Content uploaded by Kentaro Katahira
Author content
All content in this area was uploaded by Kentaro Katahira on Nov 06, 2014
Content may be subject to copyright.
1
Individual differences in heart rate variability are associated with the
avoidance of negative emotional events
Kentaro Katahira1,2,3,a, Tomomi Fujimura1,3,b, Yoshi-Taka Matsuda1,3,c, Kazuo
Okanoya1,2,3,4 and Masato Okada1,3,5
Institutional affiliation:
1 Okanoya Emotional Information Project, Exploratory Research for Advanced Technology
(ERATO), Japan Science Technology Agency, Wako, Saitama, Japan
2 Center for Evolutionary Cognitive Sciences, The University of Tokyo, Meguro, Tokyo, Japan
3 RIKEN Brain Science Institute, Wako, Saitama, Japan
4 Graduate School of Arts and Sciences, The University of Tokyo, Meguro, Tokyo, Japan
5 Graduate School of Frontier Sciences, The University of Tokyo, Kashiwa, Chiba, Japan
Present address:
a. Department of Psychology, Graduate School of Environment, Nagoya University, Nagoya,
Aichi, Japan
b. Human Technology Research Institute, National Institute of Advanced Industrial Science and
Technology (AIST), Ibaraki, Japan
c. Center for Baby Science, Doshisha University, Kyoto, Japan
2
Abstract
Although the emotional outcome of a choice generally affects subsequent
decisions, humans can inhibit the influence of emotion. Heart rate variability (HRV)
has emerged as an objective measure of individual differences in the capacity for
inhibitory control. In the present study, we investigated how individual differences in
HRV at rest are associated with the emotional effects of the outcome of a choice on
subsequent decision making using a decision-making task in which emotional pictures
appeared as decision outcomes. We used a reinforcement learning model to
characterize the observed behaviors according to several parameters, namely, the
learning rate and the motivational value of positive and negative pictures.
Consequently, we found that individuals with a lower resting HRV exhibited a greater
negative motivational value in response to negative pictures, suggesting that these
individuals tend to avoid negative pictures compared with individuals with a higher
resting HRV.
Keywords: Heart rate variability; Reinforcement learning; Emotional picture; Decision
making; Motivational value
3
Introduction
Although emotion greatly influences our behaviors, humans are endowed with an
inhibitory control capacity for suppressing the influence of emotion on behavior. Heart
rate variability (HRV) has attracted growing attention as a marker of the inhibitory
capacity trait of individuals (for reviews, see Appelhans & Luecken, 2006; Thayer &
Lane, 2009). Several studies have suggested relations between individual differences in
the resting HRV (particularly parasympathetically mediated HRV) and activities in the
prefrontal cortex (Ahern et al., 2001, Lane et al., 2009; see Thayer et al., 2012 for a
meta-analysis), which is believed to play a pivotal role in such inhibitory control. The
prefrontal inhibitory (particularly the ventromedial prefrontal cortex) processes involve
tonic inhibition of the amygdala, which predominantly responds to negative
information, such as a threat or uncertainty (LeDoux, 1996). Thus, the resting HRV
likely reflects the prefrontal inhibitory control for negative emotional events. In fact,
low-HRV individuals tend to allocate their attention to threating stimuli and display
greater emotional responses (Johnsen, Thayer, Laberg, Wormnes, Raadal, Skaret, Kvale,
& Berg, 2003; Thayer & Brosschot, 2005; Melzig, Weike, Hamm, & Thayer, 2009). The
resting HRV can be used as a marker to assess the inhibitory capacity for emotional
events.
Several studies have suggested that individual differences in the resting HRV can
predict individual differences in the emotional influence on decision making. For
instance, Harlé et al. (2010) and Dunn et al. (2012) investigated the relation between the
resting HRV and the emotional influence on decision making using the ultimatum game,
which has two players: a proposer and a responder (Harlé, Allen, & Sanfey, 2010; Dunn,
4
Evans, Makarova, White, & Clark, 2012). The proposer makes an offer for how to
divide a sum of money, and if the responder accepts the offer, then the proposal is
implemented. However, if the responder rejects the offer, then neither player obtains any
money. The rejection of an unfair offer is “irrational” unless the offer is zero because
accepting any offer higher than zero yields a higher gain than receiving no money1.
[footnote 1: Although the game runs over several trials in a typical experimental setting,
each trial is assumed to be a single-shot game, and the subjects are told that the same
proposer is never matched again. Thus, a rejection cannot change the future proposer’s
attitude and does not lead to an increase in the receiver’s long-term gain.] However, the
subjects who act as responders often reject unfair offers. This seemingly irrational
choice is thought to be driven by negative emotions, such as anger or frustration, in
response to unfair treatment. In this game, individuals with a lower HRV tended to
reject unfair offers compared with individuals with a higher HRV (Harlé et al., 2010;
Dunn et al., 2012;), which suggests that inhibitory control processes promote the
rational choice. The resting HRV was also measured in conjunction with another
decision-making task, a gambling task in which the participants were asked to decide
between a “sure” option and a “gambling” option (Sütterlin, Herbert, Schmitt, Kübler,
& Vögele, 2011). The relations between the resting HRV and the “framing effect” were
examined in these experiments. The framing effect indicates that human choice depends
on the manner in which the options are presented, even when both of the options
produce the same result (e.g., “keep 40 cents” or “lose 60 cents” when the participant is
given 100 cents). Sütterlin et al. (2011) found that greater resting HRV values were
associated with a lower framing effect, suggesting that inhibitory control suppresses the
“irrational” framing effect.
5
The goal of the present study was to extend these previous findings regarding the
relation between the resting HRV and decision making in several ways. First, previous
studies examining the HRV and decision-making tasks only discussed the immediate
influence of emotion on choice. For example, in the ultimatum game, the negative
emotional response to unfair offers can only affect the decision regarding the offer itself
and not subsequent decisions because no relation exists between trials; therefore,
learning is not required. In contrast, humans and other animals often encounter
situations in which explicit information regarding the choice-outcome relation is not
given and must be learned by trial and error (Bechara, Damasio, Damasio, & Anderson,
1994; Niv, Edlund, Dayan, & O’Doherty, 2012; Yechiam, Busemeyer, Stout, & Bechara,
2005). The first and primary goal of the present study was to investigate how the
inhibitory control capacity, as measured by the resting HRV, affects the manner in which
emotions influence subsequent decisions, up to several trials later. Second, previous
studies using the resting HRV have primarily considered negative emotions rather than
positive emotions. Although inhibitory control is primarily important for suppressing
the effect of negative emotions, positive emotions might also be suppressed in some
situations. One recent study reported that individuals with a lower resting HRV
experienced a more aroused state in response to successive positive emotional images
(Fujimura & Okanoya, 2012). In the present study, we examined how the resting HRV
predicts the effect of positive (pleasant) pictures on subsequent choice behavior and
how the resting HRV predicts the effect of negative (unpleasant) pictures on such
behavior.
To achieve these goals, we administered a decision-making task using emotional
6
pictures developed in a previous study (Katahira, Fujimura, Okanoya, & Okada, 2011).
In this task, emotional pictures (both positive and negative) and neutral pictures were
presented as the outcome of a choice, and the valence of the picture was stochastically
contingent on the choices of the participants. We used a reinforcement learning model to
extract the internal parameters of the individual participants from the choice data. This
experimental and computational paradigm provides a method to quantify how and what
aspects of emotional events affect human behavior. Specifically, using the model
parameter fit, we were able to investigate which aspects of the decision were related to
the resting HRV.
To evaluate whether the observed results arose from a purely emotional component
or were related to the cognitive processes involved in the task, we also employed a
standard stochastic decision-making task using a monetary reward. The monetary
reward in this task appeared to invoke a weaker emotional reaction than the emotional
pictures for the following reasons: (1) the outcome was not related to the actual
compensation, and the participants were informed of this fact before the task. (2)
Although risky options have been used to study an emotion during a gambling task
(e.g., the Iowa gambling task), the risks of all of the options were identical in the
present task, and no “risk-taking behavior”, by definition, existed. (3) Money is a
secondary reinforcer, whereas emotional pictures are considered primary reinforcers;
thus, pictures can more directly evoke emotional responses during the task. An
additional question is whether the resting HRV is also associated with the phasic heart
rate (HR) response to emotional pictures presented during the task. Earlier studies found
a larger phasic HR deceleration in response to emotionally negative events compared
7
with neutral events (Bradley, Codispoti, Cuthbert, & Lang, 2001). This deceleration has
been considered an orienting response that is primarily mediated by the parasympathetic
nervous system (Graham & Clifton, 1966). This question is relevant for understanding
whether the vagally mediated modulation of cardiac activity, which causes an increase
in the HRV, directly affects the influences of emotion on decision making or whether
vagally mediated inhibition is only a downstream consequence of a prefrontal inhibitory
process. If we find a significant relation between the resting HRV and the phasic HR
response to emotional pictures, it would favor the former interpretation, although it
would not be conclusive evidence of causality.
Methods
Participants
The participants included 45 healthy volunteers who were recruited from the
Saitama and Tokyo areas. All of the participants were native Japanese speakers and had
normal or corrected-to-normal vision. Four participants were excluded from the analysis
because of a recording error or operational error of the physiological recording system.
Three additional participants were excluded from the analysis because these participants
made response errors on over the 5% of the trials. The data from the 38 remaining
participants (21 females; mean age = 20.05 years old, SD = 2.56, range = 18-30 years
old) were analyzed. This study was approved by the Ethics Committee of the Japan
Science and Technology Agency (JST). The participants were provided detailed
information regarding the study, and all participants provided written consent to
participate in the study. The participants also received fixed compensation that was
8
independent of their task performance for participating in the experiment.
[Figure 1 around here]
Task
The task consisted of 80 decision-picture trials (Figure 1A), 80 money trials
(Figure 1B), and 40 passive-picture trials (not shown in the figure). The passive-picture
trial was a control trial for investigating the effect of choice. For two of the choice trial
types, the participants were faced with a choice between two possible actions that were
represented by affectively neutral fractal images. Then, for the decision-picture trials, a
picture was presented whose valence (positive, neutral, or negative) was dependent on
the choice. For the money trials, a resulting monetary outcome (+500 yen, 0 yen, or -
500 yen) was presented. In the passive-picture trials, two squares marked with “?” and
“×” were presented first, and then the participants were asked to press the button
corresponding to the position of “?”. Then, a picture was presented whose valence was
randomly determined. All three of the trial types were pseudo-randomly inter-mixed
throughout the task. The choice-outcome contingency (in the decision-picture trials and
in the money trials) was as follows: For each choice trial, one fractal image represented
the “advantageous option” and was associated with positive pictures/+500 yen at a
probability of 65%, neutral pictures/0 yen at a probability of 20%, and negative
pictures/-500 yen at a probability of 15%. Another fractal image represented the
“disadvantageous option” and was associated with positive pictures/+500 yen at a
probability of 15%, neutral pictures/0 yen at a probability of 20%, and negative
pictures/-500 yen at a probability of 65%. The advantageous and disadvantageous
options were switched between two fractal images for each trial type without any cue at
9
the 20th, 35th, and 45th trials of the decision-picture trials and the 15th, 30th, and 50th
trials of the money trials. The assignment of the fractal images to the options was
counterbalanced across the participants. We referred to a block of trials from the trial
after the switch of contingency (or from the first trial) to the trial immediately prior to
the next switch as the “choice-outcome contingency block”, or simply, “block.” The
location of the fractal images was also randomized. If a response was not made within
the time limit of 1.5 s, then a response omission was indicated to the participants, and
the trial was aborted. After presenting a white frame to indicate that the choice had been
presented for 4.0 s, an outcome image (a picture or an image of the monetary outcome)
was presented. The outcome image lasted 2 s and was followed by an inter-trial interval
with a duration of 4.5±0.5 s (drawn from the uniform distribution).
For each picture category (positive, negative, or neutral), 20 pictures were
selected from the international affective picture system (IAPS; Lang, Bradley, &
Cuthbert, 2008)2. [footnote 2: The IAPS slide numbers used in this study were as
follows: 2411, 7004, 7217, 7491, 2840, 7010, 7175, 7500, 7950, 5740, 7014, 7077,
7018, 7021, 7001, 7590, 7006, 5395, 7150, and 7026 for the neutral pictures; 1440,
5199, 5910, 5994, 7470, 1710, 5833, 7502, 8031, 5301, 1920, 2655, 5890, 1410, 5829,
2314, 7508, 5825, 1720, and 5621 for the positive pictures; and 6550, 3230, 9041, 9295,
9419, 1271, 6231, 9421, 9530, 9610, 2750, 9280, 6242, 7380, 9290, 1111, 9102, 2301,
2276, and 6560 for the negative pictures.] Their normative valence/arousal ratings were
(mean±SD) as follows: 5.03±0.21/2.92±0.85 for the neutral pictures,
7.37±0.59/5.06±0.71 for the positive pictures, and 2.70±0.36/5.26±0.79 for the negative
pictures. The negative and positive pictures were selected to be equidistant from the
10
neutral pictures in terms of valence and arousal. The pictures for the decision-picture
trials and for the passive-picture trials were randomly sampled from the same set of
pictures. Prior to performing the main task, the participants experienced one training
session, which comprised 30 trials (10 trials per trial type). Prior to the training session,
the participants were instructed to select one stimulus per trial by pressing the left or
right response button. The participants were also told that a picture or an image
indicating the monetary reward would be shown after making their choices. The
participants were instructed to look carefully at the pictures to answer questions
regarding the scenes and the people in the pictures after the entire experimental session
had finished. The participants were not told which stimulus was associated with which
outcome type; however, the participants were told that one of each pair of stimuli was
associated with a higher probability of obtaining an outcome than the other and that the
probability might change without any warning. The participants were also instructed (1)
to attempt to make a choice that would show the participants a picture that they would
want to see and that would avoid showing the participants a picture that they would not
want to see during the decision-picture trial and (2) to attempt to maximize their gain
during the money trial. Although which picture the participants wanted to see could
vary depending on the participants, we made a tentative definition such that the option
that likely produced positive pictures were “advantageous” or “optimal” based on the
assumption that the participants would prefer seeing positive to negative pictures. This
assumption is justified by the subjective valence rating reported in the Results3.[footnote
3: Katahira et al. (2011) also demonstrated that positive pictures can be appetitive and
negative pictures can be aversive when they appear as outcomes of decision making. In
contrast, the previous studies that employed a free-viewing paradigm demonstrated that
11
participants viewed positive and negative pictures equally long if they were provided
the opportunity to terminate picture presentation (Lang, Greenwald, Bradley, & Hamm,
1993). A discussion regarding this seemingly contradictory difference can be found in
Katahira et al. (2011).] The participants underwent two sessions, each consisting of 120
trials. There was a break in between the two sessions to allow the participants to rest.
Apparatus and physiological recording
The experimental events were controlled by a program written using
Presentation ver. 14.1 software (Neurobehavioral Systems, Inc., Albany, CA, USA). The
visual stimuli were presented on a 19-inch LCD monitor (E1902S, Iiyama; 1024×768
pixels, 75-Hz refresh rate). The electrocardiogram (ECG) was recorded using a data
acquisition system (MP150 system; BIOPAC Systems Inc., Goleta, CA, USA) with
electrodes placed in a Lead II configuration. The ECG signals were 0.5- to 35-Hz
bandpass-filtered, amplified using a BIOPAC amplifier (ECG100C; BIOPAC Systems
Inc., Goleta, CA), and digitized at a sampling rate of 1,000 Hz. We also recorded the
electrodermal activity (EDA), the electromyographic (EMG) activity and the respiration
rate for constructing a benchmark data set for another study, with the aim of developing
a computational method to predict choice behavior from physiological activities.
However, because the main interest of the present study was cardiac activity, we only
analyzed the ECG data for the present study.
12
Procedure
The experiment was conducted individually in an electronically shielded and
sound-attenuated room. All of the participants completed the informed consent form and
successfully participated in the experiment. To calculate the HRV during a resting
period, an ECG was measured for 5 min during the resting state. The participants were
asked to relax while sitting in a chair but not to close their eyes to avoid falling asleep.
After measuring the resting HRV, a task instruction was provided. Then, the training
session and the main task were conducted sequentially. After completing the
experimental task, the participants rated all of the pictures for valence on a scale from 1
(most unpleasant) to 9 (most pleasant) using a paper-based questionnaire. Regarding
this valence rating, the participants were asked to rate how the images made them feel
during the decision-making experiment. For convenience, the scale was converted from
-4 (most unpleasant) to 4 (most pleasant) for the analysis. A recognition test in which
the participants were asked whether each picture had appeared in the task was also
conducted. All 60 pictures we prepared were presented on the paper-based questionnaire
sheet. The pictures that were presented were determined randomly and depended on the
actual participants’ performances. The mean fraction of pictures actually presented in
the task was 0.84 (SD = 0.04, maximum = 0.92, and minimum = 0.72). The mean
correct hit rate across the participants was 0.97 (SD = 0.037, minimum = 0.85), and the
mean correct rejection rate was 0.98 (SD = 0.057, minimum = 0.71), suggesting that all
of the participants satisfactorily paid attention to the picture stimuli.
13
Physiological data analysis
The inter-beat intervals (IBIs) were derived from the ECG signals using
Acknowledge 4.1 software (BIOPAC Systems Inc., Goleta, CA, USA). The IBIs were
evaluated using an automatic threshold method and corrected when the R-wave triggers
were misplaced. To evaluate the resting HRV, the root mean square successive
difference (RMSSD; Task Force of the European Society of Cardiology and the North
American Society of Pacing Electrophysiology, 1996), which is a common measure for
the resting HRV, was calculated according to the IBI for a 5-min resting period for each
participant.
The phasic HR responses in beats per minute (bpm) during each task were
quantified relative to the pre-presentation 1-s baseline and characterized according to
two phases: the initial deceleration and the late peak acceleration (Bradley et al., 2001).
The initial deceleration was calculated as the minimum value in the 0- to 3-s interval
after the picture onset, and the peak acceleration was the maximum value in the 4- to 6-s
interval after the picture onset (Bradley et al., 2001). We analyzed the HR of the
decision-picture trials and that of the passive-picture trials and divided the participants
into a higher-HRV group and a lower-HRV group based on the median of the RMSSD.
We analyzed each HR response index using a three-way ANOVA, with the two HRV
levels (the lower-HRV group or the higher-HRV group) as the between-subjects factor
and with the picture valence (positive, neutral, or negative) and the trial type (decision-
picture trial or passive-picture trial) as the within-subject factors. The repeated-
measures ANOVAs were conducted using the package “anovakun v.4.6.1” under R
3.0.1. Post-hoc multiple comparisons were conducted using Shaffer’s modified
14
sequentially rejective Bonferroni procedure.
Model and parameter fit
To model the choice behaviors of the participants, we employed the Q-learning
model, which is a standard reinforcement learning model (Sutton & Barto, 1998;
Watkins & Dayan, 1992). The Q-learning model represents the value of each action
(selecting one fractal image) as Q-values, where Qi (t) denotes the Q-value for option i (i
= 1, 2) on trial t. The Q-values were updated according to the choice and the resulting
outcome (the outcome in this study corresponded to the picture valences or to the
monetary outcomes). Furthermore, a(t) denotes the option the participant chose on trial
t; if a(t) = i, then the Q-value corresponding to the selected option was updated as
whereas the Q-value corresponding to the unselected option did not change. Here,
represented the learning rate that determined the degree of the update,
and v(t) represented the “motivational value” for the picture presented during trial t, as
subsequently specified. Given a Q-value set, a choice was made according to the
probability of choosing option 1, which is given by the following soft-max function:
with P(a(t) = 2) = 1-P(a(t) = 1).
The model set v(t) to v(t) =
P if the picture valence on trial t was positive, v(t)
= 0 if the picture valence was neutral, and v(t) =
N if the picture valence was negative
(Katahira et al., 2011). We set the value of the neutral picture to zero. Because we did
15
not know the “motivational value” of each valence a priori,
P and
N represented free
parameters that needed to be estimated based on the participants’ choice data. We used
the model for the decision-picture trials and the money trials separately, that is, all of the
parameters and Q-values were independently used for each trial type, and the trial index
t was counted separately. For the motivational value parameters, we denoted each
parameter for the two trial types as described below. For the t-th trial of the decision-
picture trial,
and for the t-th trial of the money trial,
Because we were interested in the individual differences in the model parameters, we fit
each model parameter to each participant separately rather than pooling the data across
participants. To prevent the parameter from taking an unrealistic extreme value, we used
the maximum a posteriori approach to fit the model parameters to the choice behaviors
of the participants (Niv et al., 2012). Specifically, a beta distribution with the parameters
a = 2 and b = 2 were used as a prior for the learning rate , and a Gaussian distribution
with the parameters µ = 0 and = 25 was used for the motivational value parameter
s. It should be noted that these priors were only imposed to constrain the parameter to
a reasonable range and did not create bias toward our hypothesis. The model parameters
were optimized by minimizing the negative log posterior using the Matlab function
“fmincon”. To facilitate finding the global minimum, the algorithms were run 10 times
16
for each participant; each run was initiated from a random initial value, and the
parameter set that provided the lowest negative log posterior was selected.
Model-free analysis of choice behavior
To examine the learning performance of the participants for each decision task,
the fraction of trials in which the participants chose the advantageous option was
computed for each choice-outcome contingency block of each trial type while
discarding the first 5 trials of the block as a transient phase. Then, we performed a
statistical hypothesis test for each block of each trial type, with a null hypothesis: the
fraction is on the chance level (=0.5). We also submitted the entire fraction data to a
three-way ANOVA with the two HRV levels (the lower-HRV or higher-HRV groups) as
the between-subjects factor and with the block (first, second, third or fourth) and the
trial type (decision-picture trial or money trial) as the within-subject factors. In addition,
we computed the 95% confidence interval of the fraction of optimal choice (choosing
the advantageous option) for each trial by assuming that a choice obeys a binomial
distribution.
To measure how the valences of the previous pictures affected the next choice,
we computed the empirical probability p(switch(t)|valence(t − 1)), which is the
probability of choosing a different option in trial t than that chosen in the previous trial,
given the picture valence at trial t − 1. To examine how more than one past picture
affected choice, we also conducted a linear regression analysis (filter analysis)
according to the methods of Sugrue et al. (Corrado, Sugrue, Sebastian Seung, &
Newsome, 2005; Sugrue, Corrado, & Newsome, 2004). With the index of valence j =
17
neutral, positive, or negative, we defined the picture history by rj(t), such that rj(t) = 1 (-
1) if a participant chose option 1 (2) on trial t and the resulting picture was valence j;
however, rj(t) = 0 if the resulting picture was not valence j. The choice history was
defined by c(t) such that c(t) = 1 if a participant chose option 1, and c(t) = -1 if a
participant chose option 2 on trial t. With these quantities, the linear regression model
was provided as follows:
where kj(i) and h(i) correspond to the regression coefficients for the i-th trials into the
past, for the picture history of valence j and the choice history, respectively. M
represents the length of the choice-picture history (from the current trial to past trials)
we included, and we used M = 5. We optimized the coefficients kj(i) and h(i) so that
these coefficients minimized the sum of the squared errors between Lj(t) and the choices
made by the participants.
Results
Subjective valence rating and HRV
We examined the valence rating data (-4 = most unpleasant, 0 = neutral, and 4
= most pleasant) collected after the decision-making task. The average rating for each
picture category was 2.34 ± 0.50 (mean ± SD across the participants) for the positive
pictures, -2.29 ± 0.56 for the negative pictures, and 0.09 ± 0.27 for the neutral pictures.
An ANOVA yielded significant main effects of the picture valence category [F(2, 74)=
18
1024.99, p < 0.0001]. A post-hoc multiple comparison test revealed that the average
valence rating for all picture categories differed from each other (ps < 0.0001). No
significant correlation was found between the RMSSD (resting HRV) and the subjective
rating score (for the neutral pictures: r = -0.115, p = 0.49; for the positive pictures: r = -
0.011, p = 0.95; and for the negative pictures: r = -0.112, p = 0.50).
Task performance and HRV
The decision-making tasks were challenging because the participants should have
continued learning due to the abrupt changes in the cue assignment to the
advantageous/disadvantageous option. First, we examined how well the participants
performed the task. Figure 2A shows the time courses of the fractions of choosing the
advantageous options across the participants, with the 95% confidence intervals. If the
lower bound of the confidence exceeds 0.5 (dotted line), then the performance was
significantly above the chance level. As expected, immediately after a contingency
switched, the fractions of choosing the novel advantageous options were less than 0.5;
however, the participants improved their performance through the experience of several
trials. Consequently, the participants displayed a preference for the advantageous
option, with the exception of the second choice-outcome contingency block of the
decision-picture trial, where this tendency was weak. The fraction of the trials in which
the participants chose the advantageous option for each block is summarized in Table 1.
With the exception of the second block of decision-picture trials, the participants
displayed a significant preference for the advantageous option. Next, we examined the
relation between task performance and the RMSSD, which is a measure of the resting
19
HRV. The mean value of the RMSSD during the 5-minute resting period across the
participants was 53.47 ms (SD = 36.58). We divided the participants into two groups (a
lower-HRV group and a higher-HRV group) based on the median value of the RMSSD
(42.15 ms). The mean of the RMSSD was 26.78 ms (SD = 9.51) for the lower-HRV
group and 78.12 ms (SD = 38.72) for the higher-HRV group. A three-way ANOVA with
the two HRV levels as the between-subjects factor and the choice-outcome contingency
block and trial type (decision-picture trial or money trial) as the within-subject factors
revealed a main effect of the HRV level [F(1, 36) = 4.927, p < 0.05], which suggests the
lower-HRV individuals tended to choose advantageous options throughout the task
(Figure 2B). No main effect of the trial type [F(1, 36) = 2.294, p = 0.137] was observed.
A significant main effect of the block [F(3, 108) = 4.036, p < 0.01] was observed. A
post-hoc multiple comparison test indicated that the participants chose the advantageous
option less for the second block than the first or fourth blocks (ps < 0.05). No significant
interactions were observed.
[Table 1 around here]
[Figure 2 around here]
Model parameters and HRV
Several possible reasons explain why the lower-HRV individuals tended to choose
advantageous options more frequently. Because the choice of the advantageous option
likely produces more positive events and fewer negative events, at least three
possibilities in the underlying computing processes exist: first, the lower-HRV
individuals might easily learn from positive outcomes compared with the higher-HRV
20
individuals; second, these individuals might easily learn to avoid negative events; and
third, their learning ability might be higher for the overall task, irrespective of the
outcome. To determine which possibility best explains the data, we applied
reinforcement learning models to the participants’ choice data and estimated the
processes underlying the trial-by-trial learning dynamics (Daw, 2009). If the first
possibility were correct, then the motivational value of the positive events would
correlate with the HRV. If the second possibility were correct, then the negative
motivational value of the negative events would correlate with the HRV. In contrast, if
the third possibility were correct, then both motivational values would have relation
with the HRV, or the learning rate, which determines the magnitude of the value update
resulting from a single trial, would inversely correlate with the HRV.
Figure 2A shows the model prediction for the fraction of optimal choice (gray
solid lines; averaged across the participants). We can confirm that the model fit well to
the choice data over the entire experiment, with all model predictions in the 95%
confidence block of the fraction of the participants’ optimal choice. The results of the
correlation between the resting HRV and the model parameters (motivational value) are
shown in Figure 3. We found a significant negative correlation between the resting HRV
and the motivational value of negative pictures (
), (r = 0.416, p = 0.009; Figure
3B). There were no significant correlations for the other parameters, including the
learning rate (r = -0.021, p = 0.902 for the decision-picture trial; r = -0.055, p = 0.741
for the money trials), the motivational value parameters for the positive pictures (r =
0.002, p = 0.990; Figure 3A) or the monetary outcomes (r = -0.122, p = 0.466 for gain; r
= 0.001, p = 0.993 for loss; Figure 3C, D). To verify that the result was not influenced
21
by outliers, we conducted two additional analyses by removing potential outliers. The
first analysis detected outliers solely based on the distribution of the RMSSD using the
modified Thompson τ method (p < 0.01); consequently, five samples (participants) with
the highest RMSSD were classified as outliers. The second analysis computed the
relative jackknife influence function for each sample; thus, samples that might have an
extraordinary influence on the correlation coefficient (r) were detected (Efron, 1992).
Consequently, two samples (shown by circles in Fig. 3B) were classified as outliers. The
correlation between the RMSSD and the motivational value of negative pictures
remained significant after removing potential outliers, which were detected using either
the first method (r = 0.432, p = 0.012) or the second method (r = 0.461, p = 0.005).
These results suggest that individuals with a lower HRV are more susceptible to
negative pictures. Thus, negative pictures strongly induced avoidance behavior in
subsequent choices for individuals with lower HRV compared with those individuals
with higher HRV.
Detailed behavioral data analysis
[Figure 4 around here]
The above results indicated that individuals with a lower HRV displayed a
stronger tendency to avoid negative pictures compared with those individuals with a
higher HRV. Next, we investigated in greater detail which aspects of behavior were
reflected in this motivational value parameter. The first assumption was that lower-HRV
individuals may tend to switch their choice immediately after the negative picture
appeared. However, perhaps counter-intuitively, no correlation was observed between
22
the switching probabilities and the RMSSD (r = -0.049, p = 0.771). Another possibility
that may have explained the difference in the influence of negative pictures may be
related to trials that occurred more than one trial into the past, in which the negative
picture was presented. To evaluate these possibilities, we performed a separate
regression analysis (filter analysis) for each participant’s data in the decision-picture
trials ( Sugrue et al., 2004; Corrado et al., 2005). Figure 4 plots the regression
coefficients (kj(i)) averaged across the participants. Larger absolute values of the
coefficient are associated with larger influences of the picture valence of the past trial on
the current choice. The lower plot indicates the correlation between the regression
coefficients and the RMSSD across participants. Although none of the correlation
coefficients reached significance, we observed a mild positive correlation for values
from 2 trials back to 5 trials back (r = 0.247, 0.173, 0.191, and 0.110, respectively).
Because the motivational value in the reinforcement learning model represents the total
impact of several recent trials, the correlation between the motivational value for
negative pictures and the HRV is conceivably due to the impact of recent trials rather
than due to the immediate impact. Because the regression model included more
parameters (10) than the reinforcement learning model (3), the regression analysis had a
weak detection power, which resulted from an estimation error because of the relatively
small trial size (80 trials).
Phasic heart rate response to emotional pictures
[Table 2 around here]
[Figure 5 around here]
23
Next, we analyzed the effect of the trial types, valence, and HRV-level on the
phasic HR response to decision-outcomes. The HR response to emotional pictures is
characterized by two phases: an initial deceleration and a late acceleration (Bradley et
al., 2001). This pattern is shown in Figure 5A and B, which display the average
waveform of the HR. A similar pattern was observed for the money trial (Figure 5C);
however, we only analyzed the phasic HR response in two picture trial types because we
were interested in determining whether vagally mediated inhibition, which is quantified
by the HRV, affects the peripheral response to emotional outcomes.
The mean values of the initial deceleration and the late acceleration for each
HRV group are shown in Table 2. It should be noted that because these measures are
calculated as the minimum/maximum in the intervals (0-3 and 4-6 s, respectively), these
measures tend to have greater value than the averaged waveforms shown in Figure 5.
For the initial deceleration, no main effect of the trial type, i.e., decision-picture trial
compared with passive-picture trial [F(1, 36) = 2.235, p = 0.144], was observed, and no
interaction was found between the trial type and valence [F(2, 72) = 1.317, p = 0.274];
these findings suggest that the decision-making process had no major effect on the
phasic HR. The picture valence had a significant main effect [F(2, 72) = 22.630 and p <
0.001]. A post-hoc multiple comparison test indicated that greater deceleration occurred
after the negative pictures compared with the neutral pictures (p < 0.001) and the
positive pictures (p < 0.001). A marginal difference was observed between the positive
pictures and the neutral pictures (p = 0.079). The HRV group had a marginal main effect
[F(1, 36) = 3.458, p = 0.071], which suggests the high-HRV individuals tended to
exhibit greater deceleration compared with the low-HRV individuals. This finding might
24
be a trivial consequence, considering that we took the minimum of the HR change;
greater HRV would likely induce a greater difference in the heart rate. No significant
interactions were observed between the HRV group and the trial types or valence.
Additionally, for the late acceleration in the HR, no main effect of the trial type [F(1,
36) = 2.087, p = 0.158] and no interaction between the trial type and valence [F(2, 72) =
0.827, p = 0.442] were observed. The picture valence had a significant main effect [F(2,
72) = 18.674, p < 0.001], and a smaller acceleration occurred after the negative pictures
compared with the neutral (p < 0.001) or positive (p = 0.001) pictures. The HRV group
had no main effect [F(1, 36) = 0.596, p = 0.445]. Taken together, the present data did
not provide evidence that the individual differences in the resting HRV were related to
the effects of emotion on the HR response; however, the phasic HR response was
sensitive to the valence of the pictures, which is consistent with a previous report
(Bradley et al., 2001).
Discussion
In the present study, we investigated how individual differences in the HRV
were associated with the influences of the emotional outcome on learning during
decision making. Our main finding was that individuals with a lower HRV demonstrated
a larger negative motivational value for negative pictures, which indicates these
individuals tended to avoid negative pictures compared with higher-HRV individuals.
However, a similar relation was not observed between the HRV and the subjective
valence reports, in accordance with a previous study (Fujimura & Okanoya, 2012). This
discrepancy may be explained by the difference between the influence of emotional
25
events on subsequent behavior and the subjective feelings regarding emotional events.
In our previous study (Katahira et al., 2011), we found that the estimated motivational
value (influence on the subsequent decision) was asymmetric between positive and
negative pictures such that the negative motivational value of negative pictures was
larger in magnitude than the positive motivational value of positive pictures. However,
this asymmetry was not observed in the valence for subjective ratings. Given the
increasing notion that individual differences in the HRV are associated with a capacity
for prefrontal inhibitory control, the inhibitory process may work to suppress the
influence of the negative emotion on subsequent decisions but not on the subjective
feeling concerning the emotion. The systems that affect decision and subjective
evaluation may be at least partly segregated.
The lack of a correlation between the HRV and the model parameters for the
money trials suggested that our results concerning the relation between the HRV and the
motivational value of negative pictures were primarily related to the emotional
components rather than to the cognitive components, such as strategic reasoning for
avoiding negative consequences. Although decision-making tasks using monetary
rewards have been commonly used to investigate emotions (Bechara et al., 1994),
unlike previous studies, our monetary rewards were considered to induce less emotion
for the reasons discussed in the Introduction.
By analyzing the phasic HR response, we asked whether the vagally mediated
suppression of the HR in response to emotional pictures was related to the resting HRV.
We did not find any significant relation between the resting HRV and the valence effect
on the phasic HR response, whereas the valence effect itself was observed in both
groups, suggesting that emotional pictures were effective for inducing inhibitory cardiac
26
control in both of the groups. Our picture presentation duration was shorter (2 s) than a
typical study that investigated HR modulation to emotional pictures (6 s; Bradley et al,
2001). In contrast, the duration was longer than in Codispoti, Bradley, & Lang (2001),
which reported that the emotional content of briefly presented picture (500 ms) did not
affect HR modulation. The result in the present study demonstrating the significant
effect of valence on HR modulation even after the picture vanished (4-6 s) suggested
that the presentation duration of emotional pictures were sufficient to induce an
emotional response. The null result regarding the interaction between the HRV level
and the valence effect may have been due to a lack of statistical power because of the
small sample size (n = 19 per group); however, there is a possibility that the individual
differences in inhibitory control capacity measured by the HRV were not directly
associated with the online control of cardiac activity in response to emotional events.
Growing evidence suggests that individual differences in the HRV are associated with
the activity of the medial/ventromedial prefrontal cortex (mPFC, vmPFC, for a review,
see Thayer et al., 2012). Thus, it is possible that the degree of this inhibition affected
our behavioral results and was reflected in the resting HRV, whereas the vagally
mediated suppression of cardiac activity was not a direct cause of the individual
differences in choice behavior. However, further studies are required to determine
whether HR modulation by the vagal nervous system is only a downstream consequence
of prefrontal inhibitory processes or whether this modulation causally affects the
decision-making process through afferent feedback to the brain.
The objective of the decision-making task for our participants was clear for the
money trials (“maximize your gain”); however, the objective was rather ambiguous for
the decision-picture trials (“attempt to make a choice so that you can see a picture that
27
you want to see and attempt to avoid seeing a picture that you do not want to see”).
Therefore, the participants had no explicit requirement to attempt to suppress the
emotional response to negative pictures in the decision-picture trials. Consequently, if
the individual differences in the HRV reflect the degree of inhibitory control, then this
process was likely an automatic, unintentional suppression rather than an intentional
one. Consistent with this interpretation, the prefrontal cortex was shown to be activated
when emotion suppression was not explicitly required, and the activity of the amygdala
was negatively correlated with the prefrontal cortex (Hariri, Bookheimer, & Mazziotta,
2000).
In contrast to the negative pictures, we did not find any relation between the
HRV and any of the parameters for the positive pictures. Three possibilities may explain
this lack of any detectable relation. First, the inhibitory control capacity reflected in the
HRV may only be related to the negative emotional responses. Second, automatic
inhibition may not work for positive emotions, at least for the emotions induced by the
positive pictures that we used. Third, the nervous systems connecting positive emotional
stimuli and appetitive behavior (e.g., the nucleus accumbens) may not be under the
control of the inhibitory process of the prefrontal cortex. A neuroimaging study would
be helpful to discriminate between these possibilities.
The detailed analysis of behavior using the linear regression model suggested
that the differences between the lower-HRV individuals and the higher-HRV individuals
were related to the extent of the influence of negative pictures from several previous
trials rather than those negative pictures from the immediate past only. This result
suggests that inhibitory control capacities that are measured by the resting HRV affect
28
the ability to learn to avoid negative behavior rather than only a simple win-stay and
lose-switch strategy. The reinforcement learning model is suitable for capturing
individual differences in such learning processes. To conclude, our experimental and
modeling paradigm, combined with the measurement of the resting HRV, represents a
valuable tool for measuring how an emotional event can influence the higher-level
behavior of individuals, such as decision making, rather than simple emotional reactions
to events.
References
Ahern, G.L., Sollers, J.J., Lane, R.D., Labiner, D.M., Herring, A.M., Weinand, M.E., Hutzler, R., Thayer,
J.F., 2001. Heart rate and heart rate variability changes in the intracarotid sodium amobarbital (ISA)
test. Epilepsia 42, 912–921.
Appelhans, B. M., & Luecken, L. J. 2006. Heart rate variability as an index of regulated emotional
responding. Review of General Psychology 10, 229-240.
Bechara, A., Damasio, R., Damasio, H., & Anderson, S. W. 1994. Insensitivity to future consequences
following damage to human prefrontal cortex. Cognition 50, 7–15.
Bradley, M. M., Codispoti, M., Cuthbert, B. N., & Lang, P. J. 2001. Emotion and motivation I: Defensive
and appetitive reactions in picture processing. Emotion 1, 276–298. doi:10.1037//1528-
3542.1.3.276
Codispoti, M., Bradley, M. M., & Lang, P. J. 2001. Affective reactions to briefly presented pictures.
Psychophysiology 38, 474-478.
Corrado, G. S., Sugrue, L. P., Sebastian Seung, H., & Newsome, W. T. 2005. Linear-Nonlinear-Poisson
Models of Primate Choice Dynamics. Journal of the Experimental Analysis of Behavior 84, 581–
617. doi:10.1901/jeab.2005.23-05
29
Daw, N. D. 2011. Trial-by-trial data analysis using computational models. In M. R.Delgado, E. A. Phelps,
& T. W. Robbins (Eds.), Decision making, affect, and learning:attention and performance XXIII
(pp. 3–38). Oxford, UK: Oxford University Press.
Dunn, B. D., Evans, D., Makarova, D., White, J., & Clark, L. 2012. Gut feelings and the reaction to
perceived inequity: The interplay between bodily responses, regulation, and perception shapes the
rejection of unfair offers on the ultimatum game. Cognitive, Affective, & Behavioral Neuroscience
12, 419-429.
Efron, B. 1992. Jackknife-after-Bootstrap Standard Errors and Influence Functions. Journal of the Royal
Statistical Society. Series B, 54, 83–127.
Fujimura, T., & Okanoya, K. 2012. Heart Rate Variability Predicts Emotional Flexibility in Response to
Positive Stimuli. Psychology, 3. 578–582.
Graham, F. K., & Clifton, R. K. 1966. Heart-rate change as a component of the orienting response.
Psychological Bulletin, 65(5), 305.
Hariri, A. R., Bookheimer, S. Y., & Mazziotta, J. C. 2000. Modulating emotional responses: effects of a
neocortical network on the limbic system. Neuroreport, 11(1), 43–48.
Harlé, K. M., Allen, J. J. B., & Sanfey, A. G. 2010. The Impact of Depression on Social Economic
Decision Making. Journal of Abnormal Psychology, 119(2), 440–446.
Johnsen, B. H., Thayer, J. F., Laberg, J. C., Wormnes, B., Raadal, M., Skaret, E., Kvale, G., & Berg, E.
2003. Attentional and physiological characteristics of patients with dental anxiety. Journal of
Anxiety Disorders, 17, 75-87. doi:10.1016/S0887-6185(02)00178-0
Katahira, K., Fujimura, T., Okanoya, K., & Okada, M. 2011. Decision-Making Based on Emotional
Images. Frontiers in Psychology, 2, 311. doi:10.3389/fpsyg.2011.00311
Lane, R.D., McRae, K., Reiman, E.M., Chen, K., Ahern, G.L., Thayer, J.F., 2009. Neural correlates of
heart rate variability during emotion. Neuroimage 44, 213–222.
Lang, P. J., Greenwald, M. K., Bradley, M. M., & Hamm, A. O. 1993. Looking at pictures: Affective,
facial, visceral, and behavioral reactions. Psychophysiology 30, 261-273.
30
Lang, P. J., Bradley, M. M., & Cuthbert, B. N. 2008. International Affective Picture System ( IAPS ):
Affective ratings of pictures and instruction manual. Technical Report A-8. Technical Report A-8.
Gainesville, FL.
LeDoux, J., 1996. The Emotional Brain. Simon and Schuster, New York.
Melzig, C., Weike, A., Hamm, A., & Thayer, J. 2009. Individual differences in fear-potentiated startle as
a function of resting heart rate variability: Implications for panic disorder. International Journal of
Psychophysiology 71, 109-117. doi:10.1016/j.ijpsycho.2008.07.013
Niv, Y., Edlund, J. a, Dayan, P., & O’Doherty, J. P. 2012. Neural prediction errors reveal a risk-sensitive
reinforcement-learning process in the human brain. Journal of Neuroscience 32(2), 551–62.
doi:10.1523/JNEUROSCI.5498-10.2012
Sugrue, L. P., Corrado, G. S., & Newsome, W. T. 2004. Matching behavior and the representation of
value in the parietal cortex. Science 304, 1782–7. doi:10.1126/science.1094765
Sütterlin, S., Herbert, C., Schmitt, M., Kübler, A., & Vögele, C. 2011. Frames, decisions, and cardiac-
autonomic control. Social Neuroscience 6, 169–77. doi:10.1080/17470919.2010.495883
Sutton, R. S., & Barto, A. G. 1998. Reinforcement learning: An introduction (Vol. 1). Cambridge:
Cambridge Univ Press.
Task Force of the European Society of Cardiology and the North American Society of Pacing and
Electrophysiology, 1996. Heart rate variability: standards of measurement, physiological
interpretation, and clinical use. Circulation 93, 1043–1065.
Thayer, J. F., Åhs, F., Fredrikson, M., Sollers III, J. J., & Wager, T. D. 2012. A meta-analysis of heart
rate variability and neuroimaging studies: implications for heart rate variability as a marker of stress
and health. Neuroscience & Biobehavioral Reviews 36, 747-756.
Thayer, J. F., & Brosschot, J. F. 2005. Psychosomatics and psychopathology: Looking up and down from
the brain. Psychoneuroendocrinology 30, 1050-1058. doi:10.1016/j.psyneuen.2005.04.014
Thayer, J. F., & Lane, R. D. 2009. Claude Bernard and the heart-brain connection: further elaboration of a
model of neurovisceral integration. Neuroscience and Biobehavioral Reviews 33(2), 81–8.
doi:10.1016/j.neubiorev.2008.08.004
31
Watkins, C. J. C. H., & Dayan, P. 1992. Q-learning. Machine Learning 8, 279–292.
Yechiam, E., Busemeyer, J. R., Stout, J. C., & Bechara, A. 2005. Using cognitive models to map relations
between neuropsychological disorders and human decision-making deficits. Psychological Science
16, 973–8. doi:10.1111/j.1467-9280.2005.01646.x
Acknowledgments
This work was supported, in part, by funding from the Japan Science and Technology
Agency, Exploratory Research for Advanced Technology, the Okanoya Emotional
Information Project, and the Grants-in-Aid for Scientific Research (KAKENHI) grants
24700238 and 26118506.
32
Figures
Figure 1. Schematic of the tasks of the decision-picture trial and the money trial. The
participant chose one of two fractal images and indicated his or her choice by pressing a
corresponding key. After the choice, a white frame surrounding the chosen stimulus was
presented for 4 s. Then, for a decision-picture trial (A), a picture whose valence (neutral,
positive, or negative) depended on the choice was presented for 2 s. For a money trial (B),
a resulting monetary outcome (0 yen, +500 yen, or -500 yen) was presented for 2 s.
33
Figure 2. The learning time courses of the participants’ performance and the model
prediction. (A) The black line depicts the fractions of choosing the advantageous options
across all participants. The broken lines indicate the 95% confidence interval of the
fraction. The gray solid line depicts the model prediction for the fraction of optimal
choices. The vertical broken lines indicate the points at which contingencies changed
(thus, the advantageous options flipped). (B) The fractions of choosing the advantageous
options for each HRV-group. The convention is identical to that in (A).
34
Figure 3. Relations between the RMSSD (a measure for the HRV) and (A) the
motivational value for the positive picture (
), (B) the motivational value for the
negative picture (
), (C) the motivational value for the gain in money during the trials
(
) and (D) the motivational value for the loss in money during the trials (
).
The regression slope and correlation coefficients are shown for each figure. As an
additional analysis, Jackknife outlier detection was applied to the motivational value for
the negative picture (B) where a significant correlation was found, and the potential
outlier data points (subject) are shown using open circles (see main text).
35
Figure 4. Dependencies of the choice behavior on history. (A) The mean regression
coefficients for the history of each picture valence. The larger the absolute value of the
coefficient is, the larger is the influence of the picture valence of the past trial on the
current choice. The error bars indicate the standard error. ** p < 0.01 (different from zero,
t-test with Bonferroni correction) (B) The relations between the RMSSD and individual
regression coefficients for negative pictures.
36
Figure 5. Averaged waveforms for the phasic heart rate change in response to emotional
pictures of the decision-picture trial (A) and of the passive-picture trial (B) and to
monetary feedback in the money trial (C). The top panels show the waveforms averaged
over the lower-HRV group, whereas the bottom panels show the waveforms averaged
over the higher-HRV group. Heart rate changes (bpm) were taken as the differences from
the average of the pre-presentation 1-s baseline.
37
Tables
Table 1.
Fractions of choosing the optimal options for each block.
The fraction was calculated excluding the first five trials for each block. Standard
deviations are reported in parentheses. * p < 0.05, ** p < 0.01 (different from the chance
level (=0.5), t-test, uncorrected)
Mean values for the cardiac response to emotional pictures during the
Blocks (trial indices)
All participants (n = 38)
Lower-HRV group (n = 19)
Higher-HRV group (n = 19)
Decision-picture trials
1st (1-19)
0.69 (0.21)**
0.76 (0.17)**
0.63 (0.22)*
2nd (20-34)
0.55 (0.27)
0.57 (0.29)
0.54 (0.25)
3rd (35-44)
0.69 (0.24)**
0.74 (0.21)**
0.65 (0.27)*
4th (45-80)
0.71 (0.22)**
0.74 (0.24)**
0.68 (0.19)**
Money trials
1st (1-14)
0.75 (0.20)**
0.77 (0.20)**
0.73 (0.20)**
2nd (15-29)
0.66 (0.24)**
0.75 (0.25)**
0.58 (0.22)
3rd (30-49)
0.73 (0.16)**
0.72 (0.16)**
0.73 (0.16)**
4th (50-80)
0.67 (0.21)**
0.69 (0.24)**
0.65 (0.18)**
38
Table 2.
Mean values for the cardiac response to emotional pictures during the decision-picture
trial and the passive-picture trial.
Note. The unit is beats per minute (bpm). Standard deviations are reported in
parentheses.
Trial type
Lower-HRV group
Higher-HRV group
Negative
Neutral
Positive
Negative
Neutral
Positive
Decision-picture
trial
Deceleration
-5.08 (2.20)
-4.12 (1.51)
-4.13 (1.66)
-6.32 (2.64)
-5.13 (1.98)
-5.06 (2.07)
Acceleration
1.34 (2.72)
2.46 (1.62)
2.61 (1.95)
2.04 (3.94)
3.18 (2.78)
3.77 (3.26)
Passive-picture
trial
Deceleration
-4.72 (2.46)
-4.23 (1.58)
-3.68 (1.73)
-6.38 (2.83)
-5.23 (2.06)
-4.75 (2.15)
Acceleration
1.42 (2.65)
2.41 (1.73)
2.57 (2.35)
0.84 (3.99)
3.28 (2.84)
3.63 (4.47)
