The risk of cannabis use disorder is mediated by altered brain connectivity: A chronnectome study
Fazio et al.
Department of Neuroscience, Padua Neuroscience Center University of Padua Padua Italy
Department of General Psychiatry at the Center for Psychosocial Medicine Heidelberg University Heidelberg Germany
Department of Psychiatry and Psychotherapy, Central Institute of Mental Health, Medical Faculty Mannheim Heidelberg University Mannheim Germany
Department of Psychiatry and Psychotherapy Saarland University Saarbrücken Germany
Tri‐institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS), Georgia State University, Georgia Institute of Technology Emory University Atlanta Georgia USA
Department of Neuroradiology Saarland University Saarbrücken Germany
* CorrespondenceFabio Sambataro, Department of Neuroscience, Padua Neuroscience Center, University of Padua, Via Giustiniani 5, Padova, Padua, Italy.
Email: fabio.sambataro@unipd.it
Abstract
The brain mechanisms underlying the risk of cannabis use disorder (CUD) are poorly understood. Several studies have reported changes in functional connectivity (FC) in CUD, although none have focused on the study of time‐varying patterns of FC. To fill this important gap of knowledge, 39 individuals at risk for CUD and 55 controls, stratified by their score on a self‐screening questionnaire for cannabis‐related problems (CUDIT‐R), underwent resting‐state functional magnetic resonance imaging. Dynamic functional connectivity (dFNC) was estimated using independent component analysis, sliding‐time window correlations, cluster states and meta‐state indices of global dynamics and were compared among groups. At‐risk individuals stayed longer in a cluster state with higher within and reduced between network dFNC for the subcortical, sensory‐motor, visual, cognitive‐control and default‐mode networks, relative to controls. More globally, at‐risk individuals had a greater number of meta‐states and transitions between them and a longer state span and total distance between meta‐states in the state space. Our findings suggest that the risk of CUD is associated with an increased dynamic fluidity and dynamic range of FC. This may result in altered stability and engagement of the brain networks, which can ultimately translate into altered cortical and subcortical function conveying CUD risk. Identifying these changes in brain function can pave the way for early pharmacological and neurostimulation treatment of CUD, as much as they could facilitate the stratification of high‐risk individuals.
Article notes
Fazio G , Olivo D , Wolf ND , et al. The risk of cannabis use disorder is mediated by altered brain connectivity: A chronnectome study. Addiction Biology. 2024;29 (5): 29(5):e13395. doi:10.1111/adb.13395 38709211 PMC11072977
Footnote Group
1INTRODUCTION
Cannabis is the third most widely used controlled substance in the world, after alcohol and tobacco, with estimates from the United Nations dating back to 2020, which indicate that 209 million people have used cannabis at least once in the previous year. 1 Among 1 in 10 regular users and 1 in 3 individuals using this substance daily 2 , 3 develop a problematic use with an inability to stop using cannabis despite clinical and psychosocial problems, that is, Cannabis Use Disorder (CUD). 4
CUD has been associated with an increased risk of developing concomitant psychiatric disorders and contributes to the evolution of psychotic disorders in ultra‐high risk subjects. 5 In this regard, observational and experimental studies have confirmed the association between cannabis consume and the onset and persistence of psychotic disorders with an effect size that is related to the duration of abuse and to the potency of cannabis. 6 In particular, cannabis use was linked with an earlier onset of psychosis, greater symptom severity, higher rates of relapse, longer hospitalizations and poorer outcomes. 7 Furthermore, chronic use of cannabis was also associated with a worse prognosis, including treatment resistance for various psychiatric disorders 8 and a greater risk of manic episodes and suicide. 6
Several factors can contribute to the development of CUD in at‐risk individuals. These include genetic and psychological factors (e.g., emotional dysregulation), comorbid mental disorders (including depression, anxiety, attention deficit hyperactivity disorder [ADHD], conduct and/or personality disorders), pattern of cannabis use (frequency and duration of use, concomitant use of alcohol and tobacco and other substances), parental cannabis use and stressful life events (childhood abuse, unemployment, financial difficulties). 9
At the biological level, Koob and Volkow 10 propose a more general model of substance addiction that can be applied to cannabis addiction 11 and encompasses three stages: binge/intoxication, withdrawal/negative affect and preoccupation/anticipation. The first entails excessive impulsive and compulsive behaviours aimed at consuming cannabis despite the negative consequences associated with its use. This stage is characterized by hyperactivation of the mesocorticolimbic dopaminergic reward pathway caused by an impairment in incentive salience, which appears to drive dopaminergic signalling to maintain the drug consume upon exposure to conditioned cues and during tolerance development. These continuous mechanisms of opponent‐process responses trigger the withdrawal/negative affect stage. This stage is characterized by neurobiological changes that take place within‐systems, including decreased dopaminergic signalling in the nucleus accumbens (NAcc) and dorsal striatum that result in an elevation of reward thresholds for non‐drug reinforcers, and between‐systems, which include changes in the stress responses of the brain, including corticotropin‐releasing factor (CRF) release in the amygdala and HPA‐axis dysfunction. These changes can lead to loss of motivation for non‐drug rewards and impaired emotion regulation during acute and sustained abstinence. The last stage, preoccupation/anticipation, is involved in the re‐establishment of substance use after abstinence. At the neural level, the GABAergic and glutamatergic signalling pathways between the pre‐frontal cortex (PFC) and the brain areas mediating decision‐making, self‐regulation, inhibitory control and working memory are altered. At the behavioural level, this stage may be associated with excessive salience attribution to drug‐paired cues, decreases in responsiveness to non‐drug cues and reinforcers and reduction of the ability to inhibit maladaptive behaviour. 10 , 11 Overall, the neural changes implicated in vulnerability and the risk of developing CUD need to be further explored.
Neuroimaging studies have shown that the brain is intrinsically organized in functionally connected networks that interact with each other to determine a specific function and their spontaneous activity can be studied using resting state functional Magnetic Resonance Imaging (rs‐fMRI) rather than using specific tasks that engage few regions and can be influenced by performance. 12 Numerous studies in fMRI have shown that cannabis use can alter brain functioning, especially in networks that support working memory, attention and cognitive control processing. 13 More specifically, a recent systematic review found that a substantial number of studies (40%) conducted in cannabis users showed a higher positive resting state FC in fronto‐frontal, fronto‐temporal and fronto‐striatal networks relative to controls. 14 Moreover, alterations of functional connectivity (FC) at rest have been reported in individuals with CUD relative to healthy controls, although their direction has shown mixed. 15 These contrasting findings are due in part to the temporal characteristics of FC. Indeed, FC cannot be considered stationary but varies over time with functional coupling between regions and networks that dynamically rearrange in various connectivity modes. 16 For this reason, the use of a static analysis approach may oversimplify a model that eventually fails to capture the complexity of brain FC.
Recently, to overcome these limitations, a new analytic framework, dFNC, that investigates the temporal changes of brain connectivity over time (and, for this reason, defined as ‘chronnectome’) has been developed. 16 , 17 , 18 , 19 , 20 , 21 DFNC studies the recurring patterns of FC between different functional brain networks, namely, FC ‘states’, that are reproducible over time and between different subjects. 17 , 22 , 23
DFNC can be studied with two different and complementary methods: a cluster‐state approach and a meta‐state approach. The first is based on a deterministic assumption, with a subject dwelling in only a single state of connectivity at a given point in time, and reveals pairwise network connectivity patterns arranged in states and the changes in their temporal engagement. 16 The second arises from a probabilistic view of the connectivity states where a subject can occupy several states of connectivity at each time, ‘meta‐states’, 24 and provides estimates of the overall dynamics of FC between states with metrics of dFNC fluidity and range. 16 , 24 , 25 , 26 , 27
In this study, our major aim was to understand the role of altered dFNC in the development of CUD. For this, we wanted to identify alterations in dFNC in individuals at risk of CUD. Moreover, to confirm the possible role of these neural changes in the clinical presentation of CUD, we correlated dFNC measures with clinical variables in cannabis users. Notably, we recruited only individuals at risk of cannabis addiction: first, to understand the mechanisms associated with the development of the CUD and, second, to study the effects of cannabis use while reducing the confounding effects of clinical and psychosocial changes after the problematic use of cannabis. 28 , 29 , 30 , 31
2MATERIALS AND METHODS
2.1Participants and MRI data
Ninety‐eight subjects were initially recruited for this study. Exclusion criteria were considered: a present or past mental disorder assessed using the structured clinical interview (SCID) for the DSM‐IV‐TR; current or past substance use other than cannabis and nicotine; and history of neurological disorders, brain trauma or other concurrent drug treatment. None of the participants met the diagnostic criteria for the DSM‐IV‐TR cannabis use disorder. After recruitment, four participants were excluded for not having undergone an imaging session (n = 3) or clinical data appointment (n = 1). A final sample of 94 individuals (66 males and 28 females, aged between 18 and 30 years) was included in the study.
The following demographic data were collected: age, sex, handedness and education. All participants were also evaluated for depression, anxiety and ADHD using Hamilton Depression (HAM‐D), 32 , 33 the State and Trait Anxiety Inventory (STAI‐G X1, STAI‐G X2) 34 and the German adult ADHD symptoms self‐rated scale (Aufmerksamkeitsdefizit‐/Hyperaktivitätsstörung im Erwachsenenalter, Selbstbeurteilungsskala, ADHS‐SB). 35
Participants were stratified according to their risk of developing CUD using a screening tool, the Cannabis Use Disorder Identification Test (CUDIT‐R), with the risk of CUD defined by a score greater than or equal to 8 as reported previously, 36 , 37 thus resulting in 55 controls and 39 individuals at risk of CUD. Patterns of cannabis use (lifetime total cannabis use, age of onset, duration and current use [days/week and g/week]) were used to evaluate cannabis use (see Table 1).
| Control s n = 55 Mean ± SD | CUD at‐risk n = 39 Mean ± SD | p a | |
|---|---|---|---|
| Age (years) | 23.30 ± 3 | 23.20 ± 3.31 | 0.77 |
| Sex (M:F) | 30:25 | 36:3 | 0.006 |
| Handedness (R:L) | 52:3 | 36:3 | 0.202 |
| Education (years) | 14.90 ± 2.68 | 13.70 ± 2.57 | 0.033 |
| HAM‐D | 0.364 ± 0.729 | 1.38 ± 2.22 | 0.002 |
| STAI‐G X1 | 31.00 ± 7.47 | 36.30 ± 8.03 | 0.001 |
| STAI‐G X2 | 31.50 ± 7.74 | 35.60 ± 8.75 | 0.018 |
| ADHD‐SB tot | 6.110 ± 5.99 | 13.10 ± 6.94 | <0.001 |
| ADHS inattentive | 2.98 ± 2.55 | 5.82 ± 3.10 | <0.001 |
| ADHS impulsivity | 0.836 ± 1.13 | 2.18 ± 2.14 | <0.001 |
| ADHS hyperactivity | 1.62 ± 2.21 | 3.03 ± 2.28 | 0.003 |
| Tobacco per year (pack/year) | 0.942 ± 3.090 | 2.50 ± 4.65 | 0.054 |
| N. joints lifetime | 2000 ± 8,430 | 3,258 ± 3,949 | <0.001 |
| CUDIT‐R total | 1.040 ± 2.04 | 16.90 ± 7.64 | <0.001 |
| Age of the first consume (years) | 4.670 ± 8.17 | 17.70 ± 2.89 | <0.001 |
| Duration use (years) | 0.373 ± 1.57 | 3.41 ± 3.39 | <0.001 |
| Current use (days/week) | 0.473 ± 1.41 | 4.95 ± 2.32 | <0.001 |
| Current use (g/week) | 0.212 ± 0.779 | 5.01 ± 5.53 | <0.001 |
Consumers were asked to abstain from cannabis use for at least 24 h before clinical assessment and MRI. Participants consented to these study‐specific requirements, and no craving or other withdrawal symptoms were reported before the MRI scanning. The study was carried out in accordance with the Declaration of Helsinki, and the protocol was approved by the ethical review board of the Saarland Medical Association, Saarbrücken, Germany. Written informed consent was obtained from all participants after the study procedures had been fully explained. 38
2.2MRI data acquisition
The scans were obtained using a 3T Magnetom‐type Skyra (Siemens, Erlangen, Germany) MRI system. Images in structural MRI were acquired through rapid sequences prepared at gradient‐echo magnetization (3D‐MPRAGE) set with the following parameters: TE = 3.29 ms, TR = 1900 ms, TI = 110 ms, flip angle = 9°, FOV = 240 mm, slice plane = axial, voxel size = 0.5 × 0.5 × 0.9 mm3, distance factor = 50%, number of slices = 192.
Then, rs‐fMRI images were obtained using a BOLD echo‐planar (EPI) sequence set with the following parameters: TE = 30 ms, TR = 1800 ms, flip angle = 90°, FOV = 192 mm, slice plane = transversal, voxel size = 3 × 3 × 3 mm, distance factor = 25%, number of slices = 32, PAT factor = 2, number of measurements = 230.
2.3MRI data analysis
After discarding the first eight scans, acquired to allow the signal to reach the steady state, 222 volumes were pre‐processed.
Data Processing Assistant for rs‐fMRI (DPABI/DPARSF 5.2) was used to pre‐process the neuroimaging data. 39 A slice‐timing correction was used to correct for temporal mismatch in slice acquisition. Next, realignment was used to correct head movements through rigid‐body transformations and estimate motion extent. Functional and structural images were coregistered, segmented and normalized on a Montreal Neurological Institute (MNI) template using DARTEL. 40 Finally, a 6‐mm 3D Gaussian kernel full width at half maximum was used for smoothing.
The data were decomposed into functional networks using a spatial group‐independent component analysis (ICA), 41 implemented through the group ICA of fMRI toolbox (GIFT) 42 (https://trendscenter.org/software/gift/). The group information guided ICA (GIG‐ICA) algorithm was used for constrained ICA on the NeuroMark template, which comprises 53 intrinsic connectivity networks (ICNs) that were estimated in two large independent samples and span across seven functional network domains: subcortical (SC), auditory (AUD), sensorimotor (SM), visual (VIS), cognitive‐control (CC), default mode (DM) and cerebellar (CB). 43 To increase the reliability and stability of the estimation, the ICA was repeated 50 times using the ICASSO toolbox. 44 , 45
2.4Dynamic functional network connectivity
DFNC was estimated using the GIFT toolbox, through sliding window analysis. 22 , 46 For each subject, the 53 (one for each ICN) time‐courses (TC) were detrended, orthogonalized with respect to 12‐motion parameters, despiked using 3dDespike (https://afni.nimh.nih.gov/pub/dist/doc/program_help/3dDespike.html) and, subsequently, filtered with a fifth‐order Butterworth low‐pass filter (cut‐off of 0.15 Hz). 47 A sparse matrix of inverse covariance was calculated through a least absolute shrinkage and selection operator regression using an L1 penalty to improve the estimation of the shortest TC. 48 The matrix size resulted of 53 × 53 based on the number of ICNs. We used Gaussian windows with a size of 30 TR (54 s) and a sigma of 3 TR (5.4 s) to taper along the window edges advancing 1 TR at each step, resulting in 192 windows (see Figure 1). Then, the dFNC values were Fisher's Z‐transformed and residualized with respect to significant nuisance covariates, including age, sex, education and motion (mean framewise displacement according to Power et al. 49 ).
2.4.2Cluster states analysis
The state‐based dFNC analysis was implemented through a cluster state approach to identify dFNC states in the native state space. In this case, k‐means clustering of wFNC matrices was repeated to reduce dimensionality. Each windowed functional connectivity matrix (wFNC) was defined by the value N(N−1)/2, where N represents the number of components of interest: K means clustering consists of a dimensional reduction technique that allows us to “collapse” the space connectivity of N(N−1)/2 > 1,000 dimensions into one, thus obtaining the most recurrent connectivity pattern: a subject will therefore be in a single specific state at any given moment. 22
K‐means clustering allowed us to identify the four most recurrent connectivity patterns, minimizing the total variance within the most similar matrices, and obtaining the states. The number of clusters or states was determined by the elbow criterion. For each FC cluster state, the mean dwell time, defined as the average time that each subject spent in a specific state, measured as the number of consecutive windows, was estimated 22 (see Figure 1). Cluster states that recurred in less than 10 individuals per sample were considered not reliable for group comparisons and excluded from further analysis.
2.5Statistical analysis
Statistical analyses were performed using JAMOVI 50 (https://www.jamovi.org/). Demographics, psychometric assessments, substance use patterns and brain dynamics indices were compared between samples using two‐sample t tests. Brain dynamics indices were also correlated with clinical and substance use parameters using Pearson's correlations. Additionally, to take into account the psychopathological differences between samples, the analyses were repeated using analyses of covariance (ANCOVA) to compare group differences and partial correlations for brain‐behaviour correlations with the score of each scale (HAM‐D, STAI‐G X1 and X2, and ADHS‐SB) as nuisance covariates, respectively. Statistical significance was assessed using a threshold of p = 0.05, and a false discovery rate (FDR) correction with α = 0.05 was used for multiple comparisons. 51
3RESULTS
3.1Meta‐state analysis
The following brain dynamics indices were significantly increased in individuals at risk for CUD compared to controls (see Table 2 and Figure 2).
| t | p a | |
|---|---|---|
| Number of states | 2.35 | 0.021 |
| Change between states | 2.43 | 0.017 |
| State span | 2.72 | 0.008 |
| Total distance | 3.36 | 0.001 |
These results were confirmed by ANCOVAs with HAM‐D, STAI‐G X1, STAI‐G X2 and ADHS‐SB (see Table 3).
| HAM‐D ANCOVA | F | p | STAI‐G X2 ANCOVA | F | p |
|---|---|---|---|---|---|
| Number of states | 2.66 | 0.009 | Number of states | 2.09 | 0.039 |
| Change between states | 2.75 | 0.007 | Change between states | 2.21 | 0.030 |
| State span | 2.97 | 0.004 | State span | 2.41 | 0.018 |
| Total distance | 3.59 | <0.001 | Total distance | 3.06 | 0.003 |
All correlation analyses carried out between meta‐state measures and psychometric scores or consumption patterns were not significant.
3.2Cluster state analysis
Four cluster states were estimated. After discarding state four data for limited reliability (see before), three clusters were further analysed (see Figure 3).
3.2.1Cluster State 1
State 1 recurred 15% of the time. In this state, the within‐network correlations were highest in SC, SM and CC, and the between‐network correlations were highest in SC (negative with SM and CC), SM (negative with VI and SC and positive with CC) and CC (negative with VI and SC and positive with SM).
3.2.2Cluster State 2
State 2 recurred 12% of the time. In this state, within‐network correlations were highest in SC, SM and VI, and between‐network correlations were highest in SM (negative with SC, DM, some CC ICs and positive with VI), SC (negative with SM, VI and CB and positive with DM), VI (negative with SC, positive with SM), DM (positive with SC and some CC ICS, negative with SM) and CB (negative with SC).
3.2.3Cluster State 3
State 3 recurred 29% of the time. In this state, within‐network correlations were highest in SC, SM, VI, CC and DM, while between‐network correlations were reduced for SC (with AU, SM and CC), AU (with SC, SM, VIS, CC and CB), SM (with SC, AU, VIS, CC and CB), VI (with SC, AU, SM and CB), for CC (with SC, AU, SM and CB) and CB (with AU, VIS and CC).
When comparing the dwell time between the groups, at‐risk individuals dwelled longer only in State 3 relative to the control group (t 92 = 2.85, p = 0.005) (see Figure 4). These results were confirmed by ANCOVAs with HAM‐D (t 91 = 2.55, p = 0.013), STAI‐G X1 (t 91 = 2.54, p = 0.013), STAI‐G X2 (t 91 = 2.80, p = 0.006) or ADHS‐SB (t 91 = 2.31, p = 0.023).
The mean dwell time in Cluster 3 and the psychometric scores or the cannabis use patterns were not significantly correlated.
4DISCUSSION
The aim of this study was the identification of dFNC alterations associated with the risk of CUD. This study yielded two main findings resulting by two complimentary approaches: at‐risk individuals had a greater number of extremely volatile meta‐states and moved more frequently between them, covering a greater distance in the state space, and, at the same time within specific cluster states, they tended to stay longer in a state (Cluster State 3) with high within‐network coupling in SC, SM, VI, CC and DM, and reduced between‐network coupling.
4.2Cluster state features in the risk of CUD
These individuals persist longer in a state characterized by higher intra‐network connectivity in the subcortical, sensorimotor, visual, default mode and cognitive‐control network domains and reduced between‐network interactions. The subcortical network domain includes the nucleus accumbens, the ventral tegmental area (VTA), the basolateral amygdala, the hippocampus, the bed nucleus of the stria terminalis, the dorsolateral striatum and the globus pallidus. THC activates presynaptic CB1 receptors (CB1R) in VTA GABAergic neurons that inhibit presynaptic GABA release, and this causes increased dopaminergic firing in the VTA, which may explain the greater intrinsic activity of this network. 56 The subcortical regions are connected to the prefrontal regions through dopaminergic and GABAergic projections. The cognitive control network, including the prefrontal regions, plays a role in cognition, emotional regulation, salience and impulsivity that can be altered in cannabis use and dependence. 57 Therefore, connectivity between subcortical and cognitive control networks can be crucial for the recognition of a reward in the environment and in complex affective and cognitive behaviours, including the formation and retrieval of associative and contextual fear‐ and reward‐related memories. 57 Altered functional connectivity of the subcortical network domain has been reported in previous studies, 14 thus suggesting that changes in dynamic connectivity may be related to altered reward processing and disinhibition. Altered dynamic connectivity of sensorimotor and visual networks supports a modulatory role of cannabis use on sensorimotor function. In this regard, it is noteworthy that recent studies have shown that individuals with high THC use have more severe neurological soft signs (NSS) compared to controls. 58 This is probably caused by an impairment of the sensorimotor system through direct stimulation of CB1R and an alteration of dopamine signalling in the limbic/associative and sensorimotor cortical regions due to THC. 58 Furthermore, the activity of the visual network appears to be influenced by cannabis use and is related to the development of NSS. A recent study demonstrated that chronic cannabis consumers have greater activation during a visual attention task not only in the occipital cortex but also in various frontal and parietal regions, suggesting compensatory and adaptive processes to maintain physiological network functioning. 59 Even the default mode network, particularly in the posterior regions, has been implicated in THC dependence. 60 Hyperactivation of this network is related to greater sensitization to drug‐seeking behaviours and decreased performance on strategic decision‐making tests, 61 , 62 which supports the role of these changes in the development of addiction.
Overall, our findings are in accordance with the GABA‐glutamatergic hypothesis of CUD, which posits that THC can modulate the neuronal excitability of the brain circuits by regulating GABA and glutamate signalling. 60 , 63 CB1Rs, which are more expressed at the level of GABAergic neurons, are activated by endocannabinoids and inhibit neurotransmitter release through retrograde synaptic transmission. GABAergic interneurons, especially fast‐spiking parvalbumin (PV)‐expressing interneurons and non‐fast spiking cholecystokinin (CCK)‐positive cells, play a substantial role in the synchronization of large numbers of glutamatergic pyramid cells. CCK‐positive cells that express CB1R work as a noise filter to remove high‐frequency signals from pyramidal cells. Endocannabinoids can determine a fine‐tuning of cortical neural oscillations by modulating these interneurons. Therefore, the disruption of endocannabinoid signalling caused by CB1R downregulation in chronic cannabis users appears to disrupt excitatory‐inhibitory balance and affects dynamic connectivity, causing a major disruption in network interaction with cortical and subcortical domains that function more independently of each other. 64
4.3Strengths and limitations
This study has some potential limitations. First, the relatively small sample size reduces the potential of detecting more subtle associations and therefore the representativeness of our findings. However, our design was sufficiently powerful to detect the moderate‐higher differences with a robust correction for multiple comparisons to avoid Type II errors. Future studies with larger sample sizes are warranted for the replication of these results.
Second, anxiety and ADHD scores were higher in at‐risk subjects relative, and this could have partially contributed to our findings. Notably, most subjects had scores below the clinical cut‐off, thus suggesting a limited impact of these variables, if any, that could be efficiently accounted for using these scores as nuisance covariates. Finally, our cross‐sectional study does not allow for the evaluation of trajectory risk.
5CONCLUSION
This is the first study to evaluate dynamic connectivity in individuals at risk for CUD. We have shown that individuals at risk for CUD present, early in the course of cannabis use, a global alteration of the dynamics of brain network interplay with specific changes in subcortical and cortical circuits.
DFNC alterations can contribute to the identification of prognostic factors in individuals at risk of CUD, in whom preventive intervention can be carried out, and of novel targets for treatment. DFNC alterations could also be used as biomarkers in preclinical pharmacological trials to test drug efficacy to normalize the dFNC of at‐risk individuals' disruption to avoid their progression to addiction. Finally, longitudinal studies are needed to establish the relationship between functional connectivity dynamics and the development of cannabis use disorder to confirm their role as biomarkers of the risk.
CONFLICT OF INTEREST STATEMENT
The authors declare that no competing financial interest exists.
ETHICS STATEMENT
The study was carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki) for experiments involving humans. All participants gave their informed written consent as approved by the Ethics Review Board of the Saarland Medical Association, Saarbrücken, Germany.
ACKNOWLEDGEMENTS
We thank all study participants for their understanding and willingness to participate in this study.
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.