Internet gaming disorder (IGD) is defined as the inability to control excessive Internet game playing [1–Reference Ko, Liu, Yen, Chen, Yen and Chen3], and has been shown to share similar neuropsychological processes with drug addictions and pathologic gambling [2–Reference Grant, Potenza, Weinstein and Gorelick4]. The Diagnostic and Statistical Manual of Mental Disorders 5th Edition (DSM-5) has included IGD as a condition deserving further studies [Reference Association5] and encouraged more investigations of this disorder for possible inclusion in future editions of the DSM.
Neuroimaging studies of IGD have focused on the identification of regional alterations in structure and function [Reference Ko, Liu, Hsiao, Yen, Yang and Lin2,Reference Ko, Liu, Yen, Chen, Yen and Chen3,Reference Ding W-n, Sun J-h, Sun Y-w, Zhou, Li and Xu J-r6,Reference Du, Qi, Yang, Du, Gao and Zhang7]. However, growing evidence demonstrated that this behavioral addiction is associated with system-level alterations between brain regions rather than with the functional breakdown of isolated regions [8–Reference Zhang, Mei, Zhang, Wu and Zhang12]. Emerging concepts in cognitive neuroscience suggested that an individual's behaviors are governed by the interaction of multiple brain regions [Reference Park and Friston13]. Functional brain imaging data have revealed that the human brain is topologically organized into a set of coherent spatiotemporal Independent Component Networks (ICNs), which orchestrates disparate cognitive processes [14–Reference Fox, Snyder, Vincent, Corbetta, Van Essen and Raichle17]. Multiple resting-state functional magnetic resonance imaging (fMRI) and structural brain imaging studies have consistently found that IGD showed abnormalities in extensive ICNs, such as executive-control network (ECN), salience network (SN) [Reference Xing, Yuan, Bi, Yin, Cai and Feng18,Reference Yuan, Qin, Yu, Bi, Xing and Jin19], default-mode network (DMN) [Reference Ding W-n, Sun J-h, Sun Y-w, Zhou, Li and Xu J-r6], and sensory-motor-related brain networks [Reference Wang, Yin, Sun, Zhou, Chen and Ding8]. Although these ICNs may take charge of distinct cognitive processes, the information that they process needs to be integrated for coherent cognition, perception, and behaviors [Reference Power, Cohen, Nelson, Wig, Barnes and Church15]. Altered interactions between ICNs have been demonstrated to have great potential serving as biomarkers of IGD. For example, the imbalanced functional link between ECN and reward network in IGD can predict their online-gaming behaviors [Reference Dong, Lin, Hu, Xie and Du20], and inefficient functional interactions and reduced fractional anisotropy between ECN and SN were also detected in IGD [Reference Yuan, Qin, Yu, Bi, Xing and Jin19]. Thus, analyses of ICNs and their interactions may help to elucidate impaired network patterns in IGD [Reference Chu-Shore, Kramer, Bianchi, Caviness, Cash and Network Analysis:21].
Recently, a triple-network model regarding the abnormal interactions between ECN (whose key nodes include the dorsolateral prefrontal cortex [DLPFC], and posterior parietal cortex [PPC]), DMN (which includes nodes of the ventromedial prefrontal cortex [VMPFC] and posterior cingulate cortex [PCC]) and SN (which includes seeds of the ventrolateral prefrontal cortex [VLPFC] and anterior insula (jointly referred to as the fronto-insular cortex; FIC) and the anterior cingulate cortex [ACC]) [Reference Sridharan, Levitin and Menon22] has been proposed to characterize psychiatric and neurological disorders [Reference Menon23]. This model has also been adapted to understand the mechanism of addictive disorders [Reference Sutherland, McHugh, Pariyadath and Stein24]. SN has been reported to play a critical role in switching between ECN and DMN [Reference Sridharan, Levitin and Menon22], allocating attentional resources toward ECN and DMN, and therefore facilitating orientation to external versus internal stimuli [Reference Menon23,25–Reference Seeley, Menon, Schatzberg, Keller, Glover and Kenna28]. Consistent with the framework, a study on alcohol use disorder suggested that the impaired decision making in alcohol addicts may not only be a deficiency in either the DMN or ECN, but also a deficiency in the switching between those networks, and the key site of this impairment may be the anterior insula [Reference Arcurio, Finn and James29]. Based on the triple-network model, Lerman & Gu [Reference Lerman, Gu, Loughead, Ruparel, Yang and Stein30] proposed a resource allocation index (RAI) as SN-ECN connectivity subtracting SN-DMN connectivity, reflecting the superiority of modulation of ECN than DMN (or allocating more resources to ECN versus DMN). Using this index, they found that the increased craving for smoking in smokers under acute abstinence was negatively correlated with the reduced RAI, but not with the abnormal SN-ECN or SN-DMN connectivity. Thus, integrating the inter-connectivity between SN and DMN/ECN may offer more comprehensive information about attentional and cognitive control in addiction.
Although IGD shows similar pathogenetic processes with substance use disorders as well as pathological gambling [2–Reference Grant, Potenza, Weinstein and Gorelick4]; as a behavioral addiction, it is relatively free from the pharmacological effects of substance use [Reference Cho, Kwon, Choi, Lee, Choi and Choi31]. Whether and how the interactions between these ICNs (especially the modulation of ECN versus DMN by SN) are disrupted in IGD remains unexplored. In the current study, we aimed firstly to identify DMN, ECN and SN in adolescents with IGD and HCs. Second, given the hypothesized role of SN in toggling resources between the ECN and DMN, we explored the inter-connectivity between SN and DMN/ECN, and then examined the alterations in the modulation of DMN and ECN using the RAI introduced by Lerman & Gu [Reference Lerman, Gu, Loughead, Ruparel, Yang and Stein30]. Finally, we assessed the association of altered coupling and the severity of addiction/craving for Internet gaming within IGD.
2. Materials and methods
2.1. Participants’ inclusion criteria and clinical assessments
This study was approved by the Institutional Review Board of the State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University. Prior to the study, written informed consent was signed by all participants.
Given the higher incidence of IGD in male than female [Reference Ko, Yen, Chen, Yang, Lin and Yen32], a total of 432 males (319 individuals potential to have IGD [who play Internet gaming frequently] and 113 potential HCs [who play Internet gaming occasionally]) were recruited for initial screening via the Internet and advertisements posted at local universities. The diagnosis criteria of IGD contained: score ≥ 67 [Reference Ko, Yen, Chen, Yang, Lin and Yen32] of the Chinese Internet Addiction Scale (CIAS)[Reference Chen, Weng, Su, Wu and Yang33]; spent more than half of the online time on Internet gaming [Reference Lin, Dong, Wang and Du34]; and ≥ 14 hours spent on Internet gaming per week (with at least 2 hours spent on Internet gaming every day) as assessed by a semi-structured interview [Reference Zhang, Yao, Li, Zang, Shen and Liu9]. However, the participants with score of CIAS ≤ 60 and never having or spent ≤ 2 hours per week on Internet gaming were classified as healthy control (HC) [Reference Zhang, Yao, Li, Zang, Shen and Liu9]. Participants were excluded if they are younger than 18 or older than 30 years, left handed; score of the Fagerstrom Test for Nicotine Dependence (FTND) [Reference Fagerstrom35] > 6; score of the Alcohol Use Disorder Identification Test (AUDIT-C) [Reference Bush, Kivlahan, McDonell, Fihn, Bradley and The36] > 9; any history of other psychiatric or neurological illness, or current or previous use of illegal substances or gambling, or currently taking any psychotropic medications; not suitable for MRI scanning, or with excessive head motion. The data in this study consisted of 39 subjects with IGD and 34 matched healthy controls.
The Beck Anxiety Inventory (BAI) [Reference Beck, Brown, Epstein and Steer37], Beck Depression Inventory (BDI) [Reference Beck, Erbaugh, Ward, Mock and Mendelsohn38], and subjective craving of Internet (gaming) adapted from the Questionnaire of Smoking Urges (QSU-brief) [Reference Cox, Tiffany and Christen39], were conducted to assess the clinical situation of the subjects.
2.2. Image acquisition
Magnetic resonance imaging (MRI) was conducted with a Siemens Trio 3-Tesla scanner (Siemens, Erlangen, Germany). The resting-state fMRI data comprised 200 continuous echo-planar imaging (EPI) whole-brain functional volumes: repetition time (TR) = 2000 ms; echo time (TE) = 30 ms; flip angle (FA) = 90°; slice number = 33; field of view (FOV) = 200× 200 mm; matrix size = 64 × 64; voxel size = 3.1× 3.1 × 3.5 mm3; gap = 0.7 mm. The subjects were instructed to look at a black screen, staying awake and motionless, and not to think of anything in particular.
2.3. Preprocessing of fMRI images
Image data were preprocessed using DPARSF version 3.0 ([Reference Yan and Zang40]; http://rfmri.org/DPARSF]. Slice timing was applied to correct for within-scan acquisition time differences between slices, then images were realigned to the first volume. Subject whose head motion exceeds 3.0 mm in translation or 3 in rotation was excluded. Friston's 24-parameter model [Reference Yan, Cheung, Kelly, Colcombe, Craddock and Di Martino41] were conducted to reduce the confounds of head motion. Besides, signals from the cerebrospinal fluid and white matter were regressed out [Reference Murphy, Birn, Handwerker, Jones and Bandettini42]. The data were normalized to the MNI space by Dartel [Reference Tahmasebi, Abolmaesumi, Zheng, Munhall and Johnsrude43], and smoothed with a full width at half maximum (FWHM) Gaussian kernel of 8 mm [44–Reference Wang, Wu, Lin, Zhang, Zhou and Du47].
2.4. Group independent component analysis and network identification
The preprocessed data were further analyzed using a standard procedure in group ICA algorithm (GIFT, http://mialab.mrn.org/software/gift,version3.0a) to identify spatially independent networks [Reference Calhoun, Liu and Adalı48]. Data from all participants were concatenated into a single dataset and reduced its dimensionality using two stages of principal component analysis (PCA) [Reference Calhoun, Adali, Pearlson and Pekar49]. Then the data were separated with an infomax algorithm [Reference Bell and Sejnowski50] into 23 independent components, determined by the modified minimal description length (MDL) criteria. The analysis was repeated 20 times using ICASSO to assess the stability of Independent Components [Reference Himberg, Hyvärinen and Esposito51]. A time course for each component and its corresponding spatial map were obtained and then back-reconstructed for each participant [Reference Calhoun, Adali, Pearlson and Pekar49,Reference Meda, Stevens, Folley, Calhoun and Pearlson52]. The individual subjects’ spatial maps of each Independent Component Networks (ICNs) were converted to Z values. Hence the intensities of each spatial map indicated the relative contribution of voxels to distributed and coherent brain activity within that ICN [Reference Beckmann, DeLuca, Devlin and Smith53]. For the spatial maps of each selected ICN, voxel-wise one-sample t-test was conducted across all subjects to define brain regions that belong to the ICN.
Based on the descriptions available from reference studies [Reference Yuan, Qin, Yu, Bi, Xing and Jin19,Reference Lerman, Gu, Loughead, Ruparel, Yang and Stein30,Reference Cai, Chen, Szegletes, Supekar and Menon54,Reference Liang, He, Salmeron, Gu, Stein and Yang55], the SN/DMN were bilateral spatial maps, but the ECN could split to left and right maps. Referring to the main brain regions of SN/ECN/DMN introduced in Sridharan & Levitin [Reference Sridharan, Levitin and Menon22], the SN/DMN/right ECN (rECN)/left ECN (lECN) were identified by visual inspection based on the descriptions of these networks of previous studies [Reference Yuan, Qin, Yu, Bi, Xing and Jin19,Reference Lerman, Gu, Loughead, Ruparel, Yang and Stein30,Reference Cai, Chen, Szegletes, Supekar and Menon54,Reference Liang, He, Salmeron, Gu, Stein and Yang55]. Furthermore, spatial sorting with Multiple Linear Regression (MLR) of GIFT was completed to assess correlations between the identified ICNs and templates by Laird (with SN/DMN/rECN/lECN corresponding to IC4 (bilateral anterior insula/frontal opercula and the anterior aspect of the body of the cingulate gyrus), IC13 (medial prefrontal and posterior cingulate/precuneus areas), IC15 (right-lateralized fronto-parietal regions), IC18 (left-lateralized fronto-parietal regions) respectively) [Reference Laird, Fox, Eickhoff, Turner, Ray and McKay56].
2.5. Analyses of inter-connectivity between SN and DMN/ECN and RAI
The individual time courses for SN/DMN/rECN/lECN, obtained by back-reconstruction for each participant, were used to compute the coupling of these networks. Using the same analysis as developed by Lerman & Gu [Reference Lerman, Gu, Loughead, Ruparel, Yang and Stein30], the inter-network connectivity between SN and ECN (ZSN,ECN) and between SN and DMN (ZSN, DMN) were computed as the correlation coefficients between corresponding component time courses and then converted to z scores via Fisher's r-to-z transform. Finally, based on the crucial role of SN in switching between CEN and DMN and allocating attentional resources [Reference Sridharan, Levitin and Menon22], we defined an SN centered resource allocation index (RAI), referring to Lerman & Gu [Reference Lerman, Gu, Loughead, Ruparel, Yang and Stein30], to assess the simultaneous interactions of the SN on DMN and ECN. RAI was computed as the difference between SN-CEN connectivity and SN-DMN connectivity (RAI for the left hemisphere was computed as lRAI = zSN, lECN–zSN, DMN, and for the right hemisphere rRAI = zSN, rECN–zSN, DMN). RAI reflects superior modulation of ECN than DMN; a larger RAI indicates that SN is temporally integrated more with ECN while simultaneously dissociated from DMN [Reference Menon and Uddin25].
2.6. Statistical analyses
One sample t-test was applied across all the subjects to investigate the property of SN-DMN/ECN coupling. Independent sample t-tests were conducted to investigate the group differences in the lRAI/rRAI. To better understand which inter-network connectivity was responsible for the reduced RAI, post-hoc analyses were performed in SN-ECN connectivity (ZSN,lECN or ZSN,rECN) and the SN-DMN connectivity (ZSN, DMN). Although IGD subjects reported significant higher BDI and BAI scores than HC, given concern related to the post-hoc inclusion of covariates [Reference Kraemer57], and the significant correlations between BDI/BAI scores and hours spent on Internet gaming per week/CIAS scores/craving for Internet (gaming) (see Supplementary material Table S1), including these measures as covariates may remove variance explained by problematic Internet game-playing or IGD severity. Thus, we did not include these variables as covariates in our primary data analyses.
Pearson correlations were performed between the altered RAI and severity of IGD/score of craving within IGD to examine its relationship with clinical assessments.
3.1. Participant characteristics
There was no significant differences between the IGDs and HCs in age, education, the proportion of cigarette/alcohol users, or the frame-wise displacement (FD) of head position [Reference Power, Barnes, Snyder, Schlaggar and Petersen58,Reference Power, Mitra, Laumann, Snyder, Schlaggar and Petersen59]. As expected, IGD individuals had significantly higher scores on CIAS, craving, anxiety, depression, and more hours spent on Internet gaming (Table 1).
a Mann-Whitney U Test.
b Chi-square test.
c Data of 7 HCs were included; other HCs never played Internet games.
3.2. Independent component networks in resting state
Referring to the templates in Laird [Reference Laird, Fox, Eickhoff, Turner, Ray and McKay56], SN was consistent with its ICN4 (r 2 = .19), the identified DMN was consistent with their ICN13 (r 2 = .28), whereas the rECN and lECN were respectively consistent with their ICN15 and ICN18 (r 2 = .29 and 0.17, respectively). The Stability index (Iq) of these components were 0.98, 0.97, 0.98 and 0.97, respectively. These components are displayed at p FWE = 0.000001 of the one sample t-test in Fig. 1.
3.3. Differences in RAI/SN-DMN/rECN/lECN connectivity and its correlations with severity of addiction/craving for Internet gaming within IGD
Across all the subjects, SN-DMN connectivity was significantly negative (t (72) = −2.56, P = .013, d = −.30), and SN-rECN connectivity was significantly positive (t (72) =9.36, P < .001, d = 1.10), however SN-lECN connectivity was not significant different from zero (t (72) = .85, P = .400, d = 10). Then we focused on the generated rRAI/lRAI from the difference between SN-rECN/SN-lECN and SN-DMN connectivity, the rRAI and lRAI was significantly higher against 0 (t (72) = 13.12, P < .001, d = 1.18; t (72) = 3.30, P = .002, d = .39).
Moreover, rRAI was decreased significantly in IGD compared to HC (t (71) = −2.42, P = .018, d = −.58), but the decrease of lRAI was not significant (t (71) = −.80, P = .426, d = −.01; Fig. 2 A). The post-hoc analyses manifested that IGD showed significantly increased SN-DMN connectivity (t (71) = 3.36, P = .001, d = .80), while there was no significant alteration of SN-rECN connectivity in IGD (t (71) = 1.32, P = .190, d = .30) (Fig. 2 B).
The rRAI was negatively correlated with the scores of craving (r = −.37, P = .020; Fig. 2 C), but not with severity of addiction (CIAS: r = −.11, P = .513; hours spent on Internet gaming per week: r = .000, P = .998) in IGD. Correlation analyses between SN-DMN/SN-rECN connectivity (zSN,DMN/ZSN_rECN) and severity of addiction/craving for Internet gaming within IGD were done and the results showed no significant associations (|r|’s ≤.25, p's ≥.125; details in Supplementary material Table S2).
To the best of our knowledge, this is the first study to investigate the abnormality in network coupling of SN, ECN and DMN in IGD. Our study showed that the rRAI, an index reflecting the extent of superiority of SN-rECN connectivity (zSN, rECN) versus the SN-DMN connectivity (zSN, DMN), was decreased significantly in IGD compared with HC, which was mainly caused by significantly increased SN-DMN connectivity; and was negatively correlated with the scores of craving for Internet gaming within IGD.
Given the critical role of SN in toggling resources between the ECN and DMN, we mainly focused on the inter-network connectivity between SN and DMN/rECN/lECN, and we found that SN had significantly negative connectivity with DMN, and positive connectivity with rECN but not lECN. Previous studies have also demonstrated that the ECN was involved in exteroceptive processes related to cognitive control and goal-directed attention [Reference Dosenbach, Fair, Miezin, Cohen, Wenger and Dosenbach60], while DMN was implicated in interoceptive processes and self-referential thinking [Reference Raichle, MacLeod, Snyder, Powers, Gusnard and Shulman27]. Our findings were supported by the previous evidences that SN facilitated orientation to external versus internal stimuli and allocating attention [Reference Menon23,26–Reference Seeley, Menon, Schatzberg, Keller, Glover and Kenna28], through its positive correlation with ECN and negative correlation with DMN [Reference Sridharan, Levitin and Menon22,Reference Menon and Uddin25,Reference He, Qin, Liu, Zhang, Duan and Song61]. The rationale here is that SN and CEN are typically co-activated during cognitively demanding tasks, while SN and DMN are typically anticorrelated [Reference Greicius, Krasnow, Reiss and Menon62].
The rRAI, integrating SN-ECN connectivity (zSN, rECN) and the SN-DMN connectivity (zSN, DMN), may reflect a predominant resource allocation by SN to externally oriented cognitive and behavioral control than internally oriented processes. The reduced rRAI we found suggested that the homeostatic balance was impaired in IGD, similar with the findings in the abstinent smokers compared with the smoking-sated subjects [Reference Lerman, Gu, Loughead, Ruparel, Yang and Stein30] and in Attention-deficit/hyperactivity disorder (ADHD) compared with HC [Reference Cai, Chen, Szegletes, Supekar and Menon54]. Moreover, the only significant group difference in rRAI but not lRAI was consistent with previous results showing the causal outflow hub of the right frontal insular cortex (a key component of SN) to rECN and DMN, but not on the left hemisphere [Reference Sridharan, Levitin and Menon22]. A possible explanation is that the rECN played a particularly crucial role in the interpretation and modulation of such bodily sensations [Reference Critchley, Elliott, Mathias and Dolan63], supported by the reduced craving in cocaine dependent individuals by repetitive transcranial magnetic stimulation (rTMS) over the right, but not the left ECN [Reference Camprodon, Martínez-Raga, Alonso-Alonso, Shih and Pascual-Leone64], and the risk-taking behavior induced by a disruption of rECN [Reference Knoch, Gianotti, Pascual-Leone, Treyer, Regard and Hohmann65].
IGD showed an increase in SN-DMN connectivity but not in SN-rECN connectivity, suggested the increased resource was allocated to DMN to promote internal mental processes against the cognitive control. In line with this finding, we had found that IGD had higher functional connectivity between anterior insula and areas of DMN [Reference Zhang, Yao, Li, Zang, Shen and Liu9], and higher task-related activity in the default mode network (DMN) when making the risky decisions [Reference Wang, Wu, Lin, Zhang, Zhou and Du47]. Similarly, cannabis users also showed enhanced of functional anti-correlation between DMN and insula [Reference Pujol, Blanco-Hinojo, Batalla, Lopez-Sola, Harrison and Soriano-Mas66], as well as higher functional connectivity between the right anterior insula and components of DMN (e.g., bilateral precuneus, posterior cingulate cortex and left angular gyrus) in substance use disorders [Reference Menon23,67–Reference Zhang and Li69]. Furthermore, maladaptive interactions between the insula and DMN have been thought of as a key neural marker underlying the development and maintenance of addiction [Reference Sutherland, McHugh, Pariyadath and Stein24].
We have also found that rRAI was negatively correlated with the higher craving for internet gaming in IGD, consistent with the association between decreased RAI and increased craving for smoking by Lerman & Gu [Reference Lerman, Gu, Loughead, Ruparel, Yang and Stein30]. Furthermore, we also examined the relationship between SN-DMN connectivity (zSN, DMN) and scores of craving within IGD, and results showed no significant associations, suggesting that the dual modulations of DMN and ECN were more sensitive than individual paired network couplings in accounting for IGD, as observed in the previous study of substance addiction and ADHD [Reference Lerman, Gu, Loughead, Ruparel, Yang and Stein30,Reference Cai, Chen, Szegletes, Supekar and Menon54]. These findings may be explained by the triple-network model, proposed by Menon [Reference Menon23], that abnormal organizations of the SN, ECN and DMN are prominent characteristics of various psychiatric and neurological disorders, including addiction. The key part in this model is the improper allocation of saliency to external stimuli or internal mental events [Reference Menon23,Reference Menon and Uddin25]. Thus, in IGD, SN directed more attentional resources towards the internal state via increasing interaction with DMN, despite effort at cognitive control (SN-rECN connectivity) in an attempt to override/inhibit the enhanced craving, the dysfunction of SN in switching from DMN to ECN (decreased rRAI) made individuals perceive more pronounced self-focused thoughts related to craving and withdrawal.
It is important to note that IGD showed significantly higher anxiety and depression scores than HC, which were significantly associated with problematic Internet game-playing. Similar with findings, prior studies have reported elevated anxiety and depression scores among frequent Internet users [Reference Dalbudak, Evren, Aldemir, Taymur, Evren and Topcu70,Reference Demirci, Akgönül and Akpinar71], which were positively related with problematic Internet use [72–Reference Błachnio, Przepiórka and Pantic74]. In addition, among psychological variables considered, depression has been most strongly associated with the development of IGD [Reference Hyun, Han, Lee, Kang, Yoo and Chung75]. Thus, for individuals with IGD, higher depression and anxiety might be representative indicators of problematic Internet game-playing. Actually, we also did additional analyses with anxiety/depression and head motion as covariates, there were still marginal significant decrease in rRAI, and significant increases in SN-DMN connectivity (see Supplementary material Table S3). Further studies explicitly recruiting individuals with IGD that have low levels of anxiety and depression are needed to disentangle the effects of these variables on the coupling of these networks.
One limitation of our study is that only male subjects participated in the study. So further studies with female participants were needed to confirm and/or extend the current results. Another is that RAI is an imaging index reflecting inter-network connectivity of SN, ECN and DMN, and further analyses including other networks (e.g., those reflecting emotions) may have greater potentials to characterize abnormal behaviors in IGD. Moreover, because there was no collection of the physiological data, we can’t exclude the effect of cardiac and respiratory fluctuations on our results.
In summary, this study provided novel evidence for the triple-network model in IGD, that the interactions between ECN/DMN and SN, especially the deficient modulation of the activity of ECN versus DMN by SN played a critical role in the maintenance of addictive behaviors in IGD. Such large-scale brain network coupling may provide novel insights for understanding the neurobiological mechanisms underlying IGD and for developing effective treatment strategies for the disorder.
Informed consent was obtained from all individual participants included in the study.
Role of funding source
This study was supported by the National Natural Science Foundation of China (No. 31170990 to X.-Y.F and No. 81100992 to J.-T.Z), Project of Humanities and Social Sciences supported by Ministry of Education in China (No. 15YJC190035 to J.-T.Z), the Fundamental Research Funds for the Central Universities (2017XTCX04); the Hundred Talents Program of the Chinese Academy of Sciences (Y5CX072006 to C.-G.Y), and Open Research Fund of the State Key Laboratory of Cognitive Neuroscience and Learning (to S.Z). Y.-H.Y is supported by the Intramural Research Program of the National Institute on Drug Abuse, National Institutes of Health, USA.
Disclosure of interest
The authors declare that they have no competing interest.
We thank all subjects for participation. We also thank Dr. John M Olson for the grammar checking. All authors have approved the final manuscript.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.eurpsy.2017.06.012.