Better you lose than I do: neural networks involved in winning and losing in a real time strictly competitive game

Many situations in daily life require competing with others for the same goal. In this case, the joy of winning is tied to the fact that the rival suffers. In this fMRI study participants played a competitive game against another player, in which every trial had opposite consequences for the two players (i.e., if one player won, the other lost, or vice versa). Our main aim was to disentangle brain activation for two different types of winning. Participants could either win a trial in a way that it increased their payoff; or they could win a trial in a way that it incurred a monetary loss to their opponent. Two distinct brain networks were engaged in these two types of winning. Wins with a monetary gain activated the ventromedial prefrontal cortex, an area associated with the processing of rewards. In contrast, avoidance of loss/other-related monetary loss evoked activation in areas related to mentalizing, such as the temporo-parietal junction and precuneus. However, both types of winnings shared activation in the striatum. Our findings extend recent evidence from neuroeconomics by suggesting that we consider our conspecifics’ payoff even when we directly compete with them.


Results
Behavioral data. The ANOVA revealed a main effect of the factor outcome (F(2,136) = 54.16, p < 0.001, η 2 = 0.44). Post-hoc tests demonstrated that there was no significant difference in reaction time between gain (mean ± SD ; 198 ms ± 22) and loss cue trials (197 ms ± 21, p = 1), while both gain and loss cue trials resulted in faster reaction times than neutral cue trials (216 ms ± 17; p < 0.001 for both comparisons).  (Fig. 1c).

Discussion
The current study investigated brain activation in a strictly competitive game, in which the participant's performance had direct consequences for the outcome of the other player in the game. We particularly targeted two different types of winning in such game: Winning with monetary gain and winning without a gain, but incurring a monetary loss to the opponent. Winning trials with monetary gain of the participant were associated with increased activity in the bilateral striatum, VMPFC and middle/posterior insula. Winning which avoided a monetary loss to the participant, but was also associated with a monetary loss of the opponent, also engaged the bilateral striatum. In addition, it activated the inferior frontal gyrus (IFG), inferior parietal lobe (IPL), and precuneus. The behavioral data demonstrated that participants were faster in reaction time to monetary cues comparing to non-monetary ones, which is in line with other studies which used similar task 20,21 . Before we interpret these findings in more detail, we will discuss the main effects and interaction results of our design.
Main effect and interactions. The main effect of winns revealed, as expected, activation in regions like bilateral striatum and VMPFC, which is in line with many studies associating these areas with positive valence, in particular in the context of economic decision making (see 12 for recent meta-analysis). The interaction contrasts with self-related monetary winning (in gain frame trials) and self-related non-monetary winning (in loss frame trials) revealed activation in distinct areas. The former showed activation in VMPFC (at a cluster level correction threshold), whereas the latter indicated stronger engagement of the bilateral temporo-parietal lobe and precuneus. Overall, these results demonstrate that distinct regions were involved in the two types of winning.
The winning condition associated with monetary reward engaged more VMPFC, while the winning condition associated with avoidance of monetary loss/opponent's punishment involved the temporo-parietal region. As stated above, VMPFC engagement was repeatedly observed in numerous studies investigating primary and secondary rewards, and value computation. Thus, the increasing activation in this region when subjects increased their own monetary outcome is in line with the previous literature.
The activation in precuneus and temporo-parietal areas can be explained based on findings from social neuroscience associating these areas with empathy, mentalizing and theory of mind (for review see 22,23 ). Engagement of these functions might be related to the fact that in the monetary loss trials, subjects necessarily had to incur a loss in their opponents, in order to avoid a monetary loss in themselves. Therefore, activation in these areas may be related to a higher propensity to feel with or to adopt the perspective of the opponent's negative outcome in such a setting. Notably, previous studies have also identified MPFC to play a role in mentalizing, while this area was not activated in our study. However, a recent meta-analysis of mentalizing/theory of mind studies suggests that mPFC is predominantly related to trait and false belief statements, which might explain the lack of activation in the present setting which required state inferences 24 .
Since the focus of this work was also to investigate responses to different types of wins and losses, we will therefore now discuss the specific results of the corresponding contrasts in detail.  of numerous studies investigating the processing of primary and secondary rewards in both social and non-social contexts (for reviews see 12,[25][26][27] ). Additionally, mPFC and ACC also showed activation when a monetary gain was received. It has been suggested that these areas represent self-perception or self-knowledge in social contexts, as well as the ability to differentiate the self from other objects, and to recognize attributes and preferences related to oneself 15,28,29 .

Losses (LL[−3:0] > WG[3:0]) and (LG[0:3] > WL[0:−3]) contrasts. Contrary to our hypothesis,
we did not observe activation for the contrast LL[− 3:0] > WG [3:0] in the insula and in the ACC for self-related monetary loss. Only after reducing the threshold to 0.001 uncorrected we observed activation in bilateral IFG, TPJ and precuneus. A similar situation was given with another losing contrast LG[0:3] > WL[0:− 3], where we observed activation in bilateral insulae and left hippocampus only after reducing the threshold to 0.001 uncorrected. One explanation for a lack of significant activity for these contrasts is that the numbers of loss trials was lower than the win trials, to make the task settings believable and let the subjects get more profit. This might however have reduced the statistical power of analyses targeting higher activation in the loss trials.

Avoidance of monetary loss /opponent's punishment WL[0:−3] > LG[0:3]. This contrast
aimed to compare a situation in which participants achieved a zero payoff, but avoided a monetary loss which instead was incurred to the opponent, with an equivalent situation in terms of payoff, which however carried a monetary gain for the opponent. Interestingly, this revealed rather distinct neural networks. Additionally to activation in VS and PPC, we observed activation in temporo-parietal areas  like bilateral inferior parietal lobule, TPJ, TP, precuneus, and the IFG. The temporo-parietal areas are often described as empathy-and mentalizing-related areas that are recruited when individuals need to understand and predict other people's intentions and beliefs 14,[30][31][32] . This is particularly important in social contexts like competition, where we need to observe or are directly made aware of how our own actions and their outcomes affect others. Neuroimaging studies which used other types of competition tasks also confirmed an engagement of these areas. For example, competition was associated with activation in the right IFG, bilateral temporal lobe, bilateral fusiform and bilateral precuneus during an adapted Stroop Task 33 . Competition was also associated with activation in the inferior, parietal and medial prefrontal cortices 2 . Two more studies observed competition-related brain activation in TPJ and TP during a competitive ultimatum game 3 and a competitive domino game 4 . A recent study by Radke et al. also demonstrated activation in parietal cortices and TP when the action of a participant in the game had negative consequences for their opponents 5 .
In addition to temporo-parietal activation during other's-related monetary loss, we observed lateral prefrontal (LPFC) and bilateral IFG activation. Earlier accounts had associated these areas with distinguishing self from other 30 . More recently, several neuroimaging studies observed activation of IFG during loss aversion 34,35 , safe reward 36 and risk aversion 37 . Hence, activation in IFG during observing someone else's misfortune might represent a general mechanism of processing losses.

ROI results.
Our exploratory ROI analysis demonstrated that VMPFC was activated significantly higher during Monetary gain contrast than during Avoidance of monetary loss/opponent's punishment contrast, while activation for right PCC and right TPJ showed the opposite pattern of activation. These findings also support our hypothesis about that different brain networks involved depend on the type   of winning. The coordinates for rTPJ and rPCC were taken from the peak of activation of the contrast "Mentalizing about Others versus Mentalizing about Yourself " from the study of 38 and we observed higher activation in these regions for the contrast where the participant had the stronger negative impact on the payoff for the opponent.

Shared activation for own monetary gain and avoidance/other monetary loss. One explana-
tion for shared activation in the ventral striatum when participants won a trial and received a monetary gain, and when winning in loss frame trials (which incurred a loss to the opponent) is that this activation is associated with general reward processing 39 . According to a recent concept it may indicate an enhanced motivational value in the form of incentive salience attribution to stimuli perceived at that moment 40 . In addition, in our ROI analysis there was no difference in activation in right nucleus accumbens for different types of winning. This also suggests that activation in VS is associated with generalized aspects of winning. It might be argued that our results can also be explained in a prediction error framework, as areas such as the striatum have been associated in the coding of prediction errors 41 . However, the present design was not tailored to analyze or interpret our results within such a framework. This is so because the outcome of each trial was very ambiguous and hard to predict for the player, which likely resulted in a complex, subjective and individually varied mixture of positive and negative expectations and expectation violations which could not be modeled.
A different explanation is that a competitive situation may elicit different types of emotional reactions. Participants may experience empathy while observing failure of a group member, but failure of a rival may cause Schadenfreude, i.e. pleasure about someone else's misfortune. One possible condition when Schadenfreude may arise is when people can gain from another's misfortune 42

. Takahashi et al. (2009) demonstrated a stronger correlation between activation in ventral striatum and self-reported
Schadenfreude in a situation when misfortunes happened to envied persons 43 , and a different study concluded that the striatum plays a role in mediating the emotional consequences of social comparison during competition 8 . Furthermore, in a social group competition an increase of VS activation was observed during success of the favored team or failure of the rival team, even against a third team 6 . Similarly, the VS was activated during watching a negatively evaluated out-group member receiving pain 44 , and observing others making errors 13 .
Although we did not explicitly measure the level of Schadenfreude, pleasantness and motivation in this study, we speculate that activation in the striatum partially is related to these aspects. This interpretation is further supported by a recent study which showed that participants' self-evaluations of pleasantness were associated with activation in the VS when winning in a competitive game 45 . Additional research is needed to directly examine the link between VS with Schadenfreude and motivation during competitive interactions.
However, we need to take into account that humans are not exclusively motivated by material self-interests, but that people often also care for the well-being of others 46 . Moreover it was found that individual differences in prosocial value orientation are important for the allocation of recourses between self and others, and that amygdala, striatum and VMPFC play a critical role mediating this effect [47][48][49] . Our task design provided no choice but to punish the opponent in the loss condition, and this certainly has affected subjects with differences in prosocial orientations in a different way. Since we however did not collect data on individual differences in prosocial orientation, the question how the neural networks identified in our study are related to such differences needs to be clarified by future studies.
Taken together, this study demonstrates that two distinct brain networks are engaged when people process of two types of winning in the game, i.e., own monetary gain and others-monetary loss. A medial-frontal network demonstrated activation for own monetary gain, while a temporo-parietal network was more involved in response to others' monetary losses. Both types of winning in the game shared activation in the VS which may represent the "joy of winning" for outperforming someone else during competition. Alternatively, this may suggest that the misfortunes of opponents were treated as reward and elicited Schadenfreude.
In conclusion, the present study demonstrated that, depending on the type of winning in the competitive game, distinct brain areas are engaged in the processing. Our results provide new insights for understanding brain function during competitive contexts and fundamental features of human social interactions.

Methods
Participants. Sixty nine healthy volunteers (38 females and 31 males) participated in the experiment.
The average age was (mean ± SD) 23.8 ± 5.4 years old. All volunteers had no history of psychiatric or neurological disorders or contraindications for high-field MRI scanning. All were right-handed as assessed by the Edinburgh Handedness Inventory. All participants signed informed consent before the study and the study protocol was approved by the ethics committee of the Medical University of Vienna. The methods were carried out in accordance with approved guidelines.
Competitive Task design. We employed a Competitive Incentive Delay (CID) task, which was a modification of the Monetary Incentive Delay (MID) task introduced by Knutson and colleagues 11 . The Scientific RepoRts | 5:11017 | DOi: 10.1038/srep11017 CID differed from the MID only by the fact that participants played against another person, rather than trying to stay within a pre-set reaction time as in the original MID.
More specifically, participants were told that they were competing with another participant, to whom they were connected via the computer network. In reality, though, they were playing "against" a pre-programmed computer algorithm. To make the task more believable, all participants had taken part in practice trials, together with the experimenter and before entering the scanner. In these practice trials, experimenter and participant played the CID against each other in real time and while sitting next to each other, in front of a computer. The practice trials also served to familiarize the participants with the task and to minimize learning effects during the experiment. After entering the scanner and before the task started, the abstract silhouette of an opponent and a message that the connection with the opponent's computer had been initiated was shown on the screen. Participants did not get any personal information about their adversary. In reality, they played against a pre-set computer algorithm, and were debriefed after completion of the experiment.
The CID consisted of one scanning run lasting about 9 min, in which 72 trials were played. At the ontset of each trial, participants saw one of three geometrical cues for 250 ms. Next, they anticipated the appearance of a target square, to which they had to respond with a button press as fast as possible. During target anticipation, a fixation crosshair was shown, and the anticipation period was varied randomly between 2000-2500 ms. Immediately after disappearance of the target, feedback was presented for 1650 ms. Feedback informed participants about whether they had won or lost money during that trial, their total score, and the opponent's total score (Fig. 4a). "Monetary Gain" cues signaled the possibility of winning € 3 (a circle with three horizontal lines; 32 trials), "Monetary Loss" cues signaled the possibility of losing € − 3 (a square with three horizontal lines; 24 trials), and cues representing "no monetary outcome" (€ 0; 12 trials) were denoted by a triangle. The rationale for a larger number of gain trials was that we wanted participants to have the chance to finish the game with a net monetary gain.
To increase the competitiveness of the task, and in line with the strictly competitive task setup we intended to implement, the possible outcomes were arranged in a way that participant and "opponent" were always directly linked to each other's monetary score. I.e., if participants pressed the button in time before the go cue would disappear from the screen, they would win money, while the (alleged) opponent's payoff was zero. If they failed to respond fast enough, the opponent received the monetary gain and the participant nothing. If participants pressed the button on time after a loss cue, the opponent would lose money, but not the participant. If they missed, the opposite payoff was the case (Fig. 4b). The main overall goal of the task communicated to participants was to maximize their monetary outcome, and to receive more money than the opponent. Thus, participants were paid the final monetary revenue they had achieved after completing the task. Trial types were pseudorandomly ordered within each run. The display duration of the target cue was adapted to the participant's performance (within 80-370 ms) to ensure that all participants won in approximately 2/3 of all trial types.
Reaction times were analyzed in SPSS 20.0 (SPSS Inc., Armonk, USA) using a repeated-measures ANOVA with 3 levels for the factor outcome ("Monetary Gain", "Monetary Loss" and "Non-Monetary Outcome"). Significance was evaluated at P < 0.05. Post-hoc tests with Bonferroni correction for multiple comparisons were applied. Data are reported as means ± SD. MRI scanning. MRI scanning was conducted on a 3 Tesla TIM Trio whole body scanner (Siemens, Germany). Participants were scanned using the manufacturer's 32-channel head coil. Functional images were obtained with a single-shot echo planar imaging (EPI) sequence. The image acquisition parameters were as follows: repetition time (TR) = 1.8 s, echo time (TE) = 38 ms, flip angle (FA) = 90°, 294 whole-brain volumes (matrix size 128 × 128, FoV = 190 × 190 mm 2 , 3 mm slice thickness). For anatomical registration, we obtained high-resolution 3D T1 anatomical images after the fMRI runs (magnetization prepared rapid gradient echo sequence, TR = 2.3 s, TE = 4.21 ms, 1.1 mm slice thickness, 900 ms inversion time, 9° flip angle).
Image analysis was performed using the SPM8 software package (www.fil.ion.ucl.ac.uk/spm) implemented in MATLAB (Mathworks Inc., Natick, USA). Preprocessing included correction for slice timing differences, realignment to the first image to adjust for movement, segmentation, normalization to standard MNI space (at isotropic voxel size), and smoothing with a Gaussian filter (8 mm). The first level (individual subject) analyses were set up using the general linear model approach, with events of interest being modeled by regressors. The fixation cross interval between trials were modeled as an implicit baseline.
The anticipation-related responses for all cues were also modeled. Contrast images of these regressors from the first level were then entered into second level random-effects analyses. We used the flexible factorial design option implemented in SPM8 to compare brain activations in response to the different types of feedback.
The contrasts we assessed focused on neural activation differences during the feedback conditions. First, we calculated the contrasts (WG For all analysis, we used a family-wise error (FWE) correction at the voxel level, at a threshold of P < 0.05 and a cluster extent threshold of 5 voxels, for identifying statistically significantly activated voxels. In some cases where we had strong prior hypotheses, data were also explored at more liberal thresholds (see Results). All results are reported in accordance with recommendations from 50,51 . Exploratory ROI analysis. For the exploratory analysis of brain activation in regions associated with reward processing and mentalizing, we prepared four regions of interests. Two regions, the right PCC and the right TPJ, were defined as spheres of 8 mm radius, with the ROI center being taken from the peak of activation of the contrast "Mentalizing about Others" versus "Mentalizing about Yourself " from the study of Lombardo et al. (8 − 58 28 and 60 − 60 14, respectively) 38 . Two other ROIs, VMPFC and right nucleus accumbens, were defined in the same way from a meta-analysis 26 . The spheres were based on the peak coordinates found for the analysis of monetary outcome (2 40 − 6 and 8 14 − 4, respectively).
Mean parameter estimates within four ROI masks were extracted for the contrasts (WG[3:0] > LL[− 3:0]) and (WL[0:− 3] > LG[0:− 3]) from each individual, and entered into statistical analysis. Group differences for each contrast were analyzed in SPSS 20.0 (SPSS Inc., Armonk, USA) using a repeated-measures ANOVA with ROIs as within-subjects factors and Contrast as between-subjects factor. If the sphericity assumption was violated (significant results in Mauchly's test of sphericity), degrees of freedom were corrected using Greenhouse-Geisser estimates of sphericity. Significance was evaluated at P < 0.05. Post-hoc tests with Bonferroni correction for multiple comparisons were applied.