Temporally Dynamic Interaction between Drug Cue Reactivity and Response Inhibition: An fMRI Study among People with Methamphetamine Use Disorder

Cue-induced drug craving and disinhibition are two essential components of continued drug use and relapse in substance use disorders. While these two phenomena develop and interact across time, the temporal dynamics of their underlying neural activity and their interaction remain under-investigated. To explore these dynamics, an analysis of time-varying activation was applied to fMRI data from 62 men with methamphetamine use disorder in their �rst weeks of recovery in abstinence-based treatment program. Using a mixed block-event, factorial cue-reactivity/Go-NoGo task, and a sliding window across the task duration, dynamically-activated regions were identi�ed in linear mixed effects models (LMEs). Habituation to drug cues across time was observed in the superior temporal gyri, amygdalae, left hippocampus, and right precuneus, while response-inhibition was associated with the sensitization of temporally-dynamic activations across many regions of the inhibitory frontoparietal network. Cue-reactivity and response-inhibition dynamically interact in the parahippocampal gyri and right precuneus (corrected p-value < 0.001) regions, which show a declining cue-reactivity contrast and an increasing response-inhibition contrast. Overall, the declining craving-related activations (habituation) and increasing inhibition-associated activations (sensitization) along the task duration suggest the gradual recruitment of response-inhibition process and the concurrent habituation to drug cues in areas with signi�cant dynamic interaction. This exploratory study demonstrates the time-variance of the neural activations undergirding cue-reactivity, response-inhibition, and their interaction, and suggests potentials to assess this dynamic interaction. This preliminary evidence provides justi�cations for new avenues in biomarker development and interventions using cue exposure paradigms, which could promote habituation to drug cues and sensitization in inhibitory control regions.


Introduction
The prevalence and health burden of methamphetamine use disorder (MUD) continues to increase globally 1 and in countries such as the US, where 0.4% of the adult population suffers from methamphetamine use disorder 2 and methamphetamine-related overdose rates have tripled from 2011 to 2016 3 .This potential crisis is compounded by the fact that despite decades of research, assessments of MUD are still largely reliant on interviews, self-reported measures, and urinalysis 4,5 , and data on effective interventions for MUD remains inconsistent 6 , with growing calls to better delineate the neurobiology of MUD to identify novel treatment targets and clinically-relevant biomarkers 7 .In tandem with research elucidating the involvement of a plethora of cognitive functions in the MUD 8 , functional magnetic resonance imaging (fMRI) studies on the neurobiology of MUD have characterized a number of functional brain changes associated with cognitive alterations in the MUD, methamphetamine craving, use history and relapse risk, and treatment outcomes 9,10 .
Two central aspects of methamphetamine use disorder are a characteristic reactivity to drug cues (itself involving attentional bias towards drug cues, their increased salience, and ultimately the induction of craving) 11,12 , and failures of executive control and response-inhibition 13 .These phenomena widely

Participants
Sixty-two men with MUD (age: 32.12 ± 5.89) were recruited from addiction treatment centers in Tehran, Iran.Inclusion criteria were (1) Diagnosis of methamphetamine dependence (for at least 6 months) according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition criteria (DSM-IV TR) 42 , (2) abstinence from any substance for at least one week, with the exception of nicotine, based on self-report and con rmed by urine drug screening, (3) right-handedness, determined using the Edinburgh Handedness Inventory 43 , and (4) age between 20 and 40 years.Exclusion Criteria were (a) any comorbid axis-I disorders other than drug dependence, based on DSM-IV TR criteria, (b) ineligibility for MRI scanning (e.g., metal implants, claustrophobia), (c) history of head trauma resulting in neurological disorders.Nine participants were excluded from fMRI analyses due to excessive movement during scanning, leaving 53 individuals (see Preprocessing section for details).Demographic and behavioral data of the 53 participants who were included in the analyses are provided in Table 1.
The research protocol was designed and implemented in accordance with the Declaration of Helsinki.After a referral from treatment centers to the research team, individuals were informed regarding the aims of the project, the collected information and measures are taken to ensure anonymity, scanning procedures, the fMRI task and its potential to induce methamphetamine craving, and that they can exit the study at any point with no implications for their ongoing treatment.After the consent form was read both out loud by a psychologist and by the individual to ensure comprehension, participants provided written, informed consent prior to further screening for enrollment.The data collected from each participant was sent to the primary data analyst and anonymized before further processing.The study protocol was reviewed and approved by the ethical review board of the Tehran University of Medical Sciences with the approval code 93-02-98-23869.

Procedures and measures
Participants were abstinent prior to scanning, but were allowed to smoke.After arriving at the imaging center, participants were interviewed by two clinical psychologists, and several measures were administered prior to scanning.Collected data included demographic information, mental status examination, clinical assessments (including drug use pro le, treatment history), risky behaviors pro le, the Barratt Impulsiveness Scale-11 (BIS-11) 44 , the Depression Anxiety and Stress Scale-21 (DASS-21) 45 , and the Positive and Negative Affect Schedule (PANAS) 46 .Methamphetamine craving was assessed using a 0-100 Visual Analog Scale (VAS) before MR scanning.After scanning, participants were again assessed with PANAS and rated their craving (Table 1).To minimize the risk of drug use after the fMRI session, participants were asked to remain in the scanning center for an hour while recovering.fMRI Go-NoGo task Participants were scanned during four consecutive runs of the mixed Go-NoGo task, separated by resting blocks with a xation point.Each run included four 36-second blocks of 24 stimuli, depicting geometric Go-NoGo signs overlaid on background cues.Background images were either blank (black), neutral images, negative emotional cues, or methamphetamine-related cues.Each block contained 18 Go signs (triangles, squares, or diamonds) and 6 NoGo signs (circles).Each stimulus lasted one second and was followed by a jittered inter-stimulus interval generated using a gamma probability density function (mean = 0.5).The blocks were separated by 18-second xation periods in which a white cross was shown on a black background, so each run took 198 seconds.A total of 16 blocks were presented, four of each condition (blank, neutral, negative, drug).The total scanning duration was approximately 13 minutes (Fig. 1).
Participants were instructed to respond as fast as possible when the Go stimuli were presented and to withhold their response to NoGo stimuli.Participants underwent a training test outside the scanner and were informed that both speed and accuracy are important.
The methamphetamine cues have been evaluated in previous studies of Iranian participants 47 , and neutral and negative emotional cues were selected from the IAPS database 48 .The researchers had permission to use the utilized images.Neutral, methamphetamine, and negative cues were matched in terms of visual complexity, brightness, luminance, and color.Go-NoGo signs overlaid on background cues.Background images were either blank (black), neutral images, negative emotional cues, or methamphetamine-related cues.Each block contained 18 Go signs (triangles, squares, or diamonds) and 6 NoGo signs (circles).Each stimulus lasted one second and was followed by a jittered inter-stimulus interval generated using a gamma probability density function (mean = 0.5).The blocks were separated by 18-second xation periods in which a white cross was shown on a black background, so each run took 198 seconds.A total of 16 blocks were presented, four of each condition (blank, neutral, negative, drug).The total scanning duration was approximately 13 minutes (Fig. 1).
Participants were instructed to respond as fast as possible when the Go stimuli were presented and to withhold their response to NoGo stimuli.Participants underwent a training test outside the scanner and were informed that both speed and accuracy are important.
The methamphetamine cues have been evaluated in previous studies of Iranian participants 47 , and neutral and negative emotional cues were selected from the IAPS database 48 .The researchers had permission to use the utilized images.Neutral, methamphetamine, and negative cues were matched in terms of visual complexity, brightness, luminance, and color.

Scanning parameters
Whole-brain T2* weighted images were acquired in a 3. Pre-processing FSL (FMRIB's Software Library, www.fmrib.ox.ac.uk/fsl) version 6.0.3 was used to preprocess structural and functional data 49 .Structural data was skull-stripped to remove non-brain tissue from the structural T1-weighted images using the Brain Extraction Tool (BET).BET parameters were chosen based on each individual skull size.
Functional data were analyzed using the fMRI Expert Analysis Tool (FEAT), part of FMRIB's Software Library.The functional pre-processing included the removal of the rst ve volumes, motion correction with 6 degrees of freedom, interleaved slice-timing correction, linear Boundary-Based Registration (BBR) of functional images to the high-resolution T1 images, nonlinear registration of the T1 images to the standard Montreal Neurological Institute (MNI) space with 12 degrees of freedom, intensity normalization, smoothing with a 5-mm full-width at half-maximum (FWHM) Gaussian kernel, denoising with melodic ICA, high-pass temporal ltering (with the cut-off frequency equal to the inverse of 120 seconds).
High motion effects on fMRI time series were identi ed using the DVARS metric 50 and were regressed out in the rst-level generalized linear model (GLM) analysis."High movement subjects" were de ned as those with displacement > 4 mm and also DVARS > 75 in more than ten volumes in a single block (36sec), and were excluded from the analyses (9 subjects).

Conventional whole-brain analysis
The pre-processed functional images were analyzed in a GLM framework.Event-types were speci ed at the time of indicator onset, and the canonical hemodynamic response was used to model the regressors for the conditions of interest.The event types included Neutral Successful NoGo (NSNG), Neutral Successful Go (NSG), Drug Successful NoGo (DSNG) and Drug Successful Go (DSG), blank and negativeemotional successful Go and NoGo trials, and unsuccessful Go and NoGo trials, were included in the GLM as independent regressors.Six head motion parameters and high motion time-points extracted based on the DVARS metric were included as nuisance regressors.
To determine the time-invariant neural correlates of methamphetamine cue-reactivity, response-inhibition and response-inhibition during cue exposure, each event type was included as a single regressor and three contrasts were de ned: (DSNG + DSG) > (NSNG + NSG) to model cue-reactivity, (DSNG + NSNG) > (DSG + NSG) to model inhibition and (DSNG > DSG) > (NSNG > NSG) to model the interaction of cuereactivity and inhibition.
To calculate average activations patterns, rst-level models were then carried forward into a second-level mixed effects analysis using FMRIB's Local Analysis of Mixed Effects (FLAME) tool with a cluster de ning threshold (Z-threshold > 3.1, corrected cluster-level threshold: p < 0.001).
Temporally dynamic fMRI analysis ROI-based whole-brain analyses were performed using the Brainnetome atlas (BNA) 51 .First, the whole brain was parcellated into 246 regions based on the BNA.The BNA masks in MNI space were then registered to each subject's space using the transformation matrices derived from the pre-processing step and after determining subject-speci c masks for each ROI across the 53 subjects, the mean activations and standard errors were calculated.We then used a sliding window over the BOLD response to investigate temporal variability, with a window duration of two task runs (396 seconds) and a sliding interval equal to one run (198 seconds) leading to the extraction of three overlapping windows.We modeled each of the four event types with separate regressors across the three windows, so every participant had six beta coe cients estimated for each of the three contrasts (cue-reactivity, inhibition, and interaction).
Then, three Linear Mixed Effects (LME) models were t to the fMRI data in R, version 3.6.2 52.The models were all speci ed as "beta ~ condition * time" with the condition, time, and their interaction as xed effects and the subject as a random effect.The conditions were (DSNG + DSG) or (NSNG + NSG) in the cue-reactivity LME, (DSNG + NSNG) or (DSG + NSG) in the inhibition LME and (DSNG > DSG) or (NSNG > NSG) for the interaction LME.Time was treated as a discrete variable with integer values of 1 through 3 (for the rst through the third window) which were mean centered.For each of the three models, regions with a signi cant main effect of time and condition and those with a signi cant condition-by-time interaction were identi ed after a False Discovery Rate (FDR) correction with a threshold of p < 0.001.Lastly, for each model, the temporal activation dynamics of signi cant regions were examined by plotting the beta values for the relevant conditions across the three windows.

Correlation of dynamic activity and behavioural data
For every ROI exhibiting a signi cant condition-by-time interaction in each model, separate "beta ~ condition * time" linear models (LMs) were t for each subject and the slope of the interaction term was taken to be an index of the evolution of condition-relative individual-level dynamic activity in the ROI.For cue-reactivity LMs the slope would represent sensitization to drug versus neutral cues across time, in inhibition LMs it would represent temporally escalating activation when inhibiting versus not inhibiting pre-potent responses, and in interaction LMs it would be a subject-level re ection of an increased neural load required to successfully inhibit responses in the presence of drug versus neutral stimuli.Then, the correlations of these individual-level beta values in the signi cant ROIs extracted from LMs with behavioural data were assessed in each of the three contrasts separately.Correlations between pre-and post-scanning craving and dynamic cue-reactivity slopes, between Barratt impulsiveness sum scores and commission error rates and inhibition slopes, and between all four variables and interaction contrast slopes were estimated.The correlation of methamphetamine use duration and all three slopes was also assessed.Lastly, differences in the three slopes in each signi cant ROI between subjects who hadn't used methamphetamine in the month prior to scanning and those who had were assessed using t-tests.As tests would involve slopes from a number of regions that were dynamically active for each condition, FDR-corrected thresholds of p < 0.05 were used for these tests.

Time-invariant activations
Conventional whole-brain analysis results for the three contrasts can be viewed in Fig. 2. We illustrated whole-brain maps (Z-threshold = 3.1, corrected cluster-level threshold: p < 0.001) as well as activation changes across the 246 regions using BNA parcellation.
Whole-brain maps of activations in each of the three windows can be viewed in the supplementary materials (Fig. S1, Fig. S2, and Fig. S3).
Subcortical regions such as the amygdala and hippocampus, as well as the STG and Pcun show a decreasing response over time (habituation) to drug cues compared to neutral cues.(Fig. 4).
Since dynamic inhibition was observed in 107 regions, we only explored the temporal inhibitory and noninhibitory behavior of regions shown to be involved in response-inhibition in a recent meta-analysis of fMRI Go-NoGo tasks 53  showed increasing activations (sensitization) (Fig. 5).
Table 2 Regions with signi cant condition by time interactions in the cue-reactivity and cue-reactivity/responseinhibition interaction linear mixed effects (LME) models.All p-values are FDR-corrected.

Dynamic interaction of cue-reactivity and responseinhibition
In the LME modeling the interaction of craving and inhibitory processes, the main effect of condition (DSNG > DSG or NSNG > NSG) is highly signi cant across much of the frontal cortex, STG, IPL, Pcun, PoG, dorsal INS, MVOcC, and BG (basal ganglia) (corrected p-value < 0.001) (Supplementary Fig. S4).Regions with a signi cant main effect of time are located in the ITG, FuG, PhG, Pcun, right dorsal and ventral CG, LOcC, and hippocampus (corrected p-value < 0.001) (Supplementary Fig. S5).
For the three ROIs with a signi cant condition-by-time interaction in the cue-reactivity/inhibition interaction LME, the temporal behavior of the three contrasts across the overlapping windows is illustrated in Fig. 6.b.The PhG show decreasing estimates (habituation) for the cue-reactivity (DSNG + DSG > NSNG + NSG) and interaction (DSNG -DSG > NSNG -NSG) contrasts, but increasing values of inhibitory control (DSNG + NSNG > DSG + NSG) across time.In the right Pcun, the cue-reactivity contrast and the interaction contrast decrease (sensitization), but the response-inhibition contrast remains mostly stable.

Correlates of dynamic brain activity
None of the correlations between regional dynamic activation slopes (beta values extracted from the individual-level LMs) and behavioral or clinical variables were signi cant after FDR correction.There are

Discussion
This exploratory study is an investigation of temporally dynamic regional brain activation patterns underlying cue-reactivity, response-inhibition, and their interaction in individuals with MUD.While similar sliding window techniques are relatively common in dynamic functional connectivity analyses 54,55 and despite decades of evidence for temporal variation in regional sensitization and habituation in cognitive/affective neuroscience [31][32][33] , dynamic analyses of regional activation in addiction remain rare and this is the rst study which explored this dynamic interaction in response-inhibition in the context of cue-reactivity.

Dynamic cue-reactivity
Dynamic cue-reactivity was observed in the bilateral STG, the right amygdala, and rostral hippocampus, and the left Pcun and ITG.Many of these regions have previously been indicated in methamphetamine cue-reactivity 16,17,19 , and drug cue-reactivity more widely 56,57 .Notably, dynamic amygdala activity with a similar downward slope over time has been observed in two recent cue-reactivity studies in individuals with MUD and opioid use disorder 37,38 .A study on individuals with heroin use disorder estimating dynamic causal modeling parameters in overlapping windows has also demonstrated craving inputs to the amygdala increase during a cue-reactivity task, and that the DlPFC's (dorsolateral prefrontal cortex) modulatory impact on the connection between the VMPFC (ventromedial prefrontal cortex) and the amygdala decreases over time 58 .Ekhtiari et al. also similarly reported bilateral dynamic cue-reactivities in the STG, but they observed an initially escalating and subsequently decreasing activation whereas we observed a consistent habituation response 37 .Broadly, our results suggest generalized habituation to drug cues across the task duration.
The dynamic cue-reactivity LME showed no signi cant condition-by-time interactions in the VMPFC and the VSTR (ventral striatum), indicating a lack of dynamic activity, unlike another dynamic study using similar analytical procedures 37 .Unexpectedly, these regions also showed no static activity, potentially showing that they were not recruited by our task components.Methamphetamine use duration had a signi cant uncorrected correlation with dynamic cue-reactivity in the right Pcun, similar to a previous study in which addiction severity was found to be correlated with Pcun activation during cue-reactivity tasks 59 .

Dynamic response-inhibition
More than a hundred regions in our LME model showed dynamic response-inhibitory activity.This may not be surprising, as response-inhibition is associated with large-scale neural activity 53 and dynamic brain network recon guration 41 .Also, notable is that dynamic prefrontal activations were also observed in the response-inhibition model, whereas only FDR-uncorrected prefrontal activations were observed in the other two models (cue-reactivity and cue-reactivity/inhibition interaction).There have been reports of prefrontal sensitization to salient cues 36 , and it has been observed that the prefrontal cortex is implicated in the dysfunctional behavioral regulation seen in the MUD during response control tasks 22 .The observation of dynamic activity in prefrontal regions was expected, given their involvement in inhibitory control networks 51 and response-inhibition in substance use disorders 60,61 .
Most of the regions involved in response-inhibition in a recent meta-analysis of Go-NoGo tasks 53 had dynamic activation patterns in this study.Notably, while dynamic cue-reactivity was associated with a generalized habituation effect, these regions showed two broad temporal activation patterns.The MTG, the left INS, the right CG, the right MTG, and supramarginal gyrus, showed falling inhibitory activations while the PrG, the left SPL, and IFG (operculum) showed increasing activations (sensitization).This might re ect differences in response-inhibitory processes that these regions contribute to, such as error monitoring and attentional control 62,63 , or the involvement of these regions in other networks that interact with the response-inhibition network in individuals with substance use disorders, such as the INS in the salience network or the MFG in self-directed processing 14 .
Commission error rates have been used as a measure of response-inhibitory success in Go-NoGo tasks, and have been correlated with activities in the right SPL and DLPFC in individuals with addictive disorders 64 .In our study, commission error rates had signi cant uncorrected correlations with dynamic response-inhibition slopes commission error rates in the ITG, IPL, Pcun, and dorsal and ventral anterior CG, potentially implicating these regions in response-inhibition dysfunctions in the MUD.These regions can play an important role in the development of response control-related biomarkers in addictive disorders 65 .

Dynamic response-inhibition during cue exposure
The bilateral PhG and the right Pcun were the only regions with a dynamic interaction of responseinhibition and cue-reactivity.Several meta-analyses have demonstrated that drug cue-reactivity is associated with heightened precuneal activation 66,67 , and based on the response-inhibition literature, dopaminergic inhibition and network decoupling of precuneal activity may be important for successful response-inhibition [68][69][70] .Precuneal involvement in cue-reactivity in substance use disorders might be related to its role in the default mode network and self-referential processing in general 14 , and, interestingly, it has been argued that the Pcun might be an important node for the integration of contradictory executive control and cue-reactivity processes 71 .Considering the above, the decreasing activation associated with drug-related inhibition in the right Pcun may re ect a lessening effect of drug cues in hampering response-inhibition across the task duration.Since it appears that the responseinhibition contrast in the Pcun is mostly stable across time while cue-reactivity and interaction contrasts decline, habituation to drug cues or top-down suppression of precuneal cue-reactivity, rather than the role of the Pcun in response-inhibition per se, maybe the responsible mechanisms.
The PhG have also been implicated in substance use disorders.Addictive disorders are associated with parahippocampal gray matter changes 72 and increases in its connectivity within the default mode network 73 , both the right and the left parahippocampus generally show higher activations in response to drug-related cues compared to neutral cues 66,74,75 , and response-inhibition-associated parahippocampal dysfunction has been observed in individuals with substance use disorders compared to healthy controls 20 .As part of the default mode network and given its association with drug cue-reactivity, it was expected that similar to the Pcun, the cue-reactivity contrast in the PhG would decrease, re ecting both habituation processes and task-engagement-related suppression.Some evidence also exists for parahippocampal habituation during exposure to emotionally salient stimuli 76,77 and for the role of the parahippocampus in the extinguishing of drug cue associations 78 .However, the parahippocampus is also involved in neural networks involved in associational memory and learning 79,80 and might be activated to support learning during response-inhibition tasks 14 .Indeed, increasing parahippocampal recruitment during a learning task has been observed before 81 .These dual roles of the parahippocampus in cue habituation and learning could explain why the cue-reactivity contrast decreased while the response-inhibition contrast increased in the PhG during the task, and is supported by the observation that drug-related inhibition remained mostly stable, while inhibition during neutral cue exposure was associated with increasing parahippocampal activity.
An interesting observation in this study was the right-lateralization of dynamically active regions across the three contrasts.Some evidence exists that the right hemisphere may be more important in responseinhibitory and attentional control processes 82,83 , and right lateralization of dynamic response to salient stimuli has been observed in the right amygdala, IPL, and hippocampus 31,84 .It has been argued that while the left amygdala is involved in sustained stimulus evaluation, the right amygdala might be more specialized for dynamic stimulus processing 34 .

Limitations
While the results of this exploratory investigation are promising, several limitations are important to point out.Firstly, we included no healthy control group, and so the speci city of observed patterns to individuals with MUD is unclear.Also, all participants were men, treatment seeking individuals MUD, limiting the generalizability of our observations.Regarding the task design, an inherent limitation introduced by our use of a mixed drug cue and negative emotional Go-NoGo task is the potential carryover effects of salient cues on brain activity during subsequent blocks 85,86 .While such issues may be ameliorated by the choice of a blocked presentation of different cue types, the results are likely confounded by these effects.Lastly, we used no measure of craving across the task duration, which would have allowed the analysis of temporal correlations between craving and neural activity, as in one recent study by Murphy et al. 38 .

Conclusion
This study provides preliminary evidence that a mixed event-block Go-NoGo/cue-reactivity task can be used to assess the temporal dynamics of cue-reactivity, response-inhibition, and their interaction.The regions with a temporally dynamic response are involved in various neuro-cognitive aspects of addictive disorders.Notably, we observed dynamic amygdalar activity in both response-inhibition and cue-reactivity contrasts, and there is extensive literature on the time-variance of amygdalar activity 31,34,37,87 , and it has been recently argued that amygdalar habituation is a more reliable index than mean amplitude 88 .The interaction of cue-reactivity and response-inhibition occurred in regions in which the neural activations associated with cue-reactivity and response-inhibition followed broadly opposing slopes across time, namely in the parahippocampal regions and the precuneus, suggesting that these regions may be important hubs where response-inhibitory and cue-reactivity processes integrate.Dynamic interactions in these regions may help biomarker development and suggest new targets for interventions, and it has been suggested that failures to inhibit precuneal cue-reactivity may predict relapse 89 , and impairments of parahippocampal habituation are associated with poorer treatment outcomes in cocaine users 39 .Future studies could make use of better power analyses, exible sliding window sizes and inference methods, prospective designs, and replication across different populations and time-points to assess the stability, generalizability, and potential predictive utility of these dynamic activation and interaction parameters.

Figure 6 a
Figure 6

Table 1
Demographic and the pro le of Methamphetamine users (n=53)