Not all smokers are alike: the hidden cost of sustained attention during nicotine abstinence

The study addresses why cognitive manifestations of Nicotine Withdrawal Syndrome (NWS) vary markedly among smokers and whether heterogeneity can be objectively partitioned using concurrent stressors. Prior work shows variability in withdrawal symptoms and treatment response, implying endophenotypic diversity. The authors propose that combining a cognitive stressor (high-demand Parametric Flanker Task; PFT) with acute nicotine abstinence (a physiological stressor) will expose latent subgroups of smokers with distinct behavioral and neurobiological responses. They hypothesized that subgroups would differ in cognitive performance, task-evoked brain activation, and functional connectivity (FC) patterns precipitated by abstinence versus satiety.

The introduction synthesizes evidence that NWS symptoms differ in duration, intensity, and phenotype across individuals, and that cessation aid efficacy (bupropion, varenicline, nicotine replacement therapy) is variable. Genetically informed biomarkers such as the nicotine metabolite ratio (CYP2A6 variation) predict treatment outcomes, reinforcing heterogeneity. The allostatic load framework suggests chronic stress can impair adaptive responses to additional stressors, promoting relapse; combined stress markers (cue reactivity and abstinence) better predict relapse than either alone. Cognitive withdrawal deficits in sustained attention, response inhibition, and working memory are reported, particularly at higher task difficulties, and predict relapse. Transdermal nicotine can alleviate abstinence-induced attentional lapses (errors of omission). Functional connectivity studies show abstinence-related alterations in insula, amygdala, default mode, salience, and executive control networks, and varenicline/nicotine can modulate these circuits. Collectively, these findings motivate a dual-stressor approach to delineate smoker subgroups.

Design: Longitudinal within-subjects study with two MRI sessions per participant: nicotine sated (session 1) and ~48-hour biochemically verified abstinent (session 2). Session order was fixed (sated first) due to the parent cessation protocol; average inter-session interval 75 days (median 28 days). Participants: Fifty-nine right-handed smokers were recruited; 14 were excluded (incomplete behavioral data n=2, scanning errors n=2, abnormal MRI n=2, medical exclusions n=3, excessive motion n=5), yielding 45 participants. Exclusion criteria included psychiatric/neurological disorders or other substance abuse/dependence; all scans required negative urine drug and breath alcohol screens. Measures: During fMRI, participants performed a 25-minute Parametric Flanker Task (PFT) with four cognitive control DEMAND levels (none/low/medium/high) manipulating flanker-target conflict on a trial basis. Behavioral metrics: correct response speed (1/RT), coefficient of variation of correct response speeds (SpdCV), adjusted accuracy (Correct/(Correct+Errors of Commission)), and Errors of Omission (EOm; attentional lapses). Subjective ratings: withdrawal (WSWS), craving (TCQ-SF), and affect (PANAS) collected prior to each scan. Subgrouping: Based on adjusted accuracy in the high DEMAND PFT condition (averaged across states), participants were divided into High Task Performers (HTP; n=21) and Low Task Performers (LTP; n=24). Behavioral analysis: Mixed-design ANOVAs for Speed, SpdCV, and Accuracy with within-subject factors STATE (satiety, abstinence) and DEMAND (none, low, medium, high), and between-subject factor SUBGROUP (HTP, LTP). EOm abstinence effect computed as ΔEOm = abstinent − sated; subgroup differences assessed via Kruskal–Wallis and post-hoc Wilcoxon tests. Subjective differences assessed via A scores (abstinent − sated) with one-sample t-tests and between-group tests (parametric/nonparametric as appropriate). MRI acquisition: Data collected on Siemens Trio (n=35) and Prisma (n=10) scanners. Preprocessing and analysis with AFNI. Two parallel pipelines: (1) Task-evoked activation: Event-related GLM with regressors for DEMAND (none, low, medium, high) crossed with response type (correct, error) and motion parameters; percent signal change estimated via 3dREMLfit. Group analysis used 3dMVM with factors STATE and SUBGROUP, covariates scanner and motion (AFD), focusing on the high DEMAND [correct − errors of commission] contrast (voxelwise p=0.0001; cluster-corrected FWE α≤0.05), yielding 19 clusters for seed definition. (2) Task-regressed functional connectivity: To avoid task-activation confounds, PFT-evoked responses were regressed out using an FIR model (eight-parameter TENT over 14 TRs), and residual time series band-pass filtered (0.001–0.25 Hz) served as proxy resting-state data. Seed-based FC analyses: Each of the 19 task-derived ROIs used as seeds in whole-brain FC models (3dMVM) testing SUBGROUP main effects and SUBGROUP×STATE interactions; voxelwise p=0.001 with cluster extent k=69 (FWE α≤0.05). Significant seed–target pairs were denoted dyads; target ROIs were then used as seeds in secondary FC analyses to map broader circuits. Brain–behavior relationships: Multiple linear regression (3dRegAna) relating ΔFC (abstinent − sated) to ΔEOm, including SUBGROUP and SUBGROUP×ΔEOm terms and AFD motion covariate; six regressions corrected with voxelwise p=0.001 and cluster k=100 (FWE α≤0.05/6). Additional analysis tested correlations between ΔAccuracy and ΔFC; none were significant.

Subgrouping and behavior: Based on high DEMAND accuracy, two subgroups were identified: HTP (n=21) with 88.68% ± 5.19 SD accuracy and LTP (n=24) with 51.04% ± 4.72 SD accuracy, independent of nicotine state. HTPs exhibited greater nicotine dependence (FTND: HTP 5 ± 0.38 vs LTP 3.83 ± 0.35; p=0.03); other demographics and use measures did not differ significantly. Across the cohort, higher DEMAND reduced speed, increased variability, and reduced accuracy; abstinence decreased accuracy and increased EOm. Critically, a SUBGROUP×STATE effect emerged for EOm: HTPs showed greater abstinence-induced increases in EOm than LTPs (Kruskal–Wallis χ²=5.47, p=0.01). Within-group, EOm increased with abstinence for both HTP (W=378, p=8.76e-06) and LTP (W=432, p=9.47e-04). Subjective withdrawal, craving, and affect showed expected abstinence effects but no subgroup differences. Task-evoked activation: HTPs showed larger BOLD responses than LTPs for [correct − errors of commission] in high DEMAND across states, in bilateral insula, dorsal ACC, frontoparietal attentional regions, and right thalamus, consistent with enhanced performance monitoring and attentional control (exemplar ROI shows greater sensitivity to errors vs correct in HTPs). Task-regressed FC: Of 19 seeds, two showed SUBGROUP×STATE interactions: Dyad1: Left Precentral (seed)–Right ventral Insula (target); Dyad2: Left posterior Insula (seed)–Right middle Occipital (target). In both dyads, FC increased during abstinence only in HTPs (no significant change in LTPs). Secondary analyses indicated coactivation between dyads during abstinence in HTPs (R ventral Insula–L posterior Insula; R middle Occipital–R ventral Insula). Brain–behavior relationships: ΔFC in L Precentral–R Orbitofrontal Cortex positively correlated with ΔEOm for HTPs and negatively for LTPs. ΔFC in R ventral Insula–Left Medial/Superior Frontal Gyrus positively correlated with ΔEOm across both subgroups (main effect of ΔEOm). Additional positive ΔFC–ΔEOm relationships from the R middle Occipital seed to regions including R parahippocampal, L middle temporal, and R lingual gyri were observed. No significant correlations were found between ΔAccuracy and ΔFC in dyad circuits, supporting subgrouping based on overall accuracy rather than change-scores.

The dual-stressor approach (cognitive demand plus abstinence) unmasked robust heterogeneity among smokers. HTPs maintained superior selective attention performance (accuracy) across states but incurred a hidden cost during abstinence: increased sustained attention failures (EOm). Neurobiologically, HTPs showed greater engagement of classic performance-monitoring and attentional-control regions (insula, dACC, frontoparietal cortex, thalamus) irrespective of nicotine state, and exhibited abstinence-specific increases in FC within circuits linking precentral cortex with ventral insula and posterior insula with occipital cortex. These salience- and attention-related circuits may support selective attention under stress, but their enhanced coupling during abstinence correlates with more attentional lapses, suggesting a resource trade-off wherein maintaining attentional control during withdrawal compromises sustained attention. Brain–behavior regressions further implicated orbitofrontal and superior/medial frontal regions in linking FC changes to lapses, with subgroup-specific directions of association. The absence of subgroup differences in subjective withdrawal underscores the value of objective behavioral and neural markers. Findings help reconcile inconsistent literature on cognitive withdrawal effects by highlighting subgroup-specific vulnerabilities and suggest that endophenotypic stratification can inform mechanistic understanding and intervention targeting.

Using concurrent cognitive and abstinence stressors, the study identified distinct smoker subgroups with differential susceptibility to abstinence-induced sustained attention deficits and corresponding neural signatures in task-evoked activation and functional connectivity. HTPs, despite better selective attention performance, showed greater abstinence-related attentional lapses and increased FC within salience/attentional circuits, with FC–behavior coupling in orbitofrontal and superior/medial frontal regions. These multi-level markers of heterogeneity may aid in predicting vulnerability to NWS-related cognitive disruptions and enable targeted treatment strategies, such as tailoring nicotine replacement or neuromodulatory interventions to specific subgroups. Future research should validate these endophenotypes prospectively, test whether subgroup-informed interventions improve cessation outcomes, and examine more direct sustained attention paradigms and longitudinal trajectories during cessation.

The PFT primarily indexes attentional control; while it elicited sustained attention lapses, more direct sustained attention tasks might reveal additional subgroup differences. Session order was not counterbalanced (sated always preceded abstinent) due to the parent protocol, potentially introducing order effects. The inter-session interval was variable (mean 75 days, median 28 days). Participants were part of a larger treatment study and were not recruited specifically to maximize heterogeneity. Although task effects and abstinence effects did not directly interact, ecological validity of acute abstinence modeling required fixed timing. Finally, FC analyses used task-regressed residuals as a proxy for resting-state data, which, while mitigating task-activation confounds, is not identical to true rest.