无删除的降尺度：区分睡眠中的突触降尺度与类淋巴废物清除

Abstract The Synaptic Homeostasis Hypothesis (SHY) correctly identified that NREM sleep triggers proportional synaptic downscaling — but this core insight has been widely misinterpreted as evidence for “synaptic pruning” (deletion). This paper extends SHY by formally distinguishing two functionally independent processes that are conflated under the pruning umbrella: (P1) reversible synaptic downscaling — AMPA receptor endocytosis, spine shrinkage, proportional strength reduction ; and (P2) glymphatic clearance — AQP4-gated convective flushing of metabolic waste from the perisynaptic space. We demonstrate that key evidence cited for pruning can be reinterpreted under the P1+P2 model without additional entities, and propose falsifiable predictions (2×2 factorial AQP4/Homer1a manipulation). P1-downscaled spines enter a metastable state (P3: Metastable Downscaled State, MDS) characterized by AMPAR internalization with retained PSD scaffold, with an estimated 24-48 hour time window before probabilistic commitment toward re-expansion or irreversible degeneration. Spine fate in this window depends on P2 clearance efficacy. We discuss a major potential counter-evidence — the absence of widespread spine loss in AQP4 knockout models — noting this may reflect a measurement gap (most AQP4 studies did not perform longitudinal spine tracking) rather than falsification of the model.