The goal of this fMRI study was two-fold: first, to investigate the neural basis of cannabis approach and avoid responses in heavy cannabis users, and, second, to assess the predictive power of these neural approach-bias activations for future cannabis use and problem severity. In contrast to our hypothesis, no brain areas showed greater approach-bias activations in heavy cannabis users compared to controls. However, within the heavy cannabis users, approach-bias activations in various fronto-limbic areas were more pronounced with increased lifetime use, which is in line with previous studies on human approach-avoidance learning 
. Most important, beyond self-reports of session-induced craving, approach-bias activation in DLPFC and ACC predicted problem severity after six months. This novel finding underlines the potential of neural approach-bias activations as predictors of cannabis problem severity and identifies the DLPFC and ACC as loci for targeted interventions.
Stronger DLPFC and ACC activation during cannabis approach vs. avoid was related to decreases in cannabis related problem severity. The DLPFC is involved in regulatory self-control (i.e., providing top-down guidance to more basal cognitive processes supported by networks elsewhere in the brain 
), and hypoactivation has been linked to poor decision-making in dependent cannabis users 
. The ACC is involved in evaluative control (i.e., monitoring one’s performance and assessing salience of motivational information 
) and has been linked to deficient error monitoring associated with substance abuse 
. The DLPFC and ACC together are thought to play an important role in appropriately adjusting behavior in conflicting situations 
, which may be critical to successfully resist substance use. The observed prospective negative association between DLPFC/ACC activation and future cannabis related problems may then reflect, the fact that those heavy cannabis users with a well-developed capacity to evaluate and regulate their drug use are more likely to reduce or control their cannabis use, rather than the presence of problem severity per se.
Interestingly, it has been shown that alcohol-dependent patients can be retrained to avoid alcohol and that successful retraining improved treatment outcome 
. Recent work in our lab showed that this improvement is probably mediated by increased control over alcohol approach and avoidance responses rather than decreased strength of automatic appetitive approach tendencies. Also, the present findings indicate that both approach and avoidance tendencies towards cannabis explain unique variance in the change in problem severity. These findings underline the notion that the approach-bias observed in individuals with a SUD does not merely reflect sensitized and conditioned bottom-up drug-approach tendencies. Instead, control (or the lack thereof) over approach and avoidance behavior in a substance-specific context could be the primary mediator of the relation between approach-bias, continued substance use and substance use-related problems. Future research efforts should be aimed at confirming these inferences.
Approach as well as avoidance responses engage activation in fronto-limbic areas, with considerable overlap between these areas 
. We hypothesized that a cannabis approach-bias would result in increased activation for cannabis approach compared to avoidance in these fronto-limbic areas. Across groups, we found such differences in ventromedial prefrontal and posterior cingulate cortex. In contrast to our hypothesis, these activations were very similar for controls and heavy cannabis users, suggesting that a history of cannabis use in the present sample did not suffice to alter activation of these areas. However, within the group of heavy cannabis users, lifetime cannabis use predicted approach-bias activations in various fronto-limbic areas including the amygdala, insula, inferior frontal gyrus, medial frontal gyrus, and parahippocampal gyrus but also visual areas, precuneus, and the cerebellum. Short-term experience-related increases have been observed in all these areas during approach and avoidance learning 
. The observed correlation with lifetime use may indicate that increased front-limbic activity extends beyond short-term rapid learning processes and probably reflect increased salience and motivation for cannabis over time. Moreover, the lack of an association with cannabis problem severity and weekly cannabis use suggests that the increased fronto-limbic response to cannabis approach relative to avoidance may be a function of lifetime cannabis exposure, rather than cannabis problem severity or the direct (sub)acute effects cannabis use. Given the relatively young age and short duration of heavy cannabis use in the present sample, the association with lifetime use raises the hypothesis that group differences could be expected in more long-term cannabis users compared to controls. This further suggests that all cannabis users could develop increased salience and motivation for cannabis over time.
We have previously shown, using the same sample as in the present study, that heavy cannabis users had an approach-bias for cannabis related materials as measured with a different joystick approach-avoidance task. Moreover, the approach-bias predicted absolute levels of cannabis use after six months 
. The current study contributes the important novel finding that approach-bias activation in DLPFC and ACC explain unique variance in future cannabis use-related problems. However, the lack of differences in approach-bias activation patterns and RTs might suggest a limitation in the construct validity of the task. To the best of our knowledge, this is the first study to use the SRC combined with fMRI. Further studies are needed to verify if the SRC is a reliable task to measure the neural mechanisms underlying approach and avoidance behavior.
Some potential limitations must be taken into account. First, there were more smokers among heavy cannabis users and almost all cannabis users (90%) smoked cannabis combined with tobacco (by far the most common form of cannabis use in the Netherlands 
). Since the heavy cannabis users were relatively light smokers and nicotine use was not significantly associated with approach-bias activations in heavy cannabis users, it is unlikely that nicotine use accounts for the observed effects. However, duration of heavy cannabis use correlated with nicotine use and we cannot exclude potential confounding effect of nicotine use. A post-hoc hierarchical regression analysis was performed with FTND-score as additional covariate to verify if DLPFC and ACC activity still predicted cannabis problem severity after correction nicotine dependence. This analysis showed that nicotine dependence did not affect the predictive relationship between DLPFC and ACC activity and cannabis problem severity (DLPFC/ACC approach-bias index p
.037). Nevertheless, it may still be worthwhile to include a group of cannabis naive cigarette smokers in future studies in order to better distinguish cannabis from nicotine effects. Also, it should be mentioned that we excluded potential participants if they had a history of psychiatric disorder; a less stringently selected but more ecologically valid control group may display considerable comorbid externalizing disorders. Therefore, the extent to which the results generalize to all heavy cannabis users remains to be tested.
In summary, the current fMRI study is the first to investigate the neural mechanisms underlying the approach-bias in SUDs. In addition to and independent from self-reported clinical characteristics (including craving), cannabis-specific approach-bias activation in the DLPFC and ACC predicted the course of cannabis related problemsin heavy cannabis users. These findings highlight the importance of the approach-bias in maintenance of addictive behaviors and support a specific role for DLPFC and ACC functionality as a biomarker in the prediction of problem severity and as new loci for targeted prevention and treatment.