Prefrontal-posterior coupling mediates emotion transitions and their influence on executive function

Although emotions often result from dynamic experiences with self-regulation unfolding over time, most research has focused on responses to affective stimuli from a rather static perspective. We studied and analyzed emotion transitions, attempting to reveal brain functions related to affect dynamics. EEG responses were examined during exposure to stable versus changing emotion-eliciting images (static vs dynamic conditions) plus their impact on executive function (EF) assessed with the flanker task. During dynamic conditions, reduced prefrontal to posterior EEG coherence in the beta frequency band and greater left frontal activity occurred compared to the static conditions. Among individuals suffering higher chronic stress, subsequent EF was hindered after dynamic conditions. Furthermore, the adverse effects of emotion transitions on EF for more chronically stressed individuals were mediated by prefrontal-posterior coherence in low beta frequency band during emotional image sequences. Emotion appears to influence EF through changes in large-scale synchronization. Individuals high in chronic stress are vulnerable to these effects.


Introduction
Emotional well-being is a key component of maintaining equanimity against a backdrop of variable experiences. Affect dynamics reflect self-regulation ability and are crucial for psychological health [1][2][3] . Changing environmental conditions trigger affect dynamics, and studying the brain and behavior consequences provide insights about underlying mechanisms of cognition-emotion as they unfold over time [4][5][6][7] . In our experimental paradigm, we define emotion transition as responses to dynamic conditions. As opposed to static conditions without affect transition, dynamic conditions are defined as exposure to stimuli of varying affective valence or intensity, e.g., image sequences with changing valence and/or intensity.
Cortical synchronization is related to a greater control of the affective state 8 . Specifically, emotionally arousing/threatening stimuli and the modulation effect of executive control have been shown to influence EEG coherence between prefrontal and posterior cortical regions indexing changes in functional communication 9,10 . A reduction in the prefrontalposterior coupling may reflect a loosening of top down control from the prefrontal cortex leading to greater emotional contagion 9,11-15. Executive function (EF) may also be recruited when conflicts are detected among different emotional affect between a current stimulus and reverberation from a previous emotional stimulus. Thus, emotional transitions may decrease prefrontal-posterior coherence.
The present study investigates neural responses to dynamic emotional conditions and their influence on subsequent EF as compared to static conditions. We also examine the influence of chronic stress on these outcomes. EF, emotion, and stress interact in an adaptive feedback loop in response to environmental cues 16 . Individuals under chronic stress exhibit automatic responses that are more reactive than reflective, particularly in an unpredictable environment 17,18 , such as when presented with dynamic stimuli. Dynamic emotional stimuli can have cumulative effects which may be influenced by chronic stress. Although enhanced sensitivity to novel stimuli helps maintain response flexibility to new threats, it may add to the accumulative impact of glucocorticoids on the body potentially leading to mental illness 19 . In addition, both acute and chronic stress can affect brain structures or functions 20 , such as it influences prefrontal hemispheric imbalances in neuroendocrine regulation 21 . Moreover, besides the deficits in regulatory processes of emotional salient events, the conflicts generated by affect transition recruit more cognitive control mechanisms in the prefrontal cortex 22 . Processing dynamic emotional events likely requires more cognitive resources to regulate emotions and under chronic stress, these resources might be compromised. Therefore, dynamic stimulation might exhibit different aftereffects than purely negative events on EF.
In addition, lateralized activation of the prefrontal cortex is assessed because it has been linked to emotion-regulatory capacity 23 . Our work links brain connectivity with responses to emotional transition events and executive functioning, and examines the moderating role of chronic stress in this relation. Thus, we hypothesized decreased coherence in dynamic conditions which, in turn, would mediate the relation between emotional stimulation and subsequent EF performance. Furthermore, we expected this link to be stronger among individuals experiencing elevated chronic stress because of inflexibility and/or diminished capacity to adapt to changes in the emotionally affective environment.

Coherence in Static vs Dynamic Conditions
We examined beta frequency band [14 30] Hz, low [14 20] Hz and high [20][21][22][23][24][25][26][27][28][29][30] Hz beta frequency band coherence initially based on prior work, significant effects of condition on coherence were observed in both hemispheres. Because low beta frequency band ( [14,20] Hz) coherence had consistently strong effects and mediated the subsequent EF, we only report results from low beta frequency band coherence (Figure 1 (a-b). In the right hemisphere,   There were no main or interactive effects of chronic stress on coherence and frontal alpha asymmetry. Self-rated chronic stress (M = 15.91, SD = 6.33) was unrelated to gender, age, flanker performance or frontal-posterior coherence (with all tested frequency bands during image sequence and flanker task) and alpha asymmetry across conditions.
The coherence during image sequence was highly correlated to coherence during the flanker task, r = 0.90, t (130) = 24.16, p < .001. Furthermore, similar trend as the influence of coherence in image sequence on flanker RTs, the influence of coherence during flanker task on RTs was dependent on chronic stress in the right hemisphere, F (1, 68) = 4.89, p = .030, PCVbetween = 16.9%. Under high chronic stress, lower coherence during flanker task led to worse flanker performance; whereas higher coherence facilitated flanker performance, as depicted in Figure 2 (b). The effect of static vs. dynamic conditions on flanker performance was mediated by coherence during image sequences, only in participants suffering high chronic stress (above the mean) ( Figure 3). In the right hemisphere, the indirect path ab = 7.91, standard error = 3.20, p = .013; in the left hemisphere, the indirect path ab = 5.01, standard error = 2.48, p = .044. The dynamic conditions evoked lower coherence that resulted in slower RTs among those with high chronic stress. Among those emotion or attention related EEG (i.e., midline theta band power, theta/beta ratio, high beta/low beta ratio, and alpha asymmetry), only anterior-posterior low beta coherence during emotional conditions appeared to affect RTs.

Discussion
Prefrontal-posterior EEG coherence in beta frequency band coherence is lower during emotion transition regardless of affect (neutral or negative). Negative stimuli influence the beta frequency band coherence 10 , but this is the first demonstration that beta frequency band coherence between prefrontal and posterior cortical regions is related to dynamic changes in emotion-emotion transition. The coherence in low beta ( [14][15][16][17][18][19][20] Hz) frequency band also mediated emotional stimuli on subsequent EF among chronically stressed individuals.
In our experimental paradigm, we consider the interaction of changes in emotional stimuli and neural responses thereto as potential correlates of self-regulation. Decreases in beta frequency band prefrontal-posterior EEG coherence induced by emotional stimulation reflect a loosening of the prefrontal cortex's regulatory control of parietal regions 9,11-15 . Therefore, beta coherence plays a critical role in brain networks responding to emotion-cognition interaction. Image sequences with emotion transitions invoke more regulation effort as left frontal activity increases in comparison to responses to static neutral or negative images 23 .
Our study calls attention to the mediating role of brain synchronization in the relationship between the affective dynamism of stimulation and EF. In the present study, EF in chronically stressed individuals is hindered when preceded by reduced prefrontal-posterior neural uncoupling under emotional transitions. Residual emotion from preceding images might generate emotional conflicts, impelling more executive control mechanisms for self-regulation to manage conflicting emotional stimuli. To having greater conflict, these individuals might be less able to manage conflicts 24 . This also appears to diminish at least short term EF as indexed herein by the flanker task 25 . With the reduced coherence evoked by emotion transitions continued during the flanker task, chronically stressed individuals were more vulnerable and showed deficits in inhibitory ability during flanker task. This is salient among persons suffering greater chronic stress perhaps because of already constrained self-regulation capacity.
The primary limitations of the current study are that the duration of the induced drop in EF performance is unknown because only 6 trials of flanker task manifested aftereffect. We also do not know if other negative emotional sequences would create similar results. A valuable extension of this study would be to examine persons with self-regulatory difficulties, for instance, individuals with anxiety disorders. Reversely, manipulating self-regulation capability by EF training 25,26 or cognitive reappraisal 27 under emotion transition and its research on psychopathology are needed. Moreover, the investigation of emotion transition can be extended into other forms of dynamics, such as including more valence changes, or multiple transitions in affect stimulations. Limitations of the method (magnitude-squared coherence) include possible volume conduction artifacts. However, the relative difference of coherence among different experimental conditions is of our primary interest. In our study, no significant changes occurred in delta or theta EEG coherence comparing dynamic vs static condition during affective viewing. This also argues against general volume conduction confounds.
Emotion transitions may have more significant effects on coupling in neuronal regions and EF measures than static events without transitional affect. Brain responses to dynamic stimuli may reflect emotional regulation capacity specifically and coping skills more generally.

Participants
Participants were 40 right-handed students from Cornell University. Informed consent was obtained, and participation was compensated with course credits or $20 cash. Due to incomplete data, 33 participants were included in the final sample (51.5% females, age M = 22.40, SD = 3.80). Exclusion criteria were any open or healing wounds on the scalp, use of any medication that could affect nervous system processing and any history of neurological disorders. Participants were requested to refrain from alcohol caffeine, and other stimulants for four hours prior to the experiment. They were also asked to sleep for at least six hours the night before the experiment. The study was approved by Cornell's Institutional Review Board and was performed in accordance with the its guidelines. Informed consent was obtained from all participants.

Procedure
To measure chronic stress, we administered the Perceived Stress Scale (PSS) before EEG recordings 28 . PSS was designed to measure perceived stress level, i.e., unpredictable, uncontrollable, and overloaded. This questionnaire contains 10 items and requires respondents to rate the degree to which situations are appraised as stressful in the last month, such as: "In the last month, how often have you felt confident about your ability to handle your personal problems?" or "In the last month, how often have you felt difficulties were piling up so high that you could not overcome them?". PSS is a reliable and valid instrument for assessment of perceived stress in college students and workers 29,30 . Four conditions of image sequences were randomly presented with no image repetition to participants. For dynamic conditions, the images either transitioned from neutral to negative (dynamic-increase) or from negative to neutral (dynamic-decrease), which appeared 18 times each. For static conditions, the sequences consisted of all neutral (static-neutral) or all negative (static-negative) images, which appeared 6 times each. Each sequence involved four images with each image 4500 ms duration. First two images and last two images were from the same affect category, therefore, the transition occurred at the third image. More dynamic image sequences ensured the experiment conditions were in a dynamic setting. To counteract the balance of the amount of image sequence between static and dynamic conditions, we selected 6 sequences in dynamic conditions that were next to the static conditions. These were distributed across the experiment.
Immediately after each sequence, 12 trials of the flanker task 32 each 1200 ms (stimulus 200 ms and response window: 900, 1000 or 1100 ms) were presented to assess EF performance 33 . In the flanker task, a visual array of flag stimuli with the middle flag randomly surrounded by flags in the same direction (congruent) or in the opposite direction (incongruent). Participants responded to the direction of the middle flag as quickly and accurately as possible.

EEG Processing
EEG was recorded from a 128-channel BioSemi EEG device with digital sampling rate at 512 Hz. Coherence measures the degree of covariance between two spatially distinct signals in prefrontal and posterior regions. We applied the magnitude-squared coherence, which was calculated by the cross-spectrum divided by the product of the auto-spectrum of the two signals. It includes information of the amplitude and phase. The phase difference between the two signals plays an important role in estimating brain connectivity 34 . Following the methods described in Miskovic and Schmidt 10 and Allen et al. 35 , the coherence in beta frequency band [14 30] Hz, low beta frequency band [14 20] Hz and high beta frequency band [20][21][22][23][24][25][26][27][28][29][30] Hz and frontal alpha asymmetry at the dorsolateral region were calculated throughout the image sequence (18000 ms) for each condition. For the full description of the EEG processing, see supplement. We followed up these hypothesized effects with exploratory analyses of other EEG metrics that are related to emotion and attention.

Statistical analysis
We applied linear mixed effect models to predict brain activities by image sequence condition with each hemisphere and each frequency band. Then we added chronic stress as interaction with the condition. High and low chronic stress are plotted ±1 SD from the mean for descriptive purposes only in the Figures. Inferential analyses maintained the continuous nature of the chronic stress variable.
To explore condition effect on subsequent EF, and the mediation role of brain activities, we first predicted flanker reaction times (RTs) by condition. Then RTs was predicted by brain activities under image sequence and flanker task. Again, chronic stress was added as an interaction term.
Flanker performance was calculated as RTs on incongruent trials with RTs on congruent trials as a covariate. Trials with inaccurate responses or outlier RTs were deleted (3.14% of trials were deleted). We found that the first half of flanker trials (6 trials) followed by each image sequence had stronger interaction effect with chronic stress while the effect decayed in the end given temporal separation from the emotional stimuli. Thus, we used the initial six flanker trials following the emotional stimuli in the following analysis.
The authors declare no competing interests as defined by Nature Research, or other interests that might be perceived to influence the results and/or discussion reported in this paper.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Supplement EEG signal processing
EEG was recorded from a 128-channel BioSemi EEG device with digital sampling rate at 512 Hz. All EEG channels were referenced offline to the algebraic average of left and right mastoids and notch filtered (55~65 Hz) to remove power-line noise. EEG signals were bandpass filtered between 1~40 Hz. Bad channels were identified and spherically interpolated. Then the data were epoched from 4 seconds before and 35 seconds (total image sequence and flanker task duration) after the sequence onset. Each epoch was visually inspected. Those epochs with obvious abnormal signal segments, such as head movement, were excluded for Independent Component Analysis (ICA). Then ICA (EEGLab toolbox ICA function) was utilized to detect and remove artifact contaminated by eye movements, muscle, and cardiac artifacts. After removing the artifact components, the ICA source signals were transferred back to the original signal space, which was then used for the subsequent analysis.
The first half second of affective image viewing (0 to 0.5 s) was removed from analyses to eliminate sensory transition effect. The average number of artifact free seconds per participant was M = 79.90 (SD = 4.61), M = 79.80 (SD = 4.39), M = 238.4 (SD = 11.80) and M = 239.2 (SD = 9.20) for conditions of static-neutral, static-negative, dynamic-increase and dynamic-decrease respectively. There was no significant difference in the time length between static conditions (p = .910), and between dynamic conditions (p = .720).
The cross-spectral coherence between two channels was calculated using the following formula in (1). (Note: "# denotes the cross-spectrum, " and # denotes the auto-spectrum; The E denotes the expectation across the repeated sequences).