# TEMPORAL DYNAMICS OF REWARD PROCESSING IN HUMANS: FROM ANTICIPATION TO CONSUMMATION

EDITED BY : Ya Zheng, Ruolei Gu and Daniela M. Pfabigan PUBLISHED IN : Frontiers in Psychology and Frontiers in Neuroscience

### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88966-074-2 DOI 10.3389/978-2-88966-074-2

### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# TEMPORAL DYNAMICS OF REWARD PROCESSING IN HUMANS: FROM ANTICIPATION TO CONSUMMATION

Topic Editors:

Ya Zheng, Dalian Medical University, China Ruolei Gu, Chinese Academy of Sciences (CAS), China Daniela M. Pfabigan, University of Oslo, Norway

Citation: Zheng, Y., Gu, R., Pfabigan, D. M., eds. (2020). Temporal Dynamics of Reward Processing in Humans: From Anticipation to Consummation. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88966-074-2

# Table of Contents

*05 Editorial: Temporal Dynamics of Reward Processing in Humans: From Anticipation to Consummation*

Ya Zheng, Ruolei Gu and Daniela M. Pfabigan


Chengkang Zhu, Jingjing Pan, Yiwen Wang, Jianbiao Li and Pengcheng Wang


Qi Li, Weine Dai, Yang Zhong, Lingxiao Wang, Bibing Dai and Xun Liu


Sónia Ferreira, Carlos Veiga, Pedro Moreira, Ricardo Magalhães, Ana Coelho, Paulo Marques, Carlos Portugal-Nunes, Nuno Sousa and Pedro Morgado

*86 Working Memory Load Enhances the Attentional Capture of Low Reward History*

Yujie Wu, Tingni Li and Zhe Qu

*95 The Impact of Callous-Unemotional Traits and Externalizing Tendencies on Neural Responsivity to Reward and Punishment in Healthy Adolescents* Yonglin Huang, Tingting Wu, Yu Gao, Yuyang Luo, Ziyan Wu, Shawn Fagan, Stephanie Leung and Xiaobo Li

*109 Attention Capture of Non-target Emotional Faces: An Evidence From Reward Learning*

Xing Zhou, Bixuan Du, Zhiqing Wei and Weiqi He

*118 Regulatory Effects of Reward Anticipation and Target on Attention Processing of Emotional Stimulation*

Yujia Yao, Yuyang Xuan, Ruirui Wu and Biao Sang

# Editorial: Temporal Dynamics of Reward Processing in Humans: From Anticipation to Consummation

Ya Zheng<sup>1</sup> , Ruolei Gu2,3 \* and Daniela M. Pfabigan<sup>4</sup> \*

*<sup>1</sup> Department of Psychology, Dalian Medical University, Dalian, China, <sup>2</sup> Key Laboratory of Behavioral Science, Institute of Psychology, Chinese Academy of Sciences, Beijing, China, <sup>3</sup> Department of Psychology, University of Chinese Academy of Sciences, Beijing, China, <sup>4</sup> Department of Behavioural Medicine, Faculty of Medicine, University of Oslo, Oslo, Norway*

Keywords: reward, anticipation, consummation, temporal dynamics, individual differences

### **Editorial on the Research Topic**

### **Temporal Dynamics of Reward Processing in Humans: From Anticipation to Consummation**

People often want what they like, and like what they want. However, this lay knowledge is inconsistent with findings in reward-related disorders (Treadway and Zald, 2011; Whitton et al., 2015; Nusslock and Alloy, 2017). For example, patients suffering from depression and schizophrenia sometimes show intact hedonic responses to pleasurable stimuli, but are less willing to expend effort to acquire rewards (Culbreth et al., 2018). Drug addiction is characterized by an excessive craving for drugs, but is rarely companied by the expected positive hedonic responses (Robinson and Berridge, 1993). Individuals with anorexia-type eating disorders have normal levels of "wanting," but reduced levels of "liking" of foods (Berridge et al., 2010). These clinical observations indicate that reward processing is not a homogenous construct but consists of two mainly successive phases, anticipation and consummation, as it unfolds over time (Rangel et al., 2008; Romer Thomsen et al., 2015). Although the dissociation between reward anticipation and consummation is well-established in seminal animal models (Berridge and Robinson, 2003), more work is needed to disentangle reward anticipation and consumption in humans. The current Research Topic includes 11 original articles portraying the dynamics of reward processing in humans in terms of self-report, behavioral, or neural changes.

In this topic, three articles focus on the role of emotions and personality traits in anticipatory and consummatory phases of reward processing. First, Li X. et al. asked participants to evaluate their daily experience of hedonic feelings by using experience sampling, and found that dysphoric college students reported less state anticipatory and consummatory pleasure compared with their non-dysphoric counterparts. Their results support the view that anhedonia leads to deficits in both anticipatory and consummatory phases of reward processing. Meanwhile, Huang et al. combined the classic Monetary Incentive Delay task (Knutson et al., 2000) with functional magnetic resonance imaging (fMRI) in a sample of adolescents. They discovered that callous-unemotional personality traits were positively correlated with ventral striatum activation in the anticipatory phase, but this effect was dependent on externalizing behavior. In the consummatory phase, externalizing behavior was negatively correlated with amygdala activation during punishment receipt even after controlling for callous-unemotional traits. These results help to clarify the relationship between psychopathic traits and antisocial behavior in dysfunctional reward processing. Finally, Ferreira et al. asked students to place bids to obtain food during fMRI recording and found that chronically stressed participants proposed lower bids than non-stressed ones, but there was no behavioral and neural differences during cognitive regulation of craving.

### Edited and reviewed by:

*Paul E. M. Phillips, University of Washington, United States*

### \*Correspondence:

*Ruolei Gu gurl@psych.ac.cn Daniela M. Pfabigan d.m.pfabigan@medisin.uio.no*

### Specialty section:

*This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Psychology*

Received: *26 June 2020* Accepted: *10 July 2020* Published: *19 August 2020*

### Citation:

*Zheng Y, Gu R and Pfabigan DM (2020) Editorial: Temporal Dynamics of Reward Processing in Humans: From Anticipation to Consummation. Front. Psychol. 11:1901. doi: 10.3389/fpsyg.2020.01901*

**5**

Two articles focus on the electrophysiological correlates of option evaluation, a critical stage of reward-based decisionmaking during which individuals evaluate and assign a value to each available option (valuation). First, Wang et al. examined event-related potential (ERP) responses of probability weight and monetary magnitude during the evaluation of a risky reward. The results showed that probability weight was encoded by the P200, the medial frontal negativity (MFN), and the late positive potential (LPP) components, whereas monetary magnitude was solely encoded by the MFN. The results demonstrated distinct temporal dynamics involved in the processing of probability weight and monetary magnitude. Meanwhile, Zhu et al. investigated both ERP and oscillatory correlates underlying the evaluation of ambiguous options using an ambiguous choice task. The authors found that delta activity was enhanced for low- vs. high-ambiguity options 200–400 ms after option onset, and for high- vs. low-reward options 400–500 ms after option onset. Ambiguity and reward information were integrated during the time window of 500–600 ms as indexed by both the P3 component and delta activity. These results help clarify neural dynamics of ambiguity vs. reward processing during option evaluation.

Four articles in this volume focus on the characteristics of reward processing in terms of its subcomponents as diverse as anticipation, learning, and consummation. Yao et al. applied an emotion (vs. sex) recognition task while participants anticipated either reward or non-reward. Their results showed that reward anticipation facilitated the processing of target information only when the target was defined by the emotional arousal of stimuli. Using a visual search task, Zhou et al. investigated the effects of prior reward learning on the processing of non-target emotional faces and found that reward history had stronger effects on fearful faces than happy faces. These two studies could further our understanding of the interaction of emotions and reward. Wu et al. investigated the impact of working memory capacity on value-driven attentional capture of reward history, and found that under the memory load condition, attentional capture of target information was more likely to be distracted by low reward-value distractors. By assessing ERPs, Yu et al. were interested in how arbitrary group membership affects the processing of reward and loss feedback in a male sample. Contrary to their expectations, the authors observed no direct support for increased in-group bias in their gambling observation task. ERP results showed that their participants employed more attentional resources during outcome processing of out-group

### REFERENCES


individuals. This suggests an enhanced need for perspective taking in these cases.

Adopting a more clinically-oriented perspective, Li Q. et al. used structural equation analysis to characterize gender differences in regards to how impulsivity, coping styles, and Behavioral Inhibition/Approach System (BIS/BAS) influence internet addition in adolescents. Emotion-focused coping mediated the relationships between impulsivity/internet addition and BIS/internet addiction in girls, while problemfocused coping strategies were mostly observed to mediate the relationships between impulsivity/internet addiction and BAS/internet addiction in boys. These findings suggest that gender-sensitive training approaches should be devised to target internet addiction in adolescents more appropriately. Based on the observation that individuals undergoing evaluation of traumatic brain injury may be malingering neurocognitive deficits for compensatory benefits, Neal et al. developed a novel neural-based method for discriminating fake (i.e., simulated) from true brain injury. The authors found that individuals simulating memory deficits were characterized by delayed left frontal neural responses during recognition of studied items, which reached sensitivity of 80% and specificity of 79% in differentiating malingered from true brain injury.

In sum, the experimental findings presented in this topic shed light on the temporal dynamics (anticipation vs. consummation) of reward processing and indicate a possibility of further decomposing reward anticipation/consummation into subcomponents with distinct theoretical significance. Future research should extend existing theoretical models of reward processing by better characterizing implicated sub-processes such as stimulus-reward associations, effort computations, feedback integration, and social context effects. By addressing these issues, we may better inform more targeted prevention of and interventions for reward-related disorders.

### AUTHOR CONTRIBUTIONS

YZ conceived the idea of the Research Topic. YZ, RG, and DP have made a substantial, direct and intellectual contribution during preparation of the editorials, and approved it for publication.

### FUNDING

This work was supported by the National Natural Science Foundation of China (31971027, 31571161, and 31671173).

disparate psychological and neural mechanisms? Psychol. Med. 48, 889–904. doi: 10.1017/S0033291717002525


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Zheng, Gu and Pfabigan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Neural Dynamics of Processing Probability Weight and Monetary Magnitude in the Evaluation of a Risky Reward

*Guangrong Wang1,2 \*, Jianbiao Li2,3,4 \*, Pengcheng Wang2,5 , Chengkang Zhu2 , Jingjing Pan2 and Shuaiqi Li2*

*1 Neural Decision Science Laboratory, Weifang University, Weifang, China, 2 Reinhard Selten Laboratory, China Academy of Corporate Governance, Business School, Nankai University, Tianjin, China, 3 Department of Economic and Management, Nankai University Binhai College, Tianjin, China, 4 School of Economics, Shandong University, Jinan, China, 5Business School, Tianjin University of Economic and Finance, Tianjin, China*

### *Edited by:*

*Ya Zheng, Dalian Medical University, China*

### *Reviewed by:*

*Liang Meng, Shanghai International Studies University, China Zachary A. Yaple, National University of Singapore, Singapore*

### *\*Correspondence:*

*Guangrong Wang grongw@126.com Jianbiao Li biaojl@126.com*

### *Specialty section:*

*This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Psychology*

*Received: 14 January 2019 Accepted: 27 February 2019 Published: 26 March 2019*

### *Citation:*

*Wang G, Li J, Wang P, Zhu C, Pan J and Li S (2019) Neural Dynamics of Processing Probability Weight and Monetary Magnitude in the Evaluation of a Risky Reward. Front. Psychol. 10:554. doi: 10.3389/fpsyg.2019.00554*

Risky decision-making involves risky reward valuation, choice, and feedback processes. However, the temporal dynamics of risky reward processing are not well understood. Using event-related brain potential, we investigated the neural correlates of probability weight and money magnitude in the evaluation of a risky reward. In this study, each risky choice consisted of two risky options, which were presented serially to separate decision-making and option evaluation processes. The early P200 component reflected the process of probability weight, not money magnitude. The medial frontal negativity (MFN) reflected both probability weight and money magnitude processes. The late positive potential (LPP) only reflected the process of probability weight. These results demonstrate distinct temporal dynamics for probability weight and money magnitude processes when evaluating a risky outcome, providing a better understanding of the possible mechanism underlying risky reward processing.

### Keywords: probability weight, money magnitude, risky choice, neural dynamics, ERP

### INTRODUCTION

Risky decision-making, which involves trade-offs between lotteries with differing magnitude and uncertainty, is ubiquitous in everyday life. Therefore, when making decisions among risky rewards, it is necessary to evaluate the subjective value of each risky reward. The subjective value of a risky reward depends on its probability and magnitude (Tversky and Kahneman, 1981; Brown and Braver, 2007). Prospect theory, an influential model of risky decision-making, suggests that the subjective value of a risky outcome depends on gains or losses relative to status quo and probability weighting function (Kahneman and Tversky, 1979; Camerer, 2000).

The neurocognitive mechanisms underlying risky decision-making involve several processes: valuation, choice, and feedback (Rangel et al., 2008; Liu et al., 2012). Previous neuroscience studies focused on the choice and feedback processes of risky decision-making, but the neural correlates for valuation of risky rewards are not well understood (see review by Chandrakumar et al., 2018). The focus of this study is on the temporal dynamics of the valuation process of risky rewards using an event-related brain potential (ERP) technique.

**8**

Neuroimaging studies have demonstrated that a number of brain regions including the ventromedial prefrontal cortex, amygdala, insula, anterior cingulate cortex (ACC), striatum, parietal, and temporal cortices are implicated in risk processing (Paulus and Frank, 2006; Berns et al., 2008; Hsu et al., 2009; Blankenstein et al., 2018). Several studies investigated the neural correlates of probability and magnitude of a risky reward. Berns and Bell (2012) measured independently the neural responses to magnitude and probability of a risky outcome by displaying serially magnitude and probability information. They found that the ventral and dorsal striatum were involved in the processes of magnitude and probability, respectively. These results demonstrate a second-order decision process, in which participants integrate judgments instead of information. In a study by Smith et al. (2009), high reward elicited more activation in several brain regions including the insula, amygdala, and posterior cingulate cortex when other parameters were held constant, as opposed to low reward. Low-probability reward induced more activation in the ACC than high-probability reward when other parameters were held constant.

Eye-tracking methodology has been used to investigate processes in risky decision-making (Rayner, 1998; Glöckner and Herbold, 2011). These studies suggested that risky decisionmaking relied mainly on automatic-intuitive processes, which were partially accounted for by automatic integration or simple heuristic models (Glöckner and Herbold, 2011; Fiedler and Glöckner, 2012; Venkatraman et al., 2014; Aimone and Ball, 2016). Eye-tracking studies have focused on risky choices in which two gambles were displayed simultaneously. Such paradigms did not allow us to distinguish valuation and choice processes. Furthermore, in real world, individuals usually face risky choice options serially.

Existing event-related brain potential (ERP) studies of risky decision-making focused on responses to risk-related decision and feedback. Only a minority of ERP studies have focused on the neural response to risky options (see review by Chandrakumar et al., 2018). Both feedback-related negativity (FRN) and P300 are two important ERP components involved in the risk process. The FRN, which is often known as reward positivity (RewP) associated with outcomes processing in the context of gains in contrast with losses (Foti et al., 2012; Proudfit, 2015; Yaple et al., 2018a,b), is larger following negative feedback relative to positive feedback (Wu and Zhou, 2009; Polezzi et al., 2010; Yang and Zhang, 2011; Yang et al., 2015; Zhao et al., 2016; Kardos et al., 2017). The P300, which is thought to reflect the outcome of stimulus evaluation and decision-making, was pronounced in response to the selection of a risky option and positive feedback (Yeung and Sanfey, 2004; Sato et al., 2005; Oberg et al., 2011; Schuermann et al., 2012; Wang et al., 2015).

While previous ERP studies have yielded important insights into the neural mechanisms of risky decision-making, there are limitations. First, risky decision-making involves valuation and choice processes, with evaluation of a risky reward most relevant to the valuation process (Rangel et al., 2008; Liu et al., 2012). Previous ERP paradigms, in which the participants' task was to decide whether or not to accept a risky bet, did not allow one to distinguish among valuation and choice processes. Furthermore, evaluation of a risky reward involves its probability weight and magnitude, but previous ERP studies did not focus on these two components.

In this study, we developed a risky choice task to investigate neural mechanisms underlying probability weight and magnitude of risky rewards based on study paradigms derived from the intertemporal choice literature (Pine et al., 2009; Xia et al., 2017), given that there are a number of similarities between delay and probability discounting (Green and Myerson, 2004; Madden and Bickel, 2010; McKerchar and Renda, 2012). In our paradigm, each choice consisted of two risky options, which were first presented serially, then presented simultaneously. This allowed us to separate risky rewards valuation and selection processes. By controlling for the effects of probability and magnitude respectively, we could explore the neural mechanisms underlying probability weight and magnitude during evaluation of a risky reward.

According to previous research, several ERP components are associated with magnitude and probability processes. Based on these, we analyzed the ERP response related to probability weight and money magnitude in the evaluation of a risky reward. Since frontal P200 may be involved in stimulus evaluation and quick assessment (Boudreau et al., 2009; Lau et al., 2013), a hypothesis, in which the frontal P200 would reflect the difference between high- and low-probability rewards, was proposed. A second evaluated component was medial frontal negativity (MFN)1 , which represents the early appraisal of feedback and is more pronounced for bad outcomes compared to good outcomes (Hajcak et al., 2006; Holroyd et al., 2006; Hewig et al., 2007; Boksem and de Cremer, 2010; Huang and Yu, 2014; Umemoto et al., 2017). In this study, when the magnitude of options was held constant, the high-probability rewards were considered "good" outcomes compared to low-probability rewards. Therefore, we predicted that the MFN would reflect the difference between high- and low-probability rewards. Similar predictions for the magnitude of risky rewards were made. Furthermore, the P300 has been shown to be sensitive to outcome evaluation, including the magnitude and valence of rewards (Goyer et al., 2008; Wu and Zhou, 2009; Harris et al., 2013; Righi et al., 2014). It is possible that the P300 would also encode the probability weight of risky rewards. Therefore, we hypothesized that the P300 or a later component would reflect the process of probability weight and money magnitude.

### MATERIALS AND METHODS

### Participants

A total of 20 right-handed undergraduates were recruited. Twelve females and eight males participated. They were 20–25 years

<sup>1</sup> The MFN is a class of ERP components which include the FRN and ERN. The MFN has been proposed to reflect a motivational/affective evaluation of negative outcomes. Both feedback and stimulus information elicit MFN. The MFN in the context of feedback-locked ERPs could reflect whether outcomes matched expectations, while the MFN in the context of stimulus-locked ERPs could reflect whether the events violated (social and non-social) expectancy (Holroyd and Coles, 2002; Goyer et al., 2008; Boksem and de Cremer, 2010; Schuermann et al., 2012). This study focused on stimulus-locked ERPs.

of age with a mean age of 22.35 (*SD* = 1.59). All participants had normal or corrected-to-normal visual acuity and no history of neurological or mental disease. All subjects signed an informed consent prior to the experiment, which was performed in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Reinhard Selten Laboratory, Nankai University. The participants received an average of 65 Chinese yuan (approximately \$10) (Krajbich et al., 2012; Li et al., 2016; Yaple et al., 2017, 2018a,b).

### Task and Stimuli

According to Prospect Theory (Kahneman and Tversky, 1979), the subjective value (*V*) of a risky gamble is given by:

$$V = \sum\_{i=1}^{n} \pi\left(p\_i\right) \mu(\mathbf{x}\_i),$$

In the present study, we focused on the evaluation of risky rewards. Therefore, each risky option consisted of a risky reward and a zero reward. Therefore, the subjective value (*V*) of a risk option was expressed as:

$$V = \pi\left(p\right) \times u\left(x\right)$$

The function *π(p)* represents the subjective probability to objective probability *p*, with *u(x)* the undiscounted utility of a reward (*x*).

This study tried to explore neural processing of probability weight and money magnitude of a risky reward. Since the subjective value of a risky reward is determined by the magnitude and probability of its receipt, an experimental paradigm was designed to allow comparison based on: (1) different probabilities but same money magnitude, and (2) different money magnitude but same probability. To obtain subjective utility related to probability, two types of stimuli were considered: winning CNY50 at the probability of 0.2 (low probability, LP) and CNY50 at the probability of 0.5 (high probability, HP). Similarly, for money magnitude, we considered two types of stimuli: winning CNY15 at the probability of 0.66 (small magnitude, SM) and CNY40 at the probability of 0.66 (large magnitude, LM). In addition, in order to improve the reality of the experiment and decrease the risk that participants will be bored, some stimuli including CNY10 by 0.99, CNY30 by 0.33, CNY60 by 0.33, and CNY20 by 0.99 were defined as filling materials.

The task of the participants was to choose between two options with different magnitude and probability of occurrence. Each option of a choice was presented serially to separate decision-making and option valuation processes. At the end of the experiment, two of the participant's choices were selected at random and used for subject payment.

### Procedure

The rules of the experimental task were instructed to the participants by explaining written instructions. The task was performed in a quiet and isolated laboratory. The participants were told that they would be paid for participation after completion of the experiment. The recording session took approximately 30 min.

After 8 practice trials, a total of 100 trials were randomly divided into 2 blocks with 50 trials each. Each trial was created through the following sequence. In each trial, a cross was first displayed in the center of a screen for 800–1,200 ms. Afterward, option 1 was presented for 1,500 ms. Then, after a cross of 800–1,200 ms, option 2 was presented for 1,500 ms. Next, the choice was displayed until a response had been made. The presentation of the two options for each type of stimuli was counterbalanced in a random order across trials. Then, their choice was shown for 1,000 ms, after which a blank screen was displayed for 1,000 ms, and then the next trial started (**Figure 1**).

### Electroencephalography (EEG) Recording and Analysis

EEGs were continuously acquired at a 1,000 Hz sampling rate with a Neuroscan Synamp2 Amplifier, by using an electrode cap with Ag/AgCl electrodes mounted according to the extended international 10–20 system. The EEG signals were amplified online (band pass: 0.05–100 Hz). All rows of electrode recordings were referenced online to the left mastoid, and they were re-referenced offline to the average of the left and right mastoid. Electrode impedance was kept under 5 kΩ. Following the electrode application, the participants sat in a comfortable chair located in a shielded room and were asked to fix on the center of the computer display located 1 m away from their eyes during the experiment.

EEG epochs of 1,000 ms (from −200 to 800 ms after the onset of stimulus) were extracted offline, and the 200-ms pre-stimulus defined as baseline. Ocular artifacts were corrected. Trials contaminated by amplifier clipping, bursts of electromyographic activity or peak-to-peak deflection exceeding ±75 μV were excluded from further analysis. The remaining trials were baseline corrected. The EEG segments were averaged separately for probability type (HP vs. LP) and magnitude type (LM vs. SM). Averaged ERPs were digitally filtered with a low-pass filter at 30 Hz. Within-subject repeated measure analysis of variance (ANOVA) were used to analyze ERP data. Behavior and ERP data were statistically analyzed using SPSS (version 22, SPSS Inc., Chicago, IL, USA). A Greenhouse-Geisser correction for violation of sphericity assumption was applied when the degrees of freedom were more than one. The significance level was set at 0.05 for all analyses. To control for family-wise error for multiple *t*-tests, *p* were Bonferroni corrected.

Based on the visual inspection of the grand-average waveforms, three components were analyzed. The frontal P200 was measured as peak amplitude between 150 and 250 ms after stimulus onset at F3, Fz, F4, FC3, FCz, and FC4 (Polezzi et al., 2008; Molinaro and Carreiras, 2010; Gui et al., 2016). The MFN component was measured as peak amplitude between 250 and 350 ms after stimulus onset at F3, Fz, F4, FC3, FCz, and FC4 (Boksem and de Cremer, 2010; Schuermann et al., 2012; Huang and Yu, 2014; Xia et al., 2017). The LPP was measured as mean amplitude between 450 and 650 ms after stimulus onset at CP3, CPz, CP4, P3, Pz, and P4 (Harris et al., 2013; Righi et al., 2014; Gui et al., 2016). ERP analyses were conducted by repeatedmeasure ANOVAs, with electrode (for P200 and MFN: F3, Fz, F4, FC3, FCz, and FC4, for LPP: CP3, CPz, CP4, P3, Pz, and P4) and probability (high, low), and electrode and magnitude (large, small), respectively.

### RESULTS

### Behavioral Results

Behavioral results are shown in **Table 1**. For the choice of CNY50 by 0.2 probabilities and CNY15 by 0.66 probabilities, 44.60% of decisions chose the former. For the choice of CNY50 by 0.5 probabilities and CNY40 by 0.66 probabilities, 38.73% TABLE 1 | Behavioral results.


of decisions chose the former. The average response time was 775.98 ms (*SD* = 611.5195) and 719.52 ms (*SD* = 413.1792), respectively. For the choice of CNY50 by 0.2 probabilities and CNY40 by 0.66 probabilities, all decisions chose the latter. For the choice of CNY50 by 0.5 probabilities and CNY15 by 0.66 probabilities, 95.24% of decisions chose the former. The average response time was 550.69 ms (*SD* = 204.8855) and 590.89 ms (*SD* = 268.96), respectively.

Participants took more response time to make decision between the choices in which the expected value of two options was similar, compared to choices in which there was large difference between the expected values of two options (*p* = 0.02). Based on formal logic, when the expected value of each option of a risky choice is similar, the higher level of conflict requires more brain resources for conflict resolution, which results in more response time.

Behavioral data showed that participants chose the option with largest expected value. This is consistent with previous studies and demonstrates that participants clearly understood the experimental task.

### ERP Results for Probability Weight P200

**Figure 2** shows ERP waveforms and topographic maps for probability processes at Fz and FCz electrodes. In the frontal area, there was a significant main effect of P200 for probability levels [*F*(1, 19) = 8.309, *p* = 0.010, *η*<sup>2</sup> = 0.304], no main effect for laterality [*F*(2, 38) = 3.899, *p* = 0.051, *η*<sup>2</sup> = 0.170], and no interaction between probability levels and laterality [*F*(2, 38) = 2.160, *p* = 0.145, *η*<sup>2</sup> = 0.102] were found. In the frontalcentral scalp area, significant main effects were found for probability levels [*F*(1, 19) = 7.586, *p* = 0.013, *η*<sup>2</sup> = 0.285] and laterality [*F*(2, 38) = 5.117, *p* = 0.017, *η*<sup>2</sup> = 0.212]. There was no interaction between probability levels and laterality [*F*(2, 38) = 2.288, *p* = 0.126, *η*<sup>2</sup> = 0.107]. High-probability rewards elicited more positive P200 than low-probability ones when the magnitude was kept constant.

### MFN

As shown in **Figure 2,** in the frontal area, significant main effects of MFN were observed for probability levels [*F*(1, 19) = 10.389, *p* = 0.004, *η*<sup>2</sup> = 0.353] and laterality [*F*(2, 38) = 5.490, *p* = 0.024, *η*2 = 0.224], but no interaction was found between probability levels and laterality [*F*(2, 38) = 1.656, *p* = 0.211, *η*<sup>2</sup> = 0.080]. In the frontal-central area, significant main effects of MFN were observed for probability levels [*F*(1, 19) = 12.067, *p* = 0.003, *η*2 = 0.388] and laterality [*F*(2, 38) = 5.443, *p* = 0.020, *η*<sup>2</sup> = 0.223].

There was no interaction between probability levels and laterality [*F*(2, 38) = 3.594, *p* = 0.053, *η*<sup>2</sup> = 0.159]. These results showed that low-probability rewards elicited more negative MFN than high-probability ones, when the magnitude was kept constant.

### LPP

**Figure 3** shows ERP waveforms and topographic maps for probability processes at Pz and CPz electrodes. In the parietal area, a significant main effect for LPP was found for probability levels [*F*(1, 19) = 17.599, *p* = 0.000, *η*<sup>2</sup> = 0.481]. There were no significant main effect for laterality [*F*(2, 38) = 2.418, *p* = 0.105, *η*2 = 0.113] and no interaction between probability levels and laterality [*F*(2, 38) = 0.032, *p* = 0.935, *η*<sup>2</sup> = 0.002]. In the central-parietal area, significant main effects for LPP was found for probability levels [*F*(1, 19) = 19.374, *p* = 0.000, *η*<sup>2</sup> = 0.505] and laterality [*F*(2, 38) = 3.884, *p* = 0.043, *η*<sup>2</sup> = 0.170], but no interaction was found between probability levels and laterality [*F*(2, 38) = 0.033, *p* = 0.934, *η*<sup>2</sup> = 0.002]. These results demonstrated that high-probability rewards elicited more positive LPP than low-probability ones, when the magnitude was kept constant.

### ERP Results for Money Magnitude P200

**Figure 4** shows ERP waveforms and topographic maps for money magnitude at Fz and FCz electrodes. In the frontal area, there were no significant P200 effect for reward magnitude [*F*(1, 19) = 0.093, *p* = 0.764, *η*<sup>2</sup> = 0.005] and no interaction between magnitude and laterality [*F*(2, 38) = 0.045, *p* = 0.921, *η*2 = 0.002]. But there was significant P200 effect for laterality [*F*(2, 38) = 7.140, *p* = 0.002, *η*<sup>2</sup> = 0.273]. In the frontal-central area, no significant P200 effects were observed for magnitude levels [*F*(1, 19) = 0.095, *p* = 0.761, *η*<sup>2</sup> = 0.005] and laterality [*F*(2, 38) = 1.716, *p* = 0.198, *η*<sup>2</sup> = 0.083]. There was no interaction between magnitude levels and laterality [*F*(2, 38) = 0.059, *p* = 0.878, *η*<sup>2</sup> = 0.003].

### MFN

As shown in **Figure 4,** in the frontal area, significant main effects for MFN were found for magnitude levels [*F*(1, 19) = 6.380, *p* = 0.021, *η*<sup>2</sup> = 0.251] and laterality [*F*(2, 38) = 13.866, *p* = 0.000, *η*2 = 0.422], but no interaction was found between magnitude levels and laterality [*F*(2, 38) = 1.461, *p* = 0.246, *η*<sup>2</sup> = 0.071]. In the frontal-central area, significant main effects for MFN were found for magnitude levels [*F*(1, 19) = 5.619, *p* = 0.029, *η*<sup>2</sup> = 0.228] and laterality [*F*(2, 38) = 9.404, *p* = 0.001, *η*<sup>2</sup> = 0.331], but no interaction was found between magnitude levels and laterality [*F*(2, 38) = 0.735, *p* = 0.454, *η*<sup>2</sup> = 0.037]. Given same probability weight, small rewards elicited more positive MFN than large ones.

### LPP

**Figure 5** shows ERP waveforms and topographic maps for magnitude at Pz and CPz electrodes. In the parietal area, no significant main effect of LPP was found for magnitude levels [*F*(1, 19) = 1.937, *p* = 0.180, *η*<sup>2</sup> = 0.093] and no interaction was found between magnitude levels and laterality [*F*(2, 38) = 1.867, *p* = 0.176, *η*<sup>2</sup> = 0.089]. There was no significant main effect for

laterality [*F*(2, 38) = 1.651, *p* = 0.208, *η*<sup>2</sup> = 0.080]. In the centralparietal area, no significant main effects of LPP were found for magnitude levels [*F*(1, 19) = 1.431, *p* = 0.246, *η*<sup>2</sup> = 0.070] or laterality [*F*(2, 38) =1.804, *p* = 0.186, *η*<sup>2</sup> = 0.087], and no interaction was found between magnitude levels and laterality [*F*(2, 38) = 0.676, *p* = 0.473, *η*<sup>2</sup> = 0.034].

### Relationships Among Evaluation and Risky Decision-Making Behaviors

In the present study, there were two types of normal choices: CNY50 by 0.2 probabilities (expected value (*EV*) = 10) and CNY15 by 0.66 probabilities (*EV* = 9.9), CNY50 by 0.5 probabilities (*EV* = 25) and CNY40 by 0.66 probabilities (*EV* = 26.4). Based on behavioral results, participants were divided into groups. For the choice of CNY50 by 0.2 and CNY15 by 0.66, 9 participants (LH) almost chose the former and 11 participants (LL) almost chose the latter. For the choice of CNY50 by 0.5 probabilities and CNY40 by 0.66 probabilities, 8 participants (HH) almost chose the former and 12 participants (HL) almost chose the latter.

We conducted independent *t*-test using ERP data on Fz to analyze the correlation among valuation of risky rewards and risky decision-making behaviors (**Figure 6**). Statistical results showed that there was no significant ERP difference between LH and LL groups when they observed CNY50 by 0.2 and CNY15 by 0.66, respectively. However, when observing CNY50 by 0.5 and CNY40 by 0.66 respectively, HH and HL groups displayed different ERP valuation. For CNY50 by 0.5, there was no significant ERP difference between HH and HL groups, but for CNY40 by 0.66, two types of participants expressed different valuation. For P200 component, the mean amplitudes were 1.1040 and 4.9620 μV in HH and HL groups, respectively (*t* = −2.025, *df* = 18, *p* = 0.058). For MFN component, the mean amplitudes were −4.0098 and 0.2311 μV in HH and HL groups, respectively (*t* = −1.675, *df* = 18, *p* = 0.111). For the LPP component, the mean amplitudes were −1.8682 and 4.5597 μV in HH and HL groups, respectively (*t* = −2.437, *df* = 18, *p* = 0.025).

### DISCUSSION

Risky decision-making involves risky reward valuation, choice, and feedback processes. This study focused on risky reward valuation. This investigation assessed the neural dynamics involved in the processing of probability weight and money magnitude. ERP results demonstrated distinct temporal dynamics for probability weight and money magnitude processes. The early frontal P200, MFN, and LPP components all represented the process of probability weight; however, only the MFN component was associated with the process of money magnitude when evaluating a risky reward.

Frontal P200 revealed a significant main effect of probability weight on the frontal and frontal-central areas, but no significant main effect of money magnitude for a risky reward. Low-probability reward elicited less positive P200 amplitude when compared to high-probability reward at the same magnitude. Previous studies showed that P200, the probable sources of which may be the mesotelencephalic dopamine reward system, likely associates with stimulus evaluation and quick assessment (Boudreau et al., 2009; Chen et al., 2009). The P200 component has been shown to be involved in attention to relevant cues including reward-related stimuli (Molinaro and Carreiras, 2010; Lau et al., 2013; Gui et al., 2016). Several ERP studies, which explored the processing of reward, found that a reward condition elicited larger P200 compared to a non-reward condition (Martin and Potts, 2004; Franken et al., 2010). The relationships between low- and high-probability rewards were similar to those relationships. Schuermann et al. (2012) found that P200 was enhanced on negative feedbacks in high-risk compared to low-risk choices, which suggests that large negative prediction errors are already processed in the P200 time range. Hence, our findings are consistent and suggest that participants detected the initial feature of probability, not magnitude at the early stage of risky option processing.

The MFN component, which reflects the impact of dopamine-dependent reward signals on the ACC, may represent the evaluation of reward value (Gehring, 2002; Holroyd and Coles, 2002; Proudfit, 2015). In the present study, consistent with this classical theory, the MFN component showed significant main effects of both probability weight and money magnitude of a risky reward. Given the same magnitude, low-probability options evoked a more negative MFN as compared to high-probability options. Moreover, small magnitude induced a more pronounced MFN than large magnitude for the same probability weight. Existing studies demonstrated that the MFN component reflects the early appraisal of feedback, in which the MFN response to unfavorable outcomes is larger compared to favorable outcomes (Holroyd et al., 2006; Goyer et al., 2008; Boksem and de Cremer, 2010; Broyd et al., 2012; Huang and Yu, 2014; Umemoto et al., 2017). Our results are consistent with those findings. Since risky decision-making is ubiquitous, highprobability rewards are considered better than low-probability ones for the same magnitude. In other words, a high-probability reward is a "good" outcome, relative to a low-probability reward when the magnitude is constant. Likewise, large rewards are considered better than small rewards with the same probability weight.

The LPP component has been mainly associated with affective and emotional processing (Ferrari et al., 2011; Righi et al., 2012). Many studies have found positive and negative stimuli to elicit larger LPP amplitude than neural stimuli, which suggests that more brain resources are allocated to affective stimuli (Foti and Hajcak, 2008; Hua et al., 2014; Zhang et al., 2014; Guo et al., 2018). In this study, LPP was more positive for high-probability than low-probability reward, demonstrating that participants paid more attention to high-probability reward. The study of Harris et al. (2013) found that LPP reflected process differences between liked and disliked food items. Those results suggest LPP is related to valuation modulation. The relationship between liked and disliked foods is similar to that between high- and low-probability rewards. Wu et al. (2011) investigated the neural response to selection of risky rewards. They found that medial prefrontal cortex (mPFC) involved in the process of magnitude, and mPFC and ACC correlated with probability. Given that P300 and LPP amplitude variation is related to the striatum (Pfabigan et al., 2014) and the MFN is correlated with ACC and mPFC (Gehring, 2002; Boksem and de Cremer, 2010), their findings support our conclusions.

In summary, this study investigated neural dynamics of the processes associated with probability weight and money magnitude in the evaluation of a risky reward. ERP results demonstrated P200, MFN, and LPP components to reflect the processing of probability weight, while only the MFN component reflected the processing of money magnitude when evaluating a risky reward. These findings contribute to an understanding of the temporal course of processing probability weight and money magnitude during risky choices.

### AUTHOR CONTRIBUTIONS

GW and JL conceived and designed this study. GW, JL, PW, and CZ designed experimental stimuli and procedures. CZ and JP implemented experimental protocols and collected data. SL and GW analyzed data. GW and JL wrote the paper.

### REFERENCES


### FUNDING

This work was supported by the National Natural Science Fund of China (Grant numbers: 71673152 and 71533002), the National Social Science Foundation of China (Grant number: 16BJY035), and Shandong Provincial Natural Science Fund, China (Grant number: ZR2016GM07).


evidence. *Organ. Behav. Hum. Decis. Process.* 125, 73–87. doi: 10.1016/j. obhdp.2014.06.003


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2019 Wang, Li, Wang, Zhu, Pan and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Neural Dynamics Underlying the Evaluation Process of Ambiguous Options During Reward-Related Decision-Making

### *Chengkang Zhu1,2† , Jingjing Pan1,2† , Yiwen Wang3 , Jianbiao Li2,4 \* and Pengcheng Wang5 \**

*1 Reinhard Selten Laboratory, Business School, China Academy of Corporate Governance, Nankai University, Tianjin, China, 2 School of Economics, Institute for Study of Brain-like Economics, Shandong University, Jinan, China, 3 China Center of Social Trust Research, Fuzhou University, Fuzhou, China, 4 Department of Economic and Management, Nankai University Binhai College, Tianjin, China, 5 Business School, Tianjin University of Economic and Finance, Tianjin, China*

Ambiguous decision-making involves different processes. However, few studies have focused on the evaluation process. In this study, event-related potentials (ERPs) and event-related spectrum perturbation (ERSP) techniques were used to explore the neural dynamics underlying the evaluation process of ambiguous options through an ambiguous choice task. Some important results emerged. We found a preference for lotteries with low ambiguity regardless of reward amount, suggesting that subjects were averse to ambiguity in our paradigm. Our electroencephalography (EEG) results clarified the neural dynamics underlying the evaluation process. In the time domain, lotteries with both a larger reward and lower ambiguity elicited a larger P3. In the time-frequency domain, larger amplitudes of delta activity at 200–400 ms and 500–600 ms post-stimulus were elicited by lotteries with low ambiguity. Moreover, lotteries with a larger reward elicited larger amplitudes of delta activity at 400–600 ms post-stimulus. Our ERPs and ERSP results suggested that individuals in our paradigm evaluated ambiguity and reward separately, and then integrated their evaluation to form subjective values of different lotteries.

### Keywords: ambiguous options, evaluation, neural dynamics, P3, delta activity

### INTRODUCTION

Decision-making under uncertainty permeates our daily life. According to the precise likelihood of outcome, economists divide two types of uncertain events: risk and ambiguity. For risk, the outcome probability corresponds to a point estimation. For ambiguity, the outcome probability is either unknown (Ellsberg, 1961) or an interval estimation (Becker and Brownson, 1964; Curley and Yates, 1985; Rustichini et al., 2005). The majority of uncertainty in real situations is ambiguity. In many experiments related to ambiguity, one of the most prominent phenomena, referred to as "ambiguity avoidance," is the individual's preference for risk over ambiguity (Ellsberg, 1961; Camerer and Weber, 1992). Decision-making is a continuous process, which entails the evaluation of ambiguous options, formation of preference, choice, and learning from feedback (Wang et al., 2015). Ambiguity avoidance has been demonstrated to emerge during the evaluation process (Rode et al., 1999). During the evaluation of ambiguous options, the mean outcome and its variance are integrated to form preference. Ambiguity is supposedly averse because of its high outcome variance. However, this speculation lacks support from neural dynamic evidence of the evaluation of ambiguous options.

### *Edited by:*

*Ya Zheng, Dalian Medical University, China*

### *Reviewed by:*

*Yiping Zhong, Hunan Normal University, China Yansong Li, Nanjing University, China*

### *\*Correspondence:*

*Jianbiao Li biaojl@126.com Pengcheng Wang chengpw@126.com*

*† These authors have contributed equally to this work*

### *Specialty section:*

*This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Psychology*

*Received: 03 May 2019 Accepted: 13 August 2019 Published: 28 August 2019*

### *Citation:*

*Zhu C, Pan J, Wang Y, Li J and Wang P (2019) Neural Dynamics Underlying the Evaluation Process of Ambiguous Options During Reward-Related Decision-Making. Front. Psychol. 10:1979. doi: 10.3389/fpsyg.2019.01979*

**18**

The focus of previous neuroimaging studies has been on contrasting the neural mechanism related to decision-making under risk and ambiguity. Decision-making under ambiguity elicits greater activity in the amygdala, orbitofrontal cortex, lateral prefrontal cortex, anterior insular cortex, posterior inferior frontal gyrus, and posterior parietal cortex, and less activity in the striatum (Hsu et al., 2005; Huettel et al., 2006; Bach et al., 2009). One functional magnetic resonance imaging (fMRI) study investigated the neural representation of subjective value under risk and ambiguity (Levy et al., 2010). In that study, subjects were asked to make decisions under different levels of risk and ambiguity. Their behavioral data were used to calculate the subjective value of each option, and neural activity was measured. The results revealed that the activities of the striatum, medial prefrontal cortex, posterior cingulate cortex, and amygdala were correlated with the subjective value of risky and ambiguous options.

Event-related potentials (ERPs) and event-related spectral perturbations (ERSPs) have millisecond-level temporal resolution, which is useful in exploring the evaluation process of ambiguous options. Existing ERP studies have been mainly focused on neural correlates underlying the choice and feedback stage of decision-making under uncertainty (for reviews, see Chandrakumar et al., 2018). To our knowledge, only one ERP study has explored the neural mechanism underlying the evaluation stage (Wang et al., 2015). In their experiment, participants were asked to decide whether to bet or not, under ambiguity and risk. They made decisions after a random monetary reward was presented. They would either earn or lose the monetary reward if they decided to bet. Otherwise, they would earn nothing. The results revealed that a larger P3 was elicited by risky options compared with ambiguous options. Previous ERP studies have shed light on the neural dynamics of decision-making under ambiguity (Gu et al., 2010; Xu et al., 2011; West et al., 2014; Kóbor et al., 2015; Mussel et al., 2015; Wang et al., 2015; Endrass et al., 2016; Azcárraga-Guirola et al., 2017). However, there have been limitations. Decision-making under ambiguity includes several stages, from the evaluation of ambiguous options to feedback processing. Few ERP studies have been focused on the evaluation process. Although Wang et al. (2015) explored the neural dynamics underlying the evaluation of ambiguous options, they mainly aimed at comparing the neural mechanism of ambiguity and risk. Their paradigm did not allow one to distinguish between evaluation and choice processes. Moreover, their study did not clarify the temporal dynamics of ambiguous option evaluation, which entails processing of the level of ambiguity, reward amount, and the corresponding integration process.

In this study, we used ERPs and the ERSP technique to investigate neural temporal dynamics underlying the process of ambiguous option evaluation. Therefore, we developed an ambiguous choice task. Our task paradigm was derived from previous literature on risky choice (Wang et al., 2019), given that the evaluation of ambiguous options is somewhat similar to that of risky options (Levy et al., 2010). In our task, two ambiguous lotteries were serially presented. Subjects were then asked to choose one lottery to decide their payoff. This allowed us to separate evaluation and choice processes. No feedback was shown to the subjects to control for the learning effect. By varying the probability interval of reward, we manipulated the level of ambiguity, based on the methods of Levy et al. (2010). We set up four types of lotteries: high ambiguity with a reward of 20 Chinese yuan (CNY) (H20); high ambiguity with a reward of 10 CNY (H10); low ambiguity with a reward of 20 CNY (L20); and low ambiguity with a reward of 10 CNY (L10). For each reward, lotteries with different levels of ambiguity led to the same mean reward. Using this paradigm, we were able to clarify the integration process of ambiguity and mean reward during the evaluation of an ambiguous option.

Several electroencephalography (EEG) components in the time domain and time-frequency domain can be used to explore the process of ambiguous option evaluation. In the time domain, the relevant component during the evaluation stage is P3 (Goldstein et al., 2006; Broyd et al., 2012; Zhang et al., 2017; Zheng et al., 2017). The P3 peak at 300–600 ms post-stimulus at posterior scalp sites is associated with reward evaluation and anticipation (Pfabigan et al., 2015). Furthermore, P3 possibly encodes the subjective value of each ambiguous option. Thus, we hypothesized that both low ambiguity and reward of 20 CNY would elicit larger P3 amplitudes. In the time-frequency domain, delta power (1–4 Hz) is an index of reward processing (for review, see Knyazev, 2007, 2012). Therefore, the dynamics of delta activity when faced with four types of lotteries would reflect the sequential order of processing the ambiguity, mean reward, and integration between the two. We predicted that low ambiguity would elicit a larger delta band activity when processing the ambiguity, while the reward of 20 CNY would elicit a larger delta band activity when processing the mean reward. Moreover, we predicted that both low ambiguity and the reward of 20 CNY elicited larger delta band activity when processing the integration between ambiguity and mean reward.

### MATERIALS AND METHODS

### Participants

A total of 25 healthy volunteers (age range = 21–25 years; females = 12) from Nankai University participated in this study. Sample size was determined by power analysis. All participants were right-handed, native Chinese speakers. The participants had normal or corrected-to-normal vision and no history of psychiatric or neurological disorders. Each participant signed written informed consent forms and received a base payment of 30 Chinese yuan (CNY, roughly equal to US \$4.50) for participation, plus a bonus of 0–60 CNY based on his/her decision. The study protocol was approved by the Ethics Committee of Nankai University. The procedures were performed in accordance with approved guidelines and the Declaration of Helsinki. Materials and data related to this study will be made available upon request.

### Stimuli and Task

The participants performed an ambiguous choice task. In each trial, participants were presented with two lotteries sequentially with varying ambiguity levels and varying reward amounts. Participants had to indicate which lottery they preferred. Each lottery appeared on the screen in the form of a "pie" painted partly red and partly green (**Figure 1**). All pies contained 10 sectors. Participants were told that each image on the screen represented a physical bag containing 10 balls. The relative numbers of red and green balls were indicated by the proportions of red and green sectors. Part of the pie was hidden by a gray occluder. The probability of drawing green or red balls was therefore ambiguous. Two different occluder sizes (covering either two or six sectors) represented two ambiguity levels. For the low ambiguity level, the probability of drawing a red ball could have been anywhere between 40 and 60%. Similarly, the probability of drawing a green ball could have been anywhere between those two values. For the high ambiguity level, the probability of drawing a red ball could have been anywhere between 20 and 80%. The probability of drawing a green ball could also have been anywhere between 20 and 80%. The number under the pie represented the amount of money to be won that was associated with the target color. For half of the participants, drawing a red ball yielded a given amount of money and drawing a green ball yielded nothing. For the other half of the participants, drawing a green ball yielded a given amount of money, and a red ball yielded nothing. Two reward amounts (10 and 20 CNY) were used at each ambiguity level, to give four types of lotteries, i.e., H20, H10, L20, and L10.

At the beginning of the experiment, participants were told that each lottery corresponded with a unique bag. Therefore, we provided four sealed bags associated with four types of lotteries. Before the task, participants were asked to sign their names on the sealed bags. This method was used to ensure the relative numbers of red and green balls could not have been adjusted by experimenters during the task. At the end of the experiment, three trials were randomly selected by computers. Based on the lotteries they chose in these trials, participants then drew a ball from each corresponding bag (if two or three lotteries were the same, they would draw a ball twice or three times from the corresponding bag with replacements). In addition to their participation fee, they were paid according to the lotteries and number of balls of the target color.

### Procedure

The EEG recording was performed in a small, sound-attenuated, electrically shielded chamber. After the EEG electrodes were attached, the participants sat in a comfortable chair that was approximately 100 cm in front of a 23-inch (58.42 cm) computer monitor. Before the tasks began, all participants read the instructions carefully and were asked to take eight practice trials. **Figure 1** shows the timeline of a single trial. Each trial began with the presentation of a single centrally located white fixation cross for 600–800 ms. A black screen was then presented for 500–700 ms, followed by the first lottery for 1,000 ms. Subsequently, the second lottery was presented for 1,000 ms, after which, a black screen was presented for 500–700 ms. The order of these lotteries was counterbalanced. Thereafter, participants were asked to choose one of the lotteries to decide their payoff in that trial.

The entire experiment comprised 80 test and eight practice trials. Only the test trials were used for EEG analysis. The trials occurred within four blocks of 20 trials. Each block was separated by a break, the duration of which was determined by the participants. All 80 trials were performed within 15–25 min, during which the trials were randomly presented. The E-Prime software was used to control the display of stimuli and acquisition of behavioral data (Version 2.0, Psychology Software Tools, Inc.).

### Electroencephalography Acquisition

The EEG data were recorded continuously with a 40-channel NuAmps DC amplifier (Compumedics Neuroscan, Inc., Charlotte, NC, USA). According to the International 10-20 System, 32 active Ag/AgCl electrodes were used. The EEG was sampled at 1,000 Hz using a 22-bit A/D converter. The reference and ground electrodes were positioned at AFz, and the impedances of all electrodes were kept below 10 kΩ.

### Electroencephalography Analysis

Preprocessing of EEG data was performed with the EEGLAB 14.1.1 tool (Delorme and Makeig, 2004), implemented in MATLAB 2017a. In addition, a 0.1/30 Hz high-/low-pass filter was applied after the reference of EEG signals was reset to the average of the left and right mastoids. Individual epochs were extracted from −1,000 to 2,000 ms around the presentation of the stimulus defined as the lotteries that sequentially presented in our task. A manual artifact correction procedure was applied to eliminate trials with artifacts, based on visual inspection. Independent component analysis (ICA) was performed to remove eye movement, and the related ICA components were manually selected. Artifact-free epochs of each subject were grouped into four conditions, i.e., H20, H10, L20, and L10.

Clean EEG data were analyzed in the time domain. The 1,000-ms epochs were extracted, starting at 200 ms before the presentation of the stimulus. A 200-ms pre-stimulus period was used as baseline, and the accepted epochs were baselinecorrected. The P3 was scored as the mean voltage from 500 to 600 ms post-stimulus at Pz, corresponding to the 100-ms time window surrounding the peak.

Time-frequency analysis was performed using the Fieldtrip toolbox (Oostenveld et al., 2011) built-in ft\_freqanalysis function, based on complex Morlet wavelet convolution (1–10 cycles, 1–30 Hz, 120 spaced frequencies, 1,000 time points per epoch). The time interval of −200 to 0 ms before presentation of the stimulus was used for baseline normalization. The mean converted amplitudes within 1–4 Hz from 200 to 300 ms, 300 to 400 ms, 400 to 500 ms, and 500 to 600 ms at Pz were used to analyze the delta band power change in different time windows.

For all analyses, the values of *p* were corrected using the Greenhouse-Geisser correction when the sphericity assumption was violated. A value of *p* < 0.05 was considered significant. Significant interaction was analyzed using the simple effect model. Statistics were analyzed using the SPSS 19.0 software.

### RESULTS

### Behavior Data

The subjective value of a lottery was defined as the frequency with which it was selected by the participants. The subjective value was analyzed using two-way repeated measures ANOVA (rmANOVA) with ambiguity levels (high vs. low) and reward amounts (10 vs. 20 CNY) as within-subject factors (**Figure 2A**). A significant main effect was found for reward amount [*F*(1, 24) = 452.859, *p* < 0.001, partial *η*<sup>2</sup> = 0.950], as a higher subjective value was noted for 20 CNY (mean ± SEM = 30.860 ± 1.392) versus 10 CNY (mean ± SEM = 9.140 ± 1.119). A significant main effect was also found for ambiguity [*F*(1, 24) = 228.553, *p* < 0.001, partial *η*<sup>2</sup> = 0.905], as a higher subjective value was noted for low ambiguity (mean ± SEM = 27.840 ± 1.808) versus high ambiguity (mean ± SEM = 12.160 ± 1.492). A significant interaction effect was revealed [*F*(1, 24) = 15.600, *p* = 0.001, partial *η*<sup>2</sup> = 0.394]. For high ambiguity, a significant main effect was observed with reward amount [*F*(1, 24) = 210.251, *p* < 0.001, partial *η*<sup>2</sup> = 0.898], as a higher subjective value was noted for 20 CNY (mean ± SEM = 21.760 ± 1.007) than 10 CNY (mean ± SEM = 2.560 ± 0.663). For low ambiguity, a significant main effect was also observed with reward amount [*F*(1, 24) = 513.497, *p* < 0.001, partial *η*<sup>2</sup> = 0.955], as a higher subjective value was observed for 20 CNY (mean ± SEM = 39.960 ± 0.040) than 10 CNY (mean ± SEM = 15.720 ± 1.053). For 20 CNY, a significant main effect was observed with ambiguity level [*F*(1, 24) = 334.586, *p* < 0.001, partial *η*<sup>2</sup> = 0.933], as a higher subjective value was noted for low ambiguity (mean ± SEM = 39.960 ± 0.040) than high ambiguity (mean ± SEM = 21.760 ± 1.007). For 10 CNY, a significant main effect was also observed with ambiguity level [*F*(1, 24) = 87.662, *p* < 0.001, partial *η*<sup>2</sup> = 0.785], as a higher subjective value was noted for low ambiguity (mean ± SEM = 15.720 ± 1.053) than high ambiguity (mean ± SEM = 2.560 ± 0.663).

Reaction time was analyzed using one-way rmANOVA with conditions (two lotteries with the same ambiguity level but different reward amounts vs. two lotteries with different ambiguity levels but same reward amount vs. two lotteries with different ambiguity levels and different reward amounts, henceforth referred to as SD vs. DS vs. DD). We found no significant main effects [*F*(2, 48) = 1.818, *p* = 0.184, partial *η*<sup>2</sup> = 0.070] for the various conditions (**Figure 2B**).

### Electrophysiological Data P3

A two-way rmANOVA was performed with ambiguity (high vs. low) and reward (10 vs. 20 CNY) as factors (**Figure 3**). A significant main effect was found [*F*(1, 24) = 33.891, *p* < 0.001, partial *η*<sup>2</sup> = 0.585] for reward amount, with a larger amplitude of P3 for 20 CNY (mean ± SEM = 2.730 ± 0.376 μV) than 10 CNY (mean ± SEM = 0.403 ± 0.417 μV). Moreover, a significant main effect was also found [*F*(1, 24) = 12.367, *p* = 0.002, partial *η*<sup>2</sup> = 0.340] for ambiguity level, with a larger amplitude of P3 for low ambiguity (mean ± SEM = 2.283 ± 0.430 μV) than high ambiguity (mean ± SEM = 0.824 ± 0.373 μV). However, no significant interaction effects were found.

### Delta Activity

A two-way rmANOVA was performed on the delta power from 200 to 300 ms with ambiguity (high vs. low) and reward (10 vs. 20 CNY) as factors (**Figure 4**). Only one significant main effect was found [*F*(1, 24) = 11.822, *p* = 0.002, partial *η*<sup>2</sup> = 0.330] for ambiguity level, with a larger amplitude for low ambiguity (mean ± SEM = 2.379 ± 0.259 dB) than high ambiguity (mean ± SEM = 1.733 ± 0.218 dB). However, we found no significant main effects for reward or significant interaction effects.

A two-way rmANOVA was performed on the delta power from 300 to 400 ms with ambiguity (high vs. low) and reward (10 vs. 20 CNY) as factors (**Figure 5**). Only one significant main effect was found [*F*(1, 24) = 7.646, *p* = 0.011, partial *η*2 = 0.242] for ambiguity, with a larger amplitude for low ambiguity (mean ± SEM = 2.444 ± 0.255 dB) than high ambiguity (mean ± SEM = 1.864 ± 0.219 dB).

A two-way rmANOVA was performed on the delta power from 400 to 500 ms with ambiguity (high vs. low) and reward (10 vs. 20 CNY) as factors (**Figure 6**). Only one significant main effect was found [*F*(1, 24) = 9.846, *p* = 0.004, partial *η*2 = 0.291] for reward amount, with a larger amplitude for 20 CNY (mean ± SEM = 2.252 ± 0.200 dB) than 10 CNY (mean ± SEM = 1.502 ± 0.194 dB).

A two-way rmANOVA was performed on the delta power from 500 to 600 ms with ambiguity (high vs. low) and reward (10 vs. 20 CNY) as factors (**Figure 7**). A significant main

from 300 to 400 ms.

from 500 to 600 ms.

effect was found [*F*(1, 24) = 6.772, *p* = 0.016, partial *η*<sup>2</sup> = 0.220] for ambiguity level, with a larger amplitude for low ambiguity (mean ± SEM = 2.485 ± 0.212 dB) than high ambiguity (mean ± SEM = 1.854 ± 0.160 dB). Moreover, a significant main effect was found [*F*(1, 24) = 13.481, *p* = 0.001, partial *η*2 = 0.360] for reward, with a larger amplitude for 20 CNY (mean ± SEM = 2.658 ± 0.191 dB) than 10 CNY (mean ± SEM = 1.681 ± 0.199 dB). However, no significant interaction effects were found.

### DISCUSSION

Ambiguous decision-making involves different processes, from evaluation to feedback. Previous behavioral studies have suggested that ambiguity aversion occurs because of a lower subjective value with high variance of the mean outcome during the evaluation process. However, this speculation lacks supportive evidence from neural dynamics analyses. In this study, ERP and ERSP techniques were used to explore the neural dynamics underlying the evaluation process of ambiguous options through an ambiguous choice task. Some important results have emerged. We found a preference for lotteries with low ambiguity regardless of reward amount, suggesting that subjects were averse to ambiguity in our paradigm. Our EEG results clarified the neural dynamics underlying the evaluation process. In the time domain, both lotteries with larger rewards and those with low ambiguity elicited a larger P3. In the time-frequency domain, larger amplitudes of delta activity at 200–400 and 500–600 ms poststimulus were elicited by lotteries with low ambiguity. Moreover, lotteries with larger rewards elicited a larger amplitude of delta activity at 400–600 ms post-stimulus.

Our behavioral data showed that most participants disliked lotteries with high ambiguity and small rewards. This finding is consistent with previous studies regarding decision-making under ambiguity (Rode et al., 1999; Levy et al., 2010). In our task, the subjective value of lottery H10 was 2.56. When the ambiguity was low, the subjective value increased to 15.72. As the reward increased to 20 CNY, the subjective value increased significantly and reached 21.76 in cases of high ambiguity. Moreover, for lottery L20, the subjective value was 39.96 and significantly higher than all other lotteries. Lower ambiguity led to an increased frequency with which the lottery was chosen, and therefore, a larger subjective value. We noted that the reaction time was identical among different choice conditions (i.e., SD vs. DS vs. DD). This result suggested that the difficulty of choosing between different types of lotteries was similar in our task.

In the time domain, we observed an obvious P3 component peaking at approximately 500–600 ms following the presentation of the lotteries on the screen. This component reflects the stimulus categorization process (for reviews, see Polich, 2007) and motivational salience to the stimulus (for review, see Polich and Kok, 1995). The P3 wave has also been associated with activation of the ventral striatum (Pfabigan et al., 2015) during the evaluation process. Since the ventral striatum is a region related to reward processing (Delgado et al., 2000; Schultz, 2000; Breiter et al., 2001; Knutson et al., 2003; Tobler et al., 2007; Haber and Knutson, 2010; Kahnt et al., 2011; Sescousse et al., 2013, 2014; Wilson et al., 2018), the P3 can be an index of reward evaluation. In our study, both lotteries with low ambiguity and those with larger rewards elicited a larger amplitude of P3, indicating that P3 integrated the evaluation of ambiguity and reward. Among four types of lotteries, the amplitude of P3 for H10 was the smallest at only −0.215 μV. When the ambiguity was low, the amplitude of P3 increased to 1.022 μV. As the reward increased to 20 CNY, the amplitude of P3 increased and reached 1.863 μV in cases of high ambiguity. For the L20 lotteries, the amplitude of P3 was 3.543 μV, the highest among all lotteries. Considering our behavioral and ERP data together, we found that a larger amplitude of P3 led to an increased frequency with which a lottery was selected. Thus, we suggested that the amplitude of P3 during the evaluation process encoded the subjective value of each ambiguous lottery, and could be used to predict the subsequent choice.

In the time-frequency domain, delta activity is sensitive to reward evaluation during reward anticipation processing (for review, see Knyazev, 2007, 2012; Glazer et al., 2018). It could also be an index representing the integration of reward (Gheza et al., 2018; Zhu et al., 2019). In the present study, the dynamics of delta power for lotteries during the evaluation process reflected the subjects' processing of the elements of the lotteries (i.e., ambiguity level and mean reward) and their integration. At 200–400 ms after the stimulus, lotteries with low ambiguity elicited larger amplitudes of delta activity, indicating that individuals started to evaluate ambiguity at about 200 ms after the presentation of lotteries on the screen. This result also suggested that individuals preferred lotteries with low ambiguity to those with high ambiguity. During the next 100 ms, lotteries with a larger reward elicited larger amplitudes of delta activity, indicating that individuals evaluated reward at 400–500 ms post-stimulus. Moreover, this result suggested that individuals preferred larger rewards. At 500–600 ms post-stimulus, both lotteries with low ambiguity and those with larger rewards elicited larger amplitudes of delta activity. These results indicated that subjects integrated their evaluation of ambiguity and reward to form a subjective value of lotteries at 500–600 ms after presentation of the lotteries. This supports our findings in the time domain and supports the idea that delta activity plays a key role in P3 generation during reward evaluation (Demiralp et al., 2001; Bernat et al., 2007, 2015; Ergen et al., 2008; Ishii et al., 2009).

In summary, to our knowledge, this study is the first to investigate the neural dynamics underlying the evaluation process of lotteries under ambiguity. Our ERPs and ERSP results suggest that individuals in our paradigm evaluated ambiguity and reward separately. The ambiguity level was evaluated at 200–400 ms and the reward was evaluated at 400–500 ms after the lotteries were presented on the screen. At 500–600 ms after the stimulus, individuals integrated the evaluation of ambiguity and reward to form a subjective value of the different lotteries. These findings shed light on our understanding of the temporal course of processing ambiguous options. Furthermore, our findings provide neural dynamic evidence of the emergence of ambiguity avoidance during the evaluation process. One limitation of this study should be mentioned. Although the evaluation of ambiguity was earlier than that of reward in our task, whether individuals in other tasks evaluate ambiguity first is unclear. Future studies should explore the impacting factor of evaluation sequence during the process of ambiguous decision-making.

### DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the Ethics Committee of Business of Nankai University committee with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol

### REFERENCES


was approved by the Ethics Committee of Business of Nankai University committee.

### AUTHOR CONTRIBUTIONS

CZ, JP, PW, and JL designed the experiment. CZ and JP carried out the experiment, analyzed the data, wrote the paper, and contributed equally to this work. CZ, JP, JL, YW, and PW revised the paper.

### FUNDING

This work was supported by the National Natural Science Fund of China (Grant numbers: 71673152; 71533002; 71602051); Ministry of Education in China Project of Humanities and Social Sciences (Grant number: 18JJD630001); The National Social Science Foundation of China (Grant number: 16BJY035); and Shandong Provincial Natural Science Fund, China (Grant number: ZR2016GM07).


human functional neuroimaging studies. *Neurosci. Biobehav. Rev.* 37, 681–696. doi: 10.1016/j.neubiorev.2013.02.002


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2019 Zhu, Pan, Wang, Li and Wang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Diminished Anticipatory and Consummatory Pleasure in Dysphoria: Evidence From an Experience Sampling Study

Xu Li1,2 \*, Yu-Ting Zhang1,2, Zhi-Jing Huang1,2, Xue-Lei Chen1,2, Feng-Hui Yuan<sup>3</sup> and Xiao-Jun Sun1,2

<sup>1</sup> School of Psychology, Central China Normal University, Wuhan, China, <sup>2</sup> Key Laboratory of Adolescent Cyberpsychology and Behavior, Ministry of Education, Central China Normal University, Wuhan, China, <sup>3</sup> School of Sociology, Central China Normal University, Wuhan, China

### Edited by:

Ruolei Gu, Institute of Psychology (CAS), China

### Reviewed by:

Jinhyuk Kim, Pennsylvania State University, United States Sam Winer, Mississippi State University, United States

> \*Correspondence: Xu Li xuli@mail.ccnu.edu.cn

### Specialty section:

This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Psychology

Received: 13 June 2019 Accepted: 02 September 2019 Published: 19 September 2019

### Citation:

Li X, Zhang Y-T, Huang Z-J, Chen X-L, Yuan F-H and Sun X-J (2019) Diminished Anticipatory and Consummatory Pleasure in Dysphoria: Evidence From an Experience Sampling Study. Front. Psychol. 10:2124. doi: 10.3389/fpsyg.2019.02124 Anhedonia, the experience of diminished pleasure, is a core feature of major depressive disorder and is often present long before the diagnosis of depression. Most previous studies have investigated anhedonia with self-report measures of trait anhedonia or with behavioral paradigms using laboratory stimuli, and the real-time characteristics of hedonic processing in subclinical depression remain under-investigated. We used the experience sampling method to evaluate momentary experience of hedonic feelings in the context of daily life. Dysphoric (n = 49) and non-dysphoric (n = 51) college students completed assessments of their current positive affect (PA), as well as state anticipatory and consummatory pleasure, 3 or 4 times a day every day for 2 weeks. The results showed that dysphoric individuals reported less state anticipatory and consummatory pleasure compared with non-dysphoric individuals. Moreover, significant time-lagged associations between anticipatory pleasure and follow-up consummatory pleasure were found in the whole sample, after adjustment for current PA. The current findings thus hold considerable promise in advancing our understanding of anhedonia as well as the important role of state anticipatory pleasure in relation to depression.

Keywords: anhedonia, anticipatory pleasure, consummatory pleasure, experience sampling, dysphoria

### INTRODUCTION

Anhedonia, the diminished ability to experience pleasure, manifests as a transdiagnostic symptom among individuals with different psychiatric disorders (Thomsen et al., 2015). For this reason it has been identified by the Research Domain Criteria (RDoC) initiative as key to the investigation of behavioral and clinical symptoms across disorders (Insel et al., 2010). In the case of major depressive disorder (MDD), anhedonia is an essential diagnostic feature, and the severity of anhedonia has been found to correlate with the severity of depressive symptoms and time to remission (McMakin et al., 2012).

However, accumulating evidence suggests that anhedonia is not a unitary construct. Two independent sub-components of the hedonic processing function have been revealed: anticipatory pleasure and consummatory pleasure (Kring and Caponigro, 2010; Treadway and Zald, 2011). Anticipatory pleasure refers to pleasure derived from predicted future events; in contrast,

consummatory pleasure involves the experience of pleasure during current events. The importance of such a distinction has been clear in other disorders. For example, anhedonia in schizophrenia has been demonstrated in anticipatory but not consummatory pleasure, in patients as well as in individuals at risk for schizophrenia (Li et al., 2015). More research is warranted to investigate the distinct roles of anticipatory and consummatory pleasure and their contributions to the pathology seen in depression.

Although it is well-documented that anhedonia plays a role in depression, the role of each type of anhedonia is not wellunderstood. Most empirical studies on anhedonia in depression have primarily focused on consummatory pleasure. Depressed individuals have been shown to report reduced emotional reactivity to stimuli used in laboratory research (Bylsma et al., 2008), and to demonstrate lower positive reactivity and elevated negative reactivity toward daily events (Bylsma et al., 2011). In addition, significant aberrations of brain function related to consummatory pleasure have been reported in individuals at risk for depression (Foti et al., 2011; Bress et al., 2012). In these studies reward-related brain areas such as the caudate, putamen, and anterior cingulate cortex were less activated in individuals with depression. Such blunted reactivity has also been detected in individuals at risk for depression (Saxena et al., 2017), suggesting that depression is robustly associated with reduced behavioral and neural hedonic responsivity, i.e., reduced consummatory pleasure.

Compared to consummatory pleasure, less is known about anticipatory pleasure in depression. A few studies have investigated anticipatory and consummatory pleasure simultaneously, and these studies have produced mixed findings. In an MDD sample, Sherdell et al. (2012) observed no deficits in reward seeking behavior or in consummatory responses to rewards in a laboratory based effort-reward task, suggesting intact anticipatory and consummatory hedonic processing capacity in depression. However, studies utilizing self-report measures of anhedonia have shown both blunted anticipatory and blunted consummatory pleasure in depression (Liu et al., 2012; Sherdell et al., 2012; Yang et al., 2014). Moreover, based on self-report, diminished anticipatory pleasure in depression, rather than consummatory pleasure, was found to be a significant predictor of effort expended for future rewards (Sherdell et al., 2012). Hence, both intact and reduced hedonic capacity have been found to relate to depression. One possible explanation for the differing results may have to do with the operationalization of anhedonia. Experimental studies evaluated dynamic hedonic responses on a trial-by-trial basis with laboratory-based task, while more static estimates of hedonic capacity were obtained with self-report measures.

Importantly, however, both the experimental method and selfreport method have limitations in the study of anhedonia. On one hand, stimuli used in experimental studies (e.g., monetary rewards, positive images) are often standardized to obtain more precise control of confounding factors, but because contextual information is often removed or obscured, laboratory tasks might not be analogous to the actual changing contexts of everyday life. On the other hand, in self-report studies, participants are indicated to rate their pleasure experience in response to a hypothetical situation rather than actual situation. The self-report measures of hedonic experience rely heavily on retrospective recall and therefore scores derived from these measures might be confounded by the severity of memory deficits of the clinical sample characterized by anhedonia (Liu et al., 2012; Olsen et al., 2015). The reduced recall accuracy for specific personal memories has been suggested to be associated with reduced expectations for future events (Schacter et al., 2008). More critically, memory bias toward mood-congruent negative content is strongly related to depressive symptomatology and represents a primary mechanism in depression (Marchetti et al., 2018). Thus, retrospective reports might not provide a comprehensive representation of the fluctuating emotional experiences in daily life.

The experience sampling method (ESM) is a technique that enables real-time and repeated sampling, which could yield more reliable and accurate estimates than single-point assessment or retrospective reports (Csikszentmihalyi and Larson, 1987). Given the fact that emotional feelings are highly variable, the ESM is now generally considered as the "gold standard" to capture the dynamical signature of subjective experience (Shiffman et al., 2008). Important for the current study, the ESM has been frequently used to examine the daily fluctuations of affect in depression, with results showing that depression involves alteration in the mean level of both positive affect (PA) and negative affect (Kuppens et al., 2010; Thompson et al., 2012). The ESM makes it possible to examine how pleasure unfold in the context of daily life and thus offers insights into factors affect pathogenic processes and pathological states of depression.

Studies have started to employ ESM to investigate the temporal features of anticipatory and consummatory pleasure in the context of daily life. The first study was conducted in schizophrenia by Gard et al. (2007), in which deficits in anticipatory but not consummatory pleasure were reported, while recent ESM studies suggested intact capacity to experience pleasure during reward anticipation in schizophrenia (Gard et al., 2014; Edwards et al., 2018). However, the distinctive momentary feature of anticipatory and consummatory pleasure in relation to depression has rarely been studied. In a recent ESM study (Heininga et al., 2019), temporal consummatory pleasure of anhedonic individuals with current MDD episode was examined, the results showed that the frequency of reward experience in anhedonic individuals with current MDD episode were at equivalent levels as in healthy controls, suggesting intact consummatory pleasure in MDD with anhedonic symptoms. The first ESM study (Wu et al., 2017) investigating both daily anticipatory and consummatory pleasure of MDD, however, demonstrated that MDD patients reported blunted anticipatory and consummatory pleasure for daily activities. Although the accuracy of anticipatory ratings for pleasure could be estimated by subtracting consummatory ratings from anticipatory ratings in the Wu et al. (2017) study, this practice could not take full advantage of a key strength of ESM, which is the ability to capture the interactions between the two components of anhedonia. Cross-lagged regression models, on the other hand, could make use of the time-lagged structure of the data and thus could provide a better estimate of the moment-to-moment interplay

between anticipatory ratings and consummatory ratings. The time-lagged analysis was utilized in present study to examine the extent to which anticipatory pleasure is related to in-the-moment experience of pleasure in the flow of daily life.

Although previous studies have provided preliminary evidence that patients with MDD had reduced daily experience of pleasure, it remains unknown whether these dysfunctions of hedonic processing emerge in dysphoria, who demonstrate elevated levels of depressive symptoms, but do not meet the diagnostic criteria for MDD. Dysphoric individuals showed impairment in cognitive function, such as working memory and interference control (Owens et al., 2012); and they reported more frequent use of non-adaptive emotional regulation strategies and experienced more negative affect than non-dysphoric individuals (Quigley and Dobson, 2014). Dysphoric individual are found to be at increased risk for developing MDD in the future (Gotlib et al., 1995), especially when dysphoria is present during young adulthood (Wilcox and Anthony, 2004). A recent ESM study provided preliminary evidence for blunted consummatory pleasure in individuals with anhedonia (Heininga et al., 2017). The temporal unfolding of anticipatory pleasure and the predictive relationship between anticipatory and consummatory pleasure in dysphoria, however, have not been fully studied. Given the close relationship between dysphoria and depression, we speculate that diminished anticipatory and consummatory pleasure might be observed in people with dysphoria.

In the affect literature, pleasure has been considered a dimension of PA. Research investigating the characteristics of PA in depression has suggested that depressive symptoms are related to lower PA both in clinical (Thompson et al., 2012) and nonclinical samples (van Roekel et al., 2016), while the functioning of PA in anhedonia is not yet fully understood. Most experimental studies have examined pleasure ratings in response to stimuli without considering the affective context, especially PA. The dynamic interaction of PA, negative affect and stress experience differed between anhedonic and non-anhedonic individuals with subclinical depression (Bos et al., 2018), suggesting an interplay between anhedonia and temporal affective experience. Our study extends the prior research by considering the impact of PA on the anticipatory and consummatory components of the experience of pleasure in dysphoria.

The primary goal of the current study was to understand the nature of anticipatory and consummatory pleasure in individuals with dysphoria in college-age population by using ESM. By studying individuals at risk for depression, mechanisms that underlie the future development of MDD episode may be uncovered. Furthermore, the recruitment of dysphoric individuals could overcome some of the confounding factors when studying patients with MDD, such as medication, severity of illness and episode number. MDD has its peak incidence in young adults (Kessler et al., 2003), thus a focus of college-age sample is of great relevance. Trait dispositions in anticipatory and consummatory pleasure were assessed with questionnaires. Other data were collected using ESM. Specifically, all participants provided ratings of temporal PA, anticipatory pleasure, and consummatory pleasure in response to phone prompts 3 to 4 times a day for 14 days. It was hypothesized that dysphoric individuals would report less PA, as well as less state anticipatory and consummatory pleasure. Moreover, we explored the potential role of anticipation of pleasure in predicting current pleasure, and whether this relationship varied between individuals with and without dysphoria. Because few studies have examined the predictive relationship between anticipatory and consummatory pleasure, no hypothesis was made regarding group difference, and the relationship between anticipatory pleasure and current pleasure was examined separately in each group. This study also examined the extent to which PA influences the experience of pleasure, and whether PA changes the relationship between anticipatory pleasure and consummatory pleasure. Finally, exploratory analyses of group differences on these time-lagged associations between dysphoric and non-dysphoric groups were performed.

### MATERIALS AND METHODS

### Participants

Three hundred and forty-one college students from Wuhan, China were invited to complete the Beck Depression Inventory-II (BDI-II, Beck et al., 1996), which taps into the severity of depressive symptomology in the general population. The BDI-II is currently one of the most widely used measure for assessing depression, and it has been shown to have good sensitivity and specificity in screening MDD in college-student population (Shean and Baldwin, 2008). Participants who scored 20 (moderate depression) or higher were identified as dysphoric individuals, and participants who scored lower than 14 (minimal depression) were classified as non-dysphoric controls, the BDI-II cut-off for dysphoric and non-dysphoric individuals was consistent with previous studies with college-student sample (Lissnyder and Koster, 2010; Owens et al., 2012; Quigley and Dobson, 2014). Potential participants who met either of these cut-off criteria on the BDI-II were screened by phone to check for availability to participate in the experience sampling study. Based on the phone interview, individuals with a history of psychiatric disorders, brain injury, or substance abuse were not included. As a result, 51 dysphoric individuals and 56 nondysphoric controls were included in the present study. After baseline assessment, two participants from the dysphoric group and three from the non-dysphoric control group dropped out due to scheduling problems.

This study was approved by the Ethics Committee of the Central China Normal University. Written informed consent was obtained from all participants at the beginning of the baseline assessment. Participants were told they could stop participating at any time without penalty. At the end of the study participants received compensation of 30 RMB.

### Baseline Assessment

On the first day they arrived at our laboratory, participants were administered two self-report measures. The Temporal Experience of Pleasure Scale (TEPS, Gard et al., 2006) is a selfreport measure that assesses the trait dispositions of anticipatory pleasure and consummatory pleasure. The Chinese version of

the TEPS (Chan et al., 2012) contains 20 items and participants answered each item on a 6-point Likert scale ranging from 1 ("very false for me") to 6 ("very true for me"). Higher scores indicate greater hedonic capacity. Cronbach's α was 0.759 for the whole scale, 0.726 for the consummatory pleasure subscale, and 0.530 for the anticipatory pleasure subscale. The Snaith– Hamilton Pleasure Scale (SHAPS), developed by Snaith et al. (1995), is a self-rating scale used to measure trait consummatory pleasure in certain situations. Each item is rated on a four-point Likert scale from 1 ("definitely agree") to 4 ("definitely disagree"). The SHAPS contains 14 items, and higher scores indicate more severe deficits in consummatory pleasure. The Chinese version of the SHAPS has good internal consistency (Liu et al., 2012). Cronbach's α in the present study was 0.891.

### ESM Procedure

An Internet link containing the set of ESM assessments was sent to each participant's smart phone 3–4 times each day between 8:00 a.m. to 10:00 p.m., with a time interval of more than 90 min between two adjacent points, for 14 consecutive days. This resulted in a maximum of 56 events per person. Participants were asked to complete the brief assessments as soon as they received the link on their smartphone. If no response was given within 30 min or the participant spent more than 180 s to complete the assessments, the data were considered missing. Two participants from the non-dysphoric group were removed from the final analysis because their response rates were lower than 50%. Thus the final sample consisted of 49 participants in the dysphoric group and 51 participants in the non-dysphoric group. No significant group differences were observed in gender, age, or education (**Table 1**).

### Momentary PA

Momentary PA was measured with the 10 positive items (alert, excited, enthusiastic, attentive, proud, inspired, interested, determined, strong, active) from the Positive and Negative Affect Schedule (PANAS; Watson et al., 1988) using the structure "At


BDI-II, Beck Depression Inventory-II; TEPS, The Temporal Experience of Pleasure Scale; SHAPS, The Snaith–Hamilton Pleasure Scale.

this moment I feel. . ." For each item, participants were asked to indicate the extent to which they were experiencing a certain affect at that moment, from 1 ("not at all") to 5 ("very much").

### State Anticipatory Pleasure and Consummatory Pleasure

To assess consummatory pleasure, participants firstly were asked to indicate what they were doing at that moment, by selecting from several options (i.e., work/study, sleeping, eating, daydreaming, cleaning, playing video games, interacting on social media, shopping, watching movies, going on a date, group discussion, other) and then to rate the amount of pleasure they were experiencing from 1 ("none") to 10 ("an extremely large amount") on a visual analog scale. To measure anticipatory pleasure at the state level, participants were asked to indicate what kind of activity they would be involved in next, and then they were asked to forecast the amount of enjoyment they would derive from that activity from 1 ("none") to 10 ("an extremely large amount"). By including questions regarding current and future activities, the design helps to ensure that ratings for consummatory pleasure and anticipatory pleasure were based on daily events rather than memory representations. The use of brief measures is common in ESM research because excessive length can compromise response quality and reduce compliance (Bakker et al., 2017; Starr and Hershenberg, 2017; Edwards et al., 2018).

### Data Analysis

Group differences on demographic information and baseline self-report measures were analyzed with SPSS 23.0. The ESM data have a hierarchical structure in which within-person daily observations (Level 1; e.g., state experience of pleasure) are nested within between-person characteristics (Level 2; e.g., dysphoric status). Therefore, data were analyzed by means of multilevel modeling in Mplus version 7.4. Separate models were estimated for daily PA, state anticipatory pleasure, and state consummatory pleasure. Furthermore, all Level 1 predictors were person-mean centered (i.e., around each participant's mean score) to separate within-person effects from between-person effects (Finch and Bolin, 2017, pp 38–39). Intercepts and slopes of level 1 were modeled as random effects, allowing the intercepts and slopes to vary at between-person level (i.e., level 2) (Nezlek, 2012).

Before running models to test our hypotheses, we first ran null models in Mplus (i.e., containing no predictors at any level) with PA, state anticipatory pleasure and state consummatory pleasure as the outcome variables. This type of analysis can be used to calculate intraclass correlation coefficients (ICCs), which provides an estimate of the proportion of variance in the outcome variable accounted for by the between-person level, which reflects individual differences (vs. the within-person level, which reflects situational differences).

### Associations Between Momentary Pleasure and Dysphoric Status

To examine whether dysphoric status predicted differences in anticipatory and consummatory pleasure across daily observations, we regressed the within-person parameters on

dysphoric status (i.e., 0 = non-dysphoric group; 1 = dysphoric group) (see section "Model 1").

### Model 1

Level 1: PAij or anticipatory pleasure ij or consummatory pleasure ij = β0<sup>j</sup> + rij

Level 2: β0<sup>j</sup> = γ00+ γ<sup>01</sup> (dysphoric status <sup>j</sup>) + U0<sup>j</sup>

In the equations, i represents time point, and j represents person. The outcome at Level 1 (e.g., PA), representing the observed score of person j's at time i, was modeled as a function of a random intercept (β0j) representing the within-person mean of the corresponding outcome variable, rij represents residual errors at within-person level. At Level-2, γ<sup>00</sup> represents mean pleasure for the non-dysphoric group, and γ<sup>01</sup> represents the difference in mean pleasure between the dysphoric and nondysphoric groups. U0<sup>j</sup> represents residual errors at betweenperson level.

Next, to reveal the unique associations between momentary pleasure and dysphoria, we reran the full models, controlling for PA (see section "Model 2").

### Model 2

Level 1: consummatory pleasure ij or anticipatory pleasure ij = β0<sup>j</sup> + β1<sup>j</sup> PAij + rij

Level 2: β0<sup>j</sup> = γ00+ γ<sup>01</sup> (dysphoric status <sup>j</sup>) + U0<sup>j</sup>

β1<sup>j</sup> = γ10+ γ<sup>11</sup> (dysphoric status <sup>j</sup>) + U1<sup>j</sup>

The outcome at Level 1 (e.g., consummatory pleasure), representing the observed score of person j's score at time i, was modeled as a function of a random intercept (β0j) representing the mean pleasure of person j, and a random slope (β1j) representing the relationship between a person j's PA at time t and consummatory pleasure at time t.γ<sup>00</sup> and γ<sup>10</sup> represent the average (or fixed-effects) estimates of consummatory pleasure and PA of the non-dysphoric group, while γ<sup>01</sup> and γ<sup>11</sup> represent group differences of consummatory pleasure and PA between the dysphoric and non-dysphoric groups.

### Concordance Between Anticipatory and Consummatory Pleasure

Finally, to investigate how much the anticipated pleasure for an activity is concordant with the amount of consummatory pleasure, time-lagged associations between anticipatory pleasure and consummatory pleasure were estimated with an autocorrelation approach. In particular, we conducted multilevel modeling to predict consummatory pleasure at sampling moment t from anticipatory pleasure at t-1 (see section "Model 3"). Additionally, consummatory pleasure at t-1 and PA at t were included as covariates separately (see sections "Model 4, 5") and simultaneously (see section "Model 6") to control for potential confounding effects. Moreover, to examine whether dysphoria moderate these associations, a group variable indicating dysphoric status was added as the Level 2 predictor to test for group differences between the dysphoric and non-dysphoric groups (see **Supplementary Materials** for Model equations and further details).

### Model 3

Level 1: consummatory pleasure ij = β0<sup>j</sup> + β1<sup>j</sup> (anticipatory pleasure <sup>t</sup>−1) + rij

Level 2: β0<sup>j</sup> = γ<sup>00</sup> + U0<sup>j</sup> β1<sup>j</sup> = γ<sup>10</sup> + U1<sup>j</sup>

### Model 4

Level 1: consummatory pleasure ij = β0<sup>j</sup> + β1<sup>j</sup> (anticipatory pleasure <sup>t</sup>−1) + β2<sup>j</sup> (consummatory pleasure <sup>t</sup>−1) + rij

Level 2: β0<sup>j</sup> = γ<sup>00</sup> + U0<sup>j</sup> β1<sup>j</sup> = γ<sup>10</sup> + U1<sup>j</sup>

β2<sup>j</sup> = γ<sup>20</sup> + U2<sup>j</sup>

### Model 5

Level 1: consummatory pleasure ij = β0<sup>j</sup> + β1<sup>j</sup> (anticipatory pleasure <sup>t</sup>−1) + β2<sup>j</sup> (PA <sup>t</sup>) + rij

Level 2: β0<sup>j</sup> = γ<sup>00</sup> + U0<sup>j</sup> β1<sup>j</sup> = γ<sup>10</sup> + U1<sup>j</sup> β2<sup>j</sup> = γ<sup>20</sup> + U2<sup>j</sup>

### Model 6

Level 1: consummatory pleasure ij = β0<sup>j</sup> + β1<sup>j</sup> (anticipatory pleasure <sup>t</sup>−1) + β2<sup>j</sup> (consummatory pleasure <sup>t</sup>−1) + β3<sup>j</sup> (PA <sup>t</sup>) + rij


In the equations, the outcome at Level 1 (consummatory pleasureij), representing a person j's score on consummatory pleasure items at time i, was modeled as a function of a random intercept (β0j) and random slopes (β1<sup>j</sup> , β2<sup>j</sup> , β3j). β0<sup>j</sup> represents person j's mean score for consummatory pleasure t; β1<sup>j</sup> represents the relationship between a person j's anticipatory pleasure at time t-1 and consummatory pleasure at time t; β2<sup>j</sup> represents the relationship between person j's current consummatory pleasure and their consummatory pleasure at the prior time point in Model 4 (current PA in Model 5). β3<sup>j</sup> represents the relationship between person j's current consummatory pleasure and current PA. rij represents residual errors at within-person level.

The Level-2 parameter estimates (γ00, γ10, γ20, γ30) represent the average (or fixed-effects) estimates across participants. U0<sup>j</sup> , U1<sup>j</sup> , U2<sup>j</sup> , and U3<sup>j</sup> represent residual errors at betweenperson level.

### RESULTS

### Self-Report Measures

There were no significant group differences between the dysphoric and non-dysphoric group on the TEPS total score, t(98) = 1.49, p = 0.14, anticipatory pleasure subscale, t(83.25) = 1.47, p = 0.15, or consummatory pleasure subscale, t(98) = 1.18, p = 0.24. Moreover, no significant group difference was observed on the SHAPS total score, t(98) = 0.65, p = 0.52. These results suggested that the dysphoric group demonstrated no attenuation of trait anticipatory or consummatory pleasure.

### ESM Estimates of State Anticipatory and Consummatory Pleasure

To determine whether age, gender, education, and time of assessments covaried with our dependent variables, we regressed these variables on consummatory pleasureij, anticipatory pleasureij, and PAij separately. Results showed that those dependent variables did not vary by age, gender, education, or time. Thus, these demographic variables and time were not included as covariates in further analyses.

The ICCs from the null models revealed that 39.9% of the variance in PA, 34.9% of the variance in state anticipatory pleasure and 39.4% of the variance in state consummatory pleasure was at the between-person level, suggesting that both the temporal situation and dysphoric status play a role in momentary PA and the experience of pleasure in daily life. Several multilevel models were tested to examine the distinct contribution of dysphoric status to momentary pleasure respectively.

No significant group difference was observed in PA between the dysphoric and non-dysphoric groups (p = 0.64). However, individuals in the dysphoric group reported lower levels of state anticipatory pleasure (β = −0.77, SE = 0.23, p < 0.01) and consummatory pleasure (β = −0.82, SE = 0.23, p < 0.01) than individuals in the non-dysphoric group. Moreover, group differences between the dysphoric and non-dysphoric group in state anticipatory and consummatory pleasure remained statistically significant after including PA as a covariate (**Table 2**).

### Impact of Anticipatory Pleasure on Consummatory Pleasure

Results of tests of Models 3–6, namely time-lagged analyses examining associations between anticipatory pleasure and changes in consummatory pleasure in the dysphoric and nondysphoric groups separately, are presented in **Table 3**. Model 3 tested consummatory pleasure at the current moment as a function of anticipated pleasure at the prior time point. We found that greater anticipatory pleasure was a significant predictor of greater follow-up consummatory pleasure in the dysphoric group (β = 0.15, SE = 0.04, p < 0.01), but not the nondysphoric group (β = 0.05, SE = 0.04, p = 0.22). In addition, the significant association between anticipatory pleasure at t-1 and

TABLE 2 | Estimates of positive affect (PA), anticipatory and consummatory pleasure in individuals with dysphoria compared with non-dysphoric individuals.


consummatory at t in the dysphoric group remained unchanged after controlling for consummatory pleasure <sup>t</sup>−<sup>1</sup> (Model 4), PA <sup>t</sup> (Model 5), or both of them (Model 6), all ps < 0.01. In the nondysphoric group, no significant relationship was found between anticipatory pleasure and consummatory pleasure, all ps > 0.05.

The moderating effects of dysphoria on these associations were not significant (all ps > 0.05, see **Supplementary Table S1**). Therefore, we rerun the models 3–6 by including all participants to determine if the predictive relationships between anticipatory pleasure and consummatory pleasure were statistically significant in the whole sample. The results showed that anticipatory pleasure alone positively predict current levels of consummatory pleasure at the trend level, β = 0.08, SE = 0.04, p = 0.07; after controlling for current PA, a significant association was found between anticipatory pleasure at t-1 and consummatory at t, β = 0.04, SE = 0.02, p = 0.02. In addition, the association between current PA and consummatory pleasure was also significant, p < 0.01, indicating that a higher level of anticipatory pleasure at prior time point in conjunction with higher current PA are associated with greater experience of consummatory pleasure for all participants.

### DISCUSSION

The aim of this study was to investigate state anhedonia in subclinical depression, taking into account trait anhedonia and current PA. ESM was utilized to track the momentary experience of anticipatory and consummatory pleasure over the course of 14 consecutive days in dysphoric and non-dysphoric individuals. Multilevel analyses showed that compared to the non-dysphoric individuals, the dysphoric individuals showed less pleasure during the anticipation of future events and the engagement of on-going activities, and this pattern remained unchanged after adding temporal PA as a covariate. However, no group differences between dysphoric and non-dysphoric individuals were observed regarding trait anhedonia. Moreover, a significant predictive relationship between state anticipatory and consummatory pleasure after adjustment for temporal PA was revealed in the whole sample. Investigating the distinct nature of anticipatory and consummatory pleasure in daily life may help us to understand the mixed findings of prior studies on anhedonia in depression.

Consistent with our hypothesis, dysphoric individuals, compared to non-dysphoric individuals, reported lower levels of pleasure during anticipation of an upcoming event. This finding is consistent with experimental studies investigating anticipatory responses to laboratory stimuli in MDD (McFarland and Klein, 2009; Admon and Pizzagalli, 2015) and in dysphoria (Yuan and Kring, 2009). For example, in the study of McFarland and Klein (2009), compared to healthy controls, patients with MDD reported less PA when anticipating monetary reward. Similarly, neuroimaging studies have found blunted response during anticipation at the ventral striatum, a core region involved in reward processing, both in patients with MDD (Arrondo et al., 2015; Keren et al., 2018) and in individuals at high risk for depression (Olino et al., 2014), suggesting dysfunctional

fpsyg-10-02124 September 17, 2019 Time: 17:34 # 6


TABLE 3 | Multilevel analyses of the time-lagged associations between consummatory pleasure, anticipatory pleasure and positive affect (PA) in dysphoric and non-dysphoric groups respectively and as a whole sample.

anticipatory processing in response to laboratory stimuli. The key finding that dysphoric individuals report less pleasure when anticipating future events is also consistent with research on depressed individuals' forecasts of the intensity of their PA. In general, people are over-optimistic about their future, with evidence showing that people have a tendency to overestimate the intensity and duration of PA than that they actually experience (Wenze et al., 2012; Morewedge and Buechel, 2013). This unrealistic prediction about emotional experience, though not accurate, is beneficial in maintaining mental health. With regard to evidence for biased predictions of PA in relation to depression, more severe depressive symptoms have been shown to be associated with less optimistic bias in PA prediction on a daily and a weekly basis both in the general population (Hoerger et al., 2012; Wenze et al., 2012, 2013) and in individuals with clinical and subclinical depression (MacLeod and Salaminiou, 2001; Chentsova-Dutton and Hanley, 2010). The same pattern of reduced anticipated PA has also been found in patients with remitted MDD (Thompson et al., 2017), suggesting that reduced anticipated positive experience might represent a traitlike marker in depression.

Only recently have researchers directly investigated depressed persons' anticipatory pleasure using ESM. Greater severity of subclinical symptoms of depression in the general population are found to be predictive of reduced pleasure during anticipating positive events in daily life (Bakker et al., 2017; Starr and Hershenberg, 2017). Our findings are also consistent with Wu et al.'s (2017) study in which patients with MDD demonstrated blunted state anticipatory and consummatory pleasure in relation to daily activities. On the other hand, Wu et al. (2017) also showed that patients with MDD reported levels of trait anticipatory and consummatory pleasure that were comparable to those reported by healthy controls. Self-report measures of trait anhedonia have been criticized for assessing responses to hypothetical situations that may not be comparable to actual situations, suggesting the importance of using multiple measures of anhedonia in any given study. ESM appears to be more sensitive than other methods of measuring the hedonic function of dysphoria. Given the transdiagnostic nature of anhedonia and its importance in depression, deficits in state anticipatory and consummatory pleasure might represent a core dysfunction of hedonic processing in people with dysphoria.

Daily PA is typically measured in general terms, with only limited research on the temporal experience of pleasure in response to daily events. Our finding that dysphoric and nondysphoric individuals did not differ in their mean level of PA is consistent with study of subclinical samples (Olino et al., 2014). Moreover, in our study, dysphoric individuals, compared to nondysphoric individuals, reported less consummatory pleasure in the course of the day. Consistent with our findings, previous research showed that patients with MDD are less responsive to positive stimuli in a laboratory task (Bylsma et al., 2008). Research with patients suffering from schizophrenia has also demonstrated a dissociation between PA and consummatory pleasure, with intact PA function but altered hedonic experience in patients compared to healthy controls (Edwards et al., 2018). In a study designed to distinguish PA and consummatory pleasure, Heininga et al. (2017) found that individuals with anhedonia reported a lower level of both PA and consummatory pleasure compared to individuals without anhedonia. Future research should further investigate the roles of the temporal experience of pleasure and current PA in depression, as well as their distinct contributions to the pathology of depression.

With regard to the predictive relationship between anticipatory and consummatory pleasure, separate models conducted within the two groups revealed significant associations between anticipatory pleasure and consummatory pleasure in the dysphoric group but not non-dysphoric group. However, models including dysphoria status as predictor did not support the moderating effect of dysphoria on these associations. Anticipatory pleasure was a significant predictor of consummatory pleasure in the whole sample, after adjustment for current PA. Thus, dysphoric and non-dysphoric individuals both report higher levels of consummatory pleasure when they anticipated greater pleasure at the previous time point, although dysphoria is associated with lower levels of anticipatory and

consummatory pleasure. This result is in line with the ESM study using end-of-day dairy reports, in which it was found that the increased anticipation for positive experience and positive events of the next day predicted greater reduction of depressed symptoms in dysphoria (Starr and Hershenberg, 2017).

Levels of anticipatory pleasure are predictors of effortexpenditure for rewards in healthy volunteers (Geaney et al., 2015), and it has been suggested that depression related anticipatory pleasure deficiency is associated with impairment in translating motivation to rewarding activities (Sherdell et al., 2012; van Roekel et al., 2016; Bakker et al., 2017). For example, Bakker et al. (2017) found that the more severe the depressive symptoms, the less likely that increases in reward anticipation are to be followed by increases in PA. Moreover, the positive relationship between reward anticipation and activity engagement in controls was also diminished in individuals with dysphoria. This suggests that dysphoric individuals were not able to translate their anticipation into goal-directed behaviors to pursue the anticipated reward, which might further prevent them from experiencing pleasure. The assumption of the mediating role of approach motivation in the translational mechanism from anticipatory pleasure to the experience of consummatory pleasure remains to be tested directly in laboratory and naturalistic settings. Although further validation is required, the predictive relationships between anticipatory and consummatory pleasure in dysphoria suggest that the dysfunctional reward anticipation represents a key component of anhedonia in relation to depression.

Together, these results suggest that the lack of anticipatory pleasure may be a dominant component of impairment that affects motivated behavior and subsequent experiential feelings in depression, and deficits in anticipatory pleasure might represent a critical therapeutic target requiring further investigation. Previous studies have provided preliminary evidence for the efficacy of interventions aimed at increasing anticipatory pleasure, schizophrenia patients reported higher levels of consummatory pleasure and more frequent engagement in daily activities at the end of training (Favrod et al., 2010; Nguyen et al., 2016). Future research could examine the extent to which benefits from interventions designed to improve depressed individuals' anticipation of pleasure could translate into downstream enjoyment of pleasurable experiences.

### Limitations

Our study has several limitations. Firstly, a disproportionate number of female participants were recruited, thus our results cannot be generalized to other population. Secondly, the estimated internal consistency of anticipatory subscale of TEPS is not as strong as that of consummatory subscale in present study. Anticipatory pleasure ratings are more likely to be affected by the context and therefore show less consistency than consummatory pleasure ratings (Edwards et al., 2015). In addition, the high homogeneity of college-student participants would reduce the variability in measurements (Peterson, 2001) and may affect score reliability. Thirdly, whether there are long-lasting and additive effects of anticipatory pleasure on consummatory pleasure will need to be established by further research, given evidence that anticipatory stressor together with current stressor predicted higher levels of negative affect 2.5 h later (Scott et al., 2018). Additional limitations to this study concern methodological problems in modeling time-lagged relationships, i.e., time-interval dependency (Kuiper and Ryan, 2018). Specifically, the event at t might not be exactly the one predicted at t-1, and thus the predictive relationship between anticipatory and consummatory pleasure might not reliably assess hedonic responses to the same events. To gain a better understanding of this correspondence, further research could increase the sampling density and include a more objective measure of the context in which individuals rate their experiences (e.g., by having participants indicate whether the current activity is the one that at the previous assessment they expected to be engaged in).

### CONCLUSION

This study makes an important contribution to the literature of anhedonia in depression by investigating the momentary level of anticipatory and consummatory pleasure during daily life activities in dysphoric and non-dysphoric individuals. The results showed that dysphoric individuals were characterized with less state anticipatory and consummatory pleasure compared with non-dysphoric individuals, irrespective of the level of daily PA. Moreover, a significant predictive relationship was found between anticipatory pleasure and consummatory pleasure in the whole sample. Our findings build on previous research highlighting hedonic deficits in depression, showing that measures of state rather than trait anticipatory and consummatory pleasure provide better estimates for hedonic processing function in dysphoria. These findings contribute to the ecological validation of the two dimension construct of anhedonia and provide specific targets that can be used to refine existing therapeutic interventions.

### DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

### ETHICS STATEMENT

The studies involving human participants were reviewed and approved by the Ethics Committee of the Central China Normal University. The patients/participants provided their written informed consent to participate in this study.

### AUTHOR CONTRIBUTIONS

XL generated the idea, designed and supervised the study, and wrote the first draft of the manuscript. Y-TZ and Z-JH helped with data collection, data analysis, and manuscript writing. X-LC and F-HY

helped with data collection and data analysis. X-JS commented significantly to the draft of the manuscript. All authors contributed to and have approved the final text.

### FUNDING

This study was supported by a grant from National Natural Science Foundation of China (31700957), MOE (Ministry of Education in China) Project of Humanities and Social

### REFERENCES


Sciences (17YJC190014), self-determined research funds of CCNU from the colleges "basic research and operation of MOE" (CCNU19TD018, CCNU16A05052).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg. 2019.02124/full#supplementary-material


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Li, Zhang, Huang, Chen, Yuan and Sun. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

fpsyg-10-02124 September 17, 2019 Time: 17:34 # 10

# Discriminating Fake From True Brain Injury Using Latency of Left Frontal Neural Responses During Old/New Memory Recognition

Jennifer Neal<sup>1</sup> , Stephanie Strothkamp<sup>1</sup> , Esias Bedingar1,2, Patrick Cordero<sup>1</sup> , Benjamin Wagner<sup>1</sup> , Victoria Vagnini1,3 and Yang Jiang<sup>1</sup> \*

<sup>1</sup> Department of Behavioral Science, University of Kentucky College of Medicine, Lexington, KY, United States, <sup>2</sup> Harvard T.H. Chan School of Public Health, Boston, MA, United States, <sup>3</sup> Louisville VA Medical Center, Louisville, KY, United States

Traumatic brain injury (TBI) is a major public health concern that affects 69 million individuals each year worldwide. Neuropsychologists report that up to 40% of individuals undergoing evaluations for TBI may be malingering neurocognitive deficits for a compensatory reward. The memory recognition test of malingering detection is effective but can be coached behaviorally. There is great need to develop a novel neural based method for discriminating fake from true brain injury. Here we test the hypothesis that decision making of faking memory deficits prolongs frontal neural responses. We applied an advanced method measuring decision latency in milliseconds for discriminating true TBI from malingerers who fake brain injury. To test this hypothesis, latencies of memoryrelated brain potentials were compared among true patients with moderate or severe TBI, and healthy age-matched individuals who were assigned either to be honest or faking memory deficit. Scalp signals of electroencephalography (EEG) were recorded with a 32-channel cap during an Old/New memory recognition task in three age- and education-matched groups: honest (n = 12), malingering (n = 15), and brain injured (n = 14) individuals. Bilateral fractional latencies of late positive ERP at frontal sites were compared among the three groups under both studied (Old) and non-studied (New) memory recognition conditions. Results show a significant difference between the fractional latencies of the late positive component during recognition of studied items in malingerers (averaged latencies = 396 ms) and the true brain injured subjects (mean = 312 ms) in the frontal sites. Only malingers showed asymmetrical frontal activity compared to the two other groups. These new findings support the hypothesis that that additional frontal processing of malingering individuals is measurably different from those of actual patients with brain injury. In contrast to our previous reported method using difference waves of amplitudes at frontal to posterior midline sites during new items recognition (Vagnini et al., 2008), there was no significant latency difference among groups during recognition of New items. The current method using delayed left frontal neural responses during studied items reached sensitivity of 80% and specificity of 79% in detecting malingers from true brain injury.

Keywords: malinger, event-related potentials, EEG, traumatic brain injury, P3, late positive component, fractional peak latency

### Edited by:

Ya Zheng, Dalian Medical University, China

### Reviewed by:

Aiqing Nie, Zhejiang University, China Robert Perna, Nationaal Archief of the Netherlands, Netherlands

> \*Correspondence: Yang Jiang yjiang@uky.edu

### Specialty section:

This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Neuroscience

Received: 10 April 2019 Accepted: 02 September 2019 Published: 25 September 2019

### Citation:

Neal J, Strothkamp S, Bedingar E, Cordero P, Wagner B, Vagnini V and Jiang Y (2019) Discriminating Fake From True Brain Injury Using Latency of Left Frontal Neural Responses During Old/New Memory Recognition. Front. Neurosci. 13:988. doi: 10.3389/fnins.2019.00988

### INTRODUCTION

fnins-13-00988 September 27, 2019 Time: 7:40 # 2

Traumatic brain injuries (TBIs) are attributed to 30% of all injury deaths (Taylor et al., 2017) and affect up to 69 million individuals each year worldwide (Dewan et al., 2018). People affected by TBI can suffer from the deficits of their injury for the rest of their life, potentially causing impaired memory, sensation, thinking, movement, and mood swings (Gerberding and Binder, 2003). As severe a health concern as TBI is, it is estimated by neuropsychologists that up to 40% of individuals undergoing evaluations following TBI may be malingering deficits in order to gain a compensatory reward (Mittenberg et al., 2002). Often, those who malinger, or exaggerate symptoms of TBI can be identified by intentional poor performance on cognition tests. Studies have found that, in test seeking compensation, those with mild TBI often exhibited poorer effort and worse cognitive performance than those with moderate or severe TBI (Green et al., 2001). Uncertainty in the legitimacy of the deficits of many patients affected by TBI points to a need for development of a test to screen TBI individuals to validate their deficits, while identifying malingerers.

In recent years, many studies have attempted to find effective methods to distinguish malingering behavior. Sollman and Berry (2011) conducted a large meta-analysis of detection of inadequate effort in neurophysiological testing, which included a group of 21 studies testing memory malingering. Another study has indicated the possibility that those who malinger memory deficits can be identified by the measurable physiological differences of pupil dilatation (Heaver and Hutton, 2011). Research has pointed to identification of malingerers through results of individuals undergoing new and unique testing methods based on subject performance or data processing (McBride et al., 2011; Liu et al., 2016). A classic study examined the measurement of response latency to identify malingerers when undergoing the Portland Digit Recognition Test. Using response latency, researchers were able to successfully classify 74% of malingerers (Rose et al., 1995). The success of this study implicates mental processing time and neural latencies in detection of malingerers. The methodology of using event-related potential (ERP) data to clinically differentiate malingers from those with TBI is fairly rare. If an effective method to distinguish malingering behavior from those with TBI is found, healthcare professionals will be better prepared to treat patients with the appropriate level of care. The Test of Memory Malingering (TOMM) for malingering detection is effective (Tombaugh, 1996; Kanser et al., 2019), but can be coached behaviorally. Finding new methods of identifying malingerers is a significant area of research that holds promise for the healthcare community.

Using combined methods of ERP and reaction time (RT), Vagnini et al. (2008) developed neural and behavioral methods to identify malingerers from TBI patients. The electrophysiological activity was collected using an electroencephalography (EEG) cap during. The TOMM task is a computerized method to test a subject's memory of images shown to them. Stimulus pictures were shown on a computer screen about 65 cm from the subject. The images themselves were 8 by 10 cm on a white background with a black border. The TOMM task is able to distinguish those who feign memory impairment from those with legitimate memory impairment. If a subject's score on the TOMM task is low, it suggests an exaggeration of memory impairment symptoms (Tombaugh, 1996).

Event-related potential data are averaged EEG signals that are useful for memory task analysis because the memory recognition events were time-locked to studied (Old) and New items (Finnigan, 2002). Mean ERP amplitudes for malingerers appeared to be reduced compared to those of honest or TBI subjects. Research has documented the abnormalities of ERP data within EEG signals of those with TBI. Certain character of ERP markers is linked to TBI that impact upon many cognitive functions, including processing speed, sustained attention, performance monitoring, inhibitory control, and cognitive flexibility (Dockree and Robertson, 2011). A significant component to this particular event within the EEG signal is the P3 component. The P3 component correlates to decision making and cognition when presented with a stimulus (Patel et al., 2005). The P3a component has been found to have potential to differentiate between those with TBI and those who malinger. Motivations or overt performances to feign brain injury cannot change the character of the P3a component to match that of brain injured individuals, which sets malingerers apart (Hoover et al., 2014).

In using the convenient sample, group differences were compared using advanced fractional latency methods to test a new hypothesis that decision-making of a faker needs additional frontal processing (Tombaugh, 1997). Vagnini et al. (2008) paper utilized complicated analysis of amplitudes of frontal to posterior midline electrodes, while this study focused on latency analysis of lateralized frontal electrodes not previously examined. In comparing the latencies of each subject group, significant differences in neural processing speed can be identified and attributed to the intention to malinger deficits. In contrast to combining RTs and amplitudes of multiple midline electrodes of differences waveforms, latency specific results reveal delayed decision of MNCD could indicate significant markers to identify malingering individuals.

Here we further developed a method measuring latency of neural responses in milliseconds for discriminating true TBI from malingerers who fake brain injury. We test the hypothesis that decision-making of faking memory deficits at each visual item prolongs neural responses during memory recognition.

### MATERIALS AND METHODS

### Participants

The behavioral and EEG data were collected from 47 age- and sex-matched individuals, which were approved by the medical IRB in the University of Kentucky. The control group was healthy, honest participants (HON) with no history of brain injury instructed to perform the task honestly to the best of their ability (mean age = 36.2; n = 16). The second group (MAL) was healthy individuals with no history of brain injury instructed to malinger deficits of TBI while undergoing the task (mean age = 32.7; n = 16). The final group consisted of patients with

reported TBI instructed to perform the task honestly to the best of their ability (mean age = 40.5; n = 15) (Vagnini et al., 2008). Two participants from the TBI, one from the malingering, and four from the honest group were excluded due to excessive artifacts of EEG signals. Frontal, lateralized electrodes have more muscle artifacts compared to those at the midline electrodes. The TBI individuals ranged from moderate to severe TBI. Medical records indicate that the TBI group had a mean emergency room Glasgow Coma Scale score of 8.7 (SD = 2.9), a mean duration of loss of consciousness of 7.2 days (SD = 12.0), were an average of 13 years post-injury (SD = 7.2), and the majority (73%) were injured in moving vehicle accidents. CT and MRI scans indicated brain injury in varied locations from the brain stem, frontal, temporal, occipital, and parietal in both the left and right hemisphere (Vagnini et al., 2008).

### Procedure

The study employed a 32-electrode EEG cap on subjects while undergoing the Old/New Memory Task. Participants' performances (accuracy and reaction times) were recorded along with EEG scalp signals. Data were recorded using Neuroscan 4.5 and analyzed using EP Toolkit 2.0. This was done by comparing the results from detecting MNCD to the results of the established testing method (TOMM-C) (Tombaugh, 1996, 1997; Vagnini et al., 2008; Kanser et al., 2019).

### Task

The Old/New task began with a study phase of 100 New drawings. Stimulus pictures were displayed on a computer screen, which were presented for 5 s each during the study phase, and participants were instructed to memorize each picture. After a short break, all 100 pictures were studied again for a second time. After studying the pictures, the participants entered the test phase. Participants viewed 140 pictures, presented one time (70 old and 70 foils not yet presented to the participant). For each picture, the participant decided whether the drawing was "New" or "Old" and clicked a corresponding key on the keyboard (**Figure 1**).

### Data Analysis

Standard EEG preprocessing were performed (e.g., removing artifacts). They have been reported in detail previously (Vagnini et al., 2008). Here, we picked the largest ERP components, P3 or late positive component for latency analysis. The research done by the ABC lab has focused on the P3 component of the ERP data collected. By focusing on the P3 component of ERP data rather than the mean ERP data alone, more specific results relating to the decision of MNCD could indicate significant markers to identify malingering individuals (Levada et al., 2016).

The latency analysis utilized MATLAB in combination with the extension EEGLAB and ERPLAB (Lopez-Calderon and Luck, 2014). Bilateral anterior and posterior sites were selected for analysis of fractional peak latency, which measures latency by finding the peak amplitude and then working backward in the waveform until 50% of that peak voltage is reached. Compared to simple peak latency, this is a better method that is optimal for finding onset latency and allows for most accurate results. The peak measures were tested for between group differences with a one-way ANOVA with a significance at the 0.05 level.

To further examine the implications of peak latency differences between groups, amplitudes from −200 to 600 ms at potentially significant electrodes were also examined. Significant electrodes areas were visualized from the development of scalp topographic maps of subjects over the same time frame. The topographic maps were created based on grand averaged data of all subjects within a testing group, done with MATLAB.

### RESULTS

To test the hypothesis of frontal manipulation among healthy individuals faking brain injury, we examined latencies of P3 in several bilateral frontal electrodes (i.e., FP1, FP2, F3, and F4; See **Figure 2**). Significant group differences were found with the Old (studied) memory recognition at these frontal electrodes. In using fractional peak latency analysis on these electrodes, the fractional peak latency for each subject group was compared to the grand averaged voltage data of brain activity over the −200 to 600 ms time frame for each group. This analysis allowed for visualization of the differences in peak latency between subject groups during Old (studied) memory recognition (**Figure 3**). As well as visualization, the data was tested for significance through a one-way ANOVA with a significance at the 0.05 level. The statistical analysis yielded results of significant differences in peak latency between MAL and TBI groups at the FP1, F3, and F4 electrodes for the Old condition only (**Table 1**). The largest latency differences between true TBI and malingerers are at the F3 site. The malingerers of memory deficits are on average 88 ms longer in the left frontal site. Latencies were also examined at occipital electrodes, but no significant group differences were found.

(non-studied) images.

Topographic maps created with the resultant data indicated the brain regions with the highest average amplitude of activity during testing for subject groups, identified by dark red regions on the **Figure 4**. The information gathered from the topographic maps indicated that significant electrodes for analysis were located in the frontal and visual cortices as shown.

The significance of the latency differences can also be visualized through scatter plot (**Figure 5**). Each group is represented by color (red = malingering, green = brain injured) with each point representing an individual subject. Honest subjects were not visualized in the figure because the aim is to differentiate malingerers from brain injured individuals. The solid black line at 361 ms represents the threshold of significant group differences in latency. The red points to the right of the line represent true positive values as they are delayed malingering latencies. The red marks to the left of the line represent false negative values as they are malingerers without significantly delayed latencies. The green marks to the left of the line represent the true negative values as they are traumatic brain injured individuals with no delay in latency. The green marks to the right of the line represent false positive values as they are brain injured individuals with delayed latencies. Using these values, the sensitivity or hit rate was calculated to be 80%, meaning that in using delayed latencies, 80% of malingering individuals will be positively identified. The specificity was also calculated with these values and found to be 79%.

### DISCUSSION

We report new findings that the left frontal neural responses during recognition decision of studied visual stimuli are significantly delayed in malingerers compared to those in true patients with traumatic brain injured. The results indicate an averaged delay of 396 ms for malingerers compared to a 312 ms averaged delay for TBI individuals, marking an 84-ms difference in cognitive processing between the two groups. The results also indicate honest individuals using primarily bilateral frontal engagement when viewing both Old and New images. In contrast, malingering individuals engaged right frontal and left occipital regions in response to both New and Old images. Individuals with traumatic brain injuries engaged in distributed cortices: frontal, parietal, and mostly the right occipital visual cortex in response to both New and Old images. These differences in regional engagement between test groups are most evident from the scalp topographic maps and indicate notable differences in brain activity, not only between healthy and brain injured individuals, but also between individuals responding honestly to stimuli versus those malingering deficits.


<sup>∗</sup>p < 0.05.

The decision factors about ERP old/new effects was found to be associated with the late positive components (LPC) responses which had a left > right, centro-parietal scalp topography (Finnigan, 2002). Thus our analysis focused on lateralized electrodes. We found that only maligning group had asymmetrical frontal activity during their decision of whether to lie or not about an old-item. This frontal engagement might be the involvement of the working memory required to plan and exhibit TBI-like memory failure. Interestingly, during a modified delayed match-to-sample task, left-frontal memory related potentials during the working memory task discriminated healthy older adults and those with mild cognitive impairment patients. The LPC in the right frontal ERPs were statistically identical between normal older adults and those with early Alzheimers' disease (Li et al., 2017). Fletcher and Henson (2001) determined there are two types of working memory tasks: "delayed matching tasks" and "self-ordered tasks." The faking/fringing TBI process may require a malinger to determine if a probe stimulus matches a stimulus held in their memory similar to Old/New task in the experiment and then a self-ordered task to be honest or not for this item (Fletcher and Henson, 2001). Both the honest and TBI individuals performed to the best of their ability and merely answered whether they had seen the image previously. In contrast, the malingerers' responses required different brain engagement because of their conscious effort exhibit TBI-like behavior, which is more similar to the "self-ordered tasks." According to Petrides, individuals performing "delayed matching tasks" show engagement in the ventrolateral frontal cortex, while those performing "self-ordered tasks" show engagement in the dorsolateral frontal cortex (Petrides, 1995). These differing areas of engagement show further distinction between the neural responses of malingerers to that of honest and TBI individuals.

The left occipital engagement illustrates the visual processing of the malingerers as they viewed the images presented and determined if they had seen it before (Sehatpour et al., 2008). This engagement differs from honest participants whose engagement was focused mainly frontal bilaterally implying that honest participants engaged frontal-occipital communications differently from the malingerers because they were attempting to determine the correct categorization of the image (Old/New) while malingerers were less concerned with accuracy and more so with exhibiting a TBI-like performance.

Our present findings demonstrate a simpler way to measure neural delay that is harder to fake, which may lead to better clinical identifications of true TBI individuals from those who malinger deficits. The ERP results illustrate that the Old/New Memory task can provide clinicians with distinguishable markers in brain activity to differentiate malingerers from those with legitimate TBI. Although this form of testing is not immediate and requires the subject to perform the memory task, it yields quantifiable results to accurately identify TBI individuals and allow for them to get proper treatment without concern of exaggeration or malingering.

Although the results of analysis of P3 signatures found promising results, research has found that splitting the P3 signature into two components, P3a and P3b, could yield results more tailored to specific events (Polich, 2007). The P3a component deals specifically with detection of a stimulus, an involuntary response, while the P3b response is the conscious task-relevant processing of the stimulus (Hoover et al., 2014). Our results are consistent with the P3b component. In isolating specifically, the P3a component, comparisons

line represents the delayed latency threshold at 361 ms.

can be made between malingerers and TBI individuals that can identify malingerers. Individuals attempting to malinger cognitive impairment could not simulate comparable P3a deficits seen in those with legitimate TBI. Abnormalities of these components may signal other mental disorders (Hoover et al., 2014; Bachiller et al., 2015). Because P3a is an involuntary reaction, differences in this component specifically can be a promising identifier of malingerers.

The findings are promising, but there are several limitations to the research. First, cross-validations with independent samples are important for this type of application. Second, the current method is limited in differentiating a liar from a TBI patient, or from healthy honest, but it is not clinical diagnosis test in TBI patient. Also, the sample size of each group (12–15 individuals) is small and presents problems in attempts to generalize these results to a larger population without significant effect size. Additionally, the sample size is too small to examine sex differences in brain responses during decision-making.

Recent advancement of EEG recording makes EEG screening wireless and easier to use in the clinics. Technological advances have made the use of EEG testing and ERP analysis more accessible in clinical settings. These results illustrate the possibilities of the use of ERP analysis in TBI vetting for future studies. This experiment is the early stages of more promising and expansive results. New, independent sampling and data collection is needed to further validate these findings and achieve concrete predictive values for those with TBI and those who malinger.

In practice, identification of those malingering deficits of TBI can be useful for not only healthcare professionals, but also those involved in insurance and legal processing. Vallabhajosula (2015) discusses the implications of employing neuroscience in criminal law, specifically detailing malingering and its assessment. Malingering can have legal implications where people are able to lie or exaggerate symptoms to avoid criminal conviction or military service. It is difficult for legal professionals to identify malingerers without proof because they can be accused of defamation by the potential malingerer (Weiss and Van Dell, 2017). The consequences of malingering are great; for military settings, those malingering deficits of injury or disability to avoid military service are subject to court-marshal and punishment (Malingering 83 U. S. C. §. 883, 2016). The promising findings of definitive methods of identification of malingerers can have great use to identify those malingering deficits to avoid legal responsibilities. With effective testing allowing for differentiation between TBI individuals and those malingering deficits, neurological signatures identified through research can help identify dishonest individuals. These techniques can be put into practice in court proceedings to distinguish honest individuals from those providing false testimony.

The present results contribute to future studies in developing combined methods of differentiation between TBI individuals and malingerers. For instance, machine learning type of classifications applying frontal latencies, frontal-parietal amplitudes and task performance (accuracy and reaction times) will greatly improve the precision. Previous study has found that larger P3 amplitudes correlate to faster behavioral responses, but peak amplitude latencies do not differ for behavioral reaction times (Ramchurn et al., 2014). Using fractional peak latency to compare the P3 signatures to reaction times could yield promising results. This method could allow for another form identification of possible differences in behavioral markers during memory tasks that can differentiate test groups. Research could also be useful in exploring more detailed identification of TBI to differentiate those with mild TBI versus those with severe. The Glasgow Coma Scale was developed to determine the level of consciousness of a person after a TBI. On the scale, a score of 13–15 is classified as mild, 9–12 as moderate, and 8 or less as severe (Teasdale and Jennett, 1974). Categorizing the TBI participants in the study could yield more specific results based on the severity of their brain injury. Studies have found that mild TBI results in prolonged P3 latencies at central electrodes compared to healthy individuals (Nandrajog et al., 2017). In contrast, the results found in this study show that the moderate to severe TBI individuals have early onset peak P3 latencies compared to healthy individuals. Differences in peak latencies could indicate a discernable pattern in brain activity based on the severity of TBI. In using methods to better detail the extent of TBI, patients would be able to receive more appropriate and tailored care for their level of injury.

### DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

### AUTHOR CONTRIBUTIONS

JN was responsible for writing the manuscript and performed some data analysis and interpretation. SS was involved in the data analysis of the fractional latency and interpretation as well as helped with the writing. EB contributed to the data analysis and drafting the manuscript. PC was aided with data analysis and interpretation. BW was responsible for early data analysis and interpretation. VV designed and collected the clinical and EEG data as part of her Ph.D. dissertation. YJ contributed to the study design, EEG data collection, analysis, interpretation, and writing the manuscript.

### FUNDING

This work was supported in part by the grant funding from the Henry M. Jackson Foundation and the National Institute on Aging to YJ (1R56AG060608).

### ACKNOWLEDGMENTS

We thank Dr. S. McIlwarth for helping discussion.

### REFERENCES

fnins-13-00988 September 27, 2019 Time: 7:40 # 8


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Neal, Strothkamp, Bedingar, Cordero, Wagner, Vagnini and Jiang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Mediating Role of Coping Styles on Impulsivity, Behavioral Inhibition/Approach System, and Internet Addiction in Adolescents From a Gender Perspective

Qi Li1,2, Weine Dai1,3,4,5, Yang Zhong1,2, Lingxiao Wang1,2, Bibing Dai<sup>6</sup> \* and Xun Liu1,2

<sup>1</sup> Chinese Academy of Sciences, Key Laboratory of Behavioral Science, Institute of Psychology, Beijing, China, <sup>2</sup> Department of Psychology, University of Chinese Academy of Sciences, Beijing, China, <sup>3</sup> Center of Functionally Integrative Neuroscience and Positron Emission Tomography Center, Aarhus University, Aarhus, Denmark, <sup>4</sup> Sino-Danish College, University of Chinese Academy of Sciences, Beijing, China, <sup>5</sup> Sino-Danish Center for Education and Research, Beijing, China, <sup>6</sup> Institute of Psychology, Tianjin Medical University, Tianjin, China

### Edited by:

Daniela M. Pfabigan, University of Oslo, Norway

### Reviewed by:

Oswald David Kothgassner, Medical University of Vienna, Austria Anna Felnhofer, Medical University of Vienna, Austria

> \*Correspondence: Bibing Dai daibibing@tmu.edu.cn

### Specialty section:

This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Psychology

Received: 05 June 2019 Accepted: 08 October 2019 Published: 24 October 2019

### Citation:

Li Q, Dai W, Zhong Y, Wang L, Dai B and Liu X (2019) The Mediating Role of Coping Styles on Impulsivity, Behavioral Inhibition/Approach System, and Internet Addiction in Adolescents From a Gender Perspective. Front. Psychol. 10:2402. doi: 10.3389/fpsyg.2019.02402 Previous findings have shown that impulsivity and Behavioral Inhibition/Approach System (BIS/BAS) have substantial effects on adolescents' Internet addiction, but the mechanisms underlying these associations and gender differences in these effects have received little attention. We examined the mediating effects of coping styles from impulsivity, and BIS/BAS to Internet addiction as well as gender differences in these associations. A total of 416 Chinese adolescents were examined using a cross-sectional survey involving Young's Diagnostic Questionnaire for Internet Addiction, Barratt Impulsiveness Scale, BIS/BAS scales, and Coping Style Scale for Middle School Students. The data were analyzed using the independent sample t-test, chi-square test, Pearson correlation, and structure equation modeling. The results from the multiplegroup (by adolescent gender) structural model analysis revealed that both impulsivity (p < 0.001) and BIS (p = 0.001) directly predicted positive Internet addiction in girls, while both impulsivity (p = 0.011) and BAS (p = 0.048) directly predicted positive Internet addiction in boys. Furthermore, emotion-focused coping mediated the relationship between impulsivity and Internet addiction (β = 0.080, 95% CI: 0.023–0.168) and the relationship between BIS and Internet addiction (β = 0.064, 95% CI: 0.013–0.153) in girls, while in boys, problem-focused coping and emotion-focused coping mediated the association between impulsivity and Internet addiction (β = 0.118, 95% CI: 0.031– 0.251; β = 0.065, 95% CI: 0.010–0.160, respectively) and problem-focused coping mediated the association between BAS and Internet addiction [β = −0.058, 95% CI: (−0.142)–(−0.003)]. These findings extend our insight into the mechanisms underlying the associations among impulsivity, BIS/BAS, and Internet addiction in adolescents and suggest that gender-sensitive training approaches to decrease adolescents' Internet addiction are indispensable. These interventions should focus on the different gender predictors of adolescent Internet addiction and on the development of specific coping styles for boys and girls respectively.

Keywords: adolescents, impulsivity, behavioral inhibition/approach system, coping styles, Internet addiction, gender differences

### INTRODUCTION

fpsyg-10-02402 October 22, 2019 Time: 19:11 # 2

With the speedy development of Internet technology in recent years, people increasingly use the Internet, especially adolescents. Although the growth in Internet use facilitates adolescent life in many ways, excessive Internet use can lead to Internet addiction (Choi et al., 2009; Ko et al., 2012). Internet addiction was defined as a subset of behavioral addictions that possess the core components of addiction, such as salience, tolerance, and withdrawal (Griffiths, 2000), while Shapira et al. (2000) described Internet addiction as an impulse control disorder. In 2008, Shaw and Black (2008) further refined the conception of "Internet addiction" as "excessive or poorly controlled preoccupations, urges or behaviors regarding computer use and Internet access that lead to impairment or distress." In consideration of the variety of terminologies, the lack of consistency about the conceptualization and the diagnosis of Internet addiction, Sim et al. (2012) have suggested that the most reliable and valid criterion for conceptualizing Internet addiction is to adapt the Diagnostic and Statistical Manual of Mental Disorders, 4th Edition (DSM- IV) criteria for pathological gambling. Consistent with this notion, Young defined Internet addiction as an individual's inability to control the impulse to use their Internet use, which eventually leads to psychological, social, educational, and/or occupational problems (Young, 1998). Furthermore, she confirmed 8 symptoms of Internet addiction according to the criteria for pathological gambling in the DSM-IV: tolerance, preoccupation, withdrawal symptoms, unsuccessful attempts to decrease use, continued excessive use, compromise or loss of a significant relationship and social activities, lying about online activity, and use of the Internet to self-medicate. Meanwhile, she developed a brief eightitem Diagnostic Questionnaire for Internet Addiction (YDQ) to assess Internet addiction (Young, 1998). Although Internet addiction has not yet been classified as a disorder, neither in the DSM-V nor in the eleventh edition of the International Classification of Diseases (ICD-11), most of the items in YDQ directly corresponded to the nine diagnostic criteria for Internet Gaming Disorder in the DSM-V (American Psychiatric Association [APA], 2013). Furthermore, "chemical" addiction showed no difference with "behavioral" addiction according to DSM-V addiction criteria, and DSM-V paid more attention to personal experiences rather than drug types (American Psychiatric Association [APA], 2013). Subsequently, Internet Gaming Disorder was included as "Disorders due to addictive behaviors" rather than as an "Impulse Control Disorder" in the ICD-11 (World Health Organization [WHO], 2018).

Adolescence is a critical stage in life cycle, and can be defined as a transitional period from childhood to adulthood during which individuals experience major biological, cognitive, and socioaffective changes (Dumont and Provost, 1999). Consequently, adolescents have to cope with the most stressful life-events and the challenges caused by these changes. They face specific developmental tasks including identity construction, personal autonomy, and the redefinition of relationships with adults and peers (Borca et al., 2015). Internet addiction is highly prevalent and causes more harmful consequences among adolescents than among adults in many countries due to their active psycho-social and personality development (Morrison and Gore, 2010; Spada, 2014; Stavropoulos et al., 2017). The data from a nationally representative sample of Chinese adolescents indicated that the percentage of Internet addicts in the total sample (Internet users and non-Internet users) was 6.3% (1,523/24,013), while among Internet users, it was 11.7% (1,523/12,993) (Li et al., 2014). In addition, the prevalence of Internet addiction is much higher in Asian countries (e.g., China) than in Western countries (Li et al., 2018). Moreover, Internet addiction is significantly correlated with numerous negative consequences in adolescents, such as psychiatric disorders (e.g., depression, anxiety and obsessive-compulsive specifications), physical problems, and poor academic performance (Ko et al., 2009; Salmela-Aro et al., 2017; Przepiorka et al., 2019). More importantly, addiction and its' negative influences in adolescents could continue into adulthood (Englund et al., 2008; Stavropoulos et al., 2017). Thus, it is very important to study Internet addiction among adolescents. To promote the prevention of and early intervention for Internet addiction, it is imperative to identify risk factors and underlying mechanisms for Internet addiction in adolescents, especially among those in Asian countries.

The dual-system neurobiological model tries to explain high risk-taking behaviors in adolescents, such as substance abuse, pathological gambling, Internet gaming disorders, and so on (Casey et al., 2008). It proposes that the differential development of the reward-seeking and impulse control system in adolescents, which show heightened reward-seeking and deficient impulse control relative to children and adults, might be one of the most important factors contributing to the high incidence of adolescents' risky behaviors (Steinberg, 2010). This model goes against traditional explanations, which states that risky behaviors in adolescents are mainly due to a lag in the development of the prefrontal control system and advocates to combine the development of the subcortical reward-seeking system with that of the prefrontal control system (Rubia et al., 2000, 2006; Tamm et al., 2002; Yurgelun-Todd, 2007). In line with the dual-system model, many neurobiological models indicate that the reward system and impulse control system have equal importance in terms of accounting for adolescent's risky behaviors, including substance abuse and problematic gambling (Casey et al., 2008; Steinberg, 2008; Somerville and Casey, 2010). However, few studies have simultaneously explored the characteristics of reward processing and impulse control in the context of Internet addition among adolescents to test and extend the dual-system model for this condition.

### Impulsivity, BIS/BAS and Internet Addiction

Impulsivity is defined as a predisposition that leads to the tendency to behave prematurely and without foresight in ways that are undesirably dangerous or unsuitable to the situation (Robbins et al., 2012). Impulsivity is often associated with the inhibitory control systems due to an immature frontal lobe that causes adolescents to be at particularly high risk for Internet

addiction (Crews and Boettiger, 2009; Brand et al., 2014). Previous studies have revealed that the high level of Internet addiction among adults is positively associated with impulsivity (Meerkerk et al., 2010; Zhang et al., 2015), and adolescents with Internet addiction exhibit increased impulsivity and reduced inhibitory control capacity compared with controls (Cao et al., 2007; Choi J. S. et al., 2014; Choi S.-W. et al., 2014; Bargeron and Hormes, 2017). Many researchers have emphasized the key role of impulsivity in Internet addiction and have argued that impulsivity is an important risk factor for developing Internet addiction and a marker of susceptibility to Internet addiction (Lee et al., 2012; Wu et al., 2013; Li et al., 2016).

Gray's neuropsychological reinforcement sensitivity theory states that behavior originates from activity in at least two basic dimensions of motivation, which are independent and based on biological systems (Gray, 1994; Bijttebier et al., 2009). These dimensions reveal the function of two brain systems that govern approach and avoidance behaviors in response to different types of stimuli. The behavioral approach system (BAS) is responsible for mediating reactions to all conditioned and unconditioned appetitive stimuli and is associated with the enhancement of reward or the termination of punishment. The BAS is associated with reward seeking and high levels of BAS activation indicate higher sensitivity to reward dependence and novelty processing (Li et al., 2019). The behavioral inhibition system (BIS) has been postulated to be sensitive to stimuli of punishment or the termination of a reward. The BIS is associated with the avoidance of potentially negative or harmful consequences and high levels of BIS activation imply a proneness to loss avoidance and a tendency to display a blunted response to reward (Li et al., 2019). Although Gray's neuropsychological reinforcement sensitivity theory provided an important view for understanding and explaining addiction, previous results about the associations among BIS/BAS and Internet addiction were inconsistent both among adults and among adolescents. The BIS is neither directly nor indirectly associated with Internet addiction, whereas the BAS is associated with Internet addiction only through depression and social anxiety in adults (Fayazi and Hasani, 2017), while another study found that neither the BAS and nor BIS were related to Internet addiction in adults (Meerkerk et al., 2010). In terms of adolescents, high BAS activation rather than BIS activation could predict the occurrence of Internet addiction (Yen et al., 2012), while another study indicated that the BIS activation not the BAS activation was a significant predictor of Internet addiction (Park et al., 2013). Further, two previous studies reported that both high BAS and high BIS activation were associated with Internet addiction (Giles and Price, 2008; Nam et al., 2018). Although these results from adolescents could indicate that Gray's neuropsychological reinforcement sensitivity theory can help to understand and explain Internet addiction, the inconsistent results require more research to explore the associations between BIS/BAS and Internet addiction. In sum, few studies have explored the roles that impulsivity and BIS/BAS play in Internet addition among adolescents, especially on the basis of the dual-system neurobiological model of addiction. Furthermore, although most previous studies have shown that impulsivity and BIS/BAS contribute to Internet addiction among adolescents, little is known about the mediating and moderating mechanisms underlying these associations.

### The Mediating Role of Coping Style

Coping style refers to people's behavioral and cognitive attempts to manage specific external and/or internal demands under stress (Skinner et al., 2003). In general, coping styles can be divided into problem-focused and emotion-focused coping. Problemfocused coping refers to strategies that are directed to address the problems that cause emotional distress (e.g., problem solving, seeking help, and cognitive restructuring), whereas emotionfocused coping refers specifically to strategies that palliate negative emotions (e.g., wishful thinking, denial, and withdrawal behavior) (Compas et al., 2001). Because both coping styles show contextual and process-oriented features, they can change over time and under different circumstances (Schoenmakers et al., 2015). Furthermore, how coping styles are used depends on how an individual interprets the stressor, and different coping styles could cause different results. Taken together, adolescents with high problem-focused coping tend to find appropriate methods to address their difficult circumstances, accompanied by better adjustment (Jackson et al., 2017), while adolescents with high emotion-focused coping tend to avoid their own problems passively, accompanied by maladjustment (Carlo et al., 2012). In previous studies, researchers noted that the two types of coping should be viewed as different constructs of coping instead of opposite poles (Patterson and McCubbin, 1987; Compas et al., 2001). Although there is no direct empirical evidence supporting that coping mediates the relationships among impulsivity, BIS/BAS and Internet addiction in adolescents, some indirect evidence has implied that coping styles play a mediating role in these associations. On one hand, previous studies have found that adolescents with high impulsivity are more likely to use emotionfocused coping but less likely to engage in problem-focused coping (Connor-Smith and Flachsbart, 2007; Lee-Winn et al., 2016). The BAS not the BIS was associated with problem-focused coping, and problem-focused coping mediated the relationship between the BAS and adolescent delinquent behavior, while another study found that emotion-focused coping mediated the relationship between the BIS and adolescent problematic alcohol use (Hasking, 2007; Willem et al., 2012). On the other hand, both high emotion-focused coping and low problem-focused coping were associated with adolescent Internet addiction, while another study found that only emotion-focused coping, not problem-focused coping, increased the risk of Internet addiction in adolescence (Tang et al., 2014; Zhou et al., 2017). Thus, impulsivity and BIS/BAS may be associated with coping styles, which in turn could be associated with Internet addiction. However, to the best of our knowledge, no empirical study has directly explored whether coping styles mediate the relationships among impulsivity, BIS/BAS and adolescent Internet addiction.

### The Moderating Role of Gender

Gender is a factor that potentially moderates the links among impulsivity, BIS/BAS, coping styles and Internet addiction. First, previous studies have shown that there are significant gender differences in some of the above variables. For example, men

have a relatively higher tendency toward Internet addiction than women in a meta-analysis involving 34 global jurisdictions, suggesting that gender-related differences in Internet availability and social norms could account for the gender differences in Internet addiction (Su et al., 2019). Specifically, greater gender gap in Internet penetration are associated with larger effect sizes of gender differences in Internet addiction (B = 0.223, 95% CI: 0.086–0.360). Furthermore, the more the social norms preferentially approve men to involve in potentially addictive behaviors such as smoking and alcohol consumption, the more men exhibit higher tendencies of Internet addiction than women (Su et al., 2019). The prevalence of Internet addiction has been found to be higher in boys than in girls (Ha and Hwang, 2014). Boys have higher impulsivity than girls (Munno et al., 2016). Two previous studies found that women have higher BIS activation and higher activation of specific aspects of the BAS (reward responsiveness) than men (Jorm et al., 1998; Nam et al., 2018). Women were more likely to use emotion-focused coping and less likely to use problem-focused coping than men (Matud, 2004). More importantly, it has been suggested that there are gender differences in the pathway associations between these variables. Compared with girls, Internet addiction has a strong association with impulsivity in boys (Nam et al., 2018). A meta-analysis of 46,025 adolescents found that gender is an important factor that could moderate the relationships between coping styles and Internet addiction in Chinese adolescents (Lei et al., 2018). Furthermore, girls with emotional difficulties were more easily affected by Internet addiction than boys with similar problems (Ha and Hwang, 2014). In addition, according to the social gender role theory, men are socialized as independent and self-reliant while women are socialized as warm, supportive, compassionate, sensitive to the feelings of others, and emotionally expressive (Reevy and Maslach, 2001), which further provides potential evidence for gender as a moderator among impulsivity, BIS/BAS, coping styles and Internet addiction.

### The Present Study

The purpose of the present study was to examine how impulsivity and BIS/BAS influence Internet addiction in adolescents. Specifically, this study explored the mediating effects of coping styles on impulsivity, BIS/BAS and Internet addiction as well as the gender differences among these associations. To our knowledge, this is the first comprehensive empirical study incorporating impulsivity, BIS/BAS, coping styles and gender factors and their roles in Internet addiction. On the basis of the dual-system neurobiological model and gender social roles theory (Reevy and Maslach, 2001; Casey et al., 2008), the proposed model is presented in **Figure 1**. It is plausible to hypothesize that coping styles act as mediator among impulsivity, BIS/BAS and Internet addiction, and gender serves as a moderator among these associations in adolescents. More specifically, we want to examine whether the dual-system neurobiological model includes high impulsivity and high BIS in Internet addiction among girls, while the dual-system neurobiological model combines high impulsivity and high BAS in boys. Furthermore, we suggest that emotion-focused coping will play an important role in girls, while problem-focused coping will be central in boys. Therefore, our hypotheses are as follows: (1) boys will report greater Internet addiction than girls; (2) impulsivity and the BIS will positively predict Internet addiction in girls; (3) impulsivity and the BAS will positively predict Internet addiction in boys; (4) emotionfocused coping will serve as a mediator between impulsivity and the BIS with Internet addiction in girls; and (5) problem-focused coping will serve as a mediator between impulsivity and the BAS with Internet addiction in boys.

### MATERIALS AND METHODS

### Participants and Procedure

According to the rule of thumb approach in the structural equation modeling, 10:1 is the commonly suggested ratio of sample size to free parameters, which is often used for minimum recommendations to determine a sample size for a structural equation modeling test (Bentler and Chou, 1987; Kline, 2016). Because 42 was the largest number of free parameters in our all structural equation models, the sample size was decided to be more than 420 in the present study. A total of 450 Chinese adolescents were recruited from four public schools in Beijing City, China. Among these adolescents, 416 participants (Mage = 14.56 years, SD = 1.42 years, age range: 11–18 years) completed the questionnaires, for a response rate of 92.44%. Thirty-four cases were excluded prior to analysis due to unreturned forms. The sample included 212 girls and 204 boys; 139 (33.41%) of the students were in the 7th grade, 98 (23.56%) in the 8th grade, 102 (24.52%) in the 10th grade, and 77 (18.51%) in the 11th grade. More detailed demographic information was shown in **Table 1**. All participants were fluent in Mandarin.

### Ethics Statement

The present study was approved by the Ethics Committee of the Institute of Psychology of the Chinese Academy of Sciences. School approval and parental consent were obtained prior to originating the study. All participants were informed that they could quit the study at any time without being penalized.

### Measures

### Internet Usage

For each participant, Internet usage information was obtained using 4 questions: (1) "How many years have you used the Internet?" (2) "How many hours do you use the Internet every day?" (3) "What is your duration of online gaming every day?" and (4) "Do you often play online gaming (Yes or No)?"

### Young's Diagnostic Questionnaire for Internet Addiction (YDQ)

The YDQ was applied to assess Internet addiction. The YDQ was modified according to the DSM-IV criteria for pathological gambling and consists of 8 "yes" or "no" questions (e.g., "Do you feel the need to use the Internet with increasing amounts of time in order to achieve satisfaction?" and "Do you feel restless, moody, depressed, or irritable when attempting to cut down or stop Internet use?") (Young, 1998). Total scores were calculated according to Young's method, with possible scores for all 8 items

ranging from 0 to 8. Higher scores reflect a higher level of Internet addiction. Because the YDQ is one of the most widely used questionnaires to evaluate Internet addiction, it has also good reliability and validity in Chinese adolescents (Cao et al., 2007; Li et al., 2014). In the present study, confirmatory factor analysis of the unidimensional model indicated that the model for Internet addiction showed a good fit with the data: χ 2 /df = 1.966, p < 0.05; CFI = 0.949, TLI = 0.928, RMSEA = 0.048 and SRMR = 0.041. In the current study, Cronbach's alpha coefficient for Internet addiction was 0.66.

### Barratt Impulsiveness Scale (BIS-11)

The BIS-11 is widely used to assess participants' impulsive traits by rating their frequency of 30 items on a scale from 1 (never) to 4 (always) (Patton et al., 1995). The BIS-11 includes three impulsiveness subscales: cognitive key (e.g., "I get easily bored when solving thought problems"), motor key (e.g., "I say things without thinking"), and non-planning key (e.g., "I am more interested in the present than in the future") (Choi J. S. et al., 2014; Martinez-Loredo et al., 2015). The overall impulsiveness score is determined by summing all items, with higher scores denoting greater impulsivity. In the present study, the Chinese version of the BIS-11 was used (Lu et al., 2012), and Cronbach's alpha coefficient was 0.91.

### Behavioral Inhibition System/Behavioral Approach System (BIS/BAS) Scales

A validated Chinese version of the BIS/BAS scales was used to assess the BIS and BAS (Li et al., 2015). The BIS/BAS scales are comprised of 20 items in addition to 4 filler items and include the Behavioral Approach System Scale (BAS, 13 items) and the Behavioral Inhibition System Scale (BIS, 7 items) (Carver and White, 1994). The former scale can be divided into three subscales: drive (BAS-drive, 4 items), reward responsiveness (BAS-reward, 5 items), and fun seeking (BAS-fun, 4 items). All items were assessed on a 4-point Likert scale from 1 (totally disagree) to 4 (totally agree). Sample items are "When I see an opportunity for something I like I get excited right away (BAS)" and "Criticism or scolding hurts me quite a bit (BIS)." In the present study, the Cronbach's alpha coefficients of the BAS-drive, BAS-reward, and BAS-fun were 0.67, 0.67, and 0.62, respectively. Scores for all 13 BAS items were summed to yield a single BAS score. Only the total BAS score was used in the current study. The Cronbach's alpha coefficients for the BAS and BIS in the current sample were 0.80 and 0.58, respectively.

### Coping Style Scale for Middle School Students

Adolescent coping styles were assessed with the Coping Style Scale for Middle School Students, which has been adapted for the Chinese culture (Dumont and Provost, 1999; Zhou et al., 2017; Sun et al., 2019). This inventory was designed on the basis of Folkman's interaction theory, the self-regulation theory and a prior coping styles questionnaire (Folkman et al., 1986). It is divided into two categories based on coping style, problem-focused coping and emotion-focused coping

### TABLE 1 | Descriptive statistics among the variables (N = 416).


<sup>∗</sup>p < 0.05; ∗∗p < 0.01.

(Folkman et al., 1986), and includes 36 items rated on a 4-point Likert scale from 1 (never coping) to 4 (often coping). Problemfocused coping consists of three subscales, including problem solving (7 items, e.g., "I make a plan to solve problems and execute it step by step"), support seeking (7 items, e.g., "I strive to get advice from someone about what to do"), and reasonable explanation (5 items, e.g., "I try to change my perspective to explore the positive side of frustration"). Emotion-focused coping consists of four subscales, including tolerance (4 items, e.g., "My ability is limited, so the only things I can do about unpleasant things is tolerate them"), avoidance (4 items, e.g., "I admit that I can't deal with a problem at hand, so I will give up trying"), venting emotions (4 items, e.g., "I express emotions to reduce my unhappiness"), and fantasy/denial (5 items, e.g., "I say to myself 'this isn't real' when encountering difficulties"). The scale has high construct validity, discrimination validity, and reliability in Chinese adolescents (Chen et al., 2000). In the current study, Cronbach's alpha coefficients for problem-focused coping and emotion-focused coping were 0.88 and 0.79, respectively.

### Procedure

The participants were given a packet of questionnaires that included instructions on how to respond to the questions and assurances of anonymity as well as questions regarding their basic demographic information, including gender, age, education grade, BIS/BAS, impulsivity, and coping styles. All scales were administered to participants in their classes. Students were tested individually in their classrooms. All the questionnaires were printed in the Chinese language and took approximately 30 min to finish. No personal identifying information was collected, and all the information collected was confidential.

### Data Analysis

Because the proportion of missing data was very low (<1%), mean substitution was adopted to deal with missing data. First, SPSS 20.0 was used to compute descriptive statistics and perform correlation analyses, Chi-square test and t-tests. Next, Amos 21.0 was used to test the hypothesized models. Structural equation modeling (SEM) was conducted to test the mediating role of coping styles in the relationships among impulsivity, BIS/BAS and Internet addiction. Furthermore, to assess gender differences, a multi-group (by adolescent gender) SEM was used.

In the present study, several goodness-of-fit indices were used to test the model-data fit. The first one was the Chi-square statistic and its associated p-value. If the p-value is not significant, it may indicate good model-data fit. However, the Chi-square statistic is sensitive to sample size (Bollen, 1989). Therefore we used the Chi-square to degrees of freedom ratio (χ 2 /df) to test model fit. A χ 2 /df ratio of less than 3 shows an admissible model fit. Other substitutive indices were also employed in the current study, including the comparative fit index (CFI) (Rigdon, 1996), the Tucker-Lewis Index (TLI) (Tucker and Lewis, 1973), the root mean square error of approximation (RMSEA) (Browne and Cudeck, 1993) and the standardized root mean square residual (SRMR) (Hooper et al., 2008). A CFI and TLI larger than 0.95 and a RMSEA and SRMR less than 0.08 show good model fit (Hooper et al., 2008). For the comparison of the nested models, differences in the χ 2 (Mχ 2 ) and the degree of freedom (Mdf) were used to compare the models with the goodness of fit to determine the model that best fit the data (Satorra, 2000; Byrne and Stewart, 2006). Specifically, the standard of comparison between the two nested models is as follows: when the degrees of freedom increase without a significant increase in the corresponding Chi-square value (that is, Mχ 2 /Mdf is not significant), the better model is the one with a larger degrees of freedom. Otherwise, the smaller degrees of freedom model is better. The predictive and explanatory powers of the model were measured using path coefficients and R 2 .

### RESULTS

### Descriptive Statistics and t-Tests/χ 2 -Test

The ranges, means and standard deviations of the continuous variables are shown in **Table 1** for girls and boys separately. There were no gender differences in age, education level, network age, Internet usage time, Internet addiction, impulsivity or coping, although boys showed a longer online gaming time. In the present study, 8.5% (n = 18) of girls frequently played online games, while 34.3% (n = 70) of boys frequently played online games. These results indicate that boys are more frequent online gamers than girls (χ <sup>2</sup> = 41.27, p < 0.001). Furthermore, BIS/BAS yielded significant gender differences. Compared with girls, boys had lower scores on the BIS and the BAS scales.

### Correlation Analyses

**Table 2** presents the correlations among the variables in the current study, with girls above the diagonal and boys below the diagonal. For both girls and boys, Internet addiction was positively related to impulsivity, BIS and emotion-focused coping, while Internet addiction was only negatively related to problem-focused coping in boys but had no correlations in girls. For both girls and boys, impulsivity was significantly related to problem-focused coping and emotion-focused coping. There were significant correlations between BAS and both problemfocused coping and emotion-focused coping in boys, but these correlations were not present in girls. For both girls and boys, BIS was positively related to emotion-focused coping but was not significantly related to problem-focused coping. Furthermore, there were significant gender differences in the correlations based on a one-tailed z-test, with a stronger association between



BAS, Behavioral Approach System; BIS, Behavioral Inhibition System; <sup>∗</sup>p < 0.05; ∗∗p < 0.01.

Internet addiction and BIS (z difference = 1.73) for girls than for boys (p < 0.05), a stronger association between Internet addiction and problem-focused coping (z difference = 2.80) for boys than for girls (p < 0.01), a stronger association between impulsivity and problem-focused coping (z difference = 3.08) for boys than for girls (p < 0.01), a stronger association between BAS and problem-focused coping (z difference = 1.91) for boys than for girls (p < 0.05), and a stronger association between BIS and emotion-focused coping (z difference = 1.73) for girls than for boys (p < 0.05). The following analyses of the hypothesized models were executed based on the correlation models of these variables.

### Structural Equation Model Analyses

Before analyzing the structural equation model, five observed variables (impulsivity, BAS, BIS, problem-focused coping, and emotion-focused coping, representative of their total scores respectively) and one latent variable (Internet addiction) were used to make our model more simplified and efficient. Furthermore, the YDQ was divided into two parcels, where the sum of items 1, 3, 5, and 7 constituted the first parcel (parcel 1), and the sum of items 2, 4, 6, and 8 constituted the second parcel (parcel 2), to act as indicators of Internet addiction employing an item-to-construct balance approach (Little et al., 2002). Then, structural equation modeling with AMOS 21.0 was carried out to examine our hypothesized mediation model. The factor loadings of Internet addiction for parcel 1 and parcel 2 were 0.799 and 0.689, respectively. The results of the model showed a good fit with the data: χ 2 /df = 2.417, p < 0.05; CFI = 0.984, TLI = 0.935, RMSEA = 0.058 and SRMR = 0.024. To test whether gender moderated the path relationships among these variables, two nested models were estimated. Specifically, we examined whether the estimate of the model parameters (i.e., path coefficients) varied between girls and boys. The first model permitted the structure coefficient of the two models to be estimated freely according to gender, while the second model was administered for the structure path coefficient to be equal. The results showed that these two models were significantly different, Mχ 2 (11, N = 416) = 25.424, p = 0.008, indicating that they differed according to gender. The structural model displaying unstandardized regression coefficients between variables is presented in **Figure 2**. In addition, we utilized critical ratios of differences (CRDs) as an index to examine the differences in structural path coefficients between genders. If the CRD was larger than 1.96, then the associations between these two variables would demonstrate a significant gender difference at p < 0.05. The results showed that the structure path from impulsivity to problem-focused coping revealed a significant gender difference (CRD = 2.48, p < 0.05). More specifically, the path coefficient for girls was β = −0.30, p < 0.001, while the path coefficient for boys was β = −0.48, p < 0.001. Thus, compared with girls, impulsivity had a far greater negative influence on problem-focused coping among boys. The structure path from problem-focused coping to Internet addiction also revealed a significant gender difference (CRD = 2.51, p < 0.05). More specifically, the path coefficient for girls was β = 0.06, p > 0.05, while the path coefficient for boys was β = −0.25, p < 0.01. Therefore, problem-focused coping had a far greater negative prediction to Internet addiction among boys than among girls. Furthermore, the structure path for the path from BAS to emotion-focused coping revealed a significant gender difference (CRD = −2.73, p < 0.05). More specifically, the path coefficient for girls was β = −0.02, p > 0.05, while the path coefficient for boys was β = 0.23, p < 0.001. This result suggests that BAS had a far greater positive impact on emotion-focused coping among boys than among girls. The unstandardized regression coefficients from the multiple-group structural model analysis and the CRDs between girls and boys are presented in **Table 3**. In total, the model explained 32.8% of the variance in Internet addiction among girls and 30.7% of the variance in Internet addiction among boys.

When the final model was chosen, bias-corrected bootstrapping, a non-parametric resampling procedure, was utilized to further test the significance of the mediators. Bootstrapping has considerably greater statistical power to test indirect effects than traditional mediation analyses (MacKinnon et al., 2004). When the 95% confidence intervals (CIs) do not include zero, the indirect effect is statistically significant. In the present study, 5000 bootstrapping samples were generated to derive CIs. The results of the bootstrap analyses indicated that the specific indirect effect of impulsivity on Internet addiction through emotion-focused coping was significant, and the total indirect effects of BIS on Internet addiction and the specific indirect effect of BIS on Internet addiction through emotionfocused coping were also significant in girls. The total indirect effects of impulsivity on Internet addiction and the specific indirect effects of impulsivity on Internet addiction through problem-focused coping or emotion-focused coping were significant, and the specific indirect effect of BAS on Internet addiction through problem-focused coping was also significant in boys (see **Table 4**).

### DISCUSSION

### Gender Differences in Internet Addiction

Contrary to our hypothesis, we did not find gender differences in Internet addiction, which was in line with the results of McNicol and Thorsteinsson (2017). According to the Internet Availability hypothesis (Mann, 2005), availability is an important determinant of addictive behavior. Men have higher levels of Internet addiction than women, which may be associated in part with gender-related differences in Internet availability (Su et al., 2019). A recent meta-analysis involving 34 global jurisdictions found that the prevalence of Internet addiction in men was only slightly higher than that in women (g = 0.145) from a global perspective, and that these gender differences in Internet addiction may be partly caused by the gender-related gaps between in economy and in Internet penetration (Su et al., 2019). The fact that our sample came from the capital of China (Beijing), with a high level of economic development and Internet penetration, may be one of the reasons why there were no gender differences in Internet addiction in the present study. Furthermore, there are no gender differences

and ∗∗∗p < 0.001.



The numbers in italics in the parentheses represent the standard errors; S.E., standard error; CRD, critical ratio difference.

in network age and total Internet usage time in the present study, which could provide supportive evidence for the Internet Availability hypothesis. In addition, the YDQ was used to assess participants' generalized Internet addiction rather than specific Internet addiction in the current study, which could be another important reason why there were no gender differences in Internet addiction in our sample, because gender may show different effect sizes (magnitude and/or directionality) for specific subtypes of Internet addiction (Su et al., 2019). For example, for online gaming addiction, the rates of men vs. women were 31% vs. 13.1%, while for social networking addiction, the rates of men vs. women were 27.8 and 37.3%, respectively (Tang et al., 2017). Although the gender effect size for generalized Internet addiction was small at 0.15, it was 0.67 for online gaming and 0.10 for social networking sites in a Chinese sample. Meanwhile, there was no significant gender effect size for generalized Internet addiction (g = −0.03), but medium gender effect sizes for online gaming (g = 0.58) and social networking sites (g = −0.42) in a sample from the United States (Tang et al., 2017). In our study, boys spent more time online gaming, and the number of boys who frequently played online games was greater than the number of girls, which could provide some supportive evidence for gender-related differences in subtypes of Internet addiction.

### Direct Relations Between Impulsivity and BIS/BAS With Internet Addiction Across Genders

In the current study, we found that impulsivity and BAS could directly positively predict Internet addiction in boys, whereas impulsivity and BIS could directly positively predict Internet addiction in girls. The results in boys could provide supportive evidence for the dual-system neurobiological model that heightened reward-seeking and deficient impulse control may be a risk factor for adolescent addictive behaviors (Casey et al., 2008; Casey and Jones, 2010). Compared with children and



<sup>a</sup>The indirect effect represents the effect through all possible mediators (i.e., problem-focused coping, emotion-focused coping). CI, confidence interval; PFC, problem-focused coping; EFC, emotion-focused coping; IA, Internet addiction. <sup>∗</sup>p < 0.05, ∗∗p < 0.01.

adults, adolescents are characterized by an imbalance between early emerging "bottom–up" systems that show exaggerated reaction to motivational stimuli and later maturing "top–down" cognitive control systems (Casey and Jones, 2010). Both in boys and in girls, impulsivity is positively correlated to poor to–down cognitive control (Casey and Jones, 2010). Meanwhile, with regard to bottom-up motivational stimuli, boys could be more sensitive to reward stimuli (Steinberg et al., 2009), while girls could be more sensitive to punishment stimuli rather than reward stimuli (Pagliaccio et al., 2016). With the rapid development of the Internet technology, 85.3% of adolescents in China have access to the Internet (China Internet Network Information Center [CNNIC], 2016). As a result, adolescents could seek abundantly available rewarding stimuli conveniently on the Internet. Especially for boys with low cognitive control capacity (high impulsivity), these rewarding stimuli (e.g., online gaming positive incentive) continuously reinforce their Internet behaviors, which gradually increase their risk for Internet addiction. Compared with real world, cyberspace could provide a more convenient, anonymous, and safe social interactions environment. Girls with high BIS, who are more sensitive to punishment stimuli (e.g., criticism or scolding from other people), are prone to overuse of the Internet to "escape loneliness" and "belong to a group" instead of face-to-face interactions or offline activities (Park et al., 2013). Furthermore, low cognitive control capacity (high impulsivity) could deteriorate the negative impact of BIS on Internet use in girls. Thus, girls with high impulsivity and high BIS are more prone to Internet addiction than other girls. Taken together, the present study considers the role of gender in the associations among impulsivity, BIS/BAS and Internet addiction according to the dual-system neurobiological model, which could provide further supportive evidence for the key role of impulsivity on adolescent Internet addiction (Crews and Boettiger, 2009; Brand et al., 2014), and could explain the previous inconsistencies in the associations between BIS/BAS and adolescent Internet addiction (Yen et al., 2012; Park et al., 2013).

### Mediating Relations Among Impulsivity, BIS/BAS to Internet Addiction Across Genders

The present study developed a multi-group mediation model to illuminate the different mechanisms underlying the associations of impulsivity and BIS/BAS with Internet addiction between girls and boys. The most important and interesting results showed that different coping styles were an important mechanism through which impulsivity and BIS/BAS were associated with Internet addiction across genders. Specifically, impulsivity and BIS increased the risk of adolescent Internet addiction through enhanced emotion-focused coping in girls. However, impulsivity raised the risk of adolescent Internet addiction through increased emotion-focused coping and decreased problem-focused coping in boys. In addition, problem-focused coping mediated the associations of BAS with Internet addiction in boys. Adolescents have to cope with massive stressors caused by biological, cognitive and social changes that occur across development from childhood to adulthood. The social gender role theory indicates that appraisals of life events might differ across genders (Tamres et al., 2002; Sarrasin et al., 2014). Specifically, compared with men, women are more likely to appraise events as stressful and view stressors as threats rather than challenges. Furthermore, the social gender role theory also indicates that men are socialized as independent, self-reliant and to suppress emotions, while women are socialized as warm, supportive, compassionate, sensitive to the feelings of others, and emotionally expressive with less restrictions (Reevy and Maslach, 2001). Therefore, women are more likely to be accepted by others if they express negative emotions in social interactions than men. Consequently, when the Internet is used as a stress coping resource, girls with high impulsivity and high BIS are more likely to adopt emotion-focused coping (e.g., venting emotions, fantasy, and avoidance) to palliate eventrelated distress on the Internet, which in turn increases their risk of Internet addiction. These findings are similar to the results of a recent study showing that the combination of impulsivity and neuroticism increases the risk of emotionfocused coping, and thus exacerbates adolescent binge eating behaviors (Keough et al., 2016).

As for boys, when the Internet is used as a stress coping resource, boys with high impulsivity are more likely to use emotion-focused coping (e.g., cyberbullying and avoidance by playing online games) but less likely to engage in problemfocused coping, which is in line with previous studies

(Connor-Smith and Flachsbart, 2007; Lee-Winn et al., 2016). Previous studies have also indicated that a person with emotionfocused coping views Internet use in a more dependent manner (Lam et al., 2009; Milani et al., 2009). The virtual world created by the Internet provides them with an opportunity to escape from external stress and difficulties in real life through shortterm pleasure and relief. However, problem-focused coping could decrease the risk of Internet addiction (Al-Gamal et al., 2016). Therefore, more emotion-focused coping and less problemfocused coping increases the risk of Internet addiction.

On the other hand, boys with high BAS have high novelty seeking and fun seeking tendencies (Casey and Jones, 2010), which may cause them to use problem-focused coping frequently in stressful environments. When dealing with the external stress and difficulties that occur in real life, the Internet can be seen as a helpful resource (e.g., acquiring information, seeking others help) instead of as an escape for boys with high BAS. This problem-focused coping could decrease their risk for Internet addiction. Thus, high BAS maybe plays an indirect protective role against Internet addiction in boys, which is similar to the results of Hasking' study showing that reward responsiveness is positively associated with the use of problem-focused coping, which in turn is negatively associated with delinquent behavior in adolescents (Hasking, 2007). Taken together, combined with the direct and indirect relations between BAS and Internet addiction in the present study, these results suggest that BAS is a double-edged sword for Internet addiction in boys, which calls for the need for more studies to understand the beneficial and harmful factors related to BAS in boys.

### Implications for Theory and Practice

From a theoretical perspective, extending previous research, the present study provides empirical support for the dual-system neurobiological model among boys in the context of Internet addiction. Furthermore, our study indicates that the combination of deficient impulse control and heightened BIS rather than BAS increases the risk for Internet addiction in girls. To our knowledge, this is the first study to examine the mediating roles of problem-focused and emotion-focused coping from impulsivity and BIS/BAS to Internet addiction and the moderating role of gender among these associations, which is helpful to improve our understanding of Internet addiction among adolescents. It is especially important in cases where Internet addiction is not viewed as a disorder regardless of category by DSM-V because of insufficient evidence (Grant and Chamberlain, 2016). From a practical perspective, our findings may be helpful for providing evidence-based preventions and interventions to decrease Internet addiction among adolescents. In general, further attention should be given to developing Internet addiction prevention and intervention programs that are tailored to the different needs of girls and boys. First, when screening and choosing a target population for further prevention and intervention programs among adolescents, different combinations of risk factors (e.g., impulsivity and BAS in boys) should be adopted according to gender differences. Second, and even more importantly, our results could offer invaluable knowledge on how to prevent and intervene in Internet addiction among adolescents. Specifically, interventions to decrease impulsivity and BIS could have the potential to decrease Internet addiction in girls, while interventions to reduce impulsivity and BAS may have the potential to decrease Internet addiction in boys. In addition, the finding that coping style mediates the associations from impulsivity and BIS/BAS to Internet addiction across genders provides important implications for practice. To prevent and intervene in Internet addiction in adolescents, training techniques should be exploited to enhance adolescents' coping style skills because improving specific behaviors may be more efficient than directly altering individual dispositions (Taylor and Stanton, 2007). On the one hand, parents and practitioners should provide a supportive environment, which could play a positive role in adolescents' coping skills and consequently decrease the risk of Internet addiction in adolescents. On the other hand, coping effectiveness training has previously been confirmed to be an efficient method for improving coping skills (Folkman et al., 1991). Parents and practitioners could adopt methods to help girls decrease their use of emotion-focused coping when managing developmental tasks. In the context of stressors, it would be beneficial to improve the use of problem-focused coping and decrease the use of emotion-focused coping in boys, such as providing boys with the knowledge and practical skills to resolve their problems, seek social support and balance their emotions effectively.

### Limitations and Further Directions

Although this study revealed the gender-specific pathways from impulsivity and BIS/BAS to adolescents' Internet addiction, several limitations of the current study merit attention. First, the intention of our study is not to "psychopathologize" adolescence, but rather to explore why some adolescents are more vulnerable to Internet addiction than others. However, because Internet addiction is significantly associated with symptoms of depression, social anxiety, attention-deficit/hyperactivity disorder and other mental disorders, which were not investigated in our sample, future studies including these factors could provide stronger evidence for our results with a high capacity to control for these confounding variables. Second, although YDQ and BIS/BAS scales have good reliability and validity in Chinese adolescents (Cao et al., 2007; Li et al., 2014, 2015), the Cronbach's alpha coefficients for the YDQ and BIS scale were low in the present study, which may be caused by their limited number of items (8 and 7, respectively). Future research should adopt scales with more items to assess Internet addiction and BIS (e.g., the Young's Internet Addiction Test with 20 items; the Sensitivity to Punishment and Sensitivity to Reward Questionnaire with 48 items) (Torrubia et al., 2001; Widyanto and McMurran, 2004). Third, the present study employed a cross-sectional design. Although this study design can demonstrate strong associations among variables, it cannot provide strong evidence of causal relationships among these variables. Therefore, experimental methods and longitudinal designs could be beneficial in future research to provide more reliable conclusions about the directionality of these effects. Finally, more and more research

points toward the type of Internet application (e.g., gaming and social networking site) – not the Internet itself – as being responsible for the development of a problematic usage (van Rooij et al., 2010; Lee et al., 2015). However, the type of usage was not considered, which could pose a limitation in the present study. Furthermore, evidence from previous studies has shown that it is very important to distinguish between generalized Internet addiction and specific Internet addiction and that large gender differences exist in online gaming and social networking site use (Montag et al., 2015; Tang et al., 2017; Lopez-Fernandez, 2018). Further studies are needed to include more Internet usage information, which could provide stronger evidence of whether the results of the current study are appropriate for generalized Internet addiction or specific Internet addiction.

### DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

### ETHICS STATEMENT

The studies involving human participants were reviewed and approved by the Ethics Committee of the Institute of Psychology of the Chinese Academy of Sciences. Written informed consent

### REFERENCES


to participate in this study was provided by the participants' legal guardian/next of kin.

### AUTHOR CONTRIBUTIONS

QL and XL designed the study and wrote the protocol. QL and YZ collected the research data. BD conducted the statistical analyses and wrote the manuscript. WD and LW conducted the literature searches and created the figures. All authors approved the final version of the manuscript.

### FUNDING

This study was supported by the National Natural Science Foundation of China (Grants 31571161 and 31200782), the Humanities and Social Science Research of the Ministry of Education of China (18YJC190002), the CAS Key Laboratory of Behavioral Science, Institute of Psychology (Y5CX052003), and the China Scholarship Council.

### ACKNOWLEDGMENTS

We wish to extend our appreciations to the referees for their comments.


adolescence: Problems, progress, and potential in theory and research. Psychol. Bull. 127, 87–127. doi: 10.1037/0033-2909.127.1.87


prospective study. Arch. Pediatr. Adolesc. Med. 163, 937–943. doi: 10.1001/ archpediatrics.2009.159



Satorra, (Boston, MA: Springer US), 233–247. doi: 10.1007/978-1-4615-4603- 0\_17



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Li, Dai, Zhong, Wang, Dai and Liu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Your Performance Is My Concern: A Perspective-Taking Competition Task Affects ERPs to Opponent's Outcomes

Hao Yu<sup>1</sup> , Weizhi Nan<sup>2</sup> , Guochun Yang1,3, Qi Li1,3, Haiyan Wu1,3 \* and Xun Liu1,3 \*

<sup>1</sup> Key Laboratory of Behavioral Science, Institute of Psychology, Chinese Academy of Sciences, Beijing, China, <sup>2</sup> Department of Psychology, Center for Brain and Cognitive Sciences, School of Education, Guangzhou University, Guangzhou, China, <sup>3</sup> Department of Psychology, University of Chinese Academy of Sciences, Beijing, China

Previous research has shown that people have more empathic responses to in-group members and more schadenfreude to out-group members. As a dimension of cognitive empathy, perspective-taking has been considered to be related to the enhancement of empathy. We tried to combine these effects through manipulation of a competitive task with opponents and an in-group partner and investigated the potential effect of in-group bias or the perspective-taking effect on outcome evaluation. We hypothesized that the neural activities would provide evidence of in-group bias. We tested it with a simple gambling observation task and recorded subjects' electroencephalographic (EEG) signals. Our results showed that the opponent's loss evoked larger feedbackrelated negativity (FRN) and smaller P300 activity than the partner's loss condition, and there was a win vs. loss differential effect in P300 for the opponent only. The principal component analysis (PCA) replicated the loss vs. win P300 effect to opponent's performance. Moreover, the correlation between the inclusion of the other in the self (IOS) scores and FRN suggests perspective-taking may induce greater monitoring to opponent's performance, which increases the win vs. loss differentiation brain response to the out-group agent. Our results thus provide evidence for the enhanced attention toward out-group individuals after competition manipulation, as well as the motivation significance account of FRN.

Keywords: in-group bias, gambling task, feedback related negativity, P300, EEG

### INTRODUCTION

As an important aspect of self-representation in social life, the in-group bias refers to the behavioral pattern of people more favoring in-group members than out-group members, which also has been widely explored from both developmental and evolutionary views (Brewer, 1979; Struch and Schwartz, 1989; Van Bavel et al., 2008; Yang et al., 2014; Oestereich et al., 2019). Using a minimal group paradigm, Tajfel (1970) and Tajfel et al. (1971) found that even random group classification could elicit in-group bias and out-group discrimination in subjects. A large number of social psychology studies reported the in-group bias or intergroup discrimination respectively from different aspects, such as, developmental and evolutionary views (Brewer, 1979;

Edited by:

Daniela M. Pfabigan, University of Oslo, Norway

### Reviewed by:

Katharina Paul, Ghent University, Belgium Xiaoqin Mai, Renmin University of China, China

> \*Correspondence: Haiyan Wu wuhy@psych.ac.cn Xun Liu liux@psych.ac.cn

### Specialty section:

This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Neuroscience

Received: 30 May 2019 Accepted: 14 October 2019 Published: 25 October 2019

### Citation:

Yu H, Nan W, Yang G, Li Q, Wu H and Liu X (2019) Your Performance Is My Concern: A Perspective-Taking Competition Task Affects ERPs to Opponent's Outcomes. Front. Neurosci. 13:1162. doi: 10.3389/fnins.2019.01162

Struch and Schwartz, 1989; Van Bavel et al., 2008; Yang et al., 2014; Oestereich et al., 2019). Such in-group bias or outgroup discrimination modulates various people's social behaviors, including both helpful and harmful behaviors (Cikara et al., 2011, 2014; Dickinson et al., 2018).

Apart from social psychology studies, neuroscientific research has also provided evidence for the effect of group identity on people's emotions or action tendency. Studies have shown that racial group membership modulates brain activities in a pain empathy task (Xu et al., 2009; Contreras-Huerta et al., 2013; Fabi and Leuthold, 2018; Han, 2018). Montalan et al. (2012) adopted Minimal Group Paradigm (participants were randomly assigned to different groups: e.g., underestimator and overestimator group in the dot estimation task) to investigate the empathy preference, and found higher pain empathy to ingroup members and lower empathy to out-group members in imagined pain empathy condition. Such brain activity of empathy has been shown to be able to predict the altruism motivation (Mathur et al., 2010; Xin et al., 2018) or costly helping (Hein et al., 2010; Preis et al., 2013). Moreover, as an interesting intergroup emotion, schadenfreude (Dasborough and Harvey, 2017) has also been found in neuroscience (Steinbeis and Singer, 2014; Vollberg and Cikara, 2018). One representative work from Cikara et al. (2011) on soccer fans showed that the win of the favorite team (in-group) and the loss of the rival team (out-group) activate the ventral striatum, which is a reward-related brain region. Taken together, both social psychology and neuroimaging studies indicate that in-group bias has an impact on empathy and intergroup schadenfreude.

Previous studies from Sherif showed competition is a key element in group differentiation (Sherif et al., 1961; Sherif, 1966). Previous group manipulations usually involved competition tasks. For example, studies from Molenberghs (Molenberghs et al., 2012; Morrison et al., 2012) used a team competition task in which participants had to respond as quickly as possible after the "GO" signal. In that case, people do not need to face the opponent directly and interact with them. However, people usually need to anticipate their opponents' mind in the competition situation. Notably, the empathy level is highly correlated with perspective-taking, which is considered as a key cognitive component of empathy (Davis et al., 1996). Perspectivetaking is also a way to reduce intergroup conflicts and improve the intergroup relationship (Cohen and Insko, 2008; Shih et al., 2009; Todd et al., 2011; Böhm et al., 2018). For instance, a study showed that a perspective-taking viewing task improved subjects' liking and induced more empathic feelings toward another member of the out-group (Shih et al., 2009). Such an intergroup relationship improvement effect from perspectivetaking may be attributable to the formation of a "social bond" (Mcdonald et al., 2017). However, how a perspective-taking competition game affects the in-group or out-group's outcome processing has not been investigated. Therefore, the present study will investigate the outcome evaluation by group membership and perspective-taking competition game manipulation.

Works in the domain of outcome evaluation have identified two key related ERP components: the feedback-related negativity (FRN) and the P300 (Osinsky et al., 2016). FRN is a fronto-central negative deflection that is larger following the presentation of negative feedback (Miltner et al., 1997; Holroyd and Coles, 2002; Hajcak et al., 2005). However, other recent studies have suggested that the FRN can be conceptualized as a positive deflection that is more positive following a reward compared with nonreward outcomes, particularly by principal component analysis (PCA) ERP studies (Holroyd et al., 2008; Foti et al., 2011). FRN is thought to be associated with reward prediction errors in reinforcement learning theory (Holroyd and Coles, 2002; Yeung and Sanfey, 2004), which was challenged by a study showing that FRN reflects the salience errors (Talmi et al., 2013) or expectation (Cao et al., 2015). Further, the P300 component, which is traditionally considered as reflecting the attention process or context-updating (Zhao et al., 2017), has also been suggested to be associated with the motivational significance of reward (Wu and Zhou, 2009) or the valence of the outcome (Hajcak et al., 2005).

Interestingly, studies also showed a "mirror" performance monitoring system in which observing another's gain or loss also evokes similar FRN, which is called observational FRN (oFRN), as it applies in observation situations (Kang et al., 2010; Wang Y. et al., 2014). Fukushima and Hiraki (2009) found that significant oFRN was elicited only when humans were in observation, not computer players. Further, researchers have investigated the effect of an interpersonal relationship through the ERP correlating to outcome evaluation (Itagaki and Katayama, 2008; Leng and Zhou, 2010; Marco-Pallares et al., 2010). For example, Leng and Zhou explored the different neural responses to friends and strangers when the observer was engaged in the same gambling game and failed to find a differentiation of FRN responses between friends and strangers observations. In another study, they found FRN and P300 responses to win and loss feedbacks similarly increased (Zhou et al., 2010). In summary, FRN and P300 are considered as neural markers for empathy toward an outcome evaluation (Miltner et al., 2003; Fukushima and Hiraki, 2009). However, as far as we know, there has been no study combining group membership and perspective-taking manipulation to examine these effects on outcome empathy.

In the present study, we manipulated a temporary group identity in a competition context and utilized an interactive football game to increase the perspective-taking toward the out-group members. Then we examined the possible differential effect on the partner's and opponent's win or loss in a benefit-independent context. It sounds that the in-group member's outcome evoked a larger "empathy" effect due to in-group favoritism. However, as Galinsky et al. (2008) wrote, "understanding one's opponent is valuable for success in competitive interactions" and "get inside the head of your opponent" is crucial for social interaction. Following this view, we also expect enhanced attention regarding an out-group member's outcome. Therefore, we asked an open question regarding which effect (in-group favoritism or perspective-taking) is more prominent in a simple "gambling observing" task. We aimed to examine two possible effects: the in-group empathy bias effect (e.g., more concern about a partner's outcome or an opponent's loss per the schadenfreude effect) and the competition induced attention on opponent effect (e.g., more concern to an opponent's outcome for the interaction in a perspective-taking competition game).

### MATERIALS AND METHODS

fnins-13-01162 October 24, 2019 Time: 16:6 # 3

### Participants

Nineteen right-handed man college students with normal vision (age: 22.90 ± 0.93) from Beijing participated in this study. All participants were recruited through advertisement, with no history of neurological or psychiatric illness and no drug intake. To control the task familiarity, all participants reported have the experience of watching football matches. All procedures were approved by the institutional review boards (IRB) of the Institute of Psychology, Chinese Academy of Sciences. All participants signed the informed consent before the experiment and were paid after the experiment.

### Procedure

The interpersonal reactivity index (IRI) and the inclusion of others in the self (IOS) scale data were collected before the experimental procedure. Before the formal experimental procedure, participants were told that they would join this experiment with other three participants (actually experimenters) at the same time. To make the real participants believe this, we called them and emphasized the experiment time to make sure that everybody arrived on time in case of meeting the other experimenters in the building. Before the formal procedure, all participants were asked to take a look at the other experiment rooms, with or without participants there. Moreover, we asked them to wait for 1–2 min before the first stage if one player was late.

The formal procedure consisted of three stages, as is illustrated in **Figure 1**.

### Gambling Observation Stage

Participants were instructed that two players (the partner and one opponent) were randomly selected to participate in the gambling task, and the remaining two players needed to observe their performance. All four players practiced the gambling task and understood that the gain or loss was independent (i.e., win or loss for yourself). For observers, they are asked to answer one question to ensure their involvement in the task. The question is about who would win more money in the gambling task. The one with the correct answer will be awarded 10 Yuan. For each trial of the observation task, the face of the gambling player was presented for 900 ms, implying the one performing the gambling task in the current trial. Then the gambling task started with a fixation for 300 ms, then two cards were presented (two white rectangles with a 2.5◦ × 2.5◦ visual angle), and the player was asked to make a selection by pressing the "F" or "J" key. Feedback was presented for 1000 ms at the end of the trial (see **Figure 1C**). The probability of win/loss was equal across the partner and opponent to rule out potential confounding influences on the differential win–loss probability. There were 200 trials (50 trials for each condition: partner-win, partner-loss, opponent-win, and opponent-loss) in total, with a short break for every 40 trials in the task. The whole gambling observation task lasted around 30 min.

In the formal procedure, all instructions were presented through PowerPoint software. All aforementioned procedures were conducted by E-Prime software (Version 2.0, Psychology Software Tools, Inc.). After completing all formal procedures, participants were asked to rate the performance on a seven-point scale of all players in the tasks (foot task and gambling task). After receiving payment, the participants were also asked to report their involvement or seriousness in the observation task on an assumed 7 point scale.

### Post-experiment Rating

After the experiment, participants were instructed to recall some experimental details and provide ratings about their feelings of happiness when the other's win or loss outcome occurred. All participants made a correct recall of their performance in the experiment, and they were paid 10 Yuan by Alipay for their completion of this rating task.

### Electroencephalographic Recording and Preprocessing

During the task, the participant sat approximately 80 cm from a computer screen comfortably in an electrically shielded room. We recorded the electroencephalographic (EEG) data using a 64-channel Neuroscan system (Neuroscan Inc., Herndon, VA, United States) in the gambling observation sessions. Raw EEG data were sampled at 500 Hz/channel, referenced to the left mastoid on-line, with impedance lower than 5 k. Vertical electrooculograms (VEOG) were recorded supra- and infraorbitally at the left eye. Horizontal EOGs (HEOG) were recorded by electrodes at the left and right orbital rims. The online continuous data were digitized with a bandpass of 0.05–100 Hz.

Electroencephalographics were re-referenced to the average of the left and right mastoids and filtered with a low pass of 20 Hz (24 dB/oct) off-line (Hajcak et al., 2006). Epochs were feedbacklocked, from 100 ms before the feedback onset to 500 ms after the feedback onset. Ocular artifacts were removed from the EEGs using a regression procedure implemented with Neuroscan software (Scan 4.5). Trials exceeding the threshold of ±80 µV were excluded from further analysis. As a result, 13.4% of the epochs were rejected across participants. Trials of four conditions (partner-win, partner-loss, opponent-win, and opponent-loss) were averaged, respectively, and a −100 to 0 ms baseline was used to perform a baseline correction.

### Average ERP Analysis

Previous literature identified FRN by creating a difference wave between win and loss trials (Dunning and Hajcak, 2007; Leng and Zhou, 2010) or from the grand-averaged waveform (Luo et al., 2014). In our study, we are interested in the group effect on ERPs in both the win and the loss conditions. Therefore, we directly measured the FRN and P300 in the grand-averaged waveforms rather than the difference wave. The grand-averaged ERPs at FCz and CPz and the corresponding topography map are presented in **Figure 2**. Based on both the previous literature (Gu et al., 2011) and visual inspection of the topography map, the FRN was

shows a two-player gambling task (partner vs. participant) with the time-course of visual stimuli. First, there is a face stimulus (which has been masked to protect the privacy of the participant) indicating which player's turn it is, followed by two cards (600–1000 ms), and the player is asked to choose either of the cards. The chosen card is indicated (600–800 ms), and a win/loss outcome (1000 ms) is presented after a jitter (600–1200 ms). The example shows a "loss" outcome. In this stage, the players who are not selected to play (e.g., the participant) are asked to observe the other's performance.

detected at nine fronto-central electrodes (FP1, FPz, FP2, F1, Fz, F2, FC1, FCz, and FC2) and P300 was detected at nine centroparietal electrodes (C1, Cz, C2, P1, Pz, P2, CP1, CPz, and CP2). Because we are not interested in the electrode effect in the current study, we pooled the nine electrodes and computed the averaged FRN and P300 amplitudes. The FRN amplitude was measured for each participant as the mean amplitude within the 230–280 ms window, while the P300 was identified as the mean amplitude within the 300–450 ms window. Because we are not interested in the electrode effect in the current study, the averaged FRN and P300 amplitudes were entered into a 2 (feedback valence: win and loss) × 2 (agent: partner and opponent) repeated-measures analysis of variance (ANOVA).

### Temporospatial PCA

fnins-13-01162 October 24, 2019 Time: 16:6 # 5

It is possible that components overlapping in our grand-averaged waveforms, especially for FRN (see **Figure 2**) and the PCA, is a useful tool for the decomposition of ERPs (Foti et al., 2011). Therefore, we also applied temporo-spatial PCA to more clearly identify the FRN and P300 components. PCA Toolkit (EP Toolkit, version 2.23) and MATLAB 7.8 (MathWorks, Natick, MA, United States) were employed to conduct the PCA in this study (Dien, 2010). We first imported the averaged ERPs from the four conditions for each subject to the toolbox. After checking the data for all conditions, a two-step PCA procedure was performed as in the previous study (Zhang et al., 2013), that is, a temporal PCA was performed on all-time points from each participant's average ERPs, with promax rotation. After capturing the variances in the time domain, a spatial PCA was conducted for each of the resultant temporal factors using all of the recording electrodes with an infomax rotation. Finally, three temporal factors × three spatial factors were extracted from our ERP data based on the screen plot, yielding nine temporospatial factor combinations. For our specific interest in FRN and P300, we identified these two components and extracted the amplitudes, which were also put into a 2 (feedback valence: win and loss) × 2 (agent: partner and opponent) repeated-measures ANOVA. All ANOVAs in the current study were with Tukey post hoc testing at a significance level of 0.05. The significant p-value was set as 0.05, and the effect size was calculated using partial eta squared.

### RESULTS

### Behavioral and Psychological Data

The mean IRI score was 69 ± 8.30 (SD) and the IOS score was 4.47 ± 1.26. There was no significant difference between the performance rating for the partner (3.53 ± 1.02) and opponent (3.63 ± 0.76) in the football task, t (18) = −0.35, p = 0.73. The performance rating of the gambling task between the partner (3.73 ± 0.93) and opponent (3.42 ± 0.69) was also not significantly different, t (18) = 1.37, p = 0.19.

As an important validation index of the grouping manipulation, the self-reported involvement or seriousness score was 4.05 ± 0.91, showing a relatively high involvement in the observation task. Regarding the football task, 47.3% of the participants (9 participants) were defeated in this game with their teammate and 52.7% won (10 participants), as the random manipulation regulated. Because the winning was equal for the "partner" and "opponent" in the gambling task, 42.1% of the participants (8 participants) chose the opponent and 57.9% chose the partner in the "who wins more" question after the observation task, and the difference of choice probability was not significant. The 2 (outcome valence) × 2 (agent: partner vs. opponent) ANOVA on the happiness rating showed a significant outcome × agent interaction effect, F (1, 18) = 41.53, p = 0.001, ηp <sup>2</sup> = 0.698). The post hoc analysis showed an in-group bias and a schadenfreude effect that the happiness rating was significantly higher when the partner win (5.21 ± 0.15) than when he loses (2.68 ± 0.27), as well as when the opponent loses (4.74 ± 0.30) rather than opponent win (3 ± 0.24), ps = 0.01.

### Grand-Averaged ERP Results FRN Component

The repeated-measures 2 (outcome valence) × 2 (agent: partner vs. opponent) ANOVA showed a significant interaction outcome × agent effect, F (1, 18) = 12.73, p = 0.02, η<sup>p</sup> <sup>2</sup> = 0.415). Further analysis indicated that the FRN was more negative-going following the opponent's losses (1.85 ± 0.66 µV) than following the partner's losses (3.24 ± 0.65 µV), F (1, 18) = 8.01, p = 0.01. We did not find significant FRN results between opponent's wins (2.63 ± 0.42 µV) and partner's wins (2.63 ± 0.42 µV).

### P300 Component

The repeated-measures 2 (outcome valence) × 2 (agent: partner vs. opponent) ANOVA on the P3 amplitude also showed a significant outcome × agent interaction effect, F (1, 18) = 5.51, p = 0.031, η<sup>p</sup> <sup>2</sup> = 0.245). Further analysis indicated a smaller P300 for the opponent's losses (5.25 ± 0.81 µV) than for the partner's losses (6.24 ± 0.89 µV), F (1, 18) = 4.86, p = 0.03. The post hoc analysis also indicated a significant difference for the opponent's win (6.53 ± 1.01 µV) versus opponent's loss, F (1, 18) = 4.89, p = 0.041, while such a win vs. loss difference was not significant for the partner, p = 0.67.

### PCA ERP Results

Nine-factor combinations consisted of three temporal factors and three spatial factors (see **Supplementary Table 1**). We identified the FRN (peaked at Fz on 266 ms) and P300 (peaked at P1 on 378 ms) based on the visual inspection of the factor combinations and the previous PCA results (Foti et al., 2011; Zhang et al., 2013) (see **Figure 3**). These two PCA components were statistically analyzed as the mean amplitudes within different time windows (250–300 ms for PCA-FRN, 300– 450 ms for PCA-P300) at their peak channels (i.e., Fz and P1). Thereafter, the mean values of the amplitudes were separately subjected to repeated-measures 2 (outcome valence) × 2 (agent: partner vs. opponent) ANOVAs.

As **Figure 3** shows, we found a PCA-FRN component that was prominent in the fronto-central brain area. However, the outcome × agent ANOVA on the PCA-FRN amplitude failed to find a significant main effect or interaction effect, Fs < 2.50, ps > 0.13.

topographical maps for FRN (C) and P300 (D).

For the PCA-P300 component, the outcome × agent ANOVA showed a nearly significant main effect of the outcome, F (1, 18) = 3.39, p = 0.08, η<sup>p</sup> <sup>2</sup> = 0.158. Moreover, there was a significant outcome × agent interaction effect [F (1, 18) = 7.34, p = 0.014, η<sup>p</sup> <sup>2</sup> = 0.290], confirming the smaller PCA-P300 for the opponent's losses (4.59 ± 0.75 µV) was relative to following

the partner's losses (5.77 ± 0.80 µV), F (1, 18) = 6.85, p = 0.017. Additionally, there was only a significant PCA-P300 win vs. loss difference for the opponent, F (1, 18) = 6.72, p = 0.018.

To confirm the component identification, the correlations between the PCA components and the grand-average components were computed. The correlation analysis showed a significant correlation between FRN and PCA-FRN, with a Pearson correlation of 0.826, p = 0.01. Similarly, the P300 amplitude and PCA-P300 amplitude are also very significant (Pearson correlation r = 0.954), p = 0.01, which confirmed our PCA component analysis.

### Correlation Between Questionnaire Data and ERP Results

A previous study showed the association between the IOS scale and the oFRN component (Kang et al., 2010) and the relationship between oMFN and IRI scores (Fukushima and Hiraki, 2009). Therefore, we performed a correlation between the questionnaire (IRI and IOS) scores and the ERP amplitudes (i.e., FRN, P300, PCA-FRN, and PCA-P300). Interestingly, we also found a significant correlation between the IOS score and FRN component (see **Figure 4A**, r = 0.565), p = 0.012, and the correlation between the IOS score and the PCA-FRN component (see **Figure 4B**, r = 0.640) was also significant, p = 0.01. These correlation results indicate that participants' perspective taking increases as the self-other overlap increases. However, no correlation was found for the IRI scores or P300, and there was also no significant correlation between the win/loss or performance rating and the ERP components.

### DISCUSSION

We examined the group membership effects on the outcome evaluation. The observation of the others' win or loss can be used as a window to investigate the reaction to the ingroup or out-group members' performance. The behavioral rating on the football task and gambling showed no "ingroup favoritism." For example, the participants showed no partner vs. opponent performance difference in either task. Considering that the probability of win/loss was made random in the football task and equal in the gambling task, equal performance for the players is reasonable, and it ruled out potential confounding influences of the differential win/loss probability. However, the happiness rating after the experiment showed an in-group bias and schadenfreude effect that the happiness rating was significantly higher when the partner won or the opponent lost. Such an effect confirms the group membership manipulation was successful. Nevertheless, the rating could also be attributed to the expected effect or the "participant demand characteristics" (Nichols and Maner, 2008), that participants may think the experimenter expected them to show an in-group preference in the rating.

For the EEG results, our results first showed the same increase pattern of FRN-P300 components in outcome processing while observing the performances of the non-self agents (the component becomes more negative at opponent loss condition). These two components were further confirmed by the PCA analysis which is consistent with the findings of the previous study (Zhou et al., 2010). Moreover, the ERP data suggested that the amplitude of the FRN and P300 reflected the interaction between the outcome and agent. Interestingly, a prominent win vs. loss differentiation FRN effect was observed on the opponent, i.e., the loss feedback evoked a more negative deflection than win feedback. The win vs. loss differentiation effect of the FRN was consistent with many previous studies' (Miltner et al., 1997; Holroyd and Coles, 2002; Hajcak et al., 2005; Wei et al., 2015; Zheng et al., 2017), but we have not found this win vs. loss differentiation effect on partner's outcomes. Our FRN results seem to reflect the influences of social factors in this early stage of outcome evaluation, which was consistent with the previous study (Qi et al., 2018). However, further PCA-FRN failed to replicate this effect for non-significant interaction effects. One possibility is that the PCA factor is not large enough to reach a significant differentiation effect, even it shows a similar opponent's win vs. opponent's loss difference pattern (**Figure 3A**) as the ERP results (**Figure 2A**). Another possibility is that the early FRN is not sensitive to social relationships, as it may be

entangled with the P300 effect, which is consistent with a previous study showing an interpersonal effect on P300 but not on FRN (Leng and Zhou, 2010).

Likewise, the P300 component showed a similar win vs. loss differentiation effect on the opponent's feedback, but not for the partners. Specifically, the opponent's feedback P300 was more positive for the win trials than for the loss trials. Although whether the feedback P300 is sensitive to the outcome valence is still controversial (Yeung and Sanfey, 2004; Gu et al., 2011), we could infer that there is more "empathy" or concern about the opponent's feedbacks, as it would be consistent with other studies showing a win vs. loss P300 effect on the self's and other's feedback (Wu and Zhou, 2009; Leng and Zhou, 2010). The PCA-P300 results confirmed such greater concern about the opponent's win and loss as the differentiation between win vs. loss. Thereafter, we also observed a partner vs. opponent differential effect in the loss context, showing a smaller P300 and PCA-P300 for the opponent's losses than for the partner's losses. Such an effect, we believe, is attributed to the pronounced opponent loss P300 effect. Unlike Leng and Zhou's (2010) study which showed that P300 was independently modulated by an interpersonal relationship and outcome valence, our results showed the modulation effect of the interaction. The comparison of our ERP results with previous studies in an observation situation (Fukushima and Hiraki, 2006, 2009; Leng and Zhou, 2010; Wang Y. et al., 2014) found FRN and P300 in negative feedback trials became more negative compared to positive feedback trials, for the opponent only. Thus, we can conclude that the participants showed more empathy (perspective-taking) or concern for the opponent's outcome, which manifested a win vs. loss ERP differentiation effect.

As mentioned, a growing number of studies have suggested that outcome evaluation/empathy ERPs are influenced by the relationship between the agent and the observer (Itagaki and Katayama, 2008; Marco-Pallares et al., 2010; Leng and Zhou, 2010; Qi et al., 2018). In general, an experiment designed with close others or others with a higher self-resemblance will cause participants' larger FRN or P3 (Fukushima and Hiraki, 2009; Leng and Zhou, 2010). Together with previous evidence showing more empathy to in-group members (Xu et al., 2009; Contreras-Huerta et al., 2013), the ERP results showed no win vs. loss differentiation effect on the in-group partner, which seemed to be particularly surprising at first sight. The gender of the participants and the group manipulation may account for the partner's indifferent attitude. Because winning or losing did not change the participants' bonuses, the in-group control may not be as effective as out-group control. As several studies suggested, man subjects have a lesser empathetic response than women (Fukushima and Hiraki, 2006; Tousignant et al., 2017). More importantly, recent work showed that women's ERP response of outcome processing was influenced by a short-term induced affective preference, but not that of men (Wang Y. et al., 2014). Furthermore, previous brain imaging studies showing an in-group bias are mainly based on racial or relative long-term social identity (Xu et al., 2009; Contreras-Huerta et al., 2013; Fabi and Leuthold, 2018; Han, 2018). According to male-warrior hypothesis that males respond much stronger sense of competition and are more aggressive in social context (Björkqvist et al., 1994; White and Kowalski, 1994; Van Vugt et al., 2007; Van Vugt, 2009), the man-only participant population may lead to differentiation to opponent's results. Considering that the group identity manipulation in the current study was short-term and temporary, it is interpretable that the men would show very less concern about their partner's performance when the outcome was not related to their own selfinterest, but about the opponent's as the potential competition.

However, the brain potential responses showed win vs. loss differential effects on the opponent, which seems like an empathy effect. We noted that one previous ERP study showed both empathy and schadenfreude effects (Itagaki and Katayama, 2008). In Itagaki and Katayama's study, the other's loss elicited FRN (loss-win) under cooperative conditions (i.e., empathy), while the observation of the gain of player A also elicited an FRN in player B under antagonistic conditions (i.e., schadenfreude). Unlike their research, the observation task in our study was neither cooperative nor antagonistic, for the agent's outcome was irrelevant to self-benefit. Therefore, the observer was in a neutral position while viewing the partner's and opponent's performances, which was confirmed by the performance rating for the two agents. Thus, we did not find a schadenfreude effect that the opponent's win evoked a more negative FRN.

By contrast, we found an empathy-like pattern on the opponent's outcomes, that the opponent's loss evoked a more negative FRN and P300. We inferred that the perspective-taking strategic game leads to such an effect. Since the participants have social interaction with the opponent but not partners in the game understanding the opponent's mind is critical for winning the game. Previous work has shown the effect of perspective-taking on decreasing racial bias (Todd et al., 2011; Bimper, 2015; Todd and Simpson, 2016; Müller and Scherr, 2017) and stereotypic bias (Galinsky and Moskowitz, 2000; Wang et al., 2018; Mishra et al., 2019) and increasing the willingness for intergroup contact (Wang C.S. et al., 2014). More importantly, the effect of perspective-taking on improvement in intergroup attitudes was mediated by empathy (Vescio et al., 2003). Combined with Lamm's et al. (2014) study showing that perspective-taking increases empathy, we inferred that the interaction in the competition task reinforced an empathic-like ERP pattern toward the opponents. When participants take the perspective of the opponent, there was a greater overlap between the mental representations of the self and the agent (Davis et al., 1996). The higher self-other overlap results in empathy toward the opponent, which is also confirmed by the correlation between the IOS scores and the FRN component. Therefore, our results provided a shred of evidence for the self-other overlap framework that proposes perspective-taking induces a self-other overlap and further increases social cooperation and ultimately formation of social bonds (Galinsky et al., 2005).

Another interpretation of the win vs. loss differentiation to opponents' performance is that the motivation significance is relatively higher for participants. That is, existing studies have shown FRN was modulated by motivation level. For instance, the FRN was smaller when they observe the others' performance than FRN in joint action (Loehr and Vesper, 2015; Michel et al., 2018). In our study, to some extent, the competitive task has

reinforced the motivation to observe the opponents' outcomes due to the interactive game against opponents. Additionally, all the participants in the current study are men, who have higher competitiveness and win orientation in the sports domain (Gill, 1988; Gill et al., 1996), this competitive attitude may increase the motivation to monitor the opponent's performance. A recent study has also confirmed that people showed larger FRN in competitive instruction than in cooperative instruction (Cui et al., 2016).

We admit that the small sample size and the lack of a control group may limit generalization of the conclusion (Christley, 2010) 1 . It would also be very interesting to determine the woman's empathetic response under grouping and perspectivetaking manipulation. In the present study, our subjects only included man subjects for several reasons. First, existing studies showed a gender difference in empathy (e.g., women showed more empathetic responses to others) (Han et al., 2008; Lim et al., 2018; Yue et al., 2018) and schadenfreude (e.g., men exhibit more schadenfreude toward others) (Singer et al., 2006). Furthermore, the hormones oxytocin (Shamay-Tsoory et al., 2009; Tetsu et al., 2015) and testosterone (Christian and Shariff, 2017) are associated with schadenfreude or empathy, which also modulate the mentalizing network (Wu et al., 2018). Therefore, we only examined the two effects in man subjects to exclude confounding gender or hormonal factors and investigated the FRN-P300 effect of empathy. Future work that would extend these issues to other situations can provide further evidence about the interaction effect between perspective-taking and group identity on empathy. However, the higher perspective-taking opponent and temporary in-group partner manipulation in the current study only exhibited an empathy pattern to the opponent's outcome. Although the effect of PCA-FRN is not significant, such an effect of the more negative FRN and P300 in response to an opponent's loss versus an opponent's win was observed in the ERP results and PCA-P300. We also look forward to combining perspective-taking, EEG, source localization and connectivity in the future study (Liu et al., 2019) to further investigate this research question.

In summary, our results find neither an in-group bias in empathy nor an intergroup schadenfreude pattern as predicted by the in-group-favoritism hypothesis. Instead, the results of empathy toward the opponent's outcome are consistent with the perspective-taking and self-other overlap hypotheses.

### REFERENCES


Our results also provide a positive view of improving intergroup relationships and forming social bonds by perspective-taking or social interaction.

### DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

### ETHICS STATEMENT

The studies involving human participants were reviewed and approved by the Institutional Review Boards (IRB) of the Institute of Psychology, Chinese Academy of Sciences. The patients/participants provided their written informed consent to participate in this study.

### AUTHOR CONTRIBUTIONS

HY: whole experiment execution manuscript drafting and part of the statistical analysis. WN and GY: part of the statistical analysis and refine manuscript. QL: refine manuscript and discussion result. HW and XL: supervise whole experiment design, discussion result, and suggestions for organize manuscript draft.

### FUNDING

We gratefully acknowledge supports by the National Natural Science Foundation of China (31400963) to HW.

### ACKNOWLEDGMENTS

We thank Dr. Kai Wang, Ya Zheng, and Moqian Tian for their roles as confederates in our task. Experimental data of this study can be requested via email to HW, wuhy@psych.ac.cn.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2019.01162/full#supplementary-material

27–33. doi: 10.1002/1098-2337(1994)20:1<27::AID-AB2480200105>3.0. CO;2-Q


<sup>1</sup> Using G∗Power (Faul et al., 2007), we calculated that with a sample size of 19, with the current 2 × 2 design, we had a power of β = 0.80 to detect a very large effect, namely effects that are larger than pη <sup>2</sup> = 0.34/f.72. For a large sized effect (pη <sup>2</sup> = 0.14/f.40) the current study had a power of β = 0.35, and for a medium sized effect (pη <sup>2</sup> = 0.06/f.25) a power of β = 0.17. Hence we have to admit that the small sample size severely limited the generalization of the conclusions.



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Yu, Nan, Yang, Li, Wu and Liu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Reduced Hedonic Valuation of Rewards and Unaffected Cognitive Regulation in Chronic Stress

Sónia Ferreira1,2† , Carlos Veiga1,2† , Pedro Moreira1,2, Ricardo Magalhães1,2 , Ana Coelho1,2, Paulo Marques1,2, Carlos Portugal-Nunes1,2, Nuno Sousa1,2,3 and Pedro Morgado1,2,3 \*

<sup>1</sup> Life and Health Sciences Research Institute (ICVS), School of Medicine, University of Minho, Braga, Portugal, <sup>2</sup> ICVS/3B's – PT Government Associate Laboratory, Braga/Guimarães, Portugal, <sup>3</sup> Clinical Academic Center – Braga, Braga, Portugal

### Cognition can influence choices by modulation of decision-making processes. This cognitive regulation is defined as processing information, applying knowledge, and changing preferences to consciously modulate decisions. While cognitive regulation of emotions has been extensively studied in psychiatry, few works have detailed cognitive regulation of decision-making. Stress may influence emotional behavior, cognition, and decision-making. In addition, the brain regions responsible for decision-making are sensitive to stress-induced changes. Thus, we hypothesize that chronic stress may disrupt the ability to regulate choices. Herein, we used a functional magnetic resonance imaging task where fourteen control and fifteen chronically stressed students had to cognitively upregulate or downregulate their craving before placing a bid to obtain food. We found that stressed participants placed lower bids to get the reward and chose less frequently higher bid values for food. Nevertheless, we did not find neural and behavioral differences during cognitive regulation of craving. Our outcomes revealed that chronic stress impacts decision-making after cognitive regulation of craving by reducing the valuation of food rewards but not cognitive modulation itself. Importantly, our results need further validation with larger sample sizes.

Keywords: stress, decision-making, cognition, magnetic resonance imaging, fMRI, reward, human, food

### INTRODUCTION

Value-based decision-making is the ability to make a choice from competing courses of action/alternatives based on subjective values and possible outcomes attributed to them (Balleine, 2005). This process is carried out whenever a person chooses from different alternatives (e.g., choosing between eating an apple or an orange, or between going out or not). Different interacting systems are responsible for the valuation and action selection processes in the brain (Rangel et al., 2008). First, a valuation system computes the action values. A comparator system needs to evaluate the action values. An accumulator system receives and accumulates the value signals from the

### Edited by:

Ruolei Gu, Institute of Psychology (CAS), China

### Reviewed by: Sunny Dutra,

William James College, United States Claire Zedelius, University of California, Santa Barbara, United States

### \*Correspondence:

Pedro Morgado pedromorgado@med.uminho.pt †These authors have contributed equally to this work

### Specialty section:

This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Neuroscience

Received: 15 February 2019 Accepted: 28 June 2019 Published: 10 July 2019

### Citation:

Ferreira S, Veiga C, Moreira P, Magalhães R, Coelho A, Marques P, Portugal-Nunes C, Sousa N and Morgado P (2019) Reduced Hedonic Valuation of Rewards and Unaffected Cognitive Regulation in Chronic Stress. Front. Neurosci. 13:724. doi: 10.3389/fnins.2019.00724

**72**

**Abbreviations:** ACTH, adrenocorticotropic hormone; BAI, Beck Anxiety Inventory; BDI, Beck Depression Inventory; dlPFC, dorsolateral prefrontal cortex; fMRI, functional magnetic resonance imaging; GLM, general linear model; MNI, Montreal Neurological Institute; PSS-10, 10-items Perceived Stress Scale; vmPFC, ventromedial prefrontal cortex.

comparator system until the signal for one of the actions is sufficiently strong for the choice to be executed (Gold and Shadlen, 2007; Basten et al., 2010).

Values assigned to actions during the valuation process can be influenced by different factors such as the degree of risk or uncertainty of the action (Platt and Huettel, 2008; Rangel et al., 2008). Humans have a natural aversion to risky or uncertain choices and place less value on actions with temporal uncertain rewards or multiple sets of outcomes (Christopoulos et al., 2009; McGuire and Kable, 2012). Individuals often place higher values on immediate rewards rather than on future ones (Rangel et al., 2008). Social competition, cooperation, and concerns for the well-being of others also influence decisionmaking (Fehr and Camerer, 2007). Cognition can also influence choices through modulation of the decision-making processes. This cognitive regulation process may be defined as processing information, applying knowledge and changing preferences to consciously modulate our decisions. While cognitive regulation of emotional response has been extensively studied (Ochsner et al., 2004; Delgado et al., 2008; Wager et al., 2008), few works have detailed cognitive regulation of decision-making. A functional magnetic resonance imaging (fMRI) study where participants had to modulate their cravings for food showed that cognitive regulation affects decision-making through valuation regulation and behavioral control (Hutcherson et al., 2012). The ventromedial prefrontal cortex (vmPFC) is known to compute the value signal of decisions while the dorsolateral prefrontal cortex (dlPFC) modulates this signal during cognitive regulation tasks (Hare et al., 2009, 2011; Kober et al., 2010).

Cognitive regulation of both emotion and decision-making has a role in the treatment of several conditions (schizophrenia, bipolar disorder, depression, obesity, addiction, obsessivecompulsive disorder, and eating disorders) where emotional processing and decision-making are often impaired (Phillips et al., 2003; Ochsner et al., 2004). On the other hand, mental disorders such as schizophrenia, bipolar disorder, post-traumatic stress disorder, and depression are often associated with prolonged exposure to stress (Arnsten, 2015; Sousa, 2016). Stress impacts emotional processing leading to depressive and anxious behavior associated with alterations in amygdalaventromedial-prefrontal pathways. Moreover, stress elicits cognitive impairments namely working memory and attentional deficits, poor decision-making (e.g., decreased reward sensitivity or increased influence of immediate rewards), behavioral inflexibility, and learning deficits. These cognitive differences are associated with changes in prefrontal and hippocampal regions (Sandi, 2013; Arnsten, 2015; Chen and Baram, 2016; Sousa, 2016). Additionally, the brain regions implicated in decisionmaking processes are sensitive to stress-induced changes. In fact, changes in fronto-striatal networks involved in behavioral decisions have been reported in both humans and rodents after chronic stress (Dias-Ferreira et al., 2009; Soares et al., 2012; Morgado et al., 2012, 2015a; Sousa, 2016; Magalhães et al., 2018). Thus, stress seems to influence the quality of decisions (Starcke and Brand, 2012; Morgado et al., 2015b; Bryce and Floresco, 2016; Chen and Baram, 2016) because cognitive control is diminished (Yu, 2016).

Stress has also an impact on appetite and eating behavior (Ans et al., 2018) and is one of the factors for development of eating and obesity-associated conditions (Razzoli et al., 2017). Usually, the production of adrenocorticotropic hormone (ACTH) by the anterior pituitary gland leads to the release of cortisol in the adrenal cortex to stimulate hunger and feeding behavior. High cortisol levels are associated with high insulin concentrations resulting in increased caloric intake or food craving (Adam and Epel, 2007). Stress might boost these pathways leading to an increase in food intake and appetite for high-caloric food, or also reduced reward sensitivity to lowcaloric food (Razzoli et al., 2017; Ans et al., 2018; Berg Schmidt et al., 2018; Ferrer-Cascales et al., 2018), in agreement with previous stress decision-making studies demonstrating decreased reward sensitivity or increased influence of immediate rewards (Morgado et al., 2015b). Thus, stress seems to be associated with increased food reward sensitivity due to diminished selfcontrol during food choice associated with decreased functional connectivity between the vmPFC and dlPFC (Neseliler et al., 2017) and increased connectivity between the vmPFC and subcortical regions (amygdala and striatum) (Tryon et al., 2013; Maier et al., 2015).

Herein, we used an fMRI task to clarify the impact of chronic stress on cognitive regulation of decisions. Our task consisted of cognitively upregulating or downregulating craving before placing a bid to obtain food. In addition to brain responses, we analyzed behavioral parameters (food valuation score and reaction time) associated with the task, and blood hormonal changes after the task (insulin, cortisol, and glucose). Regarding the previous findings, we hypothesize that chronic stress may disrupt the ability of individuals to regulate their choices. We expect that cognitive regulation deficits after chronic stress manifest by changes in the prefrontal cortex (vmPFC and dlPFC). Subsequently, these deficits lead to decision-making impairments, namely increased reward sensitivity, underlying brain response alterations in prefrontal and striatal regions. Moreover, we expect that chronic stress participants present augmented levels of insulin, glucose, and cortisol after the stimulation with food pictures due to an increased reward sensitivity to food.

### MATERIALS AND METHODS

### Subjects

We enrolled in this study medical students from the School of Medicine of University of Minho, Portugal. All students were healthy Caucasians, right-handed, and had a healthy bodymass index. One group was under normal academic activities [control group, n = 14; 9 females/5 males; median (range) 23.00 (3.00) years of age; education 17.00 (3.00) years] and the other included subjects on the long period of preparation for the medical licensing exam [chronic psychosocial stress condition; stress group, n = 15; 10 females/5 males; 24.00 (3.00) years of age; education 18.00 (0.00) years]. This work was conducted 1 to 3 months before the exam, but students usually start preparing 1 year before the exam. Subjects were eligible if they were at least 18 years old, reported no history of psychiatric or neurological conditions, traumatic brain lesion, or substance abuse, and were not on any psychiatric medication. The groups were matched for gender (chi-squared test χ 2 (1) = 0.02, p = 0.893) but not for age (Mann–Whitney test U = 169.00, p = 0.004, effect size r = 0.56) and education level (U = 210.00, p = 2.579 × 10−<sup>8</sup> , r = 0.93).

### Ethics Statement

fnins-13-00724 November 21, 2019 Time: 13:24 # 3

The study was performed in accordance with the Declaration of Helsinki and was approved by Ethics Subcommittee for the Life and Health Sciences of University of Minho, Portugal, and by the Ethics Committee of the Hospital of Braga, Portugal. All subjects were provided with written informed consent following description of the study goals and procedures.

### Sociodemographic and Psychological Scales

Subjects filled a questionnaire to characterize gender, age, educational level, handedness, and ethnic origin. Weight and height were also measured to prevent the inclusion of participants with an unhealthy body mass index. Subjects were assessed with the 10-items Perceived Stress Scale (PSS-10) (Cohen et al., 1983; Morgado et al., 2013), the Beck Anxiety Inventory (BAI) (Beck et al., 1988), and the Beck Depression Inventory (BDI) (Beck et al., 1996). PSS-10 measures the extent to which participants perceived their life as unpredictable, uncontrollable, and overloaded during the previous month. The higher the score, the greater the intensity of perceived stress. BAI measures the severity of an individual's anxiety during the previous week. Scores lower than 8 indicate minimal anxiety. Scores higher than 7, 15, and 25 indicate mild, moderate, and severe anxiety, respectively. BDI measures the severity of depression and can be used as a screening tool. Scores lower than 14 indicate minimal depression. Higher scores indicate more severe depressive symptoms.

### Blood Sampling and Analysis

Before the fMRI acquisition, samples of venous blood were collected from all participants into a 5 mL potassium ethylenediaminetetraacetic acid tube and a serum tube. We repeated the collection immediately after the fMRI acquisition. Pre-scan blood samples were used to measure cortisol, glucose, insulin, and ACTH serum levels. In post-scan samples, we repeated cortisol, glucose, and insulin serum measurements (ACTH measurement was not repeated due to technical constraints). The collection took place between 2 and 7 pm which assures small variation in cortisol levels during this period (Minkley and Kirchner, 2012). ACTH was measured based on solid-phase, two-site sequential chemiluminescent immunometric assay, and insulin with solid-phase, enzyme-labeled chemiluminescent immunometric assay (IMMULITE 2000, Siemens AG, Germany). Cortisol levels were assessed with competitive immunoassay based on direct chemiluminescent (ADVIA Centaur and Centaur XP, Siemens AG, Germany). Glucose was measured based on the hexokinase-glucose-6-phosphate method (Dimension Vista, Siemens AG, Germany). Standard procedures were applied following the manufacturer instructions.

### Statistical Analysis

Data related with psychological scales, laboratory values, and behavioral parameters were analyzed using IBM SPSS Statistics (version 24.0; IBM Corporation, United States). Kolmogorov– Smirnov and Shapiro–Wilk tests were used to assess for normality in the distribution of data. Comparisons between groups were carried out by parametric t-tests or repeated measures ANOVA (F-test, with Bonferroni correction for multiple comparisons for post hoc tests], or non-parametric Mann-Whitney U-tests. Differences were considered statistically significant if p < 0.05. Effect sizes were calculated for all statistically significant results.

### fMRI Task

The task was adapted from Hutcherson et al. (2012). Subjects were instructed to fast for at least 4 h before their arrival and to eat a light meal before the fasting period to increase the valuation of food pictures. We also informed that they would remain in the laboratory for 30 min at the end of the experiment to eat the food they obtained during the fMRI task. The task consisted of two parts: a pre-scan rating task that provided us with a measure of the baseline value for food, and an in-scan bidding and regulation task that measured the food value under the influence of regulation.

During the pre-scan rating task, subjects were shown 150 pictures of different snack food items (e.g., cake, chips, and candy) and rated, at their own pace, how much they would like to eat them using a four-point scale (1, "Don't want it at all"; 4, "Want it a lot"). Our set of pictures was adapted to the Portuguese context of food.

Afterward, subjects received instructions for the in-scan bidding and regulation task (**Figure 1**). The 150 snack food pictures were shown again, separated into three trial conditions: indulge, distance, and natural. Each type of trial appeared 50 times, randomly interspersed over the scanning run. On each trial, before the food appeared, participants saw an abstract black-and-white symbol indicating the trial type (cue, 2 s). On indulge trials, subjects were instructed to try to increase their craving for the snack using any strategy they needed to. During distance trials, the instruction consisted on trying to decrease their craving. On natural trials, they had to allow thoughts and feelings to come naturally. Subjects had 4 s to look at the item and engage in the craving cognitive regulation task (hereinafter referred as cognitive regulation task). After the 4 s, subjects had 2 s to place a bid (0 €, 1 €, 2 €, or 3 €) for the right to eat that food at the end of the experiment. They were asked to treat each trial as if it were the only decision that counted. These bids allowed us to measure values expressed in behavior at the time of choice.

At the end of the experiment, food was auctioned using an adapted version of the Becker-DeGroot-Marschak auction (Becker et al., 1964; Plassmann et al., 2007). We gave 3 € to each subject to spend during the auction over a maximum of three trials. Snacks and snacks prices were randomly selected by drawing a paper from a bag. The bids on those trials during the fMRI task determined whether subjects got to eat that food.

Consider b the bid made by the subject during the fMRI task. During the auction, a random price a was drawn (0 €, 1 €, 2 €, and 3 € were chosen with equal probability). If b ≥ a, the participant got the item and spend a. If b < a, the subject did not get the item. The rules of the auction ensure the subjects' best strategy to bid their true value for each food. This was explained and emphasized during the instruction period. For auction effects, omissions resulted in a bid of 0 €.

### fMRI Data Acquisition

Each participant was scanned on a clinical approved 1.5 T Siemens Magnetom Avanto system (Siemens Medical Solutions, Germany) using a 12-channel receive-only head array coil. For the functional acquisition, a T2<sup>∗</sup> weighted echo-planar imaging acquisition was acquired: 38 interleaved axial slices, repetition time 2750 ms, echo time 30 ms, field of view 224 mm × 224 mm, flip angle 90◦ , in-plane resolution 3.5 mm × 3.5 mm, slice thickness 3.5 mm, and between-slice gap 0.5 mm. To optimize the sensitivity in the orbitofrontal cortex, a tilted acquisition in an oblique orientation of 30◦ relative to the anterior-posterior commissure line was used. In total, 650 volumes were acquired during the task. The task stimulus was presented using the fully integrated fMRI system IFIS-SA (Invivo Corporation, United States) and the same system was used to record the subject key-press responses. One high-resolution T1-weighted Magnetization-Prepared Rapid Acquisition with Gradient Echo sequence, with 1 mm × 1 mm × 1 mm voxel size, repetition time 2.73 s, echo time 3.48 ms, flip angle 7◦ , field of view 234 mm × 234 mm, and 176 slices was acquired. This anatomical sequence was used to project the functional maps.

### fMRI Data Preprocessing

The functional scans from each participant were preprocessed using the Statistical Parametric Mapping (SPM) version 12 (Wellcome Department of Imaging Neuroscience, Institute of Neurology, United Kingdom) using MATLAB version R2018a (The MathWorks Inc.,United States). The preprocessing procedures included: slice-timing correction using the first slice as reference; realignment to the mean volume of the acquisition; nonlinear spatial normalization to Montreal Neurological Institute (MNI) standard space and resampling to 2 mm × 2 mm × 2 mm voxel size; spatial smoothing with a 8 mm full-width at half-maximum Gaussian kernel; high pass temporal filtering at 128 s. Participants with more than 3 mm of movement (1 voxel) were excluded (n = 0).

### fMRI Data Analysis

For the first-level analysis, one general linear model (GLM) was computed per participant. For this GLM, the regressors of interest included: the type of cognitive regulation trial (1 – distance, 2 – natural, and 3 – indulge) and the corresponding bid (4 – bids after distance trials, 5 – bids after natural trials, and 6 – bids after indulge trials). The bid regressors were parametrically modulated by the bid value (0, 1, 2, and 3 €), the pre-rating score before the task (1 to 4), and the reaction time. Additional regressors included: 7 – the cue; 8 – the interstimulus interval; 9 – the omission bids; 10 – 16 the motion parameters estimated during the realignment step. The onset and duration of the regressors were defined accordingly to the stimulus represented in **Figure 1** with a boxcar function and the regressors were convolved with the canonical hemodynamic response function.

At the group level (second-level analysis), a random-effects analysis was performed using four different mixed-design ANOVA models: (1) represented the cognitive regulation during the task (enabled comparisons in average activation for each regulation trial between and within groups); (2) concerned the bidding/valuation during the task modulated by the bid value; (3) concerned the bidding/valuation during the task modulated by the pre-rating score; (4) concerned the bidding/valuation during the task modulated by the reaction time. Models (2) – (4) were used to test if food valuation was different between groups after distinct regulation trials. For all models, the group (stress vs. control) was introduced as the betweensubject factor and each trial during cognitive regulation (distance vs. natural vs. indulge) as the within-subject factor. Age and education were used as covariates for all models. All models were implemented with the GLMFlex toolbox<sup>1</sup> which uses partitioned error terms for within-group and between-group comparisons, enabling the estimation of all the effects of interest with a single model.

Results were considered statistically significant after correcting for multiple comparisons using cluster correction

<sup>1</sup>http://nmr.mgh.harvard.edu/harvardagingbrain/People/AaronSchultz/GLM\_ Flex.html

(minimum cluster size of 90 voxels). The minimum cluster size was determined with 3DClustSim (AFNI version 17.0.13; National Institute of Mental Health)<sup>2</sup> . This program determines a minimum cluster size with Monte Carlo Simulation to achieve a corrected significance of p < 0.05 with an initial voxel-wise threshold of p < 0.001. The Automated Anatomical Labeling plugin for SPM was used to classify the brain regions.

### RESULTS

### Psychological Assessment

fnins-13-00724 November 21, 2019 Time: 13:24 # 5

The stress group revealed higher levels of perceived stress (mean ± standard deviation 15.07 ± 5.23) than the control group (8.64 ± 5.27) as assessed by PSS-10 [t(27) = 3.30, p = 0.003, effect size d = 1.27]. No statistically significant differences were found for BAI (U = 117.50, p = 0.591) and BDI (U = 134.00, p = 0.217) between groups.

### Blood Sampling

The ACTH levels before the fMRI session were similar between the two groups (U = 81.50, p = 0.310).

Cortisol serum levels were not statistically significantly different between groups [group F(1,27) = 0.45, p = 0.509] nor within group before and after the fMRI session [group × time F(1,27) = 1.00 × 10−<sup>3</sup> , p = 0.971]. However, cortisol levels decreased in both groups after the task [time F(1,27) = 10.08, p = 0.004, effect size χ <sup>2</sup> = 0.27].

Glucose serum levels were not statistically significantly different between groups [group F(1,27) = 0.40, p = 0.531] and the pre and post-measurement were similar within groups [group × time F(1,27) = 0.18, p = 0.672]. However, glucose levels decreased in both groups after the task [time F(1,27) = 8.44, p = 0.007, χ <sup>2</sup> = 0.24].

Insulin serum levels were not statistically significantly different between groups [group F(1,27) = 0.42, p = 0.522] and the pre and post-measurement were similar within groups [group × time F(1,27) = 3.68, p = 0.066]. However, insulin levels decreased in both groups after the task [time F(1,27) = 9.21, p = 0.005, χ <sup>2</sup> = 0.25].

### Behavioral Analysis

Given the differences in age and education between groups, we used these variables as covariates when analyzing behavioral parameters.

We analyzed the reaction time between and within groups during the different regulation trials (distance, natural, and indulge). We found an interaction effect between the group and the reaction time across the different regulation conditions [group × condition F(2,50) = 4.00, p = 0.024, χ <sup>2</sup> = 0.14; **Table 1** represents the results for all between and within group factors and covariate effects]. Post hoc tests with repeated measures ANOVA demonstrated statistically significant reaction time differences within the control group [F(1.42,18.43) = 7.06, p = 0.010, χ <sup>2</sup> = 0.35, Greenhouse-Geisser correction for nonsphericity] and within the stress group [F(1.34,18.82) = 4.72, p = 0.033, χ <sup>2</sup> = 0.25, Greenhouse-Geisser correction for non-sphericity]. Paired t-tests with Bonferroni correction showed that the reaction time for natural trials was shorter than for distance [t(13) = 4.82, p = 0.001, d = 2.67] and indulge trials [t(13) = 3.07, p = 0.027, d = 1.70], and distance and indulge trials presented similar reaction times [t(13) = 1.00, p = 1.000] in the control group. However, we did not find significant statistical differences in the stress group during post-hoc analysis [1.00 ≤ t(14) ≤ 2.36, 0.099 ≤ p ≤ 1.000] (**Figure 2**).

Taking into account that different instructions were given during the pre-rating (how much the participants want the food) and the bidding (how much the participants want to pay for the food), we separately analyzed differences between groups in the valuation score across the regulation conditions (distance, natural, and indulge) for pre and post-regulation scores (**Table 1** represents the results for all between and within group factors and covariate effects). During pre-rating, we did not find statistically significant differences between groups or within group in terms of food valuation across the conditions. Moreover, the valuation score varied similarly among the conditions for both groups. However, during bidding, we found differences between groups [group F(1,25) = 6.91, p = 0.014, χ <sup>2</sup> = 0.22] but not within group in terms of food valuation across the conditions. Moreover, the valuation score varied similarly among the conditions for both groups. The stress group had lower valuation scores (1.06 ± 0.36 €) during bidding in comparison to the control group (1.50 ± 0.36 €) (**Figure 3**).

Moreover, we also studied differences between groups in the number of responses for each bidding value after each cognitive regulation trial inside the scanner. We found a significant interaction effect between the group and the bid value [group × valuation F(2.13,53.15) = 3.89, p = 0.024, χ <sup>2</sup> = 0.13, Greenhouse-Geisser correction for non-sphericity; **Table 1** represents the results for all between and within group factors and covariate effects]. Post hoc tests with repeated measures ANOVA demonstrated that the control [F(3,39) = 9.61, p = 7.000 × 10−<sup>5</sup> , χ <sup>2</sup> = 0.42] and the stress group [F(2.03,28.43) = 9.04, p = 0.001, χ <sup>2</sup> = 0.39, Greenhouse-Geisser correction for non-sphericity] had a different number of responses across the bid values. Paired t-tests with Bonferroni correction demonstrated that on average the stress participants bided more often 0 [t(14) = 3.93, p = 0.009, d = 1.78] and 1 € [t(14) = 3.38, p = 0.027, d = 1.81] than 3 €, while control subjects bided more times 1 [t(13) = 3.94, p = 0.010, d = 2.18] and 2 € [t(13) = 5.91, p = 3.090 × 10−<sup>4</sup> , d = 3.28] than 3 € (**Figure 4**).

### Neuroimaging Results

We tested for differences in blood-oxygen-level-dependent responses between stress and control groups during each cognitive regulation period/trial (natural, indulge, and distance) – model (1). No statistically significant brain regions were identified for overall differences between groups (main effect of group). When looking at the interaction effect between

<sup>2</sup>https://afni.nimh.nih.gov/

TABLE 1 | Results for statistical tests on behavioral variables associated with the functional magnetic resonance imaging task: reaction time, valuation score, and response frequency.


<sup>∗</sup>Statistical significance; <sup>a</sup>Greenhouse-Geisser correction for non-sphericity.

trial condition and group, there were also no statistically significant effects. Nonetheless, we found a main effect of the cognitive regulation condition in the left hemisphere in the superior (Brodmann area 22) and middle temporal gyrus (Brodmann area 21), the rolandic operculum, and the precentral gyrus (Brodmann area 6) [7.87 ≤ F(2,54) ≤ 13.97, p ≤ 0.001, 99 voxels, Montreal Neurological Institute peak voxel coordinates -60 -6 -4). Post hoc paired t-tests with Bonferroni correction demonstrated that the distance and indulge trials elicited lower activity than natural trials and that distance trials lead to higher responses than indulge trials in these regions [distance vs. natural t(28) = 2.97, p = 0.018, d = 1.12; distance vs. indulge t(28) = 2.68, p = 0.036, d = 1.01; natural vs. indulge t(28) = 4.97, p = 9 × 10−<sup>5</sup> , d = 1.88] (**Figure 5**).

With the models (2) – (4), we tested if food valuation/bidding behavior was associated with different brain activation between groups after each regulation condition, with parametric modulation by bid value (model 2), pre-rating score (model 3), and reaction time (model 4). No statistically significant regions were identified for overall differences between groups during bidding (main effect of group) for the models (2) – (4). Additionally, no statistically significant active regions were found for interaction effects of group and the bids after each category of cognitive regulation (group × cognitive regulation condition), and the main effect of the condition was also not statistically significant for the models (2) – (4).

### Task Validity

Given that our fMRI task was adapted from Hutcherson et al. (2012), here we compared our main results with these authors' significant findings to study the task validity. Since we observed behavioral differences between the control and stress groups, we assessed the validity of the task only with the control group.

As observed by Hutcherson et al. (2012), we also saw that the control group took longer while bidding after distance [t(13) = 4.82, p = 0.001, d = 2.67, with Bonferroni correction] and indulge trials [t(13) = 3.07, p = 0.027, d = 1.70, with Bonferroni correction] than natural trials [group × condition F(1.42,18.43) = 7.06, p = 0.010, χ <sup>2</sup> = 0.35, Greenhouse-Geisser correction for non-sphericity] (**Figure 2**). Moreover, the distance trials were also associated with the longest reaction time (778.30 ± 198.70 ms), followed by indulge (744.38 ± 125.21 ms),

and natural (652.98 ± 131.97 ms) trials. These reaction times values are consistent with the previous study.

Concerning the bid value, similarly to Hutcherson et al. (2012), we observed a main effect of the cognitive regulation condition (distance, natural, and indulge) in controls [F(1.23,16.05) = 16.88, p = 4.650 × 10−<sup>4</sup> , χ <sup>2</sup> = 0.56, Greenhouse-Geisser correction for non-sphericity]. The control participants bided higher on indulge [1.74 ± 0.43 €; t(13) = 2.89, p = 0.038, d = 1.60, with Bonferroni correction] and lower on distance [1.04 ± 0.28 €; t(13) = 4.22, p = 0.003, d = 2.34, with Bonferroni correction] compared to natural trials (1.50 ± 0.23 €). Bids after distance and indulge trials were also statistically significantly different [t(13) = 4.22, p = 0.003, d = 2.34, with Bonferroni correction] (**Figure 3**). The bid values in our study (0, 1, 2, and 3 €) were distinct from the original study (\$ 0.0, 0.5, 1.0, 1.5, 2.0, and 2.5), thus we could not compare the average bid values after each condition trial.

For neuroimaging data, we computed the contrasts among the cognitive regulation trials in the control group to compare our results with the original study: Distance > Natural, Natural > Distance, Indulge > Natural, Natural > Indulge, Distance > Indulge, and Indulge > Distance. We applied cluster correction for multiple comparisons (90 voxels as described in the section "Materials and Methods"). We found statistically significant results only for the contrasts Distance > Natural and Indulge > Natural. These results are in agreement with Hutcherson et al. (2012)'s findings if the same minimum cluster size is considered. For the contrast Distance > Natural, similarly to the original work, we also found statistically significant activation in temporal and posterior parietal regions (**Supplementary Table S1**). However, results did not show statistically significant activity in medial and ventrolateral prefrontal regions. For the contrast Indulge > Natural, we found statistically significant responses in the anterior cingulate cortex, the ventral, medial, and superior prefrontal cortex, temporal and parietal regions, and the supplementary motor area (**Supplementary Table S1**). Thus, our results are concordant with these authors' previous findings.

### DISCUSSION

We studied how chronic stress influences decision-making on food valuation after cognitive regulation (increasing/indulge or decreasing/distance food craving) in medical students. Behavioral, biochemical and neuroimaging analysis were performed to address this question. We found that stressed participants present decreased food valuation scores. This result was reinforced by a higher number of responses for the lowest bid values for food in the stress group. The biochemical analysis (serum levels of insulin, cortisol, and glucose) did not show statistically significant differences between the control and stress group. The neuroimaging results did not demonstrate statistically significant differently activated brain regions between the stressed and control participants during cognitive regulation of craving and decision-making/bidding.

Although the acute stress response is generally beneficial, i.e., promotes adaptation to stressful stimulus, prolonged activation of the stress response produces deleterious effects on the body and brain, affecting cognitive processes such as decision-making (Mcewen, 2004; McEwen and Gianaros, 2011; Sousa, 2016). One of the main findings of the present study is that stressed individuals presented lower scores during food valuation, in contrast with our initial hypothesis. This may translate a blunted hedonic capacity or reward sensitivity (Berenbaum and Connelly, 1993; Porcelli et al., 2012; Bryce and Floresco, 2016; Porcelli and Delgado, 2017; Uy and Galván, 2017), as anhedonia has been associated with higher perceived stress scale scores (Pizzagalli et al., 2007) and stress causes changes in regions related to hedonic/rewarding behavior such as the amygdala, orbitofrontal cortex, vmPFC, and ventral and dorsal striatum (Gorwood, 2008; Porcelli et al., 2012; Bessa et al., 2013; Stark et al., 2015). Moreover, the distribution of the number of responses was higher for lower bids in the stress group, i.e., stressed subjects seemed less prone to place high bids for food. A previous work including a food-related task discovered decreased reward sensitivity associated with alterations in the putamen activity after acute stress induction (Born et al., 2010). Another report pointed out that acute stress does not potentiate craving after stimulation with food pictures (Stojek et al., 2015). Moreover, animal research indicates that acute stress reduces the motivation to work for food rewards (Bryce and Floresco, 2016). Other studies have also shown that acute and chronic stress mitigate brain responses to food stimuli in reward pathways (Wierenga et al., 2018). These results support the idea that stress participants have a reduced valuation attributed to food rewards. However, other studies have shown increased sensitivity to high-caloric food rewards in stressed individuals (Razzoli et al., 2017; Ans et al., 2018; Berg Schmidt et al., 2018; Ferrer-Cascales et al., 2018). Reward processing might be different when participants are stimulated with food pictures or real food. Moreover, the inclusion of chronic or acute stress

FIGURE 3 | Representation of the normalized rating scores for food pictures for each trial condition (natural, distance, and indulge) before performing the functional magnetic resonance task (before cognitive regulation) and during the functional magnetic resonance task (after cognitive regulation) for the stress and control groups (the normalized scores represent the ratio between given score and maximum score). Before cognitive regulation, we did not find statistically significant differences between groups or within group in terms of food valuation across the conditions. After cognitive regulation, the stress group had lower average valuation scores in comparison to the control group [group F(1,25) = 6.91, p = 0.014, χ <sup>2</sup> = 0.22]. The black star represents statistically significant differences. The main bars represent the mean values and the error bars represent the standard error.

subjects bided more times 1 [t(13) = 3.94, p = 0.010, d = 2.18] and 2 € [t(13) = 5.91, p = 3.090 × 10−<sup>4</sup> significant differences. The main bars represent the mean values and the error bars represent the standard error.

models might also account for different results regarding reward sensitivity (Porcelli and Delgado, 2017). However, our results need confirmation with larger sample sizes.

While bidding, the stressed subjects did not present differential brain activity when compared to control subjects, despite the behavioral differences in the valuation score. We were expecting that poor cognitive self-control reflected in reduced prefrontal activation (Hare et al., 2009, 2011; Kober et al., 2010; Hutcherson et al., 2012) would lead to higher responses in striatal and amygdalar regions associated with increased reward sensitivity (Louis et al., 2009; Tryon et al., 2013; Maier et al., 2015; Neseliler et al., 2017). Other studies have found controversial results demonstrating that reduced striatal activity was associated with high levels of stress and increased food craving (Hommer et al., 2013). However, we did not observe cognitive differences between the groups and reward sensitivity was decreased. Our sample size may have limited the statistical power of this analysis. Thus, further research should be conducted to understand the neural correlates of decision-making after cognitive regulation since our results are not conclusive.

Our neuroimaging results did not show brain activity differences between groups during cognitive regulation. According to previous studies, the vmPFC and dlPFC are regions responsible for cognitive regulation in this decision-making context (Hare et al., 2009, 2011; Kober et al., 2010; Hutcherson et al., 2012). Thus, the absence of changes in these regions between groups in our work might indicate that the processes for cognitive regulation of food craving are not affected by chronic stress, or that our specific model of chronic stress might not lead to changes in cognitive modulation of craving. However, previous works revealed that cognitive control is diminished under stress, leading to emotional and habitual-biased decisionmaking (Yu, 2016), and increased reward sensitivity for food (Tryon et al., 2013; Maier et al., 2015; Neseliler et al., 2017). Our moderate sample size might have hindered putative differences between groups. Nonetheless, as shown in **Figures 2**, **3**, stressed participants were able to modulate their responses, demonstrating an effective cognitive regulation, although the average food bidding score was lower than in controls. Both groups were capable of effectively using cognitive regulation to change the value placed on food during regulated trials. Moreover, both groups took longer times during bidding after regulated trials. Indeed, we found that activity in the superior and middle temporal gyrus, rolandic operculum, and precentral gyrus was differently modulated by trials with cognitive regulation of craving versus non-regulated trials in both groups. Previous authors provided evidence for a functional connection between the vmPFC and the precentral gyrus during food-related decisions, and for the correlation between food ratings and the response in the middle temporal gyrus (Kober et al., 2010; Hare et al., 2011; Hutcherson et al., 2012). Moreover, the temporal gyrus is also involved in food imagery (Hommer et al., 2013). Thus, the regulatory success does not seem to be affected by stress. During cognitive control tasks, attentional narrowing might occur after stimulation with negative pictures with threat and sadness-related content (van Steenbergen et al., 2011;

The main bars represent the mean values and the error bars represent the standard error. L, left; R, Right.

Melcher et al., 2012; Papazacharias et al., 2015). Thus, the negative emotional state in the stress group (e.g., fear of falling the final exam) might have led to higher attentional focus during cognitive regulation that might compensate cognitive deficits associated with chronic stress. Nonetheless, our results need further validation with larger sample sizes to rule out a putative effect of chronic stress in cognitive regulation of craving.

Insulin, cortisol, and glucose levels are expected to decrease after fasting (Kirschbaum et al., 1997; Adam and Epel, 2007; Figlewicz, 2015; Tiedemann et al., 2017). However, peripheral concentrations of cortisol rise after stimulation with food images due to appetite enhancement, while insulin and glucose levels seem to be unaffected (Schmid et al., 2005; Schüssler et al., 2012; Kroemer et al., 2013). In our study, both groups presented a decreased in insulin, glucose, and cortisol levels after the fMRI task. Thus, the effects of fasting might have potentially surpassed the effects of stimulation with food pictures (Brede et al., 2017). Nonetheless, this hypothesis needs further testing. We were expecting increased craving in the stress group after a deficient cognitive regulation and increased reward sensitivity to food (Tryon et al., 2013; Maier et al., 2015; Neseliler et al., 2017). However, our results agree with the fact that we found reduced valuation of rewards in the absence of cognitive regulation alterations in the stress group, suggesting that overall craving was reduced. For controls, the instructions to differently modulate craving might have led to balanced changes in blood parameters after stimulation with food pictures. Thus, our results might derive from fasting since they occurred in both groups.

Importantly, our results are limited by the sample size. These results need to be replicated with larger samples to avoid false negative and positive conclusions. Moreover, the results might have been influenced by the unbalanced proportion of females and males per group, given that gender differences were found in decision-making under stress (Yu, 2016; Wemm and Wulfert, 2017). However, we focused on group differences and groups were matched for gender ratio.

Our results show that the capacity to perform cognitive regulation of craving is not impaired after prolonged stress. However, chronic stress reduces the value attributed to food rewards after craving modulation. Importantly, our conclusions are limited by the small sample size and need further validation with larger samples. These findings are relevant to guide subsequent studies on cognitive regulation of food-related decision-making for eating and obesity-associated disorders. Cognitive control techniques might be used to tackle decisionmaking impairments in these conditions (Louis et al., 2009; May et al., 2012).

### DATA AVAILABILITY

fnins-13-00724 November 21, 2019 Time: 13:24 # 10

The datasets generated for this study are available on request to the corresponding author.

### ETHICS STATEMENT

This study was carried out in accordance with the Declaration of Helsinki. All subjects gave written informed consent. The protocol was approved by the Ethics Subcommittee for the Life and Health Sciences of University of Minho, Portugal, and by the Ethics Committee of the Hospital of Braga, Portugal.

### AUTHOR CONTRIBUTIONS

SF, CV, PM, RM, AC, PMore, and CP-N acquired, analyzed, and interpreted the data and wrote the manuscript. NS and PMorg supervised the work, interpreted the data, and wrote and

### REFERENCES


reviewed the manuscript. All authors participated in the design of the experiments and approved the manuscript.

### FUNDING

This work was funded by the MyHealth project (DoIT – Desenvolvimento e Operacionalização da Investigação de Translação, Contract DoIT-13853) and the BIAL Foundation, Porto, Portugal (Grant No. PT/FB/BL-2016-206). This work was also funded by the FEDER funds, through the Competitiveness Factors Operational Programme (COMPETE), and by national funds, through the Foundation for Science and Technology (FCT), under the scope of the project POCI-01-0145-FEDER-007038. This manuscript has been developed under the scope of the project NORTE-01-0145- FEDER-000013, supported by the Northern Portugal Regional Operational Programme (NORTE 2020), under the Portugal 2020 Partnership Agreement, through the European Regional Development Fund (FEDER). PaM was supported by a fellowship of the project "SwitchBox" (Contract HEALTH-F2-2010-259772). SF was supported by a combined Ph.D. scholarship from the FCT and the company iCognitus4ALL – IT Solutions, Lda (PD/BDE/127839/2016). RM was supported by the combined FCT and Enlightenment, Lda fellowship (PDE/BDE/113604/2015) from the PhDiHES program.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2019.00724/full#supplementary-material



psychiatric disorders. Ann. N. Y. Acad. Sci. 1032, 1–7. doi: 10.1196/annals.1314. 001



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Ferreira, Veiga, Moreira, Magalhães, Coelho, Marques, Portugal-Nunes, Sousa and Morgado. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Corrigendum: Reduced Hedonic Valuation of Rewards and Unaffected Cognitive Regulation in Chronic Stress

Sónia Ferreira1,2†, Carlos Veiga1,2†, Pedro Moreira1,2, Ricardo Magalhães 1,2, Ana Coelho1,2 , Paulo Marques 1,2, Carlos Portugal-Nunes 1,2, Nuno Sousa1,2,3 and Pedro Morgado1,2,3 \*

<sup>1</sup> Life and Health Sciences Research Institute (ICVS), School of Medicine, University of Minho, Braga, Portugal, <sup>2</sup> ICVS/3B's – PT Government Associate Laboratory, Braga/Guimarães, Portugal, <sup>3</sup> Clinical Academic Center – Braga, Braga, Portugal

Keywords: stress, decision-making, cognition, magnetic resonance imaging, fMRI, reward, human, food

### **A Corrigendum on**

**Reduced Hedonic Valuation of Rewards and Unaffected Cognitive Regulation in Chronic Stress** by Ferreira, S., Veiga, C., Moreira, P., Magalhães, R., Coelho, A., Marques, P., et al. (2019). Front. Neurosci. 13:724. doi: 10.3389/fnins.2019.00724

### Approved by:

Frontiers Editorial Office, Frontiers Media SA, Switzerland

### \*Correspondence:

Pedro Morgado pedromorgado@med.uminho.pt

†These authors have contributed equally to this work

### Specialty section:

This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Neuroscience

Received: 04 November 2019 Accepted: 05 November 2019 Published: 22 November 2019

### Citation:

Ferreira S, Veiga C, Moreira P, Magalhães R, Coelho A, Marques P, Portugal-Nunes C, Sousa N and Morgado P (2019) Corrigendum: Reduced Hedonic Valuation of Rewards and Unaffected Cognitive Regulation in Chronic Stress. Front. Neurosci. 13:1252. doi: 10.3389/fnins.2019.01252 In the original article, there was an error. The psychometric scale used to measure the depression scores was the "Beck Depression Inventory" and not the "Beck Depression Inventory II."

A correction has been made to the **Materials and Methods**, subsection **Sociodemographic and Psychological Scales**:

"Subjects filled a questionnaire to characterize gender, age, educational level, handedness, and ethnic origin. Weight and height were also measured to prevent the inclusion of participants with an unhealthy body mass index. Subjects were assessed with the 10-items Perceived Stress Scale (PSS-10) (Cohen et al., 1983; Morgado et al., 2013), the Beck Anxiety Inventory (BAI) (Beck et al., 1988), and the Beck Depression Inventory (BDI) (Beck et al., 1996). PSS-10 measures the extent to which participants perceived their life as unpredictable, uncontrollable, and overloaded during the previous month. The higher the score, the greater the intensity of perceived stress. BAI measures the severity of an individual's anxiety during the previous week. Scores lower than 8 indicate minimal anxiety. Scores higher than 7, 15, and 25 indicate mild, moderate, and severe anxiety, respectively. BDI measures the severity of depression and can be used as a screening tool. Scores lower than 14 indicate minimal depression. Higher scores indicate more severe depressive symptoms."

A correction has also been made to **Results**, subsection **Psychological Assessment**:

"The stress group revealed higher levels of perceived stress (mean ± standard deviation 15.07 ± 5.23) than the control group (8.64 ± 5.27) as assessed by PSS-10 [t(27) = 3.30, p = 0.003, effect size d = 1.27]. No statistically significant differences were found for BAI (U = 117.50, p = 0.591) and BDI (U = 134.00, p = 0.217) between groups."

Lastly, a correction has been made to the **Abbreviations** section:

"ACTH, adrenocorticotropic hormone; BAI, Beck Anxiety Inventory; BDI, Beck Depression Inventory; dlPFC, dorsolateral prefrontal cortex; fMRI, functional magnetic resonance imaging; GLM, general linear model; MNI, Montreal Neurological Institute; PSS-10, 10-items Perceived Stress Scale; vmPFC, ventromedial prefrontal cortex."

The authors apologize for this error and state that this does not change the scientific conclusions of the article in any way. The original article has been updated.

### REFERENCES


Copyright © 2019 Ferreira, Veiga, Moreira, Magalhães, Coelho, Marques, Portugal-Nunes, Sousa and Morgado. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Working Memory Load Enhances the Attentional Capture of Low Reward History

### Yujie Wu1,2, Tingni Li<sup>1</sup> and Zhe Qu<sup>1</sup> \*

<sup>1</sup> Department of Psychology, Sun Yat-sen University, Guangzhou, China, <sup>2</sup> State Key Laboratory of Cognitive Neuroscience and Learning, IDG/McGovern Institute for Brain Research, Beijing Normal University, Beijing, China

Attention priority of reward history, also called value-driven attentional capture (VDAC), is different from that of saliency or contingency. The magnitude of VDAC was found to be correlated with working memory capacity, but how cognitive control interacts with the attentional allocation of reward association is not clear. Here, we examined whether the distraction by learned color-reward association would change under different working memory load conditions. Participants were first trained with color-reward associations by searching a green/red circle with low/high reward. Then, during the test session, participants needed to search a unique shape while a green/red shape was either presented as a distractor or not shown at all. To manipulate the working memory load in the test, a digital memory task was integrated with the visual search task in half of the trials (memory load condition), but not in the other half (no-load condition). Consistent results were found in two experiments that the magnitude of attentional capture caused by low-value distractors was larger under memory load condition than under no-load condition, while there was no enough evidence supporting the influence of memory load on attentional capture by high-value distractors. These results suggested that working memory load, which occupied part of cognitive resources, reduced the priority of target information and might also modulate the strength of reward association holding in working memory. These findings extend the knowledge regarding the influence of working memory load on attentional capture of reward and suggest that reward-induced distraction is dynamic and could be modulated by cognitive control.

Edited by:

Ruolei Gu, Institute of Psychology (CAS), China

### Reviewed by:

Lihui Wang, Otto von Guericke University Magdeburg, Germany Tingting Wu, Queens College (CUNY), United States

> \*Correspondence: Zhe Qu quzhe@mail.sysu.edu.cn

### Specialty section:

This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Psychology

Received: 13 June 2019 Accepted: 18 November 2019 Published: 06 December 2019

### Citation:

Wu Y, Li T and Qu Z (2019) Working Memory Load Enhances the Attentional Capture of Low Reward History. Front. Psychol. 10:2722. doi: 10.3389/fpsyg.2019.02722 Keywords: attentional capture, reward, working memory, cognitive control, load theory

### INTRODUCTION

Reward experience is appealing and powerful, which can guide our attention and shape our behavior. As we all know, attention plays an important role in selecting sensory input into awareness. In recent years, an increasing number of studies found that reward history can induce capture of attention (Raymond and O'Brien, 2009; Failing and Theeuwes, 2014), which has challenged the traditional view of dichotomy of attentional selection (Anderson et al., 2011b; Awh et al., 2012): one is top-down or goal-directed attentional selection, which means the stimuli that are related to current goal will get attention (Wolfe et al., 1989); the other one is bottom-up or stimulus-driven attentional capture, which means attention will be captured by a stimulus that is salient on physical properties (e.g., Theeuwes, 1992).

So, it is important to explore the cognitive basis of attentional priority of reward history and reveal the interactions between it and other cognitive processes.

The attentional priority of reward-related information was first observed in visual search tasks. First, participants need to search the target of certain colors and get rewards trial by trial in a training session. After that, in the test session, the target changes into a unique shape, and the former target color turns to be one of the distractors in some trials and no reward will be given anymore. As a result, the reaction times (RTs) of subjects get longer when the distractors that used to be associated with reward during training appeared compared to no rewardrelated distractor, which is called value-driven attentional capture (VDAC, Hickey et al., 2010; Anderson et al., 2011b; Awh et al., 2012; Chelazzi et al., 2013).

Value-driven attentional capture is different from goaldirected attention selection and stimulus-driven attention capture for evidence from three aspects. First, even if the color that is rewarded during training is unique among distractors, the high and low value of color make the effect of VDAC different (Anderson et al., 2011a), which means the capture is not determined by the salient color itself but by the value of reward. Second, the reward-related stimulus has no common with a current target, and its effect is context-dependent. Anderson (2015) found the effect size of VDAC depends on the situation where the association of reward value is built. Third, learning the association between a stimulus and a reward is an indispensable part of VDAC. The repeated occurrence of a target and a reward makes the brain generate a predictive signal for the target when it shows up, which modulates the priority of attention (Sali et al., 2014). Accordingly, reward history seems to attract attention in a direct and special way rather than through a goal-directed or stimulus-driven way. Although their mechanisms are different, saliency and reward can have interactions on attention. For example, Wang et al. (2013) found that VDAC was easier to appear when the reward was associated with color than with shape. However, it remains unclear how VDAC interacts with goal-directed cognitive control.

Working memory acts as an important role in goal-directed attention control (Desimone and Duncan, 1995; Baddeley, 1996). Recent studies revealed that working memory also plays an important role in VDAC. For example, Anderson et al. (2011a) found that people with lower working memory capacity exhibited stronger attentional capture by stimuli with reward history. Another eye movement study (Anderson and Yantis, 2012) further verified a negative correlation between working memory capacity and extent of VDAC. However, these studies did not manipulate the cognitive control to reveal the modulation of working memory on VDAC directly.

Working memory load will occupy cognitive resources and may influence the attentional selection in two contradictory ways. The first way is that, working memory load might impair attentional selection because the processing of distractors cannot be well inhibited by executive control. An fMRI study found that correctly allocating attention to target location needs taskrelated information with good priority in working memory; high working memory load will impair the priority of target and hence processing of distractor will get enhanced (De Fockert et al., 2001). Specifically, in a flanker task, the distraction is stronger under high working memory load than low working memory load (Lavie et al., 2004). Similarly, in a visual search task, memorizing digits at the same time makes performance more susceptible to irrelevant unique-color stimuli compared to no memory task condition (Lavie and De Fockert, 2005). The second way is that, working memory load might reduce distraction and facilitate allocation of attention to targets. SanMiguel et al. (2008) found that, in a visual classification task, working memory load will reduce the distraction of novel sound both behaviorally and as an index by an attenuation of the late phase of the novelty-P3 EEG signal. So, it is proposed that the effect of working memory load on distraction depends on the nature of the distraction (SanMiguel et al., 2008). Considering the priority of reward history is learningand context-dependent, which is different from that of saliency, the influence of working memory load on VDAC is worthy to be investigated.

In the current study, we aim to reveal the modulation of working memory load on VDAC directly. Given the previously mentioned two different ways concerning the relationship between working memory and distraction, working memory load may enlarge or reduce VDAC. However, in most studies of VDAC, the reward is associated with visual stimuli that can cause response conflict in visual attention tasks, which is different from totally task-irrelevant auditory distractions used in SamMiguel's visual classification task (2008). So, here we suppose that the effect of VDAC by visual stimuli would get stronger under working memory load condition. By combining a dual-task paradigm (Pashler, 1994; Lavie and De Fockert, 2005; Muller-Gass and Schröger, 2007) with the value-learning procedure, we manipulated working memory load through a digital memory task and measured VDAC in a visual search task to test the hypothesis.

### EXPERIMENT 1

### Materials and Methods Participants

Nineteen undergraduates participated in the formal experiment. One participant did not gaze at the fixation point during the experiment and was rejected from the analysis. Data of eighteen participants were analyzed (nine males and nine females). They were 17–22 years old with a mean age of 19.5 (SD = 1.29). All of them had normal or corrected-to-normal visual acuity and had no mental disease history. The experiment was approved by the Ethics Committee of the Department of Psychology, Sun Yat-sen University. Before the experiment, all participants signed informed consent in accordance with the Declaration of Helsinki.

### Task and Stimuli

To examine the modulation of working memory on the attentional priority of reward history, two variables, working memory load (no load/load) and association between color and reward (no reward/low reward/high reward), were manipulated

within subjects. During the training session, the associations were built through visual search and reward feedback trial by trial. In the test session, each participant finished both a single task (with no working memory load) and a dual task (with working memory load).

In the visual search task during the training session, six circles (each 2.3◦ × 2.3◦ ) with different colors (all possible colors used in training and testing session: red, green, blue, cyan, pink, orange, yellow, white) were presented with an equal distance in an imaginary ring (10◦ × 10◦ ) (**Figure 1**). Search targets were red and green circles, and only one of them would appear in each trial. In a red or green circle, there was a horizontal or vertical white bar (1.1◦ ). In circles with other colors, the bars were 45 degrees off-axial directions. Participants were required to judge the orientation of the bar in a red or green circle as soon as possible. The reward of the current trial and total amount of reward appeared after response. There were two kinds of colorreward associations which were balanced between subjects. For half participants, red color was associated with high reward and green color was associated with low reward. For the other half participants, red color was associated with low reward while green color was associated with high reward. In each trial, if the color (red/green) associated with high reward showed up and the bar inside it was correctly responded, then the participant would receive 0.275 yuan with an 80% possibility or 0.055 yuan with a 20% possibility. If the target was the color (green/red) with low reward, then the participant would receive 0.055 yuan with an 80% possibility or 0.275 yuan with a 20% possibility.

In the visual search task during the testing session, five diamonds and one circle or five circles and one diamond composed an imaginary ring in the search display. Participants were asked to judge the orientation of the bar (horizontal or vertical) in a unique shape. Bars in other shapes were 45 degrees off-axial directions. All six shapes had different colors, and in 2/3 trials, one of the distractors was red or green. In the remaining 1/3 trials, which is the no-reward condition, all colors appeared in the training session, but none of them was associated with reward before. Participants were informed that there would be no reward in the test session and the colors were irrelevant anymore.

In working memory task during the test session, for each trial, six numbers which were randomly chosen from 1 to 9 were presented in the center of the screen. No three consecutive numbers were incrementing or decrementing, and all numbers appeared with equal probabilities. The probe number was also randomly chosen from 1 to 9. Participants were asked to report whether the probe number was among the memory array.

### Procedure

Each participant was tested in a dimly lit room, sitting in front of the screen with a viewing distance of 50 cm. Before training, participants practiced for 24 trials with no reward. There were 240 trials in the training session, and 360 trials in the test session. In the test session, half trials were under single-task condition (i.e., without working memory task), and the other half were under dual-task condition (i.e., with working memory task). The two conditions were divided into two blocks separately, and the order of blocks was balanced between subjects.

In the training session, each trial began with a fixation display for 400, 500, or 600 ms randomly. Then search display appeared until response or for 800 ms. Participants were asked to press key m for a vertical bar or press key z for a horizontal bar in the target circle as soon as possible. Incorrect responses were followed by a 1000 Hz beep with 100 ms duration. After that, the feedback display showed up for 1500 ms to inform participants of the current reward and their total rewards. The inter-trial interval was 1000 ms.

In the test session, for the dual-task condition, each trial began with a fixation display (400, 500, or 600 ms duration) and then was followed by a digit array for 1500 ms. Participants were told to memorize all numbers in the array and to judge whether the probe number was one of them at the end of the trial. Then after another fixation display for 2000 ms, a search display appeared. Participants needed to press key m for a vertical bar or to press key z for a horizontal bar in the unique shape as soon as possible. After response or 1500 ms, the probe number appeared, and participants should press a for yes and k for no within 3000 ms. The inter-trial interval was 500 ms.

For the single-task condition, a search display appeared following an initial random period of fixation (400, 500, or 600 ms) and waited for response until 1200 ms. And the trial ended with an inter-trial interval of 500 ms.

### Analysis

According to the appearance of reward associated colors, the trials in the test session were classified as three conditions: high reward, low reward, and no reward. Besides, based on the singletask or dual-task condition, trials in the test session were also tagged as no-load and load conditions. A within-subject repeated measures analysis of variance (ANOVA) was used to analyze RTs and error rates. All the statistical tests were conducted by using SPSS (version 22), and the statistical threshold was set at 0.05. Post hoc analysis was performed using Fisher's Least Square Difference (LSD) test (Milliken and Johnson, 1984).

### Results

### Value-Driven Attentional Capture

In the training session, RTs of high-reward and low-reward conditions were comparable [t(17) = −0.72, p = 0.479]. We divided the training session into two phases and conducted a two-way repeated measures ANOVA to reveal the training effect. A significant main effect of training was found [F(1, 17) = 17.72, MS<sup>e</sup> = 439.59, p = 0.001, η 2 <sup>p</sup> = 0.51], with shorter RTs in the second phase (527.21 ± 40.00 ms, M ± SD) compared to the first phase (548.02 ± 31.48 ms). However, there was no significant reward effect [F(1, 17) = 0.90, MS<sup>e</sup> = 295.00, p = 0.356, η 2 <sup>p</sup> = 0.05] or interaction between reward and training [F(1, 17) = 0.02, MS<sup>e</sup> = 6.47, p = 0.882, η 2 <sup>p</sup> = 0.00]. So, participants responded faster with more training and two target colors had roughly equal attentional priority during training.

Reaction times and error rates of visual search task in the test session are shown in **Table 1**. A repeated measures ANOVA on RTs showed a significant memory load effect [F(1, 17) = 89.82, MS<sup>e</sup> = 8307.20, p < 0.001, η 2 <sup>p</sup> = 0.84], a significant reward effect [F(2, 34) = 8.53, MS<sup>e</sup> = 701.62, p = 0.001, η 2 <sup>p</sup> = 0.33],

and a significant interaction between them [F(2, 32) = 5.63, MS<sup>e</sup> = 401.80, p = 0.008, η 2 <sup>p</sup> = 0.25]. Then we analyzed the reward effects in single-task and dual-task conditions separately. Consistent with previous studies on VDAC, in the single-task condition we found a significant reward effect [F(2, 34) = 6.22, MS<sup>e</sup> = 433.78, p = 0.005, η 2 <sup>p</sup> = 0.27] and a linear trend on three reward value levels (no vs. low vs. high) [F(1,17) = 11.43, MS<sup>e</sup> = 457.22, p = 0.004, η 2 <sup>p</sup> = 0.40], which indicated that attentional capture cannot be merely explained by the selection history in training session. Post hoc comparisons showed significant longer RTs in low- and high-reward conditions compared to no-reward condition (Fisher's LSD: p = 0.012 and p = 0.004, respectively). As for the dual-task condition, the reward effect was also significant [F(2, 34) = 8.28, MS<sup>e</sup> = 669.64, p = 0.001,η 2 <sup>p</sup> = 0.33]. Post hoc comparisons showed significant longer RTs in low-reward condition compared to no-reward and high-reward conditions (Fisher's LSD: p < 0.001 and p = 0.010, respectively), but there was no significant RT difference



between no-reward and high-reward conditions (Fisher's LSD: p = 0.288).

A repeated measures ANOVA on error rates showed a significant main effect of working memory load [F(1,17) = 18.51, MS<sup>e</sup> = 52.94, p < 0.001, η 2 <sup>p</sup> = 0.52]. No significant main effect of reward [F(2,34) = 29.29, MS<sup>e</sup> = 18.06, p = 0.212, η 2 <sup>p</sup> = 0.09] or interaction between reward and memory load [F(2,32) = 3.18, MS<sup>e</sup> = 13.13, p = 0.054 < 0.01, η 2 <sup>p</sup> = 0.16] was found. Different with a previous study (Lavie and De Fockert, 2005), we found error rates was higher when there was no memory load compared to when there was memory load, suggesting that participants in the current experiment was more careful in the dual task. Since neither reward effect nor interaction was significant for error rates, there was no trade-off between error rates and RTs.

### The Interaction Between Reward and Working Memory Load

Considering that the meaning of interaction between 2-level working memory load and 3-level reward association might be complicated and indirect, to further explain the interaction, we subtracting RTs of no-reward condition from those of lowand high-reward conditions to get two magnitudes of VDAC, and then reconducted a 2 × 2 repeated measures ANOVA with factors being load and reward (**Figure 2**, left). Results showed that the interaction between working memory load and reward was significant [F(1,17) = 10.52, p = 0.005, η 2 <sup>p</sup> = 0.38], whereas neither the main effect of load [F(1,17) = 0.13, p = 0.72, η 2 <sup>p</sup> = 0.01] or the main effect of reward [F(1,17) = 1.40, p = 0.254, η 2 <sup>p</sup> = 0.08] was significant. Post hoc pairwise comparisons were

conducted by using the least significant difference (LSD) method. Attentional capture of low reward was significantly stronger under working memory load condition than under no-load condition (Fisher's LSD: p = 0.016). However, attentional capture of high reward showed no significant difference between load and no-load conditions (Fisher's LSD: p = 0.264). These results indicated that, attentional capture driven by low reward history was strengthened in the dual-task condition while attentional capture of high reward was equivalent between single-task and dual-task conditions. Besides, under no-load condition, there was no significant difference of attentional capture between low- and high-reward conditions (Fisher's LSD: p = 0.309), but under load condition, attentional capture of low reward was greater than that of high reward (Fisher's LSD: p = 0.010).

The 2 × 2 interaction between load and reward (low vs. high reward) above could not fully explain the aforementioned 2(load) × 3(reward) interaction. Considering the data patterns shown in **Table 1**, it would be helpful to do another 2 × 2 repeated measures ANOVA on no-reward and lowreward conditions (Cohen, 2008). We found a significant interaction between load and reward (no reward vs. low reward) [F(1,17) = 7.18, MS<sup>e</sup> = 217.41, p = 0.016, η 2 <sup>p</sup> = 0.30], which indicated a steeper increase of RTs with memory load for lowreward condition compared to no-reward condition.

In the digit memory task, a repeated measures ANOVA showed no significant main effect of reward [F(2,34) = 0.20, MS<sup>e</sup> = 0.001, p = 0.818, η 2 <sup>p</sup> = 0.01] on memory accuracies (92.4, 92.5, and 91.9% in no-reward, low-reward, and high-reward conditions, respectively), meaning that attentional capture by reward association did not impair the performance of memory task.

### EXPERIMENT 2

In Experiment 1, we found greater attentional capture driven by low reward history under working memory load, while no significant change of attentional capture caused by high-reward association was observed between load and no-load conditions. One might argue that, in the training or test session, participants might sometimes search for a horizontal or vertical bar instead of specific colors or shapes. To reduce this confounding factor and to verify the results in Experiment 1, we changed all orientations of bars in distractors into axial directions in Experiment 2.

### Participants

Nineteen undergraduates participated in Experiment 2. One participant's performance in the training session was poor (accuracy = 0.66), which might lead to insufficient trials to learn the association between color and reward. So, the data of this participant was excluded from further analysis. Eighteen participants (6 males and 12 females) were 18–23 years old, with a mean age of 19.94 years old (SD = 1.35). All of them had normal or corrected-to-normal visual acuity and had no mental disease history. The experiment was approved by the Ethics Committee of the Department of Psychology, Sun Yat-sen University. Before the experiment, each participant signed informed consent in accordance with the Declaration of Helsinki.

### Materials and Methods

The design, procedure, and most of the stimuli of Experiment 2 were the same as Experiment 1. The only difference was that the bar orientations inside distractors were horizontal or vertical in both the training and test sessions.

### Results

### Value-Driven Attentional Capture

To ensure the training effect of reward associations, we divided the training session into two phases as in Experiment 1and did a repeated measures ANOVA on RTs. Similar to the results of Experiment 1, we found a significant main effect of training [F(1, 17) = 7.91, MS<sup>e</sup> = 421.293, p = 0.012, η 2 <sup>p</sup> = 0.32], but no significant effect of reward [F(1, 17) = 1.27, MS<sup>e</sup> = 370.26,

p = 0.28, η 2 <sup>p</sup> = 0.07]. Different from Experiment 1, a significant interaction between training phase and reward [F(1, 17) = 5.627, MS<sup>e</sup> = 146.13, p = 0.03, η 2 <sup>p</sup> = 0.25] was found, which resulted from a greater RT decrease with training for the high-reward condition (first phase: 563.52 ± 28.42 ms; second phase: 543.16 ± 29.83 ms) compared to the low-reward condition (first phase: 561.88 ± 31.18 ms; second phase: 555.03 ± 35.98 ms).

RTs and error rates of visual search task in the test session are shown in **Table 2**. A repeated measures ANOVA on RTs showed a significant load effect [F(1,17) = 97.64, MS<sup>e</sup> = 3791.33, p < 0.001, η 2 <sup>p</sup> = 0.85], a significant reward effect [F(2,34) = 13.13, MS<sup>e</sup> = 639.96, p < 0.001, η 2 <sup>p</sup> = 0.44], and a marginally significant interaction between them [F(2,32) = 2.77, MS<sup>e</sup> = 662.64, p = 0.077, η 2 <sup>p</sup> = 0.14]. Reward effects were significant under both no-load condition [F(2,34) = 4.00, MS<sup>e</sup> = 526.21, p = 0.028, η 2 <sup>p</sup> = 0.19] and load condition [F(2,34) = 10.48, MS<sup>e</sup> = 776.39, p < 0.001, η 2 <sup>p</sup> = 0.38]. In single visual search task (no-load condition), consistent with previous studies on VDAC and Experiment 1, we also found a significant linear trend of RTs under three value levels (no vs. low vs. high) [F(1,17) = 9.15, MS<sup>e</sup> = 4199.89, p = 0.008, η 2 <sup>p</sup> = 0.35]. Post hoc comparisons showed longer RTs under high-reward condition compared to no-reward condition (Fisher's LSD: p = 0.008) and there was no significant difference between no-reward and low-reward conditions (Fisher's LSD: p = 0.247) or between low- and highreward conditions (Fisher's LSD: p = 0.138). So, for no working memory load condition, both experiments indicated a linear increment of attentional capture with increased value of reward. As for load condition, post hoc comparisons showed significant longer RTs under low- and high-reward conditions compared to no-reward condition (Fisher's LSD: p = 0.001 and p < 0.001, respectively), but no significant difference was found between low- and high-reward conditions (Fisher's LSD: p = 0.725).

A repeated measures ANOVA on error rates showed no significant main effect of reward [F(2, 34) = 2.28, MS<sup>e</sup> = 27.62, p = 0.118, η 2 <sup>p</sup> = 0.12], no significant main effect of working memory load [F(1, 17) = 0.93, MS<sup>e</sup> = 70.46, p = 0.349, η 2 <sup>p</sup> = 0.05], and no significant interaction [F(2, 34) = 1.16, MS<sup>e</sup> = 22.42, p = 0.326, η 2 <sup>p</sup> = 0.06]. So, there was no trade-off between RTs and error rates.

### The Interaction Between Reward and Working Memory Load

As in Experiment 1, we also calculated the magnitude of attentional capture by subtracting RTs of no-reward condition from those of low- and high-reward conditions and did a 2 × 2


repeated measures ANOVA (**Figure 2**, right). The main effect of load is marginally significant [F(1, 17) = 3.99, MS<sup>e</sup> = 1984.05, p = 0.062, η 2 <sup>p</sup> = 0.19], but there was no significant main effect of reward [F(1, 17) = 0.455, MS<sup>e</sup> = 682.78, p = 0.509, η 2 <sup>p</sup> = 0.03] or significant interaction [F(1, 17) = 10.52, MS<sup>e</sup> = 663.92, p = 0.228, η 2 <sup>p</sup> = 0.08]. Although the interaction between reward and working memory load was not significant, which is different with Experiment 1, we ran post hoc comparisons to make sure that this inconsistency resulted from the high-reward condition. Post hoc pairwise comparisons were conducted by using the LSD method. Consistent with findings in Experiment 1, attentional capture of low reward was significantly stronger under working memory load than under no-load condition (Fisher's LSD: p = 0.022), whereas attentional capture of high reward showed no significant difference between load and no-load conditions (Fisher's LSD: p = 0.313). Meantime, no significant difference of attentional capture was found between low and high rewards under either load or no-load condition (Fisher's LSD: p = 0.725 and p = 0.138, respectively). Compared with the results in Experiment 1, the difference of interaction between the two experiments appeared to stem from the different patterns in high-reward condition.

To better explain the marginally significant 2 × 3 interaction and data patterns in **Table 2**, we also did a 2 × 2 repeated measures ANOVA on load and reward (no reward vs. low reward) as in Experiment 1. Again, a significant interaction was found [F(1,17) = 6.36, MS<sup>e</sup> = 577.40, p = 0.022, η 2 <sup>p</sup> = 0.27], which indicated a steeper RT increase with memory load for low-reward condition compared to no-reward condition. So, combining the results of two separate ANOVAs, the interaction between load and reward could be accounted for by enhanced attentional capture of low reward under working memory load.

### DISCUSSION

Reward history can guide and attract our attention no matter what our current goal is. It is especially essential to know how attention is allocated when people are under a load of a dual task. Under certain conditions, working memory load can improve (SanMiguel et al., 2008) or impair (De Fockert et al., 2001; Lavie and De Fockert, 2005) attentional selection. The current study focused on the interaction between working memory load and VDAC. Both experiments revealed that working memory load enhanced the attentional capture of low reward history but had mild or no effect on the capture of high reward history. Our results suggest that working memory load may not only impair the maintenance of target information but also modulate learned color-reward association.

### Mechanism of Working Memory Load on Value-Driven Attentional Capture

One robust finding in both Experiment 1 and 2 is the enlargement of attentional capture of low reward history under memory load condition relative to no-load condition, which is consistent with what load theory predicts. According to the load theory (Lavie et al., 2004), when the task is relatively easy and perceptual load is low, the remaining cognitive sources will be distributed to

irrelevant distractors, and attention is in charge of eliminating them. The efficiency of the elimination process is influenced by working memory (Lavie et al., 2004), and higher working memory load would cause stronger distraction (Muller-Gass and Schröger, 2007; Lavie, 2010). For example, in a visual search task where participants needed to find a unique shape while a unique color appeared as a distractor, memorizing digits at the same time made RTs much longer than no memory task condition (Lavie and De Fockert, 2005). Working memory load will enhance the processing of distractor (De Fockert et al., 2001), which can even improve the processing of a low-contrast Gabor stimulus in the presence of collinear flanking Gabors (De Fockert and Leiser, 2014). In our study, working memory load may hamper the inhibition of learned color-reward association, which enlarged the extent of attentional capture of reward and therefore prolonged the RT to find the target.

Another possible mechanism of memory load effect on attention is that the distribution of attention is dispersive at the beginning of each trial (Eriksen and James, 1986), and it is hard to get focus due to working memory load and insufficient resources (Ahmed and De Fockert, 2012a). Ahmed and De Fockert (2012b) found improved performance during attention to a global level under working memory load. The spread of attention caused by working memory load can even reduce the inattentional blindness (De Fockert and Bremner, 2011).

In Experiment 1, the attentional capture of low reward history was greater than that of high reward history under working memory load condition. To further examine this finding, we considered a possible reason and designed Experiment 2 to test it. We speculated that in Experiment 1, the window of attention could be large and dispersive, and hence participants might in some trials search for the horizontal or vertical bars by only attending their orientations while ignoring the colors or shapes, which made the color-reward associations hard to be built and therefore less capable of capturing attention. In Experiment 2, we made color/shape the only feature defining the target in the training/test session by changing all distracting bars into axial directions (vertical or horizontal). Different from the result of more attentional capture of low reward than high reward under load condition in Experiment 1, in Experiment 2, no significant difference of attentional capture was found between low- and high-reward conditions under load condition. This inconsistent patterns between experiments indicated that high-reward association in Experiment 2 exhibited higher attentional priority compared to Experiment 1, supporting that the manipulation of changing orientations of distractors is an efficient way to strengthen reward associations. Another difference between the two experiments was the training effect of reward association. In the training session of Experiment 1, there was no direct evidence supporting that low and high rewards improved visual search performance in different degrees. However, in the training session of Experiment 2, we found a greater decrease of RT for high-reward condition compared to low-reward condition, which indicated a better training effect of color-reward associations in Experiment 2 than in Experiment 1. Taken together, the inconsistent data patterns of low and high rewards under load condition in two experiments might be due to different strengths of the high-reward association in two experiments. This also implies that association strength during training is a key experimental factor to induce an attentional capture of reward and should be taken carefully.

As for why the load effect on high reward attentional capture is not significant in both Experiment 1 and 2, there might be three possible reasons. The first is the training of reward association might be not enough, which made both attentional capture by reward history and memory load effect hard to be observed. For example, the training session might be a little short, or the associations between reward and color in the training session are relatively weak. Actually, in Experiment 1, participants could finish the training session quite well even when ignoring the colors of shapes. After changing the association strength by making color (red and green) the only feature bound with the target (or reward) in training session in Experiment 2, the mean magnitude of attentional capture by high reward increased with load numerically (although not significant). The second speculation is that attentional capture of high reward already reached the ceiling, which cannot be further enlarged by working memory load. In Experiment 2, attentional capture of both low and high rewards increased with load numerically, and besides, under the load condition, attentional capture of low and high rewards reached a same level. So the results can be a combination of load effect and ceiling effect. Finally, there is a third possibility that working memory load may impair the highreward association with color held in working memory more than low-reward association, which is a counterforce to capture enlargement caused by less inhibition. Although there is no direct evidence supporting this hypothesis, as we will discuss in the next session in detail, there do exist some findings indicating that working memory load can interfere with reward association stored in memory space.

### Working Memory and Reward Association

In the present study, we did not find a significant difference in attentional capture of high reward between no-load and load conditions. This may be because the reward association is held in working memory, and working memory load may constrain or even impair this information, which makes the facilitation of reward association on distractor processing constant or even reduced. Infanti et al. (2015) found learned feature-reward associations would interfere with mnemonic representations during encoding and holding periods of working memory when no reward was provided. This indicates that reward associations learned through training can be easily invoked by working memory and interact with contents in working memory when the reward-related feature appears. So, it is reasonable that in current experiments, when red and green colors appeared in the test session, the color-reward associations were invoked, and the priority of target information was hampered, leading to attentional distraction. As for load condition, working memory load would occupy memory space, which might interfere with the color-reward association that also stored in memory space.

Therefore, the final consequence depended on the level of reward association remained in memory space.

Besides, Gong and Li (2014) found working memory performance was improved when the items were in the high reward-associated color than those in the low rewardassociated or non-rewarded color. Their results also showed interaction between reward association and working memory contents. Different from Gong and Li 's (2014) study which focused on how the reward association influences working memory, our study addressed how working memory modulates attention allocation with reward association and added some new information to this field.

In the current study, we used no memory load rather than low memory load to shorten the duration of test session, which could avoid possible attenuation of learned reward associations along with time. A similar design (no load vs. high load) was also adopted in a previous attentional capture study (Lavie and De Fockert, 2005; Experiment 1). A limitation of such designs is that the difference between no-load condition and high-load condition would inevitably include task switching in addition to memory load, which makes task switching a potential confounding factor. Future studies are needed to investigate whether task switching and memory load have different influence on attentional distraction of reward history.

In summary, consistent results were found in two experiments that attentional capture of low-reward association was enhanced under memory load condition relative to no-load condition, while no significant memory load effect was found in attentional capture of high-reward association. We propose that working memory load, which occupies part of cognitive resources, hampers the priority of target information during the process of attentional selection. Our findings extend the knowledge of the influence of working memory load on attentional capture

### REFERENCES


of reward, suggesting that attentional distraction caused by reward association is dynamic and could be modulated by cognitive control.

### DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

### ETHICS STATEMENT

The studies involving human participants were reviewed and approved by the Ethics Committee of the Department of Psychology, Sun Yat-sen University. Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.

### AUTHOR CONTRIBUTIONS

YW and ZQ designed the research. YW performed the research and analyzed the data. YW, TL, and ZQ wrote the manuscript.

### FUNDING

This study was supported by grants from the National Nature Science Foundation of China (31271190) and the Humanities and Social Science Fund of Ministry of Education of China (19YJA190004).



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Wu, Li and Qu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Impact of Callous-Unemotional Traits and Externalizing Tendencies on Neural Responsivity to Reward and Punishment in Healthy Adolescents

Yonglin Huang1,2,3, Tingting Wu1,2, Yu Gao1,3 \*, Yuyang Luo<sup>4</sup> , Ziyan Wu<sup>4</sup> , Shawn Fagan1,3 , Stephanie Leung<sup>5</sup> and Xiaobo Li<sup>4</sup>

<sup>1</sup> Department of Psychology, Brooklyn College, The City University of New York, Brooklyn, NY, United States, <sup>2</sup> Department of Psychology, Queens College, The City University of New York, Queens, NY, United States, <sup>3</sup> The Graduate Center, The City University of New York, New York, NY, United States, <sup>4</sup> Department of Biomedical Engineering, New Jersey Institute of Technology, Newark, NJ, United States, <sup>5</sup> Department of Psychology, Pace University, New York, NY, United States

Both externalizing behavior and callous-unemotional (CU) traits in youth are precursors to later criminal offending in adulthood. It is posited that disruptions in reward and punishment processes may engender problematic behavior, such that CU traits and externalizing behavior may be linked to a dominant reward response style (e.g., heightened responsivity to rewards) and deficient punishment-processing. However, prior research has generated mixed findings and work examining both the sole and joint contribution of CU traits and externalizing problems related to functional brain alterations is lacking. In this pilot functional magnetic resonance imaging study, we measured externalizing behavior and CU traits in a community sample of adolescents (n = 29) and examined their impacts on brain activity associated with anticipation and receipt of reward and punishment using the Modified Monetary Incentive Delay task. We found that CU traits were associated with greater activation of the ventral striatum (VST) during reward anticipation. However, this effect became non-significant after controlling for externalizing behavior, indicating substantial overlap between the CU and externalizing measures in explaining VST activation when anticipating reward. In addition, externalizing behavior (but not CU) was significantly negatively associated with amygdala activation during punishment receipt, even after controlling for CU traits. The present findings extend previous evidence of hyper-responsivity to reward and hyporesponsivity to punishment in relation to psychopathic traits and antisocial behavior to non-clinical, non-incarcerated youths.

Keywords: callous-unemotional, externalizing, reward, punishment, adolescence

### Edited by:

Ruolei Gu, Institute of Psychology (CAS), China

### Reviewed by:

Joe Jacques Simon, Heidelberg University, Germany Andrea Glenn, University of Alabama, United States Xu Gong, Shenzhen University, China

> \*Correspondence: Yu Gao yugao@brooklyn.cuny.edu

### Specialty section:

This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Neuroscience

Received: 14 July 2019 Accepted: 25 November 2019 Published: 13 December 2019

### Citation:

Huang Y, Wu T, Gao Y, Luo Y, Wu Z, Fagan S, Leung S and Li X (2019) The Impact of Callous-Unemotional Traits and Externalizing Tendencies on Neural Responsivity to Reward and Punishment in Healthy Adolescents. Front. Neurosci. 13:1319. doi: 10.3389/fnins.2019.01319

### INTRODUCTION

fnins-13-01319 December 11, 2019 Time: 17:45 # 2

Early life presence of externalizing behavior and psychopathy are considered to be precursors to juvenile delinquency and later criminal offending. In adults, psychopathic traits are a constellation of personality characteristics comprised of callousness, lack of empathy, superficiality, and impulsivity (Hare, 2003). Research has extended the concept of psychopathy to youth by identifying the core traits (lack of empathy and/or guilt, shallow and limited affect), which are referred to as callous-unemotional (CU) traits (Frick, 1995; Barry et al., 2000; Frick et al., 2014). Externalizing behavior, including aggressive and rule-breaking behaviors, are behaviors that violate societal norms and infringe on others' rights (Liu, 2004; Calkins and Keane, 2009). The terms "externalizing behavior" and "antisocial behavior" have been used interchangeably by some researchers, although others argue that "externalizing behavior" should be reserved to characterize destructive behaviors exhibited by youth that are less severe than antisocial behaviors such as negative, hostile, and defiant acts (Shaw and Winslow, 1997; Liu, 2004). Externalizing behavior and CU traits are highly correlated (e.g., Charles et al., 2012; Pihet et al., 2015). Furthermore, the presence of CU traits in youth is associated with more severe externalizing behavior (e.g., conduct problems, delinquency, aggression) (see Frick and Dickens, 2006 for a review), and is especially useful for predicting a subgroup of individuals with antisocial behavior that are more serious and chronic in nature (Frick and White, 2008). In fact, the Diagnostic and Statistical Manual of Mental Disorders (DSM-5; American Psychiatric Association [APA], 2013) includes a CU specifier as a feature of conduct disorder [a disorder in youth characterized by externalizing behavior that tends to precede antisocial personality disorder (ASPD)]. ASPD is a personality disorder marked by persistent disregard and/or violation of other's rights, and is often accompanied by criminal and aggressive behavior (American Psychiatric Association [APA], 2013). While only 5% of youth exhibit a pervasive and egregious pattern of both CU traits and externalizing/antisocial behavior, by adulthood this 5% accounts for a staggering 50% of crime in the United States alone (Loeber et al., 2000; Hinshaw and Lee, 2003). Therefore, it is essential to understand the etiology of CU traits and externalizing behavior in youths to help combat the future development of maladaptive behaviors.

One potential etiological pathway that engenders externalizing behavior is hyperactivity to rewards and/or hypoactivity to punishment (Frick and Marsee, 2006; Pardini, 2006; Byrd et al., 2014). Specifically, reward oversensitivity in externalizing youth can result in persistent reward-seeking behavior (Quay, 1993), and antisocial youth may rely more heavily on acting on appetitive drives than evading punishment (O'Brien and Frick, 1996). Meanwhile, insensitivity to punishment has also been implicated in psychopathy and antisocial behavior (Dadds and Salmon, 2003; Hawes and Dadds, 2005). Deficits in punishment processing may result in failure to adopt appropriate behavior via passive avoidance learning (Newman and Kosson, 1986; von Borries et al., 2010), whereby externalizing/antisocial behavior may be the behavioral manifestation of punishment insensitivity (Lykken, 1995). Impaired punishment processing (i.e., hypo-responsivity) may lead to lower levels of anxiety and fear anticipation, thereby giving rise to psychopathic and antisocial tendencies. One conceptualization has postulated that psychopathy and externalizing/antisocial behavior may arise from deficits in the modulation of both reward and punishment systems (Patterson and Newman, 1993; Wallace and Newman, 2008; Byrd et al., 2014).

Processing of reward can be distinguished by two phases: anticipation of a reward and its delivery (Knutson et al., 2001b). Evidence from functional magnetic resonance imaging (fMRI) studies on healthy populations has suggested that similar regions are activated during reward anticipation and receipt (e.g., insula, dorsal and ventral striatum) (Silverman et al., 2015), with the ventral striatum (VST) being critically implicated during the processing of reward (Knutson et al., 2001b; Oldham et al., 2018). Similarly, an anticipatory phase and the delivery of the punishment are also included in the processing of punishment (Delgado et al., 2009): anticipation elicits activation of the VST, amygdala, thalamus, and insula (Oldham et al., 2018), while punishment receipt activates the anterior cingulate cortex (ACC), insula, thalamus, and orbitofrontal cortex (OFC) (Wrase et al., 2007). One recent study indicates that overlapping neural substrates, including VST, thalamus, and amygdala are implicated in both reward and punishment anticipation (Oldham et al., 2018). Taken together, anticipation and receipts of rewards and punishments may implicate brain regions that are largely overlapping, but also distinctive.

Altered function in many of these regions has been linked to reward and punishment processing deficits seen in antisocial and psychopathic adults (Hyde et al., 2013). Meanwhile, hyposensitivity to punishment and/or hypersensitivity to reward have largely been implicated in the development of externalizing behavior and CU traits in youths (Matthys et al., 2004; Rubia et al., 2009; Bjork et al., 2010), although abnormalities in various regions have been reported. For example, in one study, adolescent males with early onset conduct disorder showed decreased activation of the OFC during reward receipt (Rubia et al., 2009). In another study, adolescents diagnosed with either oppositional defiant disorder (ODD), attention deficit hyperactivity disorder (ADHD), or conduct disorder (CD) exhibited greater activation in nucleus accumbens (NAcc; considered to be a primary subregion of the VST) and the ACC in response to reward receipt (Bjork et al., 2010) when compared to healthy controls. In contrast, an investigation on 16- to 19- year-old adolescents with disruptive behavioral disorders (DBD) had compared DBD persisters (who showed early onset and persisted into late adolescence/adulthood), DBD desisters (who showed late onset and eventually ceased disruptive behaviors) and healthy controls (Cohn et al., 2015). They found that the DBD persisters showed blunted activation in the VST, but increased activation in the amygdala, to the receipt of the monetary gain, as compared to the other two groups. In addition, CU traits were associated with reduced amygdala activation in response to monetary receipt (Cohn et al., 2015). Finally, in boys aged 8- to 11- years with clinically significant conduct problems, no significant association was found between CU

traits and neural response to reward receipt in any of the above regions (Byrd et al., 2018). Overall, most of the research on youth samples support the proposition that CU traits and externalizing tendencies are associated with hyper-responsivity to reward, although more work is warranted to address prior mixed findings.

In regards to punishment, several studies have found impaired aversive conditioning in psychopathic, CU, and externalizing populations, which is consistent with the punishment hyporesponsivity theory. Yet, studies have also yielded conflicting results and few have looked at both CU traits and externalizing behavior together. Youth with externalizing behavior (Gao et al., 2009, 2010b), conduct disorder (Fairchild et al., 2008), and those with CU traits from the community (Fung et al., 2005) have lower skin conductance responses (SCRs), an index of autonomic arousal, to cues that signaled punishment, indicating their lack of fear for impending punishments or risks. Studies that examined responses to the receipt of punishment have yielded comparable results to those of impending punishment. Youth with DBD showed reduced eye-blink startle responses (van Goozen et al., 2004) and reduced SCRs to uncued aversive tones (Herpertz et al., 2003) than normal controls. In youth with DBD and CU traits, reduced amygdala activation in response to punishment has been reported (Finger et al., 2011), but failed to replicate in another study (Byrd et al., 2018). However, Byrd et al. (2018) did not find any association between conduct problems and reductions in amygdala activation following punishment. In contrast, Cohn et al. (2015) found that DBD persisters had increased amygdala activation to punishment feedback.

Although not specifically related to reward and punishment, there is burgeoning evidence for an interactive role of CU traits in relation to brain deficits in externalizing adolescents. CU traits in youth with disruptive behavior have been associated with atypical brain functioning, particularly reduced amygdala response to socio-affective cues (e.g., fearful expressions) (Marsh et al., 2008; Jones et al., 2009; Viding et al., 2012; White et al., 2012), which in turn may lead to increases in proactive, goal-directed, aggressive behavior observed in youths with CU traits (Lozier et al., 2014). Alternatively, youths with behavioral problems and unspecified CU traits exhibit elevated activation in the amygdala, insula, and striatum in response to socio-affective stimuli (Herpertz et al., 2008; Passamonti et al., 2010). Research has shown that while CU traits are positively correlated with externalizing behavioral problems, these variables are, respectively, negatively and positively correlated with amygdala responses to socio-affective stimuli (Lozier et al., 2014). Moreover, externalizing boys with high CU traits exhibited amygdala hypo-reactivity to fearful faces, whereas externalizing boys with low CU traits were hyperreactive (Viding et al., 2012). A more recent study reported similar results: youth with high CU traits showed amygdala hyporeactivity when making judgments about causing fear in others and CU traits moderated the relationship between externalizing behavior and both the functional connectivity and activity of the amygdala (Cardinale et al., 2018). CU traits were also found to account for structural abnormalities in the amygdala of children with externalizing problems (Cardinale et al., 2019). While preliminary, these results suggest that CU traits and externalizing symptoms may play an interactive role in predicting brain deficits in youths.

In this pilot study, we aimed to extend the prior findings to non-clinical, non-incarcerated adolescents. Not only are non-clinical, non-incarcerated youth underexplored, but further examination of this group is important in identifying potential risk factors for the development of externalizing behavior and CU traits. Externalizing behavior (via the Child Behavior Checklist, CBCL; Achenbach, 1991) and CU traits (via the Inventory of Callous Unemotional Traits, ICU; Frick, 2004) were assessed in a group of adolescents from the community. They completed a modified Monetary Incentive Delay task (MID; Knutson et al., 2001a), in which participants were shown three types of geometric cues, each associated with either potential gain, loss, or no gain or loss (neutral). Their brain activation during anticipation and receipt of monetary gain and loss were acquired. Studies on healthy adolescents and adults have shown involvement of the NAcc, caudate, putamen, thalamus, and insula during the reward and loss anticipation phases of the MID (Cho et al., 2013). We expected to see hyper-responsivity to reward and hypo-responsivity to punishment in relation to externalizing and/or CU tendencies. Given that the presence of CU traits may designate a subgroup of youth with more severe externalizing tendencies (Frick and White, 2008), our second aim was to test if CU traits would interact with externalizing behavior in predicting brain activation in response to reward and punishment. It was hypothesized that adolescents with higher levels of both would show the most aberrant neural responses (e.g., hyper-responsivity to reward and hypo-responsivity to punishment).

In addition, given that prior studies have reported high correlations between externalizing behavior and CU traits (Charles et al., 2012; Pihet et al., 2015) and that they share similar etiological profiles (Frick and Dickens, 2006), we investigated the unique contribution of CU traits and externalizing behavior to reward/punishment processing in this pilot study. We predicted that there would be a great amount of overlap between externalizing behavior and CU traits in explaining brain activation. Finally, since prior research has implicated low IQ and high social adversity in antisocial behavior (e.g., Moffitt et al., 1981; Fagan et al., 2017), we included measures of IQ and social adversity as covariates. Sex and pubertal status were also included as covariates given that sex differences in the structural and functional abnormality have been found in antisocial populations (e.g., Raine et al., 2011; Visser et al., 2013) and an increasing number of studies have illustrated the effects of puberty on brain structure (Uroševic et al., 2014 ´ ; Herting and Sowell, 2017). To the authors' knowledge, this is the first study to investigate an interaction effect of CU traits and externalizing behavior in relation to reward and punishment processing deficits in youths.

### MATERIALS AND METHODS

### Participants

Data were collected as part of the Healthy Childhood Study (HCS), an ongoing longitudinal study that follows healthy

children through development. Participants and their families were originally recruited from the metropolitan Brooklyn, New York community when children were 7- to 10- years old. The original cohort consisted of 340 participants [48.2% male, mean age = 9.06, standard deviation (SD) = 0.60] and their main caregivers. Youth participants with any history of drug use, psychiatric disorders, intellectual disabilities, or developmental disorders were excluded from recruitment. More details of the full cohort can be found in Gao and Zhang (2016).

From the original cohort, 32 adolescents were randomly selected and invited to participate in the current fMRI study when they were 11- to 14-year-old. Three were excluded due to excessive head motion (>8 mm in translation or >5 ◦ in rotation) during the task. The remaining 29 adolescents were comprised of 15 girls and 14 boys (mean age = 12.3, SD = 0.8). Twenty-three of them were right-handed. The ethnic breakdown was as follows: 58.6% Black, 24.1% Hispanic, 13.8% Caucasian, and 3.4% mixedrace/other. Caregiver participants consisted of biological mothers (86.2%) and biological fathers (13.8%).

Participants and their main caregivers were invited to the Translational and Molecular Imaging Institute of Icahn School of Medicine at Mount Sinai (ISMMS) in New York for the interview assessments, the mock scan, and the actual scan, which lasted approximately 2 h in total. Participating families were financially compensated \$150 for their participation. All procedures were approved by the Institutional Review Boards of the City University of New York and the ISMMS. Written informed parental consent and youth assent were obtained from each family before participation. After consenting, caregivers filled out the CBCL and ICU. Youths filled out the ICU and Self-Rating Scale for Pubertal Development (see below) after the brain scan.

### Measures

### Externalizing Problems

Externalizing behavior was measured via caregiver's report using the Child Behavior Checklist (CBCL; Achenbach, 1991). The CBCL contains 112 items concerning a child's behavior, including internalizing (77 items) and externalizing behavior (35 items), within the past 12 months. The externalizing subscale is comprised of the aggression (e.g., "Cruelty, bullying, or meanness to others"), and delinquency (e.g., "Breaks rules at home, school, or elsewhere") subscales. Items are rated on a 3-point scale ranging from 0 (not true) to 2 (very true or often true). The total externalizing score was computed as the sum of all relevant items for each participant. Internal consistency of the externalizing subscale (Cronbach's α) was 0.73 in our sample. The externalizing scores (**Figure 1A**; mean: 5.9, SD = 5.3) were positively skewed in our sample with a skewness of 2.39 [standard error (SE) = 0.43] and kurtosis of 8.39 (SE = 0.85).

### Callous-Unemotional (CU) Traits

Both caregivers and youth participants filled out their respective versions of the Inventory of Callous Unemotional traits (ICU; Frick, 2004). The ICU is a 24-item questionnaire developed to provide a more comprehensive assessment of CU traits, composited of the callous, uncaring, and unemotional subscales. It is a 4-point rating scale ranging from 0 (not at all true) to 3 (definitely true). The total caregiver-report and self-report CU trait scores were computed separately for each participant. One participant's self-report data in the callous subscale had one missing item, which was replaced by the average score of the other items in this subscale. Internal consistency of the caregiver-report CU scores for our sample was high (α = 0.89), while the selfreport was acceptable (α = 0.61). The greater score of the two reports (e.g., caregiver- vs. self-report) was used to compute the final CU trait score for each participant, as recommended by the ICU and Antisocial Process Screening Device Manual (Frick and Hare, 2001). The CU scores (**Figure 1B**; mean: 24.9, SD = 7.9) were positively skewed in our sample with a skewness of 0.62 (SE = 0.43) and kurtosis of −0.59 (SE = 0.85).

### Pubertal Status

Adolescents filled out the Self-Rating Scale for Pubertal Development (Carskadon and Acebo, 1993). Its rating is based on a 4-point scale: 1 ("has not yet begun"), 2 ("has barely"),

curves in black represent the fitted distributions. (C) Positive correlation between these two measures. Blue dots: boys. Red dots: girls.

3 ("definitely underway"), 4 ("seem complete"), or unknown (i.e., "I don't know"). It contains three questions for both boys and girls regarding growth in height, body hair growth, and skin changes. Girls answer two additional questions about breast growth and menstruation (and a third question about the age of menstruation, if applicable), while boys answer two additional questions regarding deepening of the voice and facial hair growth. A pubertal status score (prepubertal, early pubertal, mid pubertal, late pubertal, or postpubertal) was computed for each participant based on guidelines from (Crockett, unpublished).

In addition, participants' IQ scores and social adversity levels were acquired when they were initially recruited. Full-scale IQ was estimated using four subtests of the Wechsler Intelligence Scale for Children-Fourth Edition (WISC-IV; Wechsler, 2003): Verbal Comprehension Index (VCI), Perceptual Reasoning Index (PRI), Working Memory Index (WMI), and Processing Speed Index (PSI). The four-factor indices of the WISC-IV have high reliabilities, ranging from Cronbach's α of 0.88 to 0.94 (Kaufman et al., 2006). For this study, we assessed VCI using the Vocabulary task, WMI using the Digit Span task, PRI using the Matrix Reasoning task, and PSI using the Coding task. The total scaled score across the three subtests was then converted to the estimated FS-IQ following the Tellegen and Briggs procedure (Tellegen and Briggs, 1967).

A social adversity index was computed from the caregiver's responses to ten questions based on previous literature (Raine, 2002; Gao et al., 2010a; Zhang and Gao, 2015). A total adversity score was created by adding 1 point for each of the following variables: Divorced Parents (single-parent family, remarriage, or living with guardians other than parents), Foster Home, Public Housing, Welfare Food Stamps, Parent Ever Arrested (either parent has been arrested at least once), Parents Physically Ill, Parents Mentally Ill, Crowded Home (five or more family members per room within the home), Teenage Mother (aged 19 years or younger when the child was born), and Large Family (having five or more siblings by 3 years of age). Items were scored dichotomously with a 0 (no) or 1 (yes). Higher total scores reflect higher social adversity.

### Monetary Incentive Delay (MID) Task

We employed a modified version of the Monetary Incentive Delay Task (**Figure 2**), adapted from Knutson et al. (2001a) and Samanez-Larkin et al. (2007), to examine the reward and punishment related brain responses. In this task, a Pavlovian conditioning procedure was used with each trial including two phases of interest: anticipation and outcome. The anticipation phase begins with a geometric visual cue displayed for 2000 milliseconds (ms) followed by a 2000–2500 ms central fixation crosshair, with an average total length of 4.26 (SD = 0.01, range = 4.0 - 4.5) seconds (s). Each geometric cue was associated with a particular outcome: the circle (reward cue) was associated with potential reward (monetary gain), the square (punishment cue) was associated with potential punishment (monetary loss), and the triangle (neutral cue) was associated with no gain or loss. Participants were explicitly told the meaning of each geometric cue when the task was being explained prior to the practice session. Immediately after the anticipation phase, a pentacle appears on the center of the screen for a short period as a target and participants were required to hit the response button as soon as they detected the target. Only responses

made within the window of target duration were considered as correct responses.

For each participant, the initial target duration was set as the mean response time from the 18 practice trials. For each of the following trials, the target duration was adaptively altered based on performance on prior trials to limit the current total hit rates as 66%. A fixation cross was displayed after the target for 2000 ms minus the target duration. Then, in the outcome phase, the feedback was provided for 2000 ms, including response accuracy ("Good job!" for target hit within the time window and "Sorry you missed!" for missed target), together with trial-specific and cumulative rewards earned. If the target was hit within the time window after a reward cue (circle), participants would win \$2.00; otherwise, they would gain \$0.00. Hit after a punishment cue resulted in losing \$0.00, whereas a miss would result in losing \$2.00. Hit or miss after a neutral cue resulted in neither gain nor loss (±\$0.00). The inter-trial interval was 2000–3000 ms. There were 45 trials in each run, including 15 reward, punishment, and neutral trials each, presented in random order. Each run began with a 15 s fixation period and ended with another 15 s fixation period followed by a feedback of the total rewards earned from the current run. Each run lasted about 9.5 min. There were two runs for each participant, resulting in a total of 90 trials lasting about 19 min.

### fMRI Data Acquisition

MRI acquisitions were obtained on a 3T Siemens Magnetom Skyra scanner with a 32-channel phase-array coil at the ISMMS. Each scan session lasted about 50 min. Foam padding was used to minimize participants' head movement. All images were acquired along axial planes parallel to the anterior commissure-posterior commissure (AC–PC) line. Two runs of T2<sup>∗</sup> -weighted images for fMRI were acquired during the task with a Multi-band accelerated EPI pulse sequence<sup>1</sup> with the following parameters: 60 axial slices of 2.4 mm thick, interleaved, skip = 0 mm, TR = 1000 ms, TE = 35 ms, multi-band accel. factor = 6, echo spacing = 0.72 ms, flip angle = 77◦ , FOV = 228 mm, matrix size = 96 × 96, voxel size = 2.4 mm × 2.4 mm × 2.4 mm. Each run began with a single-band reference image that matched real brain-volumes and acquired without acceleration, followed by 540 volumes covering the task period. A pair of spin-echo echo-planar imaging (SE-EPI) reverse-phase encode field maps were acquired prior to these two runs, with TR = 8600 ms and TE = 65 ms. A high-resolution T1-weighted anatomical volume of the whole brain was acquired after the task with a magnetization-prepared rapid gradient-echo (MPRAGE) sequence with the following parameters: 176 axial slices of 1.0 mm thick, skip = 0 mm, TR = 2400 ms, TE = 1.94 ms, flip angle = 8◦ , FOV = 256 mm, matrix size = 256 × 256, voxel size = 1.0 mm × 1.0 mm × 1.0 mm.

### Procedure

The task was compiled and executed via E-prime 2.0 software (Psychology Software Tools, Pittsburgh, PA, United States). Stimuli were projected onto a screen placed at the back of the magnet bore and viewed with mirrors mounted on the head coil. Prior to scanning, the task was explained to the adolescents via written and verbal instructions, and then they performed an 18-trial practice session on a PC. A mock scanner with an identical stimulus presentation and response system was shown to each participant to help them acclimate to the MRI environment. MRI-compatible lenses were provided to adolescents who required vision correction. Their responses were collected using a fiber-optic button system with a five-button response glove (BrainLogic, Psychology Software Tools) placed on their dominant hand. Participants were required to make responses by pressing the button under their index finger. At the end of the experiment, participants were debriefed.

### Imaging Preprocessing

Image preprocessing was performed for each participant using the statistical parametric mapping package (SPM 12; Wellcome Trust Center for Neuroimaging, London, United Kingdom; RRID:SCR\_007037) and FMRIB Software Library (FSL v6.0<sup>2</sup> ; RRID:SCR\_002823). The T1 image and all EPI images were manually adjusted to align the AC-PC plane. Bias correction was performed for both T1 and EPI images. Each EPI image volume was then realigned to the first volume and six motion parameters were estimated. Fieldmap in Hz and magnitude images were generated based on the field map images to calculate the voxel displacement map (VDM). The VDM was applied to all EPI images to correct distortions. Head motion and signal drifting were further corrected using the ArtRepair software version 5b (RRID:SCR\_005990)<sup>3</sup> . A mean EPI image was calculated across all EPI images after these steps of processing. The brain was extracted from the bias-corrected T1 image and coregistered to the brain extracted from the mean EPI image using normalized mutual information. The coregistered T1 brain was normalized to a bias-corrected 12-years adolescent T1 template (Richards et al., 2016), with affine regularization as ICBM space template – European brains, and resampled to a voxel size of 2 × 2 × 2 mm. Normalized EPI images were then spatially smoothed with a Gaussian kernel of 6 mm full-width half-maximum, as recommended by Sacchet and Knutson (2013) to accurately locate the VST.

### General Linear Modeling (GLM)

The GLM was performed using SPM 12. First-level (singlesubject level) statistical analyses of event-related blood oxygenation level-dependent (BOLD) signals were conducted using GLM for each participant. For each run, three regressors were constructed based on the onset vectors of the anticipation phase in three conditions (i.e., Reward, Neutral, and Punishment), with the duration of each event modeled as the total duration of the anticipation phase in the corresponding trial. Six regressors were constructed based on the onset vectors of the outcome phase in six feedback conditions (i.e., Reward-Hit, Reward-Miss, Neutral-Hit, Neutral-Miss, Punishment-Hit, and Punishment-Miss), with the event duration modeled as

<sup>1</sup>https://www.cmrr.umn.edu/multiband/

<sup>2</sup>https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/

<sup>3</sup>https://cibsr.stanford.edu/tools/human-brain-project/artrepair-software.html

0 s. For each of these six regressors, a parametric modulator of the trial-by-trial cumulative rewards earned (demeaned) was constructed to model the influence of this information on brain responses under each feedback condition. Two additional regressors were constructed based on the onset of the targets with hit and missed responses respectively, with the event duration modeled as 0 s. All of these 11 regressors were convolved with a standard hemodynamic response function (HRF). Head motion was modeled as nuisance regressors according to Friston 24-parameter model (Friston et al., 1996), including 6 head motion parameters estimated during realignment, 6 parameters as one time-point before, and 12 corresponding squared items. Low-frequency drifts in signal were removed using a high-pass filter with a 128 s cutoff. Across two runs, one nuisance regressor to indicate runs was entered into the model. The serial correlation was estimated using an autoregressive AR(1) model. This model was estimated for each participant and the images of parameter estimates (β) values were obtained for each regressor. The β images of the three anticipationassociated and six outcome-associated regressors were used in the following analyses.

For the anticipation phase, two contrasts were defined: (1) Reward cue vs. Neutral cue, to examine the involvement in gain-related anticipation, and (2) Punishment cue vs. Neutral cue, to examine the involvement in loss-related anticipation. For the outcome phase, three contrasts were defined: (1) Reward (Hit minus Miss) vs. Neutral (Hit minus Miss), to examine the involvement in reward receipt (monetary gain), and (2) Punishment (Miss minus Hit) vs. Neutral (Miss minus Hit), to examine the involvement in punishment receipt (monetary loss). For each of these four effects, a contrast image was generated for each participant.

### Examining the Neural Responses Across the Entire Sample: Whole-Brain Analyses

A second-level group GLM was conducted to examine the neural responses associated with each of these four effects across the entire sample, regardless of the individual differences in externalizing behavior and CU traits. Both positive (increase) and negative (decrease) activation associated with each effect were examined. Age, sex, IQ, social adversity, and pubertal status were entered as covariates in the group-level GLM. In the grouplevel analysis, we used a cluster-extent thresholding approach to correct for multiple voxel comparisons. Specifically, a threshold consisted of a significance level of p < 0.001 (uncorrected) for the height of each voxel (as recommended by Woo et al., 2014), together with a contiguous-voxel extent threshold (k; estimated based on the random field theory; Worsley et al., 1992) was adopted, which resulted in a cluster-level p < 0.05 threshold.

### Extracting Neural Responses From Regions of Interest (ROI)

We selected two prior defined ROI based on previous metaanalyses for fMRI studies using the MID: ventral striatum (VST), which is associated with reward processing, and amygdala, which is associated with the processing of negative emotion (Knutson and Greer, 2008; Liu et al., 2011; Richards et al., 2013; Silverman et al., 2015; Oldham et al., 2018). These two were defined anatomically. Specifically, the VST was defined according to the Oxford-GSK-Imanova structural striatal atlases (Tziortzi et al., 2011), and the amygdala was defined according to the Harvard-Oxford subcortical structural atlases. We manually traced these two ROIs on the age-specific (12-years-old) anatomical template. Signals in each ROI were defined as the first eigenvariate of the β value from all voxels within the combined cluster of that region in the left and right hemispheres. The ROI signals were extracted from each participant's first-level contrast map for each of the five effects. The statistical analyses for neural responses in ROIs (see below) are independent of the above group-level GLM analyses across the entire sample. The whole-brain exploratory regression analysis was not conducted because our sample was highly skewed on both externalizing and CU measures and the assumption of normality of variable distribution was not met.

### Modeling the Influence of Externalizing and CU Traits on Neural Responses: Statistical Analyses Primary Measures

The gender difference in both primary measures (externalizing behavior and CU traits) and measures of no interest (age, pubertal status, IQ, and social adversity) were examined using an independent sample t-test. Bootstrapped correlations between all measures were also examined. Due to the high skewness of both primary measures, the significance level of each test were estimated using bootstrapping, a non-parametric approach to estimate the population distribution of a statistical value based on the sample distribution. Specifically, a large amount of bootstrapping samples were randomly drawn from the current sample with replacement. The distribution of a statistical value computed based on each bootstrapping sample reflect the population distribution of this value (Wright et al., 2011). The bootstrapping procedure makes fewer assumptions compared to the traditional parametric approaches, and is therefore appropriate for studies with small sample size or with non-normally distributed variables. Here, for each test, 1000 bootstrapping samples were drawn from our sample, and the bias corrected and accelerated 95% confidence intervals (BCa 95% CI) of the statistical values were estimated, which adjusted for bias and skewness in the bootstrapped distribution.

### Neural Responses

Due to the high skewness of both externalizing and CU scores in our sample, bootstrapping regression (Paparoditis and Politis, 2005) was conducted for the subsequent analyses (number of bootstrapping samples = 1,000, BCa 95% CI was estimated). We examined the direct effect of externalizing behavior and CU traits on ROI activation using a regression model with ROI activation associated with each of the contrast effects defined above as the dependent variable. Age, child sex, IQ, social adversity, and pubertal status were entered as covariates in Step 1, and the primary variable (i.e., externalizing behavior or CU traits) was

entered in Step 2 (Model 1). The amount of variance explained by the regressor of interest was estimated as the difference of R 2 between the model and the comparable model in which only all regressors of no interest were included.

If both externalizing and CU scores were significantly associated with ROI activation, we then examined the effect of each primary variable of interest and brain activation when controlling for the non-focal variable by entering it as a regressor of no interest (Model 2). We also compared the R <sup>2</sup> of this model to Model 1 (1R <sup>2</sup> = R 2 model <sup>2</sup> − R 2 model <sup>1</sup>). A reduction of R 2 (negative 1R 2 ) indicates a joint contribution of these two variables, while an augment of R 2 (positive 1R 2 ) reflects an antagonistic effect.

The joint contribution of externalizing behavior and CU traits on the regional activation was examined using a bootstrapping regression model (Model 3) with covariates of no interest (Step 1), both externalizing behavior and CU traits (Step 2) and their interaction as the predictor of interest (Step 3). The interaction term was computed as the product of these two variables (demeaned). A significant positive coefficient of the interaction term would indicate a superadditive effect of the two (i.e., individuals with higher levels in both CU traits and externalizing behavior are hyper-responsive). A significant negative coefficient of the interaction term would indicate a subadditive effect (i.e., individuals with lower levels of CU traits but higher levels of externalizing behavior are hyper-responsive). Compared to the regression model with only externalizing behavior and CU traits as regressors, an augment of R 2 (positive 1R 2 ) of the model with an additional interaction term reflects an incremental contribution of this interaction term on predicting brain activation.

### RESULTS

### Descriptive Statistics of the Measures

The group-mean externalizing and CU scores together with other measures and sex differences are reported in **Table 1**. Boys and girls did not significantly differ on either externalizing (p = 0.430) or CU (p = 0.559) scores. Therefore, sex difference was not examined in the subsequent analyses. Boys and girls were also not significantly different on any of the measures of no interest except for pubertal status (p = 0.009, with girls showing a higher level of pubertal status than boys).

Externalizing and CU scores were significantly correlated (r = 0.54, BCa 95% CI: 0.264 to 0.778; p = 0.039), see **Figure 1C**, but this correlation became non-significant when controlling all covariates (r = 0.58, BCa 95% CI: 0.154 to 0.826; p = 0.055). None of the covariates were significantly correlated with externalizing or CU measures except puberty, which was positively correlated with the CU score (r = 0.40, BCa 95% CI: 0.086 to 0.694; p = 0.031).

### Results of GLM Analysis Regardless of Individual Difference

A significant Reward cue > Neutral cue effect was shown in the VST bilaterally, while no region showed a significant Reward cue < Neutral cue effect (**Figure 3**). No significant difference between Punishment cue and Neutral cue was found in any brain region. In addition, a significant Reward (Miss minus Hit) < Neutral (Miss minus Hit) effect was shown in left inferior frontal gyrus, while no region showed a significant Reward (Miss minus Hit) > Neutral (Miss minus Hit) effect. A significant Punishment (Miss minus Hit) < Neutral (Miss minus Hit) effect was shown in hippocampus/parahippocampal gyrus bilaterally, while no region showed a significant Punishment (Miss minus Hit) > Neutral (Miss minus Hit) effect. Coordinates of the peak of each cluster showing significant activation are listed in **Table 2**.

### Results of ROI Analysis

### Reward-Related Anticipation

Associations between each primary variable of interest (i.e., externalizing behavior and CU traits) and activation in each ROI during reward anticipation are illustrated in **Figure 4A**. Activation in the VST was significantly positively associated with CU traits (β = 0.053, BCa 95% CI: 0.008 to 0.095; p = 0.046), but no significant association was found for externalizing behavior (β = 0.065, BCa 95% CI: −0.023 to 0.159; p = 0.071). When controlling for externalizing score, the CU-VST activation relationship became non-significant (β = 0.042, BCa 95% CI: −0.018 to 0.111; p = 0.166, 1R <sup>2</sup> = −0.17). Similarly, when controlling for CU, the externalizing–VST relationship was nonsignificant (β = 0.033, BCa 95% CI: −0.005 to 0.074; p = 0.188,


CU; callous-unemotional traits.

1R <sup>2</sup> = −0.17). These findings suggest that there was a substantial overlap between the contributions of CU and externalizing to VST activation. Externalizing behavior and CU traits explained 23.3 and 22.8% of the variance of the activation in the VST, respectively. The externalizing by CU interaction term in the moderation model was not significant (β = 0.08, BCa 95% CI: −0.008 to 0.020; p = 0.186).

Activation in the amygdala was not significantly associated with either externalizing behavior (β = −0.007, BCa 95% CI: −0.041 to 0.068; p = 0.822) or CU traits (β = 0.014, BCa 95% CI: −0.017 to 0.045; p = 0.366). The externalizing behavior by CU interaction term in the moderation model was not significant (β = 0.004, BCa 95% CI: −0.006 to 0.020; p = 0.358).

### Punishment-Related Anticipation

Associations between externalizing behavior/CU traits and activation in each ROI during punishment anticipation are illustrated in **Figure 4B**. Activation in the VST was not associated with either externalizing behavior (β = −0.036, BCa 95% CI: −0.103 to 0.076; p = 0.118) or CU traits (β = 0.018, BCa 95% CI: −0.035 to 0.084; p = 0.523). Externalizing behavior and CU traits explained 4.6 and 1.7% of the variance of the activation in the VST, respectively. The externalizing behavior by CU interaction effect in the moderation model was not significant (β = −0.001, BCa 95% CI: −0.016 to 0.022; p = 0.858).

Activation in the amygdala was not significantly associated with either externalizing behavior (β = −0.045, BCa 95% CI: −0.090 to 0.060; p = 0.262) or CU traits (β = 0.015, BCa 95% CI: −0.045 to 0.074; p = 0.616). Externalizing behavior and CU traits explained 6.9 and 1.1% of the variance of the activation in the amygdala, respectively. The externalizing behavior by CU interaction effect in the moderation model was not significant (β = −0.003, BCa 95% CI: −0.017 to 0.033; p = 0.677).

### Reward Receipt (Monetary Gain) Related Responses

Associations between externalizing behavior/CU and activation in each ROI during the receipt of reward (monetary gain) are illustrated in **Figure 4C**. Activation in the VST was not significantly associated with either externalizing behavior (β = 0.266, BCa 95% CI: −0.173 to 0.502; p = 0.058) or CU traits (β = 0.202, BCa 95% CI: −0.066 to 0.507; p = 0.119). Externalizing behavior and CU traits explained 17.0 and 14.4% of the variance of the activation in the VST, respectively. The externalizing behavior by CU interaction effect in the moderation model was not significant (β = −0.006, BCa 95% CI: −0.053 to 0.030; p = 0.785).

Activation in the amygdala was not significantly associated with either externalizing behavior (β = 0.142, BCa 95% CI: −0.143 to 0.241; p = 0.137) or CU traits (β = −0.019, BCa 95% CI: −0.183 to 0.144; p = 0.808). Externalizing behavior and CU traits explained 8.6 and 0.2% of the variance of the activation in the amygdala, respectively. The externalizing behavior by CU interaction effect in the moderation model was not significant (β = 0.004, BCa 95% CI: −0.046 to 0.033; p = 0.820).

### Punishment Receipt (Monetary Loss) Related Responses

Associations between externalizing behavior/CU and activation in each ROI during the receipt of punishment (monetary loss) are illustrated in **Figure 4D**. Activation in the VST was not significantly associated with either externalizing behavior (β = −0.101, BCa 95% CI: −0.289 to 0.086; p = 0.275) or CU traits (β = −0.011, BCa 95% CI: −0.170 to 0.147; p = 0.883). Externalizing behavior and CU traits explained 4.9 and 0.1% of the variance of the activation in the VST, respectively. The externalizing behavior by CU interaction effect in the moderation model was not significant (β = 0.025, BCa 95% CI: −0.030 to 0.089; p = 0.095).

Activation in the amygdala was significantly associated with externalizing behavior (β = −0.155, BCa 95% CI: −0.305 to −0.006; p = 0.042), and remained significant after controlling for CU (β = −0.226, BCa 95% CI: −0.375 to −0.091; p = 0.004, 1R <sup>2</sup> = 0.06). CU traits were not significantly associated with activation in the amygdala (β = −0.008, BCa 95% CI: −0.167 to 184; p = 0.903). Externalizing behavior and CU traits explained 14.6 and 0.1% of the variance of the activation in the amygdala, respectively. The externalizing behavior by CU interaction effect in the moderation model was not significant (β = 0.007, BCa 95% CI: −0.021 to 0.044; p = 0.583).

### DISCUSSION

The present study examined the neural mechanisms of reward and punishment processing in relation to externalizing behavior and CU traits (individually and jointly) in a community sample of adolescents. Findings provide partial support for the theories of hyper-responsivity to reward and hypo-responsivity to punishment in CU and externalizing youths.

Partially consistent with our hypothesis of hyper-responsivity to reward in youths with high externalizing behavior and/or CU traits, CU traits were positively correlated with VST activation during reward anticipation. Similar positive relationship with VST activation was also found for externalizing behavior, although it was non-significant (p = 0.071), likely due to the small sample size. We also demonstrated that externalizing behavior and CU traits share substantial overlap in predicting

TABLE 2 | Brain regions showing significant activation changes in the GLM analysis.


Regions are listed in a descending order based on their peak Z value. For a cluster with multiple local peaks, the number of voxels in the whole cluster (K) was only listed under the first local peak (also for other activation tables). The threshold was p < 0.001 for the height and cluster level p < 0.05. L, left; R, right. BA, Brodmann area.

reward anticipation (Reward cue minus Neutral cue); (B) punishment anticipation (Punishment cue minus Neutral cue); (C) reward (monetary gain) [Reward (Hit minus Miss) minus Neutral (Hit minus Miss)], and (D) punishment (monetary loss) [Punishment (Miss minus Hit) minus Neutral (Miss minus Hit)]. Solid lines: significant association. Dashed lines: non-significant association.

activation of the VST during reward anticipation, given that once the effect of the non-focal variable was taken into account, the associations became non-significant. Abnormal anticipatory responses in the VST have been suggested as a biomarker for various high-risk (e.g., impulsive, addicted) populations (see Balodis and Potenza, 2015 for a review), particularly in the context of reward anticipation. In contrast, some psychiatric (e.g., schizophrenia, depression) conditions have been associated with blunted activation in the VST (Knutson and Heinz, 2015). Taken together, hyperactivity in the VST may be potentially used to predict future problematic and risky behaviors that are reward dominant, including externalizing behavior, CU traits, substance abuse, and gambling (Blaszczynski et al., 1997; King et al., 2004).

Partly in line with our hypothesis of hypo-responsivity to punishment, we found that externalizing scores were negatively associated with the amygdala responses to monetary loss during punishment receipt (with or without controlling CU traits). Specifically, adolescents with fewer externalizing behaviors showed stronger amygdala activation when the outcome of missing the target was accompanied by a punishment as monetary loss (Miss minus Hit after the punishment cue), compared to the same outcome but with no punishment (Miss minus Hit after the neutral cue). However, for individuals with higher externalizing scores, they showed weaker amygdala activation when missing the target was accompanied by a punishment (see **Figure 4B**), suggesting that the neural processing of punishment may be suppressed in adolescents with more externalizing behaviors. These results are similar to previous research study on DBD youth (e.g., Finger et al., 2011) that found amygdala deactivation in response to punishment receipt.

In addition, no significant interaction effect between CU and externalizing was found for any of the ROIs when anticipating or receiving rewards or punishment, failing to support our hypothesis that individuals with higher levels of both externalizing behavior and CU traits would show the most salient abnormalities in brain activation. It is worth noting that our study is the first to examine the interaction effects of CU and externalizing behavior on reward/punishment anticipation and processing. Prior functional brain imaging studies that examined additive effects of CU and externalizing measures (e.g., Herpertz et al., 2008; Passamonti et al., 2010) had focused on neural processing of emotional stimuli. In addition, interactive effects of CU and externalizing measures seem to be more prominent in the amygdala activation during fear-related processing in particular (Viding et al., 2012; Cardinale et al., 2018). Finally, our null findings may also be due to low statistical power, and future studies with larger sample size should be conducted to detect if any interaction effects may exist using reward and punishmentrelated paradigms.

Alternatively, the lack of effect may be partly due to the nature of the task (e.g., more anticipation events relative to feedback), which ultimately makes it less powerful for detecting effects for reward receipt. Relatedly, the MID task has been extensively used to assess reward processing, but less for punishment processing (Oldham et al., 2018). Potentially, this task may be less sensitive to punishment anticipation than to reward anticipation. In fact, studies utilizing the MID have more often linked personality traits to VST activation to reward than to punishment (e.g., Buckholtz et al., 2010; Bjork et al., 2012).

The current research also has caveats to consider. First, the sample size is small, limiting statistical power. Moreover, the externalizing and CU scores were highly skewed in our sample. Two participants had very high scores on these measures, with one having an externalizing score (27) greater than 2 SD (but less than 3 SD) of the group mean, and the other having a CU score (41) greater than 2 SD (but less than 3 SD) of the group mean. Normative samples of adolescents aged 12–14 years have reported CBCL externalizing behavior means of 7.01 and 5.38 for boys and girls, respectively (Bongers et al., 2003). Means for CU traits (via the ICU) range from 23.45 to 31.05 in community adolescent samples (Roose et al., 2010; Feilhauer et al., 2012). Still, we caution that our results may be less generalizable to individuals with clinical diagnosis of DBD because our range of externalizing scores (expect the one with very high score) is rather limited (e.g., 0–16), whereas prior reports of externalizing behavior for adolescent in clinical samples range from 18.92 to 27.2 (Bögels et al., 2008; Bjork et al., 2010; Schaeffer et al., 2014). Similarly, ranges for CU traits in our sample (after excluding the outliers) were diverse (13–38, mean = 23.8), while detainee and offender samples report total score means ranging from 23 to 41 (Kimonis et al., 2008, 2013; White et al., 2013). Next, although we did incorporate both sexes in our study, boys and girls did not significantly differ on externalizing behavior or CU traits in our sample. This may be due to low statistical power resulted from a small sample size. Sex differences have been found in the gray matter volumes of the OFC (Raine et al., 2011) and volumetric brain asymmetries of the OFC and ACC (Visser et al., 2013) in antisocial populations. Future work with larger sex-mixed samples is needed to determine the effects of sex, especially since there are large gender gaps for prevalence rates of externalizing disorders (Newman et al., 1996; Hicks et al., 2007). Finally, we only focused on the VST and amygdala as our ROIs because of their well-documented involvement in the reward and punishment processing (Bjork et al., 2012; Cohn et al., 2015; Oldham et al., 2018). Although the involvement of other regions including the vmPFC and insula has been reported (Knutson et al., 2001b; Balodis et al., 2012; Silverman et al., 2015; Oldham et al., 2018), our exploratory whole-brain regression analysis did not reveal any significant effects for these areas in our sample.

One strength of our work is that we recruited from an ethnically diverse and mixed-sex healthy community sample. Previous studies have primarily been on male and clinical populations (e.g., Finger et al., 2011; Bjork et al., 2012; Pujara et al., 2013; Cohn et al., 2015; Byrd et al., 2018). Only recently has the focus shifted to the inclusion of community samples with both sexes and younger age groups, which is what we were able to accomplish in this study. Taken together, our results suggest that both externalizing behavior and CU traits are associated with hyper-responsivity to reward, and that externalizing behavior in particular is associated with hyporesponsivity to punishment. As both externalizing behavior and CU traits in youth are known risk factors for criminal offending in adulthood (Hinshaw and Lee, 2003; Frick and White, 2008), it may be beneficial for future work to evaluate the degree to which externalizing problems and CU traits are independently and jointly associated with neural activity. This knowledge will help us better understand the etiological basis of externalizing problems and CU traits and eventually contribute to interventions for these unwanted trajectories.

### DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

### ETHICS STATEMENT

fnins-13-01319 December 11, 2019 Time: 17:45 # 12

The studies involving human participants were reviewed and approved by the Institutional Review Board of Brooklyn College (CUNY). Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.

### AUTHOR CONTRIBUTIONS

YH and YG conceived and planned the research study. YH, SF, and SL carried out the study. YH and TW preprocessed the data. YL, ZW, and XL contributed to the imaging preprocessing pipeline. TW performed the statistical analyses. YH, TW, and YG wrote the manuscript with input from all other authors. All authors provided critical feedback to all components of the research.

### REFERENCES


### FUNDING

Research reported in this publication was supported by the National Institute of General Medical Sciences of the National Institutes of Health under Award Numbers SC3GM118233 and SC1GM127243 to YG. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

### ACKNOWLEDGMENTS

The authors would like to thank the families for their participation, and the staff at Brooklyn College Psychophysiology Laboratory and the Translational and Molecular Imaging Institute at Mount Sinai Icahn School of Medicine for their assistance.



callous-unemotional traits. Am. J. Psychiatry 166, 95–102. doi: 10.1176/appi. ajp.2008.07071050



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Huang, Wu, Gao, Luo, Wu, Fagan, Leung and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Attention Capture of Non-target Emotional Faces: An Evidence From Reward Learning

*Xing Zhou1 , Bixuan Du1 , Zhiqing Wei2 and Weiqi He1 \**

*1 Research Center of Brain and Cognitive Neuroscience, Liaoning Normal University, Dalian, China, 2 School of Psychology, Northwest Normal University, Lanzhou, China*

The aim of this study was to explore whether reward learning would affect the processing of targets when an emotional stimulus was task irrelevant. In the current study, using a visual search paradigm to establish an association between emotional faces and reward, an emotional face appeared as a task-irrelevant distractor during the test after reward learning, and participants were asked to judge the orientation of a line on the face. In experiment 1, no significant difference was found between the high reward-fear distractor condition and the no reward-neutral condition, but the response times of the high rewardfear condition were significantly longer than those of the low reward-happy condition. In experiment 2, there was no significant difference in participants' performance between high reward-happy and no reward-neutral responses. In addition, response times of the low reward-fear condition wear significantly longer than those of the high reward-happy and no reward-neutral conditions. The results show that reward learning affects attention bias of task-irrelevant emotional faces even when reward is absent. Moreover, the high reward selection history is more effective in weakening the emotional advantage of the processing advantage than the low reward.

### *Edited by:*

*Ya Zheng, Dalian Medical University, China*

### *Reviewed by:*

*Xu Qiang, Ningbo University, China Ming Zhang, Chinese Academy of Sciences, China*

> *\*Correspondence: Weiqi He weiqi79920686@sina.com*

### *Specialty section:*

*This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Psychology*

*Received: 20 February 2019 Accepted: 18 December 2019 Published: 31 January 2020*

### *Citation:*

*Zhou X, Du B, Wei Z and He W (2020) Attention Capture of Non-target Emotional Faces: An Evidence From Reward Learning. Front. Psychol. 10:3004. doi: 10.3389/fpsyg.2019.03004*

Keywords: reward learning, emotional face, task irrelevant, attentional capture, distractor

### INTRODUCTION

Attention allocation is affected by reward through modulation of visual salience and behavioral motivation (Boehler et al., 2012; Botvinick and Braver, 2015). In recent years, some researchers have posited that after reward training, even a stimulus as the non-target can still automatically capture one's attention and thus receive priority processing (Anderson et al., 2011a,b; Awh et al., 2012).

As is known, emotional stimuli always receive priority attention as compared to non-emotional stimuli (Batty and Taylor, 2003; Vimal, 2008; Hodsoll et al., 2011; Barratt and Bundesen, 2012; Ikeda et al., 2013; Schmidt et al., 2015; Pool et al., 2016; Glickman and Lamy, 2017), with the exception of instances with a high percepetion load (Yates et al., 2010; Gupta and Srinivasan, 2015). In addition, early attention bias for negative emotion (such as fearful faces), in which people can quickly detect negative and threatening stimuli, has returned relatively consistent empirical results (Hansen and Hansen, 1988; Luo et al., 2010; Pinkham et al., 2010). This attention preferential processing facilitates us to respond quickly and appropriately to negative stimuli.

When emotional pictures are presented as task irrelevant, the presence of rewards can modulate attention resources and weaken the interference of emotional distractors on target processing (Padmala and Pessoa, 2014; Yokoyama et al., 2015; Walsh et al., 2018). Kulke et al. (2018) asked participants to complete a simple perceptual task while ignoring emotional images. One group was consistently rewarded for completing tasks quickly and accurately, whereas the other group was not rewarded for their performance. The results showed that the presence of rewards could alleviate the disruptive effect of emotions on the processing target (Padmala and Pessoa, 2014; Yokoyama et al., 2015; Padmala et al., 2017). It could be that the presence of rewards enables participants to alter their coping strategies from passive into active control to cope with the changes of the scene and enhance their cognitive control (Botvinick and Braver, 2015).

The association of rewards acquired from past experience can have an important impact on the attention bias. The researchers associated high values with happy faces and low values with angry faces through a reward learning phase, to see whether the reward-stimuli association would affect the processing advantage of threatening faces. The study showed that the preferential processing of anger can be modified by reward learning rather than the impact of endogenous attention during the test (Yao et al., 2013). Reward-modulation effects learned through value association impair early visual perception and hence attention allocation to angry faces. In the later stages of emotional processing, participants employed more cognitive resources to process reward history (Chen and Wei, 2019).

We note that in studies of reward-emotional attention processing, emotional faces appeared as targets during training after the reward is learned, whereas previous studies showed that the combination of reward and goal facilitates target processing (Fan et al., 2014). However, it is still unknown whether reward has an effect on the attention capture of non-target emotional faces when the reward information is absent. In addition, in some studies, reward information (reward cues or feedback) and stimuli were presented at the same trial (Bijleveld et al., 2010; Hickey et al., 2011; Wei et al., 2014, 2015), which inevitably activated reward expectation or reaction motivation of individuals. Unlike the cue paradigm or feedback paradigm, reward learning has a strong shaping effect on individual behavior and mental processing (Libera and Chelazzi, 2014), which can result in the avoidance of the influence of reward expectation and motivation on attention processing (Hammerschmidt et al., 2018). Therefore, our study established an association between emotional faces and rewards through reward learning, and explored whether the reward learning would affect the processing of targets when the emotional stimuli appeared as task-irrelevant distractors and also when reward information is absent during the testing phase. The study consists of two experiments. Experiment 1 was designed to investigate the processing characteristics of low reward-happy and high reward-fearful faces during the test phase after establishing a learning association. Considering the different reward values, their association had different effects on attention selection (Anderson et al., 2011a). Experiment 2 was aimed at investigating the attention processing of high reward-happy and low reward-fear conditions during test phase after reward associations were established. Based on previous studies regarding the relationship between reward and emotional processing, we hypothesized that, after the reward learning, the interference effect of the emotional face would be alleviated in the testing phase when it was presented to be task irrelevant, especially in cases involving a fearful face.

### EXPERIMENT 1

### Methods

### Participants

Twenty-two students (14 female, 8 male; mean age, 20.36 years; age range, 18–24 years) from Liaoning Normal University participated in the experiment. They were all right handed and had either normal or corrected-to-normal vision. The research protocol was approved by the Research Center of Brain and Cognitive Neuroscience, Liaoning Normal University Institutional Review Board, and informed consent forms were signed by all participants. Our rationale for sample size was based on previous studies (e.g., Jahfari and Theeuwes, 2016) and obtained in G-power by setting the partial *η*<sup>2</sup> as 0.25, *α* as 0.05, and power (1 − *β*) as 0.8.

### Stimuli

Pictures depicting emotions were chosen from the China Facial Affective Picture System (Huang et al., 2011). Three types of emotion faces were used: happy (*N* = 6; male 3, female 3); fearful (*N* = 6; male 3, female 3); and neutral (*N* = 6; male 3, female 3). To ensure the consistency of material stimuli, we matched the arousal and the facial attractiveness. Twelve participants assessed the valence, arousal, and attractiveness of the emotional faces. There were significant differences in the valence of the three emotions [*F*(2,11) = 141.36, *p* < 0.001, hp <sup>2</sup> = 0.357; happy (2.89 ± 0.45), neutral (4.52 ± 0.32), and fearful (6.32 ± 0.62)]. Happy and fearful faces used here were matched on their arousal [mean (*M*) ± SD, happy (5.33 ± 0.21), fear (5.82 ± 0.35), *t*12 = 2.97, *p* = 0.94], the arousal of neutral faces was 3.72 ± 0.32. There was no significant difference in facial attractiveness [*F*(2,11) = 128.68, *p* = 0.69, hp <sup>2</sup> = 0.025; happy (4.23 ± 0.35), neutral (4.12 ± 0.32), fearful (4.08 ± 0.44)]. Luminance was controlled for in the emotional faces using a unified template in Photoshop CS6, and we added line segments in different directions (vertical, horizontal, or oblique) between the eyebrows.

### Procedure

In an electromagnetic-shielded room, the participants were seated comfortably approximately 80 cm away from a 17-inch cathode-ray tube screen display. They performed a visual search task adapted from Anderson et al. (2011a,b). The time course for the reward learning and testing phases is shown in **Figures 1A,B**. During both the training and test phases, the fixation display included a black fixation cross (0.5° × 0.5° visual angle) presented in the center of the display against a gray background, and the search display consisted of the fixation cross surrounded by six emotional faces (3.58° × 3.58° visual angle of each face). The diameter of the emotional stimulation was 10.7°.

### *Training Phase*

In the training phase, a baseline block of 120 trials was given to the participants. During the baseline, emotional faces (happy or fearful) were presented as background (there were six faces, one of which was happy or fearful), and one of the neutral faces had a horizontal or vertical line between the eyebrows. The participants were asked to ignore the face and judge the line orientation. The rewarding phase had six learning blocks (540 trials). Six emotional faces (one was a fearful or happy face; the other are neutral faces) were presented in the search display. Furthermore, the target for each trial was a unique emotional face with either a vertical or horizontal black line between the eyebrows (**Figure 1C**). The participants were required to press the F or J key as quickly as possible when judging whether the line orientation between the eyebrows was horizontal or vertical. After a correct reaction by the participants, the corresponding reward feedback and total score appeared on the screen.

Fearful faces (high reward) were followed by "+100 points" feedback at 80% percent, and the remainder was "+20 points." For happy faces (low reward), the percentages were reversed. It should be pointed out that 500 points were the equivalent of 1 Chinese Yuan Renminbi. Participants were clearly informed that their additional monetary reward was determined by the total points they earned.

### *Test Phase*

After finishing the training phase and resting for an hour, all participants completed a second visual search task in which they identified the orientation of a line between the eyebrows. During the test phase, the search display consisted of a fearful face or a happy face among neutral faces, and the target was

defined as the horizontal or vertical line between the eyebrows (**Figure 1C**). The participants were told that the experimental trials would not contain a reward. For the test phase, each trial began with a fixed display (400–600 ms). Next, a search display lasted 1,500 ms until the participant responded. An interval appeared after the response. The test contained no reward feedback, only a blank screen lasting 1,000 ms. The test contained one practice block (12 trials) and six formal blocks (each 160 trials). In the test phase, to ensure that the participants could observe the attention capture effect, 50% of the trials were no-reward distractor stimuli. In the remaining 50% of the trials, 25% were high-reward disturbance stimuli, and the remaining 25% were low-reward distractors.

### Results

### Training Results

During the training, we found that the mean accuracy (ACC) for high reward-fear [*t*(19) = −5.03, *p* < 0.001, Cohen's *d* = 0.39] and low reward-happy [*t*(19) = −8.19, *p* < 0.001, Cohen's *d* = 1.72] conditions increased as compared to baseline (baseline: fear 89.31 ± 7.89, happy 86.89 ± 6.39, neutral 90.79 ± 5.93; reward training: high reward-fear 97.25 ± 2.81, low reward-happy 97.49 ± 2.63, neutral 95.98 ± 3.93). A repeated-measures ANOVA showed that ACC differed significantly among these three conditions [*F*(2,38) = 4.04, *p* = 0.025, hp <sup>2</sup> = 0.16]. There was no significant difference between high reward-fear and low reward-happy conditions [*t*(19) = −0.79, *p* = 0.44, Cohen's *d* = 0.03]. In addition, the difference between low reward-happy and neutral did not reach significance [*t*(19) = 1.96, *p* = 0.06, Cohen's *d* = 0.03]. However, there was a significant difference between low reward-happy and neutral [*t*(19) = 2.21, *p* = 0.04, Cohen's *d* = 0.31]. The results of the response times (RTs) between reward learning and baseline were not significantly different (baseline: fear 1257.46 ± 78.71 ms, happy 1251.40 ± 88.05 ms, neutral 1266.93 ± 80.73 ms; reward training: high reward-fear, 1262.20 ± 133.1 ms, low reward-happy 1254.57 ± 113.41 ms, neutral 1304.96 ± 105 ms). The training phase data are shown in **Figure 2**.

### Test Results

Next, we examined how reward experiences affected the face section of the test. A repeated-measures ANOVA on accuracy of reward distractor types was used (high reward-fear 94.07 ± 3.23, low reward-happy 94.89 ± 2.34, neutral 94.17 ± 3.52). There was no main effect on distractor [*F*(2,38) = 2.27, *p* = 0.11, hp <sup>2</sup> = 0.09]. A repeated-measures ANOVA showed that RTs differed significantly among distractor conditions (high rewardfear 1317.49 ± 87.53 ms, low reward-happy 1256.01 ± 65.83 ms, no distractors 1304.58 ± 77.51 ms) [*F*(2,38) = 10.44, *p* < 0.001, hp <sup>2</sup> = 0.54]. A high reward-fear distractor slowed RTs relative to low-happy [*t*(19) = 3.77, *p* = 0.001, Cohen's *d* = 0.78]. The test phase data are shown in **Figure 3**.

### EXPERIMENT 2

Experiment 1 provided evidence that emotional faces interfered with target processing after reward learning. However, that part of the experiment trained only the low-happy and the high-fear emotions, and not the high-happy and low-fear emotions. In addition, high rewards can give individuals a positive experience, but we wanted to discover whether the combination of high reward-positive emotion generates a stronger attention bias than the negative emotion-low reward. Therefore, in experiment 2, we used the happy-high and fear-low reward combinations in the reward learning phase. The sample size obtained was the same as experiment 1.

### Methods

### Participants

Twenty students (12 female 8 male; mean age, 21.8 years; age range, 18–25 years) from Liaoning Normal University participated in this experiment. They were all right handed and had either normal or corrected-to-normal vision. The research protocol was approved by the Brain and Cognitive Neuroscience Research Center, Liaoning Normal University Institutional Review Board, and informed consent forms were signed by all participants.

### Stimulus, Procedure, and Data Analysis

The stimuli, apparatus, procedure, and analysis were identical to those of experiment 1 with the following exceptions. During training, participants were asked to press F or J as quickly as possible whenever they saw a fearful or happy face with either a vertical or horizontal line. Then, a visual feedback informed the participant of the reward earned in that trial, as well as about the total reward accumulated across the trial. Fearful faces low reward were followed by "+20 points" in 80%; the remainder was "+100 points." For happy faces (high reward), the percentage was reversed.

### Results

### Training Results

During the training, we compared baseline with training. Participants showed high ACC to high reward-happy [*t*(19) = −4.18, *p* = 0.001, Cohen's *d* = 0.96] and low rewardfear [*t*(19) = −4.61, *p* < 0.001, Cohen's *d* = 1.06] compared with the baseline (baseline: fear 88.63 ± 3.35, happy 92.79 ± 2.73, neutral 92.58 ± 3.59; reward training: low reward-fear 96.79 ± 2.15, high reward-happy 96.94 ± 4.38, neutral 98.1 ± 3.24). There was no significant difference in RTs between reward training and baseline except in high reward-happy compared with happy [baseline: fear 1207.67 ± 65.25 ms, happy 1184.82 ± 121.8 ms, neutral 1175.96 ± 73.53 ms; reward training: low reward-fear 1251.11 ± 89.92 ms, high reward-happy 1246.49 ± 94.57 ms, neutral 1149.89 ± 114.83 ms; *t*(19) = −2.26, *p* = 0.016, Cohen's *d* = 0.61]. A repeated-measures ANOVA showed that ACCs among rewards are significantly different [*F*(2,38) = 4.32, *p* = 0.021, hp <sup>2</sup> = 0.19]. Low reward fear ACC is lower than neutral [*t*(19) = −2.5, *p* = 0.022, Cohen's *d* = 0.58], and highreward happy is lower than neutral [*t*(19) = −2.09, *p* = 0.05, Cohen's *d* = 0.48]. There is also a significant difference in RTs for reward types after reward learning [*F*(2,38) = 16.18, *p* < 0.001, hp <sup>2</sup> = 0.47]. The training phase data are shown in **Figure 4**.

### Test Results

We wanted to know the impact of previous learning on performance when performance was not predictive of reward. An ANOVA was conducted on the ACC with reward condition. The results showed that there was no difference of reward [*F*(2,38) = 1.07, *p* = 0.19, hp <sup>2</sup> = 0.04; high reward-happy 97.11 ± 7.54, low reward-fear 96.61 ± 5.47, neutral 97.33 ± 3.41]. The analysis of ANOVA on the RT indicated that there are significant differences between reward types [*F*(2,38) = 18.76, *p* < 0.001, hp <sup>2</sup> = 0.14; high reward-happy 1189.84 ± 118.12 ms, low rewardfear 1253.07 ± 120.40 ms, no distractors 1174.35 ± 97.67 ms]. Paired comparisons showed significant differences between high reward-happy and low reward-fear (*p* = 0.001). There was a significant difference between low reward-fear and no rewardneutral (*p* = 0.001), but there was no difference between high reward-happy and no reward-neutral. The test phase data are shown in **Figure 5**.

### Between-Experiments Comparison

The ACC data were evaluated with a 2(Reward: high and low) × 2(Emotion: fearful and happy) repeated-measures ANOVA. The main effects of reward (*F* < 1) and emotion (*F* < 1) were not significant, and the interaction of reward × emotion was also not significant [*F* = 2.87, *p* = 0.109, hp <sup>2</sup> = 0.14].

We also evaluated the RT data according to a 2(Reward: high and low) × 2(Emotion: fear and happy) repeated-measures ANOVA. The main effect of reward was not significant (*F* < 1), and the mean RTs of high reward and low reward were the same. The main effect of the emotion was significant [*F*(1,36) = 28.9, *p <* 0.001], and the mean RT of the fearful face distractor was slower than that for the happy face. Critically, a significant Reward × Emotion interaction was detected [*F*(1,36) = 5.31, *p* = 0.034, hp 2 = 0.24]. The results of simple effect analysis showed that there was a significant difference in the fearful face distractor between high and low reward conditions [*F*(1,36) = 5.73, *p* = 0.029]. High reward is slower than low reward. Moreover, there were significant differences in emotion faces [*F*(1,36) = 14.61, *p* = 0.001]. The fearful distractors (1317.49 ms) were longer than the happy distractors (1189.84 ms) in the high-reward condition; there was no such significant difference in the low-reward condition (*F* < 1). The group comparison data are shown in **Figure 6**.

FIGURE 4 | Mean ACC and RTs of training phase, baseline, and training. Mean ACC to low reward-fear, high reward-happy, and no0020reward-neutral increased compared with baseline (experiment 2). \*\**p* < 0.01; \*\*\**p* < 0.001.

### DISCUSSION

In the current study, we used the reward learning paradigm to investigate the effects of reward learning on attention capture of non-target emotional faces. During the training phase, the fearful and happy faces were connected with different rewards (high vs. low rewards). The participants then completed visual search tasks in the testing phase without any reward feedback, in which emotional faces trained in reward learning were presented as non-target distractors. The results showed that after the reward training, the ACC of the emotional faces was significantly improved compared with the baseline. During the test phase, no significant difference was found between high reward-fear and no reward-neutral conditions, or between high reward-happy and no reward-neutral conditions. The RTs were shorter in the low reward-happy condition than in the high reward-fear and no reward-neutral conditions. In addition, the RT was longer for low reward-fear versus high reward-happy and no reward-neutral conditions. The results suggest that the reward selection history perhaps changed the attentional selection of non-target emotional faces after high-reward learning.

Reward has a strong effect on cognition and can allocate a mass of cognitive resources to reward-related stimuli (Wei et al., 2014). During reward training, fearful and happy faces were presented as the target background, and the attention resources of the participants could be biased toward the facial background (Langton et al., 2008; Wentura et al., 2011). During such training, participants may have implicitly learned an association between the emotional faces and the rewards. Reward feedback enhances the response motivation and improves the performance of target processing (Anderson et al., 2011b; Anderson, 2016).

In the test of experiment 1, no difference was confirmed between the high reward-fear distractor condition and the no reward-neutral condition, but the RTs of the high reward-fear condition were longer than those of the low reward-happy condition. It is well known that threatening stimuli (e.g., fear, anger) could capture the individuals' attention, despite the fact that they were irrelevant to the current goal (Batty and Taylor, 2003; Barratt and Bundesen, 2012; Ikeda et al., 2013; Bucker et al., 2014). However, the current results suggest that processing a fearful face does not generate attention disengagement difficulties. Recent studies regarding reward and emotional processing found that the processing advantage of irrelevant negative stimuli was impaired under reward conditions (Schettino et al., 2013; Padmala and Pessoa, 2014; Yokoyama et al., 2015). In Padmala and Pessoa's (2014) study, however, rewards and tasks were presented in the same stimulus sequence. In our study, reward feedback did not appear during the test. Results still show that the value association between high reward and fearful faces still weakens the processing advantage of fearful faces when the reward feedback is absent. Because of the high reward-fear association acquired, the participants may have adopted an active strategy, which weakens the distraction of irrelevant stimuli on the target searching. It further showed that reward training can effectively regulate the processing of non-target fearful faces, and that the reward-learning effect persisted even when the reward does not appear.

In the test of experiment 2, there was no difference in participants' performance between high reward-happy and no reward-neutral conditions. In addition, the RTs of the low reward-fear condition were longer than those of the high reward-happy and no reward-neutral. It is rather easy to produce perceptual priming with positive emotions in social life situations, which is very common and familiar (Öhman et al., 2001). After the learning between happy face and high reward, it may enhance positive emotional experience, causing the participant to react faster to the happy face distractor. Unlike previous studies, the fearful face was not affected by low-reward learning and still showed a negative processing bias (Itthipuripat et al., 2015; Yokoyama et al., 2015; Bourgeois et al., 2017). We supposed that this may be due to the different task paradigms used and task-irrelevant stimulate previously associated with a small reward shows weaker impact than that previously associated with a larger reward (Anderson et al., 2011a). Moreover, in the study by Yokoyama et al. (2015), tasks were relatively easy with less interference. In the current study, the target search is more difficult because of a lot of task-irrelevant stimulate.

Previous studies concerning reward learning showed that different rewards have disparate effects on attention processing after reward learning. More specifically, task-irrelevant distractor previously associated with a large reward slows visual search more than an equally salient distractor previously associated with a small reward (Anderson et al., 2011a,b; Anderson, 2013). Compared with low reward, a high-reward selection experience could alleviate the disturbance effect introduced by a non-target emotional face and promote the target recognition; this finding seems to be inconsistent with the results of previous studies (Padmala and Pessoa, 2014; Yokoyama et al., 2015). On the whole, on the one hand, individuals may reallocate more cognitive resources to evaluate high-reward selection history during the decision-making stage of the emotional processing (Chen and Wei, 2019). On the other hand, the difference between the participants in experiment 1 and experiment 2 may also lead to this result. The reward learning weakens the processing advantage of the fearful face. However, the high reward-happy faces were less likely to interfere with the goal, probably because positive emotion is more common in real life.

Previous research regarding this issue usually presented rewards and goal tasks sequentially. Such a paradigm setting would increase the motivation of participants. The present study associated rewards with emotional faces in an independent training phase. During the testing, reward-associated stimuli appeared to be task irrelevant, and the reward effect can be observed indirectly. In addition, the current study used the China Facial Affective Picture System to study the emotion processing in a Chinese cultural context. The limitation is that this study chose only two types of emotional faces (happy and fearful). Thus, the discussion of how rewards influence emotional processing is simplified. Owing to the complex nature of emotional stimuli and the difficulty of the task, the rewardlearning effect may be weakened. Future research will explore the impact of reward learning on other emotions, thereby appropriately reducing the difficulty of the task.

### CONCLUSION

The current study investigated the impact of reward learning on emotional attentional capture and provided evidence for relationships between reward learning and emotional faces. The results showed that RTs for high reward-emotional faces distracters are faster than those for low reward-emotional faces. Furthermore, no significant difference was confirmed between the high reward-fear distractor condition and the no reward-neutral condition. We speculate that reward learning affects the attention bias of task-irrelevant emotional faces even when the reward is absent. Furthermore, reward selection history influences the attention bias of emotional faces, in which emotional faces are connected to high reward. Specifically, the attention advantages of fearful faces were regulated by high reward.

### DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

### REFERENCES


### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of "Brain and Cognitive Neuroscience Research Centre, Liaoning Normal University Institutional Review Board" with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the "Liaoning Normal University Institutional Review Board."

### AUTHOR CONTRIBUTIONS

XZ and ZW designed the experiment. BD collected the data. XZ analyzed relevant data and wrote the manuscript. WH revised the paper and contributed to the interpretation of the data for the work.

### FUNDING

The work was supported by the National Natural Science Foundation of China (Grant No. 81601166).


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2020 Zhou, Du, Wei and He. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Regulatory Effects of Reward Anticipation and Target on Attention Processing of Emotional Stimulation

Yujia Yao<sup>1</sup> , Yuyang Xuan<sup>1</sup> , Ruirui Wu<sup>1</sup> and Biao Sang2,3 \*

<sup>1</sup> Zhejiang University of Technology, Hangzhou, China, <sup>2</sup> Shanghai Academy of Educational Sciences, Shanghai, China, <sup>3</sup> East China Normal University, Shanghai, China

Studies suggest that reward and emotion are interdependent. However, there are discrepancies regarding the interaction between these variables. Some researchers speculate that the inconsistent findings may be due to different targets being used. Although reward and emotion both affect attention, it is not clear whether their impacts are independent. This study examined the impact of reward anticipation on emotion processing for different targets. A cue-target paradigm was used, and behavior and eyetracking data were recorded in an emotion or sex recognition task under the conditions of reward and non-reward anticipation. The results showed that when the target was related to the emotional attribute of the stimulus, the reward promoted the processing target information, thereby generating reward-oriented attention. When the target was unrelated to the emotional attributes of the stimulus, the reward did not promote the processing target information, and at the same time, individuals had negative emotional biases toward the emotional faces. The results revealed that, in addition to affecting the attention to emotional faces independently, the target regulated the promotion of reward anticipation to emotional attention and attention bias toward negative stimuli.

### Edited by:

Ruolei Gu, Institute of Psychology, Chinese Academy of Sciences, China

### Reviewed by:

Yi Luo, Virginia Tech, United States Yu Chen, The Graduate Center, The City University of New York, United States

> \*Correspondence: Biao Sang bsang@psy.ecnu.edu.cn

### Specialty section:

This article was submitted to Decision Neuroscience, a section of the journal Frontiers in Psychology

Received: 12 June 2019 Accepted: 06 May 2020 Published: 03 June 2020

### Citation:

Yao Y, Xuan Y, Wu R and Sang B (2020) Regulatory Effects of Reward Anticipation and Target on Attention Processing of Emotional Stimulation. Front. Psychol. 11:1170. doi: 10.3389/fpsyg.2020.01170 Keywords: target, reward anticipation, emotion processing, cue-target paradigm, eye tracking

### INTRODUCTION

In the process of socialization, reward is often associated with positive emotions such as pleasure and satisfaction. Both reward and emotion have affective significance, defined as either negative or positive value to the organism (Pessoa, 2009). The relationship between reward and emotion has attracted the interest of researchers. Existing studies suggest that reward and emotion impact each other. Reward induces positive emotions (Berridge and Robinson, 2003). The reward circuit in the human brain is activated when an individual imagines a pleasant scene while reading a story (Costa et al., 2010) or stares at a photograph of a lover (Aron et al., 2005). Conversely, emotion, especially negative emotion, has an impact on reward. Chen (2013) reported that depression affects the reward system: the activation of reward-related brain areas decreases when depressed patients engage in reward processing. Similarly, during a gambling task, the activation of the reward circuit is lower in college students with high depressive symptoms than in those without such symptoms (Wei, 2015).

However, there are some discrepancies among previous studies on the interaction between reward and emotion. Some studies conclude that there is no interaction between reward and emotion. For instance, Murray (2007) noted that the neural substrates for emotion and reward

were partially non-overlapping. The researcher noted a distinction between reward processing and emotional reactions, with the amygdala playing a crucial role in the latter and only a conditional role in the former. In another study, the cue-target paradigm, in which subjects were asked to attend to valid and ignore invalid spatial cues and motivation was manipulated by varying the magnitude and valence of a monetary incentive expected by the subjects for performing well on the task (Engelmann and Pessoa, 2007), was used to examine the impact of different reward conditions on the identification of vocabulary attributes of the target, and no interaction was found between reward and emotion (Kaltwasser et al., 2013). However, other researchers consider that there may be an interaction between them. Wittmann et al. (2008) found that reward strengthened memory only in the context of positive emotion. The same cue-target paradigm was used to investigate the effect of reward on emotional face recognition. A significant interaction was found between reward and emotion, and only negative emotion processing and bias effects were regulated by reward (Wei et al., 2014). Researchers speculate that the key reason for the divergence in the above findings may be the different targets. The target in the research of Kaltwasser et al. (2013) was independent of the emotional valence of the material (judging whether the target was concrete or abstract), while the target in the research of Wei et al. (2014) was related to emotional valence (judging whether the target was positive or negative). Wei and Kang (2014) demonstrated that when emotion was associated with the target, the reward effect of an emotional face (the difference in reaction times <RTs> between the non-reward condition and the reward condition) was greater than that of a neutral face. When emotion was irrelevant to the target, this effect did not exist. However, due to the limited information provided by RTs (Armstrong and Olatunji, 2012), this speculation needs to be further validated at other levels (e.g., attention) by other technologies (e.g., eye movement).

Both reward (Carmona et al., 2012) and emotion (Armstrong and Olatunji, 2012) impact the attention process. For instance, studies have revealed that individuals allocate attention resources to reward-related stimuli (Anderson, 2013). Rewards help stimuli with insignificant features capture attention, even if the rewards subsequently disappear, or the stimulus is independent of the target (Wang, 2016), and when rewards are combined with distraction stimuli, the choice of goals may be hindered (Fan et al., 2014). Numerous studies have found attention biases toward negative stimuli in cognitive processes, which means that individuals detect negative, and threatening stimuli quickly (Jerónimo et al., 2017). Negative faces, especially threatening faces, attract attention, and prolong attention maintenance or reduce attention disengagement ability (Fox et al., 2001). Even if subjects are asked to ignore the emotional information contained in a face, this information still has an impact on the subjects' responses. Bias toward negative stimuli may occur in one or more phases of attention information processing, involving priming, assessment, or response preparation. However, attention is an important stage of information processing (visual: Deubel et al., 2000; auditory: Näätänen, 1990). The relationship between reward and emotion is likely to appear in the attention stage. Wei et al. (2014) claimed that reward anticipation, which involves waiting and eagerness for upcoming rewards (Oldham et al., 2018) in the reward-appetitive phase (Stavropoulos and Carver, 2014), could promote attention to target-related stimuli or attributes. When an emotional attribute of a stimulus is related to a task, it interacts with reward anticipation, which in turn affects the behavioral response of subjects. Nevertheless, this inference still needs to be supported by empirical research in the field of attention. Eye-tracking technology is commonly used to examine the characteristics of individual attention (Liu and Reichle, 2018; Scholz et al., 2018). An eye tracker can provide continuous dynamic information on subjects during cognitive processing at a high sampling rate; it is more conducive to directly measuring the time course of cognitive processing (especially attention processing) than RTs. Therefore, this study attempts to further address this problem with eyetracking technology.

Understanding the emotional characteristics of faces is the key to social adaptation and communication skills (Trentacosta and Fine, 2010). In daily life, facial information communicates data on more than one attribute (such as sex, skin color, or expression). When reward anticipation is attached to the emotional and sex attributes of faces, what happens to the attention process? Does the relationship between reward and emotion change depending on the target? This study tried to answer these questions. As mentioned earlier, reward anticipation processing occurs during the appetitive phase of reward processing and has a strong motivational feature that plays an important role in cognitive processes (Yan et al., 2016). Since reward is usually not given in a timely manner under real experimental conditions, they likely reveal anticipation. Previous research has also found that the expectation of reward improves the preparedness of the corresponding brain regions and promotes behavioral responses to subsequent stimulation. Motivational cues bias individuals' attention resources and target-related information processing by regulating top-down cognitive processes, thereby improving behavioral performance. This study used the cue-target paradigm and eye-tracking technology to examine whether there were differences in behavioral responses and attention characteristics on different targets (emotion recognition and sex recognition) and different reward anticipation (reward and non-reward). Based on previous studies (e.g., Kaltwasser et al., 2013), this experiment predicted that only when the goal was related to the emotion, the reward could promote the processing of emotional information.

### MATERIALS AND METHODS

### Subjects

Twenty-five students from Zhejiang University of Technology were recruited; 6 of them were excluded because of the low average ratio of valid gaze data, which was less than 70%. Nineteen subjects (9 females, aged 18 to 21 years) had an average ratio of 89.75% for valid gaze data. All subjects had normal or corrected-to-normal visual acuity and were paid with basic rewards. This study was carried out in accordance

with the recommendations of the Human Research Ethics Committee of Zhejiang University of Technology. The subjects were recruited through the campus bulletin board and provided signed, informed consent before the experiment. In addition to basic remuneration, the subjects received additional monetary awards based on their experimental performance.

### Design

This study used a 2 × 2 × 2 within-subjects design (target: emotion recognition and sex recognition) × (reward anticipation: non-reward and reward) × (emotional valence: negative and positive). Dependent variables were the subjects' responses (indexed by RT and accuracy) and attention to the pictures [measured by the first fixation ratio (FFR) and fixation duration ratio (FDR)].

Drawing on the classic paradigm in eye-movement experiments, we simultaneously presented two types of stimuli on one slide to examine attention bias. Positive and negative stimuli, such as high emotional arousal stimuli, would inevitably lead to confusion in the individual emotional experience if presented at the same time. In this study, emotion was an independent variable. Therefore, neutral stimuli were added as controls. The two levels of independent variables, positive and negative emotion, were presented in positive-neutral and negative-neutral pairs.

### Materials

The experimental materials (sample face pictures of different emotional valences are shown in **Figure 1**, detail information can be seen in **Supplementary Material**) included 24 neutral face pictures (calm), 12 positive face pictures (happy), and 12 negative face pictures (angry) (Descriptive statistic of valence, arousal and dominance of material see **Table 1**), selected from the Chinese Facial Affective Picture System (Wang and Luo, 2005). The sex ratio of each kind of facial expression was 1:1. There was a significant difference between the valences of the three types of face pictures (ps < 0.05). Additionally, there was no significant difference between positive and negative faces in arousal or dominance (ps > 0.05), while both positive and negative faces had significant differences from neutral faces in arousal and dominance (ps < 0.05).

# FIGURE 1 | Sample face pictures of different emotional valences.



### Instruments

A Tobii X50 telemetry eye tracker with a sampling frequency of 50 Hz was used to track and record the eye movements of the subjects. The program was presented on a 19-inch, 60-Hz cathode ray tube screen. The eye tracker consists of three parts: cameras that take high-resolution images of subjects' eyes and movement patterns, projectors that create a pattern of near-infrared light on the eye, and algorithms (machine learning, image processing, and mathematical algorithms) that are used to determine the eyes' position and gaze.

### Procedure

All subjects were instructed to sit approximately 60 cm in front of the display screen and to complete the experiment independently. Based on the requirements of eye-tracking experiments, fivepoint calibration was used to ensure the accuracy of eye-tracking recording before the experiment started.

The procedure was written using E-prime 2.0. The subjects were asked to complete the emotion recognition task and sex recognition task separately, with corresponding instructions before each task began. The order of the tasks and the correct responses were counterbalanced between the subjects. The cuetarget paradigm (Wei et al., 2014) was modified using the experimental procedure illustrated in **Figure 2**, in which the background was set to white, and the cues and fixations were set to black. Before the experimental procedure began, the subjects were informed of the experimental process and the meaning of the cues and feedback pictures. In the practice phase (shown in **Figure 2A**), each trial began with the fixation "+" (0.59◦ × 0.59◦ visual angle) in the center of the screen for 600 ms. Then, a cue "<sup>∗</sup> " was presented for 500 ms. The fixation appeared again for 100 ms to reset the gaze. After that, two face pictures of different sexes and expressions were presented in pairs. In the emotion recognition task, the subjects were asked to identify the location of pictures according to emotional arousal (high or low). In the sex recognition task, they were asked to identify the location of pictures according to sex (female or male). As one of the objectives in this experiment was to measure the attention maintenance of the subject, the duration of the pictures was fixed to 1500 ms even if the subject responded. Then, the fixation appeared for another 100 ms to reset the gaze, followed by feedback for 500 ms. A gray solid circle appeared on the screen if the subject's response was correct, and a gray hollow circle appeared on the screen if it was incorrect. Moreover, since the reaction rates had individual differences, the paradigm needed to feed back the RT of subjects to determine whether they would be rewarded. The average RT of every subject in the practice phase was recorded and analyzed as a baseline.

The formal experiment was divided into reward trials (as shown in **Figure 2B**) and non-reward trials (as shown in **Figure 2C**). In the reward trials, "U" was presented as a cue to represent money. The feedback varied depending on the response of the subject: if the subject responded correctly and faster than his or her baseline, a coin would be presented on the screen; if the subject reacted correctly but more slowly than his or her baseline, a gray solid circle appeared on the screen; if the subject reacted incorrectly, a gray hollow circle appeared on the screen. In non-rewarded trials, "#" was presented as a cue to represent no money. The feedback varied depending on the response of the subject: if the subject's reaction was correct, regardless of how fast the reaction, a gray solid circle appeared on the screen; if the subject's reaction was incorrect, a gray hollow circle appeared on the screen.

There were 32 trials in the practice phase, of which 16 trials were sex recognition tasks (8 reward trials and 8 nonreward trials), and 16 trials were emotion recognition tasks (8 reward trials and 8 non-reward trials). There were 128 trials in the formal phase, of which 64 trials were sex recognition tasks (32 reward trials and 32 non-reward trials), and 64 trials were emotion recognition tasks (32 reward trials and 32 nonreward trials). In the formal experiment, 8 trials with the same task (recognizing emotion or sex) were used as a block, with a total of 16 blocks. The presentation order of blocks was balanced between subjects. The cues (U or #) and the same categories of pictures (such as a happy male face) were presented randomly within blocks.

### Data Analysis

The RTs and accuracies of the subjects in the formal experiment were recorded by E-prime 2.0. E-DataAid was used to collate the data, which were exported to SPSS 22.0 for further statistical analysis. The Bayesian factor (BF10) was calculated by JASP<sup>1</sup> (Wu et al., 2018). JASP provides options for model comparison and data results output. We chose "compare to best model" for TABLE 2 | Interpretation of Bayesian factors (BF10).


model comparison and "across matched models" for calculating the effect of the data. Based on Jeffreys (1961) and Wetzels and Wagenmakers (2012), the interpretation of the Bayesian factor (BF10) is presented in **Table 2**.

The eye trajectories to faces were determined by presenting positive/negative faces and neutral faces simultaneously as two areas of interest (AOIs) on one slide. The time to first fixation (TFF) and total fixation duration (TFD) data were obtained. TFF is the time point when the gaze of the subject falls on the stimulus for the first time with a latency less than 700 ms and a duration greater than 100 ms. Researchers generally use TFF to reflect subjects' facilitated attention, which belongs to the automatic processing system and is driven by stimulation. TFF reflects the processing order of a stimulus, which means that the shorter the TFF is, the earlier the AOI is noticed, and the more sensitive or alert an individual is to the stimulus (Cisler and Koster, 2010). TFD is the sum of the fixation durations of subjects to the AOI during the entire stimulus presentation process. Researchers generally use TFD to reflect an individual's difficulty with disengagement from stimulation (indicating damage to the attention control system) or attention avoidance (reflecting activation of the attention control system), which can reflect the entire cognitive processing of stimulation (Cisler and Koster, 2010).

Time to first fixation and TFD are time variables that are easily affected by individual differences. Differences in results may be caused by differences in the attention features of individuals

<sup>1</sup>https://jasp-stats.org/


TABLE 3 | Descriptive statistics of accuracy and RT under different conditions (M ± SD).

rather than different experimental conditions. As in previous studies (Wang and Yu, 2017), RT was transformed into the change ratio of RT to exclude the influence of individual differences on the target. We reanalyze TFF and TFD in the form of ratios to define FFR and FDR as new dependent variables. FFR is defined as the ratio of the number of trials with a quicker TFF in a positive/negative AOI to the total number of trials for the same experimental condition compared with a neutral AOI. For example, there were 32 negative-neutral reward trials in the emotion recognition task. If there were 24 trials with a quicker TFF for negative faces than for neutral faces, then the FFR in negative-neutral reward trials would be 0.75. The calculation of FDR is also based on two AOIs of positive/negative and neutral faces. FDR is defined as the ratio of the total TFD in a positive/negative AOI to the sum of TFDs in two AOIs under the same experimental conditions. For example, if the total TFD of negative faces were 16000 ms and the sum of the TFDs of negative and neutral pictures were 20000 ms in the 32 negative-neutral reward trials in the emotion recognition task, the FDR would be 0.8.

The statistical analyses we used are presented as follows. First, accuracy and RT were analyzed separately using repeatedmeasures ANOVA, taking target (emotion recognition and sex recognition), reward anticipation (reward and non-reward), and emotional valence (negative and positive) as factors. Regarding the target differences, we performed a separate repeatedmeasures ANOVA for each task with reward anticipation and emotional valence as factors. Since neutral faces and emotional faces were presented at the same time, we integrated FFR and FDR into the positive condition and negative condition, taking 0.5 as the expected value to perform a one-sample t-test. Finally, FFR and FDR as eye-tracking indexes were separately analyzed using repeated-measures ANOVA, taking reward anticipation, emotional valence, and target as factors. We also performed a separate repeated-measures ANOVA for each target with experimental reward anticipation and emotional valence as factors. Where a significant difference was found between factors, Student's t-test was used.

### RESULTS

### Behavioral Data

ANOVA results (descriptive statistics are reported in **Table 3**) show that the interaction between target and emotion was significant in terms of accuracy [F(1,18) = 17.15, p = 0.001, ηp <sup>2</sup> = 0.49, and BF<sup>10</sup> = 1083.61] and RT [F(1,18) = 4.97, p < 0.05, ηp <sup>2</sup> = 0.22, BF<sup>10</sup> = 1.24]. The simple effect test found that in the sex recognition task, there was a higher accuracy (p < 0.05) and a shorter RT (p < 0.05) under the positive condition, indicating superior processing toward positive faces, and that there was no significant difference in the emotion recognition task. The main effect of emotion on accuracy was significant [F(1,18) = 8.75, p < 0.05, η<sup>p</sup> <sup>2</sup> = 0.33, and BF<sup>10</sup> = 16.46], reflecting that accuracy under the positive condition was higher than that under the negative condition. Other main effects and interactions were not significant (ps > 0.05). Separate repeated-measures ANOVA results showed that in the emotion recognition task, the main effect of reward anticipation was significant [F(1,18) = 5.18, p < 0.05, η<sup>p</sup> <sup>2</sup> = 0.22, and BF<sup>10</sup> = 2.67]. This result reflects a shorter RT under the reward condition than under the nonreward condition and indicates a behavior bias toward reward. In the sex recognition task, the main effect of emotion was significant [F(1,18) = 4.50, p < 0.05, η<sup>p</sup> <sup>2</sup> = 0.20, and BF<sup>10</sup> = 1.01], manifesting as a shorter RT under the positive condition than under the negative condition. Other main effects and interactions were not significant (ps > 0.05).

### Eye-Tracking Data

The results of the one-sample t-test (descriptive statistics are reported in **Table 4** and **Figure 3**) show that FFR was marginally significantly different under negative conditions [t(18) = 2.06, p = 0.06, d = 0.67, and BF<sup>10</sup> = 2.55], indicating an attention bias toward negative faces. FDR was also significantly different under negative conditions [t(18) = 2.58, p < 0.05, d = 0.84, and BF<sup>10</sup> = 6.13], suggesting that negative faces were able to hold attention longer than neutral faces. Under positive conditions, comparing FFR and FDR with 0.5, no significant difference could be asserted (ps > 0.05).

The ANOVA results show that the main effect of the target was significant [F(1,18) = 18.13, p = 0.000, η<sup>p</sup> <sup>2</sup> = 0.50; and BF<sup>10</sup> = 10.02] on FFR, with a lower FFR in the emotion recognition task than in the sex recognition task. The main effect of emotion was also significant [F(1,18) = 5.82, p = 0.03, ηp <sup>2</sup> = 0.24; and BF<sup>10</sup> = 1.03]; the FFR of positive faces was lower than that of negative faces, which also indicated a negative bias. Other main effects and interactions were not significant (ps > 0.05). Separate repeated-measures ANOVA results show that in the emotion recognition task, the main effect of reward anticipation was significant [F(1,18) = 5.37, p = 0.03, η<sup>p</sup> <sup>2</sup> = 0.23; and BF<sup>10</sup> = 1.89], with a significantly lower FFR under the reward condition than under the non-reward condition. Additionally,

TABLE 4 | Descriptive statistics of FFR and FDR under different conditions (M ± SD).

FIGURE 3 | This figure reveals the mean FFR (A) and FDR (B) by target, emotion type and reward anticipation. Error bars represent the standard error of the mean. These lines mainly indicate that there are significant differences under different conditions through the difference comparison test. <sup>∗</sup>Represents p < 0.05.

the main effect of emotion was significant [F(1,18) = 6.88, p = 0.02, η<sup>p</sup> <sup>2</sup> = 0.28; and BF<sup>10</sup> = 13.40], with a lower FFR for positive faces than for negative faces, while the interaction between emotion and reward anticipation was not significant. In the sex recognition task, other main effects and interactions were not significant (ps > 0.05).

The ANOVA results indicate that the main effect of the target was marginally significant [F(1,18) = 4.08, p = 0.059, η<sup>p</sup> <sup>2</sup> = 0.19; and BF<sup>10</sup> = 246.22] on FDR, with a higher FDR in the emotion recognition task than in the sex recognition task. The main effect of emotion was also significant [F(1,18) = 6.67, p < 0.05, ηp <sup>2</sup> = 0.27; and BF<sup>10</sup> = 1.07], with a lower FDR for positive faces than for negative faces. Other main effects and interactions were not significant (ps > 0.05). Separate repeated-measures ANOVA results show that in the sex recognition task, the main effect of emotion was significant [F(1,18) = 4.41, p < 0.05, η<sup>p</sup> <sup>2</sup> = 0.20; and BF<sup>10</sup> = 4.07], with a lower FDR for positive faces than for negative faces. Other main effects and interactions were not significant (ps > 0.05).

### DISCUSSION

No interaction between reward anticipation and emotion was found in the emotion recognition or sex recognition tasks in this study. However, we found that the results in the two tasks were completely different. When the target was related to emotion (emotion recognition task), consistent with the research by Wei et al. (2014), the main effect of reward anticipation was significant. This may be because reward anticipation promotes the processing of the emotional attributes of a stimulus, so in the reward trials, the subjects showed shorter RTs and higher FFRs. While the target was unrelated to emotion (sex recognition task), consistent with the research by Kaltwasser et al. (2013), the main effect of reward anticipation was not significant. The subjects focused on the sex information of the stimulus, which interfered with the automatic processing of the emotional attributes of the stimulus. The processing of the emotional attributes of a stimulus is superior to the processing of other attributes to a certain extent, and it has a certain impact on the processing of other attributes (Yang et al., 2016); thus, reward anticipation promotes the processing of emotion and sex information at the same time. In a word, it suggests that the target itself regulates the effect that reward anticipation work on the processing target attribute. According to the theory of Murray (2007), reward anticipation cues influence behavior, whose underlying mechanism comprises two systems inside the amygdala running in parallel. One system can adjust the universal arousing effect of reward anticipation, while the other links the sensory properties of reward anticipation

with emotion. Therefore, in the emotion recognition task, the two systems work together, resulting in shorter RTs and higher FFRs in reward trials. Meanwhile, the two systems of reward anticipation inside the amygdala run separately, resulting in the effect of reward anticipation being dispersed, which causes no significant main effect of reward anticipation. In other words, reward anticipation promotes emotion processing explicitly and automatically, but the processing of emotional information (such as the emotional attributes of stimuli) disperses part of the promotion effect of reward anticipation. The processing of emotion stimuli increases processing speed only when it is rewarded.

Consistent with previous studies (Hartikainen et al., 2014), this study indicated that subjects had a negative emotional bias during the face recognition process. Individuals exhibited a priority effect on unpleasant stimuli, especially threatening stimuli such as violence, bloody scenes and angry faces, which affected psychological processes and behavioral responses (Buckner et al., 2010). Similarly, the target influenced the effect of emotion on the response of the emotion process. According to the perceptual load theory, when attention resources are completely occupied by task-related content, the process of dealing with task-free interference will stop (Neumann et al., 2011), and emotional bias is affected by the perceptual load (Luo et al., 2017). Negative emotional bias appeared in the sex recognition task, indicating that some attention resources were allocated to emotional processing. However, in the emotion recognition task, compared with positive faces, subjects gazed at negative faces for a longer time but with lower processing quality. This was likely because more attentional resources were used to alert individuals to negative stimuli, and fewer cognitive resources were used for target-related processing, resulting in a higher RT (Zhu and Zhu, 2011; Ji, 2013) or lower accuracy. This indicates that the emotional valence of the stimulus might affect the overall attention processing quality. Individuals have a need to stay in a neutral state, and they may need to spend additional resources regulating the effect of negative emotion when processing stimuli. This reduces the processing speed and interference accuracy of the target-related process.

Although the target did not impact the interaction between reward anticipation and emotion, the results suggested that the main effect of the target was significant for FFR and FDR, which indicated that the target might affect the emotion attention process independently. Compared with the sex recognition task, the subjects had lower FFRs and higher FDRs in the emotion recognition task. This meant that the individuals tried to avoid emotional faces while holding a high level of attention maintenance when completing an emotion-related task. According to previous studies, if the target is valence related, the valence of faces will have additional effects on attention resource allocation (Schulz et al., 2013).

The rapid and effective identification and analysis of various types of information in complex environments are of great significance to the adaptation and development of individuals. The results of this study provide theoretical support for understanding individuals' emotion processing. This study found that reward anticipation promotes emotion processing explicitly and implicitly. We required the subjects to respond as quickly as possible during the practice trials (baseline). And in the formal trials, subjects can receive rewards only when their response was faster than the baseline, which required the subjects to pay close attention to the target. Besides, implicit processing of emotional faces included recognition of other facial cues, such as sex (Scheuerecker et al., 2007). In the emotion recognition task, in which emotion processing is explicit processing of faces, consistent with the research by Wei et al. (2014), the results show that reward anticipation promotes emotion processing; that is, reward anticipation promotes explicit emotion processing. In the sex recognition task, emotion processing is implicit processing of faces. The subjects showed higher processing quality in a shorter fixation duration for positive faces, while reward anticipation did not promote sex processing. It is inferred that this was due to the emotional content being automatically processed and interfered the effect of reward anticipation on sex processing (Rigoulot et al., 2012), and emotion processing was promoted, so we speculate reward anticipation promoted implicit emotion processing.

### CONCLUSION

This study, which adopted a cue-target paradigm to explore the role of the target in the relationship between reward anticipation and emotion, drew the following conclusions: Target status can regulate the promotion of reward anticipation to emotional attention. Reward anticipation promotes explicit emotion processing. The emotional relevance of the target can impact the orientation and maintenance of attention to emotional faces. Emotional attributes may take processing priority over other attributes to a certain extent.

### DATA AVAILABILITY STATEMENT

The database generated for this study is available upon request to the corresponding author.

### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of The ZJUT Human Research Ethics Committee. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by The ZJUT Human Research Ethics Committee.

### AUTHOR CONTRIBUTIONS

YY and YX developed the study concept and design. Data collection was performed by RW. YY and YX performed the data analysis. All authors contributed to the data interpretation and manuscript writing and approved the final version of the manuscript for submission.

### FUNDING

fpsyg-11-01170 May 30, 2020 Time: 19:19 # 8

YY received support from National Education Science Planning Projects (ECA160410) and Preliminary Research Project in Humanities and Social Sciences of ZJUT.

### REFERENCES


### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg. 2020.01170/full#supplementary-material



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Yao, Xuan, Wu and Sang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.