Stress-induced cortisol is associated with generation of non-negative interpretations during cognitive reappraisal


Thirty-six (25 female and 11 male) undergraduate or graduate students with a mean
age of 21.3 years (SD?=?1.8) participated in the present study. The participants were recruited through
poster advertisements and announcements on a university campus. Exclusion criteria
were a self-reported history of smoking, current psychiatric disorders, and use of
any medication or alcohol consumption on the day of the experiment. The study was
approved by the university ethics committee. All the participants signed a written
informed consent form and were paid for their participation.


Depressive symptoms

Depressive symptoms were measured using the Japanese version of the Center for Epidemiologic
Studies-Depression Scale (CES-D) 13], originally developed by Radloff 14]. The CES-D consists of 20 items rated on a 4-point Likert scale, with the scores
ranging from zero to 60, as in the original version. The Japanese version of the CES-D
has been demonstrated to be reliable and valid 13]. The CES-D was used to compare the levels of depressive symptoms of responders and

Negative affect

Negative affect was measured using the negative affect subscale of the Japanese version
of the Positive and Negative Affect Schedule (PANAS) 15], originally developed by Watson, Clark, Tellegen 16]. The PANAS is a self-rating scale comprising two subscales that measure positive
and negative affect. Two items in the original version were omitted from the negative
affect subscale of the Japanese version of the PANAS (PANAS-N) due to low communality
in factor analysis 15]. Thus, the Japanese version of PANAS-N consists of eight items, such as “distressed”
and “nervous”. Current negative affect was rated on a 6-point Likert scale, with the
scores ranging from 8 to 48. The Japanese version of the PANAS has been demonstrated
to be reliable and valid 15]. The PANAS-N was used to confirm that the stress induction task used in the study
enhanced negative affect.

Cortisol levels

Saliva samples were collected between 1345 and 1620 h because the diurnal variation
of cortisol is lower in the afternoon. Saliva sampling was conducted by passive drool,
i.e., participants were asked to draw saliva into their mouth for two minutes, and
then drool into a specimen tube through a 4-cm-long straw. Saliva samples were stored
in a freezer below ?20 °C between collection and assay. The cortisol level was measured
by means of an enzyme-linked immunosorbent assay (ELISA), using a commercial kit (Salimetrics,
State College, PA, USA). The inter- and intra-assay variances were 4.96 and 4.43 %,

Psychological stress induction task

A mental arithmetic task was used for psychological stress induction. Mental arithmetic
tasks have been confirmed to elicit negative affect and cortisol responses 17]. Participants were asked to serially subtract out loud 17 from 3093 for 5 min, as
quickly and accurately as possible. They calculated in front of both a video recorder
and a judge of the same sex as the participant and were informed that their task performance
was recorded by the video and evaluated by the judge. The judge asked participant
to restart the calculation in the case of miscalculations.

Experimental task

Four kinds of video clips of depression-related stressful social situations, one for
practice and three for measurement, were created for the present study. It has been
reported that viewing negative stimuli does not induce cortisol release 18], thus viewing video clips of depression-related situations would not induce cortisol
release. The scenarios of the video clips for measurement were based on the scenarios
from the Cognitive Bias Questionnaire (CBQ) 19]. The scenarios were 1) encouraged by your friends, you run for president of the campus
organization that you had joined but lost, 2) over lunch one afternoon, you talked
to a man or woman you found attractive, but the next afternoon he or she sat at another
empty table despite having noticed you, and 3) you visited a professor to discuss
a test but he ended the conversation because he was busy. Each video clip lasted for
approximately 1–3 min. They were presented on a 1.5 m?×?2 m screen placed 3.4 m away
from the participant. First, the participants were instructed to watch the video clips
passively, imagining that they were experiencing the situations for themselves. Immediately
after the offset of each video clip, the participants were asked to rate their current
depressive mood on a scale of zero (not at all depressed) to 100 (very depressed)
to check if the video clips enhanced their depressive mood. Next, they were asked
to generate as many interpretations as possible to reduce the depressive mood. Different
interpretations may be generated by the scenarios, and there may be individual differences
in which scenarios generate more interpretations in response. Because stress-induced
cortisol does not influence recall 20], the same video clips were presented to the participants twice (before and after
the stress induction task) in order to control for individual differences.

All responses were coded by two independent coders who specialize in clinical psychology
and who were not informed of the study’s hypotheses or the participants’ status. The
coding procedure was taken from Wisco and Nolen-Hoeksema 11]. All responses were presented to the coders in random order. The coders rated the
negativity of each response using a five-point scale anchored by l?=?not at all negative
and 5?=?very negative. Acceptable interrater reliability was obtained from all participants’
interpretations (scenario 1: ICC?=?.84, scenario 2: ICC?=?.84, scenario 3: ICC?=?.94).
Three responses were removed as negative interpretations due to their ratings above


An overview of the experimental procedure is shown in Fig. 1. Experiments began at 1300 h or 1500 h and lasted for approximately 90 min. On arrival,
the participants remained seated in a quiet room for 45 min. At the beginning of this
rest period, they gave written informed consent, then completed the CES-D. At +15,
+30, and +45 min with reference to the onset of the rest period, they completed the
PANAS-N and provided a saliva sample. However, we used the cortisol level at +45 min
after the onset of the rest period as the baseline cortisol level (T1) because resting
for at least 45 min before administering tasks is necessary to control for potential
confounders 21]. After the rest period, the participants viewed one practice video clip followed
by three measurement video clips, then rated their current depressive mood and generated
as many non-negative interpretations as possible. Next, a judge entered the room to
administer the mental arithmetic task. After the task, the judge left the room and
the participants remained seated for 10 min. At +0 (T2) and +10 min (T3) with reference
to the onset of the post-stress rest period, the participants completed the PANAS-N
and provided a saliva sample. Because the cortisol level was measured to allocate
the participants to either a responder or non-responder group, the timing of post-stress
saliva sampling was set to capture a rise in cortisol levels in response to a stressor,
based on the report that significant saliva cortisol elevation was observed 1–10 min
after a stressor onset, if not peak 22]. After the saliva sampling at T3, the participants again viewed the same video clips
that were presented before the stress induction task, followed by rating their current
depressive mood and generating as many non-negative interpretations as possible. The
post-stress assessment of the generation of non-negative interpretations was conducted
from 15 min after the offset of the stress induction task because the relationship
between stress-induced cortisol and enhanced emotional processing was found when measured
shortly after stress exposure 3]. Finally, the participants were were debriefed about the experiment and paid for
their participation.

Fig. 1. Overview of the experimental procedure

Data analyses

The participants were allocated post-hoc to either a responder or non-responder group
based on the cortisol response. The participants were allocated to the responder group
if their maximum cortisol level during the post-stress period (T2, T3) was higher
than the baseline cortisol level (T1). The non-responder group comprised participants
who were not included in the responder group. Age, depressive symptoms, and the sex
of the responder and non-responder groups were compared by means of the Welch’s unpaired
t-tests and Fisher’s exact test, respectively. To assess cortisol and negative affective
responses between the responder and non-responder groups, two-way Group (responder,
non-responder)?×?Time (T1, T2, T3) mixed-design analyses of variance (ANOVAs) were
used for the cortisol levels and PANAS-N scores. Due to non-normality, log-transformed
cortisol values were analyzed. However, raw data are presented in Table 1 and Fig. 2. Furthermore, the percentage increase from the baseline cortisol level (T1) to the
maximum values of the cortisol levels after stress (T2, T3) was compared between the
groups using a Welch’s unpaired t-test. To assess the associations between cortisol response and the generation of
non-negative interpretations, a two-way Group (responder, non-responder)?×?Time (pre-stress,
post-stress) mixed-design ANOVA was performed on the number of non-negative interpretations.
Because the baseline cortisol level and sex ratio were different between the responder
and non-responder groups, a one-way analysis of covariance (ANCOVA) was used, with
the number of post-stress non-negative interpretations as a dependent variable, group
as an independent variable, and sex, baseline cortisol level, and the number of pre-stress
interpretations as covariates. Pearson’s correlation analysis was used to assess the
strength of association between change in the cortisol level and change in the number
of non-negative interpretations. For all statistical analyses, the significance levels
were set to .05 (two-tailed).

Table 1. Group characteristics

Fig. 2. Cortisol levels for the responders and non-responders. Error bars indicate standard
error of the mean