Our academic experts are ready and waiting to assist with any writing project you may have. From simple essay plans, through to full dissertations, you can guarantee we have a service perfectly matched to your needs.
GET A 40% DISCOUNT ON YOU FIRST ORDER
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/biopsycho
Sustained engagement of attention is associated with increased negative
self-referent processing in major depressive disorder
, Logan T. Trujillob
, David M. Schnyera
, Christopher G. Beeversa
a Department of Psychology, The University of Texas at Austin, United States
b Department of Psychology, Texas State University, United States
Major depressive disorder
This study investigated the link between self-reference and attentional engagement in adults with (n = 22) and
without (HC; n = 24) Major Depressive Disorder (MDD). Event-related potentials (ERPs) were recorded while
participants completed the Self-Referent Encoding Task (SRET). MDD participants endorsed significantly fewer
positive words and more negative words as self-descriptive than HC participants. A whole-scalp data analysis
technique revealed that the MDD participants had larger difference wave (negative words minus positive words)
ERP amplitudes from 380 to 1000 ms across posterior sites, which positively correlated with number of negative
words endorsed. No group differences were observed for earlier attentional components (P1, P2). The results
suggest that among adults with MDD, negative stimuli capture attention during later information processing; this
engagement is associated with greater self-referent endorsement of negative adjectives. Sustained cognitive
engagement for self-referent negative stimuli may be an important target for neurocognitive depression interventions.
The way that one views oneself, one’s self-concept or self-schema, is
intricately tied with mood. Beck’s (1967) cognitive model of depression
postulates that the schema – internal beliefs and knowledge about the
self, the world, and the future – influences how life events are appraised
and interpreted. Schemas also prioritize the processing of incoming
information, such that environmental stimuli that are consistent with
one’s self-schema are attended to, processed, and subsequently recalled
more readily (Segal, 1988). A negative self-schema may result in biased
interpretation of ambiguous stimuli, or cause elaborative processing of
over-attended stimuli (Everaert, Koster, & Derakshan, 2012). This in
turn has been theorized to facilitate increased recall of negative stimuli,
resulting in negatively biased memory. Although other mechanisms
clearly also contribute to the maintenance of depression (e.g., emotional
blunting), negative self-schema may be an important mechanism
that fuels many of the negative cognitive biases thought to maintain
Consistent with the cognitive model, individuals with Major
Depressive Disorder (MDD) have been shown to display negatively
biased attention, interpretation, and memory (Everaert et al., 2012).
Further, depressed people often do not display protective positive
cognitive biases that are observed in healthy individuals (Disner,
Beevers, Haigh, & Beck, 2011; Walker, Skowronski, & Thompson,
2003). Major depression instead privileges negative processing and, as
a result, individuals with MDD are likely to view themselves as having
more negative and fewer positive characteristics than non-depressed
One method of measuring self-schema is the self-referent encoding
task (SRET; Derry & Kuiper, 1981). The SRET is a binary-choice, affective
decision-making task combined with incidental recall of SRET stimuli.
The SRET is generally a computer-based task, where positive and
negative adjectives are presented one at a time to participants who
determine as quickly as possible whether each word is self-descriptive
or not. Following presentation of the word stimuli, participants are
then asked to recall as many of the SRET stimuli as possible. Studies
have shown strong correlations between endorsement of negative
(but not positive) words and depressive symptoms (e.g., Disner,
Shumake, & Beevers, 2016); increased endorsement of negative words
(and decreased endorsement of positive words) on the SRET is also
predictive of depression symptom course (Connolly, Abramson, & Alloy,
2015; Disner et al., 2016). Responses on the SRET also appear to be
consistent over time particularly when depression symptoms remain
relatively stable (Auerbach et al., 2016; Goldstein, Hayden, & Klein,
These studies provide a clear link between negative self-referent
Received 5 June 2017; Received in revised form 6 September 2017; Accepted 6 September 2017
⁎ Corresponding author at: Department of Psychology, The University of Texas at Austin, Austin, TX 78712, United States.
E-mail address: email@example.com (J. Dainer-Best).
Biological Psychology 129 (2017) 231–241
Available online 08 September 2017
0301-0511/ © 2017 Elsevier B.V. All rights reserved.
cognition, as measured by the SRET, and depressive symptoms.
Understanding the neural architecture of negative self-referent bias is
important, as it could provide a more comprehensive understanding of
this important cognitive bias and point to translational treatment targets
for neurocognitive interventions.
In the current project, we used electroencephalography (EEG) to
measure the temporal characteristics of cognitive processes involved in
self-appraisal. Although the spatial resolution of EEG is not ideal, EEG is
extremely effective at measuring information about the time course of
cognitive phenomena (Kappenman & Luck, 2012). Thus, collecting
event-related potentials (ERPs) during the SRET can provide information
as to whether biased self-referent processing is occurring at an
early processing level; whether it occurs at a later level of cognitive
evaluation; or whether both processes contribute to this negative selfreferent
Early ERP processes include the P1 and P2, both positive deflections
in an ERP waveform thought to reflect automatic processing of attentional
information that may nonetheless be influenced by emotion
(Delplanque, Lavoie, Hot, Silvert, & Sequeira, 2004; Hajcak, Weinberg,
MacNamara, & Foti, 2012). These peaks occur between 100 and 300 ms
following a stimulus. The P2 in particular may index post-perceptual
selective attention, as it occurs late enough (peaking approximately
180 ms after stimulus onset) to be related to the association of new
information with prior comprehension (Hajcak et al., 2012;
Luck & Hillyard, 1994). As both occur early, they are understood to be
related to early attentional engagement; both are typically increased
when attending to emotional stimuli (Delplanque et al., 2004).
The late positive potential (LPP), conversely, is a component often
considered an index of cognitive evaluation and engagement with stimuli.
The LPP begins around 300 ms post-stimulus and continues up to
1500 ms (Cuthbert, Schupp, Bradley, Birbaumer, & Lang, 2000; Hajcak
et al., 2012). The LPP is often either posterior or central in localization
(Hajcak et al., 2012). The LPP increases in positive amplitude in response
to prioritization of information, indicating increased engagement,
especially to negative information. Schupp et al. (2004) demonstrated
that the LPP is greater in response to unpleasant or negative
images as compared to neutral or positive images, regardless of mood
state; others have shown that attending to non-arousing images reduces
the LPP (Hajcak, MacNamara, Foti, Ferri, & Keil, 2013). Moreover,
some studies have shown a generally diminished LPP in participants
with MDD (Blackburn, Roxborough, Muir, Glabus, & Blackwood, 1990;
Proudfit, Bress, Foti, Kujawa, & Klein, 2015; Weinberg, Perlman,
Kotov, & Hajcak, 2016).
Several studies have attempted to use EEG to identify the key ERP
components that contribute to negative self-referent processing during
the SRET. A prior study found that both current and remitted MDD
groups had increased amplitudes for negative stimuli in an early component
of attentional capture (the P2) in comparison to healthy controls
(Shestyuk & Deldin, 2010). They also found that individuals who were
currently depressed showed more positive amplitudes in the late positive
potential (LPP) for negative stimuli than the other groups. This
suggests that MDD participants selectively attended to negative information
(due to the initial P2 amplitude difference from controls) and
were engaging in increased cognitive evaluation of negative information
(due to the increases in the LPP compared to healthy controls).
Similarly, in a sample of depressed and healthy female adolescents,
depressed girls were shown to exhibit greater early (P1) amplitudes in
response to negative words, and greater later (LPP) amplitudes to negative
words (Auerbach, Stanton, Proudfit, & Pizzagalli, 2015). These
findings are consistent with the results of prior work, with the MDD
group showing early attention to negative words that continues over
the time-course of the ERP.
Further work with a large sample of younger female participants
(N = 121) found indications that risk for depression (i.e., maternal
history of MDD) was also associated with greater LPP amplitudes to
negative words when compared to those at low risk for MDD (Speed,
Nelson, Auerbach, Klein, & Hajcak, 2016). This study used a principle
components analysis (PCA), which builds components from the EEG
electrode channels that most strongly contribute to an outcome. With
the LPP described by the PCA, there was no difference between positive
and negative valence within groups; however, in response to negative
words only, the at-risk participants showed increased LPP amplitudes
and increased subsequent recall of negative stimuli compared to positive.
This study did not find differences in the earlier waveforms (P1 or
P2). An additional study investigated depressive response on the SRET
from a semantic processing perspective (i.e., the N400 waveform), arguing
that a diminished N400 suggests stronger self-reference (Kiang
et al., 2017). This study demonstrated that participants with MDD had a
diminished N400 in response to negative, but not positive, adjectives.
A recent SRET study using PCA techniques with ERPs in a large
community sample of adults (N = 128) found that individuals with
elevated depressive symptoms had enhanced negativity to both positive
and negative words in early frontal regions (Waters & Tucker, 2016). A
waveform that they believed to be an element of the late positive
complex or P300, at a similar time frame to the LPP, was attenuated in
response to all stimuli in parietal regions. Notably, these findings are in
the opposite direction to the results reported above (e.g., the LPP
measured at a similar point in time was increased in depressed female
adolescents in Auerbach et al., 2015).
In summary, these studies reveal some conflicting results in terms of
the waveforms associated with the SRET, raising a question of whether
there is attenuation or augmentation of the early selective attention
components (P1, P2) and later cognitive evaluation (LPP) in response to
negative stimuli in depressed participants relative to healthy controls.
Many of the above-reviewed studies were conducted in young, female
participants; it is important to determine which of these findings, if any,
extend to adult samples. Further, relatively few studies have been
completed in a sample with a clinical diagnosis of MDD. Additionally,
given the recent emphasis in psychology to replicate novel research
(e.g., Munafò et al., 2017), this study’s potential to independently replicate
prior work in this area is important.
In the current study, we anticipated that behavioral results would
follow in the same vein as previous work, with more endorsements of
negative words as self-referent in participants diagnosed with MDD
compared to healthy controls. Based on prior research and the cognitive
model of depression, we predicted that adults with MDD, in comparison
to healthy controls, would show early, differential attention between
negative and positive words in the P1 and P2. We also predicted that
MDD-diagnosed participants, as compared to healthy controls, would
show increased cognitive evaluation in later components (similar to the
LPP) for negative stimuli. Were this confirmed, it would imply that
differential processing of self-referential information results from both
the early components involved in perception and selective attention of
negative stimuli, and also from the way that these stimuli are elaborated,
processed, and encoded.
To better assess the full span of attentional processing in response to
word presentations, we conducted analyses using a non-parametric
technique often applied to functional neuroimaging analyses
(Nichols & Holmes, 2002), which identified spatiotemporal areas that
might be strongly differentiating between the MDD and HC groups
during self-referential processing. This technique, discussed further
below, uses randomized permutations of the data to conduct point-bypoint
t-tests, correcting for multiple comparisons, which allows spatiotemporal
areas with strong differences to rise to the forefront. This
data-driven method identifies the onset of differential between-group
responses in a manner that is conservative compared to standard
parametric approaches because, first, no assumptions of normality are
required and, second, no a priori (and possibly biased) choices of time
window or electrode region are necessary. This is in contrast to the
traditional, parametric approach to ERP analysis, which examines activity
within a limited number of electrode sites averaged across specified
time windows. Nevertheless, we also conducted a limited number
J. Dainer-Best et al. Biological Psychology 129 (2017) 231–241
of parametric analyses to allow for easier comparisons of results from
past work and the current study.
Adults were recruited from the Austin, TX community through the
use of advertisements posted on websites (Craigslist, a UT-Austin
message board, and Indeed) and fliers. Postings described a study on
“mood” and “emotional experiences” but highlighted the need for both
healthy and depressed subjects. The advertisements directed participants
to a website to determine eligibility. At this website, participants
provided informed consent for the screening and filled out a brief
survey of current mood and demographics. Research assistants and
graduate students trained on diagnostic interviewing conducted phone
screenings on eligible participants, using the Mini International
Neuropsychiatric Interview (version 6.0, Sheehan et al., 1998). The
MINI is a standardized instrument used for brief screenings to diagnose
a variety of psychiatric disorders. Research assistants took part in a
training workshop during which they learned interview skills, roleplayed
interviews, and reviewed diagnostic criteria. After the workshop,
they listened to calls conducted by experienced researchers and
had their initial screening interviews monitored for fidelity. Phone calls
were audio-recorded with consent from participants throughout the
study for fidelity analyses. An independent assessor (J.D.B.) randomly
selected and rated 20% of MDD and HC interviews. Agreement for MDD
diagnosis between study interviewers and the independent assessor was
excellent (k = 1.00, p < .0001).
2.2. Inclusion and exclusion criteria
Participants were included in the current study if they were between
the ages of 18 and 55 and spoke fluent English. Participants who scored
less than 13 or greater than 16 on the Center for Epidemiologic Studies
– Depression Scale (CESD) during online screening were invited to
complete the MINI interview over the phone. Participants with
CESD > 16 who met DSM-5 criteria for MDD (N = 22; Diagnostic and
Statistical Manual of Mental Disorders, 2013) were included in the
study. Nine participants met criteria for a current single major depressive
episode, while 12 met criteria for recurrent MDD. Participants
with MDD were included whether or not they met criteria for a current
anxiety disorder; for all participants in this group, MDD was the primary
diagnosis as assessed during the MINI. Of those diagnosed with
MDD, 12 met criteria for one or more DSM-IV anxiety or trauma-based
disorders (seven for Generalized Anxiety Disorder, two for Social Anxiety
Disorder, five for panic attacks, and four for PTSD).
Healthy control participants had a CESD < 13 and did not meet
diagnostic criteria for past or current MDD or a current anxiety disorder
(N = 24). Importantly, all participants (HC and MDD) were excluded
from the study if they met diagnostic criteria for the following disorders:
current alcohol or substance abuse or dependence; mania or
hypomania; bipolar disorder; or psychosis. Participants were also excluded
based on criteria that could affect EEG collection, including a
history of seizures or epilepsy, head trauma, current use of betablockers,
and current use of anti-psychotic drugs (Keil et al., 2014).
2.3. Sample characteristics
Data from one participant in each condition were excluded because
of poor EEG data quality. Demographics are reported in Table 1; participants
were in their mid-twenties, mostly female, and approximately
half were white (53% white, 28% Hispanic/Latino, 16% Asian, and 2%
Black). Groups did not differ significantly on the basis of age (t[41.82]
= −0.33, p = .74), gender (χ2  = 2.30, p = .32), or race (χ2 
= 0.77, p = .85). Groups did of course significantly differ on
depression severity, as measured by the CESD, t(31.6) = −14.0,
p < .001.
2.4. Power analysis
A priori power analyses were conducted to determine sample size
necessary to achieve a medium-to-large effect size (η2 = 0.22), as per
Auerbach, Stanton, Proudfit, and Pizzagalli (2015) for multivariate
ANOVA; with the current sample size, we would achieve 80% power.
The institutional review board at the University of Texas at Austin
provided approval for the study (IRB # 2014-08-0078). Participants who
were eligible for the study following phone screening were scheduled for
a 1.5-h session in the lab. Informed consent was obtained, and participants
completed several self-report questionnaires. Following this,
Easycap EEG caps were placed on the participant’s head and prepared for
EEG collection, and then participants completed experimental tasks on a
computer. Participants were paid $20 for completing the study and were
provided with a list of local mental health resources.
2.6.1. Center for epidemiologic studies – depression scale (CES-D)
The CES-D (Radloff, 1977) is a self-report scale designed to assess
depressive symptoms over the past week using 20 items. Scores may
range from 0 to 60; a score greater than 16 is often used as a cut-off for
elevated depressive symptoms (Radloff, 1977; Santor, Zuroff, Ramsay,
Cervantes, & Palacios, 1995). The CES-D was used for screening and
was assessed again during the laboratory visit when it was confirmed
that the CES-D remained above 16 for participants with MDD and below
13 for healthy controls.
2.6.2. Mood and anxiety symptoms questionnaire (30-item version; MASQ)
The MASQ (Clark & Watson, 1991) is a 90-item self-report scale
designed to measure the tripartite model of depression, anxiety, and
general distress. A short (30-item) adaptation of this questionnaire has
been developed, which maps closely onto the original questionnaire
(Wardenaar et al., 2010), and was used in the present study. The MASQ
provides sub-scales of General Distress, Anhedonic Depression, and
Anxious Arousal. The Anxious Arousal subscale was used as a covariate
in a final set of analyses to determine if anxiety symptoms altered observed
Characteristics of participants included as healthy controls (HC) and diagnosed as depressed
Age, mean (SD) HC (N = 23) MDD (N = 21)
25.3 (7.9) 24.6 (6.7)
Female 15 (65%) 16 (76%)a
White 13 (57%) 10 (48%)
Hispanic 6 (26%) 10 (48%)
None 22 15
Current medication usage 1b 6
Current SSRI, for > 10 weeks 1 4
Other antidepressant 0 2
Anti-anxiety medication 0 1
CESD, mean (SD) 4.9 (5.2) 34.7 (7.93)
MASQ, Anxious Arousal subscale 11.3 (2.0) 19.6 (6.1)
MASQ, General Distress subscale 13.9 (4.2) 32.3 (8.3)
MASQ, Anhedonic Depression subscale 24.2 (6.4) 40.5 (5.6)
a One participant in this group identified as agender. b This participant reported long-term medication usage, but otherwise met inclusion
criteria. Analyses were conducted without this participant and showed no significant
J. Dainer-Best et al. Biological Psychology 129 (2017) 231–241
2.6.3. Self-referent encoding task (SRET)
The SRET (Derry & Kuiper, 1981, see Fig. 1) is an affective decisionmaking
task where participants make binary-choice decisions about
whether positive and negative words are self-descriptive. Participants
view the words on a computer screen and make rapid judgments following
the word’s display. In this version, participants viewed 40 negative
and 40 positive words (as in Auerbach et al., 20151
from the Affective Norms for English Words (Bradley & Lang, 2010) for
a total of 80 trials.
Stimuli were presented in random order for 200 ms, followed by
1800 ms of a fixation cross. Since subjects were instructed to withhold
their motor response during this fixation cross, event-related potentials
to word presentation and decision-making remained uncontaminated
by motor responses. Only after offset of the fixation cross were participants
presented with the question prompt, “Does this word describe
you?” Participants used a Logitech gaming controller’s shoulder buttons
to respond “yes” or “no”. Although the 1800 ms period between the
stimulus offset and behavioral response allows for the recording of
neural response to stimuli, this extended period renders the reaction
time response less meaningful and difficult to interpret.2 Thus, we focus
on ERP responses to word stimuli rather than reaction time in our
analyses, which is consistent with prior work in this area (e.g.,
Auerbach et al., 2015). Participants completed several neutral practice
trials before the task began to ensure that they knew to wait to respond
until the question appeared. A jittered intertrial interval followed each
trial, between 1500 and 1700 ms in length.
After completing the task, participants completed an image-based
task for approximately 12 min. Following this distraction, they were
asked to recall as many adjectives as possible from those presented
during the SRET, within five minutes. Participants were not previously
informed that they would be asked to perform this recall task. The
primary behavioral outcome from the SRET is the number of positive
and negative words endorsed as self-referential. An alternative method
(see e.g., Goldstein et al., 2015) involves calculating processing bias
scores, with a negative score calculated as numberofnegativewordsendorsed
the reverse for positive valence. Thus, individuals with an increased
number of negative words endorsed, or a negative processing bias
closer to one, have a stronger negative processing bias.
2.7. EEG recording and data analysis
EEG was recorded using a 64-channel active electrode system placed
in the Easycap recording cap and recorded with the BrainVision
actiCHamp amplifier and PyCorder software. In addition to the 64 cap
channels, an additional four channels were collected to track vertical
and horizontal eye movements. All head channels were located based
on extended 10/20 system locations; cap sizes were chosen based on the
circumference of participant’s heads. Electrode impedances were reduced
using an electrical conducting gel to below 10 kΩ. Continuous
EEG were sampled at 500 Hz, initially referenced to Cz. Offline, data
were processed using BrainVision Analyzer 2.0 software where the data
were re-referenced to an average reference of all head channels.
Electrooculogram (EOG) channels were created by subtracting active
electrodes placed below the eyes from above-eye (Fp1 and Fp2)
sites for vertical EOG, and by subtracting sites placed outside of the left
and right canthi of the eyes for horizontal EOG. A Butterworth infinite
impulse response filter was applied to bandpass filter the data from 0.1
to 30 Hz (slope of 12 dB/oct), and sections with major artifacts identified
by visual inspection were marked and excluded from analysis.
Several participants (N = 6) had one or two faulty electrodes; these
participants had the faulty channel(s) interpolated using a linear triangulation
algorithm before further analyses were conducted. An independent
component analysis (ICA) transform was then conducted in
order to identify and remove the effects of eye blinks and eye movements,
using both vertical and horizontal EOG channels. Each participant’s
whole dataset was used to calculate the ICA matrix, and a restricted
Infomax rotation was used to decompose the ICA and remove
components relating primarily to eye blinks and eye movements (Jung
et al., 2000; Lee, Girolami, & Sejnowski, 1999).
Individual trials were split into 1500 ms epochs selected from
200 ms before stimulus onset to 1300 ms following stimulus onset.
Epochs were created separately for positively- and negatively-valenced
words. Following the creation of these epochs, intervals were further
artifact assessed using the following semi-automated criteria based on
Auerbach et al. (2015): a maximal voltage step of 50 μV/ms; a voltage
difference above 300 μV within an epoch; amplitudes above 200 μV or
below −200 μV; and periods longer than 100 ms with activity under
0.5 μV. Epochs that did not pass these parameters were rejected from
Fig. 1. The SRET (Self-Referent Encoding Task). Event-related
potential epochs are based on the moment of word
1 The following 80 words were included in the SRET: Positive words were: admired,
adorable, alive, beautiful, bold, bright, capable, carefree, confident, cute, devoted, dignified,
elated, engaged, famous, festive, friendly, gentle, grateful, happy, honest, hopeful,
inspired, jolly, joyful, lively, loyal, lucky, masterful, outstanding, proud, satisfied, silly,
surprised, terrific, thoughtful, untroubled, useful, vigorous, wise. Negative words were:
afraid, alone, angry, anguished, bored, brutal, burdened, cruel, crushed, defeated, depressed,
disgusted, disloyal, displeased, distressed, dreadful, fearful, frustrated, guilty,
helpless, hostile, insane, insecure, lonely, lost, morbid, obnoxious, rejected, rude, scared,
shamed, sinful, stupid, terrible, terrified, troubled, unhappy, upset, useless, violent.
Positive and negative words were matched for arousal, frequency, and length (Auerbach
et al., 2015). 2 Indeed, we tested whether group and valence predicted reaction times, and found no
J. Dainer-Best et al. Biological Psychology 129 (2017) 231–241
further analysis. A linear de-trend3 was then applied to the data based
on the 100 ms before stimulus onset and the 100 ms at the end of the
epoch, and a baseline correction was applied by averaging the period
from 200 ms preceding stimulus onset until that onset.
Average responses were created per-participant, for negative and
positive words separately. A minimum of 20 valid epochs were required
per participant, per each valence; all participants had at least this many.
On average, 36 (SD = 2.9) epochs per participant were included for
negative word stimuli trials, and 36.1 (SD = 2.7) epochs were included
for positive word stimuli trials. Difference waves were generated by
subtracting positive from negative word stimuli trials.
The data were examined using pointwise non-parametric randomized
permutation t-tests, which were corrected for multiple comparisons
across time and site (Nichols & Holmes, 2002; Pernet, Latinus,
Nichols, & Rousselet, 2015; Sanguinetti, Trujillo, Schnyer,
Allen, & Peterson, 2015; Trujillo, Allen, Schnyer, & Peterson, 2010).
This cluster-based method allows for the identification of differential
responses between MDD and HC groups in a manner that is more
conservative than standard t-tests, avoids putative a priori choices of
regions of analysis by utilizing the full scalp recorded data. This analysis
was performed on the difference waves (negative words minus
positive words) in a between-subjects analysis.
This non-parametric statistical method consists of a three-step process
to create an empirical null distribution for the between-group ERP
difference waves to be used for hypothesis testing the between-group
differences. First, we computed a statistical significance threshold for
the between-group difference waves at electrode and time point. As the
ERP responses were recorded from 62 channels (after the two
facial channels Fp1 and Fp2 were dropped) over a 1.5 s (1500 ms)
epoch at a 500 Hz sampling rate, this amounted to a total of
62 × 1.5 × 500 = 46,500 independent thresholds. We determined
these thresholds by computing a distribution of 20,000 between-group
t-statistics for each data point under the null hypothesis. Each t-statistic
was computed after exchanging (permuting) the data of a randomized
subset of participants in each group (the size of each subset equaled the
number of individuals in the smaller of the two subject groups, i.e.,
n = 22). If the null hypothesis is true and there are no between-group
differences, then the t-value computed after this exchange is still an
element of the null distribution (because exchanging subjects across
groups should make no difference if there truly are no between-group
differences). This process was repeated 20,000 times to create a distribution
of t-values from which we determined the two-tailed p = 0.05
threshold for each of the 46,500 data points.
Because such a large number of independent tests will inflate type-I
error, it was necessary to correct for multiple comparisons. We accomplished
this in a second step by using these significance thresholds
to determine contiguous t-statistic clusters across electrodes and time
points. We then computed the distribution of maximal t-statistic clusters
under the null hypothesis. This was accomplished by computing a
second round of 20,000 between-group data permutations, where
during each permutation new t-values were computed for each data
point. Those t-values that were above the p = 0.05 thresholds determined
in the first step of this procedure were then divided into
contiguous clusters. Data was arranged into a three-dimensional
structure (anterior-posterior electrode dimension, left-right electrode
dimension, time dimension) and t-statistic clusters were defined as
three-dimensional neighborhoods of contiguous points (26-connected
point neighborhoods). Then, for each identified cluster, we computed
its exceedance mass, defined as the summed total of t-values within the
cluster (i.e. “the integral of the statistic image above the primary
threshold within the suprathreshold cluster”; Nichols & Holmes, 2002,
p. 8). We then selected the largest exceedance mass for a given permutation,
yielding a distribution of 20,000 maximal exceedance mass
values under the null hypothesis.
In a final step, we used the previously-obtained null distribution of
maximal cluster exceedance masses to hypothesis test the cluster exceedance
masses observed in the non-permuted data. Cluster exceedance
masses calculated from the non-permuted data with sizes
greater than the null distribution’s p = 0.05 criterion exceedance mass
were considered to be significant at the two-tailed level with strong
control for type-I error. All p-values were corrected for multiple comparisons
using a step-down procedure (Holmes, Blair, Watson, & Ford,
1996). As each cluster corresponded to a spatiotemporal extent of between-group
ERP differences, this method allowed us to simultaneously
identify where on the scalp and when in time those differences were
In an effort to compare our findings with prior work, event-related
potential (ERP) components were also calculated as the mean area
under the curve for relevant electrode sites and time windows.
Consistent with previous studies (e.g., Auerbach et al., 2015), the P1,
P2, and the early late positive potential (LPP), components were calculated
using averages across Pz, POz, P1, P2, PO3, and PO4. The P1
was quantified as the mean average from 100 to 200 ms following the
stimulus; the P2, from 200 to 300 ms; and the early LPP, from 400 to
600 ms. The late LPP was calculated as the average of Fz, FCz, and Cz,
from 600 to 1200 ms following the stimulus.
Behavioral data cleaning, modeling, and visualization was conducted
in RStudio (version 1.0.136) running R (version 3.3.2) with the
following packages: dplyr (Wickham & Francois, 2015), tidyr
(Wickham, 2015), ggplot2 (Wickham, 2009), lme4 (Bates, Mächler,
Bolker, & Walker, 2015), lmerTest (Kuznetsova, Brockhoff, &
Christensen, 2016), and compute.es (Re, 2013). EEG data were prepared
in BrainVision Analyzer 2.0 and analyzed in MATLAB via inhouse
3.1. Summary statistics and analyses of behavioral data
Behavioral SRET data are summarized in Table 2. Two-way mixed
effects ANOVAs were conducted with factors of group and valence. For
processing bias, the group × valence interaction was significant, F(1,
40) = 138.8, p < .001, generalized η2 = .78, with the MDD group
having a greater negative processing bias than the HC group, t(27.3)
= −11.21, p < .001, Cohen’s d = −3.47, 95% CI [-4.46, −2.48].
Given that processing bias is a ratio, the test for positive processing bias
return inverse results with opposite signs, and is thus not repeated here.
For each valence, participants could endorsed between 0 and 40
words. In the HC group, participants endorsed from 0 to 8 negative
words, and 12–36 positive words. In the MDD group, participants endorsed
2–30 negative words, and 8–27 positive words. In the two-way
mixed effect ANOVA, the group × valence interaction was significant,
F(1, 40) = 115.3, p < .001, generalized η2 = .62, with the MDD group
endorsing more negative words than the HC group, t(20.9) = −10.08,
p < .001, Cohen’s d = −3.12, 95% CI [-4.06, −2.19], and the HC
group endorsing more positive words than the MDD group, t(40.0)
= 5.90, p < .001, Cohen’s d = 1.83, 95% CI [1.08, 2.57]. Within the
MDD group, there was a near-significant effect of valence, with more
negative words endorsed than positive, t(33.3) = 2.02, p = .051,
Cohen’s d = 0.66, 95% CI [−0.02, 1.33]. Within the HC group, there
was a significant effect of valence, with significantly more positive
words endorsed than negative, t(27.3) = −16.8, p < .001, Cohen’s
d = −4.94, 95% CI [−6.14, −3.75].
3 Given the possibility that a linear de-trend would alter differences in longer lasting,
late ERP components, all analyses were repeated without the de-trend. Results were not
substantially different except in one test indicated in the text, below; nor did the visual
inspection of waveforms reveal substantial differences.
J. Dainer-Best et al. Biological Psychology 129 (2017) 231–241
3.2. Non-parametric statistical analysis
Grand averages were generated for HC and MDD participants separately
and used to create topographic maps (see Fig. 2). The topographic
maps were used to visualize the scalp distribution of the group
by valence differences, and how they change over time. As temporal
interval increases, consistent frontal and posterior differences develop
between groups, with the MDD group demonstrating greater differences
than healthy control participants in responses to negative versus positive
words. Central differences were especially apparent from 600 to
1000 ms, indicating periods where the HC group’s responses to negative
minus positive words were increased compared to the MDD group.
Thus, these topographic maps indicated heightened differential activation
at central sites during a time window that is consistent with the
LPP. Early time periods show relatively minor differences between
groups, primarily in frontocentral regions.
We computed permutation tests using negative word stimuli minus
positive word stimuli difference waves, comparing between groups (HC
subtracted from MDD). The permutation tests were performed over all
electrode sites following data processing, using an interval from 0 to
1000 ms post-stimulus, thus encompassing the waveforms that were
significantly different between groups in previous work described
above, including P1, P2, and the LPP.
The results of the permutation tests are displayed in Figs. 3 and 4.
The upper portion of Fig. 3 depicts the ERPs averaged across all electrodes
that showed statistically significant differences between groups.
The lower portion of Fig. 3 indicates the electrodes and time periods
where the difference waveform for negative vs. positive word stimuli
was significantly different between MDD and HC groups. Darker colors,
concentrated at left-frontal and central sites, indicate where the MDD
group showed a more negative difference waveform relative to controls;
that is, where the negative minus positive difference waveform was
more negative in the MDD group compared to controls. Permutation
tests indicated significant differences from 380 to 866 ms. Lighter
colors indicate the converse: that the MDD group showed a more positive
difference waveform, primarily in the posterior sites; these differences
were most evident in a later time-period, from 380 ms through
the end of the analysis window at 1000 ms.
To facilitate interpretation of the difference waveforms observed in
Fig. 3, we also plotted the average waveforms across frontal, central
and posterior scalp locations collapsed across all electrodes that showed
significant differences in the statistical maps (i.e., those indicated in the
lower section of Fig. 3), in response to negative and positive stimuli
separately for the MDD and HC groups. These plots are shown in Fig. 4.
Beginning around 450 ms in frontal and central sites, more positive
ERPs to positive words than to negative are evident in data from the
MDD participants, whereas HC participants show little difference between
stimuli. The posterior differences were inverted relative to the
frontal effects. Beginning around 500 ms in posterior sites, the MDD
participants showed more positive ERPs to negative words than to positive;
HC participants showed the opposite effect. This pattern over
posterior sites begins late and grows more positive over time. The
waveform is less positive in response to positive words for the MDD
group, whereas it is less positive in response to negative words for the
HC group. There were not apparent differences for either group in early
responses to positive and negative words, either between or within
3.3. Relationship between behavioral and neural outcomes
Based on the results described above, we exported timepoint-bytimepoint
voltage per participant, averaged across all spatiotemporal
clusters marked as significantly different between MDD and HC groups
in the difference waves (i.e., the mean per-group of the ERPs depicted
Behavioral data for the Self-Referential Encoding Task. Number of positive and negative
words endorsed are sums; processing biases are ratios calculated as the number of positive/negative
words endorsed over the total number of words endorsed. Recall is the
number of words of that valence recalled; self-referential recall only includes words that
were endorsed during the task. HC = healthy control; MDD = Major Depressive
Disorder; RT = reaction time.
M SD M SD
Positive Processing Bias 0.92 0.091 0.45 0.16
Negative Processing Bias 0.084 0.091 0.55 0.16
Positive Words Endorsed 25.30 6.27 14.84 5.23
Negative Words Endorsed 2.09 2.19 18.89 6.99
Positive Recall 9.78 3.81 7.79 2.86
Negative Recall 6.00 3.19 8.63 3.62
Self-Referential Positive Recall 6.74 3.26 3.32 1.92
Self-Referential Negative Recall 0.78 1.00 4.37 2.61
Positive RT 409.5 183 506.4 274
Negative RT 400.9 169 497.7 272
Fig. 2. Topographies from 0 to 1000 ms, for negative minus positive, for HC minus MDD participants. Red shading indicates where the HC group showed a greater difference between
negative and positive adjectives in a given spatiotemporal area; blue shading indicates where the MDD group showed a greater difference. HC = healthy control; MDD = Major
J. Dainer-Best et al. Biological Psychology 129 (2017) 231–241
Fig. 3. Event-related potential (ERP) differences by group. Top: ERP difference waves elicited by negative minus positive words, for MDD group (solid line) and HC group (dashed line),
recorded at frontal scalp sites (left), central (middle), and posterior (right). Stimulus presentation is indicated by the solid black line at time 0; negative voltage is plotted up. ERPs are
averaged across the electrodes that showed statistically significant differences between groups. Bottom: Color values indicate significant (p < .05) t-values for clusters comparing MDD to
HC groups across positive and negative images. Clusters are arrayed by time (x-axis) and by laterality (y-axis, with left at the bottom and right on the top). Orange (primary shading)
indicates no significant spatiotemporal difference; darker colors indicate that the MDD group showed a more negative waveform, and lighter that the MDD group showed a more positive
Fig. 4. Average Event-Related Potentials for each group, collapsed across regional electrode sites that were significantly different in the permutations tests; negative is plotted upwards.
Any electrode that showed significant differences in the permutations tests is included in its regional average. The solid line shows negative words and the dashed line positive words;
difference waves for permutations tests were calculated as negative minus positive. The MDD group is in the top row, and HC in the bottom row; plots show grand averages across subjects
in that group, for significant electrode sites in that region. In frontal and central sites, the MDD group on average shows a more negative amplitude towards negative words, whereas the
HC group shows a more negative amplitude towards positive words. In posterior sites, the trend is reversed; MDD shows a more positive amplitude trend towards negative words, whereas
the HC group shows a more positive amplitude trend towards positive words.
J. Dainer-Best et al. Biological Psychology 129 (2017) 231–241
in Fig. 3). We then examined whether the ERP response to negative
words minus positive words predicted behavioral outcomes (i.e., processing
bias and endorsements).
For ERP response predicting endorsements of word stimuli on the
SRET, a linear mixed model regression with a factor of valence (positive
or negative words) and a continuous predictor of the non-parametric
test voltage values found an interaction between valence and the voltage
values, t(78) = −2.80, p = .006, Cohen’s d = −0.86, 95% CI
[−1.51, −0.21]. A more positive ERP difference was related to an
increased number of negative words endorsed, t(39) = 2.28, p = .029,
Cohen’s d = 0.5, 95% CI [0.06, 0.95]. Conversely, a less positive ERP
difference was non-significant in predicting the number of positive
words endorsed, t(39) = −1.64, p = .11, Cohen’s d = −0.41, 95% CI
[-0.83, 0.08]. These results are consistent with the above-discussed
group effects, as MDD participants are more likely to show more positive
ERP difference waves (to negative minus positive words), especially
in posterior sites.
A final model tested whether the behavioral and ERP results were
independent predictors of depression group status, using logistic generalized
linear models with group status modeled as 1 for MDD and 0
for HC. An additive model with three predictors: positive words endorsed,
negative words endorsed, and permutations test voltage values
(rescaled with a standard deviation of 1, from 0 to 4, to result in odds
ratios [OR] that were interpretable), revealed a significant effect for
number of negative words endorsed, OR = 1.66, 95% CI [1.21, 3.49],
z = 2.04, p = .04; MDD diagnosis was 1.66 times more likely for every
additional negative word endorsed. This model had a non-significant
effect of the number of positive words endorsed, OR = 0.81, 95% CI
[0.51, 1.05], z = −1.31, p = .19 and the permutations test voltage
values, OR = 0.82, 95% CI [0.10, 5.00], z = −0.23, p = .82.
These predictors were correlated with one another but measured
separate constructs. The number of negative words endorsed was
strongly negatively correlated with the number of positive words endorsed,
r = −.63, 95% CI [−.42, −.79], p < .001; it was positively
correlated with the permutations test voltage values, r = .34, 95% CI
[.04, .59], p = .03; the number of positive words endorsed was not
significantly correlated with the permutations test voltage values,
r = −.25, 95% CI [−.52, .06], p = .11. There was a marginally-significant
positive correlation between permutations test voltage values
and score on the CESD, r = .26, 95% CI [-.05, .52], p = .09.
3.4. ANOVA: early attentional components
We performed parametric analyses of variance (ANOVA) to compare
ERP responses between groups at specific time-windows; these
waveforms are shown in Fig. 5. Area under the curve was calculated for
each component and entered into a mixed-level ANOVA, with fixed
factors of group and valence, and a random factor of participant. There
was no significant interaction of group × valence for the area under the
P1 waveform, 100–200 ms following stimulus presentation, F(1, 40)
= 0.03, p = .87, generalized η2 < .001. There were no significant
main effects of group or valence: for group, F(1, 40) = 0.02, p = .90,
generalized η2 < .001, or for valence, F(1, 41) = 0.21, p = .65, generalized
η2 < .001.
Likewise, there was no interaction of group × valence for the area
under the P2 waveform, 200–300 ms following stimulus presentation, F
(1, 40) = 0.06, p = .80, generalized η2 < .001. There were also no
significant main effects of group or valence: for group, F(1, 40) = 0.12,
p = .73, generalized η2 = .003, for valence, F(1, 41) = 0.11, p = .74,
generalized η2 < .001. These results demonstrate that for these early
components, neither group showed significant effects of valence nor
significantly differed from the other.
3.5. ANOVA: late positive potential (LPP)
For the early LPP, calculated at the same sites as the P1 and P2
components, from 400 to 600 ms, there was a significant group × –
valence interaction, F(1, 40) = 7.51, p = .009, generalized η2 = .01.4
This interaction, visualized in Fig. 5, indicates that the depressed group
exhibited greater activity following negative versus positive words and
healthy controls demonstrating the opposite pattern. This difference
(i.e., amplitudes to positive − negative stimuli between groups) was a
medium to large effect (MDD: M = −0.16 μV, SD = 0.34; HC:
M = 0.15 μV, SD = 0.38). For the comparison of difference waves,
Cohen’s d = 0.85, 95% CI [0.20, 1.50].
The late LPP was calculated across Fz, FCz, and Cz from 600 to
1200 ms. A mixed-effects ANOVA with factors of group and valence
found no group × valence interaction, F(1, 40) = 0.01, p = .92, generalized
η2 < .001. After dropping the interaction, there was a significant
effect of group, F(1, 40) = 6.09, p = .018, generalized
η2 = .13, with the depressed participants showing greater amplitudes
across both valences compared to healthy controls. These group differences
are apparent in the right portion of Fig. 5. There was also a
main effect of valence, with a greater amplitude to negative words
compared to positive words, F(1, 41) = 4.89, p = .03, generalized
η2 = .04.
3.6. Anxiety as a covariate
For three of the primary analyses reported above, we repeated the
ANOVA analyses with the Anxious Arousal subscale of the MASQ as a
covariate. For the prediction of negative processing bias, the
group × valence interaction remained significant after including anxious
arousal as a covariate, F(1, 79) = 274.2, p < .001, generalized
η2 = .78. For ERP response predicting endorsements of word stimuli,
with anxious arousal as a covariate, the interaction between word valence
and the voltage values remained significant, t(77) = −2.85,
p = .006, Cohen’s d = −0.86, 95% CI [−1.50, −0.22]. Finally, for the
early LPP (from 400 to 600 ms) as the outcome, a significant
group × valence interaction remained when including anxious arousal
as a covariate, F(1, 39) = 7.51, p = .009, generalized η2 = .01.
This study examined the electrocortical corollaries of positive and
negative self-referent processing in depression using the SRET. Using a
novel analytic technique, pointwise non-parametric randomized permutations,
we compared significant spatiotemporal differences between
MDD and healthy control groups. Our primary findings indicated
that across posterior sites, beginning immediately but becoming
strongest by 380 ms and increasing until the end of the analysis window
(1000 ms), MDD participants demonstrated more positive ERP amplitudes
in response to negative words than positive words, whereas HC
participants showed an inverse configuration. This pattern of findings
may be driven by the late positive potential (LPP), which is associated
with sustained attentional engagement and increased cognitive evaluation
of negative material in MDD (Auerbach et al., 2015;
Shestyuk & Deldin, 2010). Further, these results could not be accounted
for by the presence of anxious arousal symptoms.
Findings in the current study indicate differential cortical responses
between MDD and healthy controls over centroparietal sites that are
generally consistent with previously observed LPP responses in MDD
(Auerbach et al., 2015; Shestyuk & Deldin, 2010). As indicated in Fig. 3,
these effects are most evident across central and posterior sites in a time
period suggestive of cognitive evaluation (i.e., from approximately
350–900 ms). Further, using parametric analyses, we observed a
4 For this test only, analyses with data that had not had a linear de-trend applied during
the data processing pipeline found a different result – the interaction was no longer
significant, F(1, 39) = 2.65, p = .11, generalized η2 = .003. A main effect of valence was
also not significant, F(1, 40) = 0.87, p = .36, generalized η2 = .02.
J. Dainer-Best et al. Biological Psychology 129 (2017) 231–241
group × valence interaction for the early portion of the LPP
(400–600 ms). The depressed group exhibited greater activity following
negative versus positive words and healthy controls demonstrated the
opposite pattern. This effect replicates past work (Auerbach et al.,
2015), which found depression group differences for amplitudes to
positive versus negative words during this same time window at the
same electrode locations.
Recent research in participants with MDD has shown less positive
LPP amplitudes in response to rewarding images compared to HC participants
(MacNamara, Kotov, & Hajcak, 2016; Weinberg et al., 2016).
Our findings of reduced LPPs in response to positive stimuli in the MDD
group are consistent with these findings. Indeed, recent work indicates
that MDD affects reward-processing as well as negative informationprocessing
(Proudfit, 2015). This reduction in evaluation of reward thus
fits with the RDoC concept of depression as a failure of a positive valence
system (Cuthbert & Insel, 2013; Proudfit, 2015).
For the later portion of the LPP, the parametric analysis of variance
revealed a main effect for group (not an interaction between stimuli
condition and group), where MDD participants had more positive amplitudes
to both word valences compared to healthy controls. However,
valence differences during this time period between MDD and healthy
controls were captured in the pointwise non-parametric randomized
permutation analyses, suggesting that approach may have been more
sensitive than parametric analyses. Thus, results of the non-parametric
analyses suggest that depressed individuals show greater late-stage
cognitive evaluation of negative stimuli, which is consistent with prior
work examining the LPP in depression.
In contrast to the findings at posterior locations, non-parametric
analyses also revealed negative amplitudes in frontal and some central
regions for negative versus positive words in MDD but not HC participants,
with consistent differences appearing from approximately
380 ms–866 ms. However, the functional significance of these left
frontal differences is unclear. Additionally, interpretation of these
frontal effects must be tempered by the fact that we did not employ
recording sites over frontopolar regions.
It is possible that these negative amplitudes reflect an N2 wave
(although the N2 often appears earlier in the time course). The N2 or
N200 waveform is a scalp potential with negative polarity that is located
in frontal and central regions and appears to be linked to error
monitoring, cognitive monitoring, and response inhibition (Ramautar,
Fig. 5. Parametric event-related potential components compared between groups. Negative is plotted upwards. The solid line shows negative words and the dashed line positive words.
Shaded regions indicate the time-frame for the labeled waveforms. Early attentional components (at left; as indicated, P1, P2, and early LPP) are shown from 200 ms before stimulus
presentation to 700 ms following, and are averages across Pz, POz, P1, P2, PO3, and PO4. The late LPP can be seen at right; the waveform is plotted from 200 ms before stimulus
presentation to 1300 ms following, and is an average across Fz, FCz, and Cz. The MDD group is shown at the top, and the HC group at the bottom.
J. Dainer-Best et al. Biological Psychology 129 (2017) 231–241
Kok, & Ridderinkhof, 2004; Schmajuk, Liotti, Busse, & Woldorff, 2006).
The N2 is more negative when inhibition of response is successful in a
Stop-Signal task (Schmajuk et al., 2006). Given that in our data, MDD
participants showed a more negative amplitude during this time-frame
in response to negative words, it is possible that this reflects increased
cognitive evaluation and monitoring of the correct response to these
negative stimuli. It is also possible, however, that this waveform results
from participants inhibiting their behavioral response on the task until
they were permitted to respond. Future research will need to tease apart
these competing explanations should this pattern be replicated.
Past ERP studies have also found that early attentional responses,
including the P1 and P2, differentiated MDD from HC participants
(Auerbach et al., 2015; Shestyuk & Deldin, 2010; Waters & Tucker,
2016). However, the current study did not find these distinctions in the
nonparametric analysis, nor in parametric analysis of variance performed
on time-locked waveforms, although such differences were
hinted at in the waveforms seen in Fig. 5. These findings indicate that
there were not large differences in these stages of processing of emotional
stimuli. Instead, the results of the current study seem to indicate
that the primary differences between groups were evident in later, more
elaborative stages of processing and cognitive evaluation. Given that we
based the design of our task on past work (Auerbach et al., 2015), there
were no major methodological differences that should have resulted in
this lack of early attentional difference (in the P1 or P2) between
groups. Thus, differences across studies could be due in part to developmental
stage (adolescents in prior work versus adults in current
work), symptom severity, or other factors. Future work that examines
the ERP responses to the SRET in depressed samples across the lifespan
could address this question directly.
In this study, behavioral data from the SRET consistently showed
that participants in the MDD group but not the HC group endorsed more
negative words and fewer positive words as self-descriptive compared
to healthy controls. We also found a significant relationship between
the ERP responses and the behavioral results—most prior studies have
not linked these levels of analysis. This relationship indicated that
electrical activity within frontal, central, and posterior scalp sites across
all electrodes that showed significant differences between the MDD and
healthy control groups were predictive of the number of negative words
endorsed—i.e., that a more positive amplitude to negative words (relative
to positive words) was related to an increased endorsement of
negative words. This supports the idea that the brain responses to negative
stimuli found herein are associated with self-referential processing,
rather than simply representing an ERP response to word valence.
Moreover, this relationship falls within the negative valence system
(Cuthbert & Insel, 2013; Woody & Gibb, 2015), demonstrating how
electrocortical activity in MDD may be strongly connected to negative
self-reference. Indeed, the increased cognitive evaluation of negative
stimuli may be linked to rumination, which is common in MDD (NolenHoeksema,
2000). How rumination may differ from increased cognitive
evaluation bears further scrutiny in psychophysiological work with the
Although the non-parametric analytic techniques used in this study
were conservative, and there are several strengths to our methods, it is
important to acknowledge that our sample was relatively small, which
limits our ability to detect large effects. Larger samples could also
further examine the role of anxiety symptoms or other mental illness in
self-referent processing, or allow us to fully explore potential gender
differences in the results (although results were consistent when a
covariate of gender was added). Consistent with past work, the SRET
did not include an other-reference condition, focusing solely on selfreference.
Given the nature of the stimuli, many of the healthy participants
did not endorse many negative words, and many of the participants
in the MDD group endorsed few positive words. This makes
subdividing ERPs into cells by valence and self-reference difficult, due
to empty cells for many participants. Including neutral words in future
studies using the SRET could provide an alternative ERP difference
wave model that would provide further evidence of the presence or
absence of LPP differences between groups. Additionally, although this
study did perform a structured diagnostic interview with participants,
diagnoses were performed over the phone and not by clinicians. It is
possible that inclusion in the study would have been modified slightly
had participants instead been diagnosed by clinicians.
The current study is consistent with recent efforts in other areas of
research to develop literatures that are robust and replicable, an effort
that proves increasingly important in the current neuropsychological
landscape (Munafò et al., 2017). One important aspect of this effort is
for independent laboratories to conduct replications of prior work to
determine whether previously observed results are consistent across
settings and are robust to changes in methods or samples. The current
study attempted to replicate prior findings and indeed found support for
later stage cognitive evaluation of negative information in MDD. We
believe that conducting additional replication studies for important
clinical phenomenon is a critical direction for psychopathology research
in general. Engaging in large scale, multi-site, and pre-registered
collaborative studies should be central to this endeavor (Tackett et al.,
In summary, the current study provides evidence of increased cognitive
evaluation of negative compared to positive self-referent stimuli
in major depression, without evidence of differential early attentional
engagement. These results are evident both behaviorally and in later
posterior ERP components thought to reflect cognitive evaluation.
Negative stimuli appear to capture and sustain attention among participants
with MDD to a greater degree than positive stimuli during the
later stages of information processing. Importantly, brain responses
during the cognitive evaluation stage were also predictive of the
number of negative words endorsed, even after statistically controlling
for anxiety symptoms, indicating that this component of the ERP is
related to self-referential processing. Given these results, late-stage
event-related potentials that support biased processing of self-referential
stimuli appear to be a stable feature of depression. As such, future
work should investigate whether this processing can be ameliorated
through treatment. Results from the present study indicate that interventions
should target later, more elaborative stages of information
processing and provide important direction for identifying the brain
responses that should be targeted by such treatments in major depressive
This work was supported by awards from the National Institute on
Drug Abuse (4R01-DA03245705) and the National Institute of Mental
Health (1R56-MH10865001A1) to CGB.
Auerbach, R. P., Stanton, C. H., Proudfit, G. H., & Pizzagalli, D. A. (2015). Self-referential
processing in depressed adolescents: A high-density event-related potential study.
Journal of Abnormal Psychology, 124(2), 233–245. http://dx.doi.org/10.1037/
Auerbach, R. P., Bondy, E., Stanton, C. H., Webb, C. A., Shankman, S. A., & Pizzagalli, D.
A. (2016). Self-referential processing in adolescents: Stability of behavioral and ERP
markers. Psychophysiology, 53, 1398–1406. https://doi.org/10.1111/psyp.12686.
Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting linear mixed-effects models
using lme4. Journal of Statistical Software, 67(1), 1–48. http://dx.doi.org/10.18637/
Beck, A. T. (1967). Depression: clinical, experimental, and theoretical aspects. University of
Blackburn, I. M., Roxborough, H. M., Muir, W. J., Glabus, M., & Blackwood, D. H. R.
(1990). Perceptual and physiological dysfunction in depression. Psychological
Medicine, 20(01), 95–103. http://dx.doi.org/10.1017/S003329170001326X.
Bradley, M. M., & Lang, P. J. (2010). Affective norms for English words (ANEW): Stimuli,
instruction manual, and affective ratings (technical report c-2)Gainesville, FL: The Center
for Research in Psychophysiology, University of Florida.
Clark, L. A., & Watson, D. (1991). Tripartite model of anxiety and depression:
Psychometric evidence and taxonomic implications. Journal of Abnormal Psychology,
100(3), 316–336. http://dx.doi.org/10.1037/0021-843X.100.3.316.
Connolly, S. L., Abramson, L. Y., & Alloy, L. B. (2015). Information processing biases
J. Dainer-Best et al. Biological Psychology 129 (2017) 231–241
concurrently and prospectively predict depressive symptoms in adolescents: Evidence
from a self-referent encoding task. Cognition and Emotion, 30(3), 550–560. http://dx.
Cuthbert, B. N., & Insel, T. R. (2013). Toward the future of psychiatric diagnosis: The
seven pillars of RDoC. BMC Medicine, 11, 126. http://dx.doi.org/10.1186/1741-
Cuthbert, B. N., Schupp, H. T., Bradley, M. M., Birbaumer, N., & Lang, P. J. (2000). Brain
potentials in affective picture processing: Covariation with autonomic arousal and
affective report. Biological Psychology, 52(2), 95–111. http://dx.doi.org/10.1016/
Delplanque, S., Lavoie, M. E., Hot, P., Silvert, L., & Sequeira, H. (2004). Modulation of
cognitive processing by emotional valence studied through event-related potentials in
humans. Neuroscience Letters, 356(1), 1–4. http://dx.doi.org/10.1016/j.neulet.2003.
Derry, P. A., & Kuiper, N. A. (1981). Schematic processing and self-reference in clinical
depression. Journal of Abnormal Psychology, 90(4), 286–297. http://dx.doi.org/10.
Diagnostic and statistical manual of mental disorders (5th ed.). (2013). American Psychiatric
Disner, S. G., Beevers, C. G., Haigh, E. A., & Beck, A. T. (2011). Neural mechanisms of the
cognitive model of depression. Nature Reviews Neuroscience, 12(8), 467–477. http://
Disner, S. G., Shumake, J. D., & Beevers, C. G. (2016). Self-referential schemas and attentional
bias predict severity and naturalistic course of depression symptoms.
Cognition and Emotion, 1–13. http://dx.doi.org/10.1080/02699931.2016.1146123.
Everaert, J., Koster, E. H., & Derakshan, N. (2012). The combined cognitive bias hypothesis
in depression. Clinical Psychology Review, 32(5), 413–424. http://dx.doi.org/
Goldstein, B. L., Hayden, E. P., & Klein, D. N. (2015). Stability of self-referent encoding
task performance and associations with change in depressive symptoms from early to
middle childhood. Cognition and Emotion, 29(8), 1445–1455. http://dx.doi.org/10.
Hajcak, G., Weinberg, A., MacNamara, A., & Foti, D. (2012). ERPs and the study of
emotion. In S. J. Luck, & E. S. Kappenman (Eds.). The Oxford handbook of event related
potential components (pp. 441–474). Oxford University Press.
Hajcak, G., MacNamara, A., Foti, D., Ferri, J., & Keil, A. (2013). The dynamic allocation of
attention to emotion: Simultaneous and independent evidence from the late positive
potential and steady state visual evoked potentials. Biological Psychology, 92(3),
Holmes, A. P., Blair, R. C., Watson, J. D. G., & Ford, I. (1996). Nonparametric analysis of
statistic images from functional mapping experiments. Journal of Cerebral Blood Flow
and Metabolism, 16, 7–22. http://dx.doi.org/10.1097/00004647-199601000-00002.
Jung, T.-P., Makeig, S., Humphries, C., Lee, T.-W., Mckeown, M. J., Iragui, V., &
Sejnowski, T. J. (2000). Removing electroencephalographic artifacts by blind source
separation. Psychophysiology, 37(02), 163–178. http://dx.doi.org/10.1111/1469-
Kappenman, E. S., & Luck, S. J. (2012). ERP components: The ups and downs of brainwave
recordings. In S. J. Luck, & E. S. Kappenman (Eds.). The Oxford handbook of
event-related potential components (pp. 3–30). Oxford University Press.
Keil, A., Debener, S., Gratton, G., Junghöfer, M., Kappenman, E. S., Luck, S. J., … Yee, C.
M. (2014). Committee report: Publication guidelines and recommendations for studies
using electroencephalography and magnetoencephalography. Psychophysiology,
51(1), 1–21. http://dx.doi.org/10.1111/psyp.12147.
Kiang, M., Farzan, F., Blumberger, D. M., Kutas, M., McKinnon, M. C., Kansal, V., …
Daskalakis, Z. J. (2017). Abnormal self-schema in semantic memory in major depressive
disorder: Evidence from event-related brain potentials. Biological Psychology,
126, 41–47. http://dx.doi.org/10.1016/j.biopsycho.2017.04.003.
Kuznetsova, A., Brockhoff, P. B., Christensen, R. H. B. (2016). lmerTest: Tests in linear
mixed effects models [Computer software manual]. Retrieved from https://CRAN.Rproject.org/package=lmerTest
(R package version 2.0-32).
Lee, T.-W., Girolami, M., & Sejnowski, T. J. (1999). Independent component analysis
using an extended infomax algorithm for mixed subgaussian and supergaussian
sources. Neural Computation, 11(2), 417–441. http://dx.doi.org/10.1162/
Luck, S. J., & Hillyard, S. A. (1994). Spatial filtering during visual search: Evidence from
human electrophysiology. Journal of Experimental Psychology: Human Perception and
Performance, 20(5), 1000. http://dx.doi.org/10.1037/0096-15184.108.40.2060.
MacNamara, A., Kotov, R., & Hajcak, G. (2016). Diagnostic and symptom-based predictors
of emotional processing in generalized anxiety disorder and major depressive
disorder: An event-related potential study. Cognitive Therapy and Research, 40(3),
Munafò, M. R., Nosek, B. A., Bishop, D. V., Button, K. S., Chambers, C. D., du Sert, N. P., …
Ioannidis, J. P. (2017). A manifesto for reproducible science. Nature Human
Behaviour, 1, 0021. https://doi.org/10.1038/s41562-016-0021.
Nichols, T. E., & Holmes, A. P. (2002). Nonparametric permutation tests for functional
neuroimaging: A primer with examples. Human Brain Mapping, 15(1), 1–25. http://
Nolen-Hoeksema, S. (2000). The role of rumination in depressive disorders and mixed
anxiety/depressive symptoms. Journal of Abnormal Psychology, 109(3), 504–511.
Pernet, C., Latinus, M., Nichols, T. E., & Rousselet, G. A. (2015). Cluster-based computational
methods for mass univariate analyses of event-related brain potentials/fields:
A simulation study. Journal of Neuroscience Methods, 250, 85–93. http://dx.doi.org/
Proudfit, G. H., Bress, J. N., Foti, D., Kujawa, A., & Klein, D. N. (2015). Depression and
event-related potentials: Emotional disengagement and reward insensitivity. Current
Opinion in Psychology, 4, 110–113. https://doi.org/10.1016/j.copsyc.2014.12.018.
Proudfit, G. H. (2015). The reward positivity: From basic research on reward to a biomarker
for depression. Psychophysiology, 52(4), 449–459. http://dx.doi.org/10.1111/
Radloff, L. S. (1977). The CES-D scale: A self-report depression scale for research in the
general population. Applied Psychological Measurement, 1(3), 385–401. http://dx.doi.
Ramautar, J. R., Kok, A., & Ridderinkhof, K. R. (2004). Effects of stop-signal probability in
the stop-signal paradigm: The N2/P3 complex further validated. Brain and Cognition,
56(2), 234–252. http://dx.doi.org/10.1016/j.bandc.2004.07.002.
Re, A. C. D. (2013). compute.es: Compute effect sizes [Computer software manual].
Retrieved from http://cran.r-project.org/web/packages/compute.es.
Sanguinetti, J. L., Trujillo, L. T., Schnyer, D. M., Allen, J. J., & Peterson, M. A. (2015).
Increased alpha band activity indexes inhibitory competition across a border during
figure assignment. Vision Research, 126, 120–130. http://dx.doi.org/10.1016/j.
Santor, D. A., Zuroff, D. C., Ramsay, J., Cervantes, P., & Palacios, J. (1995). Examining
scale discriminability in the BDI and CES-D as a function of depressive severity.
Psychological Assessment, 7(2), 131–139. http://dx.doi.org/10.1037/1040-3590.7.2.
Schmajuk, M., Liotti, M., Busse, L., & Woldorff, M. G. (2006). Electrophysiological activity
underlying inhibitory control processes in normal adults. Neuropsychologia, 44(3),
Schupp, H., Cuthbert, B., Bradley, M., Hillman, C., Hamm, A., & Lang, P. (2004). Brain
processes in emotional perception: Motivated attention. Cognition and Emotion, 18(5),
Segal, Z. V. (1988). Appraisal of the self-schema construct in cognitive models of depression.
Psychological Bulletin, 103(2), 147–162. http://dx.doi.org/10.1037/0033-
Sheehan, D. V., Lecrubier, Y., Sheehan, K. H., Amorim, P., Janavs, J., Weiller, E., …
Dunbar, G. C. (1998). The Mini-International Neuropsychiatric Interview (M.I.N.I.):
The development and validation of a structured diagnostic psychiatric interview for
DSM-IV and ICD-10. Journal of Clinical Psychiatry, 59(20), 22–33.
Shestyuk, A. Y., & Deldin, P. J. (2010). Automatic and strategic representation of the self
in major depression: Trait and state abnormalities. American Journal of Psychiatry,
167, 536–544. http://dx.doi.org/10.1176/appi.ajp.2009.06091444.
Speed, B. C., Nelson, B. D., Auerbach, R. P., Klein, D. N., & Hajcak, G. (2016). Depression
risk and electrocortical reactivity during self-referential emotional processing in 8–14
year-old girls. Journal of Abnormal Psychology, 125(5), 607–619. http://dx.doi.org/
Tackett, J. L., Lilienfeld, Scott, Patrick, C., Johnson, S., Krueger, R., Miller, J., … Shrout,
P. E. (2016, December 23). Clinical Science and the Replicability Crisis. Retrieved
Trujillo, L. T., Allen, J. J. B., Schnyer, D. M., & Peterson, M. A. (2010). Neurophysiological
evidence for the influence of past experience on figure-ground perception. Journal of
Vision, 10(2), 1–21. http://dx.doi.org/10.1167/10.2.5.
Walker, W. R., Skowronski, J. J., & Thompson, C. P. (2003). Life is pleasant-and memory
helps to keep it that way!. Review of General Psychology, 7(2), 203–210. http://dx.doi.
Wardenaar, K. J., van Veen, T., Giltay, E. J., de Beurs, E., Penninx, B. W., & Zitman, F. G.
(2010). Development and validation of a 30-item short adaptation of the mood and
anxiety symptoms questionnaire (MASQ). Psychiatry Research, 179(1), 101–106.
Waters, A. C., & Tucker, D. M. (2016). Principal components of electrocortical activity
during self-evaluation indicate depressive symptom severity. Social Cognitive and
Affective Neuroscience, 11(8), 1335–1343. http://dx.doi.org/10.1093/scan/nsw046.
Weinberg, A., Perlman, G., Kotov, R., & Hajcak, G. (2016). Depression and reduced neural
response to emotional images: Distinction from anxiety, and importance of symptom
dimensions and age of onset. Journal of Abnormal Psychology, 125(1), 26. http://dx.
Wickham, H., & Francois, R. (2015). dplyr: A grammar of data manipulation [Computer
software manual]. Retrieved from https://CRAN.R-project.org/package=dplyr (R
package version 0.4.3).
Wickham, H. (2009). ggplot2: Elegant graphics for data analysis. Springer-Verlag, New
Wickham, H. (2015). tidyr: Easily tidy data with ‘spread’ and ‘gather’ functions
[Computer software manual]. Retrieved from https://CRAN.R-project.org/package=
tidyr (R package version 0.3.1).
Woody, M. L., & Gibb, B. E. (2015). Integrating NIMH research domain criteria (RDoC)
into depression research. Current Opinion in Psychology, 4, 6–12. https://doi.org/10.
J. Dainer-Best et al. Biological Psychology 129 (2017) 231–241