Blunted Neurobehavioral Loss Anticipation Predicts Relapse to Stimulant Drug Use

BACKGROUND: Patients with stimulant use disorder experience high rates of relapse. While neurobehavioral mechanisms involved in initiating drug use have been studied extensively, less research has focused on relapse. METHODS: To assess motivational processes involved in relapse and diagnosis, we acquired functional magnetic resonance imaging responses to nondrug (monetary) gains and losses in detoxi ﬁ ed patients with stimulant use disorder ( n = 68) and community control participants ( n = 42). In a prospective multimodal design, we combined imaging of brain function, brain structure, and behavior to longitudinally track subsequent risk for relapse. RESULTS: At the 6-month follow-up assessment, 27 patients remained abstinent, but 33 had relapsed. Patients with blunted anterior insula (AIns) activity during loss anticipation were more likely to relapse, an association that remained robust after controlling for potential confounds (i.e., craving, negative mood, years of use, age, and gender). Lower AIns activity during loss anticipation was associated with lower self-reported negative arousal to loss cues and slower behavioral responses to avoid losses, which also independently predicted relapse. Furthermore, AIns activity during loss anticipation was associated with the structural coherence of a tract connecting the AIns and the nucleus accumbens, as was functional connectivity between the AIns and nucleus accumbens during loss processing. However, these neurobehavioral responses did not differ between patients and control participants. CONCLUSIONS: Taken together, the results of the current study show that neurobehavioral markers predicted relapse above and beyond conventional self-report measures, with a cross-validated accuracy of 72.7%. These ﬁ ndings offer convergent multimodal evidence that implicates blunted avoidance motivation in relapse to stimulant use and may therefore guide interventions targeting individuals who are most vulnerable to relapse.

https://doi.org/10.1016/j.biopsych.2023.07.020Stimulant drugs (e.g., amphetamine, cocaine) are powerfully addictive and induce chronic relapse.In the United States, in 2020, an estimated 4.6 million people were diagnosed with stimulant use disorder (SUD), representing a 60% increase from 2019 (1,2).Mortality rates from stimulant use are also rising (3,4) because deaths involving methamphetamine overdose tripled from 2011 to 2016, and cocaine consistently ranked as the second or third drug most responsible for overdose-induced deaths (5).Currently, no medications have been approved for the treatment of SUD.While contingency management offers the most effective psychosocial treatment, it does so only in the continued presence of reinforcers (6).Even after treatment, relapse to stimulant drug use remains a persistent problem because 50% to 60% of patients with SUD relapse within a year after leaving treatment (7).Therefore, reliably identifying risks for relapse and developing targeted SUD treatments remain clinical priorities.
Consistent with theories that drug cues can trigger seeking responses (8,9), neuroimaging research suggests that increased mesolimbic responses to stimulant drug cues can prospectively predict relapse in patients with SUD (10,11).Because psychostimulant drugs directly influence activity in circuits that are modulated by mesolimbic projections of midbrain dopamine neurons (12,13), adaptations of these circuits may promote drug seeking as well as relapse after abstinence.However, associations between neural responses to nondrug incentive cues and drug relapse have been explored less often.One robust and replicable paradigm for probing neural responses to nondrug incentives (i.e., uncertain monetary gains and losses) is the monetary incentive delay (MID) task (14).Several studies have used the MID task to examine incentive processing in the context of stimulant addiction [see (15)(16)(17) for reviews] but have produced mixed findings.Some studies have reported reduced striatal responses to gain cues and increased striatal responses to gain outcomes in patients with SUD relative to control participants (16,18), but others have not (19).Fewer neuroimaging studies using nondrug incentives have focused on losses than on gains (15,16,(18)(19)(20).One study that included loss incentives found that current and past cocaine users showed reduced insula activity during loss anticipation relative to control participants (21).Another study found evidence that anterior insula (AIns) responses to monetary risks could predict relapse to stimulant use (22).These neurobehavioral responses to loss may have theoretical relevance to relapse prevention because a primary feature of addiction involves continuation of drug use despite negative consequences (as specified by clinical diagnostic criteria).
In this research, we investigated whether neurobehavioral responses to nondrug incentives could predict stimulant use relapse by combining functional magnetic resonance imaging (fMRI) with the MID task (14).We scanned detoxified patients with SUD (n = 68) and healthy control participants (n = 42) and then assessed them 1, 3, and 6 months after leaving treatment (Figure 1A).Analyses focused on preregistered brain regions and task phases that are typically recruited in healthy samples during the MID task (23) (i.e., the nucleus accumbens [NAcc], AIns, medial prefrontal cortex, and ventral tegmental area) and tested for group differences in relapse at the 6-month followup.Given the dearth of research using neural responses to nondrug incentive cues to predict stimulant use relapse, we preregistered nondirectional hypotheses.Furthermore, we sought to explore the association of functional neural markers with behavior, affect, structural connectivity [diffusionweighted MRI data from the same sample previously published (24)], and functional connectivity.Finally, because research has implicated different circuits in the initiation of drug use versus relapse (17,24,25), we examined the preregistered hypothesis that neural predictors of relapse would differ from neural alterations observed between patients and control participants at baseline.

METHODS AND MATERIALS
The study was approved by the institutional review boards of the Stanford University School of Medicine and the Research and Development Office of the Veterans Affairs Palo Alto Health Care System.Prior to participating in the study, all individuals provided written informed consent.The study's hypotheses and analyses were preregistered (https:// aspredicted.org/blind.php?x=ZXV_IAM).

Blunted Loss Anticipation Predicts Stimulant Relapse
Participants Seventy-nine (n = 68 after exclusions) detoxified patients with SUD were recruited from a residential SUD treatment program at the Veterans Affairs Palo Alto Health Care System in Palo Alto, California (n = 65 total; n = 58 after exclusions) [see also (10)], as well as a residential recovery program for women at the Epiphany Center in San Francisco, California (n = 14 total; n = 10 after exclusions) from 2015 to 2020.All patients had a current diagnosis of SUD for stimulant drugs (e.g., methamphetamine, crack, and powder cocaine) and sought treatment primarily for problematic stimulant use.Patients completed a urine toxicology test prior to scanning to verify abstinence.After excluding for head motion during the task scan (n = 8), insufficient task engagement (n = 3), and being lost to followup (n = 8), the final sample included 68 patients for baseline analyses and 60 patients for analyses of relapse (see Figure S1 for details).
Healthy control participants included individuals without current or past stimulant dependence who were members of Stanford University's paid psychology experimental pool and the surrounding community (n = 42).A subset of control participants was U.S. military veterans (n = 12) (see Table S1 for demographics and Table S2 for additional clinical descriptions of patient subgroups).All participants were compensated with an Amazon gift card, which included a base payment of $100 plus an additional $0 to $30 based on task performance (i.e., the cumulative outcome of all trials on the MID task).
Task and Experimental Procedure fMRI Task.All participants performed the MID task (14) during fMRI scanning.This version of the MID task included 6 cued conditions (2$5.00, 2$1.00, 2$0.00, 1$0.00, 1$1.00, and 1$5.00) with 15 trials each, or 90 trials total, presented in a pseudorandom order.Each trial began with a cue (2 seconds), followed by a fixation cross (2 seconds), followed by a period that included a briefly presented target (160-300 ms with a jittered onset during a 2-second window), followed by presentation of an outcome (2 seconds), and then a central fixation cross presented during a variable intertrial interval (2, 4, or 6 seconds) (Figure 1B).Participants were instructed to press a button rapidly during presentation of each target to either gain or avoid losing money (Figure 1C).Adaptively timed target durations ensured hits on approximately 66% of the targets within each condition (26).Participants first performed practice trials outside the scanner, which yielded average reaction times (RTs) for setting initial target durations.After scanning, participants rated their affective responses to each of the cues (6$0, 6$1, 6$5) on valence and arousal scales (7 points ranging from either "very negative" to "very positive" or "not at all aroused" to "highly aroused," respectively) and completed additional individual difference questionnaires.Affect ratings were not reported in 4 control participants and 4 patients (1 abstained, 3 relapsed), leaving data for 38 control participants and 56 patients (n = 26 abstained and n = 30 relapsed) for this analysis.
Questionnaires.Participants also completed self-reported individual difference questionnaires including the Barratt Impulsivity Scale-11 (27), the Kirby Monetary Choice Questionnaire (21 items) (28), and the Beck Depression Inventory (29) (Tables S1 and S2).Finally, patients (but not control participants) completed a Brief Addiction Monitor questionnaire, which assesses psychological factors and behaviors relevant to substance dependence over the past month (30).
Follow-up Assessments.Patients were subsequently contacted for follow-up interviews 1, 3, and 6 months post treatment discharge.Using the timeline follow-back method (31) during follow-up assessments, two treatment outcomes were considered: a binary relapse measure at the 6-month follow-up and a continuous number of days to relapse measure (see Supplemental Methods for details).

fMRI Acquisition
Neuroimaging data were acquired on a 3T General Electric DISCover MR750 scanner with a Nova Medical 32-channel head coil at the Stanford Center for Cognitive and Neurobiological Imaging.T1-weighted structural images as well as functional T2*-weighted echo planar images were acquired for all participants.The following parameters were used for functional image acquisition: echo time = 25 ms, repetition time = 2.0 seconds; voxel dimensions = 2.9 mm isotropic; acquisition matrix = 80 3 80, number of slices = 46; phase encoding = anteroposterior; field of view = 232 3 232 mm; flip angle = 77 .These parameters were optimized to resolve subcortical fMRI responses that are typically elicited by incentive tasks (32).The MID task was performed over 2 runs (526 seconds and 596 seconds) within the same session.
fMRI Data.Analyses of preprocessed fMRI data included both whole-brain and volume-of-interest (VOI) (confirmatory within-group, exploratory between-group) approaches, and VOI approaches tested the primary predictions.Logistic regression analyses identified which neural marker predicted relapse (after correcting for multiple VOIs and task phases; n = 9; Bonferroni-corrected p , .0056).To verify the robustness of identified predictors, we implemented a "regression multiverse" analysis, which facilitated visualization of the strength of a predicted effect in the presence of all possible combinations of potential confounds (24,34).Potential confounds included age, gender, craving and negative mood (past 30 days), and years of use, based on previous work on relapse prediction (10,24) and following the practice of including less than one regressor for each 10 data points in the analysis (35) (see Supplemental Methods for details).

Blunted Loss Anticipation Predicts Stimulant Relapse
Biological Psychiatry --, 2023; -:---www.sobp.org/journalStructural Coherence.Our previous work indicated that structural coherence (indexed with fractional anisotropy and inverse radial diffusivity) of a white matter tract connecting the right AIns to the NAcc predicted relapse to stimulant use (24).
Because this effect was found in the same cohort as was used in the current study, we sought to integrate this structural predictor with identified functional predictors of relapse and to test whether they could account for overlapping or distinct  S3.The same contrasts for control participants and patients at baseline are depicted in Figure S8.R, right.
Blunted Loss Anticipation Predicts Stimulant Relapse variance in relapse.To compare structural with functional measures, we used a summary metric of right AIns-NAcc tract fractional anisotopy calculated for each participant in the current sample estimated from structural tract nodes that predicted relapse [see (24) for detailed methods].
Correlated Functional Activity (or Functional Connectivity).Auxiliary analyses assessed correlated activity (sometimes called functional connectivity) between AIns and NAcc VOIs.Correlated activity between these regions was calculated during the full length of each trial plus the hemodynamic lag (approximately 4-6; i.e., trial onset 1 16 seconds) separately for each condition.Correlation coefficients for each participant were converted to z scores using a Fisher r-to-z transformation prior to analyses, which incorporated a stringent motion censoring criterion (omitting any brain volumes with .0.5-mm displacement).

Blunted Loss Anticipation Predicts Stimulant Relapse
Biological Psychiatry --, 2023; -:---www.sobp.org/journalBiological Psychiatry confounding demographic variables (age, gender) as well as clinical variables (craving, negative mood, and years of stimulant use) further verified the robustness of the association of blunted AIns activity during loss anticipation with subsequent relapse (Figure 3F).Specifically, a multiverse regression approach quantified how the effect size of interest changed after controlling for various combinations of confounding variables (24,34).The effect of AIns loss anticipation activity on relapse status ranged from z scores of 22.03 to 22.53 across all multiverse regression models, suggesting that the association could not be explained by individual difference confounds.Furthermore, frequency of psychotropic medication use did not differ significantly between patients based on follow-up relapse status (Table S5).
Behaviorally, blunted AIns activity during loss anticipation was associated with slower response times and reduced negative arousal to anticipated loss.As expected, participants generally responded more rapidly to high-incentive (6$5) versus no-incentive (6$0) cues, suggesting increased motivation (Figure S3) (a main effect of trial type on RTs F 5,540 = 24.1,p , .001).Comparison of RTs in the patient sample revealed that patients who subsequently relapsed responded more slowly to large loss cues (2$5) than patients who abstained (t 58 = 2.68, p = .01,Cohen's d = 0.70) (Figure 4A).Specifically, slower RTs on 2$5 loss trials independently predicted relapse at the 6-month follow-up (z = 2.45, p = .014)(Figure 4B), but not in any other condition (all zs , 1.16, all ps ..25).Further shedding light on the convergence of this motivational effect with neural signatures, slower RTs on those trials were associated with reduced AIns activity during anticipation of 2$5 loss (R = 20.25,p = .029)(Figure 4C), but not in any other condition (all Rs , 0.2, all ps ..12) (Figure 4D).This

Blunted Loss Anticipation Predicts Stimulant Relapse
specific association between anticipatory AIns activity and RT remained significant when contrasting 2$5 versus 2$0 loss trials (R = 20.34,p = .004)(Figure S4).Finally, reduced AIns activity during anticipation of large losses was associated with reduced negative arousal in response to the 2$5 cues (R = 20.28,p = .019),but was not associated with general negative mood, Beck Depression Inventory scores, or craving (Figure S5), suggesting that this neural marker is a motivationally relevant affective signal.
Convergent behavioral, structural, and functional measures additively predicted relapse to stimulant use.Previous research on this sample revealed that lower fractional anisotropy of the right AIns to right NAcc tract predicted relapse (24).Because the current findings implicated AIns activity in predicting relapse, we hypothesized that individual differences in AIns activity particularly during a motivationally relevant context (loss anticipation) might be associated with structural coherence of the AIns-NAcc tract.Indeed, we found that loss anticipatory activity in the right AIns was positively correlated with right AIns to right NAcc tract coherence (R = 0.30, p = .028)(Figure 4E), and this association was specific to the large loss condition (other conditions: all Rs , 0.18, all ps ..18) (Figure 4F).
Additional analyses examined conditional correlated activity (or functional connectivity) between preprocessed activity time series averaged over the AIns and NAcc VOIs, targeting activity during large loss trials.These findings indicated that right AIns to NAcc tract coherence was associated with correlated activity between the right AIns and right NAcc during trials involving large losses (R = 0.30, p = .025)(Figure 4G).This association was marginal for the small loss trials (R = 0.24, p = .075)but not significant in other conditions (all Rs , 0.19, all ps ..16) (Figure 4H).However, right AIns-NAcc white matter tract coherence and correlated activity between the right AIns and right NAcc across the entire task were not significantly associated (R = 0.195, p = .157).These findings suggest that the pathway between the AIns and the NAcc is specifically relevant in the context of loss anticipation and subsequent avoidance (see Figure S6 for a graphical summary).
Logistic regression models estimated the relative contributions of neural and behavioral variables to the classification of relapse.Classification of treatment status was assessed using leave-one-out cross-validation.Neural and behavioral variables all contributed independently and additively to explaining variance, as well as in the accurate classification of treatment outcomes.A model combining functional activity (AIns loss anticipation), behavior (loss avoidance RTs), and brain structure (AIns-NAcc tract coherence) yielded the highest crossvalidated classification accuracy of 72.7% (Table 1 and Table S6).Cross-validated classification accuracy was lower for models that included demographic (age, gender) and/or clinical (craving, negative mood, years of use) variables, ranging from 55.3% to 65.2% (Table S7; see Figure S7 for pairwise associations).
Predictors of relapse did not differ between patients and control participants overall.Specifically, patients did not differ from control participants in AIns activity during loss anticipation in either whole-brain (Figure S8) or VOI analyses (t 108 = 21.04,p = .30)(Figure S9A).Whole-brain and VOI analyses also revealed similar neural responses to gain and loss anticipation and outcomes in patients and control participants (Figures S8-S10; Table S4).Patients showed marginally greater increases in NAcc activity in response to gain outcomes compared with control participants (t 108 = 1.80, p = .074,Cohen's d = 0.35).Although nonsignificant, this trend/ pattern is consistent with previous meta-analytic evidence of increased striatal responses to gain outcomes in patients with SUD versus control participants (16,18).Patients reported less negative arousal to large loss cues overall (t 100 = 24.87,p , .001) (Figure S3E).Behaviorally, patients also responded similarly to control participants during loss trials (t 108 = 20.98,p = .33)(Figure S3A) but responded more slowly during nonincentive trials (1$0 or 2$0, i.e., control trials; 1$0 RTs: t 108 = 23.42,p , .001;2$0 RTs: t 108 = 22.18, p = .03)(Figure S3A), consistent with less behavioral discrimination of nonincentives from incentives.

DISCUSSION
This research investigated whether neurobehavioral responses to monetary gains and losses could predict relapse to These findings add to an extensive literature linking AIns activity to general arousal (38), negative arousal (33,39,40), risk avoidance (41)(42)(43)(44)(45), incentivized inhibition (46), and addiction (47,48).Taken together, the findings specifically imply that anticipation of loss may help explain how AIns function promotes abstinence from stimulant drug use.Clinically, addiction is defined not only by initial impulsive and chronic seeking of drugs but also by continued compulsive drug use despite adverse consequences (based on criteria for substance use disorder in the DSM-5) (49).Relapse not only may be associated with immediate positive affect from use or relief from withdrawal-related negative affect, but also implies severe longterm costs (e.g., jeopardizing relationships, losing a job).Individuals who have a blunted neural response to potential loss may be less likely to weigh long-term risks against immediate benefits of continued use, thus increasing risk for relapse.These results also confirm and extend the specificity of recent findings that blunted AIns responses to monetary incentives and during risky choice can predict relapse in stimulant users (17,22).
While recent neuroimaging research has focused primarily on differential neural responses to incentives as a function of stimulant use diagnosis, differences between users and control participants may or may not confer risk for relapse.Recent structural evidence suggests a double dissociation in which the coherence of different white matter tracts is associated with SUD diagnosis (i.e., connecting the ventral tegmental area with the NAcc) versus relapse (i.e., connecting the AIns to the NAcc) (24,50).This distinction supports theories that different neurobehavioral mechanisms confer vulnerability to developing a drug use diagnosis versus relapse to use after abstinence (17,25).The current findings provide additional convergent neurobehavioral evidence showing that measures that predict relapse may not necessarily differ between patients with SUD and control participants.
These findings raise the question of whether individual differences in loss anticipation precede drug taking or instead result from chronic use.While the current design cannot rule out either possibility, the results are more consistent with the notion that decreased motivation to avoid losses precedes rather than results from chronic drug use.Neither AIns responses during loss anticipation nor response times to loss trials differed significantly between diagnosed patients and control participants.These neurobehavioral measures also did not correlate with previous years of use, which might be expected if they resulted from previous chronic stimulant use.Disentangling which neural markers precede and confer susceptibility to addiction versus result from repeated use is an important goal for future research.
This work features several novel strengths.First, inclusion of a control group and the longitudinal design ensured that neurobehavioral predictors could be specifically linked to relapse to stimulant use versus diagnosis.Second, beyond specifically linking neural activity to relapse to stimulant drug use, convergent findings within and across multiple measurement modalities bolster confidence in the association of this circuit's function with clinical outcomes.Third, functional specificity to anticipation of losses but not gains or incentive outcomes may help inform more targeted relapse predictions and interventions in the future.Fourth, quantitative model comparisons indicated that AIns activity during loss anticipation and RT in response to loss cues can significantly improve predictions, adding value to more conventional predictors of relapse.Fifth, regression multiverse analyses ruled out alternative accounts related to potential confounds from other clinical variables and individual differences.Finally, and consistent with open science practices, predictions and analyses were preregistered, with freely available data and code (https://osf.io/9usfv).
This research also has limitations.First, patients were recruited from two different facilities, which varied with respect to treatment modalities and sex composition (one primarily included men and the other included only women).Despite these programmatic differences, findings did not obviously differ between the programs (Table S2).Second, comorbid conditions that were not assessed may interact with the neurobehavioral mechanisms underlying addiction in SUD, and relapse may interact with these comorbidities.While regression multiverse analyses ruled out several potential confounds, and use of psychotropic medications did not differ between those who relapsed and those who abstained, future studies will need to assess and control for other individual and clinical confounds.Third, the degree to which blunted neurobehavioral responses to anticipated loss generalize to predict relapse to other drugs of abuse remains to be tested.Fourth, although the longitudinal design supported tracking of relapse, the amount and chronicity of subsequent relapse were not tracked after initial relapse.Future research could target these finer-grained measures of relapse to better inform harm reduction approaches.Finally, although the longitudinal design was adequately powered to Blunted Loss Anticipation Predicts Stimulant Relapse test predicted associations, generalization to broad swathes of the population awaits future verification in larger and more diverse samples.
In summary, these convergent multimodal neuroimaging findings suggest that blunted AIns activity during loss anticipation predicts relapse to stimulant drug use.Theoretically, these results may improve the specificity and interpretability of biomarkers capable of predicting and preventing relapse to stimulant use (51).Clinically, these findings have implications for leveraging neurobehavioral measures that are capable of adding value to more conventional predictors of relapse.From a policy standpoint, the findings imply that individuals who are vulnerable to relapse may benefit from clear and contingent negative consequences for resuming drug use (e.g., as implied by the success of swift and certain programs for reducing drunk driving) (52).If individuals who are vulnerable to relapse could be identified prior to leaving treatment, perhaps they could be armed with more selective strategies and tools for anticipating and forestalling the recurrence of addiction.

Figure 1 .
Figure 1.Experimental design.(A) Diagram of the experimental design and timeline.Patients were scanned while undergoing treatment and then recontacted 1, 3, and 6 months after leaving treatment to inquire about drug use status.(B) Schematic of a monetary incentive delay task trial structure.(C) Cues and their associated hit/miss outcomes.FMRI, functional magnetic resonance imaging.

Figure 2 .
Figure 2. Neural activity in response to gain and loss anticipation and outcomes.Whole-brain comparison of patients who relapsed (n = 33) vs. those who abstained (n = 27) at the 6-month follow-up for the 4 contrasts of interest.Relative to patients who abstained, patients who relapsed showed a statistically significant decrease in preregistered anterior insula volume-of-interest activity for the loss anticipation contrast.Black circles highlight preregistered volumes of interest for each contrast of interest.All axial slices are at z = 2 Montreal Neurological Institute to provide a view of the anterior insula and medial prefrontal cortex.All coronal slices are at y = 29 Montreal Neurological Institute to provide a view of the nucleus accumbens and anterior insula.Maps are thresholded at p , .01,uncorrected for purposes of visualization.Coordinates of peak activations after cluster corrections are reported in TableS3.The same contrasts for control participants and patients at baseline are depicted in FigureS8.R, right.
(B) Activity time courses during 2$5 loss trials in the AIns.The plot shows averaged activity time courses relative to the start time of the trial.Highlighted areas indicate the period corresponding to the anticipation phase of the trial after accounting for the hemodynamic lag.Lines depict sample-level mean 6 standard error.(C) Anticipatory activity in the AIns.Dots indicate average AIns activity during the anticipation period of 2$5 loss trials for each subject.The significant difference between patients who relapsed and those who abstained was not solely driven by the 2 extreme points in the relapse group because the effect remained significant even after these values were omitted (t 56 = 22.78, p = .007,Cohen's d = 0.73).(D) Coefficients 6 standard error plotted for independent logistic regression models for AIns activity during anticipation of each of the 6 trial types as the predictor variable and relapse at 6 months as the binary outcome variable.(E) Lower AIns anticipatory activity predicts shorter time to relapse.Dots indicate average AIns activity during anticipation of the 2$5 loss trials for each subject.The line indicates the fitted model estimate based on restricted mean survival (RMS) analysis.(F) Multiverse regression model showing the color representation of t statistics for logistic regression model coefficients with relapse at 6 months as the binary outcome variable and all possible combinations of regressors.Each column depicts coefficients from an independently estimated regression model.The top row depicts the predictor of interest, which is included in all iterations of the model (outlined in red).*p , .05,Bonferroni corrected.

Figure 4 .
Figure 4. Integrating behavioral, functional, and structural predictors of relapse.(A) Patients who relapsed had slower reaction times (RTs) during 2$5 loss trials compared with patients who abstained.Dots depict RT of each participant averaged over 15 trials of each trial type.(B) Coefficients 6 standard errors plotted for independent logistic regression models with RT of each of the 6 trial types as the predictor variable and relapse at 6 months as the binary outcome variable.(C) RTs on large loss trials were associated with anterior insula (AIns) anticipatory activity on those trials.(D) Coefficients 6 standard errors plotted for independent linear regression models between RT and anticipatory AIns activity on each of the 6 trial types.(E) Right AIns activity during anticipation of large losses was also associated with white matter tract right AInsnucleus accumbens (NAcc) tract coherence indexed by fractional anisotropy (FA).(F) Coefficients 6 standard error plotted for independent linear regression models between anticipatory right AIns activity on each of the 6 trial types and right AIns-NAcc tract coherence indexed by FA. (G) Correlated activity of the right AIns and right NAcc specifically during large loss trials was associated with right AIns-NAcc tract coherence indexed by FA. (H) Coefficients 6 standard error plotted for independent linear regression models between right AIns-NAcc-correlated activity during each trial type and right AIns-NAcc tract FA.Red bars in panels (B), (D), (F), and (H) represent predicted effects and highlight the specificity of the associations.*p , .05,**p , .01,† p , .1.