Interactions Between the Prefrontal Cortex and Attentional Systems During Volitional Affective Regulation: An Effective Connectivity Reappraisal Study

The procedure was compliant with the directives of the Helsinki Declaration (1975, revised 2000) and approved by the Ethical Committee of the Institute of Psychology, Jagiellonian University. All participants were warned that the film clips may present strongly affective content. 33 women volunteers (median age: 21, range 18–33) participated in the study after signing a written consent form. All volunteers were medication-free with no reported history of any neurological or psychiatric disorders or substance abuse, and had normal or corrected-to-normal vision. During the procedure, the subjects were seated in an air-conditioned soundproof cabin in front of a 24″ LCD monitor. We used the Biosemi ActiveTwo system with 64 electrodes placed on a 10–10 headcap, with four additional leads for monitoring eye movement and two more electrodes for offline linked mastoid reference.

During a prior experimental session, the participants received in-depth training in reappraisal while viewing standardized emotional pictures. They were taught how to generate more positive interpretations of the scenes in three ways: (1) by denying their reality, (2) by taking a third-person perspective, or (3) by imaging their positive outcomes. Participants were free to use any of these strategies to control their emotions in the most efficient way. During the training session, they were instructed to use reinterpretation through several trials, with feedback and shaping by the experimenter (during some trials, the subjects were asked to report their reinterpretations aloud). When a cue word (“control” or “watch”) appeared on the screen, they were taught to reappraise or to simply view the picture and respond naturally, letting their emotions arise. The training session lasted until the participants declared they are ready to proceed with the experiment based on the film clips. The subjects were also asked to implement the strategy only after the picture/film appeared on the screen and to fully watch its content, without closing their eyes or looking away.

Thirty-six 20 s-long film clips were chosen, aiming to feature balanced content, including 12 neutral (normal food, people in everyday activities or ordinary relationships, animals, beauty treatments, neutral landscapes) and 24 negative clips (disgusting food, people in dramatic situations, fights, animal slaughter, mutilations, surgical procedures, accidents). The clips were presented randomly in three runs of fixed-ordered blocks (conditions): NEU (watching neutral clips), NEG (passively watching negative clips), and REAP (reappraising negative clips). Each block consisted of four clips. During the clips, an additional hint was displayed to remind the participants about the current experimental conditions. The block design was chosen to minimize extra brain activations associated with the effort of switching between the tasks. At the end of each block, the subjects were asked to assess their current affective experience regarding valence, intensity and dominance associated with their emotions using a computerized version of the Self Assessment Manikin. Additionally, after the main procedure, the subjects were asked to estimate the negativity of each film clip and to report any failures in their reinterpretation of particular clips.

We assumed that the manifestations of voluntary emotional control will be visible in the contrast of “reappraisal” minus “passively watching negative clips” conditions (these conditions differ only in terms of the instruction received: to reappraise or to watch the clips naturally), while the manifestation of processing or monitoring of emotionally-loaded content will be visible in the contrast of “passively watching negative clips” minus “watching neutral clips” conditions. Since our main connectivity analyses were related to a few directional hypotheses we used planned t-contrasts for ultimate verification of the expected differences between conditions for both behavioral and connectivity data.

As the ability to regulate negative emotions depends on individual and situational characteristics, in order to control this variability, we calculated an additional variable: effectiveness of control. It was estimated based on the valence ratings from the Self Assessment Manikin test. For each participant we calculated the difference between mean valence ratings after “reappraisal” blocks and mean valence ratings after “passively watching negative clips” blocks. Arousal and dominance ratings were not included in the effectiveness of the control variable, as emotional control is mainly targeted towards the valence dimension. Thus, the effectiveness of control variable reflected the within-group variation in participant reappraising ability, motivation, and involvement in the procedure. This variable was used as a covariate in further analysis.

The DTF was used to conduct the connectivity analysis. The DTF method is based on the multivariate autoregressive model (MVAR). In this approach we assume that a sample of the data in k channels at a time t can be expressed as a weighted sum of p previous samples with a random component added:where \(X(t) = (X_{1} (t),X_{2} (t), \ldots ,X_{k} (t))^{T}\) is the vector of data values at the time t and \(E(t) = (E_{1} (t),E_{2} (t), \ldots ,E_{k} (t))^{T}\) is the vector of the random component values at the time t. The matrices A(j) (of size k × k) are known as the model parameters (or coefficients) and p is known as the model order.

This approach is well established in EEG data analysis and has many advantages. Most importantly, it makes it possible to construct a measure of causal relations between signals X _i (t) based on the Granger causality concept (Granger 1969). Granger causality defines a signal Y as causal for a signal X if values of X can be better predicted using previous values of both signals X and Y than using previous values of signal X alone. In particular, DTF estimates derived from MVAR parameters can be interpreted as causal relations. Another advantage of the MVAR approach is that it accounts for the whole multivariate set of signals, so the analysis is not performed separately for every pair of signals (pair-wise). This way we avoid problems which may arise in the event of a presence of common sources in the set of signals, which may lead to a misinterpretation of results (Kus et al. 2004) Moreover, because volume conduction propagates nearly instantly, it does not involve a time delay between signals and does not increase transmission in DTF results (Kamiński and Blinowska 2014). This greatly improves spatial resolution as compared to a typical EEG analysis, which has been confirmed by several neuroimaging studies in different domains (Brzezicka et al. 2010; Kamiński and Blinowska 2014).

The MVAR model can be transformed to the frequency domain where it takes a form of a linear filter H with noise E on input and signal X on its output:where X(f), A(f) and E(f) are the Fourier transform of X(t), A(j) and E(t), respectively. The matrix H(f) = A ⁻¹(f) is known as the transfer matrix. Details of the procedure can be found in Kamiński and Blinowska (1991).

The normalized DTF function is defined in the frequency domain as:where H _ij (f) are elements of the transfer matrix H. Normalized DTF γ _ij (f) describes the causal influence of channel j on channel i at frequency f. The above equation defines a normalized version of DTF, which takes values from 0 to 1 producing a ratio between the inflow from channel j to channel i to all the inflows to channel i.

The preprocessing of the EEG signal was performed using the EEGLab toolbox which included downsampling to 128 Hz. Next, beta band frequencies were extracted using zero-phase 13 Hz high-pass and 30 Hz low-pass filters. The beta band has been chosen as the flows can be characterized as moderate or long distance cortical connections. It has been proposed that such connections are based, above all, on beta oscillations (Bastos et al. 2012; Vossel et al. 2013). This is also consistent with the evidence of the crucial role of beta oscillations in attentional processes, especially in the visual domain (Gola et al. 2013; Kamiński et al. 2012; Kus et al. 2008; Wróbel 2000). Additionally, our previous studies based on the DTF measure also confirmed the importance of the beta band in cortical communication (Wyczesany et al. 2014a, b).

Since the signal subjected to the DTF analysis should not be modified by any method which alters the original correlation structure of the dataset, we did not apply any artifact correction method. Instead, the fragments of signals contaminated with artifacts were rejected. To obtain this, the 20 s-long recordings from each movie clip were divided into 2 s-long epochs which were screened for artifacts. First, the independent component analysis (ICA) was used to separate the eye movement signal, identified using criteria of scalp distribution and temporal characteristics (Jung et al. 2000). Trials in which the amplitude of eye components exceeded 30 μV were rejected. Second, trials where the amplitude on any electrode exceeded 100 μV were also rejected (average 26.6 per subject, mostly due to eye-blinks). Moreover, the epochs for which the subjects reported failures in reinterpretation were also excluded (average 1.33 film clip per subject). Due to statistical constraints, we reduced the set of simultaneously processed signals to 24 channels (provided below). 256-sample long (2 s), artifact-free epochs were analyzed. Then, in order to ensure that the remaining data length after all above processing steps was sufficient to proceed with MVAR modelling, the following formula was applied: W ≥ 10 * pM/N (where W is the required minimum window size in samples, p is the model order, M is the number of channels and N is the total number of epochs; Korzeniewska et al. 2003). DTF calculations were carried out using Multar software (University of Warsaw). The MVAR model order was set to six, according to the Akaike criterion (AIC). Normalized DTF values were then determined. In order to compare propagations for the two experimental conditions, the DTF results were integrated over the selected beta frequency band (13–30 Hz). The distributions of the DTF values were checked to identify and reject possible extremes, defined using boxplot 1.5 IQR (interquartile range). Based on brain atlas (Kaiser 2007; Okamoto et al. 2004), the following regions of interest (ROIs) were defined: anterior orbitofrontal cortex (OFC: AF7, AF8, Fpz); left/right dorsolateral PFC (LDL: F3, F5; RDL: F4, F6); anterior cingulate (ACC: FCz, Fz); right inferior frontal gyrus (RIFG: F8, AF8); left/right intraparietal sulcus (LIPS: CP1, P3, P1; RIPS: CP2, P4, P2); right temporoparietal region (RTPJ P6, P8, TP8); and occipital area (Occ: O1, O2, PO3, PO4). In spite of some concerns regarding the quality of EEG recording from deeper structures like the anterior cingulate, we decided to consider this region in our analyses, due to its importance in emotional control and affective phenomena. It was further justified by the fact, that in the classic EEG analysis techniques, the ACC activity is widely analyzed (just to mention the ERN or N2b components of evoked potentials; Herrmann et al. 2004).