This comprehensive meta-analysis examined convergent fMRI findings from tasks indexing regional brain activations within the executive function, social cognition/emotion processing, and reward processing domains in youth with cannabis use. Such an effort allowed for identification of consistent brain activation patterns that were associated with elevated propensity and severity for cannabis use, as well as their relationships with age, sex, cannabis use features, and comorbid psychopathologies in adolescents. The main findings here pertain to identifying differences in brain activation between CU and TD participants in the medial regions of the PFC during social cognition/emotion processing and to a lesser extent during executive function/cognitive control but not during reward processing. A second set of findings identified sex differences in brain activation during executive control, and brain activation differences across cognitive domains in relation to cannabis problem severity and psychiatric comorbidity. Sex differences were seen in insula activation during executive control tasks and showed reduced activation in CU, relative to TD youth, as a function of an increasing proportion of females. Regarding cannabis problem severity, activation differences between CU and TD youth were consistently observed when analyses were restricted to samples in which the majority of CU participants met DSM-IV/5 criteria for CUD or when CUD severity was considered. These analyses showed that higher CUD severity was associated with increased BOLD response in CU (relative to TD) youth, in right rmPFC and vmPFC during executive control tasks and decreased and increased BOLD response in the right insula during reward and social cognition/emotion processing tasks, respectively. Psychiatric comorbidity contributed some variance to brain outcomes with these effects varying across domains, suggesting complicated cannabis use-by-psychiatric comorbidity relationships. Supplemental analyses identified brain activation differences between CU and TD youth in the right caudate, thalamus, insula, precentral and postcentral gyri, and bilateral mPFC regions that varied as a function of the length of abstinence and showed some anatomic overlap and some regional divergence with CUD severity results. Collectively, these findings indicate that differences in brain activity between cannabis using and non-using adolescents in medial prefrontal, insula, and other cortical and subcortical regions implicated in control, social cognition/emotion processing, and motivation are present and vary as a function of sex, CUD severity, psychiatric comorbidity, and length of abstinence. Further, our results provide preliminary evidence that brain activation alterations in adolescent cannabis users with greater cannabis severity or CUD diagnoses are more consistent, more broadly distributed, and may reflect multidomain impairments across executive control, social cognition/emotion processing, and reward processing domains in these youth. Implications of these findings are discussed below.

The primary meta-analytic results from our study yielded significant group differences in the activation of regions within the medial PFC (e.g., rostral, ventral and dorsal mPFC) across all tasks, as well as individually in executive control and emotional processing domains. Specifically, across all tasks, higher vmPFC activation were observed in CU, compared to TD. When each cognitive domain was examined separately, in CU compared to TD, higher activations were observed in the right rmPFC during executive control tasks and lower activations were observed in bilateral dmPFC and dACC during social cognition and emotion processing tasks.

Our executive control domain meta-analysis identified cannabis-related activation differences in mPFC regions, consistent with prior studies in both adolescent [7,17,18,19,38] and adult [6,22] cannabis users. Given comparable between-group behavioral performance on cognitive tasks in most of these studies, the increased rmPFC activation observed in CU compared to TD youth points to inefficient cognitive processing and suggests that CU participants may have to recruit additional PFC resources to perform comparably to their non-using TD counterparts, which has been posited before [71,72,73,74,75,76]. Our executive control results may be better understood by examining them within the context of scientific findings of studies investigating changes in brain activation across childhood and adolescence during inhibitory control in normally developing youth and youth with familial SUD (e.g., individuals at SUD risk). For example, one report by Somerville et al., 2011 focused on typically developing youth suggests that activation during inhibitory control tasks tends to diminish from childhood to adolescence purportedly due to neurobehavioral maturation and improved impulse control [77]. Additionally, longitudinal studies in children of parents with and without AUD show that while unaffected youth show decreases in brain activation during inhibitory control tasks from ages 7 to 13, youth with familial AUD risk exhibit increased activation during the same task [58]. Together, these studies suggest that vulnerable youth may need to recruit ancillary brain regions/systems and expend greater effort on inhibitory control over the course of their development to perform similarly to their typically developing counterparts. By extension, this suggests that some of the mPFC activation differences we identified between CU and TD adolescents in our analyses may reflect preexisting vulnerability and also could be interpreted as a compensatory strategy to overcome possible pre-existing deficits. It is important to note that our sensitivity test restricting the executive control meta-analysis to studies where participants had very low levels of psychiatric comorbidity did not change the main results in this domain. This suggests that the variance in activation differences between CU and TD youth cannot be fully explained by the presence of comorbid psychiatric conditions, and indicates that some variance in brain activations could be related to cannabis exposure or other unmeasured confounders. Thus, our rmPFC result may be better explained in relation to cannabis use behaviors or pre-existing addiction vulnerability that is unrelated to psychiatric symptom expression in CU youth. Relatedly, during jackknife analyses, the rmPFC result identified in our main executive control meta-analysis exhibited moderate study-to-study variability, and was most consistently observed when reliability tests included studies with CUD samples. This suggests that the rmPFC activation differences during executive control tasks could reflect executive dysfunction in adolescents with more severe pathophysiology, such as those with heavier cannabis use or greater CUD severity [1,6]. Results from our executive control sensitivity tests and meta-regression analyses related to CUD diagnostic-status provide additional support for this hypothesis, which warrants further investigation.

Another possible explanation for both the executive control rmPFC finding and our other findings is that chronic cannabis exposure results in maladaptive neuroplasticity in these brain regions or that constitutional abnormalities in vulnerable youth may be further exacerbated by chronic cannabis exposure during adolescence, a sensitive developmental period. Related to this hypothesis, the spatial distribution of activation findings from the present study appear to correspond with brain regions known to have increased cannabinoid receptor 1 (CB₁R) expression and are implicated in endocannabinoid (eCB) signaling [78]. Given this, we posit that the alterations in brain activation seen in adolescent cannabis users are more likely to be found in cortical and subcortical brain regions enriched for cannabinoid receptors, and may occur as a result of homeostatic alterations in brain physiology following chronic cannabis exposure, reflecting eCB system dysregulation in these youth [79,80]. Incorporating genetic and epigenetic data collection and analysis into future neuroimaging studies of CUD populations may provide additional clues into how eCB signaling and related neural processes are altered with cannabis use. Whether brain activation differences during executive control and emotion processing in CU youth compared to their TD counterparts are attributable to common predispositional factors, cannabis-induced neuroadaptive changes, or both warrant further investigation and may inform the development of future brain-based interventions.

For tasks that focused on social cognition/emotion processing domains, we found that CU youth showed less engagement of dmPFC/dACC regions compared to TD youth. This finding was highly reliable across jackknife analyses, and was also observed in our emotion processing subdomain meta-analysis. In contrast, the dmPFC/dACC cluster was not observed in our low psychiatric comorbidity subgroup meta-analysis, the results of which yielded no significant clusters. This sensitivity test result should be interpreted cautiously as 3 of the 4 studies included in the analysis used social cognition paradigms, which contributed less variance to our main social/emotional domain meta-analytic result compared to emotion processing studies. Still, it suggests that some of the variances in BOLD response between CU and TD youth during social cognition may be related to psychiatric conditions that co-occur with cannabis use in adolescents. Decreased activation in the dorsal PFC in cannabis using youth during emotion processing may reflect either inherent or drug associated emotional deficits for this group of adolescents. Emotion processing involves brain circuitry and subjective emotional experience undergo significant neurodevelopmental changes during adolescence [81]. Prior studies have shown that in comparison to adults, adolescents experience more frequent high-intensity positive and negative emotions, greater emotional intensity, and greater instability [82] and that adolescents primarily respond to emotional cues with increased activation of the subcortical circuitry [83]. This heightened emotional reactivity during adolescence is considered adaptive in many situations related to social group cohesion, communication, and support, but may also contribute to adolescent-specific developmental vulnerability to emotional disorders [82,83]. During the transition to young adulthood, prefrontal inputs to subcortical nuclei (e.g., amygdala) increase, allowing for better regulation of emotional responses [81,82,83,84], resulting in an expected increase of activation in related cortical brain regions and greater cortical-subcortical coupling. Accordingly, our finding of decreased dmPFC/dACC function points to deficits in emotional processing in adolescent cannabis users [85], which may in turn signify impairments in navigating the social environment [86,87]. As the dorsal mPFC is believed to support healthy emotional awareness/recognition and the dACC is implicated in emotion regulation [88], abnormal functioning of these regions may reflect impaired emotion recognition and regulation in cannabis using youth. [85,86,87] Reverse directional explanations for these associations are also possible. For example, emotion processing deficits could alternatively predate cannabis exposure and predispose youth to early cannabis engagement and affect-motivated cannabis use as a maladaptive strategy to cope with the challenges of social interactions [82].

Somewhat surprisingly, we did not find differences in activation during reward processing in our primary reward domain meta-analysis. There are several possible explanations for this. First, as already suggested, cannabis use in adolescents may be predominantly related to social and emotional domains and may have little effect on the reward brain system in cannabis users that do not have cannabis dependence (conversely see [89]). It is also possible that differences among reward tasks account for differences in the activation patterns, which when combined either do not survive the statistical threshold, counteract or cancel each other. Furthermore, the lack of differences in regional activation may be accounted by individual neurodevelopmental differences that moderate the signal in some yet poorly understood fashion resulting in seemingly null differences in brain activation among participating groups. Related to this assertion, it is notable that while we did not identify reward alterations in our primary reward domain meta-analysis, both our subgroup meta-analysis restricted to studies with very low levels of psychiatric comorbidity and our meta-regression analysis focused on cannabis problem severity identified significant reward-related brain activation differences between CU and TD youth. Interestingly, both of these activation differences mapped onto the right insula and were in opposite directions, such that insula activation was higher in CU, relative to TD youth, when the meta-analytic sample was restricted to studies with limited psychiatric comorbidity, and lower in CU, relative to TD youth, when examined in relation to the proportion of CUD diagnoses among CU youth. This suggests cannabis use-by-psychiatric comorbidity interaction effects on BOLD signal are present, and further vary as a function of CUD status. The directional pattern of these relationships suggests that CU youth without comorbid/co-occurring psychiatric conditions may exhibit dysfunctional reward processing showing hyper-responsiveness in the insula during reward feedback. However, when psychiatric comorbidity is not taken into account (as in our main reward domain meta-analysis), activation differences related to reward processing between CU and TD youth are more variable, and any effects of cannabis use on BOLD signal, if present, may be obfuscated by differences in underlying brain function related to comorbid psychiatric conditions that predate cannabis exposure and/or differences in the neural response to cannabis exposure in youth with and without comorbidities. Furthermore, these results indicate that when CUD diagnostic status is considered, another pattern emerges, whereby youth with high CUD severity also exhibit dysfunctional reward processing but in the opposite direction, showing hypo-responsiveness in the insula during reward feedback. Taken together, these findings point to parametric variation in reward responsiveness among CU youth as a function of psychiatric comorbidity and CUD diagnostic status. As such, differences in sample characteristics may explain some of the variability in outcomes from fMRI studies in the literature. Further exploration of brain correlates of cannabis exposure in combined samples of cannabis using and non-using youth with and without comorbid psychiatric conditions are needed.

One of the main objectives of this study was to use meta-analysis and meta-regression approaches to investigate the influence of age, sex, psychiatric comorbidity, and cannabis use features on BOLD differences between CU and TD youth. Partially consistent with our hypotheses, our results showed distinct activation differences related to sex, CUD severity, and length of abstinence but no effect of age, and mixed findings with regard to the effects of psychiatric comorbidity on brain function. Null findings for age effects on BOLD response may be due to design and analysis differences and limited variation in mean age of study samples, with early adolescent cannabis users (ages 12–13 years) being underrepresented among studies.

Regarding our psychiatric comorbidity effects, null results on the main effect of psychiatric comorbidity were not surprising given that our main meta-analytic results across studies (collapsed across domains) were also not significant when conservative thresholding was applied. Notably, the meta-analysis across studies with CU and TD groups matched for low psychiatric comorbidity did show greater activation in CU compared to TD youth in an orbitofrontal cortical region commonly reported in CU studies [22,23], when thresholded at p < 0.025. Additionally, domain-specific effects were seen for executive control, social cognition, and reward domains when psychiatric comorbidity was controlled for in the analyses. Specifically, the executive control meta-analysis showed no change in the main rmPFC cluster, and the dmPFC/dACC cluster identified in the main social cognition/emotion processing meta-analysis was no longer present when these analyses were rerun restricting the meta-analytic sample to studies with very low psychiatric comorbidity. Additionally, a difference in right insula activation between CU and TD youth during reward feedback was identified that was not present in the main reward domain meta-analysis when this analysis was rerun in a subgroup of studies with very low psychiatric comorbidity. These findings suggest that the activation differences between CU and TD youth observed in our meta-analytic report cannot be fully explained by the presence of comorbid psychiatric conditions. Rather, they point to psychiatric comorbidity playing a contributory role to variability in BOLD signal in CU and TD youth, in combination with other factors such as cannabis exposure, sex, and unmeasured confounders. Taken together, our results provide early evidence for cannabis use-by-psychiatric comorbidity interaction effects on brain outcomes during executive control, social cognition, and reward processing in CU youth, and suggest that these interaction effects may manifest as different BOLD response patterns in CU youth with and without psychiatric comorbidities. It is important to note that our approach to measuring psychiatric comorbidity effects in this study was non-specific. Thus, the comorbidity results presented here may better reflect the influence of ‘general’ psychopathology as opposed to specific psychiatric conditions on BOLD differences between CU and TD youth. These associations warrant further investigation, and should be examined in relation to both ‘general’ psychopathology and specific psychiatric symptom clusters and disorders in future studies.

Regarding sex effects, we identified a left insula cluster that varied between CU and TD youth as a function of sex distribution showing decreased BOLD response in CU relative to TD youth in executive control fMRI studies that had a higher proportion of female participants. This suggests that sex may moderate the relationships between cannabis exposure, executive function, and brain activity during adolescence. Previous studies have shown sex differences in neuropsychological functioning in adolescent and young adult cannabis users [90]. Our results are consistent with prior studies in cannabis using adolescents and adults that have shown alterations in brain structure, activity, and connectivity in cortical and cerebellar regions rich in CB₁R in women that correlate with cognitive impairments and, in some cases, are directionally opposite from those found in men [91,92,93,94,95,96,97,98]. Of note, none of these prior reports identified sex-by-cannabis group differences in the insula, although one [98] did show higher cerebral blood flow in the insular cortex of males compared to females in both cannabis using and non-using control groups. Thus, our meta-analytic report is the first to describe sex-specific effects of cannabis exposure in this region in humans. A number of possible factors could explain this sex-by-cannabis effect. Sex differences in the sensitivity to adolescent cannabis exposure could partially explain this result. There is growing evidence for sexual dimorphism of the eCB system [99], including sex differences in CB₁R expression in the insular cortex, which has been shown across species. These eCB differences emerge during puberty, and may render cannabis using girls more vulnerable to adverse health outcomes related to cannabis exposure [99]. In support of this, recent animal studies have shown sex-specific effects of adolescent cannabis exposure on cortico-striatal-limbic circuit maturation and long-term THC-related cognitive and affective alterations extending into adulthood [100]. Our sex-related BOLD effects could also be secondary to phenotypic differences between cannabis using boys and girls. Women typically start using cannabis later than men, but progress more quickly from first use to dependence (a phenomenon known as the ‘telescoping’ effect) [101]. Additionally, women who use cannabis report greater subjective psychoactive effects, cannabis withdrawal severity, cannabis-related problems, and have higher rates of comorbid mood and anxiety disorders compared to cannabis using men [102]. These sex differences in natural history of cannabis onset and clinical presentation may indicate divergent neurobiological underpinnings in cannabis using men and women. The extent to which these behavioral differences map onto brain differences and predate or are exacerbated by cannabis exposure warrants further investigation. Lastly, these results might also be explained by general neurodevelopmental differences between boys and girls [103], which could be further exacerbated by cannabis exposure. Given the growing clinical and preclinical evidence for sex differences related to cannabis exposure, research in this area is sorely needed and may inform sex-specific interventions for women and men with CUD.

Regarding our cannabis problem severity effects, we identified rostral and ventral mPFC and right insula clusters that varied between CU and TD youth as a function of CUD severity and affected cognitive domain (e.g., executive control, emotion, and reward). During executive control, CU (compared to TD) showed increased activation of the right rmPFC and vmPFC as a function of increasing proportion of CUD diagnoses in the sample. This parallels our main findings and suggests that impaired over-engagement of the rmPFC/vmPFC is associated with greater CUD severity [104]. Activation in the right insula was also associated with CUD severity, with the direction of this effect being domain dependent. During social/emotional processing tasks, we observed an increase in BOLD response in the right insula in CU compared to TD youth that varied as a function of increasing mean CUDIT score, such that higher mean CUDIT scores in CU participants were associated with higher insula BOLD response in CU, relative to TD youth. However, during reward processing tasks, we observed a decrease in BOLD response in this region as a function of increasing proportion of CUD diagnoses in the sample, such that a higher proportion of CUD in CU participants was associated with lower insula BOLD response in CU, relative to TD youth. This suggests that greater CUD severity is associated with insula hyper-responsiveness during social interactions and emotion processing and insula hypo-responsiveness during reward processing and motivated decision making. Together these findings across domains support a cognitive control deficit, reward deficit, and stress surfeit model of CUD severity in adolescents [105], consistent with Ernst’s triadic model of addiction vulnerability [53]. Our results are consistent with previous studies in adults showing CU vs. TD differences in cognitive control and reward-related brain function that are more prominent in dependent cannabis users compared to non-dependent users [69,106].

In our abstinence subgroup meta-analyses, we found evidence for distinct patterns of activation differences across domains between CU and TD youth that varied as a function of length of abstinence. Notably, we observed what appeared to be a subcortical-to-cortical gradient in activation differences across abstinence periods with chronic CU (relative to TD) exhibiting a shift from predominantly subcortical circuitry alterations during ad lib use to predominantly cortical circuitry alterations during longer periods of abstinence (e.g., Ad-lib to 12-h: caudate and thalamus; 24-h and 48-h to 72-h: insula and IFG; and >21 days: dmPFC/dACC and precentral and postcentral gyri). These findings converge with previous studies showing state-level differences in cortical and subcortical brain function as a function of ‘sated’ versus ‘abstinent’ states and time since last use in adolescent and adult cannabis users [66,78,107]. Taken in the context of recent advances in dynamic functional connectivity research, these findings could reflect distinct brain states emerging at specific periods of abstinence. Inter-individual variability in the magnitude and of the length of time/occupancy spent in these brain states might have prognostic significance related to relapse risk during a cessation attempt, and represent neural treatment targets [108]. Regarding the spatial distribution of our results, brain regions showing abstinence subgroup effects generally did not overlap with each other, but did show some regional overlap (e.g., insula) with brain regions related to CUD severity. The extent to which these cross-sectional analyses can be compared with one another to improve our understanding of neuroadaptive changes following cannabis cessation is unclear. This represents an important area of future inquiry that requires serial MRI scans within cannabis using individuals over time.

The existing literature on medial PFC regions in adolescents suggests that these are regions that undergo protracted development from childhood to late adolescence, making adolescence a critical period of neuroplasticity [77,81,84]. Although attempts have been made to identify separate and more specific functions linked predominantly to the rostral vs. dorsal mPFC, it seems very plausible that these regions are functionally interconnected and often activate during similar cognitive processes including forward planning, executive control, self-reflection and social awareness [88,104]. Results from the current meta-analysis reveal that CU adolescents show differences in activation primarily in regions linked to executive control and emotion processing, cognitive functions that underpin more complex behaviors such as social functioning [88]. Our report of alterations in brain activation possibly link to adolescents’ behavior in social milieu but not in relation to reward processing, which suggests that cannabis use in adolescence is perpetuated and facilitated by factors predominantly related to youth’s social environment and interactions and to a lesser extent by factors related to the rewarding properties of the drug [85,86,87]. If true, this speculation is consistent with existing evidence suggesting that therapies targeting family and social dynamics show higher efficacy in treating CUD in adolescents [109,110] and may further inform strategies for designing a therapeutic intervention that focuses on building social skills to facilitate peer interactions and minimize substance use.

This meta-analytic report has a number of important limitations. As our main aim was to conduct meta-analyses and meta-regressions, our study was reliant on the study methodology, analytic techniques, and measurements used in each of the included fMRI studies. Many of these studies had small sample sizes and there was significant heterogeneity in design, behavioral measurements, and handling of psychiatric comorbidity and co-use of alcohol and tobacco which likely contributed variance to study outcomes. Consequently, many of our planned subgroup and meta-regression analyses may have been underpowered to detect subtle differences in brain activation with small effect sizes between cannabis users and non-users. Our attempts to address these issues by contacting authors and searching imaging repositories were limitedly successful but did yield an additional five studies that were included in our analysis. Changes in data management, reporting, and data sharing practices (especially of full datasets and unthresholded statistical maps) are needed to advance meta-analytic inquiry in the field of neuroimaging [111]. In parallel, a consensus core outcome set (COS) of standardized measures related to adolescent cannabis use that incorporate indices of frequency, quantity, duration, dose, addiction severity (e.g., [112]) would reduce heterogeneity across studies and facilitate meta-analytic comparison across different cannabis using populations. No studies included in this meta-analysis reported biochemically quantified delta-9-tetra-hydrocannabinol (THC) or cannabidiol (CBD) levels as an index of cannabis exposure in CU participants, limiting our ability to investigate this relevant emerging domain. This is particularly salient given preliminary data showing divergent and at times opposing effects of THC and CBD on brain activity and cognitive, emotional, and reward processing in adults [113,114]. As such, future studies should measure THC and CBD exposure from cannabis use and relate these exposures to brain changes. Another major limitation of this meta-analytic report is that the cross-sectional experimental design of most included studies prevents us from inferring causality of these results. That is, it is impossible to determine if the reported group differences in activation predate the onset of cannabis use (i.e., if these are constitutional) or if they are “caused” or exacerbated by the cannabis use. Notably, this is an inherent concern in most published studies to date. However, long term longitudinal studies, such as the ABCD study, are poised to circumvent this issue and allow independent assessments of predispositions of cannabis use separately from the effects of cannabis use. For ease of interpretation and to facilitate quantitative comparisons, we focused specifically on task-based fMRI studies investigating BOLD response differences from task contrasts. Still, given our limited understanding of the relationship between the fMRI BOLD signal and the corresponding neuronal processes, it remains challenging to interpret the directionality of the findings. We chose not to include studies that used resting-state fMRI or functional connectivity outcomes in our meta-analyses. While this decision could be viewed as a limitation, it was made after closely reviewing functional connectivity studies identified in our initial search and determining that there were too few studies and too much heterogeneity in analytic approaches across these studies (e.g., seed-based vs. independent component analysis) to allow for an appropriately powered quantitative analysis. As the number of functional connectivity studies in cannabis using adolescent samples expands this will be an important next step that can build on the findings presented here and improve our understanding of brain network alterations at rest and during cognitive processes. Despite these limitations, our study also has some relevant strengths that are worth noting. It is one of the first meta-analytic studies to characterize brain activation differences between cannabis using and non-using individuals in a developmental sample of adolescents and young adults, and relate these activation differences to CUD severity and length of abstinence. In doing so, it identifies key neural targets to guide future research and theory development related to the pathophysiology of adolescent cannabis use. An additional strength is in the use of SDM meta-analysis and meta-regression techniques which enabled us to quantitatively assess relationships between BOLD response and age, sex, cannabis features, and psychiatric comorbidity in the sample and conduct sensitivity tests to clarify the extent to which brain outcomes were related to study design and sample characteristics.