- Email: [email protected]
Reduced sensitivity to neutral feedback versus negative feedback in subjects with mild depression: Evidence from event-related potentials study
Brain and Cognition 100 (2015) 15–20
Contents lists available at ScienceDirect
Brain and Cognition journal homepage: www.elsevier.com/locate/b&c
Reduced sensitivity to neutral feedback versus negative feedback in subjects with mild depression: Evidence from event-related potentials study Li Peng a,b, Song Xinxin b, Wang Jing b, Zhou Xiaoran b, Li Jiayi b, Lin Fengtong b, Hu Zhonghua b, Zhang Xinxin b, Cui Hewei b, Wang Wenmiao b, Li Hong a,b,⇑, Cong Fengyu c,⇑, Debi Roberson d a
Brian Function and Psychological Science Research Center, Shenzhen University, Shenzhen 518060, China Research Center of Brain and Cognitive Neuroscience, Liaoning Normal University, Dalian 116029, China c Department of Biomedical Engineering, Faculty of Electronic Information and Electrical Engineering, Dalian University of Technology, Dalian 116029, China d Department of Psychology, University of Essex, Wivenhoe Park, Colchester, Essex CO4 3SQ, UK b
a r t i c l e
i n f o
Article history: Received 25 December 2014 Revised 4 August 2015 Accepted 5 August 2015
Keywords: Neutral feedback Depression FRN ACC Reinforcement learning
a b s t r a c t Many previous event-related potential (ERP) studies have linked the feedback related negativity (FRN) component with medial frontal cortex processing and associated this component with depression. Few if any studies have investigated the processing of neutral feedback in mildly depressive subjects in the normal population. Two experiments compared brain responses to neutral feedback with behavioral performance in mildly depressed subjects who scored highly on the Beck Depression Inventory (high BDI) and a control group with lower BDI scores (low BDI). In the first study, the FRN component was recorded when neutral, negative or positive feedback was pseudo-randomly delivered to the two groups in a time estimation task. In the second study, real feedback was provided to the two groups in the same task in order to measure their actual accuracy of performance. The results of experiment one (Exp. 1) revealed that a larger FRN effect was elicited by neutral feedback than by negative feedback in the low BDI group, but no significant difference was found between neutral condition and negative condition in the High BDI group. The present findings demonstrated that depressive tendencies influence the processing of neutral feedback in medial frontal cortex. The FRN effect may work as a helpful index for investigating cognitive bias in depression in future studies. Ó 2015 Elsevier Inc. All rights reserved.
1. Introduction Our complex and dynamic living environment requires human beings to learn how to predict and evaluate the consequences of actions that lead to rewards and punishments. For this reason accurate processing of external feedback is essential if humans are to optimize their behavior (Miltner, Braun, & Coles, 1997). Much research has focused on the brain activity associated with the evaluation of outcomes by examining the timing of electroencephalography (EEG) responses (Holroyd & Coles, 2002; Li, Han, Lei, Holroyd, & Li, 2011; Li et al., 2010; Ullsperger, Fischer, Nigbur, & Endrass, 2014; Walsh & Anderson, 2012). Investigations into feedback evaluation have consistently found negative deflection in the Event-Related Potential (ERP) following the ⇑ Corresponding authors at: No. 3688, Nanhai Road, Nanshan District, Shenzhen 518060, China (H. Li). LingGong Road #2, Dalian 11602, China (F. Cong) E-mail addresses: [email protected] (H. Li), [email protected] (F. Cong). http://dx.doi.org/10.1016/j.bandc.2015.08.004 0278-2626/Ó 2015 Elsevier Inc. All rights reserved.
presentation of negative feedback, an effect that has been termed Feedback Related Negativity (FRN, Cohen, Wilmes, & van de Vijver, 2011; Holroyd & Coles, 2002; Li et al., 2009; Miltner et al., 1997). This time-domain trial-averaged FRN component peaks at 250–300 ms after feedback and has a fronto-central distribution. Convergent findings from multiple methodologies suggest that the FRN is probably generated in the anterior cingulate cortex (ACC) in medial frontal cortex (Hauser et al., 2014; Warren, Hyman, Seamans, & Holroyd, 2014; but see Nieuwenhuis, Slagter, Alting von Geusau, Heslenfeld, & Holroyd, 2005). Theories concerning the functional significance of FRN have been constantly updated in the last two decades. The current most influential theoretical account of FRN comes from Holroyd and Coles’s ‘‘reinforcement learning theory of the error-related negativity (RL-ERN theory)”. According to this theory, the FRN amplitude reflects reward prediction error, i.e. a signed value corresponding to the difference between the obtained reward and the expected reward (Holroyd & Coles, 2002; Nieuwenhuis, Holroyd, Mol, &
16
P. Li et al. / Brain and Cognition 100 (2015) 15–20
Coles, 2004). This explanation has been supported by abundant evidence in FRN studies (e.g. Holroyd, Krigolson, Baker, Lee, & Gibson, 2009; Krigolson, Pierce, Holroyd, & Tanaka, 2009; Walsh & Anderson, 2012). However, several studies in the last ten years have provided evidence that the FRN conveyed an unsigned prediction error or ‘‘salience” encoding signal (Donkers, Nieuwenhuis, & Van Boxtel, 2005; Hauser et al., 2014; Oliveira, McDonald, & Goodman, 2007; Talmi, Atkinson, & El-Deredy, 2013). More recently, Sambrook and Goslin (2015) utilized the ‘‘great grand averages” approach in a meta-analysis study and showed strong effects of magnitude and likelihood on the FRN, which provided new evidence to support the RL_ERN theory. Based on the predictions of RL_ERN theory, researchers proposed that the feedback processing system, putatively indexed by the FRN component, reveals a binary way to evaluate current outcomes, i.e. whether the current outcome is worse than expected or not (e.g. Hajcak, Moser, Holroyd, & Simons, 2006; Yeung & Sanfey, 2004). However, several studies have explored the phenomenon by introducing neutral feedback and showed that neutral feedback elicited a relatively larger FRN than negative feedback (Gu, Ge, Jiang, & Luo, 2010; Hirsh & Inzlicht, 2008; Li, Baker, Warren, Li, submitted for publication; Müller, Möller, Rodriguez-Fornells, & Münte, 2005). Müller et al. (2005) first found that the FRN occurred earlier and had a higher peak in the neutral condition than in the negative feedback condition. This discrepancy between neutral and negative feedback was also observed in two studies with negative affective states (Gu et al., 2010; Hirsh & Inzlicht, 2008). To systematically explore the effect of neutral feedback on the amplitude of the FRN, Holroyd and his colleagues (2006) conducted five experiments, which included an intermediate reward condition in the first three experiments and a neutral feedback condition in the later two experiments. Their results were consistent with the now widely accepted proposals that the evaluation system classifies outcomes into two categories: the satisfied outcome and unsatisfied outcome. Hence, it remains unclear how the evaluation system works when it comes to the neutral feedback. Thus investigating the FRN effect elicited by neutral feedback may also contribute to the above-mentioned arguments concerning the theoretical account of the FRN phenomenon because neutral feedback is a special case in terms of valence and magnitude. According to the RL_ERN theory, the FRN manifests the dopamine signal transferred from the basal ganglia to the medial prefrontal cortex, more specifically, the ACC (Holroyd & Coles, 2002). So far, accumulating evidence shows that the patterns of activation in the ACC during performance monitoring vary as a function of individual differences in personality (Van Noordt & Segalowitz, 2012). Major depressive disorder (MDD) is a highly prevalent multifactorial psychiatric disorder and has been characterized as an abnormal tendency to engage in negative mood states, together with difficulty in disengaging from negative mood states (Holtzheimer & Mayberg, 2011). The impairment of ACC function by depression has been demonstrated in a number of ERP studies reporting that participants with severe MDD showed hyperactivation to internal or external error compared with the control group (Holmes & Pizzagalli, 2008; Santesso et al., 2008; but not in Foti & Hajcak, 2009; Ruchsow et al., 2004). However, it is still unclear whether or not depression influences the reward processing of neutral feedback. In fact, two FRN studies focused on individual differences in personality have found larger FRN effects following neutral feedback in subjects scoring high on neuroticism and high on trait anxiety scales (Gu et al., 2010; Hirsh & Inzlicht, 2008). Given that depression shares the same underlying biases of information processing with trait anxiety (Mathews & MacLeod, 2005) and that such biases might be predicted by neuroticism (Miller & Pilkonis, 2006), it is plausible that depression may also affect the FRN effect with neutral feedback.
To our knowledge, no study to date has focused on the FRN elicited by neutral feedback in depression. The most relevant study came from Mies et al. (2011), in which they investigated both behavioral and electrophysiological responses to feedback validity in non-medicated depressed patients. Their results found that nonmedicated depressed in-patients showed a more pronounced FRN amplitude regardless of feedback validity. It is worth noting that the invalid feedback still contained valence information in Mies et al.’s study, and thus it differed from what we call ‘‘neutral feedback” here. Moreover, in previous studies the severity of depressive symptoms or neuroticism drives different neural responses toward reward (or correct) and non-reward (or error) feedback (Foti & Hajcak, 2009; Hirsh & Inzlicht, 2008; Tucker, Luu, Frishkoff, Quiring, & Poulsen, 2003). Tucker et al. (2003) found enhanced FRN responses in moderately depressed, but not in more severely depressed patients. Foti and Hajcak (2009) showed an enhancement of the FRN to non-rewards relative to rewards that was inversely related to depression. The FRN effect of neutral feedback has also been found to vary linearly as neuroticism scores change (Hirsh & Inzlicht, 2008). The current studies seek to extend this body of research by linking the neutral FRN phenomenon with mild depression in undergraduate students to complement the studies of clinical MDD (e.g. Mies et al., 2011). Using a time estimation task, the present study investigated the outcome evaluation of neutral feedback in two groups varying in non-clinical depression. We also compared the behavioral performance of two groups in a subsequent time estimation task with real feedback. The aim of this study was to investigate whether depression attenuates or increases the performance-monitoring processing of neutral and negative feedback in medial prefrontal cortex. Given that depression is frequently associated with negatively biased information processing (cf. Mathews & MacLeod, 2005), the depressive group tested here was more likely to treat neutral feedback as negative feedback. Therefore, we hypothesized that neutral feedback would elicit comparable FRN amplitudes to negative feedback in the depressive group. We also predicted that distinctly different FRNs would be observed between neutral and negative feedback in the control group as it was in a previous study with random samples (Li et al., submitted for publication). In addition, we intended to compare the behavioral pattern of a depressive group and a control group in Exp. 2, in which real feedback was provided. 2. Methods 2.1. Participants Participants were recruited from a large group (769) of students in two universities. 38 Participants aged 18–25 years were selected for the present study. All of them were assessed for depressive tendencies using the Beck Depression Inventory (BDI; Hautzinger, Bailer, & Worall, 1994) and were measured for trait anxiety using the trait section of the State Trait Anxiety Inventory (STAI; Spielberger, Gorssuch, Lushene, Vagg, & Jacobs, 1983) one week before the experiment was conducted. 19 participants (9 females, mean age 20.6) with a BDI score higher than 10 were identified as the high BDI group and another 19 participants (10 females, average age 20.7) with BDI scores lower than 10 were selected as the low BDI group. A chi-square analysis provided no evidence of significant differences in gender ratio between these two groups, X2(1, N = 38) = .11, p = .75. The depression scores between the high BDI group (16 ± 4.49) and low BDI group (4.58 ± 2.87) were significantly different, t(36) = 9.35, p < .001, Cohen’s d = 3.03. The anxiety scores between the high BDI group (49.95 ± 7.39) and the low BDI group (32.58 ± 5.53) were also significantly different, t(36) = 8.20, p < .001, Cohen’s d = 2.66. Since no correlation was found between
P. Li et al. / Brain and Cognition 100 (2015) 15–20
anxiety scores and other variables (except depression, r = .76, p < .001), anxiety was not investigated further in this study. As paid volunteers, these thirty-eight students participated in the final two ERP experiments. All participants had normal or corrected-to normal vision, were right-handed and had no neurological or psychological disorders. This study was approved by the local ethics committee, and all subjects gave written informed consent in each experiment. 2.2. Experimental procedure In the first experiment (Exp. 1), participants were told to perform a modified time estimation task similar to that used by Miltner et al. (1997), but one that included neutral feedback. At the beginning of each trial, they heard a 1500 Hz sound as a cue that lasted for 50 ms. Then they estimated one second from the sound cue and pressed a key when they believed one second had elapsed. After they responded, a feedback stimulus appeared on the screen that told them whether their estimation was correct or wrong. There were three types of feedback stimuli in the present experiment. A circle with a check mark inside meant that their time estimation was correct. A circle with a cross mark inside meant that their time estimation was wrong. A circle with nothing inside was neutral feedback that provided no information about their action. The screen was blank between each pair of trials, and this ISI lasted randomly either 1400 ms, 1500 ms or 1600 ms. Participants were told that if their reaction time was within the time window from 900 ms to 1100 ms, they would get positive feedback, otherwise they would get negative feedback. However, this time window narrowed by 10 ms if they responded correctly on the previous trial and increased by 10 ms if they responded incorrectly on the previous trial. There were 300 trials in total and 100 trials were set to have neutral feedback. In the second experiment (Exp. 2.), participants conducted the same time estimation task as Exp. 1 but with real feedback, i.e., they got correct feedback if their estimation was within 900–1100 ms and incorrect feedback if their estimation was outside that time window. The rest of the design was kept the same as Exp. 1. There were 300 trials in total and 100 trials had neutral feedback. 2.3. Data acquisition Brain electrical activity was recorded from 64 scalp sites using tin electrodes mounted in an elastic cap (Brain Product, Munchen, Germany), with a ground electrode placed on the frontal midline and references placed on the left and right mastoids. Vertical electrooculograms (EOGs) were recorded infra-orbitally relative to the left eye. The impedances of all the electrodes were maintained below 10 kX. The EEG and EOG were filtered using a 0.05– 100 Hz bandpass and continuously sampled at 500 Hz/channel for offline analysis. Ocular artifacts were corrected using the eye movement correction algorithm described by Gratton, Coles, and Donchin (1983). Trials with EOG artifacts (mean EOG voltage exceeding 80 lV) and peak-to-peak deflection and those contaminated with artifacts due to amplifier clipping exceeding 100 lV were excluded from averaging.
17
see Holroyd & Krigolson, 2007), FRN amplitude was calculated for each feedback condition (positive, negative, neutral) using a base -to- peak algorithm according to previous studies (Hajcak et al., 2006; Holroyd, Hajcak, & Larsen, 2006; Holroyd, Nieuwenhuis, Yeung, & Cohen, 2003; Holroyd, Pakzad-Vaezi, & Krigolson, 2008; Pfabigan, Alexopoulos, Bauer, & Sailer, 2011). First, the most positive voltage before the FRN peak was taken as the base of the FRN. Then, the sample with the most negative value within a time window (200–400 ms) after feedback presentation was taken as the peak. FRN amplitude was calculated as the difference between these bases and peak amplitudes. The base to peak method is limited to measure a clear peak on the individual waveforms, thus we made the algorithm assigning 0 mV where no negative deflection was detected (c.f. Holroyd et al., 2006). All the ERP data was chosen from the FCz channel where the FRN peaks. Results were analyzed with repeated-measures analyses of variance (ANOVA), and the Greenhouse–Geisser correction was applied where appropriate. 3. Results 3.1. Behavioral data In Exp. 1, Participants received 33.3% neutral feedback, 31.3% positive feedback and 35.3% negative feedback in total. The accuracy for each participant was calculated as their real performance, i.e. how many times they made correct estimation within the time window (900–1100 ms). A two-tailed2 t-test showed that the Accuracy of the high BDI group (138 ± 16) was not significantly different from that of the low BDI group (144 ± 13), t(36) = 1.21, p = .23, Cohen’s d = 0.41. To measure the behavioral adjustment associated with each kind of feedback, a two-way ANOVA was conducted on the average absolute-change in RT (DRT) between the N trial and N + 1 trial following three types of feedback separately, with feedback valence as the independent variable and depression as the between-subject variable. Results revealed a significant main effect of feedback, F(2, 72) = 90.02, p < .001, g2 = 0.71. T-test pairwise comparisons indicated that the DRT following negative feedback (222 ± 11 ms) was significantly longer than the DRT following neutral feedback (186 ± 12 ms) and the DRT following neutral feedback was significantly longer than the DRT following positive feedback (149 ± 10 ms, all p < .001). The DRT in the two groups did not differ significantly (F(1, 36) = 1.09, p = .30) and the interaction between depression and valence did not reach significance, F (2, 72) < 1, p = .61. In Exp. 2, participants received real feedback while one-third of the total feedback was set to be neutral feedback. A t-test showed that there was no significant difference between Accuracy in the high BDI group (85 ± 46) and that in the low BDI group (104 ± 48), t(36) = 1.25, p = .22, Cohen’s d = 0.4. Given that the correct time-window was not adjusted trial by trial according to participants’ performance, some participants rarely got correct feedback in Exp. 2. The large variation of individual differences led to very different probability of feedback types, thus it was not possible to analyze the ERP data in Exp. 2.
2.4. Data analysis
3.2. FRN results
In Exp. 1, the EEGs elicited in response to the feedback were segmented (800 ms epochs from 200 ms to 600 ms) with a 200 ms baseline correction. Since there was a possible component overlap between the FRN and other components (such as P3001,
As shown in Fig. 1, a clear FRN component was observed in frontal-central sites (FCz). The scalp distributions of difference waves in the four conditions are shown in Fig. 2. A two-way ANOVA analysis was carried out on the FRN amplitude at the electrode FCz with feedback valence (positive, negative & neutral) as
1 P300 component was defined as the most positive peak within 300–600 ms timewindow at Pz. Please refer to supplementary material for the P300 results.
2
All reported pairwise t-test comparisons were two tailed.
18
P. Li et al. / Brain and Cognition 100 (2015) 15–20
4. Discussion
Fig. 1. Grand average in three conditions of two groups at FCz. ‘‘Neutral_Low, Negative_Low and Positive_Low” indicate the neutral, negative and positive feedback condition of Low BDI group, respectively; ‘‘Neutral_High, Negative_High and Positive_High” mean the neutral, negative and positive feedback condition of high BDI group, respectively.
Fig. 2. The histograms represent the base-to-peak FRN amplitudes at channel FCz following positive, negative and neutral feedback in high BDI group (High_BDI) and low BDI group (Low_BDI) respectively. The error bars represent standard error of the mean.
an independent variable and with depression (high BDI & low BDI) as a between-subject variable. The results showed that the main effect of valence was significant, F(1.67, 60.2) = 26.93, p < .001, g2 = 0.43. The follow-up pairwise t-test showed that the FRN amplitude in the negative condition ( 3.27 lV) was significantly larger than that of the FRN in the positive condition ( 1.6 lV), p < .001 and the FRN amplitude following neutral feedback ( 3.78 lV) was also significantly larger than the FRN amplitude in the positive condition, p < .001. However, there was no significant difference between the FRN amplitude in the negative and neutral conditions, p = .07. Moreover, the interaction between valence and depression also reached significance, F(2, 72) = 4.02, p < .03, g2 = 0.1. Inspection of the simple main effects of the interaction showed that the main effect of valence was significant in the low BDI group, F(2, 35) = 16.46, p < .001. The pairwise comparison t tests revealed that the FRN amplitude in the positive condition ( 1.14 lV) was significantly smaller than that in the negative condition ( 2.95 lV, p < .001) and the neutral condition ( 4.15 lV, p < .001). Moreover, the FRN amplitude in the neutral condition was significantly larger than the FRN in the negative condition (p < .005). The main effect of valence was also significant in the high BDI group, F(2, 35) = 7.72, p < .003. The pairwise comparisons revealed that the FRN amplitude in the positive condition ( 2.06 lV) was significantly different from that in the negative condition ( 3.59 lV, p < .001) and neutral condition ( 3.42 lV, p < .02), but there was no significant difference between the FRN in the negative and neutral conditions in the low BDI group (p = .67). The main effect of the between-subject variable was not significant, F(1, 36) < 1.
The present study investigated whether mild, non-clinical depression influences the FRN component elicited by negative, neutral and positive feedback in a time estimation task. Consistent with some previous studies (Gu et al., 2010; Hirsh & Inzlicht, 2008; Müller et al., 2005), a larger FRN effect was observed with neutral feedback than with negative feedback in the low BDI group. However, in the high BDI group, no difference in the size of the FRN component with negative and neutral feedback was observed. Reduced sensitivity to non-reward versus reward feedback has been associated with depression (Foti & Hajcak, 2009). However, in the present study, impaired processing of neutral feedback was observed in high BDI group, without reduced accuracy. These data raise the question of how depression influences performance and whether neutral stimuli can be distinguished from negative feedback by a reward-processing system. There is still disagreement as to whether depression increases or decreases the negative-FRN (Foti & Hajcak, 2009; Mies et al., 2011; Ruchsow et al., 2004; Santesso et al., 2008; Tucker et al., 2003). In the present study, in a non-clinical population, both groups showed equal sensitivity to negative feedback. The lack of group difference in performance monitoring has also been observed in other studies (Schrijvers et al., 2008, 2009). Reduced sensitivity to negative feedback may only occur in conjunction with severe levels of depression. Difference in the FRN effect for negative feedback has been found between severe and moderate depression (Tucker et al., 2003) and could account for the difference in findings from studies that test participants with different levels of severity of depression (Foti & Hajcak, 2009; Holmes & Pizzagalli, 2008; Olvet, Klein, & Hajcak, 2010; Ruchsow et al., 2004; Santesso et al., 2008). Additionally, in the present study, the introduction of neutral feedback might have created ambiguity in the learning context, although such ambiguity should have been equal for both groups. If one classified the neutral feedback as invalid feedback, the present findings contrast with research by Mies and colleagues, who found that clinical MDD led to a larger FRN effect for both invalid negative and invalid positive feedback (Mies et al., 2011). There are several possible reasons for the difference between our findings and theirs. Firstly, our participants were recruited from a non-clinical population, so that even those with high BDI scores were only mildly depressed. Secondly, we utilized simple feedback stimuli compared to Mies et al’s relatively complicated facial feedback, and lastly, neutral feedback in the present study contained no clear valence information at all, while the invalid feedback in Mies et al.’s study still informed subjects whether they could get a reward or not. Our results suggest that, at least in mildly depressed participants, the processing of both neutral feedback without valid reward information and negative feedback were influenced by depression. The discrepancy between the negative-FRN and neutral-FRN effects in the low BDI group compared with the high BDI group in the present study suggests that the FRN phenomenon is more complex than merely a binary pattern that indexes the reward processing system (see Hajcak et al., 2006; Holroyd et al., 2006; Yeung & Sanfey, 2004). Holroyd and colleagues proposed that the ACC utilizes dopamine reward prediction error signals for adaptive decision-making, especially when the optimal behavior is learnable (Holroyd et al., 2009). Thus neutral feedback that has no learning significance might be a special case for a reinforcement-based learning system compared with feedback that conveys learning information. In fact, the current view of a binary reward evaluation system was largely based on widely accepted knowledge of phasic
P. Li et al. / Brain and Cognition 100 (2015) 15–20
dopamine reward prediction error signals in primate studies (e.g. Schultz, 2013; Schultz, Dayan, & Montague, 1997). Despite the fact that the neutral feedback provided as much counter-reward information as negative feedback, it has been suggested that neutral feedback has less motivational significance compared with informative feedback (Bromberg-Martin, Matsumoto, & Hikosaka, 2010). The results of our previous EEG study would support such a hypothesis, since we found that different frequency bands, i.e., theta and beta-gamma, may correspond to reward prediction error signals and motivational significance for future action adjustment separately (Li et al., submitted for publication). Consistent with some previous studies (Gu et al., 2010; Hirsh & Inzlicht, 2008; Müller et al., 2005), the FRN effect observed here did not support a binary view of the reinforcement learning process. We propose that the reason for these findings, which are inconsistent with a binary interpretation, may be the fact that the feedback conveys not only the reward prediction error signal but also the motivational significance for future action modulation. These signals might be varied between different subjects as previous study showed (Gu et al., 2010; Hirsh & Inzlicht, 2008). We argue that it is not only possible but also important for organisms to distinguish ambiguous stimuli from informative outcomes (positive and negative feedback) because only the informative outcomes contain cognitive rewards that might benefit future action. A recent primate study has also shown that the dopaminergic reward system of monkeys not only processes primitive rewards but also cognitive rewards (Bromberg-Martin & Hikosaka, 2009). Therefore, we speculate that cognitive rewards might also be processed by human brains in such a trial-anderror learning task. Together with our previous work (Li et al., submitted for publication), the present study showed that neutral feedback could be distinguished from negative feedback in the medial frontal cortex of low BDI group, but that above a certain depression threshold within the non-clinical population, this distinction is no longer observable. Thus, the present study suggests that the ability to distinguish neutral from negative feedback has been affected by depression, and that there are preliminary effects of depressive tendencies on feedback processing in the general population. This hypothesis was also supported by the behavioral data that showed how participants modulated their response following external feedback. In line with the prediction of reinforcement learning theory (Holroyd & Coles, 2002) and our previous studies (Li et al., 2013), participants adjusted their responses moderately in the neutral condition. This suggests that participants may lack a clear strategy for action adjustment when confronted with neutral information. Regardless of whether we provided random feedback in Exp. 1 or real feedback in Exp. 2, there was no group difference in RT adjustment and accuracy between high BDI and low BDI participants. This null result demonstrated that participants’ time-estimation ability was stable across tasks. Despite the fact that no behavioral differences were found between the two groups, the FRN effect differed. Indeed, the present result dovetails with several previous ERP studies which did not find behavioral differences but observed notably different brain activity between depressive groups and healthy controls (Rose, Simonotto, & Ebmeier, 2006; Santesso et al., 2008). These findings may suggest that mild depression only leads to a moderate degree of biased cognitive processing of reward information. ERP data may be more sensitive to mildly depressive tendencies than behavioral performance. If these results should be confirmed by further evidence, the Neutral_FRN effect could be a helpful index for investigating cognitive bias in depression in future studies. In conclusion, the present study showed that subjects with mild depression have reduced sensitivity to neutral feedback compared
19
to negative feedback. In contrast, the neutral feedback elicited a larger FRN effect that negative feedback in a control group without depressive tendencies. We suggest that differential processing between neutral and negative feedback may play an important role for future action modulation. Acknowledgments This research was supported by the National Natural Science Foundation of China (NSFC31300872, 31271088), and MOE (Ministry of Education in China) Project of Humanities and Social Sciences (13YJC190013) and the National Student Innovation Training Program (201410165020). We are grateful to Qinglin Zeng, Tingting Ren and Ying Wu for their help with data collection. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.bandc.2015.08. 004. References Bromberg-Martin, E. S., & Hikosaka, O. (2009). Midbrain dopamine neurons signal preference for advance information about upcoming rewards. Neuron, 63(1), 119–126. Bromberg-Martin, E. S., Matsumoto, M., & Hikosaka, O. (2010). Dopamine in motivational control: Rewarding, aversive, and alerting. Neuron, 68(5), 815–834. Cohen, M. X., Wilmes, K. A., & van de Vijver, I. (2011). Cortical electrophysiological network dynamics of feedback learning. Trends in Cognitive Sciences, 15(12), 558–566. Donkers, F. C. L., Nieuwenhuis, S., & Van Boxtel, G. J. M. (2005). Mediofrontal negativities to averted gains and losses in the absence of responding. Cognitive Brain Research, 25, 777–787. Foti, D., & Hajcak, G. (2009). Depression and reduced sensitivity to non-rewards versus rewards: Evidence from event-related potentials. Biological Psychology, 81, 1–8. Gratton, G., Coles, M. G., & Donchin, E. (1983). A new method for off-line removal of ocular artifact. Electroencephalography and Clinical Neurophysiology, 55(4), 468–484. Gu, R., Ge, Y., Jiang, Y., & Luo, Y. J. (2010). Anxiety and outcome evaluation: The good, the bad and the ambiguous. Biological Psychology, 85(2), 200–206. Hajcak, G., Moser, J. S., Holroyd, C. B., & Simons, R. F. (2006). The feedback-related negativity reflects the binary evaluation of good versus bad outcomes. Biological Psychology, 71(2), 148–154. Hauser, T. U., Iannaccone, R., Stämpfli, P., Drechsler, R., Brandeis, D., Walitza, S., et al. (2014). The feedback-related negativity (FRN) revisited: New insights into the localization, meaning and network organization. Neuroimage, 1(84), 159–168. Hautzinger, M., Bailer, M., & Worall, H. (1994). The Beck Depression Inventory (BDI) [in German]. Bern: Huber. Hirsh, J. B., & Inzlicht, M. (2008). The devil you know neuroticism predicts neural response to uncertainty. Psychological Science, 19(10), 962–967. Holmes, A. J., & Pizzagalli, D. A. (2008). Spatiotemporal dynamics of error processing dysfunctions in major depressive disorder. Archives of General Psychiatry, 65(2), 179–188. Holroyd, C. B., & Coles, M. G. (2002). The neural basis of human error processing: Reinforcement learning, dopamine, and the error-related negativity. Psychological Review, 109(4), 679–709. Holroyd, C. B., Hajcak, G., & Larsen, J. T. (2006). The good, the bad and the neutral: Electrophysiological responses to feedback stimuli. Brain Research, 1105(1), 93–101. Holroyd, C. B., & Krigolson, O. E. (2007). Reward prediction error signals associated with a modified time estimation task. Psychophysiology, 44(6), 913–917. Holroyd, C. B., Krigolson, O. E., Baker, R., Lee, S., & Gibson, J. (2009). When is an error not a prediction error? An electrophysiological investigation. Cognitive, Affective, & Behavioral Neuroscience, 9(1), 59–70. Holroyd, C. B., Nieuwenhuis, S., Yeung, N., & Cohen, J. D. (2003). Errors in reward prediction are reflected in the event-related brain potential. Neuroreport, 14, 2481–2484. http://dx.doi.org/10.1097/00001756-200312190-00037. Holroyd, C. B., Pakzad-Vaezi, K. L., & Krigolson, O. E. (2008). The feedback correctrelated positivity: Sensitivity of the event-related brain potential to unexpected positive feedback. Psychophysiology, 45, 688–697. http://dx.doi.org/10.1111/ j.1469-8986.2008.00668.x. Holtzheimer, P. E., & Mayberg, H. S. (2011). Stuck in a rut: Rethinking depression and its treatment. Trends in Neurosciences, 34(1), 1–9. Krigolson, O. E., Pierce, L. J., Holroyd, C. B., & Tanaka, J. W. (2009). Learning to become an expert: Reinforcement learning and the acquisition of perceptual expertise. Journal of Cognitive Neuroscience, 21(9), 1833–1840.
20
P. Li et al. / Brain and Cognition 100 (2015) 15–20
Li, P., Baker, T., Warren, C., & Li, H. (submitted for publication). Oscillatory profiles of positive, negative and neutral feedback stimuli during a time-estimation task. Li, P., Han, C., Lei, Y., Holroyd, C. B., & Li, H. (2011). Responsibility modulates neural mechanisms of outcome processing: An ERP study. Psychophysiology, 48(8), 1129–1133. Li, P., Jia, S., Feng, T., Liu, Q., Suo, T., & Li, H. (2010). The influence of the diffusion of responsibility effect on outcome evaluations: Electrophysiological evidence from an ERP study. Neuroimage, 52(4), 1727–1733. Li, P., Shen, Y., Sui, X., Chen, C., Feng, T., Li, H., et al. (2013). The neural basis of responsibility attribution in decision-making. PloS One, 8(11), e80389. Li, P., Yuan, J., Jia, S., Feng, T., Chen, A., & Li, H. (2009). Feedback-related negativity effects vanished with false or monetary loss choice. Neuroreport, 20(8), 788–792. Mathews, A., & MacLeod, C. (2005). Cognitive vulnerability to emotional disorders. Annual Review of Clinical Psychology, 1, 167–195. Mies, G. W., vander Veen, F. M., Tulen, J. H., Bir kenhäger, T. K., Hengeveld, M. W., & van der Molen, M. W. (2011). Drug-free patients with major depression show an increased electrophysiological response to valid and invalid feedback. Psychological Medicine, 1–11. Miller, J. D., & Pilkonis, P. A. (2006). Neuroticism and affective instability: The same or different? American Journal of Psychiatry, 163(5), 839–845. Miltner, W. H. R., Braun, C. H., & Coles, M. G. H. (1997). Event-related brain potentials following incorrect feedback in a time-estimation task: Evidence for a ‘‘generic” neural system for error detection. Journal of Cognitive Neuroscience, 9, 788–798. Müller, S. V., Möller, J., Rodriguez-Fornells, A., & Münte, T. F. (2005). Brain potentials related to self-generated and external information used for performance monitoring. Clinical Neurophysiology, 116, 63–74. Nieuwenhuis, S., Holroyd, C. B., Mol, N., & Coles, M. G. H. (2004). Reinforcementrelated brain potentials from medial frontal cortex: Origins and functional significance. Neuroscience and Biobehavioral Reviews, 28, 441–448. Nieuwenhuis, S., Slagter, H., Alting von Geusau, N., Heslenfeld, D. J., & Holroyd, C. B. (2005). Knowing good from bad: Differential activation of human cortical areas by positive and negative outcomes. European Journal of Neuroscience, 21, 3161–3168. Oliveira, F. T., McDonald, J. J., & Goodman, D. (2007). Performance monitoring in the anterior cingulate is not all error related: Expectancy deviation and the representation of action-outcome associations. Journal Of Cognitive Neuroscience, 19(12), 1994–2004. Olvet, D. M., Klein, D. N., & Hajcak, G. (2010). Depression symptom severity and error-related brain activity. Psychiatry Research, 179(1), 30–37. Pfabigan, D. M., Alexopoulos, J., Bauer, H., & Sailer, U. (2011). Manipulation of feedback expectancy and valence induces negative and positive reward prediction error signals manifest in event-related brain potentials. Psychophysiology, 48, 656–664.
Rose, E. J., Simonotto, E., & Ebmeier, K. P. (2006). Limbic over-activity in depression during preserved performance on the n-back task. Neuroimage, 29, 203–215. Ruchsow, M., Herrnberger, B., Wiesend, C., Grön, G., Spitzer, M., & Kiefer, M. (2004). The effect of erroneous responses on response monitoring in patients with major depressive disorder: A study with event-related potentials. Psychophysiology, 41(6), 833–840. Sambrook, T. D., & Goslin, J. (2015). A neural reward prediction error revealed by a meta-analysis of ERPs using great grand averages. Psychological Bulletin, 141(1), 213–235. Santesso, D. L., Steele, K. T., Bogdan, R., Holmes, A. J., Deveney, C. M., Meites, T. M., et al. (2008). Enhanced negative feedback responses in remitted depression. Neuroreport, 19, 1045–1048. Schrijvers, D., de Bruijn, E. R., Maas, Y., De Grave, C., Sabbe, B. G., & Hulstijn, W. (2008). Action monitoring in major depressive disorder with psychomotor retardation. Cortex, 44(5), 569–579. Schrijvers, D., De Bruijn, E. R., Maas, Y. J., Vancoillie, P., Hulstijn, W., & Sabbe, B. G. (2009). Action monitoring and depressive symptom reduction in major depressive disorder. International Journal of Psychophysiology, 71(3), 218–224. Schultz, W. (2013). Updating dopamine reward signals. Current Opinion in Neurobiology, 23(2), 229–238. Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural substrate of prediction and reward. Science, 275(5306), 1593–1599. Spielberger, C. D., Gorssuch, R. L., Lushene, P. R., Vagg, P. R., & Jacobs, G. A. (1983). Manual for the state-trait anxiety inventory. Palo Alto, CA: Consulting Psychologists Press. Talmi, D., Atkinson, R., & El-Deredy, W. (2013). The feedback-related negativity signals salience prediction errors, not reward prediction errors. Journal of Neuroscience, 33(19), 8264–8269. Tucker, D. M., Luu, P., Frishkoff, G., Quiring, J., & Poulsen, C. (2003). Frontolimbic response to negative feedback in clinical depression. Journal of Abnormal Psychology, 112(4), 667–678. Ullsperger, M., Fischer, A. G., Nigbur, R., & Endrass, T. (2014). Neural mechanisms and temporal dynamics of performance monitoring. Trends in Cognitive Sciences, 18(5), 259–267. Van Noordt, S. J., & Segalowitz, S. J. (2012). Performance monitoring and the medial prefrontal cortex: A review of individual differences and context effects as a window on self-regulation. Frontiers in Human Neuroscience, 6. Walsh, M. M., & Anderson, J. R. (2012). Learning from experience: Event-related potential correlates of reward processing, neural adaptation, and behavioral choice. Neuroscience & Biobehavioral Reviews, 36(8), 1870–1884. Warren, C. M., Hyman, J. M., Seamans, J. K., & Holroyd, C. B. (2014). Feedback-related negativity observed in rodent anterior cingulate cortex. Journal of PhysiologyParis (September 16. pii: S0928-4257(14)00042-4). Yeung, N., & Sanfey, A. G. (2004). Independent coding of reward magnitude and valence in the human brain. The Journal of Neuroscience, 24(28), 6258–6264.
Contents lists available at ScienceDirect
Brain and Cognition journal homepage: www.elsevier.com/locate/b&c
Reduced sensitivity to neutral feedback versus negative feedback in subjects with mild depression: Evidence from event-related potentials study Li Peng a,b, Song Xinxin b, Wang Jing b, Zhou Xiaoran b, Li Jiayi b, Lin Fengtong b, Hu Zhonghua b, Zhang Xinxin b, Cui Hewei b, Wang Wenmiao b, Li Hong a,b,⇑, Cong Fengyu c,⇑, Debi Roberson d a
Brian Function and Psychological Science Research Center, Shenzhen University, Shenzhen 518060, China Research Center of Brain and Cognitive Neuroscience, Liaoning Normal University, Dalian 116029, China c Department of Biomedical Engineering, Faculty of Electronic Information and Electrical Engineering, Dalian University of Technology, Dalian 116029, China d Department of Psychology, University of Essex, Wivenhoe Park, Colchester, Essex CO4 3SQ, UK b
a r t i c l e
i n f o
Article history: Received 25 December 2014 Revised 4 August 2015 Accepted 5 August 2015
Keywords: Neutral feedback Depression FRN ACC Reinforcement learning
a b s t r a c t Many previous event-related potential (ERP) studies have linked the feedback related negativity (FRN) component with medial frontal cortex processing and associated this component with depression. Few if any studies have investigated the processing of neutral feedback in mildly depressive subjects in the normal population. Two experiments compared brain responses to neutral feedback with behavioral performance in mildly depressed subjects who scored highly on the Beck Depression Inventory (high BDI) and a control group with lower BDI scores (low BDI). In the first study, the FRN component was recorded when neutral, negative or positive feedback was pseudo-randomly delivered to the two groups in a time estimation task. In the second study, real feedback was provided to the two groups in the same task in order to measure their actual accuracy of performance. The results of experiment one (Exp. 1) revealed that a larger FRN effect was elicited by neutral feedback than by negative feedback in the low BDI group, but no significant difference was found between neutral condition and negative condition in the High BDI group. The present findings demonstrated that depressive tendencies influence the processing of neutral feedback in medial frontal cortex. The FRN effect may work as a helpful index for investigating cognitive bias in depression in future studies. Ó 2015 Elsevier Inc. All rights reserved.
1. Introduction Our complex and dynamic living environment requires human beings to learn how to predict and evaluate the consequences of actions that lead to rewards and punishments. For this reason accurate processing of external feedback is essential if humans are to optimize their behavior (Miltner, Braun, & Coles, 1997). Much research has focused on the brain activity associated with the evaluation of outcomes by examining the timing of electroencephalography (EEG) responses (Holroyd & Coles, 2002; Li, Han, Lei, Holroyd, & Li, 2011; Li et al., 2010; Ullsperger, Fischer, Nigbur, & Endrass, 2014; Walsh & Anderson, 2012). Investigations into feedback evaluation have consistently found negative deflection in the Event-Related Potential (ERP) following the ⇑ Corresponding authors at: No. 3688, Nanhai Road, Nanshan District, Shenzhen 518060, China (H. Li). LingGong Road #2, Dalian 11602, China (F. Cong) E-mail addresses: [email protected] (H. Li), [email protected] (F. Cong). http://dx.doi.org/10.1016/j.bandc.2015.08.004 0278-2626/Ó 2015 Elsevier Inc. All rights reserved.
presentation of negative feedback, an effect that has been termed Feedback Related Negativity (FRN, Cohen, Wilmes, & van de Vijver, 2011; Holroyd & Coles, 2002; Li et al., 2009; Miltner et al., 1997). This time-domain trial-averaged FRN component peaks at 250–300 ms after feedback and has a fronto-central distribution. Convergent findings from multiple methodologies suggest that the FRN is probably generated in the anterior cingulate cortex (ACC) in medial frontal cortex (Hauser et al., 2014; Warren, Hyman, Seamans, & Holroyd, 2014; but see Nieuwenhuis, Slagter, Alting von Geusau, Heslenfeld, & Holroyd, 2005). Theories concerning the functional significance of FRN have been constantly updated in the last two decades. The current most influential theoretical account of FRN comes from Holroyd and Coles’s ‘‘reinforcement learning theory of the error-related negativity (RL-ERN theory)”. According to this theory, the FRN amplitude reflects reward prediction error, i.e. a signed value corresponding to the difference between the obtained reward and the expected reward (Holroyd & Coles, 2002; Nieuwenhuis, Holroyd, Mol, &
16
P. Li et al. / Brain and Cognition 100 (2015) 15–20
Coles, 2004). This explanation has been supported by abundant evidence in FRN studies (e.g. Holroyd, Krigolson, Baker, Lee, & Gibson, 2009; Krigolson, Pierce, Holroyd, & Tanaka, 2009; Walsh & Anderson, 2012). However, several studies in the last ten years have provided evidence that the FRN conveyed an unsigned prediction error or ‘‘salience” encoding signal (Donkers, Nieuwenhuis, & Van Boxtel, 2005; Hauser et al., 2014; Oliveira, McDonald, & Goodman, 2007; Talmi, Atkinson, & El-Deredy, 2013). More recently, Sambrook and Goslin (2015) utilized the ‘‘great grand averages” approach in a meta-analysis study and showed strong effects of magnitude and likelihood on the FRN, which provided new evidence to support the RL_ERN theory. Based on the predictions of RL_ERN theory, researchers proposed that the feedback processing system, putatively indexed by the FRN component, reveals a binary way to evaluate current outcomes, i.e. whether the current outcome is worse than expected or not (e.g. Hajcak, Moser, Holroyd, & Simons, 2006; Yeung & Sanfey, 2004). However, several studies have explored the phenomenon by introducing neutral feedback and showed that neutral feedback elicited a relatively larger FRN than negative feedback (Gu, Ge, Jiang, & Luo, 2010; Hirsh & Inzlicht, 2008; Li, Baker, Warren, Li, submitted for publication; Müller, Möller, Rodriguez-Fornells, & Münte, 2005). Müller et al. (2005) first found that the FRN occurred earlier and had a higher peak in the neutral condition than in the negative feedback condition. This discrepancy between neutral and negative feedback was also observed in two studies with negative affective states (Gu et al., 2010; Hirsh & Inzlicht, 2008). To systematically explore the effect of neutral feedback on the amplitude of the FRN, Holroyd and his colleagues (2006) conducted five experiments, which included an intermediate reward condition in the first three experiments and a neutral feedback condition in the later two experiments. Their results were consistent with the now widely accepted proposals that the evaluation system classifies outcomes into two categories: the satisfied outcome and unsatisfied outcome. Hence, it remains unclear how the evaluation system works when it comes to the neutral feedback. Thus investigating the FRN effect elicited by neutral feedback may also contribute to the above-mentioned arguments concerning the theoretical account of the FRN phenomenon because neutral feedback is a special case in terms of valence and magnitude. According to the RL_ERN theory, the FRN manifests the dopamine signal transferred from the basal ganglia to the medial prefrontal cortex, more specifically, the ACC (Holroyd & Coles, 2002). So far, accumulating evidence shows that the patterns of activation in the ACC during performance monitoring vary as a function of individual differences in personality (Van Noordt & Segalowitz, 2012). Major depressive disorder (MDD) is a highly prevalent multifactorial psychiatric disorder and has been characterized as an abnormal tendency to engage in negative mood states, together with difficulty in disengaging from negative mood states (Holtzheimer & Mayberg, 2011). The impairment of ACC function by depression has been demonstrated in a number of ERP studies reporting that participants with severe MDD showed hyperactivation to internal or external error compared with the control group (Holmes & Pizzagalli, 2008; Santesso et al., 2008; but not in Foti & Hajcak, 2009; Ruchsow et al., 2004). However, it is still unclear whether or not depression influences the reward processing of neutral feedback. In fact, two FRN studies focused on individual differences in personality have found larger FRN effects following neutral feedback in subjects scoring high on neuroticism and high on trait anxiety scales (Gu et al., 2010; Hirsh & Inzlicht, 2008). Given that depression shares the same underlying biases of information processing with trait anxiety (Mathews & MacLeod, 2005) and that such biases might be predicted by neuroticism (Miller & Pilkonis, 2006), it is plausible that depression may also affect the FRN effect with neutral feedback.
To our knowledge, no study to date has focused on the FRN elicited by neutral feedback in depression. The most relevant study came from Mies et al. (2011), in which they investigated both behavioral and electrophysiological responses to feedback validity in non-medicated depressed patients. Their results found that nonmedicated depressed in-patients showed a more pronounced FRN amplitude regardless of feedback validity. It is worth noting that the invalid feedback still contained valence information in Mies et al.’s study, and thus it differed from what we call ‘‘neutral feedback” here. Moreover, in previous studies the severity of depressive symptoms or neuroticism drives different neural responses toward reward (or correct) and non-reward (or error) feedback (Foti & Hajcak, 2009; Hirsh & Inzlicht, 2008; Tucker, Luu, Frishkoff, Quiring, & Poulsen, 2003). Tucker et al. (2003) found enhanced FRN responses in moderately depressed, but not in more severely depressed patients. Foti and Hajcak (2009) showed an enhancement of the FRN to non-rewards relative to rewards that was inversely related to depression. The FRN effect of neutral feedback has also been found to vary linearly as neuroticism scores change (Hirsh & Inzlicht, 2008). The current studies seek to extend this body of research by linking the neutral FRN phenomenon with mild depression in undergraduate students to complement the studies of clinical MDD (e.g. Mies et al., 2011). Using a time estimation task, the present study investigated the outcome evaluation of neutral feedback in two groups varying in non-clinical depression. We also compared the behavioral performance of two groups in a subsequent time estimation task with real feedback. The aim of this study was to investigate whether depression attenuates or increases the performance-monitoring processing of neutral and negative feedback in medial prefrontal cortex. Given that depression is frequently associated with negatively biased information processing (cf. Mathews & MacLeod, 2005), the depressive group tested here was more likely to treat neutral feedback as negative feedback. Therefore, we hypothesized that neutral feedback would elicit comparable FRN amplitudes to negative feedback in the depressive group. We also predicted that distinctly different FRNs would be observed between neutral and negative feedback in the control group as it was in a previous study with random samples (Li et al., submitted for publication). In addition, we intended to compare the behavioral pattern of a depressive group and a control group in Exp. 2, in which real feedback was provided. 2. Methods 2.1. Participants Participants were recruited from a large group (769) of students in two universities. 38 Participants aged 18–25 years were selected for the present study. All of them were assessed for depressive tendencies using the Beck Depression Inventory (BDI; Hautzinger, Bailer, & Worall, 1994) and were measured for trait anxiety using the trait section of the State Trait Anxiety Inventory (STAI; Spielberger, Gorssuch, Lushene, Vagg, & Jacobs, 1983) one week before the experiment was conducted. 19 participants (9 females, mean age 20.6) with a BDI score higher than 10 were identified as the high BDI group and another 19 participants (10 females, average age 20.7) with BDI scores lower than 10 were selected as the low BDI group. A chi-square analysis provided no evidence of significant differences in gender ratio between these two groups, X2(1, N = 38) = .11, p = .75. The depression scores between the high BDI group (16 ± 4.49) and low BDI group (4.58 ± 2.87) were significantly different, t(36) = 9.35, p < .001, Cohen’s d = 3.03. The anxiety scores between the high BDI group (49.95 ± 7.39) and the low BDI group (32.58 ± 5.53) were also significantly different, t(36) = 8.20, p < .001, Cohen’s d = 2.66. Since no correlation was found between
P. Li et al. / Brain and Cognition 100 (2015) 15–20
anxiety scores and other variables (except depression, r = .76, p < .001), anxiety was not investigated further in this study. As paid volunteers, these thirty-eight students participated in the final two ERP experiments. All participants had normal or corrected-to normal vision, were right-handed and had no neurological or psychological disorders. This study was approved by the local ethics committee, and all subjects gave written informed consent in each experiment. 2.2. Experimental procedure In the first experiment (Exp. 1), participants were told to perform a modified time estimation task similar to that used by Miltner et al. (1997), but one that included neutral feedback. At the beginning of each trial, they heard a 1500 Hz sound as a cue that lasted for 50 ms. Then they estimated one second from the sound cue and pressed a key when they believed one second had elapsed. After they responded, a feedback stimulus appeared on the screen that told them whether their estimation was correct or wrong. There were three types of feedback stimuli in the present experiment. A circle with a check mark inside meant that their time estimation was correct. A circle with a cross mark inside meant that their time estimation was wrong. A circle with nothing inside was neutral feedback that provided no information about their action. The screen was blank between each pair of trials, and this ISI lasted randomly either 1400 ms, 1500 ms or 1600 ms. Participants were told that if their reaction time was within the time window from 900 ms to 1100 ms, they would get positive feedback, otherwise they would get negative feedback. However, this time window narrowed by 10 ms if they responded correctly on the previous trial and increased by 10 ms if they responded incorrectly on the previous trial. There were 300 trials in total and 100 trials were set to have neutral feedback. In the second experiment (Exp. 2.), participants conducted the same time estimation task as Exp. 1 but with real feedback, i.e., they got correct feedback if their estimation was within 900–1100 ms and incorrect feedback if their estimation was outside that time window. The rest of the design was kept the same as Exp. 1. There were 300 trials in total and 100 trials had neutral feedback. 2.3. Data acquisition Brain electrical activity was recorded from 64 scalp sites using tin electrodes mounted in an elastic cap (Brain Product, Munchen, Germany), with a ground electrode placed on the frontal midline and references placed on the left and right mastoids. Vertical electrooculograms (EOGs) were recorded infra-orbitally relative to the left eye. The impedances of all the electrodes were maintained below 10 kX. The EEG and EOG were filtered using a 0.05– 100 Hz bandpass and continuously sampled at 500 Hz/channel for offline analysis. Ocular artifacts were corrected using the eye movement correction algorithm described by Gratton, Coles, and Donchin (1983). Trials with EOG artifacts (mean EOG voltage exceeding 80 lV) and peak-to-peak deflection and those contaminated with artifacts due to amplifier clipping exceeding 100 lV were excluded from averaging.
17
see Holroyd & Krigolson, 2007), FRN amplitude was calculated for each feedback condition (positive, negative, neutral) using a base -to- peak algorithm according to previous studies (Hajcak et al., 2006; Holroyd, Hajcak, & Larsen, 2006; Holroyd, Nieuwenhuis, Yeung, & Cohen, 2003; Holroyd, Pakzad-Vaezi, & Krigolson, 2008; Pfabigan, Alexopoulos, Bauer, & Sailer, 2011). First, the most positive voltage before the FRN peak was taken as the base of the FRN. Then, the sample with the most negative value within a time window (200–400 ms) after feedback presentation was taken as the peak. FRN amplitude was calculated as the difference between these bases and peak amplitudes. The base to peak method is limited to measure a clear peak on the individual waveforms, thus we made the algorithm assigning 0 mV where no negative deflection was detected (c.f. Holroyd et al., 2006). All the ERP data was chosen from the FCz channel where the FRN peaks. Results were analyzed with repeated-measures analyses of variance (ANOVA), and the Greenhouse–Geisser correction was applied where appropriate. 3. Results 3.1. Behavioral data In Exp. 1, Participants received 33.3% neutral feedback, 31.3% positive feedback and 35.3% negative feedback in total. The accuracy for each participant was calculated as their real performance, i.e. how many times they made correct estimation within the time window (900–1100 ms). A two-tailed2 t-test showed that the Accuracy of the high BDI group (138 ± 16) was not significantly different from that of the low BDI group (144 ± 13), t(36) = 1.21, p = .23, Cohen’s d = 0.41. To measure the behavioral adjustment associated with each kind of feedback, a two-way ANOVA was conducted on the average absolute-change in RT (DRT) between the N trial and N + 1 trial following three types of feedback separately, with feedback valence as the independent variable and depression as the between-subject variable. Results revealed a significant main effect of feedback, F(2, 72) = 90.02, p < .001, g2 = 0.71. T-test pairwise comparisons indicated that the DRT following negative feedback (222 ± 11 ms) was significantly longer than the DRT following neutral feedback (186 ± 12 ms) and the DRT following neutral feedback was significantly longer than the DRT following positive feedback (149 ± 10 ms, all p < .001). The DRT in the two groups did not differ significantly (F(1, 36) = 1.09, p = .30) and the interaction between depression and valence did not reach significance, F (2, 72) < 1, p = .61. In Exp. 2, participants received real feedback while one-third of the total feedback was set to be neutral feedback. A t-test showed that there was no significant difference between Accuracy in the high BDI group (85 ± 46) and that in the low BDI group (104 ± 48), t(36) = 1.25, p = .22, Cohen’s d = 0.4. Given that the correct time-window was not adjusted trial by trial according to participants’ performance, some participants rarely got correct feedback in Exp. 2. The large variation of individual differences led to very different probability of feedback types, thus it was not possible to analyze the ERP data in Exp. 2.
2.4. Data analysis
3.2. FRN results
In Exp. 1, the EEGs elicited in response to the feedback were segmented (800 ms epochs from 200 ms to 600 ms) with a 200 ms baseline correction. Since there was a possible component overlap between the FRN and other components (such as P3001,
As shown in Fig. 1, a clear FRN component was observed in frontal-central sites (FCz). The scalp distributions of difference waves in the four conditions are shown in Fig. 2. A two-way ANOVA analysis was carried out on the FRN amplitude at the electrode FCz with feedback valence (positive, negative & neutral) as
1 P300 component was defined as the most positive peak within 300–600 ms timewindow at Pz. Please refer to supplementary material for the P300 results.
2
All reported pairwise t-test comparisons were two tailed.
18
P. Li et al. / Brain and Cognition 100 (2015) 15–20
4. Discussion
Fig. 1. Grand average in three conditions of two groups at FCz. ‘‘Neutral_Low, Negative_Low and Positive_Low” indicate the neutral, negative and positive feedback condition of Low BDI group, respectively; ‘‘Neutral_High, Negative_High and Positive_High” mean the neutral, negative and positive feedback condition of high BDI group, respectively.
Fig. 2. The histograms represent the base-to-peak FRN amplitudes at channel FCz following positive, negative and neutral feedback in high BDI group (High_BDI) and low BDI group (Low_BDI) respectively. The error bars represent standard error of the mean.
an independent variable and with depression (high BDI & low BDI) as a between-subject variable. The results showed that the main effect of valence was significant, F(1.67, 60.2) = 26.93, p < .001, g2 = 0.43. The follow-up pairwise t-test showed that the FRN amplitude in the negative condition ( 3.27 lV) was significantly larger than that of the FRN in the positive condition ( 1.6 lV), p < .001 and the FRN amplitude following neutral feedback ( 3.78 lV) was also significantly larger than the FRN amplitude in the positive condition, p < .001. However, there was no significant difference between the FRN amplitude in the negative and neutral conditions, p = .07. Moreover, the interaction between valence and depression also reached significance, F(2, 72) = 4.02, p < .03, g2 = 0.1. Inspection of the simple main effects of the interaction showed that the main effect of valence was significant in the low BDI group, F(2, 35) = 16.46, p < .001. The pairwise comparison t tests revealed that the FRN amplitude in the positive condition ( 1.14 lV) was significantly smaller than that in the negative condition ( 2.95 lV, p < .001) and the neutral condition ( 4.15 lV, p < .001). Moreover, the FRN amplitude in the neutral condition was significantly larger than the FRN in the negative condition (p < .005). The main effect of valence was also significant in the high BDI group, F(2, 35) = 7.72, p < .003. The pairwise comparisons revealed that the FRN amplitude in the positive condition ( 2.06 lV) was significantly different from that in the negative condition ( 3.59 lV, p < .001) and neutral condition ( 3.42 lV, p < .02), but there was no significant difference between the FRN in the negative and neutral conditions in the low BDI group (p = .67). The main effect of the between-subject variable was not significant, F(1, 36) < 1.
The present study investigated whether mild, non-clinical depression influences the FRN component elicited by negative, neutral and positive feedback in a time estimation task. Consistent with some previous studies (Gu et al., 2010; Hirsh & Inzlicht, 2008; Müller et al., 2005), a larger FRN effect was observed with neutral feedback than with negative feedback in the low BDI group. However, in the high BDI group, no difference in the size of the FRN component with negative and neutral feedback was observed. Reduced sensitivity to non-reward versus reward feedback has been associated with depression (Foti & Hajcak, 2009). However, in the present study, impaired processing of neutral feedback was observed in high BDI group, without reduced accuracy. These data raise the question of how depression influences performance and whether neutral stimuli can be distinguished from negative feedback by a reward-processing system. There is still disagreement as to whether depression increases or decreases the negative-FRN (Foti & Hajcak, 2009; Mies et al., 2011; Ruchsow et al., 2004; Santesso et al., 2008; Tucker et al., 2003). In the present study, in a non-clinical population, both groups showed equal sensitivity to negative feedback. The lack of group difference in performance monitoring has also been observed in other studies (Schrijvers et al., 2008, 2009). Reduced sensitivity to negative feedback may only occur in conjunction with severe levels of depression. Difference in the FRN effect for negative feedback has been found between severe and moderate depression (Tucker et al., 2003) and could account for the difference in findings from studies that test participants with different levels of severity of depression (Foti & Hajcak, 2009; Holmes & Pizzagalli, 2008; Olvet, Klein, & Hajcak, 2010; Ruchsow et al., 2004; Santesso et al., 2008). Additionally, in the present study, the introduction of neutral feedback might have created ambiguity in the learning context, although such ambiguity should have been equal for both groups. If one classified the neutral feedback as invalid feedback, the present findings contrast with research by Mies and colleagues, who found that clinical MDD led to a larger FRN effect for both invalid negative and invalid positive feedback (Mies et al., 2011). There are several possible reasons for the difference between our findings and theirs. Firstly, our participants were recruited from a non-clinical population, so that even those with high BDI scores were only mildly depressed. Secondly, we utilized simple feedback stimuli compared to Mies et al’s relatively complicated facial feedback, and lastly, neutral feedback in the present study contained no clear valence information at all, while the invalid feedback in Mies et al.’s study still informed subjects whether they could get a reward or not. Our results suggest that, at least in mildly depressed participants, the processing of both neutral feedback without valid reward information and negative feedback were influenced by depression. The discrepancy between the negative-FRN and neutral-FRN effects in the low BDI group compared with the high BDI group in the present study suggests that the FRN phenomenon is more complex than merely a binary pattern that indexes the reward processing system (see Hajcak et al., 2006; Holroyd et al., 2006; Yeung & Sanfey, 2004). Holroyd and colleagues proposed that the ACC utilizes dopamine reward prediction error signals for adaptive decision-making, especially when the optimal behavior is learnable (Holroyd et al., 2009). Thus neutral feedback that has no learning significance might be a special case for a reinforcement-based learning system compared with feedback that conveys learning information. In fact, the current view of a binary reward evaluation system was largely based on widely accepted knowledge of phasic
P. Li et al. / Brain and Cognition 100 (2015) 15–20
dopamine reward prediction error signals in primate studies (e.g. Schultz, 2013; Schultz, Dayan, & Montague, 1997). Despite the fact that the neutral feedback provided as much counter-reward information as negative feedback, it has been suggested that neutral feedback has less motivational significance compared with informative feedback (Bromberg-Martin, Matsumoto, & Hikosaka, 2010). The results of our previous EEG study would support such a hypothesis, since we found that different frequency bands, i.e., theta and beta-gamma, may correspond to reward prediction error signals and motivational significance for future action adjustment separately (Li et al., submitted for publication). Consistent with some previous studies (Gu et al., 2010; Hirsh & Inzlicht, 2008; Müller et al., 2005), the FRN effect observed here did not support a binary view of the reinforcement learning process. We propose that the reason for these findings, which are inconsistent with a binary interpretation, may be the fact that the feedback conveys not only the reward prediction error signal but also the motivational significance for future action modulation. These signals might be varied between different subjects as previous study showed (Gu et al., 2010; Hirsh & Inzlicht, 2008). We argue that it is not only possible but also important for organisms to distinguish ambiguous stimuli from informative outcomes (positive and negative feedback) because only the informative outcomes contain cognitive rewards that might benefit future action. A recent primate study has also shown that the dopaminergic reward system of monkeys not only processes primitive rewards but also cognitive rewards (Bromberg-Martin & Hikosaka, 2009). Therefore, we speculate that cognitive rewards might also be processed by human brains in such a trial-anderror learning task. Together with our previous work (Li et al., submitted for publication), the present study showed that neutral feedback could be distinguished from negative feedback in the medial frontal cortex of low BDI group, but that above a certain depression threshold within the non-clinical population, this distinction is no longer observable. Thus, the present study suggests that the ability to distinguish neutral from negative feedback has been affected by depression, and that there are preliminary effects of depressive tendencies on feedback processing in the general population. This hypothesis was also supported by the behavioral data that showed how participants modulated their response following external feedback. In line with the prediction of reinforcement learning theory (Holroyd & Coles, 2002) and our previous studies (Li et al., 2013), participants adjusted their responses moderately in the neutral condition. This suggests that participants may lack a clear strategy for action adjustment when confronted with neutral information. Regardless of whether we provided random feedback in Exp. 1 or real feedback in Exp. 2, there was no group difference in RT adjustment and accuracy between high BDI and low BDI participants. This null result demonstrated that participants’ time-estimation ability was stable across tasks. Despite the fact that no behavioral differences were found between the two groups, the FRN effect differed. Indeed, the present result dovetails with several previous ERP studies which did not find behavioral differences but observed notably different brain activity between depressive groups and healthy controls (Rose, Simonotto, & Ebmeier, 2006; Santesso et al., 2008). These findings may suggest that mild depression only leads to a moderate degree of biased cognitive processing of reward information. ERP data may be more sensitive to mildly depressive tendencies than behavioral performance. If these results should be confirmed by further evidence, the Neutral_FRN effect could be a helpful index for investigating cognitive bias in depression in future studies. In conclusion, the present study showed that subjects with mild depression have reduced sensitivity to neutral feedback compared
19
to negative feedback. In contrast, the neutral feedback elicited a larger FRN effect that negative feedback in a control group without depressive tendencies. We suggest that differential processing between neutral and negative feedback may play an important role for future action modulation. Acknowledgments This research was supported by the National Natural Science Foundation of China (NSFC31300872, 31271088), and MOE (Ministry of Education in China) Project of Humanities and Social Sciences (13YJC190013) and the National Student Innovation Training Program (201410165020). We are grateful to Qinglin Zeng, Tingting Ren and Ying Wu for their help with data collection. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.bandc.2015.08. 004. References Bromberg-Martin, E. S., & Hikosaka, O. (2009). Midbrain dopamine neurons signal preference for advance information about upcoming rewards. Neuron, 63(1), 119–126. Bromberg-Martin, E. S., Matsumoto, M., & Hikosaka, O. (2010). Dopamine in motivational control: Rewarding, aversive, and alerting. Neuron, 68(5), 815–834. Cohen, M. X., Wilmes, K. A., & van de Vijver, I. (2011). Cortical electrophysiological network dynamics of feedback learning. Trends in Cognitive Sciences, 15(12), 558–566. Donkers, F. C. L., Nieuwenhuis, S., & Van Boxtel, G. J. M. (2005). Mediofrontal negativities to averted gains and losses in the absence of responding. Cognitive Brain Research, 25, 777–787. Foti, D., & Hajcak, G. (2009). Depression and reduced sensitivity to non-rewards versus rewards: Evidence from event-related potentials. Biological Psychology, 81, 1–8. Gratton, G., Coles, M. G., & Donchin, E. (1983). A new method for off-line removal of ocular artifact. Electroencephalography and Clinical Neurophysiology, 55(4), 468–484. Gu, R., Ge, Y., Jiang, Y., & Luo, Y. J. (2010). Anxiety and outcome evaluation: The good, the bad and the ambiguous. Biological Psychology, 85(2), 200–206. Hajcak, G., Moser, J. S., Holroyd, C. B., & Simons, R. F. (2006). The feedback-related negativity reflects the binary evaluation of good versus bad outcomes. Biological Psychology, 71(2), 148–154. Hauser, T. U., Iannaccone, R., Stämpfli, P., Drechsler, R., Brandeis, D., Walitza, S., et al. (2014). The feedback-related negativity (FRN) revisited: New insights into the localization, meaning and network organization. Neuroimage, 1(84), 159–168. Hautzinger, M., Bailer, M., & Worall, H. (1994). The Beck Depression Inventory (BDI) [in German]. Bern: Huber. Hirsh, J. B., & Inzlicht, M. (2008). The devil you know neuroticism predicts neural response to uncertainty. Psychological Science, 19(10), 962–967. Holmes, A. J., & Pizzagalli, D. A. (2008). Spatiotemporal dynamics of error processing dysfunctions in major depressive disorder. Archives of General Psychiatry, 65(2), 179–188. Holroyd, C. B., & Coles, M. G. (2002). The neural basis of human error processing: Reinforcement learning, dopamine, and the error-related negativity. Psychological Review, 109(4), 679–709. Holroyd, C. B., Hajcak, G., & Larsen, J. T. (2006). The good, the bad and the neutral: Electrophysiological responses to feedback stimuli. Brain Research, 1105(1), 93–101. Holroyd, C. B., & Krigolson, O. E. (2007). Reward prediction error signals associated with a modified time estimation task. Psychophysiology, 44(6), 913–917. Holroyd, C. B., Krigolson, O. E., Baker, R., Lee, S., & Gibson, J. (2009). When is an error not a prediction error? An electrophysiological investigation. Cognitive, Affective, & Behavioral Neuroscience, 9(1), 59–70. Holroyd, C. B., Nieuwenhuis, S., Yeung, N., & Cohen, J. D. (2003). Errors in reward prediction are reflected in the event-related brain potential. Neuroreport, 14, 2481–2484. http://dx.doi.org/10.1097/00001756-200312190-00037. Holroyd, C. B., Pakzad-Vaezi, K. L., & Krigolson, O. E. (2008). The feedback correctrelated positivity: Sensitivity of the event-related brain potential to unexpected positive feedback. Psychophysiology, 45, 688–697. http://dx.doi.org/10.1111/ j.1469-8986.2008.00668.x. Holtzheimer, P. E., & Mayberg, H. S. (2011). Stuck in a rut: Rethinking depression and its treatment. Trends in Neurosciences, 34(1), 1–9. Krigolson, O. E., Pierce, L. J., Holroyd, C. B., & Tanaka, J. W. (2009). Learning to become an expert: Reinforcement learning and the acquisition of perceptual expertise. Journal of Cognitive Neuroscience, 21(9), 1833–1840.
20
P. Li et al. / Brain and Cognition 100 (2015) 15–20
Li, P., Baker, T., Warren, C., & Li, H. (submitted for publication). Oscillatory profiles of positive, negative and neutral feedback stimuli during a time-estimation task. Li, P., Han, C., Lei, Y., Holroyd, C. B., & Li, H. (2011). Responsibility modulates neural mechanisms of outcome processing: An ERP study. Psychophysiology, 48(8), 1129–1133. Li, P., Jia, S., Feng, T., Liu, Q., Suo, T., & Li, H. (2010). The influence of the diffusion of responsibility effect on outcome evaluations: Electrophysiological evidence from an ERP study. Neuroimage, 52(4), 1727–1733. Li, P., Shen, Y., Sui, X., Chen, C., Feng, T., Li, H., et al. (2013). The neural basis of responsibility attribution in decision-making. PloS One, 8(11), e80389. Li, P., Yuan, J., Jia, S., Feng, T., Chen, A., & Li, H. (2009). Feedback-related negativity effects vanished with false or monetary loss choice. Neuroreport, 20(8), 788–792. Mathews, A., & MacLeod, C. (2005). Cognitive vulnerability to emotional disorders. Annual Review of Clinical Psychology, 1, 167–195. Mies, G. W., vander Veen, F. M., Tulen, J. H., Bir kenhäger, T. K., Hengeveld, M. W., & van der Molen, M. W. (2011). Drug-free patients with major depression show an increased electrophysiological response to valid and invalid feedback. Psychological Medicine, 1–11. Miller, J. D., & Pilkonis, P. A. (2006). Neuroticism and affective instability: The same or different? American Journal of Psychiatry, 163(5), 839–845. Miltner, W. H. R., Braun, C. H., & Coles, M. G. H. (1997). Event-related brain potentials following incorrect feedback in a time-estimation task: Evidence for a ‘‘generic” neural system for error detection. Journal of Cognitive Neuroscience, 9, 788–798. Müller, S. V., Möller, J., Rodriguez-Fornells, A., & Münte, T. F. (2005). Brain potentials related to self-generated and external information used for performance monitoring. Clinical Neurophysiology, 116, 63–74. Nieuwenhuis, S., Holroyd, C. B., Mol, N., & Coles, M. G. H. (2004). Reinforcementrelated brain potentials from medial frontal cortex: Origins and functional significance. Neuroscience and Biobehavioral Reviews, 28, 441–448. Nieuwenhuis, S., Slagter, H., Alting von Geusau, N., Heslenfeld, D. J., & Holroyd, C. B. (2005). Knowing good from bad: Differential activation of human cortical areas by positive and negative outcomes. European Journal of Neuroscience, 21, 3161–3168. Oliveira, F. T., McDonald, J. J., & Goodman, D. (2007). Performance monitoring in the anterior cingulate is not all error related: Expectancy deviation and the representation of action-outcome associations. Journal Of Cognitive Neuroscience, 19(12), 1994–2004. Olvet, D. M., Klein, D. N., & Hajcak, G. (2010). Depression symptom severity and error-related brain activity. Psychiatry Research, 179(1), 30–37. Pfabigan, D. M., Alexopoulos, J., Bauer, H., & Sailer, U. (2011). Manipulation of feedback expectancy and valence induces negative and positive reward prediction error signals manifest in event-related brain potentials. Psychophysiology, 48, 656–664.
Rose, E. J., Simonotto, E., & Ebmeier, K. P. (2006). Limbic over-activity in depression during preserved performance on the n-back task. Neuroimage, 29, 203–215. Ruchsow, M., Herrnberger, B., Wiesend, C., Grön, G., Spitzer, M., & Kiefer, M. (2004). The effect of erroneous responses on response monitoring in patients with major depressive disorder: A study with event-related potentials. Psychophysiology, 41(6), 833–840. Sambrook, T. D., & Goslin, J. (2015). A neural reward prediction error revealed by a meta-analysis of ERPs using great grand averages. Psychological Bulletin, 141(1), 213–235. Santesso, D. L., Steele, K. T., Bogdan, R., Holmes, A. J., Deveney, C. M., Meites, T. M., et al. (2008). Enhanced negative feedback responses in remitted depression. Neuroreport, 19, 1045–1048. Schrijvers, D., de Bruijn, E. R., Maas, Y., De Grave, C., Sabbe, B. G., & Hulstijn, W. (2008). Action monitoring in major depressive disorder with psychomotor retardation. Cortex, 44(5), 569–579. Schrijvers, D., De Bruijn, E. R., Maas, Y. J., Vancoillie, P., Hulstijn, W., & Sabbe, B. G. (2009). Action monitoring and depressive symptom reduction in major depressive disorder. International Journal of Psychophysiology, 71(3), 218–224. Schultz, W. (2013). Updating dopamine reward signals. Current Opinion in Neurobiology, 23(2), 229–238. Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural substrate of prediction and reward. Science, 275(5306), 1593–1599. Spielberger, C. D., Gorssuch, R. L., Lushene, P. R., Vagg, P. R., & Jacobs, G. A. (1983). Manual for the state-trait anxiety inventory. Palo Alto, CA: Consulting Psychologists Press. Talmi, D., Atkinson, R., & El-Deredy, W. (2013). The feedback-related negativity signals salience prediction errors, not reward prediction errors. Journal of Neuroscience, 33(19), 8264–8269. Tucker, D. M., Luu, P., Frishkoff, G., Quiring, J., & Poulsen, C. (2003). Frontolimbic response to negative feedback in clinical depression. Journal of Abnormal Psychology, 112(4), 667–678. Ullsperger, M., Fischer, A. G., Nigbur, R., & Endrass, T. (2014). Neural mechanisms and temporal dynamics of performance monitoring. Trends in Cognitive Sciences, 18(5), 259–267. Van Noordt, S. J., & Segalowitz, S. J. (2012). Performance monitoring and the medial prefrontal cortex: A review of individual differences and context effects as a window on self-regulation. Frontiers in Human Neuroscience, 6. Walsh, M. M., & Anderson, J. R. (2012). Learning from experience: Event-related potential correlates of reward processing, neural adaptation, and behavioral choice. Neuroscience & Biobehavioral Reviews, 36(8), 1870–1884. Warren, C. M., Hyman, J. M., Seamans, J. K., & Holroyd, C. B. (2014). Feedback-related negativity observed in rodent anterior cingulate cortex. Journal of PhysiologyParis (September 16. pii: S0928-4257(14)00042-4). Yeung, N., & Sanfey, A. G. (2004). Independent coding of reward magnitude and valence in the human brain. The Journal of Neuroscience, 24(28), 6258–6264.