Electrophysiological correlates of resisting craving in adults with tobacco use disorders

Abstract

Background:

Tobacco use disorder (TUD) is the largest premature cause of death in the world. While most adults with TUD have attempted to quit, only less than 10% quit for at least 6 months. Craving is one of the main causes of smoking relapses. Event-related potentials (ERPs) have been extensively studied using cue-provoked paradigms. However, little is known about ERPs related to resisting craving, which seems important to investigate since such activity might represent treatment targets to promote smoking cessation.

Aims:

The goal of this work was to investigate the ERP correlates of resisting craving as compared to craving during a cue-provoked paradigm.

Methods:

Fifty-three adults with TUD were assessed on urinary cotinine, tobacco dependence, anxiety, and impulsivity. They were presented with cigarette-related stimuli during two conditions (Crave and Resist craving). Craving ratings were collected on visual analog scales. EEG was recorded using 64 Ag–AgCl electrodes.

Results:

Results revealed that P3 and early late positive potential (LPP) amplitudes were smaller during the Resist than the Crave conditions. P1 and late LPP amplitudes were also smaller but did not survive correction for multiple comparisons. Furthermore, anxiety level was related to the P3 amplitude during the Resist and Crave conditions, but did not survive correction for multiple comparisons.

Conclusion:

This line of work contributes to the identification of potential brain targets to help patients with TUD resist craving.

Introduction

Tobacco use disorder (TUD) is still the largest preventable cause of diseases and premature death in the world, despite a large variety of widely available treatments (Centers for Disease Control and Prevention, 2015; Danaei et al., 2009; Drope et al., 2018; Siddiqi et al., 2020; World Health Organization, 2019). While most adults with TUD wish to quit smoking and have attempted to quit, only less than 10% successfully quit for at least 6 months (e.g., Ahluwalia et al., 2018; Babb et al., 2017; Caraballo et al., 2017; García-Rodríguez et al., 2013; Prochaska and Benowitz, 2016; Walton et al., 2020).

Craving is one of the main causes of smoking relapses (e.g., Killen and Fortmann, 1997). Thus, one of the key factors to prevent relapses is the ability to resist craving. It has long been shown that smokers usually report elevated craving levels in response to smoking cues (e.g., Carter and Tiffany, 1999; Littel and Franken, 2007). Brain activity related to craving has also been extensively studied using cue-provoked paradigms and short-term abstinence (e.g., McClernon and Gilbert, 2004). For instance, the use of electroencephalographic (EEG) event-related potentials (ERPs) indicates that smokers displayed enhanced P3 and late positive potential (LPP) amplitudes when they were presented with tobacco-related stimuli (Littel and Franken, 2007, 2011; Nickel et al., 2023; Piasecki et al., 2017). Furthermore, smokers displayed enlarged early LPP when they were asked to imagine how pleasant it would be to smoke (Littel and Franken, 2011). However, when smokers were asked to regulate their craving (Nickel et al., 2023) or to distract their attention (i.e., focus on the color of smoking-related stimuli; Littel and Franken, 2011), they displayed decreased early and late LPP amplitudes. In sum, most studies focused on craving, but little is known about brain activity related to resisting craving, which is important to investigate since such activity might represent treatment targets to promote smoking cessation and prevent relapses (e.g., Hartwell et al., 2011).

The main goal of this work was to investigate the ERP correlates of resisting craving and craving during a cue-provoked paradigm. Based on previous studies discussed above (Littel and Franken, 2007, 2011; Nickel et al., 2023; Piasecki et al., 2017), indicating enhanced P3 and LPP associated with presentation of smoking-related stimuli or craving induction as compared to various control conditions (e.g., to regulate craving), our main hypothesis was that the P3 and LPP (especially the early LPP) would be smaller during the Resist than the Crave condition. Since there is little knowledge about resisting craving and its related brain activity, participants were invited to suppress craving as best they could without specific guidance (Hartwell et al., 2011), and several ERPs were measured to capture waveforms from P1, P3, early LPP, and to late LPP. We also explored potential relationships between these ERPs and patients’ characteristics of urinary cotinine level, average number of cigarettes smoked per day, nicotine dependence, anxiety, and impulsivity levels.

Materials and methods

Participants

Fifty-three adults with TUD (American Psychiatric Association, 2013) participated in this study (38 men, 15 women; aged 41.7 ± 13.2 years old; six left-handed). They were treatment seekers and in the Preparation stage to quit smoking as assessed by the Prochaska level of readiness questionnaire (Prochaska and DiClemente, 1982) and had previously tried to quit at least once. They smoked at least 10 cigarettes per day for at least 1 year. They had negative results on urine test for drugs using MedTox and on alcohol breath test using Alco-sensor FST from Intoximeters^® (Ontario, Canada ) prior to the experimentation. Finally, two participants reported having asthma, and one of them used a corticosteroid drug. They all provided written informed consent prior to their participation. The study was approved by the local Institutional Review Board.

Clinical assessment

Participants were asked about the average number of cigarettes they smoked per day in the last year. They reported smoking 17.6 cigarettes per day in the last year (SD = 6.4). We also measured the urinary cotinine levels, a primary metabolite of nicotine and a sensitive marker of nicotine exposure. It is also a more stable measure throughout time (less influenced by recent intake) and longer lasting in the body than nicotine (days vs hours; Raja et al., 2016). Bioanalyses were performed at the Centre de Recherche du CHUM in the Pharmacokinetics core facility for cotinine content using high-performance liquid chromatography and a high-resolution mass spectrometry validated method. Total creatinine levels were analyzed by standard biochemical method, and a cotinine-to-creatinine ratio was calculated to correct for the effect of urine volume and dilution variations among samples. Indeed, the concentration of a cotinine urine sample can fluctuate depending on the fluid quantity a person has consumed, and the cotinine-to-creatinine ratio controls for such variations and thus provides a more accurate measure of nicotine exposure than urine cotinine alone. Urinary cotinine data from two participants are missing due to aberrant values. The average urinary cotinine was 48.3 ng/ml (SD = 23.4). The reported average number of cigarettes smoked per day and urinary cotinine levels positively correlated (r = 0.334, p = 0.016).

Participants were also assessed on tobacco dependence, anxiety, and impulsivity with the Fagerstrom Test for Nicotine Dependence (FTND, Heatherton et al., 1991), the Beck Anxiety Inventory (BAI, Beck et al., 1988), and the Barratt Impulsiveness Scale (BIS, Barratt and Patton, 1983), respectively. Furthermore, they were instructed to smoke normally prior to the laboratory session and were abstinent from smoking approximately 1.5 hours before the EEG acquisition. Similarly, they did not consume caffeine at least 1.5 hours before the EEG acquisition. The FTND scores were an average of 5.7 (SD = 1.7), classified as moderate to high dependence level; the BAI scores of 7.1 (SD = 9.3), classified as minimal to mild anxiety level, and the BIS scores of 61.4 (SD = 10.2), suggesting high impulsivity level.

ERPs during a cue-provoked paradigm with Crave and Resist conditions

EEG was recorded using 64 Ag–AgCl electrodes (BrainAmp, Brain Products GmbH, Gilching, Germany). The electrodes were placed over the scalp according to the international 10–20 system with reference to FCz. Data were collected at a sampling rate of 5 kHz and an online filter of 0.01–250 Hz.

The cue-provoked paradigm consisted of two blocks of smoking-related stimuli. The first block consisted of the Crave condition, and participants were instructed to “Let yourself crave.” The second block consisted of the Resist condition, and participants were instructed to “Try by any means to resist craving.” Each block included 56 different smoking-related stimuli (e.g., individuals smoking, hands holding cigarettes). The order of these stimuli was randomized. Each stimulus was presented for 7 seconds and separated by a black screen of 500 milliseconds. Participants rated their craving level on a visual analog scale (VAS) on the computer screen at every four stimuli. The VAS scores were converted to a 0–100 range, with 0 representing no craving and 100 representing maximum craving. Craving ratings are missing for one participant due to technical errors. Participants had a short break between the two blocks and halfway through each block. Overall, the cue-provoked paradigm lasted an average of 16 minutes.

ERP data were re-referenced to the average, resampled to 500 Hz and filtered between 0.2 and 50 Hz (named dataset1, better suited for ERPs) and between 1 and 50 Hz (named dataset2, better suited for ICA; Winkler et al., 2015). Channels identified as bad through visual inspection were interpolated. Both datasets were epoched [−300, 3000] ms around the onset of each stimulus, and excessively noisy epochs were discarded before running the independent component analysis (ICA). We used dataset1 to run the ICA and transferred the ICA weights to dataset2. ICA decomposition was computed using the “runica” algorithm with PCA size of [nb channels – nb interpolated channels]. ICLabel was used to label the components (Pion-Tonachini et al., 2019). Noisy ICA components were rejected based on visual inspection and ICLabel results. Epochs were baseline corrected [−300, 0] ms and rejected if the point-to-point difference was higher than 50 μV/ms or if the amplitude at any point was higher than 70 μV. Each participant’s grand average waveforms were then visually inspected. First, we identified the time windows used by previous studies with designs similar to ours (Littel and Franken, 2007, 2011; Robinson et al. 2016; Nickel et al., 2023). Second, we sought within the ranges of these time windows reported in the literature those encompassed the best ERPs displayed by our participants. Specifically, we visually inspected individual ERP waveforms to adapt the time windows that reflect the greatest number of our participants. This two-step procedure was intended to ensure that the selected time windows captured the relevant neural activity for our participants while remaining grounded in established component definitions. ERPs were thus computed as the mean wave in the following windows and regions of interest (ROIs): P1 [100–200] ms time window of the electrodes O1, O2, Oz, PO7, PO8, P7, and P8; P3 [300–470] ms time window of the electrodes O1, O2, Oz, PO3, POz, PO4, P3, P1, Pz, P2, and P4; early LPP [450–750] ms time window of electrodes CP1, CPz, CP2, P1, Pz, and P2; and late LPP [750–1300] ms time window of FC3, FC1, FC2, F1, Fz, F2, and FC4. The percentage of participants whose peak occurred within these time windows and ROIs is as follows: 94.4% for P1, 92.5% for P3, 90.6% for early LPP, and 75.5% for late LPP. Then the grand ERP for each condition (Crave, Resist) was computed by averaging across all participants. ERP analyses were conducted in EEGLAB v2021.0 (Delorme and Makeig, 2004).

The normality of residuals was assessed using the Anderson–Darling test (ad.test, package nortest) and visual inspection of quantile–quantile plots (qqnorm, qqline). Comparisons between the Crave and Resist conditions were performed with paired t-tests when the normality assumption was met, or with Wilcoxon signed-rank tests otherwise. Cohen’s d was performed to estimate effect size when the normality assumption was met, or with the correlation coefficients r (wilcox_effsize, package rstatix) otherwise. We entered ERP measures and scores of urinary cotinine levels, average number of cigarettes smoked per day, FTND, as BAI and BIS into Pearson’s r correlations, controlling for age. Correlations were bootstrapped 1000 times for robustness (95% confidence intervals). Statistical significance was set at p ⩽ 0.05. Corrected p-values were Bonferroni-adjusted for multiple comparisons (based on the number of comparisons for each analysis, e.g., if there were 20 comparisons, the uncorrected p-value was multiplied by 20 to obtain the adjusted p-value). Statistical analyses were conducted in R v4.5.1 (R Core Team, 2025) and RStudio v2025.09.0+387 (Posit Team, 2025).

Results

Craving ratings on VAS during the cue-provoked paradigm with Crave and Resist conditions

Participants’ craving ratings were different between the Crave and Resist conditions (t(51) = 5.19, p = 0.0000028, Cohen’s d = 0.73), indicating lower craving ratings during the Resist than the Crave condition, as shown in Figure 1. Specifically, the mean craving ratings measured on VAS scales (converted to a 0–100 scale) were 56.1 (SD = 17.4) during the Crave condition and 45.6 (SD = 17.3) during the Resist condition.

Figure 1.

Craving ratings during the Resist and Crave conditions in a cue-provoked paradigm.

ERPs during the cue-provoked paradigm with Crave and Resist conditions

There were differences in ERPs between the Crave and Resist conditions for P3 (t(52) = 2.74, uncorrected p = 0.0084518, corrected p = 0.0338071, Cohen’s d = 0.38; Figure 2) and early LPP (W = 1007, z = 2.58, uncorrected p = 0.0099906, corrected p = 0.0399624, r = .35; Figure 3). These ERPs were smaller in the Resist than in the Crave condition. P1 and late LPP were also smaller in the Resist than in the Crave condition but were not statistically different (P1: t(52) = 2.31, uncorrected p = 0.0250313, corrected p = 0.1001252; Cohen’s d = 0.32; Figure 4; late LPP: t(52) = 2.29, uncorrected p = 0.0260429, corrected p = 0.1041716, Cohen’s d = 0.31; Figure 5). We then examined whether these ERP amplitude decreases could be explained by an effect of time. We entered amplitudes across both conditions (Combined) into a linear mixed-effects (LME) model without autoregressive (AR) correlation, and for each condition (Crave, Resist) into separate LME models with AR(1). For P1 and the early LPP, there were no significant decreases (P1: Combined, slope = −0.0015 µV/trial, p = 0.354, Crave, slope = 0.0106 µV/trial, p = 0.019, Resist, slope = 0.0107 µV/trial, p = 0.030; early LPP: Combined, slope = −0.0006 µV/trial, p = 0.695, Crave, slope = 0.0209 µV/trial, p < 0.001, Resist, slope = 0.0099 µV/trial, p = 0.020). For P3, there were significant decreases when combining conditions and during the Crave condition, but not during the Resist one (Combined, slope = −0.0034 µV/trial, p = 0.031, Crave: slope = 0.0122 µV/trial, p = 0.011, Resist slope = 0.0021 µV/trial, p = 0.657). Finally, for the late LPP, there was a significant decrease when combining conditions, but not within conditions (Combined, slope = −0.0047 µV/trial, p = 0.007, Crave: slope = −0.0059 µV/trial, p = 0.270, Resist, slope = 0.0006 µV/trial, p = 0.912).

Figure 2.

P3 amplitudes during the Crave and Resist conditions in a cue-provoked paradigm. (a) Violin plots for Crave and Resist conditions showing the distribution of mean P3 amplitudes (µV). Dots represent individual participants connected across conditions. A significant difference between conditions is indicated by *p < 0.05. (b) Grand average ERP waveforms time-locked to cue onset at ROI electrodes are shown in the scalp inset in the upper-right corner of the graph. Crave (black) and Resist (grey) conditions are displayed with shaded areas representing ± 1 SEM, corrected for within-subject variability (Morey, 2008). The dashed line illustrates the difference wave (Resist–Crave), with shaded areas indicating the 95% CI. The dotted-line box indicates the 300–470 ms time window used to calculate the mean amplitude and conduct statistical analysis. Of note, the time window of P3 ([300–470] ms) is unbalanced (more time on the right side of the peak than the left one is captured); however, this was the window that captures the greatest number of participants.

Figure 3.

Early LPP amplitudes during the Crave and Resist conditions in a cue-provoked paradigm. (a) Violin plots for Crave and Resist conditions showing the distribution of mean early LPP amplitudes (µV). Dots represent individual participants connected across conditions. A significant difference between conditions is indicated by *p < 0.05. (b) Grand average ERP waveforms time-locked to cue onset at ROI electrodes are shown in the scalp inset in the upper-right corner of the graph. Crave (black) and Resist (grey) conditions are displayed with shaded areas representing ±1 SEM, corrected for within-subject variability (Morey, 2008). The dashed line illustrates the difference wave (Resist–Crave), with shaded areas indicating the 95% CI. The dotted-line box indicates the 450–750 ms time window used to calculate the mean amplitude and conduct statistical analysis. The waveform of this ROI (CP1, CPz, CP2, P1, Pz, and P2 electrodes) is noisy soon after time zero, especially during the Crave condition. We inspected individual data per participant and did not identify a subset of data that could explain the source of noise. This may reflect some non-stimulus-evoked activity created by maxillary muscle artifacts, non-stimulus-evoked residual neuronal activity (e.g., sustained attention), and/or activity related to preparatory processes, particularly seen during the Crave condition.

Figure 4.

P1 amplitudes during the Resist and Crave conditions in a cue-provoked paradigm. (a) Violin plots for Crave and Resist conditions showing the distribution of mean P1 amplitudes (µV). Dots represent individual participants connected across conditions. No significant difference was observed (ns, p > 0.05). (b) Grand average ERP waveforms time-locked to cue onset at ROI electrodes are shown in the scalp inset in the upper-right corner of the graph. Crave (black) and Resist (grey) conditions are displayed with shaded areas representing ±1 SEM, corrected for within-subject variability (Morey, 2008). The dashed line illustrates the difference wave (Resist–Crave), with shaded areas indicating the 95% CI. The dotted-line box indicates the 100–200 ms time window used to calculate the mean amplitude and to conduct statistical analysis.

Figure 5.

Late LPP amplitudes during the Crave and Resist conditions in a cue-provoked paradigm. (a) Violin plots for Crave and Resist conditions showing the distribution of mean late LPP amplitudes (µV). Dots represent individual participants connected across conditions. No significant difference was observed (ns, p > 0.05). (b) Grand average ERP waveforms time-locked to cue onset at ROI electrodes are shown in the scalp inset in the lower right corner of the graph. Crave (black) and Resist (grey) conditions are displayed with shaded areas representing ±1 SEM, corrected for within-subject variability (Morey, 2008). The dashed line illustrates the difference wave (Resist–Crave), with shaded areas indicating the 95% CI. The dotted-line box indicates the 750–1300 ms time window used to calculate the mean amplitude and conduct statistical analysis.

Correlations between ERP during the cue-provoked paradigm with Crave and Resist conditions and patients’ characteristics

The P3 amplitude during the Resist and Crave conditions positively correlated with anxiety, but these correlations did not survive multiple comparisons (Resist: r = 0.348, uncorrected p = 0.0114448, corrected p = 0.2746759; Crave: r = 0.316, uncorrected p = 0.0226660, corrected p = 0.5439846). There were no other correlations between ERP measures and patients’ characteristics (all uncorrected p > 0.05).

Discussion

This study investigated ERPs while adults with TUD were instructed to resist craving as compared to letting themselves crave in a cue-provoked paradigm. Results revealed that participant’s craving ratings, as well as ERPs of P3 and early LPP were smaller during the Resist than the Crave conditions. This was not observed for the earlier ERP (P1) or the later ERP (late LPP).

The P3 amplitude was smaller during the Resist as compared to the Crave condition. It has long been proposed that the P3 represents the motivational salience of smoking cues (e.g., Warren and McDonough, 1999), that is, the smoking-cue-elicited P3 indexes an approach-oriented incentive motivational state along with changes in craving level. For instance, smoking cue-elicited P3 positively (e.g., Littel and Franken, 2007; Piasecki et al., 2017) but also negatively correlated with craving reactivity (Littel and Franken, 2010). Furthermore, smokers who did not wish to quit displayed diminished P3 when they were not craving, but enhanced P3 when they were craving; whereas this pattern was not observed in smokers who wished to quit (Donohue et al., 2016). Such seemingly discrepant findings may be partially explained by data indicating that the P3 amplitude is sensitive to the motivational significance of the eliciting cue, regardless of the emotional valence (Briggs and Martin, 2008, 2009; Schupp et al., 2000; Weinberg and Hajcak, 2010). Thus, the P3 amplitude might depend on how a given smoker perceives smoking-related stimuli (of positive or negative valence), how much they wish to quit, and how motivated they are to resist craving. Here, participants were treatment seekers and in the Preparation stage of quitting smoking, as measured by the Prochaska level of readiness questionnaire (Prochaska and DiClemente, 1982). They might have perceived smoking-related stimuli as more negative than positive during the Resist condition than during the Crave condition, which may reflect the observed differences in P3 amplitude.

The early LPP amplitude was also smaller during the Resist as compared to the Crave condition. Early LPP is known to reflect attention toward motivationally salient stimuli (e.g., Hajcak et al., 2009). In smokers, larger early LPP was elicited during observation of smoking-related stimuli than neutral stimuli (Littel and Franken, 2011; Nickel et al., 2023). Notably, the LPP amplitude can be influenced by reappraisal of stimuli. For instance, early LPP was enhanced when smokers imagined how pleasant it would be to smoke the depicted cigarettes, but did not differ when they reappraised (distraction, rational) these cigarettes as compared to when they passively observed them (Littel and Franken, 2011). Also, the LPP amplitude was modulated (increased or decreased) by reappraisal of the arousing content of emotional stimuli. As an example, individuals displayed smaller LPP when they directed their attention to less arousing than highly arousing content of negative stimuli (e.g., Dunning and Hajcak, 2009). Our results may indicate that smokers decreased their motivated attention to the smoking-related stimuli during the Resist condition compared to the Crave condition.

There were no significant correlations between the studied ERPs and patients’ tobacco consumption or dependence level. This is in line with most previous work that did not observe correlations between ERPs and the number of cigarettes smoked per day, duration of cigarette smoking (Anokhin et al., 2000; Bloom et al., 2013; Guney et al., 2009; Hedges and Bennett, 2014; Piasecki et al., 2017), or dependence level (e.g., Bloom et al., 2013; Littel and Franken, 2010; Piasecki et al., 2017). It has been suggested that such a lack of correlation may be attributable to the selected inclusion criteria (e.g., consumption of ⩾10 cigarettes per day) that restrict the full continuum of smoking and dependence level (e.g., Robinson et al., 2016).

We also observed that among the 53 participants, 10 of them reported a greater level of craving during the Resist condition as compared to the Crave condition. The only factor that was notable in this sample of 10 participants was sex: they were all men. We did not observe differences in age, urinary cotinine levels, average number of cigarettes smoked per day in the last year, FTND, BAI, or BIS scores in this sample.

It is noteworthy to take into consideration that the order of the Crave and Resist conditions was fixed. We used this to maximize the induction of craving before smokers underwent the Resist condition. This order also led to a longer time since participants had smoked during the Resist than the Crave conditions. Furthermore, from an ethical standpoint, it is better to present the resisting condition after the craving one to limit potential craving induced at the end of the experiment. One may argue that ERPs are generally smaller for stimuli that are not novel. There were no time effects for the early LPP, but we cannot rule out that a smaller P3 might not have been in part due to an order effect. Previous works also observed smaller P3 and early LPP amplitudes when participants were instructed to regulate their food craving as compared with control conditions (e.g., sensory reward), and these effects were not attributed to order effects (Svaldi et al., 2015; Meule et al., 2013).

The current study has limitations that should be acknowledged. Future studies should include ratings on the studied stimuli (e.g., valence, arousal) prior to the cue-provoked paradigm because smokers might not perceive depicted cigarettes in the same way. Also, the use of an eye tracker would contribute to explore whether smokers look at the same components of the depicted cigarettes during the Crave and Resist conditions. The inclusion of non-smoking related stimuli (e.g., food-related) would refine the specificity of ERPs differences related to craving and resisting craving. We did not ask smokers to report their strategies to resist craving. Instead, we used the same instructions as in Hartwell et al. (2011): “allow yourself to crave when you see the smoking related pictures” and “resist the urge to smoke when you see the smoking pictures by any means you find helpful.” The strategies used by their smokers were distraction (40%), contemplating the adverse effects of smoking (22%) and the benefits of quitting (18%). It would be interesting to see whether the strategy preferred by a given smoker influences ERP amplitude (e.g., whether those who choose distraction would show lowered earlier than later ERP, whether those who choose to contemplate the adverse effects of smoking would show smaller later than earlier ERP). Finally, participants did not consume caffeine at least 1.5 hours before the experiment, but 45 out of the 53 participants reported taking caffeine during the day, before the experiment. Caffeine intake has been related to larger P3 amplitude, and its peak concentration is 40–50 minutes after intake (e.g., Wicht et al. 2022). We explored the potential influence of caffeine on our ERP amplitudes and did not find any ERP that would be considered as outliers (+/−3 SDs); however, we cannot rule out a potential influence of caffeine on the observed larger ERPs during the Crave than the Resist condition.

Conclusions

In sum, findings from our work showed that smokers, when instructed to resist craving as compared to crave, displayed lowered ERP amplitude, within the same time windows (P3 or early LPP), while presented with cigarette stimuli, whereas earlier and later ERP (P1 and late LPP) were less influenced by such instructions. These results suggest suppressed appetitive motivation and attention while resisting craving. This work contributes to characterize brain activity related to resisting craving (e.g., Hartwell et al., 2011; Nickel et al., 2023). A better understanding and characterization of the brain substrates related to resisting craving will contribute to develop new therapeutic avenues for smoking cessation, such as the use of transcranial magnetic stimulation to modulate ERPs, especially the late LPP amplitude.