Predicting nicotine dependence severity via state-dependent topological features of dynamic brain networks.

BACKGROUND

The dynamics of large-scale brain networks in tobacco use disorder (TUD) remain poorly understood. Dynamic functional connectivity (dFC) captures time-varying interactions among brain regions and provides a framework for examining network-level alterations in TUD. This study investigated dFC states and their topological properties to identify neural correlates of nicotine dependence.

METHODS

Resting-state functional magnetic resonance imaging (rs-fMRI) data were obtained from 46 individuals with TUD and 48 healthy controls (HCs). Regional time series were extracted using the Dosenbach-160 atlas and assigned to six large-scale functional networks. dFC states were identified using a sliding-window approach with k-means clustering, with the optimal number of clusters (k = 2) determined by the elbow criterion. State-specific graph-theoretical metrics were calculated, and inter-network connectivity differences were assessed using Network-Based Statistics (NBS). Multivariate support vector regression (SVR) with leave-one-out cross-validation (LOOCV) was used to predict nicotine dependence severity, measured by the Fagerström Test for Nicotine Dependence (FTND). Exploratory analyses examined associations between dynamic network measures and Reasons for Smoking Questionnaire (RRSQ) scores.

RESULTS

Two dFC states were identified: a relatively weakly connected State 1 and a relatively strongly connected State 2. Compared with HCs, individuals with TUD spent more time in State 1, showed longer dwell time, and exhibited fewer state transitions. In State 1, TUD patients showed a more randomized network topology, reflected by reduced clustering coefficient and increased global efficiency. NBS analysis revealed reduced connectivity within the sensorimotor network and weakened coupling between sensorimotor, cognitive control, and default mode networks. SVR analysis demonstrated that State 1 topological metrics significantly predicted FTND scores (r = 0.507, permutation p = 0.030). Furthermore, correlation analyses showed that the fractional occupancy of State 1 was positively associated with the pharmacological dimension of the RRSQ (PFDR = 0.028), suggesting a link between physiological nicotine dependence and increased occupancy of this weakly connected state.

CONCLUSION

TUD is characterized by constrained brain network dynamics and prolonged occupancy of a hyper-integrated, metabolically costly baseline state. This maladaptive pattern is associated with the pharmacological dimension of smoking motivation, and its state-specific topological features predict individual nicotine dependence severity. Taken together, these findings link network efficiency imbalance to physiological dependence and suggest that dynamic network metrics may serve as potential neuroimaging biomarkers for assessing mechanisms and clinical severity in TUD.