Daily cannabis use in adolescents who smoke tobacco is associated with altered late-stage feedback processing: A high-density electrical mapping study

Abstract

Impairments in feedback processing, often associated with risk-taking behavior, may have implications for development of substance abuse in adolescents. The most commonly used substances by adolescents include tobacco and cannabis, with some individuals using both substances, potentially heightening risk. Our objective was to examine feedback processing and impulsivity in adolescents who smoke cigarettes and use cannabis daily (N = 21), comparing them with adolescents who smoke cigarettes daily and use cannabis occasionally (N = 18) and non-smoking (N = 27) adolescents. To do this, the Balloon Analog Risk Task (BART) with concurrent EEG was used to measure risk- related feedback processing, and impulsivity was measured using the Barratt’s impulsiveness scale (BIS-11). It was found that adolescent daily tobacco/cannabis smoking was associated with higher BIS-11 scores, shortened feedback-related-negativity (FRN) latencies and reduced P300 amplitudes. In addition, FRN latencies during win conditions were inversely associated with tobacco-use severity, indicated by scores on the Fagerstrom Test for Nicotine Dependence (FTND), and with BIS-11 scores. Adolescents with daily tobacco and daily cannabis use show altered feedback processing and higher impulsivity. Future work should disentangle whether the effect reflects risk, consequences of use or both.

Introduction

Impaired feedback processing is common in individuals with substance-use disorders (SUDs) (Forster et al., 2017; Verdejo-Garcia et al., 2018). Feedback processing can be considered a form of performance monitoring, as information gleaned from the results of one’s behaviors (feedback) may then be used to correct mistakes and continuously improve behavior, increasing task efficiency (Masters et al., 2009). Feedback processing is tightly related to both learning (Luft et al., 2013) and decision-making (Walton et al., 2004). Difficulties in feedback processing may impair decision-making, contributing to impulsive behaviors and drug use in individuals with SUDs (Garavan and Stout, 2005). Feedback processing impairment is often observed in adults who use substances such as alcohol or cocaine (Fein and Chang, 2008; Morie, 2016; Morie et al., 2014; Parvaz et al., 2015), but less is known about adolescents who use substances. Impaired feedback processing may contribute to impulsivity and more severe drug use behaviors, including tobacco and cannabis use, which are commonly consumed by adolescents (Miech et al., 2019; Palmer et al., 2009). Here, we examine feedback processing in adolescents who do and do not use tobacco and cannabis and we examine links to impulsivity and substance use severity.

Feedback processing impairment may be particularly detrimental in adolescents (Geier, 2013). Elevated impulsive and risk-taking behaviors are common in this age group (Kann et al., 1993), and increased impulsivity is linked to adolescent substance use (Schepis et al., 2008). Impulsive behavior in adolescents may be driven by developmental imbalances between subcortical regions promoting reward-seeking behavior that mature more rapidly and regions promoting top-down prefrontal control, such as the anterior cingulate cortex (ACC) and orbitofrontal cortices, that mature less rapidly (Galvan et al., 2006). Thus, adolescents may be more likely to engage in sensation-seeking behaviors, including substance use. Adolescents who consume marijuana (Dougherty et al., 2013) and cigarettes (Reynolds et al., 2007) have demonstrated elevated self-reported impulsivity, and our group has reported similar observations (Hammond, In Press). However, how impulsivity relates to brain processes, particularly those underlying feedback processing, in adolescents who use tobacco or cannabis is not well understood. Such information could contribute further to understanding of how the temporal dynamics of feedback processing may relate to risk-taking behaviors.

Adolescents may engage in risk-taking behaviors for different reasons, with positive reinforcement motivations (e.g., for excitement) and negative reinforcement motivations (e.g., to relieve uncomfortable states like dysphoria or stress) constituting two domains (Bergevin et al., 2006; Farhat et al., 2021). Impulsivity may also relate to these domains differentially (e.g., positive and negative urgencies (Cyders et al., 2014)). Thus, understanding how impulsivity relates to different aspects of processing of rewards (theoretically linked to positive reinforcement processes) and losses (theoretically linked to negative reinforcement processes) may provide insight into factors that may represent targets for interventions aimed at preventing adolescent risk-taking and related harms.

Laboratory measures suggest that cannabis-using adolescents engage in riskier behavior during the Balloon Analog Risk Task (BART) (Hanson et al., 2014). Data on BART-assessed risk-taking in tobacco-smoking adolescents is less consistent, with some studies reporting increased risk-taking (Lejuez et al., 2003) and other work indicating increased avoidance of risk (Dean et al., 2011). Understanding feedback processing in adolescents who use substances and how these processes relate to impulsive tendencies is important for advancing treatment and prevention efforts in this high-risk group. The emerging legalization of cannabis in multiple jurisdictions, and widespread vaping of either tobacco, cannabis or both in adolescent populations, (Miech et al., 2017; Ramamurthi et al., 2018), highlights the importance of understanding how use of these substances, separately and in combination and across levels of severity, may be associated with feedback processing in adolescents.

Event-related potentials (ERPs) are a useful method for interrogating the neural correlates of feedback processing and may reveal between-group processing differences that are not apparent in self-reports or laboratory performance measures. ERP amplitudes may indicate degree of underlying brain systems engagement, while ERP waveform latency may indicate processing speed (McCarthy and Donchin, 1981). Both aspects of the ERP are important to consider in populations that use substances, as both tobacco and cannabis may affect integrity and speed of processing (Bates et al., 1995; Fridberg et al., 2013; Knott et al., 1999). Two useful ERP components for the purposes of examining feedback processing are the feedback-related negativity (FRN) and the P300. Data suggest that in response to surprising feedback, dips in dopamine release produce downstream effects which are manifested in the ERP (Hauser et al., 2014; Holroyd and Coles, 2002). The FRN is a negative deflection at fronto-central recording sites that reaches a maximum approximately 200 and 300 ms after feedback stimuli. This component is widely considered to be an automatic perception of a difference from what is expected (Bellebaum et al., 2010). Investigations of the FRN in addictive disorders in adolescents implicate feedback-processing alterations in groups at risk for addictions (Crowley et al., 2009b; Morie et al., 2018) and in adolescents with problematic internet use (Yau et al., 2015). Investigations by our group into the FRN have also indicated that the FRN changes with age, with larger, longer-latency waveforms in younger individuals (10–12) compared to older individuals (15–17) (Crowley et al., 2013). This work suggests that amplitude and latency of the FRN may show reductions as feedback processing develops in adolescence.

The second useful component for examining feedback processing is the P300. The P300 may reflect more elaborate appraisal of the meaning of feedback and consideration of behaviors on the next trial (Twomey et al., 2015). The P300 is sensitive to details of the feedback, including valence (Yeung and Sanfey, 2004), with some work indicating greater ERP magnitudes corresponding to greater magnitudes of rewarding feedback (Padron et al., 2016). Like with the FRN, reduced P300 amplitudes are seen in substance-using populations (Parvaz et al., 2012), adolescents with problematic internet use (Yau et al., 2015), and populations at high risk for substance use (Crowley et al., 2009a; Morie et al., 2018). In a sample of individuals with alcohol use disorder, P300 amplitudes elicited during BART performance were associated with family history density of alcoholism (the number of first-degree family members reporting alcohol abuse/dependence). This suggests that later stages of feedback may be an endophenotypic marker of SUDs (Fein and Chang, 2008). The collected literature on the FRN and P300 lends evidence to the conclusion that feedback processing is altered in individuals with, or at risk for, SUDs. These feedback-processing alterations may underlie increased risk-taking leading to substance use and SUDs (Garavan and Stout, 2005). We extend this literature by investigating feedback processing and impulsivity in cannabis- and tobacco-using adolescents.

The present study examined feedback processing, indicated by both behavior on the BART and amplitudes and latencies of the FRN and P300, among three different adolescent groups: those using cannabis daily and tobacco daily, those using tobacco daily with occasional cannabis use, and adolescents without tobacco or cannabis use. The BART is a reliable way to investigate the FRN and P300 in response to feedback in drug-using populations (Euser et al., 2013; Lejuez et al., 2003; Lejuez et al., 2002), and previous work from our group has demonstrated its utility in examining the FRN and P300 in adolescents prenatally exposed to cocaine (Crowley et al., 2009a). During the BART, participants can win money or points by choosing to inflate a balloon with a specified number of pumps. A greater number of pumps includes a proportionally greater reward, but also a greater likelihood that the balloon will explode, resulting in no money or points for that trial. We hypothesized that youth who used both tobacco and cannabis would show higher risk-taking, indicated by more pumps on the BART, and blunted feedback processing, indicated by smaller FRN and P300 amplitudes and shorter latencies during both wins and losses. Dimensionally, we predicted that self-reported impulsivity, assessed using the Barratt Impulsiveness Scale (BIS-11), would be inversely associated with FRN/P300 amplitudes and latencies. We also hypothesized that more severe substance-use patterns, indicated by scores on the Cannabis Use Disorder Identification Test – Revised (CUDIT-R) and the Fagerstrom Test for Nicotine Dependence (FTND), would correlate inversely with amplitudes of the FRN and P300. We further postulated this relationship would be moderated by self-reported impulsivity.

Experimental Procedures

Participants

The sample consisted of 66 physically healthy adolescents between the ages of 14–21 years (17.64, SD = 1.2). All participants were recruited from local high schools in the greater New Haven metropolitan area and via flyers, peer-referral, and advertisements, and given a phone screening. Participants provided consent/assent, and participants under age 18 also had a parent/guardian provide consent. This study was approved by the Yale University School of Medicine Human Investigation Committee. Twenty-one participants reported smoking daily both cannabis and tobacco (high cannabis using tobacco smoking (H-CUTS), 18 reported smoking daily tobacco with occasional cannabis use (low cannabis using tobacco smoking (L-CUTS), and 27 were age-, gender-, and grade-level-matched non-smoking typically developing adolescents (non-smoking). All participants reporting tobacco use consumed cigarettes or cigarettes/cigars. All participants were free of serious mental illness (psychosis, autism, bipolar disorders) and had no history of lifetime or current DSM-IV-TR diagnosis of dependence on another psychoactive substance (other than cannabis and tobacco in the case of the smoking participants). Additional exclusionary criteria included: neurologic conditions (e.g. seizures, migraines), head trauma with loss of consciousness > 2 minutes, use of any psychoactive drugs including anxiolytics and antidepressants (unless the adolescent had been taking the prescribed medication consistently for 3 months). Other than medication information for purposes of exclusion, we did not collect structured data on depressive, attention-deficit or anxiety disorders. All adolescents completed a breathalyzer for alcohol use and a urine drug screen and were monitored for signs/symptoms of intoxication and rescheduled if impaired.

Procedures

Telephone-screening interviews were administered to adolescents and their parents/guardians prior to study entry. Participants who met inclusionary criteria, and whose parents provided consent if they were under age 18, were then scheduled for a single 3-hour study session. In the session, they completed self-report questionnaires, behavioral assessments, and urine testing. They then performed the BART during electroencephalography (EEG).

Balloon Analog Risk Task

The BART utilized here was identical to that described previously (Yau et al., 2015). The task included 60 trials presented with E-prime v2.0 (Psychology Software Tools, Inc.). Each trial displayed a balloon in the middle of the screen connected to a pump with a number dial below. Only one balloon was presented per trial, each having a maximum breaking point of 128 pumps. Rather than sequentially pumping the balloon (as in the standard BART (Lejuez et al., 2002) which arguably may not be an accurate measure of the impulsive processes underlying risk-taking behavior as it gives the individual time to reflect upon their decision, participants selected the target number of pumps (corresponding to how much risk) at the beginning of each trial. For each pump, participants could obtain one point; however, if the balloon popped, the accrued points for the trial would be lost. Participants were instructed to try to win as many points as possible, though no extra payments were given for points accrued.

In order to provide sufficient numbers of popped balloons for FRN responses, a 2-block hybrid task was adopted. The first block of 30 balloons followed the “pop” points of the original BART (Lejuez et al., 2002), with balloon explosion probabilities following a normal distribution spread. The probability of the second block of 30 balloons popping was rigged such that there was a 50% chance of reward and 50% chance of loss across the task as long as the number of pumps was not extremely low (below 10) or extremely high (above 118). This ensured that a sufficient amount of “losing” trials (over 15 trials) would incur to allow for FRN recordings, while preserving the behavioral task, visually comparable to the original BART in the first 30 trials.

Measures

Sociodemographics and IQ

Sociodemographics assessed included age, gender, race/ethnicity, school involvement and grade level. The Wechsler Abbreviated Scale of Intelligence (WASI) (Weschler, 1999) Vocabulary and Matrix subtests were used to estimate Full-Scale IQ based on standard norms.

Impulsivity

The BIS-11 is a 30-item self-report questionnaire (Patton et al., 1995). It includes three subscales that assess components of impulsivity: (1) attentional impulsivity (8 items); (2) motor impulsivity (11 items); and (3) non-planning impulsivity (11 items). The BIS-11 has good divergent and convergent validity and test-retest reliability (Patton et al., 1995).

Substance-Use Frequency

Substance-use frequency for cannabis, tobacco, alcohol and other drugs was measured using the Timeline Follow-back (TLFB). The TLFB assessed past-90-day use patterns (Lewis-Esquerre et al., 2005; Sobell and Sobell, 1992).

Cannabis-related Problem Severity

Cannabis-related problem severity was assessed with the CUDIT-R (Adamson et al., 2010), an 8-item self-report measure assessing symptoms of DSM-5 cannabis-use disorder over the past six months. Total scores for the CUDIT-R range from 0 to 32, with a score of 13 suggested as a threshold for moderate-to-severe CUD.

Tobacco-related problem severity

Tobacco-related problem severity was assessed with the FTND (Prokhorov et al., 1996), a 7-item instrument assessing symptoms of tobacco-use disorder that has been adapted for youth populations. Total scores for the FTND range from 0 to 9, with a score of 3 or more suggested as a threshold for moderate-to-severe tobacco-use disorder.

Biochemical assays for cannabis, tobacco and other substance use

A urine sample was collected at the beginning of the study visit to provide a biochemical assessment of drug use. Qualitative urine screening using an immunoassay-based point of care test kit assessed for five different drugs of abuse (cannabinoids, cocaine, opioids, methamphetamines, benzodiazepines). A semi-quantitative urine cotinine test assessed levels of cotinine in all participants. Urine samples of participants whose screen was positive for cannabinoids were sent to Quest diagnostics laboratory where mass spectroscopy was used to characterize quantitative urine cannabinoid levels (11-Nor-9-carboxy-Δ9-tetrahydrocannabinol (THC-COOH), creatinine corrected, ng/dL).

EEG recording and analysis

Participants were seated 1 m in front of the computer, and their heads measured to determine appropriate electrode net size. Event-related potentials were acquired from a 128-channel sensor net of carbon fiber electrodes with Ag/Cl coating. The nets were soaked in a potassium chloride solution for 10 minutes beforehand to ensure low impedance without the need for abrading the participant’s scalp. EEG data were recorded and processed using NetStation v4.4 software package (EGI, Inc.) and EGI high-impedance amplifiers, sampling at 250Hz (EGI, Inc., Series 300 amplifier) with online filters set at 0.1–100Hz. All electrode recordings were referenced to the Cz electrode which were then re-referenced offline for data analysis.

Data were filtered offline at 0.1–30Hz and segmented into epochs that covered a time-frame of 100ms pre-stimulus baseline and 600ms post-stimulus interval for analysis. Bad eye channels were manually marked and interpolated by surrounding channels. Epochs with any eye-blink or eye-movement (threshold 150μV) or containing more than 10 bad channels were rejected. Epochs with fewer than 10 bad channels resulted in a procedure that involved interpolation of those bad channels with two surrounding channels and then re-analyzing the segment to replace the information in the epoch with the bad channels. Single-trial EEG data were re-referenced from the Cz electrode to an average reference as the latter better represents a true zero (Junghofer et al., 1999). The baseline was corrected to 100ms pre-stimulus. ERPs were obtained by averaging the single-trial data for the two blocks of the BART. After ocular artifact correction (OAR) was applied (Gratton et al., 1983), only subjects with at least 10 artifact-free trials were included in the overall statistical analysis (N=66).

ERPs were computed by analyzing the mean amplitudes and mean latencies in a time-window post-onset of feedback for loss (balloon pop) and reward (no pop, points accrued) trials respectively (Yeung and Sanfey, 2004), at both fronto-central (FCz) and posterior (Pz) scalp locations. Frontal FRN amplitude was defined as the mean ±25ms from the negative peak amplitude between 200–350ms, averaged from a group of electrodes over the midline of the fronto-central regions (channel numbers: 16, 18, 4, 11, 12, 19, 10, 5, corresponding to electrode FCz). The P300 component for the purposes of this study was defined as the mean ±25ms from the positive peak amplitude between 250–350ms, averaged from a group of electrodes over the posterior region (channel numbers: 72, 71, 75, 76, corresponding to electrode Pz). Mean amplitudes and latencies were extracted using the Netstation statistical extraction tool.

Statistical analyses

For all statistical analyses, an alpha level of .05 was employed. All analyses were computed separately using IBM SPSS Statistics Analytic software V26.0 (IBM, Armonk, NY). Group differences between H-CUTS, L-CUTS, and non-using adolescents were calculated with a one-way ANOVA. For indices of tobacco-use and cannabis-use severity across the two groups who used substances, the H-CUTS group was compared to the L-CUTS group using independent sample t-test for continuous variables and Chi-squares for categorical variables.

The mean number of pumps across balloons during the BART was analyzed for the behavioral measure of risk-taking and served as the primary dependent variable. Additionally, the deliberation time (i.e. the amount of time to decide on the number of pumps made) was examined. These measures, as well as FRN and P300 amplitudes and latencies, were examined with a 2×3 repeated-measure ANOVA with group (H-CUTS, L-CUTS, non-using) as the between-subject factor and condition (win, loss) as the within-subject factor. Gender and age were included as covariates, as previous work in our laboratory has demonstrated that age (Crowley et al., 2013) and gender (Crowley et al., 2009a) may influence amplitude or latency of the FRN and P300 in adolescents. Bivariate correlation analyses using Pearson’s correlation coefficients were used to examine associations between BIS-11 scores and ERP measures in all groups. Correlations were also performed between indices of drug use severity (CUDIT/FTND scores, urine cotinine levels and THC/Creatine ratios) and ERP measures in the substance-using groups. A Bonferroni correction was applied for the number of comparisons.

Exploratory moderation analyses were performed using SPSS PROCESS (Hayes, 2013) with bootstrapping. Specifically, the moderating effect of impulsivity on the relationships between ERP measures indices and substance-use severity was evaluated. The model was tested with 5000 resamples and 95% confidence intervals (95% CI). Significance was set at p = 0.05 and when the 95% CI did not include zero. The moderation analysis utilized the default model 1 embedded within SPSS PROCESS (Hayes and Matthes, 2009) to examine the direct and indirect paths through impulsivity and substance-use severity. Age and gender were included as covariates.

Results

Demographics

Sixty-six individuals completed questionnaire and EEG measures. The sample was similar to and overlapping with that described previously (Hammond, In Press), with two fewer non-smoking adolescents who did not complete EEG recordings and three additional smoking adolescents who completed recordings. Groups consisted of 21 H-CUTS, 18 L-CUTS, and 27 non-smoking adolescents. There were no differences between groups in age (F = .562, p = .57), gender (F = .09, p = .90) or ethnicity/race (F = 1.49, p = .23). Though in the average range, non-smoking adolescents had higher full-scale IQs (107.54, SD = 10.79) than did L-CUTS (99.81, sd = 10.63) or H-CUTS (97.07, SD = 10.22) adolescents (F = 5.43, p <.01). Non-smoking adolescents also had lower BIS-11 scores (58.91, SD = 9.55) compared to L-CUTS (67.43, sd = 11.11) and H-CUTS (68.99, sd 12.21) adolescents (F = 5.35, p < .01). Non-smoking adolescents reported fewer days of alcohol use in the past month (F = 9.23, p < .01). Table 1 illustrates detailed demographic and substance-use information between all groups (smoking comparisons included when relevant).

Table 1:

Demographic information

	Non-smoking (N = 27), 18 male		L-CUTS (N = 18), 13 male		H-CUTS (N = 21), 15 male		Total
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	F	p
Age	17.41	1.20	17.85	1.20	17.84	1.20	17.64	1.20	0.87	0.42
FSIQ	107.54	10.79	99.81	10.63	97.07	10.22	102.61	11.39	5.43	0.01
Past 30 days: Alcohol use	0.13	0.37	2.65	2.83	1.98	2.48	1.32	2.24	9.23	<0.01
Past 30 days: Heavy Drinking	0.01	0.06	1.94	2.63	1.58	2.37	0.97	2.01	6.57	<0.01
BIS-11 Total	58.91	9.55	67.43	11.11	68.99	12.21	63.96	11.54	5.35	0.01
									t	p
Cotinine Level (semi-quantitative); >3 indicates smoking	1.00	0.00	5.67	0.84	5.24	1.17	5.44	1.05	1.65	0.20
Breath CO Level	0.96	0.34	4.59	3.12	4.76	5.03	4.68	4.23	0.02	0.90
Past 30 days: Tobacco Use, Days	N/A	N/A	28.11	3.51	25.71	8.63	26.81	6.79	1.21	0.28
CPD	N/A	N/A	8.12	3.99	9.43	6.63	8.84	5.57	0.51	0.47
CUDIT Score	N/A	N/A	8.50	7.30	15.67	6.82	11.74	7.92	11.65	<0.01
FTND Score	N/A	N/A	3.81	1.42	4.40	1.59	2.23	2.33	0.26	0.61
THC/Creatinine	NA	N/A	107.73	123.78	145.28	134.67	131.79	135.02	0.56	0.45
Days since last Cannabis use	N/A	N/A	124.22	276.41	0.60	0.50	57.69	195.13	4.21	0.05

L-CUTS = Low Cannabis Using Tobacco Smoking; H-CUTS = High Cannabis Using Tobacco Smoking; FSIQ = Full Scale IQ, BIS-11 = Barratt Impulsiveness Scale, CO= Carbon Monoxide, CPD = Cigarettes per day, CUDIT = Cannabis Use Disorder Identification Test, FTND = Fagerstrom Test for Nicotine Dependence; THC = Tetrahydrocannabinol; SD = Standard Deviation

ERP analyses

Behavior

There were no differences between groups on mean deliberation times (F = .72, p = .49), (HC mean = 3718.6, SD = 914.5; L-CUTS mean = 4071.6, SD = 837.6; H-CUTS mean = 4169.63, SD = 1250.4) numbers of popped balloons (F = 2.51, p = .12), (HC mean = 12.3, SD = 2.9; L-CUTS mean = 13.2, SD =3.1; H-CUTS mean = 11.3, SD = 3.5) or numbers of pumps (F = 2.46, p = .093) (HC mean = 49.9, SD = 14.9; L-CUTS mean = 55.3, SD =12.9; H-CUTS mean = 44.1, SD = 18.0). Figure 1 illustrates the FRN and P300 components for wins and losses in the non-smoking, L-CUTS, and H-CUTS groups.

Figure 1:

FRN (frontal channels) and P300 (posterior channels) amplitude for non-smoking, L-CUTS (low cannabis using tobacco smoking), and H-CUTS (high cannabis using tobacco smoking) adolescents during win and loss conditions during the task. The central channel map highlights channels that display the electrode locations for the FRN and P300. The insets depicting the balloon demonstrate what participants saw during a loss (red square) or win (green square) condition.

FRN

The ANOVA for the amplitude of the FRN revealed significantly larger amplitudes for the loss condition than the win condition (F_1,63=5.42, p<.03, eta² = .082). However, there was no significant effect for group (F_1,63=1.25, p=.29, eta²= .040) and no significant interaction of condition and group (F_1,63=.057, p=.56, eta²= .018). The ANOVA for FRN latency revealed no difference between conditions (F_1,63=.22, p=.63, eta²= .004), though there was a significant difference in latency between groups (F_1,63=3.49, p < .04, eta²= .11). There was no interaction (F_1,63=.14, p=.86 eta²= .005). Post-hoc analyses revealed that FRN latency in the H-CUTS group (mean win = 291.6, SD = 38.00; mean loss = 276.15, SD = 18.2) was smaller than the latency values for the FRN in the non-smoking group (mean win = 309.15 SD = 26.8; mean loss = 291.1, SD = 26.5, p < .02).

P300

The RM-ANOVA for the P300 (loss versus win) was not significant for condition at the p = .05 level (F_1,63= 3.31, p= .07, eta² = .062), and there was no interaction of condition by group (F_1,63= .49, p = .61, eta²= .016). However, there was a main effect of group on P300 amplitude (F_1,63= 3.42, p < . 04, eta²= .11). Post-hoc analyses revealed significantly smaller P300 amplitudes in the H-CUTS group (mean win = .41 SD = 2.5; mean loss = 1.5, SD = 2.4) compared to the non-smoking group (mean win = 1.92, SD = 2.1; mean loss = 2.45, SD = 1.7) (p < .04). Latency analyses for the P300 revealed no effect of condition (F_1,63= .77, p= .38 eta²= .013) or group (F_1,63= .98, p= .37, eta²= .032), and no interaction (F_1,63= .74, p= .48, eta²= .025).

Associations between ERP, self-report and behavioral measures

Across all participants, scores on the BIS-11 were negatively associated with FRN latency during the win condition (r = −.28, p < .03). Correlations between ERP measures (FRN and P300 amplitude and latency in response to losses and wins) and indices of drug-use severity in substance-using groups were examined, and a strict Bonferroni correction applied due to the number of comparisons. Scores on the FTND were negatively associated with FRN latency during win conditions (r = −.43, p = .008). Scatterplots illustrating these relationships can be found in figure 2. CUDIT scores, urine cotinine levels, and THC/creatinine ratios were not correlated with any ERP measures (all p values > .1).

Figure 2:

Scatterplots illustrating the relationships between FRN latencies for measures of tobacco-use severity and impulsivity.

Exploratory analysis: Moderation of tobacco-use severity by impulsivity

The presence of correlations between FRN latencies, BIS-11 scores and tobacco-use severity indexed by the FTND allowed for the development of a moderation model, which posited that the level of impulsivity would moderate the relationship between FRN latency and tobacco-use severity in both groups of smoking adolescents during the win condition. The model approached significance when only L-CUTS and H-CUTS adolescents were included, (R^2 = .18, p = .07), and the moderation was not significant (R^2 change = .0096; figure 3). When examining the full sample, a moderating effect of impulsivity on the relationship between FRN latencies for losses and tobacco-use severity was observed. The model indicated that impulsivity moderates the relationship between the speed of feedback processing capability (indicated by the latency of the FRN during the loss condition) and tobacco-use severity, with low levels of impulsivity associated with stronger inverse associations between FRN latencies during losses and tobacco-use severity. (For full details, please see supplementary materials).

Figure 3:

Model of potential moderation of FRN latency during wins and nicotine severity. Moderation was not significant.

Exploratory analysis: Examination of P300

By request of a reviewer, we additionally examined the P300 with a longer time-window that encompasses a larger portion of the waveform, from 300–550 ms. However, these additional analyses revealed that the RM-ANOVA for the P300 (loss versus win) was not significant for condition at the p = .05 level (F_1,63= 1.3, p= .24, eta² = .026), and there was no interaction of condition by group (F_1,63= .26, p = .77, eta²= .01). There was no main effect of group on P300 amplitude in this time-window (F_1,63= .69, p < . 50, eta²= .01). Similarly, latency analyses for the P300 revealed no effect of condition (F_1,63= 2.1, p= .15 eta²= 034.) or group (F_1,63= 1.68, p= .19, eta²= 05.), and no interaction (F_1,63= . 34, p= .71, eta²= .01).

Discussion

Feedback processing, including early stages reflecting “automatic” responses (FRN) and later stages reflecting the meaning of feedback (P300), may be impaired in adolescents who use both cannabis and tobacco. H-CUTS adolescents showed greater reductions in P300 amplitudes and reduced FRN latencies. Further, impulsivity was associated with FRN latencies during feedback in all groups, implying that the differences observed between groups may reflect an increased risk-taking propensity in the H-CUTS group. FRN latencies were also associated with greater tobacco-use severity. Implications are discussed below.

Baseline differences in substance use between groups, including days consuming alcohol, suggest heightened substance-related risk-taking in H-CUTS adolescents. BIS-11 scores were higher in H-CUTS and L-CUTS adolescents relative to non-smoking adolescents, which is consistent with previous findings of heightened impulsivity in cannabis-using adolescents (Beaton et al., 2014) and in those at elevated risk for substance use (Verdejo-Garcia et al., 2008). As expected, this pattern was also observed in the previous report involving largely the same sample (Hammond, In Press). Impulsive tendencies are common among adolescents who use substances (Geier, 2013). There was also a difference in IQ between non-smoking adolescents and H-CUTS and L-CUTS adolescents, and this is consistent with findings that have indicated that individuals who smoke show lower scores on tests of cognition (Weiser et al., 2010). Inclusion of the IQ score as a covariate in the RM-ANOVA analyses did not change the significance of the results. However, there were no differences in BIS-11 scores between L-CUTS and H-CUTS adolescents. It is possible that some other factors influence use severity of cannabis use in tobacco-smoking adolescents, such as availability of cannabis (Gillespie et al., 2009) and/or factors related to socioeconomic status. Investigations using other questionnaires that address different facets of impulsivity (for example, the UPPS-P (Watts et al., 2020)) may have resulted in other findings. Further, the lack of behavioral differences between any of the groups on the BART suggests that risk-taking between these groups did not differ in the laboratory setting, at least behaviorally. It is possible that the modified version of the BART employed (which had participants select the number of pumps per trial and not perform the pumps themselves) resulted in this null finding. However, ERP findings suggested that neural correlates of responses to feedback differed between groups. Findings of ERP/neuroimaging differences with the absence of behavioral differences is common among child and adolescent populations that are at high risk of substance use (Morie et al., 2019), potentially indicating increased effort to perform at similar levels to controls. Investigations into adolescent marijuana use specifically have also revealed indications of enhanced BOLD response (Tapert et al., 2007) despite lack of performance differences on an inhibitory control task. Work examining performance monitoring related to tobacco smoking has also indicated alterations in error-related ERP components in the absence of between-group performance differences (Rass et al., 2014). These findings emphasize the power of neuroimaging measures to identify subtle processing differences that may not be observable in participant behavior, and offer insight into processes that may perpetuate drug seeking and taking.

ERP findings between groups suggest that early, automatic feedback detection indicated by the FRN is altered between substance-using and non-using adolescents, with shorter latencies for the FRN seen in H-CUTS adolescents. Shorter ERP latencies may reflect faster processing of a stimulus, as indicated by work on the P300 (McCarthy and Donchin, 1981) and FRN in people with gambling problems (Oberg et al., 2011). In the past, cannabis-using populations have shown latency differences in the P300 elicited by auditory or visual odd-ball tasks, implying faster processing (de Sola et al., 2008; Kempel et al., 2003; Solowij et al., 1995). Shorter latencies in this group, however, may also imply potentially more disinhibited responding (Fridberg et al., 2013), and this is supported by the finding that FRN latencies were negatively associated with scores on the BIS-11.

While latency differences were not seen in the P300, there were between-group differences in the amplitudes of the P300, which appeared driven by smaller amplitudes in the H-CUTS group. Similar findings were also seen in adolescents with prenatal cocaine exposure (Morie et al., 2018). It is possible that adolescents who smoke tobacco and have more severe cannabis use have particularly impaired feedback processing or are temperamentally more impulsive, and these tendencies may influence high-risk decisions such as cannabis use. This notion is consistent with work in adults with chronic cannabis use, many of whom also smoked tobacco, and this research found blunted neural activations to reward and loss feedback (Enzi et al., 2015). Taken together, the literature coupled with this most recent work suggest that daily use of tobacco and cannabis may accompany increased automatic processing speed of feedback and be coupled with impaired abilities to use information about feedback, potentially leading to impaired decision-making. Future work investigating acute effects of cannabis or tobacco use on feedback processing or longitudinal studies of risk-taking behaviors in adolescence may further shed light on these processes and their clinical correlates.

The associations observed here between self-reported tobacco-use severity and feedback responses may also contribute to understanding feedback processing in substance-using adolescents, particularly use of tobacco. FTND scores were negatively associated with FRN latencies during win conditions, suggesting potentially faster, less robust feedback processing as a function of tobacco-use severity. This finding strengthens the theorized association between reduced integrity of feedback processing and increased risk-taking behaviors in adolescents, particularly substance-using ones (Webber et al., 2017). The tobacco-use-severity/FRN-latency finding for losses is consistent with effects of nicotine on processing speed (Bates et al., 1995). Taken together, our findings suggest that tobacco use is associated with faster processing speed, but concurrent tobacco and cannabis use may blunt feedback processing as indicated by reduced P300 amplitudes, potentially contributing to impulsivity and poorer decision-making.

The three-way association between FRN latency during win conditions, BIS-11 scores and FTND scores allowed for an exploratory examination of the moderating effect of impulsiveness on the relationship between feedback-processing speed indicated by FRN latency and tobacco-use severity. The moderation model, which tested if impulsivity moderated the relationship between feedback processing speed and tobacco-use severity, approached significance when only smoking adolescents were included, and reached statistical significance in the entire sample. These exploratory analyses suggest that impulsivity moderated the relationship between FRN latencies during loss processing and tobacco-use severity, suggesting that the speed of processing during negative feedback is related to increased tobacco-use severity in individuals with less impulsivity. In other words, if an adolescent shows less impulsivity, increased severity of tobacco use is associated with increased feedback processing speed. However, findings from the sample of solely the tobacco-smoking adolescents suggest that those high in impulsivity show shorter FRN latencies. This may suggest that in less impulsive individuals, increased tobacco use is associated with patterns seen in more impulsive individuals. Larger samples in future studies may further investigate and clarify these possible relationships.

The strengths of this work, including high-density EEG recording in three well-characterized, sizable groups of adolescents (H-CUTS, L-CUTS and non-smoking) are tempered by some limitations. The measurement of urine cotinine was semi-quantitative. Further, all adolescents who reported tobacco smoking used cannabis to some extent, preventing us from examining tobacco smoking free of cannabis use. Additionally, there was no independent investigation of a cannabis-only-smoking group, reflecting difficulties in identifying and recruiting such a cohort. It is possible that the results reported are a function of co-use of tobacco and cannabis, as individuals who use both typically report more problems than those who use only one substance (Peters et al., 2012). Future work with more precise measurements of urine cotinine, with larger samples and the addition of solely tobacco-smoking and solely cannabis-smoking adolescents, may help extend the current findings. In addition, alcohol use differed between groups, with both H-CUTS and L-CUTS participants reporting more days of alcohol use and more days of binge-drinking in the past month. While findings did not change with the inclusion of alcohol use in the past month as a covariate, differences in alcohol use may have contributed to findings. We did not collect data regarding socio-economic status, and socio-economic status may contribute to risk-taking behaviors or tendencies to use tobacco or cannabis. Additionally, while individuals with serious mental illness (autism, schizophrenia) were excluded, we did not collect structured data on depressive, attention-deficit or anxiety disorders, which may have influenced our findings. Given the cross-sectional nature of the study, the findings cannot speak to the cause and effect of the observed differences in feedback processing. Finally, we did not assess individuals who may have vaped or consumed edible cannabis. Further studies should examine youth who vape nicotine or cannabis products or consume edibles. Finally, while we aimed to examine risk-taking using a modified BART that allowed for a suitable number of trials by manipulating probability, the use of a novel variant of the BART may reduce generalizability and comparability with other work that has employed the traditional BART.

Despite these limitations, our findings illustrate that later stages of feedback processing in adolescents are blunted in those who use both tobacco and cannabis, but without indicating whether the effect is a result of use or if the effect is indicative of individuals who have greater proclivities to use or co-use substances. Further, more severe tobacco use was associated with early stage feedback-processing latency differences. These findings may have relevance to the vaping of nicotine and cannabis products and legality of cannabis products in more states and potentially nationwide. Adolescents with higher risk-taking profiles may use these substances in combination more often as availability increases. Our findings suggest combined use of tobacco and cannabis is coupled with specific impairment in stages of feedback processing. Longer-term outcomes of this pattern of use may contribute to increased risk-taking behavior and decrements in efficient decision-making, and future studies should examine this possibility.