Event-related potentials to task-irrelevant sad faces as a state marker of depression

Highlights

• Negative bias was present in automatic face processing in depression.

• Negative bias was found in P1 amplitude to sad faces.

• Negative bias normalized in symptom reduction.

Abstract

Negative bias in face processing has been demonstrated in depression, but there are no longitudinal investigations of negative bias in symptom reduction. We recorded event-related potentials (P1 and N170) to task-irrelevant facial expressions in depressed participants who were later provided with a psychological intervention and in never depressed control participants. Follow-up measurements were conducted for the depressed group two and 39 months later. Negative bias was found specifically in the depression group, and was demonstrated as enlarged P1 amplitude to sad faces, which normalized in the follow-up measurements when the participants had fewer symptoms. Because the P1 amplitude recorded at the baseline did not differ between the depression group that recovered and the group that did not recover after the intervention, this brain response did not show potential as a biomarker for treatment response. It could have potential, however, to serve as a state-marker of depression.

Introduction

Depression is a common and highly recurrent disorder, which is most typically characterized by lowering of mood and reduction of energy and enjoyment (World health organization, 2010). According to Aaron Beck’s cognitive model of depression (1976, Beck, 1967, 1987), depressed individuals have a cognitive bias in information processing that predisposes them to selectively attend to negative stimuli. It could be a vulnerability factor that can affect the onset and recurrence of depression episodes (1976, Beck, 1967, 2008). Relevantly for the current study, this bias is suggested to occur also in automatic information processing facilitating the processing of negative stimuli already at early processing phases (Beck, 2008).

Negative bias in depression postulated in Beck’s theory has been demonstrated empirically with different types of stimuli (for reviews, see Mogg & Bradley, 2005; Peckham, McHugh, & Otto, 2010), especially with facial expressions (Gollan, Pane, McCloskey, & Coccaro, 2008; Gotlib, Krasnoperova, Yue, & Joormann, 2004; Naranjo et al., 2011, for a review see, Delle-Vigne, Wang, Kornreich, Verbanck, & Campanella, 2014). Negative bias in these studies was found as bias in attention or in memory to sad faces.

Processing of facial expressions has been studied widely with event-related potentials (ERPs), which give accurate timing for the brain activity related to different processing stages in face perception. The first ERP component that is modulated by facial expressions is P1, and attentive negative bias in emotional face processing in depression has been found in P1. When participants evaluated emotion intensity in faces, sad faces elicited larger P1 responses than happy or neutral faces in the depressed group but not in the control group (Dai & Feng, 2012; for absent negative bias in P1, see Dai, Wei, Shu, & Feng, 2016; Zhao et al., 2015). Negative bias is also demonstrated in subliminally presented, but attended, faces. Sad faces elicited a larger P1 response compared to neutral faces in the depressed group while controls had a smaller P1 for sad faces compared to neutral faces (Zhang, He, Chen, & Wei, 2016).

There is also evidence for depression-related negative bias in the N170 ERP component (Zhang et al., 2016; Zhao et al., 2015), which reflects structural feature processing in faces, including facial expression processing (Batty & Taylor, 2003). In an attentive condition, including a condition where subliminally presented faces are presented, N170 was larger to sad faces than to happy and/or neutral faces in the depressed participants, whereas in the control participants the N170 was the largest to happy faces (Zhang et al., 2016; Zhao et al., 2015). Furthermore, a direct comparison between the N170 in the depressed and control groups showed that the responses to sad faces were larger in depressed participants compared to control participants (Wu et al., 2016), reflecting mood-congruent bias in facial emotion processing. However, sometimes no differences between depressed and non-depressed have been found in N170 responses to facial expressions (Jaworska, Blier, Fusee, & Knott, 2012).

Beck’s cognitive model of depression suggests that negatively biased cognitive schemas function as automatic information processors (Beck, 2008). However, there is very little information on unattended or task-irrelevant processing of facial expressions in depression, especially with brain activity measurements that can reveal the time course for the processing (i.e. electroencephalography or magnetoencephalography studies, MEG; however, for a study applying functional magnetic resonance imaging, fMRI, see Suslow et al., 2010). In one study, ERPs were recorded for changes in emotional faces in depressed and control participants while participants attended faces with different colors (Chang, Xu, Shi, Zhang, & Zhao, 2010). In that study, the oddball condition was applied, in which the visual mismatch negativity (vMMN) component indexing cortical change detection is elicited (for reviews, see Kremláček et al., 2016; Stefanics, Astikainen, & Czigler, 2014). vMMN is calculated as the difference between responses to repeatedly presented standard stimuli and responses to rare deviant stimuli. In study by Chang et al. (2010), vMMN was found in two latencies reflecting mainly modulations in N170 and the following P250 component. Chang et al. observed smaller-amplitude vMMNs to happy and sad faces in the depressed group compared to the controls, thus showing no evidence of preattentive negative bias but instead an overall weakened cortical change detection related to facial expressions. However, in the study schematic faces, which have inevitably low ecological validity, were applied raising the question whether more naturalistic stimuli could reveal depression-related negative bias in task-irrelevant processing of facial expressions. In another study, task-irrelevant MEG responses were measured in participants with depression symptoms (dysphoric) and non-depressed controls to sad and happy faces presented in an oddball condition (Xu et al., 2018). Dysphoria-related negative bias was only found in later processing phase (M300 response), but no group differences were found in M100 or M170 responses, which correspond to P1 and N170 responses in ERPs, respectively.

Whether negative cognitive bias in depression is a trait-like characteristic or is state-dependent, that is, changes along with the degree of depressive symptoms, is unclear. Behavioral studies that have found similar processing bias in depressed and sub-clinically depressed participants (Dai et al., 2016) or in depressed and remitted participants (Joormann & Gotlib, 2007) or no change in negative bias in follow-up after remission (Bouhuys, Geerts, Mersch, & Jenner, 1996), have interpreted the result as reflecting a trait. However, some fMRI studies have shown normalization of brain activity for sad facial expressions after cognitive behavioral therapy (CBT; Fu et al., 2008) or after antidepressant treatment (Victor, Furey, Fromm, Ohman, & Drevets, 2010) suggesting state-dependency. Further support for state-dependency comes from an ERP study that found a correlation between depression symptom scores and negative bias in the N170 response (Wu et al., 2016) and from another study that found negative bias only in recurrent depressed individuals but not in first-episode depressed individuals suggesting that negative bias is associated with illness progression (Chen et al., 2014).

Brain responses to emotional faces may also have potential as indicators of treatment response. fMRI studies have shown that brain activation to sad expressions is associated with cognitive therapy treatment outcome in depressed participants (Costafreda, Khanna, Mourao-Miranda, & Fu, 2009; Fu et al., 2008). Costafreda et al. (2009) found that brain activity patterns related to sad facial expression processing, distinguish clinically remitted patients from non-remitted patients. Fu et al. (2008) included healthy control participants in comparisons and found better treatment response for patients who initially showed the most similar activity pattern to healthy controls in sad face processing. To best of our knowledge, ERP studies investigating treatment effect correlates of facial expression processing have not been reported, although they could be similarly feasible as the fMRI studies.

We aim to demonstrate automatic negative bias in depression reflected by ERP responses to pictures of real faces. If we find a negative bias related to depression, we will study the stability of the bias over time. In addition, we will investigate whether ERPs recorded in depressed participants for facial expressions can distinguish between those who recover and those who show no recovery after a brief psychological intervention.

We investigated P1 and N170 amplitudes to happy, sad and neutral faces presented in an oddball condition in which emotional faces were presented infrequently. The oddball condition was expected to be beneficial, because the responses to infrequent deviant stimuli could be expected to be enlarged compared to the frequently presented standard stimuli. We applied a stimulus condition where the identity of the faces, and thus, low-level visual features, changed trial-by-trial. Participants were instructed to attend to an audiobook during the face presentation. Since our adaptive behavior relies largely on preattentive cognition (Näätänen, Astikainen, Ruusuvirta, & Huotilainen, 2010) and cognitive negative bias is expected to exist already in the level of automatic processing (Beck, 2008), it is important to investigate task-irrelevant emotional face processing in depression.

Two groups of participants, depressed and age- and gender-matched non-depressed control participants, were enrolled in the study. We measured brain responses in the depressed group at three timepoints: at the baseline when all the participants were currently depressed and at 2-month (2-m) and at 39-month (39-m) follow-up measurements. At the 2-m measurement, approximately half of the depressed participants had received a brief psychological intervention for depression, and they were expected to have less depression symptoms (the other half of the group had been the first two months on a wait-list to receive the same intervention and they got the same intervention after the 2-m measurement). At the 39-m measurement, all of the depressed participants had received the intervention. Since it is very probable that some of the participants will have fewer symptoms after the intervention, this design allows us to study changes in brain responses in relationship to changes in depression symptoms. Clinical outcomes were assessed with questionnaires after the intervention for both groups to further divide the depressed participants into groups of recovered and non-recovered.

Based on previous attentive studies, we expect larger P1 and N170 amplitudes to sad faces compared to neutral and happy faces in the depressed group (Dai & Feng, 2012; Zhang et al., 2016; Zhao et al., 2015), reflecting negative bias in information processing in depression (1976, Beck, 1967, 1987). In addition, larger N170 responses to happy faces than to neutral faces are expected in the control group (Astikainen & Hietanen, 2009; Astikainen, Cong, Ristaniemi, & Hietanen, 2013; Zhao et al., 2015). Based on previous studies, we also hypothesize a positive correlation between depression symptom scores and negative bias (Chen et al., 2014; Wu et al., 2016) and normalization of the responses when depression symptoms are reduced (Fu et al., 2008; Victor et al., 2010). Furthermore, we expect depressed participants who benefit less from the brief psychological intervention to show more pronounced initial negative bias compared to those who respond better to the intervention, while those who recover would show similar processing compared to the controls (Fu et al., 2008).