Exploring the Utility of a Functional Magnetic Resonance Imaging (fMRI) Cannabis Cue-Reactivity Paradigm in Treatment Seeking Adults with Cannabis Use Disorder

Abstract

Background:

Functional magnetic resonance imaging (fMRI) studies examining cue-reactivity in cannabis use disorder (CUD) have either had small sample sizes or involved non-treatment-seeking participants. As a secondary analysis, we administered an fMRI cue-reactivity task to CUD participants entering two separate clinical trials (varenicline or repetitive Transcranial Magnetic Stimulation—rTMS) to determine the task activation patterns for treatment-seeking participants with CUD. We aimed to determine the activation patterns for the total sample and determined behavioral correlates. We additionally compared studies to determine if patterns were consistent.

Methods:

Treatment-seeking participants with moderate or severe CUD had behavioral craving measured at baseline via the short form of the Marijuana Craving Questionnaire (MCQ-SF) and completed a visual cannabis cue-reactivity task during fMRI (measuring the Blood-Oxygen-Level-Dependent—BOLD response) following 24-hours of cannabis-abstinence.

Results:

Sixty-five participants were included (37-varenicline, 28-rTMS; 32% female; mean-age 30.4±9.9SD). When contrasting cannabis-images vs. matched-neutral-images, participants showed greater BOLD response in bilateral ventromedial-prefrontal, dorsolateral-prefrontal, anterior cingulate, and visual cortices, as well as the striatum. There was stronger task-based functional-connectivity (tbFC) between the medial prefrontal cortex and both the amygdala and the visual cortex. Craving negatively correlated with BOLD response in the left ventral striatum (R²=−0.32; p=0.01) in the full sample. There were no significant differences in either activation or tbFC between studies.

Discussion:

Among two separate treatment-seeking groups with CUD, there was increased cannabis cue-reactivity and tbFC in regions related to executive function and reward processing. Cannabis-craving was negatively associated with cue-reactivity in the left ventral striatum.

Clinicaltrials.gov identifiers:

NCT02892110, NCT03144232.

INTRODUCTION

Drug craving and the related response to drug cues (cue-reactivity) are the behavioral constructs in addictions that have been studied most thoroughly. Craving has been a frequent proximate target in clinical trials and has clear clinical relevance across addictions(1-7). Neuroimaging studies, predominantly using functional magnetic resonance imaging(fMRI), exploring the neural substrates of cue-reactivity, have consistently found activation of incentive salience related structures in response to drug cues relative to neutral cues(8). Several studies have related drug cue-reactivity in these regions to clinically relevant outcomes such as relapse to substance use(9), and a number of studies have demonstrated that effective pharmacologic treatments both modulate drug cue-reactivity within these regions (demonstrate target engagement) and group-level target engagement is associated with better clinical outcomes(10-16). Relatively little is known, however, about whether the same relationships exist in Cannabis Use Disorder (CUD), and only a single trial has prospectively linked fMRI cue-reactivity to clinical outcomes in CUD(17).

There is extensive literature suggesting that heavy cannabis users display behavioral cannabis cue-reactivity(18). A series of neuroimaging studies have extended these behavioral findings and have demonstrated that cannabis users display the characteristic increase in activation in incentive salience related structures in response to cannabis cues(19-27). When taken together, these experiments suggest that addiction severity is associated with the degree of fMRI activation(17,20,28), and neural activation correlates with clinically relevant behavioral variables such as craving(19,22-26,29). Though only two studies have examined task-based network connectivity among individuals with CUD, both(26,30) found increased connectivity between striatal and prefrontal areas when participants viewed cues. These seminal investigations have provided insights into the neural basis of cue-reactivity in cannabis users, however, only two of the above studies recruited treatment-seeking participants(24,29), and both had small sample sizes, leaving this group studied minimally. Understanding the task-activation and functional-connectivity in this clinical population can potentially lead to future target engagement studies in CUD, which could lead to further treatment development.

To fill this gap, as a secondary analysis, we analyzed baseline fMRI cue-reactivity data from two randomized-controlled treatment trials that each recruited and rigorously screened participants with CUD and tested the clinical efficacy of varenicline and repetitive transcranial magnetic stimulation(rTMS) respectively. Using this combined dataset we sought to determine the task-activation and task-based-functional-connectivity patterns in treatment-seeking participants with CUD along with their behavioral correlates to see if they were consistent with the existent literature (hypothesizing there would be consistency). We further compared activation and functional-connectivity between studies to determine if the findings are generalizable (hypothesizing they would be). Given this was a secondary analysis that was not pre-planned, we took an exploratory whole-brain approach with the hopes of taking advantage of the larger sample size to maximize the utility of these results for future investigations, such as whether the studied interventions (rTMS and varenicline) effect fMRI-cue-reactivity (to be reported separately).

METHODS AND MATERIALS

Overview and participant evaluations:

We included baseline behavioral and fMRI data from two clinical trials that recruited treatment-seeking participants with moderate or severe CUD who were interested in reducing their cannabis use. The two trials investigated the potential therapeutic effects of varenicline(NCT02892110)(31) and rTMS(NCT03144232)(32) respectively, and we included the N=37 available baseline fMRIs from the varenicline trial (conducted between February 2017 and November 2018), and the N=28 baseline fMRIs collected at the Medical University of South Carolina (MUSC) site from the rTMS trial (conducted between August 2017 and May 2019). All of the included fMRI’s were performed on the same 3-Tesla MRI scanner at MUSC. This sample of convenience was not sub-sampled or matched on any variable. All research related activities were approved by the MUSC Institutional Review Board and were conducted in accordance with the declaration of Helsinki. Participants in both trials were recruited via media and print advertisements from the greater Charleston, SC, area, underwent a brief phone screen assessing eligibility, and if eligible for the study, were invited for an in-person screening and enrollment visit. After reviewing and signing informed consent, participants underwent a similar screening procedure in both trials which included evaluation by a licensed clinician with a medical history and examination, the Mini International Neuropsychiatric Interview (MINI(33)), 28-days’ worth of Time-Line Follow-Back (TLFB(34)), the Marijuana Problem Scale (MPS(35)), and urine drug testing (UDT; Alere Toxicology, testing for amphetamines, benzodiazepines, cannabis, cocaine, and opiates).

Inclusion criteria for both studies (with divergent criteria in parenthesis) included: a) age 18-55 (18-60 for rTMS); b) currently meeting DSM-5 criteria for ≥moderate CUD and cannabis use ≥3-days-per-week in the last 30 days (≥5-days-per-week for rTMS); c) interest in quitting or decreasing cannabis use; and d) sufficient intellectual level and command of the English language to provide consent and complete assessments, which was verified clinically. Additionally, participants in the varenicline study had to have a body mass index between 18 and 35kg/m² and a weight greater than 50kg for pharmacokinetic reasons. Exclusion criteria (with divergent criteria in parenthesis) included: a) being pregnant or breastfeeding; b) currently meeting DSM-5 criteria for ≥moderate non-cannabis/tobacco substance use disorder, though occasional use of other substances was not exclusionary; c) current unstable psychiatric, neurologic, or general medical condition; d) lifetime history of bipolar or psychotic disorder; e) active suicidal ideation or a suicide attempt within the past 90-days (120-days for varenicline); f) unstable dosing for central nervous system medications (no central nervous system active medications for rTMS); and g) contraindications for MRI such as claustrophobia, or implanted metal. Additionally, participants were excluded from the rTMS study if they had a history of seizure and were excluded from the varenicline study if they had taken an investigational agent in the last 30-days or were enrolled in another clinical trial within 60-days.

Participants meeting the above inclusion/exclusion criteria were invited back for a scanning visit where they were asked to abstain from the use of cannabis or other drugs for at least 24 hours (verified by self-report and a Confirm Biosciences saliva test for amphetamines, benzodiazepines, cocaine, cannabis, and opiates). Behavioral craving was assessed using the short form of the Marijuana Craving Questionnaire (MCQ-SF)(36), prior to scanning in the case of the varenicline study, and approximately 20-minutes after scanning in the case of the rTMS study. Of note the MCQ-SF was collected only once (as opposed to before and after the fMRI). Urine drug testing was also performed.

Imaging Procedures:

For this investigation, MRI was conducted using a 32-channel head coil with a 12-m gradient-echo, echoplanar imaging (EPI) sequence on a 3-Tesla Prisma^fit MRI scanner (Siemens, Erlangen, Germany). We first collected a Scout image to align subsequent image acquisitions. Next, we collected an anatomical image (MPRAGE, voxel-size=1mm³, repetition-time=2300ms, echo-time=2.26ms, inversion-time=900ms) for functional alignment and transformation to standard Montreal Neurological Institute (MNI) space. Then we collected functional data while participants completed a visual cannabis cue task (51-slices, multiband-factor=3, repetition-time=1200ms, echo-time=30ms, flip-angle=65 degrees, voxel-size=2.8x2.8x2.8mm, 601-volumes). Forward and reverse spin-echo sequences were collected for distortion correction.

We employed a previously developed block-design cannabis cue task(21). During the task, a series of cannabis-images, matched-neutral-images, blurred-images, or a fixation crosshair were presented on a projection screen visible to the participants via a mirror attached to the head coil. Images were presented during six 120-second epochs, each consisting of four repetitions of 24-second blocks of an image type, followed by a 6-second rating period. Each image block consisted of 12-pictures presented for two-seconds each. During each rating period participants rated their current urge to use cannabis on a 5-point Likert scale (1-None to 5-Extreme) using a button-push hand pad. The images consisted of 36 cannabis-images and 36 matched-neutral-images of non-food objects or plants (matched on color, hue, and visual complexity). For the cannabis-images, there were two-blocks of ‘passive’ cannabis-images which included paraphernalia or a cannabis plant, and one-block of ‘active’ cannabis-images depicting individuals smoking or handling cannabis or paraphernalia. There were coinciding blocks of ‘active’ and ‘passive’ matched-neutral-images. See the supplemental analysis section for further details on a comparison between cue-reactivity in ‘active’ and ‘passive’ images. Blurred images and fixation trials were used as contrasts to evaluate attention and non-cannabis specific effects, and the six-second rating period allowed for a normalization of the hemodynamic response in-between blocks.

Imaging pre-processing and task activation modelling procedures:

We performed fMRI preprocessing using Analysis of Functional Images (AFNI) version 6.33(37). Anatomical data were fed through AFNI’s @SSwarper to calculate non-linear transformations into MNI space. We used afni_proc.py to perform standard preprocessing, which included despiking, slice timing adjustment, motion correction, distortion correction using images with reverse phase encoding, anatomical alignment, and MNI normalization. All spatial transformations were concatenated and applied in a single step to reduce interpolation errors. Next, the MNI-space functional data were passed through ICA-AROMA(38) using the ‘non-aggressive’ setting. Output data were then entered into a second afni_proc.py function call, that performed blurring (6mm FWHM Gaussian kernel) and scaled the data such that each voxel had a mean of 100.

To determine the activation associated with each type of image, we used a general linear model (GLM) approach using AFNI’s 3dREMLfit, which accounts for autocorrelations in the fMRI time-series(39). Task event onsets were modeled in seconds and convolved with an estimate of the hemodynamic response (double gamma ‘SPMG1’) function with the appropriate duration (24-seconds for images, 6-seconds for ratings). Motion parameters and their derivatives were also included in the model, as high pass filtering with polynomial detrending up to order 5. As ICA-AROMA was used for denoising, a threshold of 0.8mm motion displacement was used to censor volumes in the model. Additionally, volumes in which more than 50% of the voxels were identified as outliers were also censored in the model.

Group statistics from the beta estimates and t-statistics produced at the first level were calculated using a mixed modeling approach in 3dMEMA(40). The smoothness of the data was calculated from the residual time-series using the more accurate Auto-Correlation Function (ACF) metric(41). These values were averaged across all participants. We then used Monte Carlo estimation in order to determine cluster thresholds corresponding to a p_FWE<0.05 for a voxel level threshold of p<0.001, bi-sided(42). Our primary contrast of interest was that of cannabis-images vs. matched-neutral-images. In addition, we examined activity in response to cannabis-images/matched-neutral images vs. blur-images and fixation cross. We performed each of the contrasts above in each study individually, and then contrasted the two studies. We additionally performed each contrast in male and female participants and contrasted the two sexes.

After the basic GLM was completed using a whole brain-approach contrasting cannabis-images and matched-neutral-images, we extracted the beta values (corresponding to % signal change) from significant clusters, including the regions around the maximum activation point of multiple whole brain points of greater activation that were consistent with other cue-reactivity trials. This was done using the same threshold as in the primary analysis. These included the clusters identified as the left and right ventral striatum, the ventromedial prefrontal cortex(vmPFC) / anterior cingulate cortex(ACC), the left and right visual cortex, and the right dorsolateral prefrontal cortex. These specific clusters were chosen based on their consistency with other reports(43).

Task-Based Functional Connectivity analysis:

We used the CONN toolbox to determine the changes in functional-connectivity between when participants were viewing cannabis-images to when they were viewing matched-neutral-images. First, we used the same denoised data that were entered in the GLM above, applied smoothing in CONN, used default filtering settings, and regressed motion parameters and their derivatives(38,44) as well as the main effect of each of the task events. The data was band passed between 0.008 and 0.09 Hz. This is the default setting in CONN18b, and serves to reduce noise in the timeseries, while preserving the low frequency signals of interest. Time-series were extracted from the unsmoothed data using the following full MNI ROI set provided with CONN18b, which provides 132 ROIs across cortical and subcortical regions. These ROIs are derived from the FSL Harvard-Oxford cortical and subcortical atlases(45-48) which combined manual segmentations to produce probabilistic ROIs that preferentially cover gray matter regions. We excluded the cerebellar ROIs (n=27), as these areas were not consistently captured across all subjects, leaving 105 ROIs in the analysis. For a full list of ROIs used, see Supplemental Table-2.

Functional connectivity was calculated using bivariate regression for each subject, with hemodynamic response function weighting and then combined in a group level analysis. We contrasted the connectivity during cannabis-images with matched-neutral-images and corrected for multiple comparisons using the strict p_FDR<0.05 analysis level correction, which accounts for both the number of targets and source ROIs(105). Like the task-activation approach above we performed additional contrasts between study and sex. Finally, we extracted the correlation coefficients (Rz-scores) from connections between the following data-driven correlations: vmPFC and (left and right) amygdala; vmPFC and occipital; vmPFC and left-lateral-occipital; vmPFC and right-lateral-occipital, and; left-parietal and occipital to relate to behavioral data. Of note the vmPFC and occipital structures included larger bilateral regions, whereas the other analyzed ROI’s represented lateralized structures.

Non-imaging data analysis procedures:

We approached the behavioral data by first comparing demographic descriptors (age, sex, race/ethnicity, and educational achievement); age of first cannabis use; illness severity (DSM-5 Criteria and MPS); cannabis craving (MCQ-SF and hand-pad urge rating); and cannabis use (TLFB) between studies. We then calculated Spearman rank correlation coefficients between craving (MCQ-SF total score), CUD related problems (the MPS), cannabis use (number of cannabis use sessions over the 7-days prior to scanning), the extracted activation %signal-changes in the six data-driven clusters above, and the extracted connectivity Rz-scores from the five data-driven connectivity pairs above (see Supplemental Table-3 for correlation matrices). We further explored significant associations between cluster activation or Rz-scores and behavioral data, using a GLM. Continuous and normally distributed outcomes (MCQ-SF, MPS) were modeled assuming a Gaussian distribution. The number of cannabis use sessions in the prior 7-days was modeled using negative binomial regression. In all models, the predictors included the MRI-variable, study, sex, and tobacco use status, and; the dependent measure was the behavioral variable of interest. Residual normality was assessed in Gaussian models. There was minimal missing data and so missing data was simply left out of analyses. Analyses were not pre-registered, nor were the findings corrected for multiple comparisons. Subsequently the results should be considered exploratory. Statistical significance, when reported, was based on two-tailed tests with an alpha=0.05. All statistical analyses were run using SAS University Edition (Cary, NC, USA).

RESULTS

Demographic and descriptive data (Table-1):

Table-1:

Sample Descriptive Statistics:

	Total Sample	rTMS	Varenicline
Simple Descriptive Variables:
Age	30.4 ± 9.9	31.3±11.2	29.8±8.8
Sex	44 Men 21 Women	18 Men 10 Women	26 Men 11 Women
Race and Ethnicity: AA = African American; C = Caucasian; and O = Other	20 AA; 42 C; 3 O	7 AA; 18 C; 3 O	13 AA; 22 C; 2 O
Marital Status	9 Married 55 Non-Married	2 Married 26 Non-Married	7 Married 29 Non-Married
Education level (4-year college degree or more education)	41 < 4-year 24 ≥ 4-year	18 < 4-year 10 ≥ 4-year	23 < 4-year 14 ≥ 4-year
Smoked cigarettes on at least 14 of the previous 28 days	20 Smokers 45 Non-Smokers	9 Smoker 19 Non-Smoker	11 Smoker 26 Non-Smoker
Age of First Cannabis Use:	15.5 ± 3.9	15.0 ± 2.4	16.0 ± 4.7
Diagnostic and Statistical Manual Criteria (DSM-5) and Marijuana Problem Scale Values
Number of DSM-5 Cannabis Use Disorder Criteria met*	7.4±1.7	8.0±1.6	6.9±1.7
Percent meeting the DSM-5 ‘Craving’-criteria	57 (87.7%)	25 (89.3%)	32 (86.5%)
DSM-5 Cannabis Use Disorder Categorical	56 Severe 9 Moderate	26 Severe 2 Moderate	30 Severe 7 Moderate
Marijuana Problem Scale score (MPS)	9.2±5.8	7.8±5.4	10.2±6.0
Short Form of the Marijuana Craving Questionnaire (MCQ) and fMRI hand-pad craving values:
MCQ-SF-Total	46.3±15.5	45.4±16.3	47.0±15.2
MCQ-SF Compulsivity	9.1±4.5	9.1±5.3	9.1±4.0
MCQ-SF Emotionality	10.8±5.4	9.9±5.5	11.4±5.2
MCQ-SF Expectancy	12.6±4.5	11.8±4.6	13.2±4.5
MCQ-SF Purposefulness	13.8±4.9	14.5±5.2	13.3±4.7
fMRI Handpad (cannabis – neutral)	1.2±1.0	1.4±1.0	1.1±0.9
Cannabis Use variables:
Urine cannabinoids Level (ng/ml)	337.4±471.5	346.4±557.3	330.3±400.6
Creatinine (mg/ml)	139.7±86.4	140.2±71.3	139.3±98.1
Creatinine corrected cannabinoids level (ng/mg)	3.5±4.18	3.1±4.4	3.8±4.0
# days using cannabis / last 7	5.1±2.3	4.9±2.4	5.2±2.3
# days using cannabis / last 28	24.1±5.2	23.3±6.2	24.6±4.4
# cannabis use sessions / last 7	15.6±16.4	17.6±19.8	14.1±13.4
# cannabis use sessions / last 28	86.5±67.2	89.9±84.1	84.3±54.6

All values reported ± Standard Deviation (SD). *p=0.01

The final sample included a total of 65 participants (twenty-eight in the rTMS study and thirty-seven in the varenicline study). The average age of included participants was 30.4±9.9SD and consisted of 32.3% women. There were no significant differences in age, sex, or other demographic variables between the two studies. There was however a significant between-study difference in the average number of DSM-5 criteria met with the rTMS-group meeting more DSM-5 criteria.

Task activation modelling:

Our primary contrast of interest was cannabis-images vs. matched-neutral-images. There was greater neural activation during cannabis-images in the bilateral ventromedial prefrontal cortices, anterior cingulate cortices, striatum, visual cortices, and dorsolateral prefrontal cortices relative to matched-neutral-images. During matched-neutral-images, there was greater activation in bilateral cuneus, right Rolandic operculum, superior temporal gyrus, and right middle cingulate cortex relative to cannabis-images (Figure-1, Figure-2, Supplemental Table-1, and Supplemental Figure-1). There were no differences between study or sex contrasts in the whole-brain analysis. However, the extracted percent-signal change between the cannabis-cues vs. matched-neutral-cues contrast in both the left and right ventral striatum differed significantly between the two studies with a higher percent signal change in the rTMS study.

Figure-1: Subtraction maps contrasting cannabis-images and matched-neutral-images in:

a) the combined sample (N=65); b) the varenicline sample (N=37), and; c) the rTMS sample (N=28). All comparisons used a voxel threshold of p<0.001 and cluster significance of p<0.05 family wise error (FWE) corrected. Critical T-statistic for the subtraction map is 3.4491 for p <0.001, two-sided. Significant clusters had to exceed 31 voxels. There were no significant differences when contrasting images b) and c) finding that the subtraction maps of the two samples did not differ significantly.

Figure-2:

Correlation between the left ventral striatal cluster (derived from the whole brain analysis) and Marijuana Craving Questionnaire Short Form (MCQ-SF) score for the combined sample (R²=0.25; p=0.01).

Task-based functional connectivity modelling:

During our primary comparison of interest (cannabis-images vs. matched-neutral-images) there was greater connectivity between the vmPFC and the amygdala (left and right); the vmPFC and the occipital cortex; the vmPFC and the left-lateral-occipital cortex; the vmPFC and the right-lateral-occipital cortex, and; the left-parietal cortex and the occipital cortex (Figure-3). All connectivity changes were verified as positive (greater connectivity) when viewing cannabis-images relative to matched-neutral-images. There were no differences between study or sex contrasts.

Figure-3:

Task Based Functional Connectivity contrasting connectivity during cannabis-images (both active and passive) relative to matched-neutral-images for the combined sample. All values represent Rz values between the denoted regions of interest reported ± Standard Deviations (SD). See Supplemental Table-3 for the Conn ROI atlas that was used. These regions were significant at p_FDR<0.05 analysis level, multiple comparison correction in the full model.

Covariate analysis between imaging and behavioral findings:

The MCQ-SF total score was negatively correlated with activation in the left ventral striatum (rho=−0.26; p<0.04) and right visual cortex (rho=−0.29; p<0.02). The total number of cannabis use sessions over the 7-days prior to scanning was negatively correlated with activation in the right DLPFC cluster (rho=−0.27; p=0.03). When controlling for sex, study, and tobacco smoker status, left striatal activation was still significantly associated with the MCQ-total score (R²=0.32; p=0.01). The MCQ-SF was negatively correlated with left ventral striatum reactivity in each study separately (rho=−0.36 in the rTMS study and rho=−0.17 in the varenicline study). The correlation between the right visual cortex and the MCQ-SF; and the correlation between the R-DLPFC and cannabis use sessions were no longer significant when controlling for sex, study, and tobacco smoker status. There were no significant correlations between any of the examined behavioral variables and task-based functional connectivity Rz-values.

DISCUSSION

In this secondary analysis study, we present preliminary cue-reactivity data from two cohorts of treatment seeking participants with CUD who were entering treatment trials. When combining the two cohorts, we found task-activation in structures consistent with the existent cue-reactivity literature across substance use disorders including greater bilateral BOLD activation in the ventromedial prefrontal cortices, dorsolateral prefrontal cortices, anterior cingulate cortices, striatum, and visual cortices when contrasting activation during cannabis-images and matched-neutral-images. BOLD activation in the left striatum negatively correlated with spontaneous craving as measured by the MCQ-SF collected on the same day. We also found that there was greater task-based-functional-connectivity between the medial prefrontal cortex and both the amygdala and visual cortices when participants were viewing cannabis relative to matched-neutral images. We did not find any significant differences in activation or connectivity between study or between sex. We explore each of these findings in the context of the existent literature in CUD, and other substance use literature where appropriate.

The main finding of this investigation is that in this relatively-large group of treatment-seeking participants with CUD there was robust neural activation in incentive-salience and visual cortices during cue-reactivity, and; that activation in the left ventral-striatum negatively correlated with baseline behavioral cannabis craving. Our findings are important for two reasons. First, our findings strengthen the results of studies in heavy cannabis users that were not treatment-seeking given our similar findings in treatment-seeking participants with CUD. Second, our results provide a starting point for future interventional trials for both predictive and target engagement goals in CUD. Ideally future investigations will have repeated fMRI observations and can explore test-retest consistency and whether interventions with behavioral effects are able to engage circuitry in a meaningful fashion (similar to the approach of Karoly and colleagues in non-treatment seeking adolescents(49)). The finding that left-ventral striatum activation negatively correlated with behavioral craving supports the findings of(28) and contrasts the findings of(29). Of note there were methodologic differences between studies, which include a whole brain approach in the present study and an ROI approach in the two other studies. Additionally, the present study included treatment-seeking participants with CUD with 24-hours of abstinence, while the other treatment-seeking trial(29) scanned participants while they were using cannabis ad libitum and found a positive correlation, and the non-treatment seeking trial(28) required at least 24-hours of abstinence and found a negative correlation. The period of abstinence prior to scanning may subsequently be a critical component of the relationship between striatal activation and craving.

We found higher connectivity between the vmPFC and both the visual cortex and the amygdala when comparing connectivity during cannabis-cues and matched-neutral-images. Our findings contrast somewhat with the two other trials examining task-based-functional-connectivity in cannabis users(26,30), which found greater connectivity between the nucleus accumbens and the ACC, striatum, and cerebellum(30), and; between the dorsal striatum and the middle frontal gyrus(26). As is the case with our BOLD contrast findings, the methodology of each of those investigations was different from the present investigation in terms of population (treatment vs. non-treatment seeking), and analysis technique (ROI vs whole-brain). Interestingly, in the present study connectivity increased between the vmPFC and both the visual cortex and amygdala, which provides early evidence that not only is there greater activation of both the visual and incentive salience circuitry, but also more connectivity between these networks. This greater connectivity is consistent with the attention bias found in substance use disorders(50).

We found no sex differences on any of our imaging or task-based-functional-connectivity contrasts. Cue-reactivity studies examining sex differences in BOLD activation are limited as most studies are underpowered or do not report sex effects(51). To our knowledge only one other study examined sex differences in cue-induced neural activation in cannabis users(52) and also did not find a sex-difference in neural activation. However, they did demonstrate correlations between baseline craving and the bilateral insula and left lateral OFC in female-participants, and between craving and the striatum in male-participants. Other measures of brain function (e.g., electroencephalogram, glucose metabolism) have been used to investigate sex differences in cannabis users, and have also shown mixed results(53). Given the notable sex differences in clinical and behavioral characteristics of cannabis use(54,55) yet the dearth of statistically powered research on sex differences in brain function, further research in this area is warranted.

Despite the strengths of this investigation including rigorous screening and enrollment procedures with licensed-clinicians, the inclusion of a relatively large sample of treatment-seeking participants, the use of two independently recruited cohorts entering distinct treatment paradigms (potentially representing heterogenous groups of CUD participants), and, the use of data-driven whole brain analyses (to confirm regions of interest for future investigations rather than relying upon them in the present investigation), there are several limitations that warrant mention. Limitations include: a) that this was an exploratory secondary analysis of data coming from trials with slightly different procedures, for example the MCQ-SF was performed prior to the scan in one trial and following the scan for the other; b) that this investigation had only a single cross sectional evaluation point so test-retest reliability cannot be concluded, nor can any conclusions regarding the correlation with behavioral outcomes be drawn; c) there was no healthy control group to compare our findings with, and so we can only conclude that these treatment seeking participants with CUD displayed this activation pattern, rather than that the observed task activation patterns are characteristic of treatment seeking participants with CUD; d) that there were only 21 female-participants in the sample and so the analysis may have been under-powered to detect sex-effects, and; e) we did not systematically collect data on the ecological validity of the scanner-cues, the number of previous quit attempts each participant had, or each participants goal for the study. These limitations reduce the impact and generalizability of our findings, and subsequently further experimentation is needed.

In summary, we found that a cohort of treatment-seeking participants with CUD displayed higher neural activation in reward and incentive salience regions when viewing cannabis images compared to when viewing matched non-cannabis neutral images. These findings are largely consistent with the findings of other cue-reactivity trials in other SUDs, as well as those studies in participants with heavy cannabis use. We also found that participants had increased task-based-functional-connectivity between the salience network and visual and limbic systems. The largely consistent findings between these two separately recruited samples (and the consistency of this investigation with other investigations in SUDs) support the potential utility of this imaging paradigm to measure cannabis cue-reactivity in treatment-seeking participants with CUD. This paradigm may subsequently have validity as a treatment target in a similar fashion to other cue-reactivity paradigms(9). However, further testing is needed to determine if within subject findings are consistent in terms of test-retest, and ability to change in target-engagement studies, as well as clinical trials.