Thirty healthy, German-speaking volunteers (mean age: 21.74 years; range 19–24 years; 15 males) with no prior psychological or neurological history were included in the current study. Participants took part in one testing session that included psychometric testing, one functional neuroimaging task and a T1-weighted structural image acquisition. Two participants were excluded from analysis since one of them had completely missing and the other person very low in-scanner performance (e.g., more than 20% misses in each run). All participants were further right-handed, had normal or corrected-to-normal vision, and provided written informed consent as approved by the local ethics committee (Ethikkommission der Nordwest- und Zentralschweiz).

Participants included in this study completed a battery of standardized tests comprising verbal and non-verbal IQ [German version of the Vocabulary and Matrix reasoning subtests of the WAIS-IV (Petermann, 2012), present mood (EWL; Janke, 1978)], behavioral and emotional functioning (YSR; Achenbach, 1991), psychopathic traits (YPI; Andershed et al., 2007), and handedness (EDI; Caplan and Mendoza, 2011). The YPI, YSR, and EWL were missing for one person. The resulting behavioral group characteristics are provided in Table 1.

The neuroimaging session included event-related functional neuroimaging during the performance of an emotional number Stroop task. Additionally, T1-weighted structural images were acquired for each participant. The emotional number Stroop task was adapted and modified based on a design described by Hart et al. (2010) (see trial design in Figure 1). We decided to use a number Stroop task as it has particularly been developed for use in the MR environment and has previously successfully been implemented in neuroimaging research studies (e.g., Blair et al., 2007; Hart et al., 2010). Each trial started with an emotional prime of either neutral or negative valence, presented for 150 ms. Negative (Neg) or neutral (Neu) primes were first followed by an item of the number Stroop task presented for 1,500 ms, before a short relaxation period of 350 ms ended the trial. During the number Stroop task, participants were presented with an array of 1, 2, 3, or 4 digits and were asked to indicate through button press the number of items presented. The number of items was either congruent (C) in relation to the printed digits (e.g., the digit 4 in an array of 4) or incongruent (IC) with the printed digits (e.g., digit 4 in an array of 3). Star shaped stimuli (S) were used as a control condition (no interference of digit and item number) and null trials (trials with a black screen instead of the Stroop trial) were added during the randomization process. Emotional stimuli were adapted from the Developmental Affective Photo System (DAPS; a child-appropriate picture system Cordon et al., 2013), which uses part of the IAPS (International Affective Picture System Lang et al., 2008) commonly used in adults. We implemented DAPS images because this task was designed to ultimately be employed in children and adolescents with psychiatric disorders. However, given that all images remained part of the IAPS system, the chosen stimuli were considered suitable for both adult and adolescent populations. A list of the images used is provided in Supplemental Information 1. In combination with the negative or neutral primes, the following combinations of prime and trial condition were possible: negative-congruent (Neg_C), negative-stars (Neg_S), negative-incongruent (Neg_IC), and neutral-congruent (Neu_C), neutral-stars (Neu_S), neutral-incongruent (Neu_IC). Prior to study start, our task was behaviorally tested in adults. Pilot data assessment indicated a significant emotion by cognition interaction for reaction time and accuracy measurements (see Supplemental Information 1.2).

Prior to the start of the experiments, an optimal stochastic trial order allowing for a rapid event-related design was determined using optseq2, a tool for automatically scheduling events for rapid-presentation event-related fMRI experiments (for further information see http://surfer.nmr.mgh.harvard.edu/optseq/ or experiments using similar designs Ferri et al., 2012; Kuhlmann et al., 2016). We administered a total of 300 Stroop trials (100 for each C/S/IC) and 50 null trials. All 300 Stroop trials were preceded by either neutral or negative primes (50:50). Total scan time was about 11.5 min. The complete experiment was performed over the course of 2 runs. At the end of the neuroimaging session, participants were further asked to perform valence ratings of the images presented in the scanner using a Likert scale from −2 to 2 (with −2 representing very negative valence, 0 being neutral, and 2 indicating high attractiveness of the stimulus). For each participant the mean scores of the negative images affect rating were used as a covariate of no interest within the group analysis to account for differences in valence judgments between the young adults (see also Supplemental Information 2.1 Emotional Valence Rating).

Whole brain blood oxygen level-dependent (BOLD) fMRI data and T1-weighted mprage images were acquired on a Siemens 3T MR imaging system (Siemens Prisma, Erlangen, Germany) and a 20-channel phased-array radio frequency head coil. For the fMRI task a rapid event-related stochastic design with TR = 2,000 ms, TE = 30.0 ms, FOV = 192 mm; image matrix = 64 × 64 mm; voxel size = 3 mm and number of slices = 37 was used. We further acquired a high resolution T1-weighted structural image using the following specifications: TR = 1,900.0 ms; TE = 3.42 ms; FOV = 256; image matrix = 256 × 256; voxel size = 1 mm. T1-weighted mprage structural neuroimaging data was used for co-registration and to calculate the total intracranial volume (TIV). Men and women are known to differ in overall brain size (Leonard et al., 2008; Luders et al., 2009; Giedd et al., 2012). This was also true for the present sample where TIV significantly differed between males and females [males M = 1,529.52 ± 87.22/females M = 1,360.72 ± 91.99; t₍₂₈₎ = 5.16, p < 0.001]. Likewise, socioeconomic status, sex, and age all correlate with TIV (Luders et al., 2009, 2014; Taki et al., 2011; Jednorog et al., 2012). This was accounted for by using a covariate of no interest in consequent random effects analyses. Therefore, TIV was extracted through the voxel-based morphometry toolbox (VBM8; http://dbm.neuro.uni-jena.de/vbm) as implemented in SPM8 and executed in MATLAB (Mathworks, Natick, MA).

All functional MRI data was analyzed using SPM8 (http://www.fil.ion.ucl.ac.uk/spm/). Preprocessing included slice timing correction, realignment, co-registration to the structural images, segmentation of the structural image, normalization to the Montreal Neurologic Institute (MNI) standard brain, and smoothing using an 8 mm full-width at half maximum Gaussian kernel. During single subject analysis, the following regressors were built: Neg_C, Neg_S, Neg_IC, Neu_C, Neu_S, Neu_IC. Contrast images were created to investigate (1) the main effect of emotion (Neg>Neu trials), (2) the main effect of cognition (IC>C or IC>S trials), and (3) the influence of emotion on cognition along with increasing cognitive demand as based on two-sample t-tests comparing Neg_C vs. Neu_C, Neg_S vs. Neu_S, and Neg_IC vs. Neu_IC trials.

Due to the challenges in capturing the intricate nature of emotion–cognition interactions, the majority of publications in this field have based their interpretation on a priori based regions of interests only. Here, we present both small volume peak-level FWE-corrected findings at p < 0.05 for the main regions of interest [i.e., amygdala, insula, and/inferior frontal junction/precentral gyrus according to previous literature (Gray et al., 2002; Etkin et al., 2006; Blair et al., 2007; Hart et al., 2010; Gu et al., 2013); defined anatomically using the automated anatomical labeling atlas (Tzourio-Mazoyer et al., 2002)] and uncorrected, exploratory, whole brain findings (p < 0.001).

In order to further characterize the effects of cognitive load on the neural basis of emotion–cognition interactions, we further extracted mean peak activation scores as based on FWE-corrected findings from the two main contrasts targeting the emotion (Neg>Neu trials) and cognition (IC>S trials) networks of the brain. More specifically the signal change at the local peak activation scores for bilateral amygdala, right insula, and bilateral precentral gyrus were extracted using the marsbar toolbox (Brett et al., 2002) and further assessed using paired-samples t-tests.

In-scanner performance was assessed by computing the mean accuracy and reaction time in response to Neg_C, Neg_S, Neg_IC, Neu_C, Neu_S, and Neu_IC Stroop trials. For both accuracy and reaction times two separate 2 (emotion: Neg, Neu) by 3 (task: C, S, IC) repeated measures ANOVAs were performed in order to investigate the main effect of task, main effect of emotion, and the influence of emotion on task.

The 2 (emotion: Neg, Neu) by 3 (task: C, S, IC) ANOVA on task accuracy (i.e., correctly answered Stroop trials) indicated a significant main effect of emotion [F_{(1, 29)} = 7.34, p = 0.011] and a main effect of cognition [F_{(2, 28)} = 12.38, p < 0.0001]. Bonferroni corrected post-hoc tests indicated that the main effect on emotion was constituted by lower accuracy following negative primes (compared to neutral primes) during Stroop task. Furthermore, significant differences in accuracy derived from the incongruent condition compared to the congruent (p < 0.0001) and stars condition (p = 0.002). However, the difference between congruent and stars conditions did not reach significance (p = 1.00). Finally, the emotion by cognition interaction did not reach significance [F_{(2, 28)} = 0.33, p = 0.722].

The 2 × 3 ANOVA implementing reaction time revealed a main effect of emotion [F_{(1, 29)} = 11.93, p = 0.002] and a main effect of cognition [F_{(2, 28)} = 80.27, p < 0.001]. Bonferroni corrected post-hoc tests revealed significant reaction time differences for incongruent compared to congruent (p < 0.0001), congruent compared to stars (p = 0.019), and incongruent compared to stars (p < 0.0001) conditions. The emotion by cognition interaction did not reach significance [F_{(2, 28)} = 0.54, p = 0.590]. An overview of the in-scanner performance as based on the mean accuracy and reaction time for the whole group is given in Table 2.

The fMRI result part is organized in line with our a priori listed main aims: (1) testing the activation of the emotion network by use of the negative prime (Neg>Neu trials); (2) assessing the activation of the cognition network by comparing an incongruent to a neutral Stroop condition (IC>S); (3) evaluating the emotion–cognition interaction dependent on increased cognitive load (Neg_C vs. Neu_C, Neg_S vs. Neu_S, and Neg_IC vs. Neu_IC trials). First, testing the (1) emotion processing network revealed that trials with a preceding negative prime compared to those with a preceding neutral prime led to significant increases in activation in known emotion processing areas of the brain (Phan et al., 2002), including insula, amygdala, and prefrontal cortices (for an overview of activated areas see Table 3, Figure 2). Secondly, testing the (2) cognition network by use of the Stroop task (IC>S) revealed activations in areas including left precentral gyrus (FWE-corrected) and uncorrected within further areas including the superior frontal brain regions, temporal cortex, and insula (Table 3, Figure 2).

Both control conditions (IC>S and IC>B) were contrasted with the IC condition. However, we decided to focus on IC>S trials for definition of the cognition network since more prefrontal activation was evoked, potentially due to different effects of the primes on congruent as opposed to stars trials. This procedure is in line with similar previous fMRI Stroop publications (for a review see Laird et al., 2005). Finally, (3) the influence of emotion on cognitive control was measured by contrasting Stroop trials with a prior negative prime to those Stroop trials following a neutral trial. Two-sample t-tests for negative vs. neutral trials were calculated for Stroop trials with increasing cognitive load (from congruent, to stars, and incongruent condition) and revealed differential activations for each of the three contrasts reflecting the influence of emotion on cognitive task performance as modulated by differential cognitive load. More specifically, for the contrast “Neg>Neu trials” during the congruent Stroop task condition, significant increases in activations were identified in bilateral amygdala, right insula, and right precentral gyrus (FWE-corrected) and on a whole-brain uncorrected level within further regions of the occipital, temporal, and inferior/middle frontal gyrus. For the opposite contrast “Neg < Neu trials” during the congruent Stroop trial, FWE-corrected activation within left precentral gyrus was observed. For the contrast “Neg>Neu trials” during the stars Stroop task condition, significant FWE-corrected findings were identified in right amygdala, while further cluster of activations were located within bilateral inferior occipital cortex (uncorrected whole brain approach). For the opposite contrast “Neg < Neu trials” significant activation was located within right precentral gyrus (FWE-corrected) and using a whole brain approach further clusters were located within occipital and superior frontal brain regions, left precuneus of the superior parietal lobe. Finally, for the contrast “Neg>Neu trials” during the incongruent Stroop task condition significant clusters of activations were based in inferior frontal and middle frontal brain regions (uncorrected findings only), while the opposite contrast of “Neg < Neu trials” during the incongruent Stroop task condition led to activity within left insula (FWE-corrected) as well as in further occipital brain regions, left pre- and postcentral gyrus, when using an uncorrected whole-brain approach (Table 4, Figure 3).

In line with previous work, negative compared to neutral primes prior to a Stroop task affected task performance, which resulted in decreased task accuracy and increased reaction times (Gray et al., 2002; Mitchell et al., 2006; Blair et al., 2007; Padmala et al., 2011). Furthermore, more incorrect answers were given during incongruent trials compared to congruent (Etkin et al., 2006) or stars trials, while the latter two showed similar accuracy measurements. Participants were fastest in responding to congruent trials, followed by stars trials and eventually incongruent trials, which reflects the so-called Stroop effect (Stroop, 1935; Blair et al., 2007; Hart et al., 2010). We propose that in this study reading the number was a more automatized/faster process compared to counting the actual stimuli, thus resulting in shorter reaction times for congruent compared to stars trials. Interestingly, congruent and stars trials had equivalent numbers of correctly answered trials, although participants responded faster to congruent than to stars trials. We therefore conclude that the behavioral advantage of the congruent condition only affected reaction time (increased speed), but not accuracy. However, behavioral analysis failed to observe an interaction effect between emotional priming and task difficulty in accuracy and reaction times. In line with the dual competition model by Pessoa (2009) and supported by previous studies (Hart et al., 2010; Melcher et al., 2011; Padmala et al., 2011) the interference of negative primes on task performance was expected to augment with higher cognitive task load due to competing mechanisms. However, in line with previous evidence (e.g., Blair et al., 2007; Van Dillen et al., 2009), we did not observe a significant interaction effect in the present analysis.

Here, we demonstrate that the influence of emotion on cognition is neurally reflected within brain regions including prefrontal cortex, amygdala, and insular cortex. Furthermore, activation in these brain regions shows an attenuating trend when increasing cognitive demand. Likewise, neural activation in the left inferior frontal junction/precentral gyrus, as indicative of cognitive control, is increased during neutral prime trials compared to trials following a negative prime. Prefrontal brain regions, amygdala, and insula have all consistently been identified as relevant for emotion–cognition interactions or emotional conflict resolution (Gray et al., 2002; Beer et al., 2006; Etkin et al., 2006; Blair et al., 2007; Hart et al., 2010; Gu et al., 2013; Buhle et al., 2014). Likewise, similar areas are activated during tasks requiring cognitive reappraisal (an emotion regulation strategy in which the stimulus meaning is reinterpreted to downregulate the emotional valence; Ochsner et al., 2002, 2004; Ochsner and Gross, 2005). Thus, we suggest that the here presented emotional number Stroop task activates similar areas within the neural network that are required for deliberate cognitive reappraisal.

The prefrontal cortex can functionally and cytoarchitectonically be subdivided into distinct sub-regions, several of which are of relevance to affective processing, cognition, or both (Pessoa, 2008). Overall, prefrontal brain regions are commonly linked to attention, working memory, goal-directed behavior (e.g., cognitive control or decision making (Pessoa, 2008; Stokes et al., 2013; Leech and Sharp, 2014), and affective processing (Phan et al., 2002). From an evolutionary perspective, early research has suggested that the evolution of the human prefrontal cortex, particularly its expansion in volume, may reflect the development of more complex social behavior (Dunbar and Shultz, 2007). While such an interpretation may be too simplistic, still researchers commonly agree that the distinct parts of the prefrontal cortex are recruited by different high-level cognitive demands (Eickhoff et al., 2016). Based on animal and human studies, a functional and cytoarchitectonical subdivision of the medial prefrontal cortex may at least result in areas including the orbitofrontal cortex (BA11), ventral prefrontal cortex and prefrontal pole (BA10), and dorsomedial prefrontal cortex and frontal pole (BA9; Eickhoff et al., 2016). All of these areas are strongly interconnected and associated with various other circuitries of the brain, including the limbic network (Reid et al., 2016). Finally, particularly the right inferior frontal gyrus has been suggested to be a crucial hub during inhibitory processing and may consequently be an area affected in response control disorders (Aron et al., 2014).

Emotion processing is generally assigned to medial prefrontal brain regions, whereas the amygdala, anterior cingulate cortex, or insula are thought to possess a more distinct function within emotional tasks (Phan et al., 2002). The amygdala is one of the most traditionally viewed emotion and motivation processing center. With its relatively small structure, the amygdala nevertheless comprises a multitude of anatomical connections allowing many intricate functionalities (Janak and Tye, 2015). While abundant research has linked the amygdala to affective processing and particularly fear conditioning, strong evidence points toward a more integral role of the amygdala as a key node for valence processing during different aversive states, including fear, anxiety, or reward processing (Murray et al., 2014; Janak and Tye, 2015). Here we demonstrated that the amount of amygdala activation obtained during emotion–cognition interaction was highest during Stroop trials with lowest cognitive demand and decreases with increasing task difficulty. This finding is in line with Etkin et al. (2006) as well as Blair et al. (2007) that found decreases in neural activation within the amygdala during concurrent task with increasing cognitive demand. In line with previous suggestions (Etkin et al., 2006), it could be concluded that amygdala activation mirrors the amount of emotional conflict, rather than the resolution of such. Importantly the amygdala is strongly interconnected with areas of the prefrontal/orbitofrontal cortex (Davidson et al., 2000). This bi-directional connection allows regulatory processes important for mental well-being. For example, early deprivation or life stress may lead to disruption in amygdala-prefrontal coupling and patients with symptoms including anxiety, posttraumatic stress disorder, or heightened aggression oftentimes show structural and functional impairments within these circuitries (Gee et al., 2013).

The present task closely resembles two prior fMRI study designs (Blair et al., 2007; Hart et al., 2010). With the exception of adaptations that account for the age of the participants being tested (i.e., use of only negative stimuli and age-appropriate images), it may be considered a replication study. The present manuscript used an adapted version of an affective number Stroop task as implemented by Hart et al. (2010) and resulted in comparable findings. Reproducibility of scientific studies is crucial in order to inform about the robustness of an observed phenomenon (Martin and Clarke, 2017). Comparing our findings more closely to these two prior studies confirm the following main findings: (1) In line with both studies, negative primes slowed the participants' reaction time during the number Stroop task and are thus confirmed to interrupt goal-directed processing; (2) In line with Blair et al. (2007) increasing cognitive demand led to decreases in emotion-related brain regions (e.g., amygdala, insula, prefrontal cortex). Hart et al. (2010), observed the same trend however, no neural decreases in dorsolateral prefrontal areas during incongruent trials. Therefore, the authors concluded that high cognitive demand may override the attenuation effect in the prefrontal cortex. In contrast to Hart et al. (2010), we only observed decreases in neural activation with increasing cognitive demand. It remains to be investigated whether such a difference may be due to the number and characteristics of participants tested (N = 14; 5 males in Hart et al. (2010)/N = 30; 15 males in the present study) or are potentially due to the difference in stimuli choice and/or slightly longer presentation (an additional 500 ms) of the Stroop trial in our study. It is important to note that the slightly longer Stroop presentation rate was chosen due to the aim of consequently applying this task in younger participants.

The insula is a functionally heterogeneous brain region which is situated in the depth of the Sylvian fissure and may be divided into three sections: a dorsal anterior, a ventral anterior, and a posterior part (Nieuwenhuys, 2012; Uddin et al., 2014). The anterior insular cortex is, mostly bilaterally, connected to limbic and prefrontal brain regions (e.g., the amygdala), while the posterior part is more strongly interconnected with parietal, occipital and temporal parts of the brain (Kurth et al., 2010; Nieuwenhuys, 2012). The insula has shown to be activated during a wide range of functions, including auditory processing, vestibular and somatosensory functions, the perception of pain and temperature, viscerosensation, taste, olfactory processing, somatomotor control and motor plasticity, speech production, cognitive control, bodily awareness, as well as emotion processing (Nieuwenhuys, 2012). Importantly, according to research the anterior insula has a critical role during the regulation of social behavior, since its structure and function is altered in individuals with social disorders (including disruptive behavior disorder Sterzer et al., 2007; Raschle et al., 2015). Here our results are in line with findings assigning a critical role for the insula in emotion–cognition interactions (Hart et al., 2010; Shackman et al., 2011; Gu et al., 2013).

An intact integration and healthy balance of competing emotion-cognition processing is crucial for our everyday functioning and an imbalance, as for example observed in individuals with emotion processing deficits, is linked to different mental health disorders (Monk, 2008). For example, faulty integration or regulation of emotion-cognition processes may result in heightened violence and aggression (Davidson et al., 2000). Therefore, it has been suggested that individuals with heightened aggression traits, as for example observed in children and adolescents with disruptive behavior disorders, may show impairments in these prefrontal circuitries responsible for successful emotion–cognition interaction and regulation. In fact, various structural and functional neuroimaging studies pinpoint areas of the limbic and prefrontal network to be disrupted in aggressive individuals (Raschle et al., 2015; Rogers and De Brito, 2016). Consequently, we conclude that future studies may implement the here presented design in order to further characterize aggressive youths and potentially impact individualized classification and treatment approaches in health and disease.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.