‐related processes in a variety of different interference paradigms combined with stop trials (e.g., flanker, Simon, and Stroop tasks), including interference not only at the level of response representations. Their results showed that motor inhibition and the processing of conflicts possibly share a common underlying mechanism. However, these results may also suggest a global inhibition process that includes both the stopping of motor responses and the suppression of inappropriate representations due to conflicting stimuli.

One way to test if both functions are dissociable and how they interact in interference‐laden situations that require motor inhibition is to create experiments in which both are tested simultaneously. For example, differential brain activations related to motor inhibition and performance monitoring in the context of erroneous responses have already been documented [e.g., Chevrier et al., 2007; Matthews et al., 2005; Rubia et al., 2003]. In these stop‐signal studies, the rIFC and MCC have been associated with motor inhibition and error monitoring, respectively. However, most studies so far did not investigate brain areas associated with concurrent monitoring processes before a response, as seen with preresponse conflict monitoring in contrast to postresponse error monitoring. Beyond, real‐world examples of inhibition often additionally require a flexible and immediate change of motor responses.

Hence, the aim of this study was to investigate those brain regions recruited during conditions that involve preresponse conflict and motor inhibition. To do so, a flanker and a stop‐change paradigm [de Jong et al., 1995; Logan and Burkell, 1986] were combined. With stop‐change trials, a signal requires not only to inhibit an initiated response, but also to subsequently execute another one. By combining these paradigms, we examined conflict via the congruency of flanking stimuli (congruent vs. incongruent) and inhibition via the occurrence of within‐trial response changes (keep response vs. stop‐change). In this way, we controlled for differences concerning (visual) stimulation and response execution. These factors usually differ between stop and go trials in stop‐signal or go‐nogo paradigms. Additionally, we took advantage of the availability of an overt behavioral performance with all trial types. A second question was if specific subregions of the MCC and IFC would be involved in conflict monitoring and motor inhibition. Although this might seem trivial given the size and cytoarchitectonical differences within these regions, up to now it has not been clear which subregions are indeed associated with these cognitive processes. Because we specify our hypothesis regionally, rather than on a voxel‐by‐voxel basis and since our experiment probes multiple concurrent, possibly partially overlapping processes, a spatial independent component analysis (ICA) was used to decompose the functional magnetic resonance imaging (fMRI) data [Calhoun et al., 2001; McKeown et al., 1998]. The strength of this method is that it detects groups of brain regions with the same temporal pattern of hemodynamic signal changes, thus providing functionally integrated networks in a data‐driven way, i.e., without the need to specify a spatial/temporal model a priori. ICA has successfully been used to provide an anatomically and functionally plausible segmentation of resting state fMRI data into networks that map onto task‐related activations [e.g., Abou‐Elseoud et al., 2010; Kiviniemi et al., 2009; Smith et al., 2009], and has also been used to assess these networks in event‐related fMRI of complex executive tasks [Eichele et al., 2008; Danielmeier et al., 2011; Huster et al., 2011]. We used the same approach here, and performed a high model order ICA focusing on MCC and IFC components; we then deconvolved hemodynamic responses (HRs) for each participant and independent components (IC) to recover individually and regionally specific empirical estimates of the HR. Thereby we account for differences of hemodynamic responses (HRs) between regions and individuals [e.g., Aguirre et al., 1998; Handwerker et al., 2004; Miezin et al., 2000; Schactar et al., 1997]. Further, we estimated the means of the HRF amplitude for each experimental condition rather than working with correlation‐based contrast maps.

We predicted the following behavioral and neural patterns due to differentially involved cognitive functions. First, potentially proving the appropriateness of our experimental manipulations we expected to replicate well‐known behavioral effects: longer reaction times (RT) and increased error rates with incongruent trials, longer RTs in post‐error trials, longer SSRTs for incongruent stop‐change trials. Second, segmentations of the MCC and IFC into separate ICs according to their differential functional roles were anticipated. Third, if inhibition and conflict are dissociable functions they should differentially load onto ICs and anatomical regions. More specifically, with specific conflict contrasts we expected activations in the aMCC or pMCC, whereas inhibition contrasts should lead to activations in a rIFC network also comprising regions of the fronto‐subcortical circuit [as for example the subthalamic nucleus (STN), the pallidum, and the thalamus].

METHODS

Participants

Eighteen healthy participants (9 men; mean age: 25.2 years; range: 21–32 years, standard deviation: 3.1 years) took part in this study. All participants were right handed according to the Edinburgh Handedness Inventory [Oldfield, 1971]. None of them had a history of psychiatric or neurological disorders. All participants had normal or corrected‐to‐normal vision. Prior to fMRI measurement, participants were fully informed about the schedule and goals of the study and provided written informed consent in accordance with procedures approved by the Ethics committee of the Medical Faculty, University of Münster. The study was carried out in accordance with the ethical principles laid down in the current revision of the Declaration of Helsinki.

Stimulation Procedure

Participants performed a combined flanker‐stop‐change paradigm. On each trial, three colored arrow pairs were presented on a screen. Participants were asked to respond by button press as fast and accurate as possible. The direction of the centrally presented target (an arrow pair) indicated the hand to respond with. The right and left presented flanking arrow pairs were to be ignored. In 50% of all trials, these flanking arrow pairs pointed to the same direction as the target (congruent trials), in the remaining 50% they pointed to the opposite direction (incongruent trials). After a certain stimulus onset asynchrony (SOA), the initial color of the arrows (S1) changed to another one (S2). In 25% of the trials, the color of S2 instructed the participants to stop their initiated response and to change the response hand (stop‐change trials). In the other 75% of the trials, the color did not indicate the necessity to change (keep‐response trials, see Fig. 1 for an overview of experimental conditions). At the beginning of an experimental session the SOA was set to 250 ms and was subsequently adapted by an algorithm that tracked the individual responses to the stop‐change trials. The SOA was adjusted by adding 50 ms after every second correct or subtracting 50 ms following every failed stop‐change trial to reach an error rate of about 1/3 in stop‐change trials. Trials were presented in a pseudo‐randomized order ensuring that for each block of 48 trials all conditions were shown with the above mentioned ratios. After each block of 48 trials, participants received a feedback about their RT and their accuracy. Triggered by the MR scanner, trials were “jittered” with a random onset delay between 0 and 2 s to improve temporal sampling. After this variable onset delay, the S1 started and S2 followed with the dynamic SOA and was presented for 200 ms. Finally, a fixation cross appeared on the screen until the total trial duration reached 4 s. The block wise feedback was presented for a total of 12 s. The total number of trials was 528.

Figure 1.

Conditions, associated processes, and contrast‐calculation. A: Depicted are all four conditions for an initially required right hand response. With the keep‐response condition, the response execution has to be continued after the color change. In contrast, due to a different color change in the stop‐change condition, the initiated response has to be stopped and changed into a response execution with the other hand. B: For each condition, a short overview of the involved processes is given. Sensory‐motor processes are assumed to be equal, whereas executive functions are thought to differ. C: Because of the differently involved processes with the four conditions, conflict detection and motor inhibition functions can be isolated.

Stimuli had a viewing size of 4 × 1 degrees of visual angle and were projected via a projector‐mirror system. The experiment was controlled by the Presentation software package (Presentation 12.00 Neurobehavioral System, Albany, USA). The experiment lasted for about 38 min.

Image Acquisition

Data acquisition was performed on a 3 Tesla MRI scanner (Philips Medical System) equipped with the standard bird cage head coil at the Institute for Clinical Radiology, University of Münster, Germany. Twenty‐eight slices (3.6 mm thickness, without gap) were obtained in an ascending order parallel to the anterior/posterior commissure using a single‐shot echo‐planar imaging (EPI) sequence (repetition time: 2 s; echo time: 38 ms; flip‐angle 90°, field of view: 230 mm, in‐plane resolution: 3.6 × 3.6 mm²). A total of 1134 T2*‐weighted MR images, were acquired. Additionally, whole‐head EPIs prior and after the functional scanning were recorded to aid the post processing of the functional time‐series data.

Behavioral Analyses

Mean RTs for correct and incorrect trials (congruent keep‐response, incongruent keep‐response, congruent stop‐change, incongruent stop‐change) were calculated. Repeated‐measures analyses of variance (ANOVA) were calculated on the RTs with the factors response (keep‐response, stop‐change) and congruency (congruent, incongruent) for both correct and incorrect trials. Error rates were calculated and tested for statistically significant effects via another repeated‐measures ANOVA with the above mentioned factors. Pearson product–moment correlations were computed to assess the relationship between RTs and response rates for all conditions. For congruent and incongruent stop‐change trials the stop‐signal reaction time (SSRT) was computed. The SSRT represents the time that is needed to inhibit an initiated response in successful stop trials. The SSRT is estimated based on the independent race model by Logan and Cowan [ 1984] as the average of the observed SSRTs per SOA and participant. Differences between the SSRTs were assessed with a t test. In addition to the SSRT, another measure was calculated for each condition to assess inhibitory control: the Z‐transformed relative finishing time (ZRFT = (RT − SOA − SSRT_av)/SD_RT; SSRT_av = average of all observed SSRTs that correspond to 0.15 < p(response) < 0.85, SD_RT = standard deviation of the reaction time distribution per participant). The resulting values represent unit‐less numbers distributed around zero. The advantage of this measure is that it accounts for differences between conditions in the mean RT as well as for its variance [for more details see Band et al., 2003]. Finally, posterror slowing was examined by comparing the RTs of posterror and postcorrect trials for all conditions. Therefore, a repeated‐measures ANOVA with the factors adjustment (posterror, postcorrect), congruency (congruent, incongruent) and response (keep‐response, stop‐change) was performed. For statistical analyses, the PAWS software package (SPSS, Chicago, USA) was used.

Preprocessing of fMRI

The image time‐series of single subjects were preprocessed using SPM 5 (Wellcome Institute, London, UK). First, to correct for head movements all images were realigned to the first volume. Second, a slice time correction was performed. Then, the images were coregistered with an individual whole‐head EPI and normalized to the MNI reference space. The resulting voxel size was set to 3 × 3 × 3 mm³. The normalized data were then smoothed with a Gaussian kernel with 8 mm FWHM.

Independent Component Analysis

The group ICA procedure as proposed by Calhoun et al. [ 2001, 2006] was performed using the academic freeware GIFT (version v1.3h, http://icatb.sourceforge.net) running in Matlab (MathWorks, Natick, MA). GIFT generates aggregate data via temporal concatenation containing the individual data from all participants (see Supporting Information Fig. 1 for a schematic overview of the ICA, deconvolution and inference procedures). To do this, each subject's data set was prewhitened and a dimension reduction via temporal principal component analysis (PCA) to 100 dimensions was applied. To form the group data, the individual principal components were temporally concatenated and then reduced in a second PCA step. Then, a spatial ICA was performed using the Infomax algorithm [see decomposition step in Supporting Information Fig. 1; Bell and Sejnowski, 1995]. The resulting outputs are group component maps and group time course maps. Robust components were identified by running ICASSO [determination of the ICA algorithmic reliability or stability; Himberg et al., 2004] 100 times with random initial values and identifying centroids by means of canonical correlations based on a clustering algorithm. As last step of the group ICA, individual independent component (IC) maps and time courses were back‐reconstructed by multiplying the corresponding data with the respective portions of the estimated demixing matrix. The group average maps were inspected by three independent raters to identify MCC and IFC related ICs with significant t‐statistics at an uncorrected threshold of t > 5 (corresponding to a false‐positive discovery rate of 1% across components). In 83.3% of the inspected components there was a perfect agreement of the raters concerning relevant (MCC and IFC) components. Only these components were included for further data analyses.

Deconvolution and Inference

For the recovery of the hemodynamic response (HR) the deconvolution method suggested by Eichele et al. [ 2008] was used. The estimation of an empirical HR has the advantage that it can deal with intersubject and interregional variability [Handwerker et al., 2004]. The empirical event‐related HRs were deconvolved for each participant and IC separately by forming the convolution matrix of all trial onsets with a duration of 20 s and multiplying the pseudo‐inverse of this matrix with the high pass filtered (128 s) and normalized IC time course. Single‐trial amplitudes were recovered by fitting a design matrix containing separate predictors for each trial onset (congruent keep‐response, incongruent keep‐response, congruent stop‐change, incongruent stop‐change) convolved with the estimated HR onto the IC time course, thereby estimating the scaling coefficients (β) using multiple linear regression. For each participant, condition, and component the single‐trial amplitudes were then averaged and entered into (one sided) paired t tests under the assumption of zero magnitude, testing for two conflict detection related contrasts (conflict 1 = incongruent keep‐response—congruent keep‐response and conflict 2 = incongruent stop‐change—congruent stop‐change) and for two motor inhibition related contrasts (inhibition 1 = congruent stop‐change—congruent keep‐response and inhibition 2 = incongruent stop‐change—incongruent keep‐response). For this statistical analysis the software package PASW Statistics 18 (SPSS, Chicago, USA) was used. Effects were considered significant at P < 0.05. Figure 1 illustrates the concept behind the calculated contrasts.

RESULTS

Behavioral Performance

The RTs (mean, ± standard deviation) were 566 ± 77 ms for congruent keep‐response, 604 ± 71 ms for incongruent keep‐response, 521 ± 64 ms for congruent stop‐change, and 548 ± 69 ms for incongruent stop‐change trials. The typical incongruency effect was observed: incongruent RTs were significantly longer than congruent ones (F(1, 17) = 35.868, P < 0.001). In addition, RTs differed significantly between keep‐response trials and stop‐change trials (main effect: response; F(1, 17) = 10.097, P < 0.01). Participants exhibited the following error rates: 4.1% with congruent keep‐response, 7% with incongruent keep‐response, 41.57% with congruent stop‐change, and 34.68% with incongruent stop‐change trials. No correlations were found between the conditions' RTs and error rates supporting the notion that subjects did not exert speed–accuracy trade‐offs. There were significantly more errors with stop‐change trials than with keep‐response trials (main effect: response; F(1, 17) = 287.64, P < 0.001). Incongruent keep‐response trials elicited more errors than congruent keep‐response trials whereas in trials with stop‐change this relation was reversed (interaction: response × congruency; F(1, 17) = 13.746, P < 0.01). RTs of erroneous responses for keep‐response trials (645 ± 98 ms for congruent, 622 ± 140 ms for incongruent keep‐response trials) differed significantly from those with stopping and changing a response (498 ± 66 ms for congruent stop‐change, and 564 ± 69 ms for incongruent stop‐change trials; main effect: response; F(1, 17) = 40.185, P < 0.001). Error RTs for congruent trials were slower in keep‐response trials, whereas in stop‐change trials they were faster (interaction: congruency × response, F(1, 17) = 15.156, P < 0.001) than incongruent trials. In congruent stop‐change trials the SSRT (mean ± standard deviation) was 320 ± 47 ms, whereas in incongruent stop‐change trials the SSRT was 347 ± 43 ms and was significantly prolonged (t(17) = 3.508, P = 0.01). These differences in inhibition time were also confirmed with another measure accounting for possible differences between conditions with respect to the mean RT, its variance, and the variance of the SSRT. The ZRFTs (mean ± standard deviation) of the congruent stop‐change and the incongruent stop‐change conditions were 0.24 (± 0.21) and −0.14 (± 0.07), respectively. This effect proved to be significant (t(17) = 7.683, P = 0.001).

Last but not least, a significant post‐error slowing was observed (main effect: adjustment; F(1, 17) = 10.667, P < 0.005). An overview of the behavioral performance measures is given in Tables I and II.

Table I.

Mean RTs (in ms) and standard deviations (SD) for correct and incorrect trials, as well as the mean SSRTs and SDs for stop‐change trials

Task	RT mean (SD)	RT of errors	Errors (SD)	SSRT mean (SD)	SOA mean (SD)	SOA of errors (SD)
Congruent keep‐response	566 (77)	645 (98)	4.5 (2.92)		201 (13)	202 (13)
Incongruent keep‐response	604 (71)	622 (140)	7 (3.65)		203 (13)	210 (19)
Congruent stop‐change	521 (64)	498 (66)	41.57(11.14)	376 (47)	193 (16)	214 (15)
Incongruent stop‐change	548 (69)	564 (69)	34.68 (9.5)	417 (48)	212 (11)	213 (11)

Given are also the means and SDs of error percentages and the SOAs between S1 and S2

Table II.

Mean RTs (in ms) and standard deviations (SD) for all trials separated into postcorrect and posterior trials

Task	RT mean (SD)
	Postcorrect	Postincorrect
Congruent keep‐response	562 (79)	581 (70)
Incongruent keep‐response	600 (72)	620 (78)
Congruent stop‐change	512 (63)	541 (81)
Incongruent stop‐change	541 (74)	555 (66)

Independent Component Structure and Assessment of Differences Concerning Experimental Conditions

Five different ICs were identified: two represented MCC networks (see Fig. 2 for a depiction of the anatomical locations, the corresponding HRs, and the bar charts of the significant differences of the conditions for each IC; and see also Table III for an overview of brain regions within each network) and three rIFC networks (see Fig. 3 and Table IV). Note: as a consequence of the modeling assumptions of ICA the fMRI signal observed at a given location is determined by the interaction of the IC weights (displayed as maps) and the values of the IC time courses.

Figure 2.

On the left side those two network maps are shown including regions of the MCC. In the middle row, group averages of the estimated hemodynamic responses are depicted. Error bars indicate standard errors of the mean (SEM). On the right, bar charts of the significant differences can be found.

Table III.

MCC networks and relevant brain regions associated with the experimental conditions

Network	h.a.	Cluster			Local maxima
		Cluster name	Hemisphere	Volume (mm3)	Subpeak	BA	t values	x	y	z
IC 1		pMCC	bi	34,000		24	26.4	−3	19	29
		Superior temporal gyrus	r	2,200		22	7.2	59	3	3
		Superior temporal gyrus	l	1,500		22	6.9	−59	−3	3
	−	Angular gyrus	r	2,100		40	8.0	56	−48	36
	−	Inferior parietal lobule	l	6,400		40	7.0	−48	−48	36
	−	Superior occipital gyrus	r	2,000		19	7.4	24	−78	23
	−	Middle occipital gyrus	l	2,100		18	7.1	−30	−87	4
	−	Precuneus	bi	900		7	6.6	6	−47	41
	−	Postcentral gyrus	l	1,900		3	6.3	−56	−12	48
	−	IFC, p. orbi.	r	2,000		47	6.3	42	26	−6
	−	Insula	l	3,100		13	6.2	−36	27	7
	−	Superior temporal gyrus	r	3,000		22	6.0	62	−12	−2
	−	Middle temporal gyrus	l	1,300		21	5.1	−62	−40	8
	−	Superior frontal gyrus	bi	6,200		6	5.8	−27	−9	61
					Middle frontal gyrus	10	5.5	24	52	−10
IC 2		aMCC	bi	42,400		32	23.8	3	33	12
					Middle frontal gyrus	10	8.9	27	39	26
		Insula	l	10,500		13	6.0	−30	20	−9
		IFC, p.orb.	r	2,400		47	7.5	36	23	−14
	−	Middle frontal gyrus	l	13,900		10	10.8	−30	47	0
	−	Middle frontal gyrus	r	9,900		9	9.8	30	18	46
	−		r		Superior frontal gyrus	8	6.3	9	46	39
	−	Inferior temporal gyrus	l	3,300		37	9.1	−59	−53	−5
	−	Putamen	r	800		n.a	9.0	21	8	−8
	−	Middle temporal gyrus	r	3,600		37	6.2	54	−60	3
	−	Superior temporal gyrus	r	2,000		22	7.6	59	−11	6
	−	Lingual gyrus	l	1,800		18	7.0	−18	−58	0
	−	Precuneus	r	1,700		7	6.9	9	−50	58
	−	Postcentral gyrus	r	1,700		3	6.2	27	−35	54
	−	Postcentral gyrus	l	600		3	5.2	−45	−18	48

Columns depict ICs, hemodynamic activity (h.a.), anatomical labels of a cluster, hemisphere (l = left, r = right; bi = bilateral cluster), volume of a cluster, anatomical label of subpeaks with a minimal distance of 10 mm, Brodmann Areas (n.a. = not applicable), t scores for the peak voxel from a random effects analysis, and MNI coordinates for the local maxima and minima. If not noted otherwise, regions show significant increases in hemodynamic activity. Regions with a decreased hemodynamic activity are marked with −. Abbreviations: BA = Brodmann Area, SMA = supplementary motor area, pACC = perigenual Anterior Cingulate Cortex, PCC = posterior cingulate cortex, sACC = subgenual Anterior Cingulate Cortex, p. orb. = pars orbitalis, p. triang. = pars triangularis, p. operc. = pars opercularis.

Figure 3.

Depicted on the left are those network maps representing regions of the rIFC. The estimated hemodynamic responses and SEM are shown in the next column. The bar charts display significant differences between conditions. Error bars indicate standard errors of the mean (SEM).

Table IV.

rIFC networks and relevant brain regions associated with the experimental conditions

Network	h. a.	Cluster			Local maxima
		Cluster name	Hemisphere	Volume (mm3)	Subpeaks	BA	t values	x	y	z
IC 3		IFC, p. triang.	r	41,100		45	17.6	53	27	15
		IFC, p. triang.	l	12,000		45	9.9	−48	27	12
	−	Middle occipital gyrus	r	6,000		19	10.5	33	−78	7
	−	Angular gyrus	r	5,100		39	8.5	56	−63	31
	−	Angular gyrus	l	9,100		39	8.6	−56	−60	33
	−	Insula	l	12,500		13	8.2	−42	12	−1
	−	Middle frontal gyrus	r	3,500		9	6.2	39	25	40
	−	Precentral gyrus	l	5,800		4	7.9	−39	11	44
	−	pre‐SMA	r	3,100		6	5.6	12	5	47
	−	Thalamus	r	2,900		n.a.	7.4	12	−20	1
	−				STN	n.a.	6.7	12	−6	−5
	−	Pallidum	l	2,600		n.a.	6.3	−21	−6	−2
	−	Precuneus	l	3,100		7	6.3	−9	−62	39
	−	MCC	l	1,700		24	6.1	−9	22	27
	−	Superior temporal gyrus	r	1,300		22	5.8	50	−26	12
	−	Temporal pole	r	1,100		21	5.5	59	8	−18
IC 4		IFC, p. operc.	r	35,700		44	14.3	53	7	26
					Rolandic operculum	n.a.	6.7	45	−5	14
					Putamen	n.a.	5.3	33	3	5
		IFC, p. operc.	l	6,200		44	8.8	−39	4	22
	−	Insula	r	5,500		13	9.9	42	11	−13
	−	Hippocampus	bi	2,300		n.a.	9.1	24	−9	−12
	−	Superior frontal gyrus	l	5,500		11	9.0	−18	55	−13
	−	Thalamus	l	4,800		n.a	8.4	−21	−17	6
	−	Caudate	r	900		n.a	7.5	6	12	8
	−	IFC, p. operc.	l	1,600		44	7.0	−59	16	27
	−	Rolandic operculum	r	800		n.a.	6.9	36	−20	18
	−	Superior occipital gyrus	l	300		18	6.3	−12	−83	21
IC 5		IFC, p. orb.	r	29,200		47	14.9	48	43	−5
					Insula	13	8.0	45	−9	9
					Putamen	n.a.	9.0	24	20	2
		IFC, p. operc.	l	4,900		44	5.6	−53	15	5
					Insula	13	7.8	−33	18	2
		Middle frontal gyrus	l	4,300		46	7.6	−21	48	25
		Inferior parietal lobule	l	7,300		40	5.6	−50	−42	38
					Superior temporal gyrus	42	5.5	−53	−40	16
		Superior frontal gyrus	r	2,500		10	5.4	6	59	14
		Supramarginal gyrus	r	700		40	5.6	56	−39	41
	−	Middle temporal gyrus	r	35,200		21	12.7	53	−21	−9
	−	PCC	bi	7,100		23	11.4	9	−45	35
	−	Middle frontal gyrus	l	8,200		10	10.8	−36	50	6
	−	IFC, p. orb.	r	5,600		47	8.2	48	23	−11
	−	Precentral gyrus	r	4,000		6	6.5	50	10	50
	−	SMA	r	2,600		8	6.2	9	23	54

Columns depict ICs, hemodynamic activity (h.a.), anatomical labels of a cluster, hemisphere (l = left, r = right; bi = bilateral cluster), volume of the a, anatomical label of subpeaks with a minimal distance of 10 mm, Brodmann Areas (n.a. = not applicable), t scores for the peak voxel from a random effects analysis, and MNI coordinates for the local maxima and minima. If not noted otherwise, regions show significant increases in hemodynamic activity. Regions with relative hemodynamic activity decreases are marked with −. Abbreviations: BA = Brodmann Area, SMA = supplementary motor area, pACC = perigenual Anterior Cingulate Cortex, PCC = posterior cingulate cortex, sACC = subgenual Anterior Cingulate Cortex, p. orb. = pars orbitalis, p. triang. = pars triangularis, p. operc. = pars opercularis.

IC1

This component predominantly showed activations in the pMCC that extended to the presupplementary motor area (pre‐SMA) and to the posterior cingulate cortex. Somewhat weaker activations were also observed in the superior temporal gyrus. Within this network the right angular gyrus, the bilateral inferior parietal lobes, the bilateral occipital‐, the bilateral superior temporal‐, as well as the right middle frontal‐, and the left superior frontal gyrus were deactivated. Three of the computed contrasts showed significant differences between conditions: the activation of the associated regions were modulated by conflict 1 (t(1,17) = 1.864, P < 0.05), as well as by both inhibition‐related contrasts (inhibition 1: t(1,17) = −5.273, P < 0.001; inhibition 2: t(1,17) = 2.203, P < 0.05).

IC2

A second network was primarily located in the anterior MCC (aMCC) which included further activations in the pregenual anterior cingulate cortex (pACC), the right orbito frontal‐, and the right middle frontal gyrus. Deactivated regions were the bilateral IFC, the bilateral middle frontal‐, the bilateral superior frontal‐, the bilateral middle temporal‐, as well as the left inferior temporal gyri. Further deactivations were found in the bilateral putamen, and in the bilateral postcentral gyrus. Activity was significantly modulated by both inhibition contrasts (inhibition 1: t(1,17) = 4.585, P < 0.001; inhibition 2: t(1,17) = 2.033, P < 0.05).

IC3

This network indicated strongest activations in the pars triangularis of the rIFC and bilateral insulae. Deactivations were observed in broader brain regions including the right STN, the left pallidum, the right thalamus, the precentral gyrus, and the right SMA. Please refer to Table IV for the whole list of deactivations. This network exhibited significant associations with differences of the conflict 1 (t(1,17) = 2.506, P < 0.05) and the inhibition 1 contrasts (t(1,17) = 2.487, P < 0.05).

IC4

The fourth network predominantly activated the pars opercularis of the rIFC and to a lesser degree also the lIFC pars opercularis. These activations extended toward the bilateral precentral gyri, and to the right rolandic operculum. Additionally, the right putamen was activated. Deactivations were observed in parts of the right lateralized basalganglia, as well as in the thalamus. For further deactivations see Table IV. Significant modulations were found for both inhibition related contrasts (inhibition 1: t(1,17) = 3.839, P < 0.001; inhibition 2: t(1,17) = 2.079, P < 0.05).

IC5

A fifth network mainly covered the rIFC's pars orbitalis. Further activations were seen in the right putamen, in the bilateral insulae, in frontal regions (right orbito‐, left middle‐, right superior frontal gyrus), as well as in parts of the temporal lobe (in left superior temporal‐, right middle temporal gyrus), and in the right supramarginal gyrus. Deactivations were observed in the right middle temporal gyrus as well as in the bilateral posterior cingulate cortex. Additional weaker deactivations are listed in Table IV. IC 5 was significantly modulated by both inhibition related contrasts (inhibition 1: t(1,17) = 3.117, P < 0.01; inhibition 2: t(1,17) = 3.38, P < 0.01).

DISCUSSION

Results show a segmentation of the MCC and the IFC corresponding to ICs 1‐5. These ICs reflect a differentiated engagement of anatomically plausible subregions of the MCC, known as aMCC and pMCC [Vogt et al., 2003], and the IFC, namely the pars opercularis, pars triangularis, and the pars orbitalis. Hence, this study reveals the anatomy of functionally coupled networks via ICs each indicating brain regions sharing the same hemodynamic response pattern over time. In each IC the MCC's and IFC's subregions are the regions exhibiting strongest activations, emphasizing their prominent roles. Furthermore, the structure of the functionally coupled networks comprised regions known to be anatomically connected as well, as was seen with ICs evolving from the pars opercularis' and pars triangularis' coactivations. These consisted of the striatum, the pallidum, the thalamus and motor regions altogether corresponding to a well‐known fronto‐subcortical circuit [Mink, 1996; Nambu et al., 2002]. However, exclusive links of specific networks to single contrasts were only shown for the pars opercularis‐ and pars orbitalis networks which both were related to inhibition processes. Indeed, many experimental manipulations are not capable of clearly isolating single (cognitive) functions. With the inhibition contrasts, for example, several unaccounted mechanisms may subserve the stop‐change process: beyond motor inhibition involved in stopping the motor response, motor planning for the initiation of the subsequently required response might play a role as well. This observation has two possible reasons: first, exclusive links of one network/region to one experimentally isolated cognitive function may physiologically not be plausible; second, other latent processes may be involved.

The anatomical subregions of the MCC revealed by ICs 1 and 2 are known to differ in cytology, and in the architecture of connections [Vogt et al., 2003]. With respect to cognitive control, such a dichotomy of rostral and dorsal activations has repeatedly been discussed under aspects of functional and anatomical dissociations, with conflicts typically activating caudal and dorsal regions whereas error‐ or feedback‐related activations are detected rostrally [e.g., Braver et al., 2001; Garavan et al., 2003; Hester et al., 2004; Taylor et al., 2006; Ullsperger and Cramon, 2001]. The finding of two subregions in the present study also argues for such a functional dichotomy. While the IC 1 network (pMCC) is a candidate for a region reflecting conflict detection, IC 2 (aMCC) does not necessarily reflect the detection of conflicts per se. For instance, two monkey electrophysiology studies showed clearly that aMCC cells of the monkey (the rostral cingulate motor area) do not exclusively respond to conflict detection [Ito et al., 2003; Nakamura et al., 2005] but demonstrate functional heterogeneity. A link between aMCC dysfunction and inattention was demonstrated by Bush [ 2009], who proposed especially aMCC hypoactivations to play a crucial role in the production of inattention seen with ADHD patients. Besides the activation of the aMCC in IC 2, a part of the dorsolateral prefrontal cortex (DLPC) is activated as well in this network. This region is widely known to allocate attention to relevant stimuli [Botvinick et al., 2001; Kerns et al., 2004; Miller and Cohen 2001; Ridderinkhof et al., 2004a]. Similarly, Orr and Weissman [ 2009] dissociated subregions of the MCC subserving the detection of conflict, and control of attention in association with the DLPC.

A similar anatomical parcellation as with the MCC was found for the rIFC. Three ICs (ICs 3‐5) included the pars opercularis, the pars triangularis, and the pars orbitalis, respectively. As was already shown by Brodmann (Brodmann Areas 44, 45, 47), these subregions differ cytoarchitectonically. Especially the pars opercularis and pars triangularis are in the focus of studies on inhibition [e.g., Aron et al., 2003; Chambers et al., 2007]. Indeed, areas from two functionally extracted networks (ICs 3, 4) exhibit activity within a fronto‐subcortical circuit (e.g., the rIFC, the striatum, the pallidum, the thalamus and motor regions) comprising regions known to constitute an inhibition network [Mink, 1996; Nambu et al., 2002]. The main differences between these two networks mainly persist in the coactivation of the primary motor cortex, probably the hand motor area, found with IC 4 but not with IC 3. In other words, even though IC 3 and IC 4 both likely reflect aspects of inhibition, IC 4 seems to be more tightly coupled to motor planning.

In contrast to ICs 3 and 4, IFC areas of IC 5 show coactivations with other regions, as for example the superior temporal gyrus and the inferior parietal lobule, indicating a different functional role. Such patterns are in agreement with other inhibition tasks. For example, Rubia et al. [ 2001] proposed that motor attention/control processes underlie activations of the inferior parietal lobule [see also Rushworth et al., 1997, 2003].

Another noteworthy finding concerns the occurrence of coactivations of the pre‐SMA with the pMCC and the pars triangularis of the rIFC. Both regions exhibited significant contributions to both the processing of conflicts and when exerting inhibition. Recent results suggest that the pre‐SMA is not associated with the processing of conflicts in itself, which was widely believed due to its association with interferences‐laden contrasts, but is more directly tied to motor inhibition [Huster et al., 2011; Sharp et al., 2010]. On the other hand, our data might suggest that the pre‐SMA serves as an interface between neural systems monitoring for conflicts and those exerting the actual inhibitory mechanisms. This notion, however, has to be addressed more directly in the future.

With regard to the conflict‐related experimental manipulations, our data showed the well‐known prolonged RTs [e.g., Eriksen and Schultz, 1979; Ridderinkhof et al., 1995, 1997], specific activations of the pMCC reflecting conflict detection, as well as activations of the pars triangularis potentially associated with conflict resolution. The specific activation of the pMCC (dorsal part) fits with the above depicted functional dichotomy. In general, in accordance with the conflict monitoring theory [e.g., Botvinick et al., 2001, 2007] an activation of the MCC was expected. Such associations have regularly been observed in flanker [Botvinick et al., 1999; Bunge et al., 2002; Casey et al., 2000, Durston et al., 2003] as well as in other interference tasks [e.g., Barch et al., 2001; Kerns, 2006]. Besides the expected engagement of the MCC as a region detecting conflicts, activation of the rIFC‵s pars triangularis was observed as well. An association of the rIFC with conflict processing is already debated under aspects of conflict resolution via inhibition [Aron et al., 2004; Eriksen and Schultz, 1979; Ridderinkhof et al., 2004a, 1999; Verbruggen et al., 2006]. For example, Desimone and Duncan [ 1995] stressed the notion of a joint attention and inhibition mechanism as a control process following the detection of conflicts. In previous studies, activations of the rIFC with conflict were thus linked to intervening inhibitory processes [Bunge et al., 2002b, Hazeltine et al., 2000, 2003]. In line with this conception, Bunge et al. [ 2002a] reported the rIFC, the right insula, and anterior parts of the right middle frontal gyrus to be associated with conflict‐related RT slowing within conflict trials believed to reflect enhanced conflict resolution. Furthermore, delta plots (a tool for the analysis of RT distributions) are proposed to reflect response inhibition in conflict tasks [Ridderinkhof et al., 2004b]. With two studies of Forstmann et al. [ 2008a, b], it was shown that the activation of the rIFC covaried with parameters of this RT analysis technique. The idea of conflict resolution via inhibition is now also strengthened by the fact that the activations of the pars triangularis were embedded within a coherent network resembling the known inhibition network. Surprisingly, the present results do not reveal MCC or rIFG activations for the second conflict contrast as were seen with conflict 1 (that is the “typical” conflict effect in flanker studies, see Fig. 2). In the first place, slight behavioral differences between these contrasts are apparent demonstrating that processes may not be fully equivalent. Furthermore, MCC activations seen with both stop‐change conditions, that are most likely related to the change (see further discussion), might blur the conflict 2 effect.

Turning to the experimental manipulations of inhibition, relevant processes recruited two rIFC networks (ICs 3 and 4) already being in the focus of inhibition‐related research, including one network (IC 4) that is suspected to implement a subsequent change of response execution. These activations are consistent with models of stop‐change paradigms and add strong support to the view, that the sole activation of an alternative response is insufficient to stop and change an initiated response [Boucher et al., 2007; Camalier, et al., 2007; Verbruggen et al., 2008, 2009]. Specifically, de Jong et al. [ 1995] suggested a selective inhibition mechanism for stop‐change processes. The constellations of deactivations and activations of IC 4—including the striatum—might hint to an “indirect” pathway that is proposed for selective inhibition [Aron and Verbruggen, 2008]. Compared to ICs 3 and 4, Stevens et al. [ 2007] presented a similar inhibition network with a go‐nogo task, including the IFC, basal ganglia, thalamus, and premotor areas.

Noninhibitory processes in association with the rIFC have been shown in many studies. Stevens et al. [ 2007], for example, found a rIFC network (similar to IC 5) including inferior parietal regions in a go‐nogo task not related to motor inhibition per se. Recent studies on inhibition specifically propose the involvement of the ventral attentional system including the rIFC to be also engaged in stop‐signal trials [Hampshire et al., 2010; Sharp et al., 2010]. Furthermore, the aMCC has been shown to be preferentially associated with errors and feedback, thus in the context of inhibition tasks this region likely reflects the allocation of attentional resources or the evaluation of task relevant behavior. The coactivation of IC 5 (pars orbitalis) and IC 2 (aMCC) points to additional requirements of attention (e.g., motor attention, and attention towards stimulus features) in inhibitory processes.

However, the activation of the pMCC with the inhibition contrasts was somewhat unexpected. Its engagement could imply that both, the IFC as well as the MCC, are involved in the implementation of inhibition. On the other hand, this finding could also indicate the involvement of conflict processes due to the change of responses. Tasks incorporating a response change [e.g., Brown, 2009; Brown and Braver, 2005] may evoke conflict processes during a stop‐change event. That is, the rapid change of the target stimuli in stop‐change conditions is thought to lead to a “temporally dynamic” conflict between the initially activated response representation and the second exclusive response representation. Therefore, it can be argued that when inhibition processes are embedded in contexts that require immediate, adaptive response changes, dynamic conflicts can arise leading to midcingulate engagement.

In conclusion, many real‐world situations force human beings to differentiate between relevant and irrelevant information to guide performance, and often require selective stops and changes of motor responses to meet challenges of the environment. The current results illustrate the functional anatomy of the MCC and rIFC and highlight their involvement in larger functional networks. Beyond, it was shown how closely both anatomical structures are linked. Here, the pre‐SMA might serve as a bridging element for the integration of conflict and inhibition related processes. Therefore, for a successful performance with complex task constellations, conflict monitoring, conflict resolution, and motor inhibition on their own as well as their interplay are essential. This study provides clear evidence of this important yet complex interaction. In future studies, relevant questions have to be addressed concerning the precise functional differences and the interplay of the identified IFC‐ and MCC‐related networks, the timing of the respective neurocognitive processes, as well as their connectivity.

Supporting information

Additional Supporting Information may be found in the online version of this article.

Suppporting Information Figure 1.