Covariance-based analysis of spindle-band EEG during declarative and non-declarative odor cueing in sleep

Sleep supports memory consolidation through the reactivation of neural circuits engaged during learning. Targeted Memory Reactivation (TMR), in which memory-associated sensory cues are presented during sleep, can enhance declarative memory retention. However, the neural signatures supporting odor-cued reactivation remain incompletely characterized.

Here, we analyzed high-density electroencephalography (EEG) recordings from a TMR paradigm designed to dissociate neural responses associated with declarative and non-declarative odor cueing during sleep. EEG epochs were examined across fast- (12.5-16 Hz) and slow-spindle (9-12.5 Hz) frequency bands, channel subsets (all, frontal, central, and posterior), and multiple post-cue time windows (0-2, 0-4, and 0-7 s). Using within-participant machine learning based on Riemannian geometry, we classified EEG epochs elicited by a declarative memory-associated odor (Odor D) vs. vehicle control, and by a non-declarative odor associated with a motor task (Odor M) vs. vehicle.

Decoding performance relative to permutation-derived chance showed condition-dependent patterns across frequency bands, time windows, and channel subsets. Across analyses, decoding tended to be higher in the declarative condition (Odor D) than in the non-declarative condition (Odor M), with the strongest effects observed in central channels. Channel-level contribution analysis further indicated more spatially structured covariance patterns during Odor D over central regions, whereas contributions during Odor M were more diffuse and less consistent. These effects were modest and did not survive correction for multiple comparisons.

These results suggest that declarative odor cueing during sleep is associated with more structured spindle-band EEG patterns than non-declarative cueing, particularly over central channels, although effects were modest and did not survive correction for multiple comparisons. A key finding is a difference in central-channel contribution patterns between conditions, consistent with memory-related neural modulation. These findings also highlight the potential of covariance-based decoding approaches for probing distributed sleep EEG dynamics, warranting further validation in larger samples.