We studied neural correlates of games experience during a first-person shooter game with fMRI. We found a relative deactivation of the caudate nucleus as well as the medial orbitofrontal cortex, as compared to the ongoing gameplay, when subjects failed in the game. This is in accordance with the reported role of those structures in reward-prediction error .
Negative reward prediction error is a decrease in activity observed when outcomes are more negative than expected or an anticipated reward is omitted [25, 26]. Haruno and Kawato  demonstrated that caudate nucleus activation is correlated with reward-prediction-error during reward feedback. They argue that the caudate nucleus, like ventral striatum, is mainly engaged in the learning process involved in comparing actual and predicted rewards. In our study, deactivations occurred whenever the subject did not receive an expected reward, e.g. was eliminated instead of eliminating an opponent. A potential confound is a spill-over from motor structures in the dorsal striatum. However dorsal striatum did not yield a significant signal. Moreover the deactivation was apparent also during success events, which are linked to high motor activity.
In response to success events we found no activation of the structures responsible for reward processing. The majority of midbrain dopamine neurons show rather stereotyped, phasic activations following temporally unpredicted rewards, even if the delivery itself is to be predicted . Although in our study the rate of success events was higher than of failures, there was a high level of temporal uncertainty included; therefore, we would expect reward system activation. Instead, we rather observed deactivation relative to baseline in the ROI analysis. This may be explained by rather tonic than phasic activation of the reward system present during the game as found by Koepp et al. . Consequently this activation was not visible in the studied contrasts. Indeed the narrative of the game may lead to rewarding experience. In order to test this hypothesis, future studies could introduce a control game containing less rewarding phases and less stress on goal directed behavior. Przybylski et al.  suggested that reward in the game is not associated with the violent content, but factors such as experience of autonomy and competence in game play. In accordance with these behavioral findings, we found no activation of the reward system in response to violent events (even though they served as the direct aim in the game and players collected points for each time they eliminated an opponent). In our coding system, we considered killing an opponent as a rewarding event and did not differentiate how realistic the interaction was (compare ). Moreover, other factors such as motor activity may covary with different game events. This may explain the deactivations of premotor cortex associated with failure events (see Table 2) as sign of reduced activity after being killed. A more complex model of game enjoyment, that is not restricted to game violence and the direct aim of the game, may account for the broader spectrum of game experience (see ).
Subjects with larger right temporal pole (rTP) response to failure reported a decrease of negative affect after game play. TP activations have been frequently observed in simple emotional tasks, such as emotional face perception, as well as in complex emotional tasks, such as theory of mind, in particular with socially important narratives; moreover, the TP responded to tasks that require one to analyze other agent's emotions, intentions or beliefs (see ). Interestingly, although TP activations are frequently listed in fMRI results, their function is rarely discussed [49, 50]. Olson et al.  suggested that the TP is involved in both social and emotional processes via binding complex, highly processed perceptual inputs to visceral emotional responses.
In our study, the rTP activation explained the individual differences in the increase of negative affect due to game playing: it was less active in response to failure in the game in those subjects who reported higher negative affect after the experiment. In the other subjects, rTP activation allowed them to evaluate the failure events in a broader cognitive and social frame of the game and protected them from affect and mood decreases. Indeed, patients with the atrophy of right but not left anterior temporal lobe present changes of mood including depression, apathy and irritability . Moreover, neuroticism scores reflected TP involvement during the perception of negative emotions . In a similar vein, Liu et al.  demonstrated that the TP was associated with an evaluation of wrong in-game decisions. We postulate that the TP can be involved in the evaluation of our own emotions in a broader social context, allowing to control our own affect in social situations, rather than assessing the intentions of others using theory of mind.
To further elucidate the function of the temporal pole and caudate nucleus in the individual appraisal of failure, we explored the prediction of single items by localized responses and found a profile of relevant adjectives. As compared to the rTP, the lTP showed less predictive power for the item and both caudate nuclei showed just a random association with the behavioral measures. The predicted adjectives (e.g. 'irritable' or 'hostile') were in agreement with the feeling of anger. In humans, uni- or bilateral anterior temporal lobe damage can lead to the Klüver-Bucy syndrome (see ). It was first described in monkeys, where it encompassed fear and anger and led to severe socio-emotional disorders . Moreover, in a multivariate analysis of a community sample, Ferguson, Olson, Kutner & Warner  found the only form of aggression being predicted by video game violence was in response to anger (see Table 4 in ). This poorly specified feeling of anger may represent a central affect dimension in violent games and be controlled by right temporo-polar areas.
Unlike negative affect, positive affect change was not correlated with the change of brain responses. This is in accordance with the postulated model of independent positive and negative affect dimensions . Moreover, it supports the claim of Coventry and Constable  who recommended measuring both positive and negative affect in gaming behavior. However, we chose a violent video game as an experimental paradigm and evaluated brain activation in response to violent events only. This may have biased us towards the experience of negative affect and also contribute to the lack of neural correlates to change in positive affect. It is less likely that the experience of being scanned rather than the game itself was responsible for the decrease in positive affect. Other researchers found no effect of scanning on the positive affect rating of PANAS (e.g. ). Moreover, violent games are known to evoke negative affective experiences, including aggressive feelings and thoughts .
Due to its location, the TP is an area susceptible to artifacts in fMRI due to the adjacent air-tissue transitions that lead to magnetic field inhomogeneities . Therefore its activation can often be missing in neuroimaging data. We applied multiecho EPI sequence with alternating phase encoding polarity  and dephasing reduction  which allowed us to obtain robust signal from this region . This improved technology lends additional credibility to the presented findings and may explain the lack of findings in previous fMRI studies.
The peak t-values at both temporal poles were low but the cluster extent was rather large. This finding suggests that the networks subserving the affective evaluation have a large variability across subjects or that the processing is distributed across extended structures. The compartmentalization at this level may be not as strict as in lower-level, motor or sensory areas (see ). This is in accordance with variability of other high-level functions such as language production . Affective evaluation and its cognitive consequences may therefore be considered to emerge from distributed networks. Additional clusters implicated orbitofrontal and premotor structures as part of such extended network subserving the subjective evaluation. The reward areas and ACC failed to show significant associations with the affective measures and thus appear as distinct functional units.
Gaming behavior and reactivity of reward system are characterized by inter-individual variety. The Reinforcement Sensitivity Theory posits that a neurobiological system, the Behavioral Activation System, defines individual differences on the subject's sensitivity and reactivity to appetitive stimuli associated with mesocorticolimbic structures . For instance, Bühler et al.  demonstrated the difference in processing of cigarette reward by occasional and dependent smokers: the former group demonstrated stronger reactivity of the mesocorticolimbic system for monetary than for cigarette reward, the latter responded equally to both reward types. Reuter et al.  reported a reduction of ventral striatal and ventromedial prefrontal activation in the pathological gamblers that was negatively correlated with gambling severity. Such variances of reactivity were not expected to specifically bias reward during gameplay or its relation to subjective experience. In larger study populations, the variables affecting reactivity in the reward system should be considered, which may prompt further insight into addictive video playing.