Impact of acute stress on neural indices of positive and negative reinforcement processing in cannabis users

Abstract

Objective:

Chronic cannabis use is maintained in part through dysregulated stress and reward response systems. Specifically, stress-related negative affect is thought to act as a salient motivator for chronic substance use. Models of addiction posit that the transition from positive to negative reinforcement motives for substance use is a key mechanism of disordered use. However, research in substance-using samples has not assessed stress-related neural processing of both positive and negative reinforcement.

Method:

Therefore, the current study utilized laboratory stress induction to examine how stress affects the reward positivity, an event-related potential sensitive to both positive (RewP) and negative (relief-RewP) reinforcement, in 87 cannabis users (58.10% Female, Mage = 19.40) varying in Cannabis Use Disorder (CUD) severity and, as part of larger study aims, history of traumatic brain injury (TBI). We predicted greater CUD severity would be associated with a blunted RewP and enhanced relief-RewP, particularly after stress induction, independent of TBI status.

Results:

Findings indicated that CUD severity was not associated with RewP/relief-RewP amplitude regardless of acute stress. Exploratory analyses revealed however, that among those with history of traumatic brain injury (TBI+), CUD severity was associated with greater stress-elicited blunting of the RewP and enhancement of the relief-RewP.

Conclusion:

Although initial findings contradict current allostatic models of addiction, exploratory findings suggest that history of traumatic brain injury, and potentially other confounding variables related to increased risk of TBI-experience, may influence the extent to which stressful experiences modulate the neurophysiology of both positive and negative reinforcement reward processing in CUD.

Introduction

Cannabis use is increasing in the United States, in part due to legalization efforts and greater cultural acceptance (Carliner et al., 2017). Considering daily cannabis use has increased twofold since 2002 (Compton et al., 2019), it is expected that disordered use will also increase (Budney et al., 2019; Hasin, 2017). Cannabis use disorder (CUD) is chronic, highly distressing, and poses an interpersonal burden (Hasin, 2016). Unfortunately, there are no FDA-approved medications for CUD and extant psychotherapeutic approaches have modest efficacy (e.g., abstinence rates ~20%; Gates et al., 2016), creating an urgent and growing unmet public health need. Trends toward increased legislation allowing for cannabis use is likely to continue the upward trajectory of daily-use prevalence rates. Therefore, it is critical to better understand the neural mechanisms of CUD to improve identification of novel intervention targets.

As with other substance use disorders (Baker et al., 2004; Koob & Schulkin, 2019), negative reinforcement processes appear central to the onset and maintenance of CUD (Zehra et al., 2018). For example, greater CUD severity is associated with frequent use of cannabis to cope with stress-related negative affect (Farris et al., 2016; for a review, see Hyman & Sinha, 2009). One longitudinal study of regular cannabis users found that self-reported coping motives predicted greater incidence and persistence of cannabis dependence over a three-year period (van der Pol et al., 2013; van der Pol et al., 2015), suggesting using marijuana to cope with, or reduce, negative affect may influence disordered use trajectory. Moreover, CUD severity is associated with decreased neural activation to non-drug rewards, such as less dopamine reactivity in the ventral striatum (Volkow et al., 2014) and blunted activation of the nucleus accumbens (Martz et al., 2016; Spechler et al., 2020), both of which are central nodes of the reward processing circuit. Most recently, Crane and colleagues (2021) found that occasional cannabis users demonstrated enhanced neurophysiological reactivity to monetary gains compared to non-users, but this effect was not found for users with a CUD history, suggesting that enhanced non-drug reward processing may characterize recreational rather than disordered users. In support of this suggestion, blunted reactivity to monetary gains in the time-frequency domain was associated with greater prior CUD severity, consistent with theory and prior findings. Taken together, both self-report and neural evidence suggest that onset and maintenance of cannabis dependence is associated with cannabis use to diminish stress-related negative affect, as well as decreased reactivity to non-drug rewards.

In contrast to the neurobiology of non-drug positive reinforcement (e.g., monetary wins on a guessing task), negative reinforcement processing has received less attention, a significant limitation given its central role in seminal theories of addiction (Baker, 2004; Koob & LeMoal, 1997). The reward positivity (RewP) is one promising neurophysiological marker of reward processing that may have utility for evaluating both negative and positive reinforcement processing. The RewP is characterized by a fronto-central positive deflection that occurs approximately 250 ms after reward feedback presentation (Holroyd et al., 2008). Relative to the feedback related negativity (FRN; Hajcak et al., 2006; Holroyd et al., 2006), which is part of the response system to punishment/absence of reward, the RewP represents activation of neural circuits responsible for processing reward feedback (for a review, see Proudfit, 2015). Importantly, the RewP is sensitive to the relative value of feedback after a response, and not simply whether the response was correct or incorrect (Gehring & Willoughby, 2002; Nieuwenhuis et al., 2004) or the salience of the feedback (Mulligan & Hajcak, 2018). As recently suggested by Albanese and Hajcak (2021), the RewP can therefore theoretically be examined following positive reinforcement (i.e., receiving money) or negative reinforcement (i.e., the absence of a punishment), which is ideal for studying altered reward processing to distress-related relief and non-drug rewards. In fact, there is already evidence that the RewP can be extracted from a modified doors task in which the ‘losing’ option is an aversive shock and the ‘winning’ option is the absence of shock (Mulligan & Hajcak, 2018), providing initial support for the use of the RewP to evaluate negative reinforcement feedback processing.

Although, to our knowledge, the RewP to negative reinforcement has not been evaluated in substance-using samples, the impact of acute stress on drug cue reactivity, reflecting a negative reinforcement process, has been assessed in several addicted populations. Stress-related increases in drug cue reactivity across brain regions responsible for craving and incentive value (e.g., anterior cingulate cortex, medial prefrontal cortex) have been found in cocaine and nicotine dependent samples (Dagher et al., 2009; Duncan et al., 2007). Specific to cannabis use, Macatee et al. (2019) found that cannabis users reporting poor distress tolerance demonstrated an enhanced neurophysiological response to cannabis cues after the stress induction, which correlated with multiple indicators of CUD severity and negative reinforcement-related craving. With respect to non-drug reward processing, the impact of acute stress on reward-related neural circuits, including the RewP, has been evaluated in healthy samples. For instance, a recent study of 26 adults found an attenuated RewP to monetary rewards during an acute stressor relative to a control condition (Burani et al., 2021). Further, Ethridge and colleagues (2020) found a blunting of monetary gain-related activity in the delta frequency power domain. Taken together, these studies suggest that acute stress can differentially impact reward processing, blunting responding to positive reinforcers and increasing reactivity to negative reinforcement cues (e.g., drug cues in addictive populations).

To summarize, behavioral and neural data are supportive of negative reinforcement models of CUD, highlighting coping-motivated cannabis use/cannabis cue reactivity during acute stress as well as blunted reactivity to non-drug rewards in dependent vs. non-dependent cannabis users. However, this work is limited in that (1) the impact of acute stress on positive reinforcer responding has never been tested in a cannabis-using sample, and (2) neural markers of negative reinforcement have only been assessed indirectly via cannabis cue reactivity during acute stress. The RewP is a neurophysiological measure of domain-general reward responsivity (i.e., across reinforcement types) that is sensitive to acute stress, making it an ideal neurophysiological measure for probing expected alterations in reward processing in CUD. Although acute stress modulation of the RewP and its domain-general nature has been evaluated in healthy samples, no research to date has compared the impact of acute stress on the RewP to both positive and negative reinforcement in a substance-using population. Examination of neurophysiological reactivity to attainment of non-drug rewards as well as avoidance of aversive stimuli, both before and after induction of acute stress, will allow for a theory-driven test of the differential alterations in both types of reward processing thought to occur in CUD.

To address current gaps in the literature, the current study examined differences in reward processing to positive and negative reinforcement before and after a stress induction in regular cannabis users. Cannabis users varying in CUD severity were recruited to complete positive reinforcement (i.e., winning vs. losing money) and negative reinforcement (i.e., avoiding vs. receiving an aversive noise blast) versions of a guessing task both before and after a laboratory stressor while electroencephalography (EEG) was recorded. Notably, as part of the larger study’s aim examining traumatic brain injury (TBI)-status on neurophysiological responding in regular cannabis users, a large portion of the current sample reported experiencing a TBI of at least mild severity. Moderate to severe TBI status is associated with greater motivation to use cannabis (Hawley et al., 2018), in addition to aberrant neurophysiological outcomes (Amyot et al., 2015). Considering a majority of the sample experienced a mild TBI, current hypotheses focused on negative reinforcement and allostatic models of addiction independent of TBI status.

Consistent with negative reinforcement and allostasis models of addiction (Baker et al., 2004; Koob & Le Moal, 2001), we hypothesized that 1) individuals with greater, relative to lower, CUD severity would exhibit an enhanced RewP to safety-related feedback, 2) greater CUD severity would be associated with a blunted RewP to natural rewards (i.e., money), and 3) the aforementioned associations between CUD severity and reward processing would be amplified after a stress induction.

Materials and Methods

Participants

English-speaking participants (18 – 30 years old) who endorsed current, regular cannabis use (i.e., >= three days per week for at least the past three months) were recruited from a large southeastern university and nearby community. As part of a larger study examining the effects of traumatic brain injury on neural correlates related to cannabis use, additional inclusion criteria required half the sample previously experience at least one mild, closed head repeated traumatic brain injuries (TBI+) and a control group consisting of individuals with no history of TBI (TBI−). Exclusion criteria were assessed with an online survey completed during the lab appointment, in addition to an in-person screening interview done by a trained research assistant and reviewed by a clinical psychology graduate student. Exclusion criteria consisted of experiencing a TBI within the past year, alcohol use >= three times per week for at least the past three months, and individuals who met criteria for Alcohol Use Disorder. Finally, participants who endorsed bipolar or psychotic-spectrum diagnoses, imminent risk for suicide, or participation in psychotherapy in the past three months were excluded.

Sample size was determined based on a power analysis conducted as part of the larger study’s primary hypothesis concerning expected TBI group differences in CUD severity. Previous studies utilizing samples of individuals with traumatic brain injuries (Dimoska-Di Marco et al., 2011) and cannabis problems (Simons & Carey, 2002) have found medium effect sizes. Power analysis suggested recruiting 71 participants to detect necessary effects at a power level > .90 and a Type 1 error rate < .05. Eighty-seven participants (M_age = 19.40, SD_age = 1.37) total were recruited and completed the baseline appointment. A majority of participants identified as female (57.5%, n = 50) and White (80.5%, n = 70), followed by Black/African American (10.3%, n = 9), Asian (4.6%, n = 4), Native Hawaiian or Other Pacific Islander (1.1%, n = 1), and Other (e.g., biracial; 3.4%, n = 3). Participants reported using the following substances at least one time per week in the last three months: Cocaine (3.5%), sedatives (1.2%), amphetamines (2.4%), methamphetamine (1.2%), hallucinogens (1.2%). Notably, frequent alcohol and nicotine use were endorsed in this sample, with 40.2% (n = 35) of participants reporting at least weekly alcohol use and 10.5% (n = 9) reporting current cigarette use. Participants reported, on average, four symptoms from the CUD section of the Structured Clinical Interview for the DSM-5 (see Measures section below), consistent with a diagnosis of moderate CUD (M_{CUD Symptoms} = 4.48, SD_{CUD Symptoms} = 2.23). Finally, participants reported experiencing their TBI around four years prior to beginning the current study (M = 4.44, SD = 3.32). TBI experiences ranged from head injuries from playing sports (e.g., football) to physical altercations.

Measures

Cannabis Use Disorder.

Participants were screened for past year CUD via the cannabis use section of the Structured Clinical Interview for the DSM-5-Research Version (SCID-5-RV; First et al., 2015). Interviews were administered by either a graduate level clinician or trained postbaccalaureate research assistant. All research staff who administered the SCID-5-RV completed extensive training on the SCID, including multiple practice sessions, watching videos demonstrating SCID administration, and reviewing SCID diagnoses on a weekly basis with a licensed clinical psychologist. Total number of CUD criteria was used as a measure of CUD severity.

Timeline Follow-back (TLFB).

The TLFB (Sobell & Sobell, 2012) was used to determine the number of cannabis use days in the past month in the current sample.

Marijuana Smoking History Questionnaire (MSHQ).

The MSHQ (Bonn-Miller & Zvolensky, 2009) is a 21-item measure that examines typical method of marijuana use, lifetime quit attempts, smoking frequency in the past 30 days, as well as age at first use and age at regular use.

Depression Anxiety Stress Scales (DASS).

The DASS (Lovibond & Lovibond, 1995) is a 21-item measure that assesses past-week anxiety, depressive, and stress-related symptoms. Previous research has verified the reliability of the DASS in clinical (Brown et al., 1997) and non-clinical (Lovibond & Lovibond, 1995) samples. In the current study, the DASS demonstrated good internal consistency for the anxiety (α = .78) and depression (α = .88) symptom subscales.

Procedure

Eligible participants who scheduled an appointment were asked to refrain from using marijuana and other illicit substances (e.g., alcohol) at least 24 hours and caffeine/nicotine use at least one hour prior to their appointment date. Upon arrival, participants were asked whether they had abstained from marijuana or any other illegal or illicit substances for 24 hours prior to the appointment time. All participants who reported marijuana or illicit substance use within 24 hours were rescheduled. Next, participants were administered the CUD section of the SCID-5-RV. A battery of other clinician-assessed and self-report measures were administered as part of a larger study. Following self-report measures, participants completed several tasks while EEG was recorded. EEG recording was conducted in a dimly lit, sound absorbing room and stimulus presentation was administered on a 21” CRT color monitor with a Dell OptiPlex 780 computer running E-Prime version 2.0.8.90. Participants viewed stimuli approximately 100 cm from the monitor, subtending a visual angle of 3.5°, and completed a baseline resting task, the doors task before and after stress induction, and finally a post-stressor resting task (see Figure 1). Recording time was between 1.5 and 2 hours. All procedures were approved by the university’s Institutional Review Board.

Figure 1.

Procedural figure demonstrating laboratory task flow. Reward ERP measurement was taken during the pre- and post-stress doors task, while subjective negative affect (NA) measurement was taken during the MMST.

Resting and Post-Stress Resting Task –

Participants completed a resting task prior to and immediately after the lab-induced stressor (see below) while electrocardiograph data were collected. Participants were instructed to stare at a white fixation point presented on the screen and to not look away for two minutes, after which they were shown instructions to close their eyes for two minutes. The resting tasks allows for the examination of heart rate change from baseline through stress-onset, as well as a sufficient period post-stress to examine sustained physiological response to the stressor.

Doors Task –

The doors task is often used to examine differences in neural processing of reward vs. punishment feedback (Proudfit, 2015). Participants were presented with two identical doors on the screen and given directions to select either the left or right door by clicking the left or the right mouse button, respectively. After selecting a door, a fixation cross appeared for 1000 ms followed by a 2000 ms feedback stimulus consisting of either a green up arrow (indicating reward) or a red down arrow (indicating punishment). The feedback stimulus was followed by another 1000 ms fixation cross, after which the participant was instructed to click any mouse button to go to the next trial. The current study consisted of 64 trials divided equally by feedback type (reward vs. punishment) to examine differences in neural response to positive and negative reinforcement. Similar to the tasks used by Mulligan and Hajcak (2018), both negative reinforcement and positive reinforcement conditions consisted of 32 trials each, divided equally into two 16 trial blocks. In each block during the negative reinforcement condition, there were 8 trials in which the red down arrow (i.e., punishment) indicated impending receipt of an aversive (i.e., 50 ms, 105 dB) white noise blast, and 8 trials in which the green up arrow (i.e., reward) indicated safety from the noise blast, yielding a total of 16 punishment and 16 safety trials across the two negative reinforcement blocks. In each block of the positive reinforcement condition, there were 8 trials where red down arrows represented losing $0.15 and 8 trials where the green up arrows represented gaining $0.30, yielding a total of 16 gain trials and 16 loss trials across the two positive reinforcement blocks. In the current study, it was equiprobable that participants would receive either the red or green arrow within each block. Block type (i.e., positive or negative reinforcement condition) order was counterbalanced, with half the participants receiving an ABAB order and the other half receiving BABA order.

Mannheim Multicomponent Stress Test (MMST) –

Stress induction was achieved via the MMST, which is a five-minute computerized task that simultaneously incorporates cognitive (i.e., difficult and quick mental arithmetic), emotional (i.e., presentation of aversive images), acoustic (i.e., increasing volume of white noise), and motivational (i.e., losing money for each mental arithmetic error) stressors. Specifically, the MMST utilizes the Paced Auditory Serial Addition Task (PASAT-C, Lejuez et al., 2003), which instructs participants to view numbers between 0 and 20 that appear on the screen. The task requires participants to sum the number quickly and repeatedly on the screen with the preceding number, using the mouse to select their answer on a keyboard presented below the numbers. Concurrently, participants hear white noise increasing in volume, see aversive pictures in the background of the screen, lose money for each answer they get wrong, as well as hear a loud explosion for each wrong response. The MMST has demonstrated significant subjective and physiological stress reactivity in multiple samples (Macatee et al., 2019; Reinhardt et al., 2012). Importantly, the task starts with 65 seconds of just the white noise and aversive pictures which has intermittent duplicates that the participant is required to click on when they appear. This forces participants to attend to the images at the beginning of the task, thus making it more difficult to ignore them during the subsequent 4 minutes PASAT portion of the task. In total, the MMST lasts 305 seconds – 65 seconds for initial picture recognition, then 4 minutes of the PASAT for a total duration of 305 seconds. In the current study, subjective stress reactivity was quantified by averaging five negative affect words (anxiety, frustration, irritability, difficulty concentrating, physical discomfort) that were measured on a 0-100 visual analogue scale (α = .78 - .81) both before and after the MMST.

Task Delivery and Psychophysiology Measurement –

A Dell OptiPlex 780 computer and Neuroscan Acquire software were used to collect all EEG data. ERP responses were measured via two 64-channel Neuroscan SynAmps RT amplifiers connected to a 64-channel Brain Vision actiCap (sampling rate = 1000 Hz, online analog bandpass filter = 0.05 – 100 Hz). A midline ground (AFz) and online reference electrode (FCz) were used during EEG recording and were re-referenced to the average of the mastoid sensors (TP9 and TP10) offline. Electrodes placed approximately 1 cm above and below the right eye, as well as on the outer canthus of each eye, measured vertical and horizontal electrooculogram (EOG) activity, respectively. EKG electrodes to measure heart rate activity were placed on the inside of the left and right upper forearms, adjacent to any prominent veins to reduce noise. High-chloride (10%) Abrasive Electrolyte-Gel was used to fill each electrode and EEG recording was done while impedance values were below 10 kohms.

Data Preprocessing

First, data were downsampled to 250 Hz, after which high-pass (0.1 Hz; ripple = .05 dB, attenuation = 80 dB) and low-pass (30 Hz; ripple = .01 dB, attenuation = 40 dB) FIR filters were applied. Data were then re-referenced to the average mastoids, and the implicit online reference (FCz) was regenerated. Next, epochs were created from −200ms to 800ms surrounding feedback stimulus onset. Artifact detection and rejection was done with the EEGLAB plugin Fully Automated Statistical Thresholding for EEG artifact Rejection algorithm (FASTER; Nolan et al., 2010), and the resulting cleaned epochs were then baseline-corrected (see Supplemental material for more detail on FASTER processing). Invalid participant data at pre- and/or post-stress were discarded due to technical errors (n=1), poor signal quality (n=1), or poor mastoid signals (n=5).

Epochs were averaged separately based on feedback type (i.e., reward vs. punishment) nested within reinforcement type (i.e., negative vs. positive reinforcement) and stress context (i.e., pre- vs. post-stressor). See Figure 2 and 3 for waveforms representing responses to reward and punishment for both reinforcement conditions at pre- and post-stressor, as well as topographical difference maps.

Figure 2.

Stimulus-locked PCA-derived TF3SF3 factors (upper) and raw waveforms (lower) at FCz for pre-stress (left) and post-stress (right) during the win-lose money condition of the Doors Task. Topographical maps for TF3SF3 waveforms show the factor amplitudes at each electrode at the peak time point (300ms). Topographical maps for raw waveforms show average scalp activity between 200 and 348 ms with color bar scale in microvolts.

Figure 3.

Stimulus-locked PCA-derived TF3SF3 factors (upper) and raw waveforms (lower) at FCz for pre-stress (left) and post-stress (right) during the safety-noise condition of the Doors Task. Topographical maps for TF3SF3 waveforms show the factor amplitudes at each electrode at the peak time point (296ms). Topographical maps for raw waveforms show average scalp activity between 200 and 348 ms, with color bar scale in microvolts.

EKG data were first exported as text files and analyzed by Kubios Premium (Tarvainen et al., 2014) using automatic beat correction. Kubios Premium allows for time-domain (e.g., heart rate) analysis by identifying the peak amplitude for each heartbeat (i.e., the R wave peak) within a specified time period. Two research assistants verified all R peak locations, as well as manually marked data as noise in which an R peak could not be identified. Extraneous R peaks or R peaks that were located in the wrong location were deleted or moved to a more accurate location, respectively. Participant data with 75% noise were removed from analysis (n = 5). Kubios Premium was used to analyze three timepoints: Baseline resting heart rate, heart rate during the MMST, as well as post-MMST heart rate.

Temporospatial PCA

To isolate the RewP component from overlapping ERPs (e.g., P3), a temporospatial PCA was conducted using the ERP PCA toolkit, version 2.90 (Dien, 2010). PCAs were conducted separately on positive reinforcement and negative reinforcement conditions, with four trial types entered in each PCA (i.e., pre-stress reward, pre-stress punish, post-stress reward, post-stress punish). First, a temporal PCA was run to isolate distinctive temporal patterns of electrocortical activity occurring from −200ms to 800ms. The timepoints from each participant’s average ERPs were used as variables, with participants, trial types, and electrode sites used as observations. Promax rotation was utilized. For the positive reinforcement PCA, 7 temporal factors were extracted after examination of the Scree plot (Cattell, 1966). For the negative reinforcement PCA, 8 temporal factors were extracted. For each temporal factor, factor scores were computed for every combination of participant, trial type, and electrode site, reflecting the variance in the raw data captured by that temporal factor. A subsequent spatial PCA was conducted for each of the temporal factors to dissociate components with comparable time courses but distinct spatial distributions. For each spatial PCA, electrode sites were used as the variables, with participants, trial types, and temporal factor scores used as observations. Infomax rotation was utilized. For the positive reinforcement PCA, 3 spatial factors were extracted for each of the 7 temporal factors, ultimately producing 21 temporospatial factors. For the negative reinforcement PCA, 3 spatial factors were extracted for each of the 8 temporal factors, ultimately producing 24 temporospatial factors. The covariance matrix and Kaiser normalization were used for each PCA. Of the temporospatial factors, 11 and 9 factors explained at least 1% of the total variance in the positive reinforcement and negative reinforcement data, respectively. These factors were converted into microvolt-scaled waveforms reflecting the portion of the raw data accounted for by that factor by multiplying the factor loadings, scores, and their standard deviations (Dien et al., 2003). Factor waveforms were then inspected for resemblance to the RewP component expected in the Doors task.

In both negative and positive reinforcement PCAs, a frontal-central positivity (TF3SF3) functionally (reward > punishment) and topographical (maximal at frontal-central sites) and temporal (i.e., peaked ~300ms) characteristics consistent with the RewP component emerged (see Figure 2 for factor waveforms). RewP factor amplitudes in each reinforcement condition were generated for every participant using the peak values for each trial type (i.e., pre-stress reward, pre-stress punish, post-stress reward, post-stress punish). To maintain clarity when differentiating between PCA-related reward components in the negative and positive reinforcement conditions, we designated the RewP factor as the reward component of the positive reinforcement condition and relief-RewP factor as the reward component of the negative reinforcement condition.

Data Analytic Plan

Given the current study’s nested data structure and presence of missing data (see Data Preprocessing section), a linear multi-level model with all fixed effects was conducted to test hypotheses. Stress (pre-MMST vs. post-MMST), Condition (Negative reinforcement vs. Positive reinforcement), and Feedback (Reward vs. Non-Reward) were entered as within-subject factors. CUD severity was entered as a continuous between-subjects variable and TBI history (absent vs. present) was entered as a categorical between-subjects factor. All main and interaction effects were entered. To test the prediction that greater severity of CUD would predict an enhanced and blunted Δrelief-RewP (No noise blast – noise blast)/ΔRewP (win money – lose money), respectively (hypotheses 1&2), the CUD*Condition*Feedback interaction was examined. To test the third hypothesis that a laboratory stressor would amplify predicted changes in negative and positive reinforcement processing, the CUD*Stress*Condition*Feedback interaction was examined. Finally, CUD*TBI interactions were examined to test if hypothesized CUD effects varied by TBI group. Significant interactions were probed with follow-up testing of effects at increasing levels of CUD severity.

Transparency and Openness

We report how we determined our sample size, all data exclusions, all manipulations, and we follow journal article reporting standards (Kazak, 2018). All data, analysis code, and research materials are available upon request. Data were analyzed using IBM’s SPSS Version 26 (Hayes et al., 2012). This study’s design and its analysis were not pre-registered.

Results

Descriptives

Refer to Table 1 for all sample descriptives. T-tests comparing CUD symptoms, t(85) = 0.34, p = .74, self-reported stress reactivity, t(52.48) = −0.87, p = .39, and sustained cardiovascular reactivity to the MMST, t(83) = −0.95, p = .35, in TBI+ vs. TBI− subsamples were all non-significant.

Table 1.

Sample descriptive statistics (N = 87)

		Mean	SD
Demographics
Age		19.4	1.368
Biological Sex
	Male	36	41.90%
	Female	50	58.10%
Sexual Orientation
	Heterosexual	72	82.80%
	Homosexual	4	4.60%
	Bisexual	9	10.30%
	Other	2	2.30%
Race
	White	70	80.50%
	Black	9	10.30%
	Asian	4	4.60%
		1	1.10%
	Native Hawaiian or Other Pacific Islander
	Other	3	3.40%
Ethnicity
	Non-Hispanic	59	67.80%
	Hispanic or Latino	28	32.20%
Level of Education
	Some high school	2	2.30%
	High school diploma	36	41.40%
	Some college	47	54.00%
	Bachelor’s degree	2	2.30%
TBI Experience
Age at TBI		15.38	2.75
No TBI		56	64.37%
Mild TBI		27	31.03%
Moderate TBI		7	8.75%
Cannabis Use
# Days used in past month		23.72	6.8
Age at first use		15.45	1.61
Age at onset of regular use		17.31	1.58
Years of regular use		2.06	1.36
CUD Diagnosis Met		90.6%	--
CUD criteria (past 12 months)		4.48	2.23
Mild CUD		25	28.7%
Moderate CUD		31	35.6%
Severe CUD		23	26.3%
Lab Stressor
	Subjective Reactivity
	Pre-MMST affect	24.51	17.21
	Post-MMST affect	58.3	22.47
	Subjective Reactivity	33.78	20.09
	Cardiovascular Reactivity
	Pre-MMST heart rate (bpm)	67.21	8.31
	MMST heart rate (bpm)	71.68	12.86
	Post-MMST heart rate (bpm)	69.2	10.54
	Sustained heart rate to stressor	2.27	6.07

Manipulation Checks

Stress Induction.

As expected, subjective negative affect increased significantly from pre- to post-MMST, F(1,85)=243.27, p<.001, η_p²=.74, indicating that the lab stressor successfully elicited subjective negative affect. Similarly, mean HR significantly varied from the resting baseline through the MMST and post-MMST period, F(2,166.82)=18.11, p<.001. Bonferroni-corrected pairwise comparisons revealed that mean HR was significantly higher during the MMST compared to the resting baseline, t(166.86)=6.02, p<.001, as well as the post-MMST period, t(166.86)=3.13, p=.006, indicating that the lab stressor successfully increased cardiovascular activity. Further, mean HR during the post-MMST period was significantly higher than the resting baseline, t(166.86)=2.86, p=.015, indicating ongoing, stressor-related residual cardiovascular activation during the post-stressor period.

Effect of Reward vs. Punishment Feedback Before and After Acute Stress.

To evaluate the overall effects of feedback, reinforcement condition, and acute stress, a model with only within-subject effects was examined. A significant main effect of feedback, F(1,648)=75.95, p<.001, revealed the expected amplitude increase to reward (M=4.52, SE=0.46) vs. non-reward (M=2.04, SE=0.46) trials, and the non-significant condition*feedback interaction, F(1,648)=1.52, p=.22, showed that this reward effect did not differ by negative vs. positive reinforcement. The stress*feedback interaction, F(1,648)=5.32, p=.021, was significant, which revealed an overall decrease in the reward effect from pre, Mdiff=2.94, SE=0.34, t(648)=8.68, p<.001, to post-stressor, Mdiff=2.01, SE=0.36, t(648)=5.61, p<.001. The non-significant stress*condition*feedback interaction, F(1,648)=2.33, p=.13, suggests that the stress-induced blunting of the reward effect did not differ by reinforcement condition.

Acute Stress Modulation of Reward Processing in CUD

The hypothesized CUD*condition*feedback, F(1,624)=0.01, p=.94, and CUD*stress*condition*feedback, F(1,624)=0.24, p=.63, interactions were non-significant. No other interactive effects between CUD and within-subject factors were significant, ps>.10. Likewise, TBI interactions with all within-subject factors were non-significant, ps>.22. However, a significant CUD*TBI*stress*condition*feedback interaction was found, F(1,624)=4.19, p=.041.¹ To examine this interaction, the CUD*stress*condition*feedback interaction was examined in the TBI− and TBI+ groups (see estimated marginal means in Figure 4). Plotted EMMs suggested that, in the TBI+ group, CUD severity was associated with greater stress-related blunting of the RewP and enhancement of the relief-RewP, in line with hypotheses. Consistent with this, the CUD*stress*condition*feedback interaction was significant, F(1,236)=4.26, p=.04, in the TBI+ group. In contrast, the TBI− group EMMs were only partially consistent with hypotheses, showing positive CUD severity associations with stress-elicited enhancement of the relief-RewP and RewP. Consistent with this, the CUD*stress*condition*feedback interaction was non-significant, F(1,388)=0.24, p=.63, in the TBI− group. However, despite an apparent trend of stress-related enhancement of overall RewP amplitudes with increasing CUD severity, the CUD*stress*feedback interaction was also non-significant, F(1,388)=2.60, p=.11, in the TBI− group.

Figure 4.

ΔReward-No Reward amplitude for both positive (Panels A and C) and negative (Panels B and D) reinforcement condition compared between pre- and post-stress at Mild and Severe Cannabis Use Disorder (CUD) diagnosis. Panels A and B show the Δreward-no reward amplitude at Mild and Severe CUD for individuals with previous history of TBI (TBI+), while Panels C and D show this interaction for individuals without TBI history (TBI−). Error bars indicate 95% confidence intervals.

A series of follow-up analyses were conducted to test for the robustness of the CUD*TBI effect. To evaluate the impact of psychiatric medication use, the 10 participants (n=5 in each TBI group) currently taking psychiatric medication were removed from the sample and models were re-run. The CUD*TBI*stress*condition*feedback interaction remained significant, F(1,548)=6.50, p=.011 (see Figure S1 in the Supplemental Material). Similarly, the CUD*stress*condition*feedback interaction remained significant in the TBI+ group, F(1,200)=4.97, p=.027, and non-significant in the TBI− group, F(1,348)=1.53, p=.22. Further, the CUD*stress*feedback interaction became marginal but remained non-significant in the TBI− group, F(1,348)=3.57, p=.06. To test for possible sex differences, Sex, Sex*CUD, and Sex*TBI between-subjects terms, along with their interactions with all within-subject factors, were added to the model. In this model, the CUD*TBI*stress*condition*feedback interaction remained significant regardless of whether participants currently taking psychiatric medication were retained in the sample. All lower- and higher-order Sex*feedback interaction terms were non-significant, ps>.10. Finally, to determine if results were robust to the influence of current anxiety/depressive symptoms, DASS-anxiety and DASS-depression scores were entered as between-subjects covariates in separate models, along with their interactions with all within-subject factors. In these models, the CUD*TBI*stress*condition*feedback interaction again remained significant independent of DASS-anxiety and DASS-depression scores.

Discussion

Overall, results were partially consistent with negative reinforcement models of addiction. Specifically, these models (e.g., Baker, 2004; Koob & LeMoal, 1997) predict that drug use becomes increasingly maintained via negative rather than positive reinforcement as disordered use develops. Further, addiction is thought to be characterized by amplified reward responsivity to cues predictive of relief (e.g., drug cues) and diminished to non-drug related rewards, particularly in the context of heightened stress (Adinoff, 2004). Similar to Burani et al.’s (2021) finding of blunted RewP amplitude from pre- to post-stress in a sample of healthy adults, current results showed an overall blunting of RewP amplitude from pre- to post-stress regardless of reinforcement condition. Contrary to prominent theories of addiction, we did not find a relationship between CUD severity and blunting of positive reinforcement or potentiation of negative reinforcement processing regardless of acute stress.

The absence of main effects of CUD severity on reward processing was unexpected. According to negative reinforcement models, fluctuations in neural activation to drug or non-drug reward stimuli are most apparent during periods of deprivation and/or acute stress (Koob & LeMoal, 2008). Abstinence after chronic substance use is related to biological changes reflective of an increased withdrawal state; in cannabis users, this state has been shown to peak around two to six days after abstinence (Budney et al., 2003). Participants in the current study were only required to abstain from cannabis for 24 hours before the lab appointment, which may not have been sufficient enough time for a deprivation state to develop. Indeed, research examining withdrawal trajectory and severity in disordered alcohol (Jasinka et al., 2014), cocaine (Denomme & Shane, 2020) and heroin (Lou et al., 2012) users consistently finds increased drug-related cue activity in regions of the brain associated with reward and craving, relative to non-drug cues. Future research may find the expected blunting and amplification of positive and negative reinforcement processing, respectively, in a sample of regular cannabis users who are asked to abstain from marijuana for a longer period of time (e.g., one week).

Null findings in the current study contrast with allostasis models of addiction. Specifically, disordered substance use is thought to alter neural reward-related pathways, such that substance-related cues produce potentiated responses and non-drug related rewards (e.g., natural rewards) are blunted in the addicted state, particularly during withdrawal or acute stress (Koob & Volkow, 2010). Thus, we hypothesized that disordered cannabis users would be more sensitive to cues indicative of relief from states of high negative affect, while simultaneously being at a higher threshold of sensitivity for naturally occurring rewards. In the current study, however, we did not observe the expected difference in reward responsivity between negative and positive reinforcement as a function of cannabis severity either before or after acute stress administration. In light of these findings, exploratory analyses examining the interactive nature of TBI status on this relationship revealed how TBI history may partially explain these null effects.

Importantly, the experience of traumatic brain injury (TBI) has been shown to be both an outcome and predictor of disordered substance use (West, 2011; Graham & Cardon, 2008). Recent studies have shown TBI status appears to sensitize reward processing to cocaine in rodents (Cannella et al., 2020), as well as alter neural systems related to reward in people who experienced a TBI early in life, which may contribute to disordered substance use (Cannella et al., 2019). Specific to cannabis, TBI status may influence motivations to use cannabis; one study found that individuals with at least moderate to severe TBI reported using cannabis after their TBI, in part, as a way to cope with ongoing stress or anxiety (Hawley et al., 2018). Despite evidence that cannabis use may decrease overall mortality (Nguyen et al., 2014) or have beneficial cognitive effects in TBI+ individuals (Hergert et al., 2021), TBI status may influence reward processing in disordered cannabis users. Considering exploratory analyses examining TBI+ individuals supported hypotheses, the role of TBI status may influence reward responsivity in disordered cannabis users. Evidence is mixed regarding the effect of TBI on ERP status (for a review, see Gaetz & Bernstein, 2001), however there is evidence that mild TBI status on neurophysiological activity normalizes after a few months to one year (Nuwer et al., 2005; for a review, see Amyot et al., 2015). Therefore, it does not seem likely that mild TBI status, of which the current sample was primarily composed, would influence neurophysiological activity, especially since all TBIs occurred at least one year prior to participant lab visit in the current study.

Alternatively, it is possible that a third variable is responsible for the observed TBI moderation of CUD’s association with stress modulation of reward processing. Specifically, an underlying personality trait that predisposes individuals towards both TBI and externalizing behaviors (e.g., risk-taking, substance use) may be influencing current results. Disinhibition is highly heritable and has previously been found to predict reward dysregulation (Dawe et al., 2004). For instance, in a substance using sample, Joyner et al. (2019) found that disinhibition interacted with a blunted RewP to predict substance use problems, in line with allostatic models of disordered substance use. Thus, although speculative, it may be that TBI status is a proxy variable for a disinhibited temperament. Future research is needed that explicitly measures TBI-associated individual differences to better understand the underlying mechanisms of TBI’s moderating effect on the relationship between CUD severity and reward processing.

Finally, TBI status primarily affected the relationship between CUD severity and stress modulation of the positive reinforcement RewP, such that there is an observed negative association for TBI− but a positive association for TBI+. In other words, for people with no history of TBI, as disordered cannabis use becomes more severe there appears to be greater responsivity to positively reinforcing stimuli (i.e., receiving money) after acute stress, whereas this responsivity is blunted in people with TBI history, consistent with allostatic models of addiction (Koob & LeMoal,2001). Notably, stress enhancement of responsivity to negative reinforcement (i.e., avoidance of a loud noise blast) was found for severe CUD in both TBI+ and TBI− individuals. Taken together, failing to find a feedback interaction in the TBI− subsample may be consistent with incentive salience models of addiction, which posit that cross-sensitization of stress and reward-relevant neural circuits occurs as addiction develops (Berridge & Robinson, 1998), whereas the observed pattern in the TBI+ group was consistent with allostatic models (Koob & LeMoal, 2001).

There are several limitations in the current study that must be addressed. First, the study sample was predominantly White, precluding generalization of findings to non-White individuals. Second, the current study did not include a control condition in which there was no lab-induced stressor. Future research may take this into account by having participants complete a comparable non-stressful control version of the MMST to ensure that stress-related, and not task novelty or fatigue effects, are driving current findings. Third, while subjective ratings to the stress task were available for analysis, there were no data pertaining to sustained subjective response to the stress task. While subjective initial stress response did not interact with CUD severity in the positive or negative reinforcement conditions to predict RewP/relief-RewP amplitude, it is possible that subjective sustained stress response moderation effects may have emerged. An additional limitation to the current study is that initial power analyses were completed based on the expected effect size of the larger study’s primary hypothesis, which was that TBI group differences (i.e., TBI severity), would predict CUD severity. In turn, the initial power analysis is not applicable to determine power for testing the current study’s primary hypothesis. Further, it is important to note that TBI severity in the current sample was disproportionately skewed towards those who only experienced a mild TBI (79.41%), while a smaller proportion endorsed a TBI of at least moderate severity (20.59%), and zero participants with severe TBI history. In light of current findings, it is possible that not only TBI status, but TBI severity interacts with CUD symptoms to influence neurophysiological reward responsivity. Although unlikely, it is possible that the 24-hour abstinence period may have induced withdrawal symptoms in participants with more severe cannabis use disorder. Considering the current study did not take into account potential withdrawal symptoms during the EEG lab visit, future studies attempting to replicate current findings may benefit by examining whether the 24-hour abstinence period induces withdrawal symptoms sufficient enough to affect results. Finally, the current study lacked a control group of cannabis users who do not meet criteria for a CUD as well as a non-stressful control task condition. Notably, only 9.2% (n = 8) individuals did not meet criteria for CUD in the current sample, while 28.7% (n = 25) met for mild CUD, 35.6% (n = 31) met for moderate CUD, and 26.4% (n = 23) met for severe CUD (see Table 1). A control group comprised of non-disordered cannabis use individuals would allow for greater confidence that findings are a result of a) disordered cannabis use rather than occasional/recreational use, and a non-stressful control task would ensure that b) acute stress, rather than expectancies, fatigue, or habituation, is responsible for changes in reward processing.

Findings from the current study may have useful clinical implications. Broadly, the RewP/relief-RewP may be used as neurophysiological indices to examine the underlying stress+reward processing relationship in the brain. Further, physiological sensitivity to acute stress may be an important individual difference in severe CUD relevant to stress-induced alterations in reward processing, which may be helpful in risk identification efforts. Finally, current results provide a foundation from which future research may examine how differential neurophysiological reactivity to positive vs. negative reinforcement influence drug-seeking behavior during periods of stress often experienced by those with CUD (e.g., withdrawal during a quit attempt).