Emotional Interference in Early Adolescence: Positive Reinforcement Modulates the Behavioral and Neural Effects of Negative Emotional Distracters

Abstract

Limited research has examined functioning within fronto-limbic systems subserving the resistance to emotional interference in adolescence despite evidence indicating that alterations in these systems are implicated in the developmental trajectories of affective disorders. This study examined the functioning of fronto-limbic systems subserving emotional interference in early adolescence and whether positive reinforcement could modulate these systems to promote resistance to emotional distraction. Fifty healthy early adolescents (10–13 years old) completed an emotional delayed working memory (WM) paradigm in which no distractors (fixation crosshair) and emotional distracters (neutral and negative images) were presented with and without positive reinforcement for correct responses. WM accuracy decreased with negative distracters relative to neutral distracters and no distracters, and activation increased in amygdala and prefrontal cortical (PFC) regions (ventrolateral, dorsomedial, ventromedial, and subgenual anterior cingulate) with negative distracters compared with those with no distracters. Reinforcement improved performance and reduced activation in the amygdala, dorsomedial PFC, and ventrolateral PFC. Decreases in amygdala activation to negative distracters due to reinforcement mediated observed decreases in reaction times. These findings demonstrate that healthy adolescents recruit similar fronto-limbic systems subserving emotional interference as adults and that positive reinforcement can modulate fronto-limbic systems to promote resistance to emotional distraction.

Introduction

The ability to sustain attention on goal-directed behavior while resisting interference from distracting emotional information (e.g., focusing on one’s teacher and ignoring an angry classmate) is critical for adaptive behavior. In adults, the ability to successfully prevent emotional interference appears to rely on complex interactions between frontal and limbic systems (e.g., Dolcos et al. 2006; Dolcos and McCarthy 2006). However, limited research has examined functioning within fronto-limbic systems subserving the resistance to emotional interference in adolescence (cf., Vetter et al. 2015; Ladouceur et al. 2018). As such, it is unclear whether findings from the adult literature extend to adolescents. Replication with adolescents could have important clinical implications in light of evidence suggesting that the altered functioning of fronto-limbic systems may contribute to the developmental trajectories of risk for affective disorders (Joormann et al. 2007; Ladouceur et al. 2009, 2013; Ladouceur 2012). Furthermore, given that emotional interference is a disruptive characteristic of affective disorders in adults (Jones et al. 2015, 2016) and adolescents (Colich et al. 2017), identifying variables or procedures that can be used to buffer against interference-related declines in goal-directed behavior is an important yet largely unexplored area of research (cf., Ladouceur et al. 2018).

According to some theories, emotional information is prioritized over concurrent goal-directed activity (Mogg et al. 2000; Armony and Dolan 2002), because emotion information contains signals closely linked to survival (LeDoux 2000). Thus, emotion information is highly motivationally salient and captures attention (Anderson and Phelps 2001; Piech et al. 2011). Some models posit that emotional information can cause interference with ongoing goal-directed activity because it competes for resources at both perceptual and executive levels of processing (i.e., the dual competition model; Pessoa 2009). This competition for limited resources arises because emotional information is integrated with cognitive control processes through complex interactions between subcortical and prefrontal regions (Pessoa 2015). Using cognitive tasks modified to include task irrelevant emotional distracters, numerous studies have provided evidence that emotional stimuli tend to “hijack” attention more easily than nonemotional stimuli resulting in disrupted goal pursuit and less optimal task performance (e.g., slower reaction times and reduced accuracy; Dolcos and McCarthy 2006; Anticevic et al. 2010; Denkova et al. 2010). The magnitude of the interference effect varies as a function of how arousing the emotional information is and its task relevance such that highly arousing, task-irrelevant emotional information results in a greater diversion of executive resources and stronger negative impact on performance, particularly in the context of mental illness (Schweizer et al. 2019).

One task that has been used to examine the functioning of fronto-limbic systems underlying attentional control in the context of emotional distracters is the emotional delayed working memory (EDWM) task. The EDWM task is a modified version of a visual delayed match-to-sample task that includes a visual probe held in working memory (WM) during a delay period and typically involves the presentation of no distracter, a neutral distracter, or an emotional distracter (e.g., negatively valenced pictures). In adults, recent meta-analyses of emotional interference (Hung et al. 2018; Schweizer et al. 2019) have found that processing negative emotional distracting information is consistently associated with increased activation in the temporo-occipital lobe, including the fusiform gyrus (i.e., the ventral visual stream), the ventrolateral prefrontal cortex (VLPFC), orbital frontal cortex (OFC), and the amygdala, and decreased activation in the dorsolateral prefrontal cortex (DLPFC). This pattern of activation is typically interpreted as an indication that the resources allocated to the DLPFC—implicated in active maintenance of task-relevant information—is momentarily commandeered due to the salience and biological relevance of the visual affective stimuli, as processed by the amygdala and OFC (Dolcos et al. 2008). Consistent with this assertion, studies demonstrate that greater activation within the amygdala (Anticevic et al. 2010; Dolcos et al. 2013) and decreased activation in the DLPFC during the delay period (Dolcos and McCarthy 2006; Dolcos et al. 2008) are associated with poorer WM accuracy. Furthermore, evidence consistently points to activation with the VLPFC serving a key role in coping with emotional interference. Increased VLPFC activation is associated with decreased subjective ratings of distractibility and the emotionality of negative affective stimuli (Dolcos and McCarthy 2006), as well as increased WM accuracy (Dolcos et al. 2006, 2013; Dolcos and McCarthy 2006; Anticevic et al. 2010; Iordan and Dolcos 2017).

The majority of research into the functioning of fronto-limbic systems subserving resisting emotional interference has been conducted in adults. Investigating the functioning of these systems with different age groups is an important step needed to firmly establish the generality of emotional interference effects and presumed brain mechanisms. Given evidence that prefrontal cortical (PFC) regions and their connectivity to subcortical regions, such as the amygdala, undergo important maturational changes during adolescence (Gee et al. 2013; Casey et al. 2019), while subcortical regions implicated in emotional processing are particularly active compared with younger children and adults (Guyer et al. 2016), it is possible that adolescents could show greater emotional interference associated with less activation in PFC regions than adults. Limited evidence is emerging which supports the generality of emotional interference effects and the presumed brain mechanism in adolescents. Vetter et al. (2015) used a perceptual discrimination task to examine emotional interference. Preventing emotional interference was operationalized as the ability to correctly compare two shredded images to determine their equality while ignoring emotional stimuli. Preventing emotional interference was contrasted against attending to emotional stimuli, operationalized as determining the equality of intact emotional stimuli while ignoring shredded emotional stimuli. Attending to emotional stimuli relative to resisting emotional interference was associated with increased activation in the ventral visual stream, amygdala (to negative stimuli only), and VLPFC, and with reduced activation in the DLPFC and inferior parietal lobe. Similarly, in our preliminary study conducted in a separate small and under-powered sample, we have demonstrated that processing emotional distracters is associated with greater activation in the amygdala, ventromedial prefrontal cortex (vmPFC) extending into the dorsomedial prefrontal cortex (dmPFC), and the VLPFC during the delay period of the EDWM task adapted for use with adolescents (Ladouceur et al. 2018). Taken together, these studies begin to demonstrate that the same pattern of fronto-limbic interactions observed in the adult literature is replicating in adolescence. While illustrative, neither study examined brain-behavior associations. Specifically, no identified research in adolescence has examined associations between amygdala functioning and WM performance despite evidence from the adult literature indicating such an association exists (Anticevic et al. 2010; Dolcos et al. 2013). Thus, the first goal of the current investigation is to replicate and extend findings in young adolescents by examining the functioning of fronto-limbic systems underlying resistance to emotional interference and examining brain-behavior associations to elucidate potential brain mechanisms underlying resistance to emotional interference.

Given that susceptibility to emotional interference is a deficit present in both adults (Jones et al. 2015, 2016) and adolescence (Colich et al. 2017) diagnosed with affective disorders, identifying variables or procedures that can be used to prevent interference and associated declines in goal-directed behavior are an important yet largely unexplored area of research. In particular, research indicates that affective disorders are associated with biased processing of emotional stimuli (Ladouceur et al. 2005; Monk et al. 2008; Phillips et al. 2008; Britton et al. 2013) and abnormal recruitment of top-down cognitive control, particularly in the presence of interfering emotional stimuli (e.g., fearful images, Colich et al. 2017) and internal emotional states (e.g., anxious arousal, Jones et al. 2015, 2016), which are associated with deficits in goal-directed behavior. Hence, there is a need to identify ways of promoting attentional control of emotion. As reviewed, evidence suggests that emotional responses and emotional interference can be mitigated by increased recruitment of top-down cognitive control regions (Dolcos et al. 2006, 2013; Dolcos and McCarthy 2006; Anticevic et al. 2010). A number of animal and human neuroimaging studies have shown that positive reinforcement (i.e., receipt of a reward following a specific response) boosts behavioral performance on cognitive control tasks such as WM (Watanabe 1996; Pochon et al. 2002; Beck et al. 2010; Jimura et al. 2010; Bahlmann et al. 2015; Iordan and Dolcos 2017; Bruening et al. 2018) and that such improvement in performance is associated with increased neural activation in dorsal and lateral PFC regions. Such findings suggest that positive reinforcement for correct responding on the EDWM task could boost behavioral performance by counteracting the effects of emotional interference. There are a number of potential routes through which positive reinforcement may exert influence. These include reduced activation in brain regions supporting the processing and coping with emotional distracters (e.g., amygdala, vmPFC and VLPFC), which was observed in our preliminary study (Ladouceur et al. 2018), increased activation in attentional control regions (lateral prefrontal cortex), or both. As such, it is likely that reduced activation in emotional processing regions and/or increased activation in prefrontal cognitive regions occurring as a function of reward could mediate changes in behavior performance. Accordingly, the second goal of this study was to test the extent to which providing positive reinforcement (i.e., monetary rewards for correct responses) would improve task performance and modulate activation in PFC and subcortical regions involved in emotional interference in a larger sample of adolescents.

We hypothesized that: 1) we would replicate in adolescents the patterns of activation associated with emotional interference observed in adults; 2) positive reinforcement would be associated with reduced activation in limbic and ventral neural regions (i.e., amygdala and VLPFC); 3) greater amygdala activation during negative emotional distracters would be associated with poorer accuracy and slower reaction times; and 4) the magnitude of the decrease in amygdala activation, in response to emotional distracters, due to positive reinforcement would mediate changes in task performance.

Materials and Methods

Participants

One hundred and one healthy youths (52% girls) aged between 10 and 13 years old and with no history of psychiatric or neurological disorders participated in the study. Participants were recruited from the community through advertisements and flyers. They were recruited to be within a narrow age-range due to the aims of the larger longitudinal project in which participants were enrolled. Thirty-eight participants were excluded for having excessive motion (i.e., >25% of scans with motion >5 mm and/or 3 standard deviation [SD] intensity shifts) as determined by ArtRepair (Mazaika et al. 2005); 13 were excluded for poor behavioral performance (<65% accuracy). Analyses were conducted on the remaining 50 participants (mean age: 12.00 years; SD = 1.06, range: 10.11–13.91 years, 58% female, 72% Caucasian, 18% African American, and 5% unknown).

Exclusion criteria included the presence of psychiatric disorder, visual disturbance (<20/40 Snellen visual acuity), being pregnant, MRI contraindications (e.g., metal in the body, claustrophobia), or taking oral steroids. Parental informed consent was provided and adolescent assent was obtained prior to scanning. The study was approved by the University of Pittsburgh Institutional Review Board. Participants received performance-based earnings in addition to compensation for their participation.

Screening Measures

The Columbia Diagnostic Interview for Children (Shaffer et al. 2000) a computerized structured interview used with adolescents and their parent about their child, was used to screen participants for the presence of psychiatric diagnosis based on criteria from the DSM-IV (American Psychiatric Association 1994).

Pretask Practice

Immediately prior to the scanning session, participants read instructions and completed a shortened version of the task in a mock scanner to familiarize them with the EDWM task. The practice session also included training participants to map numbers 1–2 to corresponding buttons on the response glove as well as exposing participants to the sounds of the scanner and teaching them how to remain still during the scan.

fMRI EDWM Paradigm

Participants completed two fMRI paradigms during a 90-min MRI scan at the Magnetic Resonance Research Center, University of Pittsburgh Medical Center Health System, USA. The focus of the current investigation was their performance on an adapted version of the EDWM task (Anticevic et al. 2011). The EDWM task included 80 trials of a version of a delayed match-to-sample task (Sternberg 1969) with two geometric shapes and two potential distracter types presented during the delay or the maintenance period of the task: emotionally negative images and visually complex neutral images (see Fig. 1). Trials began with the presentation of the memoranda (3 s) followed by a fixation cross (1 s), a delay period (fixation cross (10 s) or distracter (3 s) and fixation cross (7 s)), the probe (2.5 s), inter-trial stimulus (in blocks without positive reinforcement: fixation cross (7 s); in blocks with positive reinforcement, feedback (0.5 s): “Correct: win $1” written in green following correct responses or “Wrong: no money” written in red following incorrect responses or “No response” written in white following omissions and fixation cross (6.5 s). A PC running E-prime software (psychology software tools (PST)) controlled stimulus display. A color high-resolution LCD projector projected visual stimuli onto a rear screen at the head of the scanner bore, viewable via a mirror attached to the head coil. Responses were recorded using a PST glove.

Figure 1.

Illustration of the adapted version of the EDWM task with and without positive reinforcement for correct responding on trials with no, neutral or negative distracters. ITI: interstimulus interval.

The 80 trials of the EDWM task were divided according to distracter type: 30 negative distracter trials, 30 neutral distracter trials, and 20 no distracter trials, which were blank trials with a fixation cross used to estimate distracter-free maintenance activity. The trials were grouped into eight blocks of 10 trials each. The first two blocks included trials with no distracters (block 1 without positive reinforcement and block 2 with positive reinforcement). The following six blocks included 30 neutral and 30 negative trials presented in a fixed random order within each block. Three blocks were presented without positive reinforcement (blocks 3, 5, and 7), and three blocks included positive reinforcement (blocks 4, 6, and 8). The even-numbered blocks included the same trials and timing as the odd-number blocks but with the added instructions and feedback pertaining to reinforcement. Each trial lasted 24 s with each block lasting 4.3 min. As in Anticevic et al. (2010), the memory sets were constructed from complex geometric shapes (Attneave and Arnoult 1956) created using a MATLAB algorithm designed to generate geometric shapes that were not easily described and encoded verbally (Collin and McMullen 2002). The negative and neutral distracters were a subset of digitized slides from the international affective picture system (IAPS) stimulus set (Lang et al. 2008) selected for use with children and adolescents (McManis et al. 2001). All distracters were presented at the center of the screen, with a visual angle of 8.5°. At the end of the scan, participants were informed that the monetary reward received was to be added to their participant payment card.

Prior to the start of the experiment, each participant was presented with instructions explaining the task and completed a practice session. They were instructed to try to remember two shapes presented on the screen and to keep their eyes looking at the center of the screen and to not respond once the shapes disappeared. They were also told that a single shape would then be presented for a brief time and to press a button with their index finger (yes) if this shape matched one of the shapes previously presented and to press a button with their middle finger (no) if this shape did not match any of the shapes. They were also informed that at some point during the task, graphic images would be presented. In order to examine the influence of positive reinforcement on the performance of the EDWM task, each block was repeated, but the distracters (neutral and negative) within the blocks were randomized, and blocks 2, 4, 6, and 8 included instructions at the beginning of the block indicating to the participants that they would receive a monetary reward ($1) for each correct response.

Manipulation Check: Postscan Valence and Arousal Rating Task

In order to account for the possibility that the emotional distracters were not perceived as emotionally salient, participants performed a computerized valence and arousal rating task immediately following exiting the scanner. The task consisted of viewing a series of 80 IAPS pictures (38 negative and 42 neutral) that were presented in a random order and included pictures used as emotional distracters in the EDWM task (36%) and pictures that were randomly selected from the IAPS series (64%). Participants were asked to use the Self-Assessment Manikin (Bradley and Lang 1994; Lang et al. 1999) rating scale which is a nonverbal pictorial assessment technique, to rate their affective reactions to the pictures in terms of valence (1 = positive to 9 = negative) and arousal (1 = excited and 9 = calm). Only ratings of images used in the EDWM task were analyzed.

MRI Data Acquisition

Scanning was performed on a 3 T Siemens Biograph mMR scanner. Functional T2^* weighted images were acquired using an echo-planar imaging sequence (repetition time (TR): 2000 ms; echo time (TE): 26 ms; flip angle: 90°; FOV: 205 × 205 mm; acquisition matrix: 64 × 64; slice thickness: 3.1 mm; number of slices: 38 slices). Field map images were also acquired during the scanning session (TR: 40 ms; TE: 4.92 ms; TE2: 7.38 ms; flip angle: 35°; field of view (FOV): 256 × 256 mm; acquisition matrix: 128 × 128; slice thickness: 3.1 mm; number of slices: 38). Both the functional T2^* weighted and field mapping images were acquired as axial slices aligned with the anterior commissure–posterior commissure (AC–PC) line at midline. Due to a change in parameters used for the field map acquisition that occurred after study recruitment had commenced, field maps were not collected for 12 participants. A high-resolution, T1-weighted magnetization-prepared rapid-acquisition gradient echo anatomical scan was acquired in the axial plane for each participant at the beginning of the functional scanning session (slice thickness-1 mm, voxel size = 1 × 1 × 1 mm³, 160 slices, TR = 2300 ms, TE = 2.47 ms, flip angle = 9^o, matrix = 256 × 256, FOV = 256 mm).

Data Analyses

Behavioral Data

Accuracy (% hits + % correct rejections/2) and reaction times (correct trials only) were computed for each condition for each participant. Data were analyzed using a repeated measures analysis of variance in SAS V.9.4 with emotional distracter type (no distracter, neutral or negative) and reinforcement condition (with or without) as within-subject factors and an alpha criterion set at P < 0.05. Post hoc comparisons were performed using false discovery rate (FDR) corrections (Benjamini and Hochberg 1995).

fMRI Data

Image preprocessing was performed with Statistical Parametric Mapping software (SPM8; Wellcome Trust Centre for Neuroimaging). These steps included correction for differences in acquisition time between slices, correction for motion (realignment using six-parameter rigid body), and unwarping using a field map, coregistration of the unwarped image to the participant’s structural image. For the 12 participants’ missing filed maps, the unwarping preprocessing step was not conducted; the realigned, motion corrected images were coregistered to the participant’s structural image. Images were normalized using the “unified segmentation” approach, including the segmentation of the structural image and registration into MNI space (ICBM template), and then the registration of functional images to MNI space was performed based on the parameters of the structural registration. Finally, normalized images were smoothed using a 6-mm FWHM Gaussian kernel.

For the first-level analysis, individual effects were estimated using the general linear model (GLM) approach implemented in SPM8. We used a finite impulse response (FIR) approach to analyze task-related blood oxygen-level-dependent (BOLD) responses. This allows the statistical model to determine the shape of the hemodynamic response function across each 24 s trial. The FIR model was specified with a 2 s sampling rate resulting in an HRF amplitude being estimated for each TR (i.e., 12 timepoints). We modeled the onsets of each distracter type with and without positive reinforcement (i.e., six conditions). Only trials with correct responses were used in analyses (Without reinforcement: no distracter Mdn = 9, range: 4–10; neutral Mdn = 12, range: 7–15; negative Mdn = 11, range: 6–14. With reinforcement: no distracter Mdn = 9, range: 6–10; neutral Mdn = 13, range: 9–15; negative Mdn = 12, range: 8–15). An FIR approach was used because multiple psychological processes are unfolding in succession and the main process of interest, that is, emotional interference, is hypothesized to have a sustained effect whereby emotional processing continues into another cognitive phase of the task. Furthermore, modeling only the delay period would have made it difficult to interpret relative “deactivations” in the BOLD response.

Second-level voxel-wise analyses testing Hypothesis 1 and Hypothesis 2, described below, were restricted to regions-of-interest based on the reviewed literature. An anatomically based region-of-interest region of interest (ROI) mask of the prefrontal cortex was created with the automated anatomic labeling atlas (Tzourio-Mazoyer et al. 2002) of the WFU Pickatlas toolbox (Maldjian et al. 2003). A single mask encompassing the ventromedial (BA 10 and 11), ventrolateral (BA 45 and 47) and the dorsolateral (BA 9 and 46), regions of the prefrontal cortex was created. Broadman areas were dilated prior to inclusion in the mask.

Due to our a priori interest in using the amygdala in subsequent mediation analyses and to prevent concerns related to “double dipping,” we extracted averaged activation across all voxels located within an anatomically defined amygdala ROI, for each combination of distracter type and reinforcement condition for each TR (12), for each hemisphere, and for each subject. The amygdala was anatomically defined using an anatomical mask created using the WFU Pickatlas toolbox (Maldjian et al. 2003). Amygdala activation was winsorized prior to conducting brain behavior associations. To our knowledge, there are no established models proposing dynamic shifts in left and right amygdala functions across development. Nevertheless, several meta-analytic reviews of emotional information processing in adults indicate that the left amygdala is more frequently activated than the right amygdala, to negative stimuli (Wager et al. 2003; Baas et al. 2004; Fusar-Poli et al. 2009). Thus, given evidence that the amygdala and its connectivity to PFC regions are still developing during childhood and adolescence (Gee et al. 2013; Casey et al. 2019), it is possible that neurodevelopmental changes during adolescence may result in differential programming, which we wanted to capture by testing left and right amygdala independently.

In supplementary analyses, we also conducted whole-brain analyses that were not restricted these masks (see Supplemental Tables S1 and S2).

Hypothesis Analysis Strategy

Hypothesis 1

To address our first hypothesis pertaining to the effects of emotional distracters on fronto-limbic activation, a full factorial GLM focusing on distracter type (no distracters, neutral distracters and negative distracters—without reinforcement) by time (12 TRs) interaction was performed using GLM Flex Fast2 (http://mrtools.mgh.harvard.edu/index.php?title=GLM_Flex_Fast2). The Type I error rate was controlled using AFNI’s 3dClustSim using the autocorrelation function (-ACF) option (Cox et al. 2016). Smoothness was estimated from the level 2 residuals, and a voxel-wise threshold of P < 1 × 10⁻⁷ was used for cluster forming to increase the spatial/anatomical specificity of the resulting clusters in accordance with recommendations made by Woo et al. (2014). These methods also address issues raised by Eklund et al. (2016). Results indicated that an extent threshold of 1 voxel corresponded to a cluster-wise family-wise error rate of P < 0.05. The small extent threshold is a result of the very stringent cluster forming threshold given that the ROI mask’s total volume (total volume = 16 068 voxels) well exceeds the minimum cluster size for 3dClustSim (128 voxels). Averaged voxel parameter estimates over time were extracted (based on the significant clusters from the interaction maps) for each subject for all conditions.

Using SAS software version 9.4, we examined the significant distracter by time interaction for each of these significant clusters using mixed effects analyses where condition and time were modeled as repeated factors with residuals modeled as unconditional for condition and autoregressive (AR[1]) for time. The only purpose for computing these models was to conduct simple contrasts using the appropriate degrees of freedom from the mixed models; we do not report statistics related to the overall interaction effects because these estimates for the voxel-wise analyses limited to the prefrontal cortex would be inflated due to “double dipping.” A priori simple contrasts (negative > neutral and negative > baseline distracters) of average activation in the delay period (8–16 s post-trial onset) were conducted on least square mean estimates. Parallel mixed effects analyses were conducted for the anatomically defined left and right amygdala clusters.

Hypothesis 2

To evaluate the effects of positive reinforcement on fronto-limbic activation during the negative distracter trials, a full factorial GLM focusing on negative distracters with and without positive reinforcement by time (12 TRs) interaction was performed using GLM Flex Fast2. The Type I error rate was controlled using the same method as Hypothesis 1. Smoothness was estimated from the level 2 residuals and a voxel-wise threshold of P < 0.001 was used for cluster forming (see Supplemental Methods for the rationale supporting varying cluster-forming thresholds). Results indicated that an extent threshold of 75 voxels corresponded to a cluster-wise family-wise error rate of P < 0.05. Averaged voxel parameter estimates over time were extracted (based on the significant clusters from the interaction maps) for each subject for all conditions. Using SAS V9.4, we examined the significant condition × time interaction for each of these significant clusters using a mixed effects analysis where condition and time were modeled as repeated factors with residuals modeled as unconditional for condition and autoregressive (AR[1]) for time. A priori simple contrasts (without > with reinforcement) of average activation in the delay period (8–16 s) were conducted on least square mean estimates. Parallel mixed effects analyses were conducted for the anatomically defined left and right amygdala clusters.

Hypothesis 3

To evaluate associations between amygdala activation during the delay period and performance (accuracy and reaction times) during negative distracters, two mixed effects analyses were conducted in SAS V9.4 with predicting performance from amygdala activation (8–16 s). We allowed for differences in the association to vary as a function of reinforcement condition by including an amygdala x condition interaction. Residuals were modeled as unconditional for condition.

Hypothesis 4

Mediational analyses were conducted using the MEMORE macro for SAS developed by Montoya and Hayes (2017) to determine whether changes in amygdala activation during negative distracter trials due to reinforcement would mediate changes in task performance. This macro was used to estimate the total, direct, and indirect effects of X (reinforcement: with reinforcement vs. without reinforcement) on Y (performance: change in reaction time, change in accuracy) through the mediator M (i.e., change in left/right amygdala activation due to reinforcement) in a path-analytic form using ordinary least squares regression. This macro implements the method described by Judd et al. (2001) to test mediation in within-subject designs. Inferences about the statistical significance of the indirect effect were based on confidence intervals generated using bootstrapping approaches (10 000 samples).

To correct for multiple comparisons for tests of Hypotheses 1 and 2, the significance level was corrected using FDR q < 0.05. FDR correction was calculated using SAS’s multitest procedure which uses the raw P-values from simple contrasts to determine the FDR using the linear step-up method of Benjamini and Hochberg (1995). To correct for the four mediations tests conducted in Hypothesis 4, all tests of indirect effects were conducted using a 97.6% confidence interval which corresponds to an FDR q < 0.05 (Benjamini and Yekutieli 2001; Narum 2006).

Delay Timing Sensitivity Analyses

Previous studies have varied in the window used to define the delay period (Dolcos and McCarthy 2006; Anticevic et al. 2010; Ladouceur et al. 2018). The inconsistency in the timing of the delay period calls into question, when during the delay period are emotional interference effects most likely to be observed. In previous studies examining emotional interference on cognitive control regions, the DLPFC and lateral frontopolar cortex (LFPC) demonstrate maximum deactivations between 4.4 s (Anticevic et al. 2010) and ~4.5 s (Dolcos and McCarthy 2006) postdistracter onset based on the graphs of the time-series in both papers. We conducted sensitivity analyses to determine if differential distracter impact on prefrontal cognitive control regions are more apparent early within the delay period (4-s postdistracter onset = 8-s postmemoranda onset), which should correspond to the peak window of difference among conditions, relative to later during the delay period (6–10 s postdistracter onset = 10–14 s postmemoranda onset). Given evidence that individual differences in sustained processing of emotional information in the amygdala exist, particularly among those with affective disorders (Siegle et al. 2002), we also examined whether changes in amygdala activation due to reinforcement during the time frame most indicative of sustained processing of the negative distracters prior to probe onset (10–14 s postmemoranda onset) would be a stronger mediator of changes in task performance than the entire delay and probe period (8–16 s postmemoranda onset).

Results

Distracter Ratings

Our manipulation check of distracter type indicated that participants rated negative distracters as more negatively valenced, t(48) = 14.12, P < 0.001, Cohen’s d = 2.38, and arousing t(48) = −9.02, P < 0.001, Cohen’s d = −1.66 than neutral distracters (valence: M (SD)_negative = 7.0 (1.1) vs. M (SD)_neutral = 4.6(0.9); arousal: M (SD)_negative = 5.1 (2.2) vs. M (SD)_neutral = 8.0 (1.1)).

Task Performance

Accuracy

Figure 2A shows mean percent accuracy for each distracter type with and without positive reinforcement. Results from the repeated measures ANOVA indicated significant main effects of reinforcement, F(1247) = 28.98, P < 0.001, and distracter F(2247) = 32.94, P < 0.001, which were not qualified by a reinforcement × distracter interaction F(2245) = 0.01, P = 0.987. Participants were more accurate on trials with reinforcement, M (SE) = 83.6 (1.27), relative to those without reinforcement, M (SE) = 80 (1.27), t = P < 0.001, Cohen’s d = 0.76. Furthermore, participants were less accurate in the face of negative distracters, M (SE) = 75.4 (1.38), relative to neutral distracters, M (SE) = 81.2 (1.38), t = 4.52, P < 0.001, Cohen’s d = 0.64, and were more accurate when no distracters were present, M (SE) = 85.8 (1.38), relative to neutral distracters, t = 3.58, P < 0.001, Cohen’s d = 0.51, and negative distracters, t = 8.10, P < 0.001, Cohen’s d = 1.14.

Figure 2.

Bar charts of (A) mean percent accuracy and (B) mean reaction times on the EDWM task depicting the main effects of distracter type and reinforcement. Bars with differing letters within an effect are significantly different from one another. W/R.: with reinforcement; W/O R.: without reinforcement.

Reaction Time

Figure 2B shows mean reaction times for each distracter type with and without positive reinforcement. Results from the repeated measures ANOVA indicated a significant main effect of reinforcement, F(1247) = 52.65, P < 0.001, on reaction time, and distracter F(2247) = 3.84, P = 0.032. Main effects were not qualified by a reinforcement × distracter interaction F(2245) = 0.18, P = 0.833. Relative to trials without reinforcement, M (SE) = 1456 (28.2), trials with reinforcement, M (SE) = 1330 (28.2), was associated with faster reaction times, t = 7.26 P < 0.001, Cohen’s d = 1.03. Furthermore, participants demonstrated longer reaction times in the face of negative distracters, M (SE) = 1408 (29.5) t = 2.24, P = 0.026, Cohen’s d = 0.32; and neutral distracters M (SE) = 1410 (29.5) t = 2.33, P = 0.021, Cohen’s d = 0.33, relative to when no distracters were present M (SE) = 1360(29.5). Equivalent reaction times were observed for negative distracters and neutral distracters, t = 0.09, P = 0.927, Cohen’s d = 0.01.

Neuroimaging

Examination of Brain Regions Responsive to Differences Between No Distracter, Neutral Distracter, and Negative Distracter Conditions Without Reinforcement

Table 1 and Supplemental Table S3 list the brain regions demonstrating significant distracter type (no distracter, neutral distracter and negative distracter) × time (12 TRs) interaction from the region-of-interest analysis, and Figure 3 shows their anatomical locations. Probing the interaction indicated that the right amygdala, left subgenual anterior cingulate cortex (sgACC) (BA 25), left vmPFC (BA 10), bilateral dmPFC (BA 9), left VLPFC (BA 45/47), and right VLPFC (BA 47) were more active during the delay period in the negative distracter condition relative to the no distracter condition (see Fig. 3, FDR < 0.05). Across all regions-of-interest, activation during the delay period, in the negative distracter condition, did not statistically differ from activation during the neutral condition when controlling for multiple comparisons. However, as shown in Supplemental Table S4, when a less conservative testing approach was taken—analyzing the time points of the delay period showing the highest statistical significance for the contrasts of interest as in Dolcos and McCarthy (2006)—activation in the negative condition differed from activation in the neutral condition for the bilateral amygdala, left sgACC, left VMPFC, and bilateral dmPFC. Other cognitive control regions in the dorsal and lateral prefrontal cortex did not differ between conditions during the delay period (FDR q > 0.05). However, sensitivity analyses indicated that, early during the early delay period (8-s postmemoranda onset), bilateral DLPFC and bilateral LFPC were significantly deactivated in the negative distracter versus no distracter trials. Moreover, activation in right LFPC showed significant differences between negative and neutral distracter conditions. During the later delay period (10–14 s postmemoranda onset) there were no significant distracter type-related differences (see Supplementary Table S5).

Table 1.

Brain regions demonstrating a distracter type × time interaction effect from the region-of-interest analysis

Region	BA	x	y	z	Cluster	Peak F-value	Simple contrasts between conditions for the delay period
							Neg > No	Neg > Neu	Neu > No
							Cohen’s d	Cohen’s d	Cohen’s d
Amyg (A)		−23	−5	−15	365	2.33	0.25	0.29*	—
Amyg (A)		23	−5	−15	358	2.68	0.43**	0.24	—
sgACC	25	−2	13	−9	17	5.52	0.44**	0.29*	—
vmPFC	10/11	2	48	−9	29	5.23	0.31*	0.11	—
vmPFC	11	−4	50	−11	72	6.05	0.38**	0.19	—
dmPFC	9/10	−8	60	34	1164	10.31	0.54**	0.28	—
VLPFC	45/47	−42	23	−15	527	6.16	0.41**	0.08	0.34**
VLPFC	47	38	32	−12	54	5.02	0.37**	0.02	0.38**
VLPFC	45	55	37	11	172	6.43	0.25	0.05	0.22
LFPC	10	32	59	10	263	5.4	−0.08	−0.18	—
LFPC	10	−36	51	7	70	4.42	0.03	−0.13	—
DLPFC	8/9	44	20	47	273	5.11	−0.17	−0.17	—
DLPFC	9	−42	25	38	268	6.97	0.09	−0.08	—

^* P < 0.05 uncorrected.

^** P < 0.05 FDR corrected.

Note: A = anatomically defined region; Neg = negative; Neut = neutral; No = no distracter; Amyg - amygdala; sgACC - subgenual anterior cingulate cortex; vmPFC - ventromedial prefrontal cortex; dmPFC - dorsomedial prefrontal cortex; VLPFC - ventrolateral prefrontal cortex; LFPC - lateral frontopolar cortex; DLPFC - dorsolateral prefrontal cortex; BA - Brodmann area.

Figure 3.

Task-evoked time courses of anatomically extracted amygdala and PFC regions on correct trials. Event-related time courses for prefrontal regions are shown based on the distracter type (no distracter, neutral distracter and negative distracter) × time (12 TRs) interaction region-of-interest analysis. Activation maps highlight significant interactions (voxel-wise threshold of P < 1 × 10⁻⁷, and an extent threshold of 15 voxels) overlaid on the skull stripped Colin brain (ch2better.nii). Plots show changes in mean parameter estimates for average activation for the baseline/no distracter (black), neutral distracter (blue), and negative distracter (red) conditions. Bar graphs reflect mean parameter estimates during the delay period (8–16 s postmemoranda onset) for each distracter type. Bars with different letters differ at FDR q < 0.05. Bars with a †differ at P < 0.05 uncorrected for multiple comparisons. The gray rectangular boxes above the x-axis indicate the onset and duration of the memoranda, distracters, postdistracter fixation, and probe. dmPFC - dorsomedial prefrontal cortex; sgACC - subgenual anterior cingulate cortex; vmPFC - ventromedial prefrontal cortex; LFPC - lateral frontopolar cortex; VLPFC - ventrolateral prefrontal cortex; DLPFC - dorsolateral prefrontal cortex; BA - Brodmann area; a.u. - arbitrary units; L. - left; R. - Right.

Examination of Brain Regions Responsive to Positive Reinforcement During the Presentation of Negative Distracters

Table 2 lists the brain regions demonstrating significant reinforcement (negative distracters without reinforcement, negative distracters with reinforcement) × time (12 TRs) interaction during negative distracters, and Figure 4 shows their anatomical locations. Probing brain regions demonstrating a significant reinforcement × time interaction indicated that the right amygdala, right dmPFC, left dmPFC, left VLPFC (BA 45), and left VLPFC (BA 47) were less active during trials with reinforcement relative to trials without reinforcement (see Fig. 4, FDR < 0.05). DLPFC activation did not differ with reinforcement relative to trials without reinforcement during the delay period (FDR > 0.05). However, sensitivity analyses did indicate that early in the delay period (8-s postmemoranda onset), the DLPFC was significantly deactivated during trials without reinforcement, relative to trials with reinforcement (i.e., relatively greater activation with reinforcement relative to without reinforcement). During the later delay period (10–14 s postmemoranda onset), there were no significant differences during trials with reinforcement versus trials without reinforcement (P > 0.10). Secondary analyses indicated similar differences between reinforcement conditions for the neutral distracters with the exception of the amygdala and relatively few differences for no distracters (see Supplemental Fig. S1).

Table 2.

Comparisons of BOLD signal during correct trials of the EDWM task: effect of reinforcement × time interaction for negative distracter trials

Region	BA	x	y	z	Cluster	Maximum F-value	Simple contrasts between conditions for the delay period W/R. < W/O R.
Amygdala (A)		−23	−5	−15	365	1.26	0.28*
Amygdala (A)		23	−5	−15	358	1.39	0.41**
dmPFC	9	−2	52	31	373	5.65	0.70**
dmPFC	10	2	53	14	114	4.42	0.51**
VLPFC	47	−46	23	−13	209	5.37	0.49**
VLPFC	45	−59	22	16	110	4.46	0.58**
DLPFC	8	−44	27	43	142	4.5	0.13NS

^* P < 0.05 uncorrected.

^** P < 0.05 FDR corrected.

Note: A = anatomically defined region; W/R. = with reinforcement; W/O R. = without reinforcement; L = left; R = right. dmPFC - dorsomedial prefrontal cortex; VLPFC - ventrolateral prefrontal cortex; DLPFC - dorsolateral prefrontal cortex; BA - Broadman Area.

Figure 4.

Task-evoked time courses of anatomically extracted amygdala and PFC regions on correct trials. Event-related time courses for prefrontal regions are shown based on the negative distracters condition (with and without positive reinforcement) by time (12 TRs) interaction region-of-interest analysis. Activation maps highlight significant interactions (voxel-wise threshold of P < 1 × 10⁻³, and an extent threshold of 75 voxels) overlaid on the skull stripped Colin brain (ch2better.nii). Plots show changes in mean parameter estimates for average activation for the with reinforcement (green) and without reinforcement (red) conditions. Bar graphs reflect mean parameter estimates during the delay period (8—16 s postmemoranda onset) for each reinforcement condition. Bars with different letters differ at FDR q < 0.05. Bars with a †differ at P < 0.05 uncorrected for multiple comparisons. The gray rectangular boxes above the x-axis indicate the onset and duration of the memoranda, distracters, postdistracter fixation, probe, and feedback. dmPFC - dorsomedial prefrontal cortex; DLPFC - dorsolateral prefrontal cortex; VLPFC - ventrolateral prefrontal cortex; BA - Brodmann area; a.u. = arbitrary units; L - left; R - right; W/R.: with reinforcement; W/O R.: without reinforcement.

Brain-Behavior Associations

Examination of Accuracy/Amygdala Associations During Negative Distracters

As show in Figure 5, increased right amygdala activation during correct trials predicted worse behavioral accuracy across reinforcement conditions, β = −0.22, SE = 0.10, F(1,90) = 5.05, P = 0.027, (reinforcement condition, F(1,51) = 4.51, P = 0.039; reinforcement condition × right amygdala interaction, P > 0.10). The left amygdala was not associated with behavioral accuracy, β = −0.11, SE = 0.10, F(1,85) = 1.35, P = 0.248 (reinforcement condition, F(1,49) = 6.48, P = 0.014; reinforcement condition × left amygdala interaction P > 0.10).

Figure 5.

Individual differences in WM performance as a function of average amygdala activation 8–16 s postmemoranda onset. (A) The main effect of right amygdala activation on WM accuracy (percent correct) across reinforcement conditions. (B) The simple slopes depicting the nature of the significant amygdala × reinforcement condition interaction on WM reaction times. W/R. - with reinforcement; W/O R. - without reinforcement.

Examination of Reaction Time/Amygdala Associations During Negative Distracters

Activation in the right amygdala to correct trials was not associated with reaction time, β = 0.06, SE = 0.08, F(1,76) = 0.48, P = 0.492 (reinforcement condition, F(1,51) = 31.64, P < 0.001; reinforcement condition × right amygdala interaction P > 0.10). The association between left amygdala activation and reaction time varied as a function of condition, F(1,62) = 6.81, P = 0.011. As shown in Figure 5, increased left amygdala activation was associated with slower reaction times in the positive reinforcement condition, β = 0.33, SE = 0.09, t(48) = 3.41, P = 0.001, whereas left amygdala activation was not associated with reaction times in the without reinforcement condition, β = −0.02, SE = 0.09, t(48) = −0.22, P = 0.827 (condition F(1,48) = 32.22, P < 0.001; left amygdala F(1,63) = 5.01, P = 0.029).

Examination to Determine If Changes in Right Amygdala Activation Resulting from Positive Reinforcement Mediate Changes in Accuracy Resulting From Positive Reinforcement

Controlling for the effect of reinforcement and average right amygdala activation, decreases in right amygdala activation as a function of reinforcement were not associated with improvements in accuracy occurring due to reinforcement (see Table 3). Consistently, tests of the indirect effect indicated that changes in right amygdala activation as a function of reinforcement did not mediate changes in behavioral accuracy due to reinforcement (indirect effect = −0.13, Monte Carlo Standard Error (MCSE) = 0.83, 97.6% Monte Carlo Confidence Interval (MCCI) [−2.18, 1.91]). As shown in Table 3, time sensitivity analyses did not indicate that sustained right amygdala activation (10–14 s postmemoranda onset) was a stronger mediator than right amygdala activation 8–16 s postmemoranda onset (indirect effect = −0.73, MCSE = 1.34, 97.6% MCCI [−4.05, 2.30]).

Table 3.

Change in amygdala activation due to reinforcement as a mediator of performance for different windows within the delay period

	DV	Predictor(s)	Delay period (8–16 s)		Delay period (10–14 s)
			Β	97.6% CI	Β	97.6% CI
Eq. 1	Accuracy	Condition	−5.47*	(−9.76, −1.17)	−5.47*	(−9.76, −1.17)
Eq. 2	Right Amyg	Condition	0.03*	(0.00, 0.05)	0.04*	(0.02, 0.06)
Eq. 3	Accuracy	Condition	−5.34*	(−10.09, −0.58)	−4.73*	(−10.08, 0.60)
		Δ Right Amyg	−5.23	(−78.13, 67.67)	−17.99	(−93.77, 58.00)
		x̅ Right Amyg	7.74	(−98.38, 113.86)	−7.78	(−107.70, 92.14)
Eq. 3 R2			0.04		0.01
Eq. 3 F			0.03		0.18
Eq. 1	Reaction Time	Condition	140*	(88.3, 192.0)	140*	(88.3, 192.0)
Eq. 2	L. Amyg	Condition	0.02*	(0.00, 0.05)	0.04*	(0.01, 0.07)
Eq. 3	Reaction Time	Condition	129*	(76.5, 181.4)	117*	(63.6, 171.3)
		Δ Left Amyg	496*	(−179.7, 1172.6)	628*	(50.0, 1206.0)
		x̅ Left Amyg	−1037	(−2344.2, 270.7)	−479	(−1623.6, 665.9)
Eq. 3 R2			0.10		0.12
Eq. 3 F			2.70*		3.30*

Note: ^*P < 0.050. Amyg = Amygdala. Eq. = equation. Condition = without reinforcement—with reinforcement.

Examination to Determine If Changes in Right Amygdala Activation Resulting from Positive Reinforcement Mediate Changes in Accuracy Resulting from Positive Reinforcement

Controlling for the effect of reinforcement and average right amygdala activation, decreases in right amygdala activation as a function of reinforcement were not associated with improvements in accuracy occurring due to reinforcement (see Table 3). Consistently, tests of the indirect effect indicated that changes in right amygdala activation as a function of reinforcement did not mediate changes in behavioral accuracy due to reinforcement (indirect effect = −0.13, MCSE = 0.83, 97.6% MCCI [−2.18, 1.91]). As shown in Table 3, time sensitivity analyses did not indicate that sustained right amygdala activation (10–14 s postmemoranda onset) was a stronger mediator than right amygdala activation 8–16 s postmemoranda onset (indirect effect = −0.73, MCSE = 1.34, 97.6% MCCI [−4.05, 2.30]).

Examination to Determine If Changes in Left Amygdala Activation Resulting from Positive Reinforcement Mediate Changes in Reaction Times Resulting from Positive Reinforcement

Controlling for the effect of reinforcement condition and average left amygdala activation, participants who exhibited a decrease in left amygdala activation as a function of reinforcement demonstrated marginally faster reaction times in the with reinforcement compared with without reinforcement conditions (see Table 3 and Fig. 6A). However, tests of the indirect effects indicated that changes in left amygdala activation as a function of reinforcement did not mediate changes in reaction times due to reinforcement (indirect effect = 11.19, MCSE = 8.84, 97.6% MCCI [−3.92, 36.11]). As shown in Table 3, timing sensitivity analyses indicated that change in amygdala activation due to reinforcement in the 10–14 s window mediated change in reaction time due to reinforcement (indirect effect = 22.69, MCSE = 12.49, 97.6% MCCI [0.77, 56.38]). As shown in Figure 6B, participants who experienced a decrease in left amygdala activation as a function of reinforcement demonstrated faster reaction times in the with reinforcement compared with without reinforcement conditions.

Figure 6.

Within participant mediation model testing whether changes in left amygdala activation is a mediator of positive reinforcement induced changes in WM reaction times. (A) Changes in average amygdala activation 8–16 s postmemoranda onset as the potential mediator. (B) Changes in sustained amygdala (10–14 s postmemoranda onset) as the potential mediator. c, total effect; c′, direct effect; IE, indirect effect. W/R. - with reinforcement; W/O R. - without reinforcement.

Discussion

The goals of the current study were to 1) determine whether patterns of fronto-limbic activation occurring in the context of resisting emotional interference seen in adults are present in young adolescents and 2) test the extent to which providing positive reinforcement would modulate fronto-limbic systems to reduce emotional interference and associated declines in goal-directed behavior. Main findings from the first goal suggest that typically developing youth exhibit largely similar patterns of behavioral performance, neural activity, and brain-behavior associations as those documented in adults during emotional interference. In particular, we found that: 1) WM was more impaired by emotional distracters relative to neutral and no distracters; 2) emotional distracters were associated with increased activation in the amygdala, and VLPFC, and decreased DLPFC activation relative to no distracters; and 3) individual differences in the degree of WM impairment due to emotional distracters were associated with increased amygdala activation. Findings from our second goal suggest that positive reinforcement can improve WM performance and can modulate fronto-limbic systems involved in emotional interference. Specifically, we found that: 1) positive reinforcement for correct responding was associated with both higher accuracy and faster reaction times across all distracter types; 2) positive reinforcement was associated with decreased activation in the right amygdala, left VLPFC and bilateral dmPFC, and increased DLPFC activation (8-s postmemoranda onset) relative to non-reinforced responding during the presentation of negative distracters; 3) amygdala activation was associated with lower WM accuracy (right amygdala) across reinforcement conditions, and slower reaction times for trials with reinforcement (left amygdala); and 4) decreases in sustained left amygdala activation (10–14 s postmemoranda onset) mediated observed decreases in reaction times. These findings are discussed in further detail below.

Neural Networks Underlying Emotional Interference

Our results are supportive of our primary goal to replicate and further clarify the neural networks underlying resistance to emotional interference in adolescents. Consistent with the extant literature in adults and a recent study in adolescence, we observed that emotional interference is associated with increased activation in the ventral visual stream (see Supplemental Table S1 and Fig. S2), and the amygdala (Iordan et al. 2013; Vetter et al. 2015; Hung et al. 2018; Schweizer et al. 2019). In this context, increased ventral visual stream and amygdala activation to negative distracters likely reflects the bottom-up detection of highly salient emotional information. In addition, we also observed increased sgACC activation to negative distracters, which might reflect the conscious processing of negative emotional stimuli (Laxton et al. 2013; Huebl et al. 2016). The motivational salience of the IAPS images was likely determined by the dmPFC, whose activity has been shown to support the representation, evaluation, and extraction of emotional information from affective stimuli (Taylor et al. 2003; Kober et al. 2008; Peelen et al. 2010; Etkin et al. 2011; Skerry and Saxe 2014). Based on the graduated pattern of activation in the dmPFC, it can be inferred that negative distracters were coded as more motivationally salient than neutral or no distracters, and hence, are likely to cause interference. Consistent with this interpretation, negative distracters were associated with greater impairment in WM accuracy than neutral and no distracters and slowed reaction times more than the no distracter condition. Furthermore, individuals with greater right amygdala activation in response to negative distracters demonstrated lower WM accuracy. Thus, individuals for whom negative distracters were affectively salient were more susceptible to emotional interference.

We observed evidence that the aforementioned bottom-up processes were concurrently associated with the modulation of brain regions supporting the top-down management of competing goals, specifically in the LFPC and DLPFC. A recently articulated framework of frontopolar cortex function (Mansouri et al. 2017) posits that the medial frontopolar cortex tracks the value of the stimulus, which enables the ability to disengage cognitive control from the current task and to redistribute resources to processing other goals of greater value. The graduated activation pattern in the vmPFC following the onset of negative distracters (i.e., alternate goals) suggests that negative distracters were being monitored as motivationally salient events that warranted exploration (Dolcos et al. 2004). According to the theory, these goals would have competed for primacy in WM, where the LFPC is hypothesized to enable the monitoring of several competing goals in order to facilitate the re-engagement of one as a replacement of the current goal if indicated, whereas the DLPFC is thought to recruit and implement cognitive control to optimize the performance of the current goal (Mansouri et al. 2017). As such, decreased LFPC and DLFC activation may reflect the “hijacking” of attentional resources by motivationally salient emotional stimulus (i.e., a competing goal). Interestingly, the time-course of activation indicates that the deactivation observed at 8-s postmemoranda onset was followed by a subsequent BOLD response, potentially reflecting processing of the distracter in WM, which was followed by another BOLD response corresponding to the onset of the probe (see Fig. 3). This pattern of deactivation then reactivation is consistent with the neurocomputational model of LFPC neuronal functioning when cognitive branching, i.e., perform tasks related to one goal, while keeping in WM information related to a secondary goal that needs to completed (Koechlin and Hyafil 2007). During the branching processes, it is possible that memoranda could be lost resulting in decreased WM accuracy (Anticevic et al. 2010). These findings replicate previous results in both adults (Anticevic et al. 2010; Iordan et al. 2013; Vetter et al. 2015; Hung et al. 2018; Schweizer et al. 2019) and adolescents (Vetter et al. 2015).

We also observed increased activation in the VLPFC to distracters (neutral and negative) relative to no distracter trials, which is consistent with this region’s hypothesized role in the top-down inhibition of emotional interference as indicated by its negative associations with subjective distractibility and emotionality of negative affective stimuli (Dolcos and McCarthy 2006; Iordan et al. 2013). Consistent with this interpretation, reaction times were slower and WM accuracy was lower when neutral distracters and negative distracters were presented versus no distracters (Anticevic et al. 2010; Ladouceur et al. 2018).

Thus, overall our findings are consistent with the emerging literature on emotional interference in adolescents which indicate that adults and young adolescents activate comparable networks underlying resistance to emotional interference. Future work is needed to determine the temporal sequencing of activation within identified regions to further elucidate the functional network underlying emotional interference.

The Impact of Positive Reinforcement on Emotional Interference

Our results also provided strong support for a second goal to evaluate the extent to which providing positive reinforcement would modulate fronto-limbic systems to reduce emotional interference and associated declines in goal-directed behavior. Our behavioral results indicated that positive reinforcement increased accuracy and decreased reaction times across all distracter types, which indicates that it nonspecifically reduced interference from all distracters. These results are consistent with neuropsychological studies showing that WM performance is faster and more accurate when it is rewarded than when positive reinforcement is absent or low. At the neural level, we observed that positive reinforcement of correct responding resulted in reduced activation in the right amygdala, left VLPFC, and bilateral dmPFC. As previously described, these aforementioned regions were specifically elevated during processing of emotional relative to no distracter trials. Decreases in these regions are consistent with decreased processing of the negative emotional stimuli such that they were no longer salient and did not require the recruitment of inhibitory control, or appraisal. These findings are consistent with our preliminary study (Ladouceur et al. 2018), but contrary to the results of a recent meta-analytic study (Parro et al. 2017), which observed increased activation in the fronto-parietal cognitive control network regions under conditions of positive reinforcement. Of note, most of the previous studies examined the impact of reward on cognitive control in the absence of negative emotional images and most were conducted in adults. Few studies have examined the impact of reward on negative distracter processing in youth (Kaltwasser et al. 2013; Wei and Kang 2014; Padmala et al. 2017; Ladouceur et al. 2018). The only other neuroimaging study to date reported decreased activation in the lateral occipital cortex, anterior insula, and dorsal anterior cingulate during cued reward trials, which was interpreted to reflect that reward decreased the processing of negative emotional distracter images (Padmala et al. 2017). However, in Padmala et al. (2017), amygdala functioning was not modulated by reward. In contrast, in the current investigation, not only was amygdala activation modulated by reward, but also the magnitude of the decrease in left amygdala activation observed as a function of reinforcement mediated the observed speeding up of decision response times to the WM probe. These findings are consistent with data demonstrating an attenuation of amygdala activation and improved WM accuracy when attention is directed toward nonemotional details of negative internal distracters (i.e., negative memories) relative to the emotional aspects of negative memories in the context of an EDWM task (Iordan et al. 2019). Similarly, attenuation of the amygdala response to negative images has also been observed when visual attention is consumed elsewhere, even when the control of visual attention does not require large shifts in the visual field (Pessoa et al. 2002). Thus, positive reinforcement appears to have altered the impact of negative emotional distracters through attentional filtering (Vuilleumier 2005; Iordan et al. 2019). Consistent with this interpretation, we observed that positive reinforcement was associated with decreased activation in the lateral occipital cortex during the delay period (see Supplementary Fig. S3). However, we did not observe increased activation in fronto-parietal regions implicated in top-down attentional modulation under reinforcement; this may have been due to type II error resulting from conservative statistical thresholding. Alternatively, attending to a nonemotional portion of the negative images may have required the same level of attentional control as processing the negative images resulting in no differences in fronto-parietal activation as a function of condition. Unfortunately, we did not collect eye-tracking during the task, which would have clarified where participants were focusing their attention during the positive reinforcement condition. It is also possible that increased dopamine resulting from reinforcement resulted in enhanced spatial tuning for the memoranda and the suppression of processing negative images (Vijayraghavan et al. 2007). However, this cannot be verified given that we did not alter or directly assess dopaminergic functioning.

Of note, unlike previous investigations (e.g., Padmala and Pessoa 2011; Padmala et al. 2017) examining the impact of reinforcement on cognitive control, we did not observe increases in the ventral striatum or other reward-related regions during the task. This may have occurred because of differences in task design. Specifically, participants were told prior to a block of trials whether or not the upcoming block was a reward block or a nonreward block. The blocked design is likely to instantiate the slow tonic release of dopamine likely to facilitate sustained attention processing (Schultz 2016). In contrast, studies that have employed a cue to indicate reward versus nonreward trials within a block and where there is increased uncertainty whether or not the reward would be received (i.e., greater uncertainty about whether or not the participant is likely to get the trial correct) more reliably demonstrate activation within the ventral striatum (e.g., Padmala and Pessoa 2011; Padmala et al. 2017). These types of designs with trial-specific cues may more strongly instantiate the phasic release of dopamine resulting in a better ability to detect BOLD activation within dopamine producing brain regions using fMRI (Schultz 2016). Importantly, the only other identified study employing this blocked approach only detected changes in reward-related regions when individual differences in sensitivity to reward were examined (Locke and Braver 2008). Future studies are needed to examine the impact of task design differences and individual differences in sensitivity to reward on cognitive control in the context of preventing emotional interference (cf., Beck et al. 2010).

We also observed differential associations between WM performance and activation in the left versus right amygdala. There is no settled understanding of hemispheric specialization within the amygdala. Some evidence suggests that the left amygdala is more involved in processing threat, emotional arousal, salience, and sustained emotional processing (Phelps et al. 2001; Baas et al. 2004; Costanzo et al. 2015). Whereas the right amygdala has been implicated in processing animal stimuli, emotional ambiguity and unconscious emotional information (Baas et al. 2004;Mormann et al. 2011 ; Wang et al. 2017). It is possible that left amygdala activation was associated with slower reaction times because anxiety is more likely to interfere with processing efficiency (i.e., reaction times) rather than accuracy (Eysenck et al. 2007). In contrast, greater right amygdala activation may indicate that individuals who were more effected by the negative IAPS images pertaining to animals experienced greater emotional interference causing participants to respond incorrectly (Mormann et al. 2011; Wang et al. 2017). These suppositions should be taken with caution, especially in light of recent research demonstrating that the choice of motion correction and smoothing parameters can also have a strong impact of whether bilateral, or hemispheric specific activation is found (Murphy et al., 2019).

Limitations

The findings of the current study should be interpreted in light of the following limitations. First, there was a significant loss of fMRI data (~50%) due to excessive movement and poor behavioral accuracy. This data loss likely occurred due to the age of participants combined with relatively long duration of the task (30 min). This attrition may limit the generalizability of the findings to youth with the capacity to stay still for longer periods of time. Second, we were unable to include sufficient trials to be able to model both correct and incorrect responding given the constraints of time, thus limiting our ability to examine task differences associated with performance. Third, we did not collect measures related to the subjective distractibility of the distracters, limiting our ability to draw strong inferences related to VLPFC functioning. Finally, findings are limited to the effects of negative distracters and to early adolescence. Follow-up assessments are under way and will provide clues as to whether these findings can be generalized to older adolescents.

Conclusions

In summary, findings from this study provide evidence indicating that the patterns of neural activation in fronto-limbic regions associated with emotional interference obtained with adults are also present in young adolescents. This study also demonstrates that positive reinforcement (i.e., monetary rewards for correct responses) can be used to modulate fronto-limbic systems associated with resisting emotional distraction. Specifically, we show that positive reinforcement generally improves task performance and reduces activation in limbic, medial, and ventrolateral PFC regions in a group of healthy youth. Importantly, findings demonstrate, for the first time, that the modulation of amygdala activation by reinforcement mediates changes in behavior. Future research examining the effective connectivity of key regions (e.g., the amygdala, VLPFC, vmPFC and lateral occipital cortex) within the network and testing interactions with age, which have been shown to be associated with changes in attentional control (Luna et al. 2010) and reward processing (de Macks et al. 2011), respectively, will provide further knowledge about the possible underlying mechanisms. It will also be possible to examine to extent to which age-related changes in functional connectivity between the ventral striatum and VLPFC in trials with reinforcement (Davidow et al. 2018) may underlie the impact of reinforcement on emotional interference. Such research will determine the extent to which it may be possible to harness adolescent neural response to the effects of positive reinforcement on PFC function to counteract negative environmental influences and facilitate coping.

Supplementary Material

Supplementary material can be found at Cerebral Cortex online.

Funding

National Institute of Mental Health (R01MH099007; PI: Ladouceur); National Institutes of Health (UL1TR001857).

Notes

The authors would like to thank Drs. Anticevic and Barch for allowing them to adapt the EDWM task for adolescents and for their feedback on the design of the study. They also thank the Pitt Clinical and Translational Science Institute (CTSI) for their help with recruitment as well as the children and their families for participating in this research study. Conflict of Interest: None declared.