Sleep spindles enhance latent working memory representations

Abstract

Working memory (WM) allows recently encountered information to be maintained over short periods, yet neural mechanisms supporting this short-term accessibility remain debated. Recent work suggests that WM representations can be maintained without persistent neural firing, in latent synaptic states often termed "activity-silent" memory. Such states can be reactivated by brief perturbations of cortical networks ("pinging"), providing a way to probe otherwise hidden representations. If such latent states rely on synaptic plasticity, they may be sensitive to sleep-dependent recalibration of cortical circuits. Here, we test whether sleep spindles, transient thalamo-cortical oscillations linked to synaptic plasticity, shape post-sleep accessibility of WM representations. Thirty participants performed a visual WM task before and after a daytime nap while high-density EEG was recorded. During the post-nap task, brief visual impulses were used to probe latent WM states, enabling multivariate decoding of memorised content. We show that longer spindle duration during prior NREM sleep predicts both improved post-sleep WM performance and higher-fidelity impulse-evoked decoding of item-specific representations, even when controlling for pre-sleep WM ability. These relationships were selective to spindles; slow-oscillation duration and other oscillatory metrics showed no corresponding effects. Together, our findings link sleep spindle dynamics to the neural readout of activity-silent WM and suggest that sleep-dependent synaptic recalibration optimises cortical circuits for subsequent, short-term information processing.