Participants

Through online advertisements and fliers, we recruited a total of 40 male smokers. Nicotine dependence was assessed by DSM-5 and the Fagerström Test for Nicotine Dependence (FTND). Other inclusion criteria were as follows: age between 18–50 years; smoking at least 10 cigarettes per day for 2 years or more; and right-handedness. The exclusion criteria were as follows: discomfort with the EEG experimental environment; receipt of any smoking cessation treatment in the previous 3 months. Owing to the low percentage of female smokers in China (2.1%), only male smokers were recruited. For more details about the recruitment, see the supplementary materials available at https://dx.doi.org/10.1192/bjo.2022.641.

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human participants/patients were approved by the Medical Research Ethics Committee of the First Affiliated Hospital of the University of Science and Technology of China (2022KY-123). All participants signed an informed consent form.

Guided imagery script generation

Personalised scripts describing experiences of cigarette use and neutral relaxation were generated using the Scene Construction Questionnaire (SCQ) during the interview session.^{Reference Sinha6} For smoking-related scripts, participants described a recent situation when they most wanted to smoke and finally smoked a cigarette (e.g. seeing someone else smoking). All situations that contained elements of negative emotions and stress were excluded (e.g. a cigarette after a quarrel). For neutral-related scripts, participants described a recent situation that put them in a calm and peaceful state (e.g. reading a book). All situations that involved cigarettes, friends associated with smoking behaviour or those involving high arousal were excluded. As a manipulation check for situation content, each situation was evaluated twice, first by the researcher and second by two objective independent raters.^{Reference Sinha6} For more details, see the supplementary materials. After participants had described each situation and the researcher had checked the content of the situation, participants were given a list of phrases that describe physiological/bodily sensations and were told to circle all of the physiological/bodily sensations that they would experience in the situation in question.^{Reference Sinha6} For each participant, a total of four smoking-related and four neutral-related descriptions were completed. The descriptions were edited into the final scripts and audio-taped to a standard length of 60 s. Although these scripts contained individual elements and were therefore personalised, they were of uniform format.^{Reference Sinha6}

Imagery training

All participants underwent imagery training, which has been found to effectively improve imagery performance.^{Reference Sinha6} Participants were asked to practise imagining five training scripts, which were derived from the Imagery Script Development Procedures Manual and were audio-taped to a standard length of 60 s. At the end of each training trial, participants were asked to rate vividness on a 0–10 visual analogue scale. The result is shown in Supplementary Fig. 1. The baseline imagery ability of the participants was also assessed (see the supplementary materials).

Experimental procedure

The experimental procedure is summarised in Fig. 1. During the interview session, participants provided basic demographic information, including age, years of education, number of cigarettes smoked per day and years of smoking, completed the questionnaires (DSM-5 nicotine dependence, FTND and SCQ) and underwent imagery training. During the EEG experimental session, all participants completed the guided imagery cue reactivity task. We used a block design in the EEG study. There were two blocks (smoking and neutral block) and the four trials in each block were presented in pseudorandom order. Each trial contained a baseline period (5–15 s), a read-imagery period (60 s) and a quiet-imagery period (30 s). During the baseline period, participants closed their eyes and waited for the audio-recording. During the read-imagery period, participants were asked to imagine themselves in the situation being described. During the quiet-imagery period, participants continued to imagine until they were told to stop. Participants then opened their eyes and completed ratings for cigarette craving and vividness on a 0–10 visual analogue scale. To avoid carryover effects, that is, interference caused by smoking-related cues, the smoking block was always presented after the neutral block. Before the EEG experimental session participants received a telephone call instructing them not to smoke for at least 2 h before the session began. Details on ensuring and validating smoking deprivation before the EEG experimental session are given in the supplementary materials.

Fig. 1

Experimental procedure in our study.

FTND, Fagerström Test for Nicotine Dependence; SCQ, Scene Construction Questionnaire; EEG, electroencephalogram. During the read-imagery period, participants were asked to imagine themselves in the situation being described; during the quiet-imagery period, participants continued to imagine until they were told to stop.

EEG acquisition

The EEG data were recorded using a SynAmps RT amplifier (NeuroScan, Charlotte, NC, U.S.). According to the international 10–20 system, 64 electrodes were placed on the participant's scalp. Electrooculogram (EOG) signals and mastoid signals were recorded using four electrodes (HEOL, HEOR, VEOL and VEOU) and two electrodes (M₁ and M₂) respectively. The tip of the participant's nose was used as the location for the reference electrode. To prevent interference from electromagnetic noise, AFz was selected as the ground electrode. The impedance of all electrodes was kept under 5 kΩ. The raw EEG signals were digitised at a sampling rate of 1000 Hz.

EEG data preprocessing

Preprocessing of the raw EEG data was conducted using EEGLAB toolbox (version 14_1_1b) in MATLAB 2020a for Windows (Mathwork Inc., Natick, USA). The raw EEG data were bandpass filtered between 0.1 and 80 Hz, epoched from −1 to 90 s relative to the beginning of the cue onset, and baseline corrected using the interval from −1 to 0 s; 1 to 89 s of each epoch was selected and a conventional recursive least squares algorithm was used to correct for blink artefacts.^{Reference Bu, Liu, Gou, Gan, Cheng and Liu27} Then, the EEG data were segmented into 1 s epochs, and epochs containing amplitude changes exceeding ±100 mV were rejected. For the EEG microstate analysis, data were downsampled at 250 Hz, bandpass filtered between 2 and 20 Hz and re-referenced to the average reference.

EEG microstate analysis

First, we calculated the global field power (GFP), which represents the standard deviation of the potentials of all electrodes in the topographical map at each time point, by using the following equation:

$$\matrix{ {{\rm GFP} = \sqrt {\displaystyle{{\mathop \sum \nolimits_{i = 1}^N {( {\mu_i-\bar{\mu }} ) }^2} \over N}} } \cr } $$

where N is the number of electrodes, μ _i is the potential of electrode i at a given time point and $\bar{\mu }$ is the mean of the potentials of all electrodes. In the GFP time series, troughs represent a transition from one type of topographical map to another.^{Reference Michel and Koenig14} The topographical maps at the GFP peaks had the highest signal-to-noise ratio and were selected as the initial prototype maps (Supplementary Fig. 2(a)).^{Reference Schumacher, Peraza, Firbank, Thomas, Kaiser and Gallagher28}

Second, at the individual level, initial prototype maps were submitted to k-means clustering, with k ranging from 2 to 8. The Krzanowski–Lai (KL) criterion was used to determine the optimal number of microstate classes for further analysis. The optimal k was chosen as the k corresponding to the second KL maximum value.^{Reference Michel and Koenig14} Consistent with previous studies, the optimal number of microstate classes was four in our study. We thus obtained individual topographical maps for each participant in the two conditions. For each condition, the individual topographical maps were then averaged across participants using a permutation algorithm.^{Reference Schumacher, Peraza, Firbank, Thomas, Kaiser and Gallagher28} We thus obtained two sets of topographical maps at the condition level (Supplementary Fig. 2(b)). After that, the overall topographical maps were calculated by averaging the condition-level maps.

Third, the overall topographical maps were fit back to the original GFP peaks for each participant. Based on the maximal spatial correlation between topographies, each GFP peak was identified as one of four classes (Supplementary Fig. 2(c)). Microstate parameters (duration, occurrence and contribution) were then calculated for each class. Specifically, duration represents the average length of time for each class, occurrence represents the total number of each class in 1 s and contribution represents the percentage of the time covered by each class.^{Reference Michel and Koenig14}

Statistical analyses

Paired t-testing was conducted to compare the differences between conditions for craving and vividness ratings. Repeated measures analysis of variance (rm-ANOVA) was performed with the factors ‘condition’ (smoking versus neutral) and ‘microstate class’ (classes A, B, C and D) for duration, occurrence and contribution. Post hoc paired t-testing was performed for each class only when the interaction effect was significant. To prevent type I error caused by the multiple-comparison problem, the false discovery rate (FDR) was used to correct significance. Changes in behavioural (e.g. craving ratings) and EEG data (e.g. microstate parameters) were measured by absolute changes (smoking minus neutral) only when significant differences were observed between conditions. All correlation analyses were measured using Pearson's correlation coefficient. The mediation analysis was performed using the bootstrapped method in the PROCESS macro for SPSS for Windows.^{Reference Bolin29} Two-tailed P-values <0.05 were considered significant.