Hierarchical Neural Variability Reveals Adaptive and Maladaptive Mechanisms of Non-Suicidal Self-Injury

Non-suicidal self-injury (NSSI) is a prevalent and clinically urgent condition in adolescence, marked by impairments in cognitive flexibility—a capacity thought to be supported by neural variability. However, most research to date has focused on static brain features, leaving the role of variability largely unexplored. Importantly, variability-based metrics have been emerging as promising biomarkers in psychiatry, offering superior test–retest and intersession reliability compared with traditional static metrics. Leveraging resting-state fMRI data from 160 psychiatric patients with NSSI and 50 psychiatric controls, we examined neural variability at two hierarchical levels—connectivity and network topology—and evaluated their clinical relevance using cross-sectional and longitudinal data. Patients with NSSI consistently exhibited heightened variability at both the local and global level, and these measures demonstrated significant discriminative power in classifying NSSI from controls (AUC = 0.75). The functional implications, however, diverged between two levels. Greater connectivity-level variability was associated with better emotional and attentional functioning and predicted larger reductions in NSSI behaviors over three months, suggesting an adaptive or compensatory role. In contrast, greater topology-level variability was linked to higher impulsivity, and predicted poorer behavioral improvement, reflecting maladaptive instability of large-scale organization. These findings reveal that elevated neural variability in NSSI encompasses both adaptive and dysregulated dynamics: increases in local connectivity variability may promote flexible adaptation, whereas excessive large-scale reconfiguration indicates loss of global control. Together, our findings indicate that disruption of the balance between local flexibility and global stability may constitute a neural mechanism underlying self-regulatory dysfunction in NSSI and highlight variability-based metrics as potential biomarkers and intervention targets.