Reward sensitivity modulates connectivity among reward brain areas during processing of anticipatory reward cues

Reward sensitivity, or the tendency to engage in motivated approach behavior in the presence of rewarding stimuli, may be a contributory factor for vulnerability to disinhibitory behaviors. Although evidence exists for a reward sensitivity‐related increased response in reward brain areas (i.e. nucleus accumbens or midbrain) during the processing of reward cues, it is unknown how this trait modulates brain connectivity, specifically the crucial coupling between the nucleus accumbens, the midbrain, and other reward‐related brain areas, including the medial orbitofrontal cortex and the amygdala. Here, we analysed the relationship between effective connectivity and personality in response to anticipatory reward cues. Forty‐four males performed an adaptation of the Monetary Incentive Delay Task and completed the Sensitivity to Reward scale. The results showed the modulation of reward sensitivity on both activity and functional connectivity (psychophysiological interaction) during the processing of incentive cues. Sensitivity to reward scores related to stronger activation in the nucleus accumbens and midbrain during the processing of reward cues. Psychophysiological interaction analyses revealed that midbrain–medial orbitofrontal cortex connectivity was negatively correlated with sensitivity to reward scores for high as compared with low incentive cues. Also, nucleus accumbens–amygdala connectivity correlated negatively with sensitivity to reward scores during reward anticipation. Our results suggest that high reward sensitivity‐related activation in reward brain areas may result from associated modulatory effects of other brain regions within the reward circuitry.


Introduction
Reward sensitivity is a trait that predisposes to a variety of disinhibition disorders, including attention deficit hyperactivity disorder, psychopathy, drug abuse and addiction, pathological gambling, and eating disorders (see Bijttebier et al., 2009 for review). Behavioral studies have associated this trait with enhanced reward processing and learning, a preference for immediate reward, and lower inhibitory control in reward contexts (Corr, 2004;, as well as active avoidance under punishment contingencies (Gray, 1981(Gray, , 1991Smillie & Jackson, 2005). The brain regions of the dopaminergic reward system are thought to constitute the neural substrate for individual differences in reward sensitivity (Gray, 1991;Depue & Collins, 1999;Pickering & Gray, 2001). The neural structure of the dopaminergic reward system forms a loop in which dopaminergic midbrain areas, such as the ventral tegmental area (VTA) and substantia nigra (SN) complex, send projections to limbic and prefrontal brain areas, and receive afferent fibers from most of these areas (D€ uzel et al., 2009;Haber & Knutson, 2010). Substantial literature links this system to motivation and goal-directed behaviors, and the system is thought to modulate diverse cognitive processes that allow the attainment of reward and the relief from punishment [see Berridge & Robinson (1998) for a review].
Individual differences in reward sensitivity have been associated with the structural and functional variability of definite rewardrelated areas within the dopaminergic system. For example, individuals with high reward sensitivity show diminished striatum volume (Barr os-Loscertales et al., 2006), increased white-matter tract strength between the nucleus accumbens (NAcc) and amygdala (Cohen et al., 2009), more random resting-state neural dynamics (or irregular fluctuating time series) in the NAcc and orbitofrontal cortex (Hahn et al., 2012), and increased NAcc and midbrain responses to reward anticipation (Beaver et al., 2006;Carter et al., 2009;Hahn et al., 2009;C amara et al., 2010). Although these studies provide evidence that reward sensitivity modulates the structure and functioning of brain reward areas, the role of this trait in the connectivity between these regions remains unclear.
In this study, we investigated how individual differences in reward sensitivity modulate the activity and functional connectivity of reward brain areas during the processing of valence and incentive magnitude in a monetary incentive delay (MID) task. This paradigm involves approach and active avoidance processes that are supposedly mediated by individual differences in reward sensitivity according to Gray's model (Arnett & Newman, 2000;Avila, 2001;Smillie & Jackson, 2005). Previous functional magnetic resonance imaging (fMRI) studies have been focused on the relationship between reward sensitivity and reward cues involving the NAcc, midbrain, orbitofrontal cortex, and amygdala (Beaver et al., 2006;Carter et al., 2009;Hahn et al., 2009). However, no previous fMRI studies have considered the involvement of reward sensitivity in brain reactivity to both approach and active avoidance cues, as others have investigated behaviorally (Arnett & Newman, 2000;Smillie & Jackson, 2005). On the basis of previous reports, we hypothesized that there was an enhanced response of brain reward areas (e.g. NAcc and orbitofrontal cortex) during the processing of the motivational valence of cues by individuals with stronger reward sensitivity. Moreover, we explored the relationship between reward sensitivity and brain areas involved in processing the incentive magnitude of stimuli independently of their valence. Finally, we studied the regional brain connectivity among reward-related areas associated with individual differences in reward sensitivity.

Participants
Forty-four male undergraduates (age, 23.4 AE 4.1 years; years of education, 13.8 AE 2.2) participated in this fMRI study. Participants were physically and psychologically healthy, with no history of mental disorders, head trauma, or drug abuse. All participants provided written informed consent prior to participation. The study was approved by the ethical committee of the University Jaume I. All study procedures conformed with the Code of Ethics of the World Medical Association (Declaration of Helsinki; printed in the British Medical Journal, 18 July 1964).

Measure of reward sensitivity
All participants completed the Sensitivity to Punishment and Sensitivity to Reward Questionnaire (Torrubia et al., 2001) as a measure of reward sensitivity. The mean sensitivity to reward (SR) score was 11.72 (standard deviation, 4.65; range, 2-24), and scores followed a normal distribution (Kolmogorov-Smirnov test: D = 0.113, P > 0.10). Thus, these scores were consistent with those obtained from other samples (Torrubia et al., 2001;Barr os-Loscertales et al., 2006. The Sensitivity to Punishment and Sensitivity to Reward Questionnaire has been translated into 15 languages, and is widely used to assess reward sensitivity in adults  and children (Luman et al., 2012). Previous studies have shown that the SR scale has good content validity and strongly correlates with other measures of reward sensitivity, such as reward responsiveness, drive, fun-seeking, novelty-seeking, and impulsivity scales [see Torrubia et al. (2008) for review].

Experimental design and stimuli
The goal of our experiment was to analyse the association between individual differences in reward sensitivity and the functional activity and effective connectivity of reward brain areas during the anticipation of monetary incentives. We used an adaptation of the MID task described by Knutson et al. (2001Knutson et al. ( , 2003, including all high and low reward and punishment conditions (Fig. S1). Before entering the scanner, all participants were given instructions on the task and completed a practice session. The practice session was thought to minimize later learning effects, and provided an estimate of each individual's reaction time (RT), to standardize task difficulty within the scanner. For each participant, the median RT of correct trials during the practice session was implemented as a cut-off RT in the main experiment. All participants were initially paid €20 for their participation. At the end of the experiment, participants received an individually adjusted bonus, depending on their performance in the experimental task.
Inside the scanner, participants performed two 8-min runs of the MID task. Each run consisted of 60 trials, giving a total of 120 trials. There were four kinds of event, defined by a high reward cue, a low reward cue, a high punishment cue, and a low punishment cue. Each trial consisted of one of those cues presented for 500 ms, followed by a black screen that appeared for a variable duration (2000-2250 ms), and then by a white target square that appeared for 100 ms, to which participants had to respond by pressing a response button as quickly as possible. After the participant responded, a black screen appeared for a variable duration (2000-4000 ms), followed by a feedback screen (duration of 1500 ms) that notified participants of whether they had won or lost money during that trial and indicated their cumulative total at that point. As previously noted, each event was defined by the initial appearance of a different cue: a circle with two horizontal lines, indicating the possibility of winning €3 (high reward cue; n = 24); a circle with one horizontal line, indicating the possibility of winning €0.20 (low reward cue; n = 24); a square with two horizontal lines, indicating the chance to avoid losing €3 (high punishment cue; n = 24); and a square with one horizontal line, indicating the chance to avoid losing €0.20 (low punishment cue; n = 24). Therefore, the cues informed participants of the potential valence of the outcome (reward or punishment) and its incentive magnitude (high or low). A triangle (n = 24) was the cue for non-incentive trials in which participants neither won nor lost money. Participants had to respond after each incentive cue, but they did not respond to non-incentive cues, because these cues were not followed by a target stimulus (white square). We modified the original MID task in this way in order to perform comparisons without disentangling reward anticipation from action preparation. These comparisons may produce interesting results when the effects of modulation of brain processing by reward sensitivity are analysed, as the effects of individual differences on instrumental approach and active avoidance behavior may arise from the joint effects of valence (Hahn et al., 2009) and motor responses in our regions of interest (ROIs), e.g. the striatum (Guitart-Masip et al., 2011). On the other hand, it is probably mediating the differences showed in previous behavioral studies on reward sensitivity [see Pickering & Gray (2001) and  for reviews]. Additionally, we could isolate the motivational effects in our study by means of a factorial design with two valence conditions (reward and punishment) and two incentive magnitude conditions (high and low), as motor effects are controlled for by the motivational conditions. Trial types were pseudo-randomly ordered within each run. The intertrial interval was randomized between 2000 and 4000 ms. Participants were instructed to respond as quickly as possible to target stimuli to achieve the rewards or avoid the punishments. The task was programmed and presented with PRESENTATION software (Neurobehavioral Systems, Albany, CA, USA). Visual stimuli were displayed in the scanner with Visuastim goggles (Resonance Technology, Northridge, CA, USA). Stimulus presentation was synchronized with the scanner acquisition with a SyncBox (Nordic NeuroLab, Bergen, Norway), and behavioral task performance was recorded with a ResponseGrip (Nordic NeuroLab).

fMRI preprocessing and analysis
The analyses focused on changes in the blood oxygen level-dependent contrast during the anticipatory cue periods. Data were preprocessed and analysed with the Statistical Parametric Mapping (SPM) 5 software package (Wellcome Trust Centre for Neuroimaging; http:// www.fil.ion.ucl.ac.uk/spm/). The first two scans for each participant in each run were excluded from the analyses, to discount any artefacts related to the transient phase of magnetization. For preprocessing purposes, the time series of voxels were interpolated intravolume to a middle slice (in terms of acquisition time), to correct the acquisition of non-simultaneous slices (slice order: ascending interleaved). Later, motion correction was carried out by taking the first image of the first session as the reference image, obtaining a subsequent realignment average image, and using this average image as a reference for the other session's motion correction. This correction was made with a six-parameter rigid body transformation. An anatomical image for each participant was coregistered to his average functional image with a rigid body transformation. Then, the anatomical acquisition was segmented and normalized. This normalization was completed according to the Montreal Neurological Institute (MNI) template by applying an affine transformation followed by non-linear deformation with the basis functions defined in the SPM program (Ashburner & Friston, 1999). Computed transformation parameters from the anatomical image normalization after segmentation were applied to each participant's functional time series (voxels rescaled to a final voxel size of 3 mm 3 ). Finally, the images were spatially smoothed with a 6-mm isotropic Gaussian kernel.
Significant hemodynamic changes among the conditions were examined with the general linear model (Friston et al., 1995). In the first-level (within-subjects) analysis, a statistical model was computed for each participant by applying a canonical hemodynamic response function combined with its time derivative. The fMRI time series data were high-pass-filtered to eliminate low-frequency components. The four conditions of interest (high reward cue, low reward cue, high punishment cue, and low punishment cue) were modeled as separate regressors in a general linear model. Furthermore, we modeled separate regressors for the eight outcomes (win or loss in each incentive trial) and the non-incentive cue. The six motion correction parameters from each participant were included in the model as 'nuisance' variables. Finally, statistical contrast images were generated to obtain the brain activation for anticipatory periods.

ROIs
Predefined ROIs included the NAcc, amygdala, medial orbitofrontal cortex (mOFC), and midbrain, based on previous studies of reward sensitivity (Beaver et al., 2006;Hahn et al., 2009). All of these structures were defined according to the AAL (Tzourio-Mazoyer et al., 2002) or the Wake Forest University PickAtlas (Maldjian et al., 2003) for self-defined ROIs. Discrete ROIs were defined for the amygdala and the mOFC, the latter including the bilateral rectus gyrus from the AAL toolbox (Tzourio-Mazoyer et al., 2002). The NAcc was defined as a 6-mm-radius sphere at AE10, 8, À4 [x, y, z; MNI coordinates based on Cools et al. (2002) and Barr os-Loscertales et al. (2006)], whereas the midbrain was defined as a 6-mmradius sphere at 0, À20, À12 [x, y, z; MNI coordinates based on Telzer et al. (2010)], which mainly includes the VTA-SN complex. fMRI analysis: overall task activations fMRI analyses were conducted to study brain areas responding to both the reward and incentive magnitude conditions, depending on reward sensitivity. Theoretical models of personality have proposed that reward sensitivity modulates both signals of reward and signals of relief from punishment (Gray, 1991;Pickering & Gray, 2001). In addition, previous studies have shown that some dopaminergic brain areas respond to the salience of stimuli independently of their valence (Knutson et al., 2005;Jensen et al., 2007). Thus, it is possible that reward sensitivity modulates brain areas responding to both reward and incentive magnitude. To study brain activity related to reward cues and high incentive cues, we performed a two-way [valence (reward, punishment) 9 incentive magnitude (high, low)] repeated-measures ANOVA, in a second-level random-effects analysis with the contrast images (high reward, low reward, high punishment, and low punishment) extracted from the first-level analysis.
Moreover, we implemented a conventional subtraction analysis between the reward cues (high and low) and non-incentive cue (neutral triangle). The objective of this contrast was to study the whole process of reward anticipation, including motivation and motor preparation components, given the interest in these in the analysis of individual differences in personality based on previous comments (see Experimental design and stimuli). We hypothesized that this comparison might change the effect of individual differences on brain activity during reward anticipation, allowing the study of modulation of the ROIs by reward sensitivity. We performed a comparison of reward cues and the non-incentive cue for each participant in the first-level analysis, and used the resulting contrast images in a onesample t-test in the second-level analysis. Reported results were those that survived a small volume correction (SVC) with a statistical significance threshold of P < 0.05 [family-wise error (FWE) corrected].

Analysis of effects of reward sensitivity on task-related activations
To analyse the modulatory effect of personality on brain activation, three multiple regression analyses were performed between SR scores and the resulting contrast images obtained in the first-level analysis: (i) reward vs. punishment; (ii) high incentive vs. low incentive; and (iii) reward vs. neutral. The nuisance effects of age were regressed out. Analyses were carried out on each ROI with SVC, with a statistical significance threshold of P < 0.05 (FWE corrected).

Functional connectivity analysis: psychophysiological interaction (PPI)
Following previous studies on personality (Haas et al., 2006;Cremers et al., 2010), connectivity analyses were performed to study the relationship between reward brain networks and reward sensitivity. Once we had identified the ROIs that showed effects of reward sensitivity on task-related activations, we performed PPI analyses (Friston et al., 1997), using these ROIs as source (seed) regions to study whether connectivity among these areas and the other ROIs was also related to reward sensitivity. These connectivity analyses were performed for the same contrasts of interest (psychological variables: reward vs. punishment, high incentive vs. low incentive, and reward vs. neutral) used to study task-related activations associated with SR scores. This resulted in a total of six independent PPI analyses: two regions (right NAcc and midbrain; see Results), each with three psychological variables (each contrast of interest). For each participant, we extracted the time series from the first eigenvariate of all active voxels within the right NAcc and midbrain ROIs (seed regions). Then, the time series were deconvolved, and each PPI was calculated as the element-by-element product of the deconvolved time series and a vector representing the psychological variable (Gitelman et al., 2003). These products were subsequently reconvolved with the hemodynamic response function and entered as regressors in a first-level analysis together with the physiological variable (the time series extracted from the seed region) and the vector of the psychological variable.
Then, we performed second-level analyses, including the PPI regression coefficients (changes in connectivity) in: (i) a one-sample t-test to assess positive or negative changes in connectivity at the group level in each described PPI; and (ii) multiple regression analyses with SR scores as a regressor of interest and age as a covariate, to investigate the relationship between each measure of connectivity change and SR scores. Once again, the analyses were restricted to the ROI (SVC, P < 0.05, FWE corrected).

Behavioral data analyses
The percentage of hits (successful responses) and mean of RTs were recorded for each participant. The hits and RTs for each incentive condition were used to perform two different 2 9 2 [valence (reward, punishment) 9 incentive magnitude (high, low)] repeated-measures within-subjects ANOVAs to study cue-related effects on behavioral performance. To investigate personality effects, we used correlations and partial correlations with performance variables (hits and RTs).

Behavioral results
Mean RTs and percentages of hits are shown in Fig. 1 and Table 1. The repeated-measures ANOVA for RTs showed main effects of valence (F 1,43 = 5.69, P = 0.022) and incentive magnitude (F 1,43 = 5.87, P = 0.02), indicating faster RTs for reward than for punishment conditions, and for high than for low incentive conditions. These main effects were qualified by the significant valence 9 incentive magnitude interaction (F 1,43 = 12.27, P = 0.001), indicating that participants responded faster after high reward cues than for the rest of the conditions. The repeated-measures ANOVA for hits also showed significant main effects of valence (F 1,43 = 15.65, P < 0.001) and incentive magnitude (F 1,43 = 9.24, P = 0.004), but the effect for the valence 9 incentive magnitude interaction did not reach significance (P > 0.1). Post hoc analyses indicated that the percentage of hits was higher for high than for low incentive cues and for reward than for punishment cues.
The results of Pearson correlations and partial correlations between SR scores and performance are shown in Table 1. Only the correlation between SR scores and RTs for high incentive cues reached significance when the effect of low incentive cues was controlled, confirming that individuals with higher SR scores responded faster in high incentive conditions.

fMRI results
The results from the overall task (Fig. 2) showed stronger NAcc activation for reward cues than punishment cues (right: 9, 12, 0; Z- SD, standard deviation. The significant correlation at P < 0.05 (two-tailed) is presented in bold. *Partial correlation of reward cues' related variable controlled by punishment cues. † Partial correlation of high incentive cues' related variable controlled by low incentive cues. There were no significant associations for each independent condition without taking into account the effects of lower-level conditions. Fig. 1. Mean RTs and accuracies during MID task completion for each condition. Left: mean RTs for each incentive condition. Right: percentage of hits for each incentive condition. Dark bars represent high incentive conditions, and white bars represent low incentive conditions. Post hoc analyses showed that the high reward condition yielded a higher percentage of hits and lower RTs than the rest of the incentive conditions (P < 0.05). score = 5.35; P < 0.05, FWE corrected; cluster size, 486 mm 3 ) (left: À9, 6, 0; Z-score = 3.97; P < 0.05, FWE corrected; cluster size, 405 mm 3 ), and for high incentive cues than low incentive cues (right: 15, 12, À6; Z-score = 4.07; P < 0.05, FWE corrected; cluster size, 270 mm 3 ). In addition, the midbrain was more activated in response to high incentive cues than in reponse to low incentive cues (À6, À21, À12; Z-score = 2.92; P < 0.05, FWE corrected; cluster size, 81 mm 3 ). These results are in agreement with those of previous studies that have shown midbrain response according to stimulus incentive magnitude (Knutson et al., 2005). Finally, the comparison of reward cues and the non-incentive cue showed activation in the bilateral NAcc (right: 9, 12, 0; Z-score = 6.57; P < 0.05, FWE corrected; cluster size, 999 mm 3 ) (left: À9, 6, 0; Z-score = 6.21; P < 0.05, FWE corrected; cluster size, 1161 mm 3 ) and midbrain (À3, À24, À9; Z-score = 4.09; P < 0.05, FWE corrected; cluster size, 378 mm 3 ), in agreement with the greater response of these areas to high reward cues. Whole brain voxel-wise results from the overall task are summarized in the Supporting Information (Table S1 and Fig. S2, S3, S4).

Effects of reward sensitivity on task-related activations
Multiple regression analyses showed that SR scores correlated positively with right NAcc activation (12, 6, À6; Z-score = 3.20; P < 0.05, FWE corrected; cluster size, 108 mm 3 ) and left midbrain activation (À3, À18, À15; Z-score = 4.07; P < 0.05, FWE corrected; cluster size, 135 mm 3 ) for reward cues as compared with punishment cues (Fig. 3a). Furthermore, SR scores did not correlate with activation in the ROIs for the comparison either between high incentive cues and low incentive cues, or between reward and nonincentive cues. Thus, these findings showed modulation of the NAcc and midbrain activation by reward sensitivity during reward processing, as shown in previous studies (Carter et al., 2009;Hahn et al., 2009). Therefore, these ROIs were used for later PPI analysis, as previously described (see Materials and methods). No other positive or negative correlations were found.

PPI results
One-sample t-test analyses did not show any significant main effects (positive or negative) for any ROI. Nevertheless, we found that reward sensitivity modulated changes in connectivity among rewardrelated brain areas under incentive processing. More specifically, we found a negative association between midbrain-mOFC connectivity and SR scores for high vs. low incentive cues (0, 36, À15; Zscore = 3.73; P < 0.05, FWE corrected; cluster size, 297 mm 3 ; Fig. 3b). Thus, the connectivity between the midbrain and mOFC during incentive processing is dependent on individual differences in reward sensitivity. In order to study whether this effect was driven by reward or punishment cues, we performed two PPI analyses with high vs. low reward cues and high vs. low punishment cues separately as psychological variables. We did not observe any significant effect of SR scores for these two contrasts at predefined statistical thresholds. Thus, we may conclude that the reported effects result from the high incentive condition rather than being driven by either of the valence conditions. Likewise, no positive or negative correlations with SR scores and brain connectivity were found regarding valence processing. Additionally, analysis of connectivity when we compared reward cues and the non-incentive cue showed that connectivity between the NAcc and left amygdala was negatively associated with SR scores (À21, 0, À15; Z-score = 3.23; P < 0.05, FWE corrected; cluster size, 189 mm 3 ; Fig. 3c). This finding represents modulation of connectivity between the NAcc and left amygdala by reward sensitivity during the processing of reward cues as compared with neutral cues.

Discussion
In this study, we have shown that individual differences in reward sensitivity modulate neural connectivity between the midbrain and mOFC under high incentive conditions, independently of the anticipation of possible wins or losses. We also found that activity in the NAcc and midbrain is stronger for individuals with higher SR scores, which is consistent with previous reports (Carter et al., 2009;Hahn et al., 2009). Crucially, our results showed that the trait of reward sensitivity modulates brain activity but also connectivity among reward-related brain regions.
In our study, SR scores were linked to increased activity in the NAcc and midbrain during the processing of reward cues as compared with punishment cues. These results replicated those of a previous study using the MID task, in which reward and punishment conditions were included (Carter et al., 2009). In contrast, in disagreement with our hypothesis, we did not find an association between NAcc response and SR scores for reward cues as compared with neutral cues, an extension of results reported by Hahn et al. (2009). One explanation for this negative result may be that the effects of individual differences in reward sensitivity on striatum activity were only driven by the motivation component of reward anticipation, and not by its motor preparation component. That is, the anticipation of a motor response when a reward cue was present did not modulate the association between reward sensitivity and the NAcc, or at least not in the same direction. Thus, this result implies an importance of motivational contingencies in earlier behavioral studies in which RTs under reward conditioning were modulated by reward sensitivity (see . Future studies targeting these interaction effects could better clarify the neural basis of modulation of motor or cognitive responses by reward sensitivity under reward cueing. On the other hand, in a previous study, Hahn et al. (2009) used a modified version of the MID task in which only reward (not punishment) conditions were included, and their results may have involved different contextual effects for modulation of the previously described brain activation by reward sensitivity (Patterson & Newman, 1993;. Moreover, the difference in effects of reward sensitivity on NAcc activity between our research and the study of Hahn et al. (2009) may be related to previous findings with the MID task that demonstrated NAcc modulation by available alternative incentives, with the worst available alternative being an anchor for NAcc activation (Cooper et al., 2009). Therefore, modulation of dopaminergic activity by reward sensitivity during reward anticipation may be dependent on the referenced worst available alternative, inducing different contextual effects in different event designs. On the other hand, the NAcc has been shown to be involved in both approach and active avoidance behaviors (Salomone et al., 1997). Our results could be interpreted as primary modulation of the NAcc by reward sensitivity during approach anticipation as compared with active avoidance, or as opposite modulation of the NAcc by reward sensitivity during approach and active avoidance behaviors. In future studies, it will be important to consider both contextual and condition effects to analyse how the reward sensitivity specifically modulates the response of the NAcc and midbrain to reward cues.
The crucial result of our study is that reward sensitivity modulates neural dynamics among reward brain areas. Higher SR scores were associated with relatively less connectivity between the midbrain and mOFC during processing of high incentive cues. That is, the activity of the midbrain during the processing of high incentive cues seems to be more dependent on the mOFC in individuals with lower reward sensitivity. The mOFC is involved in processing reward outcomes (Haber & Knutson, 2010), and lesions to it cause increased reward sensitivity (Bechara et al., 2000). Previous results with the MID task related the midbrain to the cue's incentive magnitude independently of its valence (Knutson et al., 2005). Overall, the caudal VTA might contribute to enhance learning in the novelty-processing and/or reward-processing contexts (Krebs et al., 2011). Animal studies have shown that the mOFC is implicated in the regulation of dopaminergic neuron activity Tong et al., 1996Tong et al., , 1998Aston-Jones et al., 2009), in that electrical stimulation of the orbitofrontal cortex induces both inhibitory and excitatory responses in dopaminergic neurons (Lodge, 2011). Specifically, Sesack et al. (2003) reported that glutamatergic neurons from the prefrontal cortex selectively target dopaminergic mesocortical neurons and GABAergic mesoaccumbens neurons, suggesting that prefrontal cortex glutamatergic firing leads to inhibition of mesoaccumbens dopaminergic neurons, whereas prefrontal cortex hypofunction may promote subcortical dopaminergic transmission. Therefore, the effect of the mOFC on midbrain activity may reflect individual differences in reward sensitivity during the processing of high incentive stimuli. Moreover, we should note that cues involve reward and active avoidance anticipation, two processes that were suggested to be subserved by reward sensitivity and the behavioral activation system from the reinforcement sensitivity theory. Further studies may contribute to clarifying the role of this coupling in the relevancy effect of salient stimuli and the maintenance of reward-seeking and active avoidance behavior in individuals with strong reward sensitivity (Takahashi et al., 2009).
The connectivity between the right NAcc and left amygdala during anticipation of reward cues was modulated by individual differences in reward sensitivity, involving motor preparation for response that correlated negatively with SR scores. This indicates that participants with high reward sensitivity had relatively less connectivity between the NAcc and amygdala when processing reward cues. The amygdala is a brain area composed of a group of nuclei involved in emotional learning and expression (Cardinal et al., 2002). Despite this area being classically linked to fear and anxiety processing, it is actually thought to play a more general role in encoding and updating the motivational and affective value of stimuli (Cardinal et al., 2002;Gottfried et al., 2003;Morrison & Salzman, 2010;Seymour & Dolan, 2008). The amygdala may contribute to goal-directed behavior though direct projections to the NAcc and other regions of the striatum (Friedman et al., 2002;Fudge et al., 2002;Haber & Knutson, 2010), and through projections to dopaminergic areas in the midbrain (Cardinal et al., 2002;Pauli et al., 2012). This network is thought to be important for learning stimulus-reward associations (Everitt et al., 1989;Murray, 2007;Pauli et al., 2012) and maintaining a representation of affective or rewarding properties of conditioned cues (Cardinal et al., 2002). Thus, lower connectivity between the NAcc and amygdala in individuals with greater reward sensitivity may represent lower flexibility in updating reward value. Consistent with this result, previous findings showed that amygdala lesions promote the selection of immediate rather than larger delayed rewards (Winstanley et al., 2004), reduce aversion to monetary loss (De Martino et al., 2010), increase risk choices when considering potential gains (Weller et al., 2007), and increase the selection of high reward but ultimately high punishment decks in the Iowa Gambling Task (Bechara et al., 1999). Thus, the amygdala seems to be crucial for appropriate decision-making, and its impairment may cause impulsive choices. Finally, the lack of modulation of NAcc-amygdala connectivity by reward sensitivity when reward cues are compared with punishment cues may be explained by the supposed role of the amygdala in processing both reward and punishment stimuli.

Limitations
A limitation of this study is inherent to the interpretation of PPI analyses. The PPI in itself is insufficient to assess the direction of effects. This is an important limitation, considering, for example, the argued roles of the mOFC and amygdala in regulating the activity of the midbrain and NAcc, respectively. Nonetheless, other studies, applying different methodologies, have provided more direct evidence for a top-down regulatory role in these networks Tong et al., 1996Tong et al., , 1998Aston-Jones et al., 2009;Stuber et al., 2011). On the other hand, the ROI definition of the midbrain (6-mm sphere) may include non-dopaminergic neurons in the region. We used this approach to study the midbrain because of the impossibility of uniquely selecting dopaminergic neurons of the VTA-SN complex in fMRI analyses. Likewise, it is important to note that the midbrain effects may not be exclusively mediated by dopaminergic neurons. Finally, we must be cautious in interpreting results obtained with the neutral condition as the control condition (i.e. NAcc-amygdala connectivity). These results may be driven by anticipation of incentive, preparation of motor responses for the attainment of objectives, or both. However, the neutral condition in this design did not involve a motor response, for two reasons: first, preparation of motor responses was better controlled by the other incentive conditions; and second, the study had the secondary objective of analysing modulation of reward response anticipation by reward sensitivity. This was of particular interest, given the focus of our research group (see Avila & Parcet, 1997, 2001Avila, 2001;Avila et al., 2003), in that both processes involve the striatum (Guitart-Masip et al., 2011).
To summarize, in this study we have replicated previous findings showing that reward sensitivity modulates brain activity in the NAcc and midbrain. In addition, we have demonstrated that reward sensitivity also modulates connectivity of the midbrain and NAcc with the mOFC and amygdala respectively. Our results suggest that high reward sensitivity-related activation in reward brain areas may partially result from associated diminished modulatory effects of other brain regions within the reward circuitry.

Supporting Information
Additional supporting information can be found in the online version of this article: Fig. S1. Task structure. Fig. S2. Brain areas showing significant activity for anticipation of reward vs. punishment. Fig. S3. Brain areas showing significant activity for anticipation of high incentive vs. low incentive. Fig. S4. Brain areas showing significant activity for anticipation of reward vs. the non-incentive condition. Table S1. (a) Brain areas showing significant activity for anticipation of reward vs. punishment. (b) Brain areas showing significant activity for anticipation of high incentive vs. low incentive. (c) Brain areas showing significant activity for anticipation of reward vs. the non-incentive condition.