What evidence supports synaptic strengthening during sleep, and how does it reconcile with downscaling findings?

Why strengthening and weakening can both be true

The apparent contradiction comes from treating sleep as one uniform state. In fact, sleep is made up of distinct phases, and the balance of strengthening and weakening shifts with them. One review describes sleep as coordinating both strengthening and weakening of connections through an interplay of NREM (deep, slow-wave) and REM sleep, with the two phases contributing differently to memory processing [1]. Another line of work reaches a similar conclusion: across sleep, either strengthening or weakening can occur, depending on the circuit and the phase [4].

The key idea is that these processes serve different jobs. Strengthening stabilizes and stores what was learned; weakening prevents the system from accumulating too much excitation. A review of offline plasticity describes potentiation and depression of synaptic transmission being induced in parallel during sleep, which the authors call "offline synaptic plasticity" to distinguish it from the changes that happen right after a learning event [2]. In that framing, the two directions are not in conflict; they are two outputs of the same overnight processing.

How strengthening is thought to work: replay

The strongest case for strengthening rests on replay. Neuronal activity patterns from the awake period are repeated in the hippocampus during sleep, and this replay is thought to coordinate reactivation across the brain, leading to memory consolidation [2]. When that replay is disrupted, long-term memory formation is blocked, which is consistent with replay doing real work rather than being incidental [2]. The proposed mechanism is that replay triggers plastic changes in synaptic transmission across multiple brain regions, providing a cellular substrate for memory [2].

At the systems level, this fits with how deep sleep is organized. During NREM sleep, the coupling of slow oscillations, spindles, and sharp-wave ripples is described as facilitating the transfer of memory representations from the hippocampus to the cortex [1]. That transfer is a plausible route by which a replayed pattern becomes a more permanent connection. It is worth keeping the language careful here: these are proposed mechanisms and consistent findings, not a fully traced causal chain, and one review notes that the precise relationship between systems-level reorganization and local synaptic refinement remains unclear, partly because of technical limits [1].

How downscaling works, and why it does not erase memories

Downscaling is the sleep-time trimming of connections, and the evidence suggests it is more selective than a blanket reset. In mice during slow-wave-sleep-like activity, researchers showed an input-specific downscaling of subthreshold synaptic inputs, meaning weakening was targeted to particular inputs rather than applied uniformly [5]. That selectivity matters: a mechanism that scaled everything down equally would threaten the memories just strengthened, whereas input-specific weakening can spare the connections that carry meaningful signals.

The homeostatic logic is that waking life strengthens connections broadly, and sleep needs to bring overall activity back toward a stable range. Work in a fruit-fly model supports this link between downscaling and sleep regulation: when a protein called APP was overexpressed, the flies showed defects in both synaptic downscaling and the sleep rebound that normally follows sleep deprivation, and both defects could be restored by inhibiting a calcium-signaling molecule called IP3R [3]. That study is about a disease model and does not address strengthening or consolidation, so it speaks to the downscaling side of the ledger rather than the memory-storage side. Its value is showing that downscaling and sleep homeostasis share molecular machinery, which is consistent with downscaling being a regulated sleep function rather than passive decay [3].

What is still unresolved about the balance

The reconciliation is conceptually clean, but the details are not settled. One open question is timing and targeting: how the brain decides which synapses to strengthen and which to weaken within the same night is not established. The reviews describe the balance as coordinated between sleep phases [1][4], but the rules that assign a given connection to one fate or the other are not pinned down. Another gap is scale: linking what happens at individual synapses to the large-scale reorganization of memory across brain regions remains difficult to observe directly, and technical limitations are explicitly cited as a barrier [1].

There is also a question of whether the two processes are truly simultaneous or simply both present across a night. The evidence supports both occurring during sleep [2][4], and it supports downscaling being input-specific [5], but that does not yet establish a single unified mechanism. A reasonable reading of the current evidence is that strengthening and downscaling are complementary and jointly managed, with the precise division of labor still an active question rather than a solved one [1][2].

Sources used in this answer

Related research questions