Human mesostriatal response tracks motivational tendencies under naturalistic goal conflict

Abstract

Goal conflict situations, involving the simultaneous presence of reward and punishment, occur commonly in real life, and reflect well-known individual differences in the behavioral tendency to approach or avoid. However, despite accumulating neural depiction of motivational processing, the investigation of naturalistic approach behavior and its interplay with individual tendencies is remarkably lacking. We developed a novel ecological interactive scenario which triggers motivational behavior under high or low goal conflict conditions. Fifty-five healthy subjects played the game during a functional magnetic resonance imaging scan. A machine-learning approach was applied to classify approach/avoidance behaviors during the game. To achieve an independent measure of individual tendencies, an integrative profile was composed from three established theoretical models. Results demonstrated that approach under high relative to low conflict involved increased activity in the ventral tegmental area (VTA), peri-aquaductal gray, ventral striatum (VS) and precuneus. Notably, only VS and VTA activations during high conflict discriminated between approach/avoidance personality profiles, suggesting that the relationship between individual personality and naturalistic motivational tendencies is uniquely associated with the mesostriatal pathway. VTA–VS further demonstrated stronger coupling during high vs low conflict. These findings are the first to unravel the multilevel relationship among personality profile, approach tendencies in naturalistic set-up and their underlying neural manifestation, thus enabling new avenues for investigating approach-related psychopathologies.

Introduction

Motivation is a basic survival mechanism, necessary to assess environmental cues of reward or punishment and facilitate approach or avoidance behaviors, ultimately promoting adaptive interactions with the environment. Accordingly, inappropriate motivational behavior is known to result in psychopathological symptoms, such as excessive drug consumption in addiction (Diekhof et al., 2008), generalized avoidance behavior in anxiety (McNaughton and Corr, 2008) or altered incentive drives in affective disorders (Gonen et al., 2014).

In recent years neuroimaging studies of motivation have elucidated several stages in this complex stimulus–response scenario. Findings regarding the initial valuation and processing of reinforcements point first and foremost to the relevance of the dopaminergic mesostriatal pathway, including the ventral striatum (VS) and ventral tegmental area (VTA) (Haber and Knutson, 2009), in both animals (Cardinal et al., 2002) and humans (Liu et al., 2011). Accordingly, a recent functional magnetic resonance imaging (fMRI) study demonstrated that the facilitating effects of incentive motivation involved the caudate and putamen (Miller et al., 2014).

Nonetheless, despite our growing knowledge of motivational processing, including cost-benefit valuation during behavioral decision making (Basten et al., 2010; Park et al., 2011), surprisingly few studies examined the neural processes underlying the behavioral phase of incentive motivation: approach or avoidance (Bach et al., 2014). The paucity of neuroimaging studies investigating motivational behavior could result from the conventional operationalization used in functional imaging, of static stimuli presented in a trial-by-trial manner (c.f. Liu et al., 2011), which does not allow for characterization of dynamic ongoing motivational behavior.

The long established link between personality factors and individual tendencies to approach rewards or avoid punishments makes the depiction of dynamic motivational behavior even more important. It was proposed that individual sensitivity to reward and punishment (RS and PS, respectively) forms the main building blocks of personality structure, and manifest in behavioral tendency to approach or avoid (Torrubia, 2001) via dedicated neuro-behavioral processes (Corr and McNaughton, 2012). Two additional psychological models related to motivational tendencies are (i) the ‘Big-Five’ phenomenological model (Costa and McCrae, 1992), consisting of five broad personality traits, in which Extraversion was associated with approach and Neuroticism with avoidance (Cunningham et al., 2011; Quilty et al., 2014) and (ii) the ‘Tridimensional model’, composed of three main trait profiles that correspond to neurochemical systems, including Reward Dependence (RD) related to approach tendency and high dopamine vs Harm Avoidance (HA) related to avoidance and high serotonin (Cloninger, 1987). However, differential neural manifestations of personality related to actual motivational behavior have not been adequately portrayed (Smillie, 2008).

Our main objective in this study was to develop a task which manipulates motivational behavior in a real-life naturalistic manner, allowing delineation of the dynamic motivational experience, and shedding light on the way personality profile is reflected in motivational behavior. To broaden the range of individual behavior, we applied goal conflict (GC) events during the game in which approaching a reward is threatened by a possible punishment. Such events accentuate individual differences in approach or avoid motivational behavior by increasing the complexity of motivational decision making. Additionally, we composed an integrative personality profile by characterizing our study sample using all three approaches to personality mentioned above. Altogether, we aimed to depict the behavioral and neural manifestation of naturalistic approach behavior, specifically under GC, as a function of our integrative personality profile.

We developed a novel interactive game scenario intended to be played during fMRI scan. The game was designed to manipulate motivational behaviors in a real-life manner by creating a dynamic, rapidly changing environment with changing levels of GC. To provide such an environment, GC was either high (HiGC), where approach behavior (toward a reward) was highly threatened by punishing cues, or low (LoGC), where no concrete threats were present on the screen.

Based on previous investigations of incentive processes described above, along with evidence relating mesostriatal dopaminergic transmission in the VTA–VS to exertion of effort (Haber and Knutson, 2009; Salamone and Correa, 2012), we hypothesized that approach behavior under HiGC, associated with greater incentive motivation, would specifically involve activation in the VTA and VS. Previous animal studies related this VTA–VS dopaminergic pathway to variations in personality traits such as extraversion or RS, commonly reflected in elevated behavioral facilitation (Gray, 1982; Depue and Collins, 1999). Furthermore, as dopamine transmission in the VTA was shown to reflect individual differences in human effort-based behavior (Treadway et al., 2012b), we predicted that these mesostriatal regions would reflect individual differences in personality as well, so that individuals with stronger personality markers of approach tendency (as depicted by the integrative personality profile), would apply more approach behaviors during the game and show greater VTA–VS activation during HiGC situations.

Methods

Participants

Fifty-five healthy subjects (28 ± 5 years old, 26 women) participated in the study. Subjects had at least 12 years of education, no reported history of psychiatric or neurological disorders, no current use of psychoactive drugs and no family history of major psychiatric disorders. The study protocol was approved by the TLVMC Ethics Committee. All subjects provided written informed consent before participation, and were paid for participation. Data from nine subjects was discarded from the imaging analyses due to technical issues and exaggerated head motions. Thus, 55 and 46 (28.37 ± 4.55 years old, 22 women) valid datasets were included in the behavioral and imaging analyses, respectively.

Dynamic motivation task

The punishment, reward and incentive motivation game (PRIMO)

We developed a novel ecological interactive computer game (Java1.6, Oracle, Redwood-Shores, CA &Processing package, http://www.processing.org), constructed to manipulate approach and avoidance behaviors during a naturalistic GC, while enabling dissociation of incentive behavior. The goal of the game was to earn money by catching coins and avoiding balls. To ensure incentive motivation and promote approach behavior, subjects were told, as part of the experimental manipulation, that at the end of the experiment they would be paid the total amount earned in the game. However, at the end of the game the incentive manipulation was revealed, and all subjects were paid a fixed amount, higher than the possible final gain in the game, and equal to the standard payment for participation in an fMRI experiment in our center.

There were two ways to gain or lose money—‘controlled’, where the player actively approached coins and avoided balls, and ‘uncontrolled’, where the player was hit by random coins and balls (Figure 1). Each coin-catch resulted in a 5-point gain and each ball hit resulted in a loss of five points, regardless of controllability. To create an ecological environment, the difficulty level of the game was modified every 10 s according to the local and global performance of the player. By dynamically adjusting the difficulty level and actively balancing the number of uncontrolled events, the game was tailored to match each player’s skills and all event types occurred roughly at the same frequency. To prevent overlap of the hemodynamic response function, each trial was separated by a jittered inter-stimulus interval (ISI), which varied randomly between 550 and 2050 ms (Clark et al., 2001). To construct HiGC and LoGC trials, the number of obstacles (i.e. balls) placed between the player and the falling coin changed in each trial. Trials with 0–1 balls between the player and the falling coin were defined as LoGC trials, while trials with 2–6 balls between the player and the coin were defined as HiGC. A slight bias toward the controlled reward outcome was intentionally predesigned in order to keep subjects’ incentive motivation throughout the game. The game was played for four sessions of 6 minutes (each as a separate scan run), starting with 1 minute fixation point to establish a baseline condition. At the end of each session subjects rated their feelings and attention toward each condition of the game on a 9-point Likert scale. Prior to scanning, subjects were introduced to the game and played it for 1 minute to ensure that instructions were fully understood.

Fig. 1.

Schematic presentation of the PRIMO game. The goal of the game is to earn money by catching coins and avoiding balls. To ensure incentive motivation, subjects were told they would be paid the total amount they earned throughout the game. There are two ways to gain or lose money: controlled, the player actively approaches coins and avoids balls (upper panel); no control, an animated figure throws random coins and balls that hit the player (lower panel). Subjects are explicitly told at the beginning of the game that these events are uncontrollable (although it is still possible for them to move). Controlled and no control events were pseudo-randomly presented throughout the paradigm to ensure similar occurrence of each trial across subjects as well as a similar final score. For controlled conditions: each trial begins when a coin appears at the top of the screen (upper left panel); the subject chooses to either approach and move toward the coin or avoid and move away from balls (motivational behavior: upper right panel, duration: 1000–3000 ms). For no control conditions: each trial begins when a flying figure appears on the screen (lower left panel), the player is then anticipating the coin/ball that falls and ‘chases’ him (motivation anticipation: lower right panel, duration: 1000–3000 ms). Trials were separated by 0.5–5.5 s ISI).

Behavioral classification analyses

One of the ecological features of the game was that the motivational behavior played in each trial was sequentially chosen by the subject according to his\her individual motivational tendency and specific game course. Therefore, labeling of the events into approach or avoidance was performed retrospectively using a regularized logistic regression classification model, which was generated in two independent stages. First, three independent judges manually labeled all related events using an in-house labeling software and based on strict criteria (see Supplementary Information). Next, an independent researcher, unaware of the manual classification criteria, was asked to identify the multivariate features that best explained these labels. The data were randomly divided into three uneven parts (train = 60%, cross validation = 20%, test = 20%) and rigorous data mining algorithms and mathematical transformations were applied to the training dataset to uncover related features (Figure 2A and B). These were cross-validated, using various feature space subsets within the confinements of the training dataset (n > 6000). In addition, misclassified events were further explored using different Machine Learning (ML) approaches (linear and non-linear supervised classification) that enabled us to uncover stronger features (Figure 2B). Finally, a stage of feature elimination was performed in order to both improve understanding of the various uncovered ‘rules’ and compare them with the initial manual classification labeling (Figure 2C, for full details regarding the development of this classification algorithm, see Supplementary Information).

Fig. 2.

Data visualization of selected output features, mined metrics and their classification power. (A) Graphical representation of one complete game session, where the x-axis represents the total session time (6 min), while the y-axis shows selected game features: difficulty, controlled (Con) punishment, uncontrolled (NoC) punishment, NoC reward, Con reward, outcome (loss in red or win in green) and keyboard input (movement to the right is presented in orange and to the left in blue). The pseudo random manner of event distribution is evident from NoC and Con events. It is easy to spot the non-linear fashion of task difficulty (top panel: larger is more difficult). In addition, punishment events become faster (width of area represents duration on screen, while height represents number of punishment objects at any given time); as a function of difficulty, however, their density is not dependent on difficulty. Finally, it is clear that as the game increases in difficulty the density of keyboard interaction increases. (B) Graphical visualization of one controlled event, where the y-axis represents time in milliseconds and the x-axis represents the width of the computer screen on which the paradigm was projected. The blue line represents the player’s movement vector, while the $ sign and red circles represents the reward and punishment stimuli, respectively. Input bar is shown on the right, where black sections represent idle events (i.e. no input is registered) and orange and blue sections represent right and left key events, respectively. (C) Distance from reward vs range game features are shown for events that ended with reward miss. Combination of these features demonstrated an efficient and simple classification of motivational behavioral space (approach and avoidance). Green and red represent approach and avoidance, respectively.

Conditions of interest

The Punishment, Reward and Incentive Motivation (PRIMO) game was built to target spontaneous approach behavior in the presence of incentives and threats. As such, each controlled trial was defined as the period where a reward was present on the screen, that is, when a coin appeared at the top of the screen, and lasted until it had been caught, or had disappeared at the bottom of the screen. For uncontrolled conditions, each trial was defined as the period between the appearance of the flying figure on the screen and the hit of the coin. By subtracting these uncontrolled conditions from the controlled ones, we were able to control for the anticipation and hedonic emotional responses to the reinforcing stimuli and extract the neural substrates of incentive motivational behavior. Within these controlled periods, we contrasted ‘HiGC vs LoGC approach’ conditions for specific characterization of approach behavior under high conflict.

Functional MRI

For fMRI acquisition and data preprocessing details, see Supplementary Information.

Whole-brain analyses

The initial experimental design included three main factors, each with two levels: motivation (approach/avoid), controllability (uncontrolled/controlled) and conflict (high/low). However, considering the small number of avoidance trials played in the game and the large difference in the occurrence of approach vs avoidance trials (see Results), we decided to focus entirely on approach behavior, and discarded the analysis of avoidance data which seemed less reliable due to these limitations. Therefore, a random-effects (RFX) General linear Model (GLM) analysis was conducted with the following predictors: ‘approach under LoGC’, ‘approach under HiGC’ and ‘uncontrolled reward’. The predictors were convolved with a standard canonical hemodynamic response function. Whole-brain analyses were applied under the following contrasts of interest: approach (both HiGC and LoGC) vs uncontrolled reward, for the neural correlates of overall dynamic approach behavior; and HiGC vs LoGC approach for specific characterization of approach behavior under HiGC. Because a greater number of movements (executed via button presses, each 5–10 pixel wide) were performed in the controlled vs uncontrolled conditions; subjects’ button presses throughout the game were regressed out of the model to avoid confound of BOLD activations related to actual movements (for details see Supplementary Information).

Region of Interest (ROI) analysis

Differences in neural patterns between the two personality groups were first evaluated in the two hypothesis-driven motivational facilitation nodes: the VTA and VS. Bilateral 3 mm diameter ROIs were functionally extracted from the contrast of HiGC vs LoGC approach surrounding the peak of activation in the bilateral VS and VTA regions. To validate the specificity of the VTA ROI within the midbrain and following Murty et al. (2014), we used it as a seed for functional connectivity (FC) analysis in an independent resting state scan performed on the study sample as part of the scanning session (see Supplementary Information for details). Repeated-measures Analysis of Variance (ANOVA) was conducted on the beta weights of HiGC approach for each ROI to assess group differences in BOLD activation. Effect size was calculated using Cohen’s d. Next, an exploratory post hoc analysis was conducted for all regions with significant activations under the contrast of HiGC vs LoGC approach (Table 1). ROIs were extracted in the same manner (3 mm sphere surrounding peak of activation) and comparisons were made between groups for each ROI using independent t-tests. The significance threshold for post hoc comparisons was corrected for multiple comparisons using Bonferonni correction and was set to P = 0.0036 (0.05 × 12 comparisons).

Table 1.

Peak of activations obtained from the whole-brain contrast of HiGC vs LoGC approach conditions

Regiona	Peak voxel (x, y, z)b			t(45)	P valuec
HiGC > LoGC
L VTA	−13	−14	−6	6.334	<0.001
L occipital (BA 17-18)	−13	−89	−9	6.118	<0.001
Bilateral precuneus	−10	−71	45	8.483	<0.001
R occipital (BA17-18)	14	−86	−6	6.827	<0.001
R premotor	23	−5	51	6.046	<0.001
L VS	−16	1	3	5.175	<0.001
R pulvinar	17	−26	9	4.873	<0.001
R VTA	11	−17	−3	4.848	<0.001
L premotor	−22	−8	57	4.721	<0.001
L pulvinar	−13	−23	9	4.094	<0.001
PAG	−4	−29	−3	4.092	<0.001
R VS	23	7	0	3.893	<0.001
R MFG	29	34	36	3.851	<0.001
LoGC > HiGC
R STG angular	41	−11	3	−5.723	<0.001
L STG angular	−52	−14	9	−5.623	<0.001
L vlPFC	−49	22	21	−5.208	<0.001
mPFC	2	43	9	−4.666	<0.001
R vlPFC	41	46	9	−4.443	<0.001
R SFG	14	28	39	−4.420	<0.001
PCC	−4	−32	30	−3.998	<0.001
R IFG	38	16	24	−3.893	<0.001

R = right; L = left.

^aMinimal cluster = 100 voxels.

^bTalairach coordination.

^cWhole-brain map threshold: P < 0.001, FDR corrected.

Functional connectivity psychophysiological interactions analysis

A whole-brain psychophysiological interactions (Friston et al., 1997; O’Reilly et al., 2012) random-effects GLM analysis was performed in order to inspect task-specific changes in the connectivity between HiGC and LoCG with the VTA as a seed region. Regressors included the following: the psychological variable (original regressor of the specific experimental condition); the physiological variable (time course activity in the seed ROI) and the interaction variable (an element-by-element product of the psychological and physiological variables).

Skin conductance

Skin conductance (SC) signal was used to validate that the incentive (controlled) conditions of the PRIMO game were experienced as such. We expected participants to exhibit greater SC, due to greater arousal and motivational significance, during controlled vs uncontrolled conditions. For details regarding acquisition and analyses of SC, see Supplementary Information.

Integrative account of personality profile

Personality questionnaires

Following the fMRI scan, all subjects completed three personality questionnaires: (i) the ‘Neuroticism-Extraversion-Openness Five-Factor-Inventory’ (NEO-FFI) (Costa and McCrae, 1992) is a widely used self-report personality inventory of 60 questions. This model includes the following traits: neuroticism, extraversion, openness, agreeableness and conscientiousness. (ii) The ‘Tridimensional Personality Questionnaire’ (TPQ) (Cloninger, 1987) is partly based on biological premises, relating each personality trait to a specific neurotransmitter activity. It includes 100 true/false questions and three axes: novelty seeking (NS), HA and RD. (iii) The ‘Sensitivity to Reward and Punishment Questionnaire (SPSRQ) (Torrubia, 2001) is a 24-item questionnaire measuring both sensitivity to punishment (SP) and sensitivity to reward (SR). This is an extensively validated tool to assess individual differences in motivational tendencies (Smillie and Jackson, 2005).

Data driven clustering

To obtain an integrative view of individual differences combining the three different theoretical models of personality, we divided our sample into two groups aiming to create approach vs avoidance tendency groups. We performed a two-stage non-hierarchical K-means cluster analysis creating two distinct clusters (using SPSS20 for Windows). First, we performed the analysis based on normalized scores of all traits from the three personality models. We then excluded the traits which did not return a significant difference between the two clusters in the ANOVA conducted as part of the K-means procedure. Excluded traits were openness and conscientiousness from the NEO-FFI, the NS from the TPQ, and the RS from the SPSRQ. Next, we performed a second K-means cluster analysis (again, into two clusters) which included the following traits: neuroticism, extraversion and agreeableness from NEO-FFI, RD and HA from TPQ, and PS from SPSRQ. Finally, we compared the behavioral and neural patterns between clusters. Behavioral patterns were modeled as indices of overall approach, calculated as the proportion of approach trials played from all trials in the game, and HiGC approach, calculated as the proportion of HiGC approach trials played from all HiGC trials in the game.

Results

The PRIMO game validation

We first validated the subjects’ engagement and affective experience in the game according to our design. To this end, we used self-ratings of emotion and attention and physiological measures of SC. Figure 3 shows that indeed controlled conditions had greater emotional value as reflected in ratings, and greater motivational significance as reflected by SC response (for more details, see Figure 3A-C and Supplementary Information).

Fig. 3.

Validation of the PRIMO game. (A) Subjects rated the attention paid to each of the game’s elements after each session. There was no difference in the attention level across sessions (F(3,159) = 2.679, P < 0.068). There was a significant difference between the attention given to the controlled (rewards and punishments, Con) conditions and the no control (NoC) ones (F(3,159) = 44.115, P < 0.001, N = 55). (B) Emotional ratings: subjects rated rewards positively and punishments negatively. Within each valance, the Con conditions were rated higher than NoC ones (F(9,477) = 2.353, P < 0.013; t(53) = −2.484, P = 0.016, N = 55). (C) A significant difference in phasic Skin Conductance Response (SCR) between Con and NoC conditions of the game revealed controlled conditions had greater motivational significance than uncontrolled (t(18) = 3.97, P = 0.0003, N = 19). (D) Average number of approach and avoidance behavior by session. There was a significant difference in the occurrence of approach trials between sessions, with less trials in sessions 1 and 3 relative to sessions 2 and 4 (F(3,52) = 6.580, P = 0.001, N = 55). (E) Significant differences were also found for approach behavior between HiGC and LoGC conditions, with more approach behaviors played under LoGC than under HiGC (F(1, 54) = 143.708, P < 0.001, N = 55). Error bars in all graphs represent SEM.

Motivational behavior

Individual motivational behavior was retrospectively classified into approach or avoidance trials using a machine learning–based approach (see Methods and Supplementary Information). Following classification, repeated-measures ANOVA was conducted to examine the number of approach and avoidance trials over game sessions. There were more approach than avoidance trials across sessions [main effect of behavior: F(1,54) = 2437.75, P < 0.001, mean (s.d.): 38.53 (3.63) and 5.51 (3.21) for approach and avoidance, respectively], confirming that the incentive motivation manipulation designed to encourage approach behavior was successful. There was, however, a significant difference in the occurrence of approach trials between sessions, with less trials in sessions one and three relative to sessions two and four [main effect of session: F(3,52) = 6.580, P = 0.001, mean (s.d.): 36.4 (7.26), 41.49 (6.86), 36.47 7.70) and 39.75(9.57) for sessions 1–4, respectively; Figure 3D]. These differences in event occurrence were the result of the changing difficulty level throughout the game, which adjusted itself continuously with respect to individual performance. Thus, session 1 began with a very low difficulty level to allow an easy start and training phase (the first session was regarded as a training session and was not included in the imaging data analyses), while in session 3, following a well-trained and successful session 2, there was a peak in difficulty level across subjects, which probably resulted in more cautious behavior and thus less approach trials.

Significant differences were also found for approach behavior across HiGC and LoGC conditions, with more approach behaviors under LoGC than under HiGC [main effect of conflict: F(1,54) = 143.708, P < 0.001, mean (s.d.): 15.291 (0.344) and 22.236 (0.471) for approach behavior under HiGC and LoGC, respectively; Figure 3E]. This effect was stable, without significant differences between sessions. Importantly, the proportion of approach trials played by the subjects under LoGC conditions was 93%, while only 64% of the HiGC trials resulted in approach behavior, confirming a clear distinction between conflict levels. Accordingly, a smaller number of avoidance trials were played under LoGC conditions than under HiGC [main effect of conflict: F(1,54) = 55.043, P < 0.001, mean (s.d.): 13.45 (7.96) and 7.27 (5.18) for avoidance behavior under HiGC and LoGC, respectively].

Integrative profile of individual differences

K-means cluster analysis of scores from three personality questionnaires resulted in two clusters of 21 and 25 subjects each, which included differential representation of traits related to either approach or avoidance tendencies. Cluster 1, hereby called the ‘approach group’, was composed of 21 subjects with high scores in approach tendency traits, namely extraversion, agreeableness and RD (final cluster centers: 0.701, 0.489 and 0.431, respectively); and low scores in traits related to avoidance tendency: HA, neuroticism and PS (final cluster centers: −0.826, −0.755 and −0.691, respectively; Figure 4A, right panel). In contrast, the second cluster, hereby called the ‘avoidance group’, showed an opposite pattern, including 25 subjects with high scores in traits related to avoidance tendency (final cluster centers: 0.694, 0.635 and 0.581 for HA, neuroticism and PS, respectively), and low scores for traits related to approach tendency (final cluster centers: −0.589, −0.411 and −0.363 for extraversion, agreeableness and RD, respectively; Figure 4A, left panel). For full details regarding the K-means procedure, see Supplementary Information.

Fig. 4.

Individual differences. (A) K-means cluster analysis divided the study sample into an approach tendency group (left panel), holding subjects with high scores in the traits of extraversion, agreeableness and reward dependence, and low scores in the traits of HA, neuroticism and punishment sensitivity, and vice versa for the avoidance tendency group (right panel). Graphs depict mean Z-score for each trait in each cluster. (B) Significant difference in approach behavior between groups: subjects from the approach tendency group played more trials of overall approach as well as approach under HiGC than subjects from the avoidance tendency group (overall approach: t(34.96) = 2.536, P = 0.016; HiGC approach: t(39.79) = 2.021, P = 0.05).

Difference in behavioral tendencies

As hypothesized, the comparison of behavioral patterns between the two groups examined using normalized indices of approach and HiGC approach (for index calculation, see Methods) confirmed the relationship between personality traits and motivational behavior. Overall, the approach group chose to approach more trials in the game than the avoidance group [t(34.96) = 2.536, P = 0.016, mean (s.d.): 0.366 (0.54) and −0.308 (1.18) for approach and avoidance groups, respectively; Figure 4B]. Similarly, the approach group chose to play more approach trials under HiGC as well [t(39.79) = 2.021, P = 0.05, mean (s.d.): 0.303 (0.68) and −0.25 (1.15) for approach and avoidance groups, respectively; Figure 4B).

Neural correlates of approach under HiGC

To examine the neural correlates of dynamic approach behavior (regardless of conflict), we first contrasted the controlled approach trials with uncontrolled rewarding trials. This contrast was intended to extract the incentive component in approach behavior from the anticipation and valuation of a reward. For more details regarding this analysis, see Supplementary Information.

To evaluate the neural patterns of approach under HiGC, we compared whole-brain activations within the controlled approach conditions, between approach under HiGC and approach under LoGC. This contrast revealed that HiGC approach elicited greater activations in the VTA and VS as hypothesized, as well as in the peri-aquaductal gray (PAG), pulvinar, bilateral premotor cortices, middle frontal gyri, high visual areas and extensive activation of bilateral precuneus (Figure 5A, Table 1). In contrast, the LoGC approach condition elicited activations in the medial pre-frontal cortex (mPFC), anterior and posterior cingulate cortex (ACC, PCC), the lateral PFC (bilateral inferior frontal sulci), bilateral posterior insula and superior temporal gyri (Figure 5A, Table 1).

Fig. 5.

Approach under HiGC. (A) The contrast approach conditions under HiGC vs LoGC revealed a set of regions related to approach behavior under high and low risk of punishment, respectively. The HiGC condition elicited activations in the PAG (1), VTA (2) and VS (3), premotor cortices (4, only the right premotor is shown here) and middle frontal gyri (MFG) (5, only Right MFG is shown here) and in high visual areas (6) including bilateral cuneus and precuneus (7) The LoGC condition elicited activations in the mPFC/ACC (8) as well as the PCC (9) and ventro-lateral PFC (10). (B) BOLD differences between personality groups in ROIs related to approach under HiGC: subjects from the approach tendency group (green bar) demonstrated higher activity in the ventral striatum (VS, Talairach coordinates: [−16, 1,3] and [23,7,0] for left and right VS, respectively) and the VTA (Talairach coordinates: [−13, −14, −6] and [11, −17, −3] for left and right VTA, respectively) than subjects from the avoidance tendency group (red bar; t(44) = 2.123, P = 0.039 and t(44) = 2.051, P = 0.046 for VS and VTA, respectively; error bars represent SEM). No differences between groups were evident in other regions during approach under HiGC, as well as under LoGC (N = 46).

Individual differences in neural activation

According to our study hypothesis, we examined whether beta values in the VTA and VS during HiGC approach condition would differentiate between the two personality groups. As expected, results revealed a significant difference between groups (t(44) = 2.051, P = 0.046 and t(44) = 2.123, P = 0.039 for VTA and VS, respectively; Figure 5B). Moreover, in both ROIs the beta values during HiGC approach were higher for the approach group compared with the avoidance group (Figure 5B). Importantly, no differences were found in activation of these regions under LoGC conditions (t(44) = 0.444, P > 0.65 and t(44) = 0.931, P > 0.35 for VS and VTA, respectively). Finally, to examine further potential regions that could discriminate between the two groups, post hoc comparisons were conducted for all elicited regions in the HiGC vs LoGC approach contrast. No significant differences were found in any of the other regions (Table 2), including the PAG, which is located in great proximity to the VTA, or the mPFC, which was previously indicated as being related to individual differences in motivational tendencies (Kolling et al., 2012). Of note is that in addition to the VS and VTA, the right inferior frontal gyrus (RIFG) demonstrated a trend toward group differences that did not survive correction for post hoc multiple comparisons (t(44) = −2.299, P = 0.026, uncorrected; Table 2).

Table 2.

Post hoc comparisons of beta weights from regions obtained in the whole-brain contrast of HiGC vs LoGC approach conditions between approach and avoidance tendency related groups (gr)

Region	Approach gr		Avoidance gr		t(44)	P value	Effect sizea
	mean	sd	mean	sd
Betas during HiGC vs baseline
PAG	0.085	0.057	0.070	0.069	0.790	0.434	0.236
Pulvinar	0.101	0.066	0.072	0.056	1.638	0.109	0.481
mPFC	−0.119	0.056	−0.120	0.044	0.079	0.937	0.023
PCC	−0.071	0.065	−0.092	0.072	1.024	0.311	0.305
precuneus	0.136	0.058	0.132	0.060	0.215	0.831	0.064
Superior Temporal Gyrus (STG)	0.083	0.048	0.053	0.057	1.826	0.690	0.555
ventro-lateral Pre Frontal Cortex (vlPFC)	−0.102	0.065	−0.091	0.076	−0.514	0.610	−0.151
Premotor cortex	0.139	0.142	0.085	0.113	−0.121	0.904	0.419
High visual areas (BA18-19)	0.064	0.043	0.055	0.045	1.463	0.151	0.216
R Superior Frontal Gyrus (SFG)	−0.071	0.077	−0.045	0.074	−1.169	0.249	−0.345
R MFG	0.040	0.048	0.039	0.068	0.012	0.990	0.004
R IFG	−0.073	0.077	−0.024	0.068	−2.299	0.026	−0.676
Betas during LoGC vs baseline
PAG	0.044	0.064	0.045	0.060	−0.094	0.926	−0.028
Pulvinar	0.059	0.050	0.052	0.058	0.467	0.643	0.139
mPFC	−0.085	0.059	−0.076	0.047	−0.538	0.593	−0.158
PCC	−0.044	−0.040	0.069	0.049	−0.239	0.813	−2.536
Precuneus	0.070	0.054	0.051	0.047	1.255	0.216	0.369
STG	0.037	0.043	0.024	0.056	0.923	0.361	0.270
vlPFC	−0.058	0.060	−0.037	0.050	−1.326	0.192	−0.389
Premotor cortex	0.091	0.061	0.079	0.089	0.526	0.602	0.158
High visual areas (BA18-19)	0.019	−0.007	0.054	0.045	1.779	0.082	−1.094
R SFG	−0.038	0.064	−0.019	0.061	−1.064	0.293	−0.314
R MFG	0.019	0.056	−0.007	0.054	1.608	0.115	0.475
R IFG	−0.034	0.041	−0.006	0.053	−2.013	0.050	−0.603

R = right; L = left.

^aEffect size measured by Cohen’s d.

Mesostriatal functional connectivity under HiGC

FC between the VTA and VS differed between HiGC and LoGC conditions, with higher coupling during HiGC (for details see Figure 6 and Table 3). Furthermore, VTA–VS FC during HiGC showed a trend toward differentiation between personality groups, with stronger VTA–VS coupling in the avoidance group [t(45) = −1.777, P = 0.082, mean (s.d.): 0.254 (0.094) and 0.759 (0.081) for approach and avoidance groups, respectively].

Fig. 6.

PPI analysis from the VTA. (A) The VTA region revealed in the HiGC vs LoGC contrast was further used in a PPI analysis comparing its task-related functional connectivity during HiGH vs LoGC. (B) Whole-brain analysis (presented here at a threshold of P < 0.001, uncorrected, with a minimal cluster size of 50 contiguous anatomical voxels) revealed increased coupling with the left VS during HiGC. (C) Parameter estimates, represented by the PPI beta values, are shown at the right panel (mean VTA–VS connectivity: 0.053 and −0.02 for HiGC and LoGC, respectively; P < 0.001, uncorrected, with minimal cluster size of 10 contiguous functional voxels).

Table 3.

Functional connectivity using psycho-physiological interaction during HiGC vs LoGC conditions

Region	Peak voxel (x, y, z)a			t(45)	P value	Cluster size
L VS	−15	−46	−8	4.452	<0.0001	72
L parahippocampal gyrus	−15	2	−2	4.086	<0.0001	53

L = left.

^aCoordinates are of peak connectivity, given according to Talairach space with their t and P values. ROI analysis was conducted for the VS which demonstrated both peak voxel of q(FDR) < 0.05 and minimal cluster size of 10 contiguous functional voxels. The L parahippocampal region did not survive these restrictions.

Discussion

In this study, we introduced a novel gaming platform providing a naturalistic environment which manipulates approach behavior and enables characterization of the neurobehavioral correlates of real-life like motivational behavior under changing conflict levels. This unique platform allowed us to demonstrate the relationship between two major mesostriatal dopaminergic nodes and motivationally induced behavioral facilitation: the VS, which was previously related in humans to representation of reinforcement’s saliency and value (Litt et al., 2011; Liu et al., 2011; Miller et al., 2014) and the VTA, related to effort-based motivation (Puryear et al., 2010; Salamone and Correa, 2012). Both nodes along with the PAG and precuneus were more active under HiGC than under LoGC situations during the game. Additionally, we related different behavioral and neural patterns of approach under HiGC to specific personality profiles standing for approach and avoidance tendencies, integrated from phenomenological, behavioral and neurochemical models of personality. As predicted, this multilevel approach revealed that the VS and VTA play a key role in differentiation of motivational tendencies, demonstrating higher activation when pursuing approach behavior during HiGC, and corresponding to higher behavioral approach tendency in the game.

We clearly demonstrated that it is possible to modulate motivational approach behavior in a conflictual gaming situation. As expected, motivational behavior was affected by the GC level, varying by the amount of threat while approaching an incentive reward. Under LoGC, subjects chose to approach the reward in 93% of the trials, whereas under HiGC conditions subjects only approached 64% of the trials, indicating that the motivational complexity indeed led to variability in motivational decisions. Contrasting approach under HiGC vs LoGC conditions indicated the VTA/PAG, VS and the precuneus involved in HiGC approach behavior. Of note is that animal studies have long demonstrated the VTA and VS as major facilitators of incentive motivation via the mesostriatal dopaminergic pathway (Haber and Knutson, 2009; Salamone and Correa, 2012), suggesting these may be the main nodes underlying facilitation of approach behavior in this study. Recordings from dopaminergic VTA neurons in monkeys have been shown to reflect the ratio between costs and benefits in which sustained activation precedes uncertain rewards (Fiorillo et al., 2003). Similarly, the VS is known to process motivational information and facilitate behavioral output via the dorsal striatum (Sesack and Grace, 2009). Indeed, we found enhanced activation in the VTA and VS, as well as strengthened VTA–VS connectivity during HiGC compared with the LoGC trials, suggesting that the HiGC conditions required investment of more effort in order to facilitate behavior. Another region substantially elicited in the HiGC condition was the precuneus, known to be activated when subjects make choices which may result in lower gain or punishment (Paulus et al., 2003; Roy et al., 2011). The precuneus was also indicated as related to the ‘prevention system’ (Strauman et al., 2012) conceptualized as a goal-attainment harm-prevention regulatory system by the ‘Regulatory Focus Theory’ (Higgins, 1997), and its activation during HiGC may represent this role. However, because HiGC conditions also involved the appearance of more balls on the screen, precuneus activation may result from the richer visual scene during HiGC. Intriguingly, no activation was elicited in the medial pre-frontal regions during HiGC approach, but rather under LoGC conditions, which mostly elicited activations in the mPFC/ACC and the PCC (Table 2). During LoGC conditions, the probability of gaining the reward was high and did not require much effort, therefore the net rewarding value was probably perceived as higher. This may explain the greater activation in the mPFC, known to respond to the value of abstract rewards such as monetary gain (Kuhnen and Knutson, 2005; Oya et al., 2005). Activation in this specific region has been suggested to underlie integration of value across different stimulus dimensions (e.g. magnitude, probability) (Clithero and Rangel, 2014). Thus, in contrast to the VTA and VS for which the incentive, motivational value was higher during HiGC conditions (i.e. harder to achieve the reward) (Salamone and Correa, 2012), for the mPFC, the LoGC reward may carry higher or more distinctly positive value (in terms of cost-benefit). In accordance, we found a trending relation between activity in the mPFC during LoGC and emotional ratings of controlled reward (r = 0.355, P = 0.015, uncorrected). As previously suggested (Gray and McNaughton, 2000; Liu et al., 2011), this finding could indicate that the role of the mPFC in the reward circuit is more associated with the reward’s value rather than regulation of VS/VTA facilitation signals.

Our integrative personality clustering confirmed that high scores in approach related traits (extraversion, reward dependence and agreeableness) were manifested in increased tendency to approach behavior during the game, both generally and under HiGC conditions. Furthermore, decreased tendency to approach was manifested by high scores in avoidance-related traits (harm avoidance, neuroticism and punishment sensitivity). Such a clear relationship between personality traits and behavioral tendencies has been long discussed (Depue and Collins, 1999; Gray and McNaughton, 2000; Higgins, 2011), yet in this study it is demonstrated using an original integrative profiling of individual differences derived from three theoretical models. Interestingly, during HiGC trials, the approach group had greater activation in both VTA and VS compared with the avoidance group, while no difference was found between the groups in the activity of the mPFC. With relation to individual tendencies, this finding provides a neural distinction between bottom-up processing in the VTA and VS and top-down processing in the medial prefrontal region. Previous studies of individual differences in motivational tendencies have highlighted the PFC, which is generally related to regulation and integrative high-level processing (Roy et al., 2012), as an important region underpinning such variations, further assigning PFC lateralization to positive vs negative affectivity (Spielberg et al., 2008, 2013). Here we did not find evidence of such involvement, suggesting that individual differences in approach behavior might result from differential activation of the subcortical regions responsible for motivation-related behavioral facilitation. Still, it could be that the stronger activation of the mPFC region during reward evaluation (in the LoGC condition) may have overshadowed its relation to individual differences during HiGC approach. Another possible involvement of the PFC lays in the marginal group difference we found in the RIFG, known as a cognitive control region during visuospatial tasks (Stephan et al., 2003). Future studies should investigate the contribution of this region to individual differences in motivational behavior.

The significance of bottom-up processing to individual differences further implies that motivational tendencies and their neural correlates reflect the innate temperamental parts of personality structure, compatible with Cloninger’s tridimensional model (Cloninger, 1987). Yet it was only the integrative personality profile we constructed that demonstrated such multilevel relations of personality and neurobehavioral patterns, indicating that the tridimensional model itself could not account for individual differences during actual motivational behavior. Pointing to bottom-up temperamental parts of personality rather than the acquired top-down influences is an important distinction: abnormal top-down regulatory mechanisms (such as the PFC) have been suggested to underlie motivation-related psychopathological conditions (Phillips, 2008; Choi et al., 2011); yet our data suggest that the incentive-related low level facilitators of the dopaminergic mesostriatal pathway (such as the VS or VTA) may serve as stronger candidates for variations in motivational behavior tendencies. This observation is supported by the trending difference between groups in VTA–VS connectivity, with the avoidance group showing stronger coupling during HiGC, corresponding to the well- documented compensatory brain connectivity in neurological conditions such as multiple sclerosis (Droby et al., 2015) or Parkinson’s disease (Yan et al., 2015), as well as in psychological conditions like alcoholism (Jung et al., 2014).

To conclude, using an interactive naturalistic motivational paradigm we have shown that specific personality traits previously associated with approach and avoidance tendencies were in fact related to different behavioral patterns under naturalistic conditions, as well as to specific activation patterns in two main motivational nodes: VTA and VS. Understanding the manifestation of individual differences in approach behavior may improve our ability to treat specific pathological conditions such as excessive goal-directed behavior in manic episodes (Johnson et al., 2012; Gonen et al., 2014), and may similarly offer new avenues of treatment in other psychopathologies involving motivational processes, such as effort-based decision making in major depressive disorder (Treadway et al., 2012a) or behavioral flexibility in OCD (Hendler et al., 2014).