Sleep as a Structural Imperative: A Systems-Level Model for Synaptic Maintenance in Plastic Brains

Eric Veien, Ph.D., Independent Researcher

eric.veien@gmail.com

Abstract

Sleep remains an evolutionary enigma despite decades of research. We propose the Sleep-Phase Induced Network maintenance (SPIN) model: a systems-level framework positing that sleep, especially slow-wave sleep (SWS), is a structural necessity for maintaining synaptic integrity in plastic brains. Without coordinated activity, synapses decay passively. Waking experience and REM sleep selectively reinforce meaningful connections, while SWS provides global synchrony that maintains the entire network. This interplay naturally explains memory stability, sparse coding, aging-related declines in plasticity, and developmental critical periods. SPIN also inspires robust continual learning strategies for artificial neural networks. We outline testable predictions and therapeutic implications, reframing sleep as an evolutionary requirement for preserving network architecture.

Introduction

Decades of intensive research have underscored the critical role of sleep in various aspects of brain function, from mood regulation to cognitive performance. Yet, the question of its deep, evolutionary conservation across virtually all animals with complex nervous systems—why organisms risk predation and forgo foraging to enter a state of profound vulnerability—remains fundamentally unanswered. While various theories have proposed roles such as energy conservation [1], synaptic homeostasis [2], and memory consolidation [3], none alone provides a comprehensive, unifying explanation for sleep's undeniable biological necessity and its unwavering evolutionary pressure.

Here, we present the Sleep-Phase Induced Network maintenance (SPIN) model [Figure 1], a unifying systems-level framework that posits sleep, particularly slow-wave sleep (SWS), as an unavoidable structural imperative for any plastic, memory-bearing brain. Our model fundamentally shifts the perspective from sleep being merely beneficial to being a critical, active process essential for maintaining the very architecture of neural networks.

Unlike the prevailing Synaptic Homeostasis Hypothesis (SHY) [2], which primarily emphasizes a global downscaling of synaptic weights for energy efficiency and to prevent saturation, the SPIN model focuses on the active, global reinforcement critically needed to preserve weak yet meaningful synapses from intrinsic decay [2]. It complements glymphatic clearance hypotheses by embedding waste removal within a broader, more fundamental process of structural preservation [17, 18]. Furthermore, SPIN expands upon active systems consolidation theories by providing a compelling, systems-level rationale for the co-evolution and necessary interplay of both selective memory replay (via waking experiences and REM sleep) and global synaptic synchrony (during SWS) [3, 9, 11].

The core tenet of the SPIN model is that synaptic connections are inherently unstable, subject to a continuous, passive decay process unless actively maintained [6, 7, 8]. During wakefulness, neural activity locally reinforces frequently used connections, but those encoding less accessed or long-term memories are vulnerable to loss. We propose that SWS provides a global, time-locked opportunity for widespread cortical synchrony [9, 12, 13], which effectively reactivates and thereby preserves even these weak or rarely used connections from degradation. Complementing this, REM sleep serves a vital role in selectively reactivating salient memory traces—especially those that are emotionally charged or novel—effectively "tagging" them for prioritized structural reinforcement during subsequent SWS cycles [3, 11, 21].

This dynamic interplay between intrinsic decay and periodic, coordinated reinforcement leads to several profound and testable consequences: it naturally explains long-term memory preservation [3], and accounts for the observed decline in plasticity with age [4]. It also provides a parsimonious explanation for the natural emergence of sparse neural network architectures [14, 15, 19], as only functionally relevant connections are maintained over time. Beyond fundamental neuroscience, the SPIN model offers crucial insights for the design of robust artificial neural systems, where challenges in continual learning and maintaining sparse representations present striking parallels to biological memory stability [19]. By integrating diverse phenomena under a single, cohesive structural imperative, the SPIN model offers a powerful new lens through which to understand the pervasive mystery of sleep.

Theory / Model Description

1. Intrinsic Synaptic Instability

Cortical neurons are constantly active, even in the absence of external stimuli [5]. This spontaneous activity allows neurons to "probe" their environment through random firing and dendritic exploration, forming transient connections. When two or more neurons repeatedly fire together, the resulting synapses are strengthened [6]. However, if shared activity ceases, these connections begin to decay passively. This decay is not pathological; it is the default state of unused or infrequently used synapses in a dynamic neural network. Experimental evidence supports the idea that synaptic strength is dynamic and that weakening occurs in the absence of reinforcement [7,8].

2. Global Synaptic Reinforcement by SWS

SWS is characterized by large-amplitude, low-frequency oscillations that sweep across the cortex [9]. These global waves synchronize activity across vast neuronal populations, including those that may not have fired together recently. This synchronized firing allows otherwise weak or decaying synapses to be co-activated and reinforced. In this way, SWS serves as a global maintenance mechanism, rescuing distributed memory traces from degradation without needing to reactivate specific experiences. Cortical reactivation patterns during SWS correlate with memory performance and consolidation [10,11]. Moreover, stimulation synchronized with SWS enhances memory retention [12], suggesting that global rhythmic activity serves a functional role in synaptic preservation. This idea aligns with extensive work demonstrating that network oscillations organize information flow and synaptic interactions across scales [13].

We can formalize this mechanism using a minimal model. In cortical networks, synaptic strength evolves over time as:

ds/dt = –λs + R(t)

where λ is the passive decay rate of the synapse, and R(t) is a reinforcement function. During wakefulness and REM, R(t) is driven by selective reactivation of relevant patterns—via external input or internal replay—thus preserving salient synapses. During SWS, R(t) becomes a global, periodic synchronizing signal:

R(t) = ρ Σ δ(t − tn)

Here, δ(t − tn) represents periodic bursts of reinforcement during slow waves, and ρ is the magnitude of the global reinforcement signal. This built-in hierarchy ensures that rarely reactivated synapses decay naturally, while meaningful connections survive through combined selective and global reinforcement.

Crucially, in cortical networks, we propose that the net magnitude of R(t) over time is slightly lower than λ unless reinforced by reactivation during wakefulness or REM dreaming [Figure 1]. This built-in imbalance ensures that unused or dormant synaptic traces decay gradually, serving as a natural "cleanup" process. As a result, the brain avoids storing obsolete or redundant patterns indefinitely. This dynamic explains why rarely accessed long-term memories eventually fade, and why cortical architectures remain sparse, robust, and metabolically efficient—at the inherent cost of memory permanence.

Importantly, in the SPIN model, waking activity and REM sleep serve fundamentally equivalent roles as selective reinforcement mechanisms. Both processes reactivate specific neural patterns that strengthen synapses encoding salient information. Waking reinforcement depends on real-time sensory input and behavior, while REM provides an internally generated rehearsal that replays, recombines, or simulates experiences independent of immediate external stimuli. This equivalence simplifies the reinforcement hierarchy: active experience during wakefulness and internal reactivation during REM both bias which traces are strongest, while SWS delivers the global, non-specific synchrony needed to maintain the structural integrity of the entire cortical network, including weak or dormant connections.

3. Emergence of Sparsity

Because synapses require co-activation or SWS reinforcement to persist, only functionally relevant connections are maintained over time. Unused or redundant synapses decay, leading to sparse neural representations [14]. This sparsity is not imposed through competition or pruning, but rather emerges from passive decay in the absence of coordinated reinforcement. Sparse coding improves signal-to-noise ratios and is metabolically efficient [15]. The SPIN model explains how these coding advantages arise naturally from a decaying, reinforcement-based system. Computational models in both biological and artificial systems have demonstrated that sparse representations can emerge from synaptic competition and decay-like dynamics, further supporting the plausibility of this mechanism [14, 19].

4. Aging and Plasticity

As the brain ages, a growing proportion of synapses become entrenched through years of reinforcement. SWS continues to support these stable circuits, but it becomes increasingly difficult for new connections to gain a foothold. This dynamic explains the age-related decline in cognitive flexibility and learning capacity [4]. Age-related reductions in SWS amplitude and continuity compound the issue [16]. Within the SPIN framework, this tradeoff between memory stability and plasticity is not a defect but an expected outcome of long-term synaptic dynamics. Thus, within SPIN, cognitive aging reflects a natural tradeoff: stable memories accumulate, but network flexibility declines.

This principle also helps explain the existence of developmental critical periods—time windows during which synaptic plasticity is elevated and rapidly reinforced to shape long-term circuitry. As reinforcement mechanisms saturate and network configurations solidify, the capacity for plasticity narrows. SPIN suggests that sleep plays a pivotal role in these windows, helping stabilize emergent patterns before they become integrated into the entrenched architecture. For example, in sensory systems like vision, if coordinated activation is absent during these early stages—such as in cases of prolonged deprivation—the affected synapses receive insufficient reinforcement relative to decay. Over successive sleep cycles, these unused pathways are pruned, resulting in lasting functional deficits even if normal input resumes later [20]. Thus, SPIN frames critical periods not as rigid developmental programs, but as dynamic interactions between activity-dependent reinforcement and the slow erosion of unused circuits.

5. Evolutionary Consequences

Plastic neural systems inherently require a global synchronizing mechanism to preserve structural integrity. Sleep—specifically SWS—evolved alongside increasing brain plasticity to meet this structural need [2]. The global, energy-efficient reinforcement provided during SWS allows large and complex cortical networks to maintain stability without excessive metabolic cost. This capacity for system-wide maintenance removes key constraints on synaptic number and network complexity, leaving the physical limits of skull and braincase size as the primary evolutionary boundary for further cortical expansion. This framework explains why SWS is consistently observed in species with highly plastic brains [17], and how it supported the rapid evolution of large, sophisticated neocortices in mammals, especially primates.

6. Role of REM Sleep and Waking Experience in Memory Selection

While SPIN posits SWS as the essential mechanism for global, non-specific preservation of all synapses, waking experience and REM sleep provide the targeted, selective reinforcement that determines which connections are strengthened enough to persist. During wakefulness, real-world sensory input and behavior-driven activity continually strengthen frequently used connections and reinforce developing circuits, ensuring that actively engaged pathways persist and are preserved by global synchrony during SWS.

Complementing this, REM sleep likely serves as an offline selective reactivation process. It replays emotionally salient, novel, or otherwise behaviorally relevant memory traces, marking them for prioritized reinforcement during subsequent SWS cycles (Figure 1). REM dreams, therefore, are not random byproducts but structured replays that refine and consolidate significant experiences.

Together, this arrangement forms a robust maintenance loop: waking experience and REM both provide targeted reinforcement of important synapses, while SWS supplies the global, non-specific synchrony needed to preserve the entire cortical network. Notably, some memory traces—particularly those strongly marked by emotional salience, such as fear—may be maintained by supplementary mechanisms that operate independently of SWS, contributing to their unusual persistence. This integrated cycle explains how the brain retains meaningful memories and adaptive circuitry within a self-pruning, decaying network and aligns with empirical findings that hippocampal and cortical replay patterns emerge during both REM and quiet wakefulness — reinforcing SPIN’s core principle that selective reactivation occurs in multiple phases and feeds into global SWS stabilization.

Discussion

The SPIN model proposes that sleep, specifically slow-wave sleep, is a biological necessity for maintaining the architecture of plastic neural networks. Rather than viewing sleep as merely advantageous, SPIN reframes it as an evolutionary solution driven by the intrinsic instability of synaptic connections. It differs from the Synaptic Homeostasis Hypothesis (SHY) by focusing on active, global reinforcement instead of downscaling; while SHY highlights synaptic weakening to maintain efficiency, SPIN explains how meaningful but weak connections persist through synchronized reinforcement during SWS. It complements glymphatic clearance models by situating waste removal within a broader structural maintenance function. SPIN also provides a systems-level rationale that connects active systems consolidation with the global network reinforcement that preserves memory-bearing connections.

By recognizing the functional equivalence of waking activity and REM sleep as sources of selective synaptic reinforcement, the SPIN framework unifies memory maintenance across both external and internal experiences. This perspective suggests that REM does not introduce a qualitatively unique mechanism but extends waking reinforcement into an offline cycle, ensuring that emotionally salient or weakly reinforced memories can be internally reactivated even without ongoing sensory input. This reactivation, in turn, biases which traces benefit most from the global stabilizing effect of subsequent SWS cycles. This refined view aligns with empirical evidence of hippocampal and cortical replay occurring during both quiet wakefulness and REM, reinforcing SPIN's central principle: selective reactivation plus global synchronization preserves memory architecture in a naturally decaying neural network.

By unifying processes such as synaptic maintenance, memory consolidation, sparsity, developmental critical periods, and age-related plasticity under the mechanism of periodic global reinforcement, the model integrates phenomena that were previously considered in isolation. It explains why weak yet significant synapses persist, why sparse coding naturally emerges, and why aging brains retain memories but lose flexibility. It accounts for memory distortions and reconfigurations, as seen in false memories, within the same decay-reinforcement cycle, especially when reactivations occur during SWS.

SPIN reframes forgetting as a necessary outcome of efficient network dynamics: the passive synaptic decay rate slightly exceeding average SWS reinforcement ensures unused traces fade, preventing clutter and maintaining relevance. However, this also means that memories not emotionally or behaviorally reinforced are inherently at risk unless periodically revisited.

This framework clarifies why early-life experiences have lasting impact. During critical periods, circuits are highly plastic but must be engaged to stabilize. SPIN explains how lack of input leads underused pathways to decay beyond recovery during sleep, causing permanent deficits. Thus, the timing and richness of experience, coupled with regular network maintenance, shape which connections endure.

Finally, waking experiences strengthen frequently used and developing connections, and REM sleep likely acts as a selector for relevance, reactivating salient memories that SWS then preserves globally. This selective-reactivation and global-reinforcement sequence (schematized in Figure 1) explains the deep evolutionary conservation of both REM and SWS in animals with complex nervous systems, grounding sleep’s multiple roles within a single structural imperative.

Predictions and Implications

The SPIN model makes testable predictions: manipulating synaptic decay rates should directly determine how strongly a network depends on SWS for structural integrity. Closed-loop modulation of global synchrony during SWS should predictably strengthen or weaken memory retention, offering novel therapeutic strategies for aging, neurodegeneration, and neurodevelopmental conditions such as autism. In particular, autism may benefit from sleep-based interventions or treatments that adjust synaptic decay kinetics. Recent studies demonstrate that closed-loop auditory or electrical stimulation can enhance slow oscillations and improve memory consolidation in humans [24, 25], supporting SPIN’s prediction that modulating global synchrony can alter synaptic preservation. Beyond biology, SPIN outlines a principled blueprint for continual learning in artificial systems — especially sparse, energy-efficient architectures. Incorporating periodic, system-wide reactivation phases could protect rarely used representations without costly retraining. By uniting biological necessity with machine learning design, SPIN directly addresses catastrophic forgetting and points toward scalable, self-maintaining networks.

During the preparation of this work the author used ChatGPT 4o as an intellectual partner during the development of SPIN, as well as for writing and editing recursively with the author. After using this tool/service, the author reviewed and edited the content as needed and takes full responsibility for the content of the publication.

References

  1. Zepelin H, Rechtschaffen A. Mammalian sleep, longevity, and energy metabolism. Brain Behav Evol. 1974;10(6):425–470.

  2. Tononi G, Cirelli C. Sleep and synaptic homeostasis: a hypothesis. Brain Res Bull. 2003;62(2):143–150.

  3. Diekelmann S, Born J. The memory function of sleep. Nat Rev Neurosci. 2010;11(2):114–126.

  4. Burke SN, Barnes CA. Neural plasticity in the ageing brain. Nat Rev Neurosci. 2006;7(1):30–40.

  5. Yuste R. From the neuron doctrine to neural networks. Nat Rev Neurosci. 2015;16(8):487–497.

  6. Hebb DO. The Organization of Behavior: A Neuropsychological Theory. Wiley; 1949.

  7. Stettler DD, et al. Axons and synaptic boutons are highly dynamic in adult visual cortex. Neuron. 2006;49(6):877–887.

  8. Loewenstein Y, et al. Multiplicative dynamics underlie the emergence of the log-normal distribution of spine sizes. J Neurosci. 2011;31(26):9481–9488.

  9. Steriade M, et al. Thalamocortical oscillations in the sleeping and aroused brain. Science. 1993;262(5134):679–685.

  10. Wilson MA, McNaughton BL. Reactivation of hippocampal ensemble memories during sleep. Science. 1994;265(5172):676–679.

  11. Rasch B, Born J. About sleep’s role in memory. Physiol Rev. 2013;93(2):681–766.

  12. Marshall L, et al. Boosting slow oscillations during sleep potentiates memory. Nature. 2006;444(7119):610–613.

  13. Buzsáki G. Rhythms of the Brain. Oxford University Press; 2006.

  14. Olshausen BA, Field DJ. Sparse coding of sensory inputs. Curr Opin Neurobiol. 2004;14(4):481–487.

  15. Laughlin SB, Sejnowski TJ. Communication in neuronal networks. Science. 2003;301(5641):1870–1874.

  16. Mander BA, et al. Sleep and Alzheimer’s disease. Trends Neurosci. 2013;36(12):714–724.

  17. Cirelli C, Tononi G. Is sleep essential? PLoS Biol. 2008;6(8):e216.

  18. Xie L, et al. Sleep drives metabolite clearance from the brain. Science. 2013;342(6156):373–377.

  19. Babadi B, Sompolinsky H. Sparseness and expansion in sensory representations. Neuron. 2014;83(5):1213–1226.

  20. Hubel DH, Wiesel TN. The period of susceptibility to the physiological effects of unilateral eye closure in kittens. J Physiol. 1970;206(2):419–436.

  21. Walker MP, Stickgold R. Sleep, memory and plasticity. Annu Rev Psychol. 2006;57:139–166.

  22. Cairney SA, et al. The role of REM sleep in emotional memory. Neurobiol Learn Mem. 2018;155:127–137.

  23. Sara SJ. The locus coeruleus and noradrenergic modulation of cognition. Nat Rev Neurosci. 2009;10(3):211–223.

  24. Ngo HVV, Martinetz T, Born J, Mölle M. (2013). Auditory closed-loop stimulation of the sleep slow oscillation enhances memory. Neuron 78(3):545–553.

  25. Papalambros NA, Santostasi G, Malkani RG, Braun R, Weintraub S, Paller KA, Zee PC. (2017). Acoustic enhancement of sleep slow oscillations and concomitant memory improvement in older adults. Front Hum Neurosci 11:109.

Figure Legend

Figure 1. SPIN model schematic.

This figure illustrates how synaptic strength evolves for two example synapses across sequential cycles of wakefulness, REM sleep, and slow-wave sleep (SWS). Synapse A (green) represents an actively used connection: it is strengthened during wakefulness through real-world experience and further reinforced by selective reactivation during REM sleep. Each SWS phase globally stabilizes both strong and weak connections, preserving the network’s structural integrity. In contrast, Synapse B (red) represents an unused connection: it passively decays during wake and REM phases, receiving only partial stabilization during SWS. This demonstrates that SWS alone slows synaptic decay but cannot fully maintain unused connections without selective reactivation. Phase shading indicates Wake (yellow), REM (orange), and SWS (blue). The model shows how the interplay of selective and global mechanisms preserves meaningful circuits while allowing unused connections to fade, supporting sparse, energy-efficient coding.