FMRI activation to cannabis odor cues is altered in individuals at risk for a cannabis use disorder

Abstract Introduction The smell of cannabis is a cue with universal relevance to cannabis users. However, most cue reactivity imaging studies have solely utilized visual images, auditory imagery scripts, or tactile cues in their experiments. This study introduces a multimodal cue reactivity paradigm that includes picture, odor, and bimodal picture + odor cues. Methods Twenty‐eight adults at risk for cannabis use disorder (CUD; defined as at least weekly use and Substance Involvement Score of ≥4 on the Cannabis sub‐test of the Alcohol, Smoking and Substance Involvement Screening Test) and 26 cannabis‐naive controls were exposed to cannabis and floral cues during event‐related fMRI. Between‐group differences in fMRI activation and correlations were tested using FMRIB’s Local Analyses of Mixed Effects and corrected for multiple comparisons using a voxelwise threshold of z > 2.3 and a corrected cluster threshold of p < .05. Results Both visual and olfactory modalities resulted in significant activation of craving and reward systems, with cannabis odor cues eliciting a significantly greater response in regions mediating anticipation and reward (nucleus accumbens, pallidum, putamen, and anterior insular cortex, supplementary motor area, angular gyrus and superior frontal gyrus) and cannabis picture cues eliciting a significantly greater response in the occipital cortex and amygdala. Furthermore, the CUD group showed significantly increased activation in the ventral tegmental area (VTA), the insula, and the pallidum compared to controls. Within the CUD group, activation in the insula, anterior cingulate, and occipital cortex to bimodal cannabis cues was significantly correlated with self‐reported craving. Conclusion Our multimodal cue reactivity paradigm is sensitive to neural adaptations associated with problematic cannabis use.


| INTRODUC TI ON
Craving is fundamentally associated with the transition from recreational drug use to problematic, compulsive drug-taking behavior (Robinson & Berridge, 1993). Biobehavioral models purport that craving develops through classical conditioning, whereby repeated exposure to environmental cues leads to hypersensitivity to the motivational effects of drugs and drug-associated stimuli through the neuroadaptation of dopaminergic reward structures (Robinson & Berridge, 1993. While some individuals can use drugs without becoming addicted, for others this sensitization of reward structures within the dopamine system intensifies ordinary 'wanting' into excessive drug craving (Robinson & Berridge, 1993. Although cannabis has less potential for addiction compared to other substances such as opioids, an estimated 30% of regular cannabis users will become dependent (Hasin et al., 2015). Everitt and Robbins (2005) state that in the early stages of addiction, a drug is voluntarily taken for its rewarding effects, but a loss of control eventually renders this behavior habitual or compulsive.
This shift from voluntary to compulsive behavior is proposed to reflect a transition from prefrontal to striatal neural network control.
FMRI studies have provided complementary insight into the neural mechanisms of cannabis craving in humans. Increased activation in response to visual cannabis cues compared to neutral cues has been observed in the ventral tegmental area (VTA), anterior cingulate cortex (ACC), orbital frontal cortex (OFC), striatum, insula, cerebellum, thalamus, pre-and postcentral gyri, inferior parietal lobe, and superior temporal gyrus (Cousijn et al., 2012;Filbey, Schacht, Myers, Chavez, & Hutchison, 2009). In addition, activation in the occipital cortex, hippocampal regions, superior temporal pole, and middle occipital gyrus has been shown to be positively correlated with subjective reports of cannabis craving (Charboneau et al., 2013), suggesting these regions may be sensitive to individual differences in addiction severity.
In addition to measures of craving, fMRI activation to cannabis cues has been studied in the context of other problems related to cannabis use. For example, fMRI activation in the nucleus accumbens (NAc) and OFC following exposure to cannabis cues was significantly positively correlated with the Marijuana Problem Scale (Filbey et al., 2009). More recently, a hierarchical linear regression analysis showed that fMRI activation in the putamen at baseline was an independent predictor of the total Cannabis Use Disorder Identification Test (CUDIT) score at a three-year follow-up visit, over and above behavioral measures including baseline cannabis use and problem severity, baseline alcohol use and problem severity, baseline nicotine dependence, baseline number of cigarettes per day, baseline craving, and baseline lifetime use of other psychotropic substances (Vingerhoets et al., 2016). Notably, the putamen has a very high expression of CB1R and is adjacent to the pallidum, which has the highest expression of CB1R of all structures implicated in reward and addiction.
Because drug craving often persists (or can resurface) long after drug use has stopped, craving is strongly associated with relapse. cue reactivity has been found to predict treatment outcome and relapse in cigarette, alcohol, and heroin addiction (Grusser et al., 2004;Janes et al., 2010;Marissen et al., 2006;Payne, Smith, Adams, & Diefenbach, 2006), and although limited research has been conducted using cannabis cues, it may also play a role in cannabis use disorders. Understanding factors that contribute to relapse will be critical not only to understanding the process of addiction, but also to developing effective therapies (Jasinska, Stein, Kaiser, Naumer, & Yalachkov, 2014).
Most cannabis cue reactivity studies have utilized pictures, auditory imagery scripts, or tactile cues. While these cues induced craving, there is undoubtedly wide individual variability in their direct relevance across participants. The cues used in experiments are critical, because it is well known that context plays a large role in the expression of sensitization, and individuals with substance use disorders tend to experience craving most strongly when they are in particular drug-associated contexts (Anagnostaras & Robinson, 1996;Anagnostaras, Schallert, & Robinson, 2002;Robinson, Browman, Crombag, & Badiani, 1998;Stewart & Vezina, 1991). Thus, for example, visual stimuli depicting paraphernalia an individual has never used in a room he/she has never entered may lead to an fMRI brain response that is weaker than would be expected, given the degree of neural sensitization that is neuroanatomically present.
A cannabis paradigm that utilizes a cue with more universal relevance would improve our ability to study the neurobiological basis of craving and its role in the development of addiction and vulnerability to relapse. We propose that capitalizing on the unique odor of cannabis will bring us a step closer to this goal. The behavioral evidence available suggests that olfactory cues in combination with visual, tactile, and/or auditory cues can produce or increase craving. In one study (Haughey, Marshall, Schacht, Louis, & Hutchison, 2008), participants were exposed to a used pipe or bong and asked to focus on it, smell it, and imagine what it would be like to smoke cannabis out of it. Subjective craving following this cue exposure was shown to increase craving over and above the baseline measurement obtained after 5 days of abstinence in daily cannabis users (Haughey et al., 2008). Another study used virtual reality simulations including audio, visual, olfactory, and vibrotactile stimuli. Participants exposed to a "party room" of people smoking cannabis and to a room containing cannabis-related paraphernalia reported higher drug craving and attention to cannabis-related cues. Importantly, once they left the cannabis rooms, they returned to baseline in terms of craving/ thoughts about smoking (Bordnick et al., 2009). However, neither of these studies were designed to look at the specific role of olfaction, nor at how the integration of multisensory cues modulates the neural craving response.
Our experiment tested a new multimodal cannabis cue reactivity paradigm that included unimodal pictures, unimodal odors, and bimodal cues combining pictures and odors, and examined whether odor stimuli activated mesocorticolimbic regions to a greater degree than picture stimuli. Next, we tested whether our various combinations of cannabis stimulus types and contrasts showed increased activation in mesocorticolimbic regions in CUD participants compared to controls. We hypothesized greater activation to cannabis cues in the CUD group but no significant group difference in response to neutral non-cannabis (flower) cues on the basis of incentive sensitization theory, positing sensitization as a response to cues predicting drug availability (Robinson & Berridge, 2008). Next, we examined whether higher activation of brain regions involved in craving (prefrontal cortex, anterior cingulate, orbital frontal cortex, hippocampus, and insula) (Koob & Volkow, 2016) was correlated to higher self-reported craving measured after the cue reactivity fMRI scan.
Lastly, we were interested in exploring whether degree of brain activation to cannabis cues would be related to shorter delay in actual use of cannabis in the 24 hr following the end of the cue-exposure paradigm among those at risk for CUD who are cannabis-deprived prior to the visit. To test this, participants were contacted the day after their MRI visit and asked about their cannabis use in the past 24 hr.
Inclusion in the CUD group was determined using the Cannabis subtest of the Alcohol, Smoking and Substance Involvement Screening Test (ASSIST; Humeniuk et al., 2008); participants who qualified as moderate to high risk for a cannabis use disorder (ASSIST Substance Involvement Score ≥ 4) and also reported weekly to daily cannabis use in the prior year were enrolled in the CUD group. Inclusion in the control group was based on self-report of no lifetime history of cannabis use. Participants in both groups were additionally screened using a semi-structured interview and excluded for the following: received a diagnosis of or received treatment for schizophrenia or other psychotic disorder, bipolar disorder, or depression within the past 6 months; reported high risk alcohol use (CAGE score > 2), or reported Moderate to High Risk use of other illicit substances (ASSIST Substance Involvement Score ≥ 4 for each substance reported, e.g., inhalants, cocaine); or reported current psychotropic medication, significant neurological medical history, clinically diagnosed hyposmia or anosmia, MRI contraindications, or left-handedness.
Details on illicit substance use in our participants are included in the Supporting information section. Current tobacco use was not an exclusionary criterion for enrollment in either group and therefore was not assessed during screening. Following data acquisition, two participants (1 CUD, 1 control) were excluded from analyses due to subsequent report of psychotropic medication use, one CUD participant was excluded because his permanent retainer caused severe signal drop-off, and another CUD participant was excluded because of technical issues with the olfactometer. The final sample included 25 controls and 25 CUD participants.

| Procedures
The following study procedures were approved by the University of Washington Human Subjects Division Institutional Review Board and informed written consent was obtained from all participants.
Participants were compensated for their participation, receiving $50 at the end of the research visit and an additional $25 after completing the follow-up phone interview one day later.

| Substance use questionnaires
All participants were asked to refrain from using cannabis for at least 48 hr prior to the research visit. Three CUD participants (out of 25) abstained from cannabis use for less than 48 hr (34.35 hr, 46.06 hr, 47.60 hr).
Before entering the scanner, participants were screened to ensure they did not exhibit symptoms of upper airway breathing disorders or acute cold symptoms. Then, they were asked to complete questionnaires pertaining to current and prior use of substances, specifically of cannabis, alcohol, and tobacco. The Marijuana Craving Questionnaire-Short Form (MCQ-SF) provided a measure of participants' level of self-reported craving for cannabis at the time of the research visit (Heishman et al., 2009). The Cannabis Use Disorder Identification Test-Revised (CUDIT-R) was used to assess severity of problematic cannabis-related behaviors (Adamson et al., 2010).
Detailed information was also collected regarding participants' individual histories of cannabis use, including age of first use, frequency of use, and duration of frequent use (total number of years with at least weekly use). The Tobacco sub-test of the ASSIST was used to collect information on concurrent tobacco use (Humeniuk et al., 2008). The Alcohol Use Disorder Identification Test (AUDIT) was administered to quantify participants' alcohol use and dependence symptoms (Saunders, Aasland, Babor, de la Fuente, & Grant, 1993). In a follow-up phone call approximately 24 hr after the end of the MRI visit, participants were asked: (a) "Have you used marijuana since you had your brain scan?" (b) "What time did you begin using marijuana?" (c) "How much marijuana have you used since you had your brain scan at [time the participant's scan ended]?" Participants were notified in advance that they would be receiving a follow-up phone call but were not given advance information about the content of the phone call to avoid influencing their post-visit behavior. A full summary of participants' demographic information and substance use measures is available in Table 1; additional details about participants' patterns and quantities of cannabis use are available in Supporting information.

| fMRI scan
A fast event-related fMRI scan, adapted from Gottfried and Dolan (2003), was collected as part of a longer imaging protocol. Before and after the beginning of the fMRI task, participants were administered the Visual Analog Scales (VAS) over the intercom and asked to rate how much they agreed with the VAS statements on a scale from 1 to 10. The VAS statements were: (a) "I crave marijuana right now." (b) "I want to use marijuana right now." (c) "Marijuana sounds very appealing to me right now." After the participants exited the scanner (approximately 15 min after completing the fMRI task), they were asked to rate how pleasant they found the cannabis odorant (range of −3 unpleasant to 3 pleasant) and how much the odorant smelled like cannabis (range of 0 = not at all to 6 = immediately recognizable).
During the fMRI task, participants were exposed to unimodal and bimodal stimuli that included pictures of cannabis products/par- Odorant pathways merged at the nosepiece delivery tube, which was mounted on the head coil adjacent to the participant's nose, which was located approximately 125 mm away routed through 1/8″ in diameter tubing. Care was taken to verify that the flow rate did not differ upon switching between the control and odorant paths to within the precision of the flowmeters when tested with minimal (<1 m) tubing lengths. The total flow rate was 1.75 L/m, and the steepness of stimulus onset was 30 ms.
Stimuli were presented with Presentation™. Participants were instructed to prepare to breathe in when they saw a yellow cross-hair and to inhale through their nose when the green cross-hair appeared.
They were informed that sometimes they would be able to smell a flowery smell, sometimes they would smell a cannabis-like smell, and sometimes they would smell neither. They were also informed that there was no THC in the cannabis odorant (Faria, Han, Joshi, Enck, & Hummel, 2020). During each 2-s trial, a 1,000 ms yellow warning cue appeared, signaling the participant to pause their breathing and prepare to sniff once the green cross-hair appeared. The green cross-hair was presented in synchrony with the 850 ms baseline stimuli (plain air; n = 22), odor stimuli (O; n = 44) or picture stimuli (P; n = 44) or odor + picture stimuli (OP; n = 44), followed by a 250 ms interstimulus interval. A sniff was made on every trial, regardless of odorant presence. See Figure 1 for an example OP cannabis trial. Within each stimulus modality, half the trials presented cannabis images and/or odors and the other half presented flower images and/ or odors. See Figure 2 for example stimuli. Participants were also exposed to incongruent OP stimuli (e.g., cannabis odor paired with a picture of roses). Incongruent trials were not included in this report.

| Olfactometer details
The olfactometer used in this experiment was built in the Instrument Development Laboratory at the University of Washington Center on Human Development (see Kleinhans et al., 2018, for more detail). The olfactometer design was based on Lorig, Elmes, Zald, and Pardo (1999), with a modification of the odorant cylinders/ manifold and nosepiece. The intent of the olfactometer design was to allow rapid switching between olfactory stimuli without interrupting the flow of air. By locating the solenoid valves outside of the scanner room, the design also ensures that participants do not receive any auditory cues to indicate the changing olfactory stimulus. 0.05 ml of each odorant was dropped onto 1-inch diameter filter papers and placed in their respective odorant chamber of the olfactometer. Participants were not exposed to the odorants prior to the fMRI task.  (Avants et al., 2011).

| MR data acquisition
Higher-level analyses were carried out using FLAME (FSL's Local Analysis of Mixed Effects) stage 1 and stage 2 . Whole-brain voxelwise comparisons of fMRI activation values in CUD and control participants were performed using t-and F-statistics, which were then converted to z-scores by means of a probability integral transformation and thresholded using clusters determined by z ≥ 2.3 and a (corrected) cluster significance threshold of p < .05 (Woolrich, Behrens, Beckmann, Jenkinson, & Smith, 2004). Additional a-priori small volume region of interest analyses were conducted in the right and left nucleus accumbens, right and left pallidum, and the VTA. Correlations between the crave VAS rating acquired after the fMRI task and activation to cannabis cues were also conducted in the CUD group, and thresholded using clusters determined by z ≥ 2.3 and a (corrected) cluster significance threshold of p < .05 (Poline, Worsley, Evans, & Friston, 1997).

| VAS rating
Visual Analog Scales ratings were collected before and after the fMRI cue reactivity tasks. Independent samples t tests were used to analyze group differences on the VAS. For the ratings collected before the fMRI scan, the average "crave" rating scores for the control group was

| Post-scan cannabis use
Cannabis use disorder participants were queried regarding their cannabis use within the first 24 hr after leaving the research visit.
Of the 25 included CUD participants, 13 reported using cannabis within 24 hr of the study visit. Of those participants who re-

| FMRI motion analyses
An independent samples t test was conducted to assess head motion during fMRI acquisition. No significant group differences were found. The absolute root mean square for the CUD group was M = 0.332, SD = 0.140 and for the control group was M = 0.323, SD = 0.128, with p = .820.

| Unimodal cues
Group activation maps for the CUD group were similar for unimodal visual and olfactory cannabis cues compared to the baseline condition. Both cue types activated the bilateral prefrontal cortex, insular cortex, nucleus accumbens, striatum, thalamus, VTA, substantial nigra, cerebellum, and occipital lobe (Figure 3).
When the flower cues were included as the control condition, the activation maps differed considerably from when baseline was modeled as the control condition ( Figure 3)

| Bimodal cues
Cannabis use disorder group activation maps for the bimodal cue condition compared to baseline were spatially similar to the unimodal olfactory cue compared to baseline. Activation included the bilateral prefrontal cortex, insular cortex, nucleus accumbens, striatum, thalamus, VTA, substantial nigra, cerebellum, and occipital   Figure 5).
A detailed report of the significant activation can be found in Table 2.

| Unimodal cues
The CUD group showed greater activation than the control group for the cannabis odor > baseline in the VTA, substantia nigra, In addition, there were no instances of greater activation in the control group than the CUD group for any of the unimodal contrasts ( Figure 6).

| Bimodal cues
For the bimodal cannabis cue > baseline contrast, the CUD group showed significantly greater activation in the same regions that were significant in the unimodal cannabis odor > baseline control condition. These regions included the frontal pole, striatum, and bilateral insula. The region of interest analyses yielded significant activation in the left pallidum (p = .0229), the VTA (p < .0149) but not the nucleus accumbens (p > .05; Figure 6). For the bimodal cannabis cue > bimodal flower cue contrast, the CUD group showed increased activation in the superior parietal cortex (Figure 7). There were no significant differences in the ROIs for this contrast. In addition, there were no instances of greater activation in the control group compared to the CUD group. A detailed report of the significant group differences can be found in Table 3.

| Correlations with VAS Craving rating in the CUD group
A correlation analysis found significant associations between activation to bimodal cannabis stimulus cues and self-reported craving measured directly after the fMRI task. For the contrast bimodal cannabis > bimodal flower, higher levels of activation within the cingulate gyrus, left insular cortex, and occipital cortex were associated with higher levels of self-reported craving following cue exposure ( Note: R = right, L = left. x, y, and z coordinates are in Montreal Neurological Institute space. "Peak region" and "other regions" are labeled using the Harvard-Oxford Cortical Structural Atlas, the Harvard-Oxford Subcortical Structural Atlas, the Cerebellar Atlas, and the Duke Midbrain Atlas. p-values are based on a whole-brain cluster correction for multiple comparisons.

TA B L E 2 (Continued)
planned correlation between fMRI activation to cannabis cues and time interval between the MRI visit and subsequent cannabis use.

| D ISCUSS I ON
This study utilized a new multisensory cannabis cue reactivity paradigm to determine the utility of including odor stimuli in paradigms designed to identify brain regions impacted by problematic cannabis use and associated with self-reported craving. In addition, we tested whether our experimental paradigm increased self-reported craving and resulted in cannabis consumption soon after completing the research visit. Participants were screened and excluded for comorbid substance dependence, history of severe psychiatric disorders, and psychotropic medication use to minimize the influence of psychotropic medication or drugs other than cannabis on our brain measures. We found evidence that multisensory cannabis cues activate reward-related circuitry and are particularly useful for identifying brain regions that are sensitive to individual difference in craving. In addition, although participants reported a significant increase in self-reported craving following cue exposure, most did not go on to consume cannabis soon after the research visit.

| Odor versus picture cues
This study utilized both visual and odor cannabis cues to determine whether cue modality modulated activation within reward circuitry in CUD. To our knowledge, odor stimuli have not been previously utilized in cannabis cue reactivity imaging studies; therefore, the strengths and weaknesses of each stimulus modality for detecting brain changes associated with CUD are unknown. In the contrast identifying brain regions that showed greater activation to the cannabis picture stimuli, we observed significantly greater activation in the entire occipital cortex, the inferior temporal lobes, and the cerebellum. In addition, there was significantly greater activation in the bilateral amygdala, which has been associated with negative emotions and stress (Koob & Volkow, 2016). Notably, the contrast designed to test for greater activation to odor cues compared to picture cues found odor activated brain regions more closely associ-  degree than picture stimuli, and thus are a useful addition to cue reactivity paradigms designed to detect neural changes associated with problematic cannabis use.

| Brain response to cannabis cues in CUD varies by cue type and baseline
We compared brain activation to cannabis cues to a simple baseline condition (i.e., a fixation cross) and to a closely related stimu-  et al., 2016) and sex (Wetherill et al., 2014).
Our comparison of cannabis cues to our simple baseline stimuli in CUD participants yielded widespread robust activation in mesocorticolimbic, insular, cerebellar, parietal and occipital regions to unimodal and bimodal cannabis cues. However, when flower stimuli were the comparison condition, significant activation to cannabis cues was no longer present in several brain regions that are rich in cannabinoid 1 receptors (Parsons & Hurd, 2015) and associated TA B L E 4 fMRI correlations with self-reported craving

F I G U R E 8
Relationship between fMRI activation to bimodal cannabis cues and self-reported craving in the CUD group. Clusters signify brain regions showing a significant (p < .05, corrected) correlation. Scatter plots depicting the relationship between craving and activation (labeled using a white box) are provided for descriptive purposes only. For each participant, a mean z-score was obtained by averaging the z-score of all the voxels within the mask defined by the significant group cluster and plotted against their VAS craving score Insula with reward and addiction, namely the VTA, pallidum, and nucleus accumbens. Instead, greater response to cannabis cues was primarily observed in temporooccipital regions associated with object processing and the dorsal attention network.
The contrast comparing bimodal cannabis cues to bimodal flower cues showed significantly greater activation than what was observed with baseline as the control condition. Specifically, activation in additional regions including the OFC, medial prefrontal cortex, anterior cingulate, insula, and amygdala were observed. In addition, this contrast exclusively yielded activation that showed a significant correlation with self-reported craving in the insula and the anterior cingulate, regions that are putatively associated with addiction-related preoccupation/anticipation (Koob & Volkow, 2016). Correlations between activation and craving were also observed in the visual cortex, a region that is not typically considered part of reward circuitry, yet is overwhelmingly reported as activated in addiction literature (Charboneau et al., 2013;Hanlon, Dowdle, Naselaris, Canterberry, & Cortese, 2014). It is likely that sensory processing is altered in individuals who are addicted to drugs, via cognitive processes known to impact activity in primary visual cortex (Hanlon et al., 2014).
Notably, consistent with the unimodal cue contrasts, activation in the VTA, nucleus accumbens, and pallidum was not significantly higher in response to the cannabis cues relative to the flower stimuli.
These findings suggest that the VTA, nucleus accumbens, and pallidum have a generalized hyperresponsivity to pleasant, rewarding stimuli, while other brain regions such as the OFC, amygdala, insula, and anterior cingulate may show a specific enhancement to cannabis cues. These results are partially consistent with work by Wetherill et al. (2014), who found that activation to sexual cues and cannabis cues were similar to each other, and when directly compared, did not result in any significant differences. The authors of this study concluded that chronic drug use did not result in the devaluation of natural rewards (Wetherill et al., 2014). For further information on brain activation to the flower stimuli in the CUD group, see Figure   S1 and Table S1. to fruit was not because enhancement of the mesocorticolimbic reward system is specific to cannabis cues, but because fruit cues do not reliably activate reward circuitry (Goldstone et al., 2009;Mehta et al., 2012) but see (Frasnelli et al., 2015). The results of the contrast comparing fruit cues to neutral object cues found that increased activation to fruit cues was limited to the thalamus, claustrum, and the cerebellum. Similarly, the non-cannabis-using controls showed greater activation to fruit cues in the posterior cingulate, superior temporal gyrus, thalamus, and cerebellum. These brain regions are not strongly associated with reward processing. Failure to activate the reward system in the Filbey study may be due to the type of food selected as stimuli. Work done by our group previously showed that neural circuits engaged in reward circuitry are selectively attuned to high-calorie food that are perceived as fattening (Schur et al., 2009).
In our study, fattening food, including candy, desserts, pastries, and high-fat savory foods such as pizza, hamburgers, chicken wings, and other fried foods showed robust activation in the midbrain, nucleus accumbens, and other regions associated with reward while nonfattening food, including fruits, vegetables, salads, low-fat meats, and seafood, only showed increased activation in the occipital lobe when compared to neutral objects (Schur et al., 2009). In light of this, future studies comparing food-based rewards might consider stimuli depicting high-calorie food as a natural reward.
In support of our a-priori hypothesis, the CUD group showed increased activation of mesocorticolimbic regions in CUD participants compared to controls. Notably, our CUD participants showed significantly increased neural sensitization in the nigrostriatal pathway, which is a dopaminergic pathway involved in habit formation, and in the VTA and pallidum, which are part of a pathway involved in the reinforcing effects of drugs and relapse (Ahrens, Meyer, Ferguson, Robinson, & Aldridge, 2016;Prasad & McNally, 2016). Regions involved in anticipation of reward and craving, the insula and prefrontal cortex (Koob & Volkow, 2016), also showed significantly higher levels of activation to the cannabis cues compared to our control participants and were correlated to self-reported craving. Overall, our findings are partially consistent with the incentive sensitization theory, proposing that sensitization of reward circuity from substance abuse generates increased incentive salience of drug cues (Robinson & Berridge, 2008). While we found sensitization of reward circuitry to cannabis cues, this sensitization was also present to our control stimulus, suggesting that sensitization may be generalized across rewarding stimuli.

| Limitations
Although our participants were screened for problematic drinking behaviors using the CAGE, many participants in the CUD group still had elevated alcohol use scores on the AUDIT. To better isolate specific cannabis-related effects, a sample of cannabis-using individuals who do not consume alcohol would be optimal, albeit atypical. Alternatively, including a second control group that is cannabis-naïve but is matched on alcohol use (e.g., AUDIT score) to the CUD group would allow us to identify whether cue reactivity to cannabis cues is impacted by comorbid problematic alcohol consumption. A similar approach would also aid in differentiating effects due to cannabis versus tobacco. In addition, education level, socioeconomic status, psychophysiological confirmation of normal olfactory function, and other factors, such as smell sensitivity, that impact olfactory perception were not collected in the present study. Thus, it is possible that observed group differences seen here may be due in part to differences in these potentially confounding variables. Our CUD group did not undergo a clinical diagnostic interview to assess for the presence or absence of Cannabis Use Disorder. Thus, while all participants in the CUD group meet criteria for "at risk" use, this does not preclude them from also meeting DSM-V criteria for Cannabis Use Disorder. In addition, we did not test the reliability and validity of the post-fMRI interview questions, which asked participants to report their substance use over the preceding 24 hr. However short-term recall of substance use using similar measures has been shown to be both reliable and valid in confidential research contexts (Babor, Steinberg, Anton, & Del Boca, 2000;Kypri et al., 2016;Laforge, Borsari, & Baer, 2005;Simons, Wills, Emery, & Marks, 2015) Finally, we did not observe significant group differences when flowers were the baseline condition relative to the cannabis cues.
It is possible that an experimental design that increased cognitive expectancies related to cannabis use (e.g., telling participants that the odorant contained THC (see, e.g., Faria et al., 2020) may have elicited a more robust signal to the cannabis cues in reward regions.

| CON CLUS IONS
Our study found that exposure to both visual and odor-based cannabis cues resulted in significant activation in the neural circuitry involved in craving and reward, specifically the VTA and pallidum and the insula. Although both modalities were sensitive to brain changes associated with problematic cannabis use, a greater neural response was observed to the odor cues in brain regions mediating anticipation and reward, suggesting that cannabis odor stimuli would be a valuable addition to cue reactivity studies.

ACK N OWLED G M ENTS
The authors would like to thank Jason Kilmer at the University of Washington, who generously contributed money toward the MRI scans. We would also like to thank Garth Terry, MD, for allowing us to use, with modifications, his cannabis quantification measure.
Additional funding for this study was provided by the Washington State Dedicated Marijuana Fund administered through the Alcohol and Drug Abuse Institute at the University of Washington.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
NK was responsible for the study concept and design. JS assisted with data collection and calculating summary statistics on our behavioral data and questionnaires. FR built the olfactometer and supervised the odor delivery during the fMRI task. BD and BD from the