Evidence that Alzheimer’s disease is a disease of competitive synaptic plasticity gone awry

Abstract

Mounting evidence indicates that a physiological function of amyloid β (Aβ) is to mediate neural activity-dependent homeostatic and competitive synaptic plasticity in the brain. I have previously summarized the lines of evidence supporting this hypothesis and highlighted the similarities between Aβ and anti-microbial peptides in mediating cell/synapse competition. In cell competition, anti-microbial peptides deploy a multitude of mechanisms to ensure both self-protection and competitor elimination. Here I review recent studies showing that similar mechanisms are at play in Aβ-mediated synapse competition and perturbations in these mechanisms underpin Alzheimer’s disease. Specifically, I discuss evidence that Aβ and ApoE, two crucial players in Alzheimer’s disease, co-operate in the regulation of synapse competition. Glial ApoE promotes self-protection by increasing the production of trophic monomeric Aβ and inhibiting its assembly into toxic oligomers. Conversely, Aβ oligomers, once assembled, promote the elimination of competitor synapses via direct toxic activity and amplification of “eat-me” signals promoting the elimination of weak synapses. I further summarize evidence that neuronal ApoE may be part of a gene regulatory network that normally promotes competitive plasticity, explaining the selective vulnerability of ApoE expressing neurons in Alzheimer’s brains. Lastly, I discuss evidence that sleep may be key to Aβ-orchestrated plasticity, in which sleep is not only induced by Aβ but is also required for Aβ-mediated plasticity, underlining the link between sleep and Alzheimer’s disease. Together, these results strongly argue that Alzheimer’s disease is a disease of competitive synaptic plasticity gone awry, a novel perspective that may promote Alzheimer’s disease research.

Introduction

Aβ is the main component of amyloid plaques found in the brains of Alzheimer’s disease, a neurodegenerative condition affecting tens of millions of people worldwide. The invariable association of Aβ pathology with Alzheimer’s disease and the discovery that genetic mutations affecting Aβ production predictively result in the development of early-onset Alzheimer’s disease (also known as familial Alzheimer’s disease) strongly indicate that Aβ likely plays a central role in Alzheimer’s disease pathogenesis. Decades of research into the activity and regulation of Aβ also suggest that it plays a physiological role in the healthy brain. Based on many converging lines of evidence from several different fields, I have proposed that a physiological function of Aβ in the brain may be to mediate neural activity-dependent homeostatic and competitive synaptic plasticity, including activity-dependent synapse competition [1]. In this process, Aβ acts not only as a reporter of circuit-level neural activity in the brain but also as an activity level-dependent effector that directly mediates homeostatic and competitive synaptic plasticity. I propose that Aβ remains as monomer-related neuroprotective species when circuit-level activity and Aβ concentration are both low, but forms toxic oligomeric species when high-level circuit activity increases Aβ production and concentration, which then lead to synaptic competition and the weakening or even elimination of weak synapses. In prototypical cell competition, besides the action of anti-microbial peptides the similarities of which to Aβ I have previously reviewed, many other mechanisms are also activated in both the steps of self-protection and the steps of competitor killing [1]. Recent studies show that similar mechanisms, including the regulation of Aβ monomer production and assembly into oligomers by astrocytic ApoE [2–24] and the amplification of “eat-me” signals displayed at weak synapses by Aβ oligomers [25–41], may also be at work during neural activity-dependent axon/synapse competition. These studies also show that sleep is induced by Aβ oligomer injection into the brain and crucial to both homeostatic and competitive synaptic plasticity [42–54], potentially explaining the link between sleep disturbance and Alzheimer’s’ disease. In this article, I will review these findings and discuss how they have shed new light on the physiological and pathological functions of Aβ, ApoE, and other genes in the healthy and the Alzheimer’s disease brain and how they have provided new insight into the regulation of normal brain synaptic plasticity as well as the pathogenesis of Alzheimer’s disease.

Relationship between roles played by ApoE and AICD in competing synapse self-protection and in Alzheimer’s disease

In cell competition, competing cells activating pathways to kill competitors must also at the same time protect themselves from the activity of these same pathways. In the simplest form of cell competition such as anti-microbial peptide-mediated competition, bacterial cells engaging competition are known to up-regulate the expression of the so-called immunity proteins, which, both membrane-bound and secreted, can bind to anti-microbial peptides and impede their aggregation into toxic oligomers in the immediate vicinity of these cells [1]. If Aβ acts similarly in axon/synapse competition, similar mechanisms may also exist for protection against toxic Aβ species. Indeed, a large body of evidence indicates that both ApoE and amyloid precursor protein (APP), likely among others, are involved in self-protection during axon/synapse competition [2–24, 55–67]. For example, studies show that ApoE promotes Aβ degradation and suppresses Aβ oligomerization [2–24] and that APP intracellular cleavage products promote mechanisms that repairs DNA damage induced by reactive oxygen species and neural activity during axon/synapse competition [55–67]. Furthermore, compromised functions in these processes have been linked to increased risk to Alzheimer’s disease. Below, in this section, I will discuss these results and highlight how these findings help unifying our understanding of the functions of Aβ, APP, and ApoE under physiological and pathological conditions.

Regulation of Aβ oligomerization and degradation by ApoE and its role in Alzheimer’s disease

ApoE is an extracellular lipid-binding protein with prominent roles in brain lipid metabolism [68, 69] and the ApoE4 allele is one of the highest genetic risk factors in late-onset Alzheimer’s disease development [70, 71]. Several lines of evidence indicate that ApoE proteins, including those induced in neurons during heightened neural activity, may participate in axon/synapse competition by regulating both neuronal self-protection and competitor elimination (for roles of ApoE in competitor elimination, see discussion in the next section) [72–77]. First, ApoE shows several properties and activities that befit such a role. For example, ApoE expression is induced in cortical neurons by glutamate [72], consistent with its regulation by neural activity. In the brain, while ApoE is well known to be most highly expressed in astrocytes, under high activity conditions such as kainic acid treatment, strong ApoE expression is also induced in neurons [73, 74]. In vitro, ApoE-containing lipoproteins protect neurons from receptor pathway-mediated apoptosis [75]. In vivo, neuronal expression of ApoE3 (but not ApoE4) also protects neurons from excitotoxic injury [76]. Furthermore, human ApoE expression in C. elegans significantly attenuates Aβ-induced neurodegeneration [77]. All these findings support the interpretation that ApoE may play a protective role in Aβ-mediated axon/synapse competition. (A note on the ApoE4 allele: while known to be the highest genetic risk factor for late-onset Alzheimer’s disease [70, 71], studies show that this allele may have allele-specific effects on several processes related to axon/synapse competition, including, for example, APP transcription and Aβ secretion [21], synaptogenesis [24], and NMDA receptor activity [78]. Some studies suggest that ApoE4 may have an enhancing effect on brain functions such as visual working memory [79], while other studies show that it impairs neuronal insulin signaling [80] and leads to the generation of allele-specific toxic proteolytic products [81]. As such, it is at present unclear if the ApoE4 mutation is purely pathological or may potentially confer advantage in specific aspects of brain function).

Several potential mechanisms may explain the protective effects observed of ApoE (Fig. 1). First, at the level of direct physical interactions, ApoE has been found to strongly influence the process of Aβ aggregation and as a result the amount of toxic Aβ species that may be produced. For example, early studies show that purified ApoE binds to Aβ peptides with significant affinity [2–6] (but also [7]). Recent investigations into the effects of ApoE on the different phases of Aβ oligomerization further show that it likely has differential effects on the production rate of different species of Aβ associated with different levels of toxicity. For example, studies analyzing the kinetics of Aβ oligomerization have consistently shown that purified ApoE slows down the growth of toxic Aβ oligomer species [8–10]. An EM study similarly shows that ApoE prevents Aβ oligomer formation [11]. Other studies show that ApoE may promote the formation of amyloid filaments or high-molecular-weight oligomers that are less toxic [12–15]. In addition, ApoE-related clusterin has also been found to similarly sequester Aβ oligomers [82]. Together, these results suggest that ApoE may potentially play a neuroprotective role during Aβ-mediated axon/synapse competition. Interestingly, of the different ApoE alleles, ApoE4 appears to have a high level of activity in stabilizing soluble toxic Aβ oligomers [8, 10], which may contribute to the increased risk of Alzheimer’s disease development associated with this allele.

Figure 1. Mechanisms by which astrocytic ApoE may provide protection during Aβ-mediated axon/synapse competition.

A. Several studies (see text for details) show that ApoE may both inhibit the formation of toxic Aβ species, potentially at steps of both primary and secondary nucleation, and promote the formation of less toxic high-molecular -weight (HMW) Aβ oligomers. B. There is also evidence indicating that ApoE may inhibit the transcription and secretion of inflammatory cytokines as well as promote phagocytosis and Aβ degradation by both activating cell surface receptors and promoting cholesterol efflux. Figure created with BioRender.com.

In vivo, the effects of manipulating ApoE levels on Aβ are more complex and harder to interpret and may include both direct and indirect effects (see also below on roles of ApoE in glial phagocytosis and Aβ clearance). Nonetheless, several studies show that ApoE appears to decrease Aβ toxicity in vivo [16–20]. For example, early studies show that genetic removal of all ApoE from the brain appears to drastically reduce the total amounts of amyloid plaques from an APP overexpression mouse brain [16], while recent studies show that these manipulations have stage-specific effects on amyloid pathology [17, 18]. Most notably, new studies show that a key effect of ApoE removal in vivo may be in the reduction of amyloid plaque compaction, an effect that may be deleterious since it may result in increased amounts of diffuse toxic oligomers [19]. Other studies also show that genetic removal of all ApoE from APP overexpression mice leads to elevated levels of soluble Aβ in the brain [20]. Altogether, in vitro and in vivo findings suggest that ApoE may potentially serve neuroprotective roles in activity-induced axon/synapse competition by physically interacting with Aβ and modulating the kinetics of Aβ oligomer and other aggregate formation. Besides these lines of evidence, ApoE has also been found to specifically promote in human neurons the transcription of APP and the secretion of Aβ monomers [21] (A significant number of studies show that Aβ monomers and low-concentration oligomers are neuroprotective and positively regulate synapse formation, function as well as learning and memory [83–94]). ApoE also increases APP interaction with β- and γ-secretases and promotes Aβ production through regulating cholesterol [22, 23]. Thus, ApoE may potentially protect competing axons/synapses additionally through up-regulating Aβ monomer levels. Lastly, ApoE has been found to promote synaptogenesis through CREB activity independent of APP transcription [24]. These lines of evidence thus further support the interpretation that ApoE may normally serve a neuroprotective role in axon/synapse competition in the brain. Surprisingly, of the different ApoE alleles, ApoE4 appears to have the highest potency in stimulating APP transcription and Aβ monomer secretion in neurons [21, 24]. It has been suggested that the chronic effects of ApoE4 on APP and Aβ may be linked to the increased risk associated with the ApoE4 allele [24].

Unlike bacterial cell competition, eukaryotic cell competition, including axon/synapse competition, involves not only cellular components that directly compete with one another but also third-party glial cells that actively regulate this process [1]. These glial cells, such as microglia and astrocytes, not only express cytokines that control the timing and other aspects of axon/synapse competition, but also perform phagocytic functions such as dendritic spine pruning and Aβ clearance that control the execution and conclusion of this process. As such, controlling the activity of glial cells is likely also a key node in the regulation of axon/synapse competition. Indeed, a large body of evidence indicates that Aβ oligomers and monomers appear to not only act directly on competing neuronal structures but also engage and modulate the activity of glial cells during axon/synapse competition [1]. Importantly, recent studies show that, besides Aβ, ApoE also strongly impacts the activity of microglia and astrocytes and may also serve a neuroprotective role in axon/synapse competition through this indirect route [95–100] (reviewed in [68, 69]).

A prominent mechanism by which microglia and astrocytes promote axon/synapse competition in the nervous system is through regulating inflammatory cytokines [1]. For example, in the retinogeniculate system, astrocyte-derived cytokine IL-33 has been found to be essential for neural activity dependent elimination of excess synapses [101]. At the vertebrate neuromuscular junction (NMJ), a classic model of activity-dependent axon competition, TNFα is also essential for activity-dependent elimination of supernumerary motor axon terminals [102]. In both the retinocollicular system and the somatosensory cortex, TNFRSF21 (or DR6), a TNFα receptor superfamily member, similarly promotes activity-dependent axon competition [103]. As mentioned, several studies suggest that, besides acting on neurons, ApoE may also protect competing axons/synapses by inhibiting inflammatory cytokine production in glial cells and as such dampen competition [95–100]. Early studies show that ApoE strongly inhibits inflammatory activation of microglia in vitro [95]. Recent studies also show that ApoE possesses potent anti-inflammatory activity against macrophages, downregulating macrophage expression of pro-inflammatory genes while upregulating anti-inflammatory genes [96]. Furthermore, ApoE may also dampen glial cell inflammatory activity through well-established mechanisms such as promoting the efflux of cholesterol and other lipid molecules (reviewed in [1]). Consistent with this interpretation, deficiency in ApoE has been found to result in increased production of proinflammatory cytokines in vivo [97]. Deficiency in the ApoE receptors VLDLR or ApoER2 similarly results in enhanced proinflammatory phenotype in macrophages [96, 98]. Like Aβ, the anti-inflammatory effects of ApoE also appears to be concentration dependent, with high concentrations of ApoE promoting rather than suppressing inflammatory response in macrophages [99, 100]. Besides regulating glial expression of cytokines during axon/synapse competition, ApoE may also protect competing axons/synapses by regulating a second function of glial cells in this process, phagocytosis. For example, converging studies suggest that Aβ oligomers may directly promote glial phagocytosis and the elimination of competitor axons/synapses during competition [1]. Indeed, large numbers of studies show that ApoE promotes Aβ degradation by glial cells and may thus be protective to competing axons/synapses [39, 104–109]. ApoE has been found to enhance Aβ trafficking and degradation by neprilysin in microglia through the modulation of microglial cholesterol levels [104, 105]. There is also evidence that ApoE may also promote Aβ degradation by astrocytes [106] (but see also [7]). In accordance with this interpretation, several groups have shown that ApoE is a strong ligand for the microglia specific Trem2 receptor [39, 107–109], the activation of which has been shown by several groups to both promote microglial phagocytosis and inhibit inflammation [39, 107, 108, 110–112]. Thus, these results support the interpretation that, like Aβ, low concentrations of ApoE may also serve a neuroprotective role in axon/synapse competition through regulating glial cell activity. Altogether, the experimental results discussed above suggest that ApoE may play a neuroprotective role in axon/synapse competition through a multitude of mechanisms including impeding toxic Aβ species formation, promoting Aβ clearance, and suppressing inflammatory cytokine production, and that the perturbation of this role of ApoE may contribute to the development of Alzheimer’s disease. Indeed, the ApoE4 allele has been found to be associated with high levels of soluble Aβ oligomers and microglial inflammatory activation in the human brain [12, 113–115] as well as perturbed activity and immune regulation of human microglia and astrocytes [116–122]. This suggests that compromises in the normal function of ApoE as discussed may play a role in the high risk associated with the ApoE4 allele.

Regulation of DNA damage repair by APP/AICD and its role in Alzheimer’s disease

Neural activity promotes Hebbian as well as non-Hebbian synaptic plasticity in the brain, both of which involve production of reactive oxygen species (ROS) [123–130], a well-known inducer of DNA damage. Neural activity also induces the transcription of plasticity related genes, the activation of which requires double strand DNA break at gene regulatory regions [131–134]. Thus, neural activity, while promoting plasticity, also induces DNA breaks or damages that are detrimental to the nervous system. In this regard, it is interesting to note that studies have long shown that APP, the precursor protein for Aβ, has an at times puzzling function in DNA damage repair [55–67]. For example, studies have shown that large numbers of genotoxic agents induce the cleavage of the APP protein, leading to the release of a C-terminal proteolytic product (AICD) and an associated FE65 protein [55, 56], in a process dependent on the gamma-secretase complex [55, 135]. APP cleavage releasing AICD in turn promotes the nuclear translocation of FE65 [57–60], which in the nucleus, binds to a Tip60-TRRAP histone acetyltransferase (HAT) complex, a complex that mediates histone H4 acetylation and DNA repair protein loading at DNA damage sites and plays an evolutionarily conserved role in DNA damage repair from yeast to mammals [61–63]. FE65 also binds to and stabilizes the Bloom syndrome protein (BLM) [64], a DNA helicase with double strand break resection activity that promotes DNA repair by homologous recombination, especially at actively transcribed genes [65, 66]. Furthermore, the AICD/FE65 complex has also been found to regulate gene expression and decrease mitochondrial superoxide production [67]. Thus, these findings suggest that besides producing Aβ peptides that may directly mediate homeostatic and competitive plasticity, APP may potentially also participate in this process by producing an intracellular proteolytic product that promotes DNA damage repair and thus plays a potential protective role during synapse competition (Fig. 2). This may help reconcile the functional implication of APP in regulating DNA damage repair.

Figure 2. APP/AICD may be protective during axon/synapse competition.

By producing an intracellular cleavage product (AICD), APP may potentially promote nuclear as well as potentially mitochondrial DNA repair through the adaptor protein FE65 and the TIP60/TRRAP histone acetylase complex. Figure created with BioRender.com.

Studies show that DNA damage in interphase cells and post-mitotic neurons leads to increased dynamics of both cytoplasmic and nuclear microtubules that promotes DNA repair [136–138]. Endogenous tau proteins have also been found to directly bind to and protect neuronal DNA [139], while tau knockdown exacerbates double strand DNA breaks [140]. These findings thus raise a possibility that the nuclear phospho-tau that is prominently observed in brain neurons in tauopathies such as Alzheimer’s disease [141–145] may be linked to the regulation of DNA damage repair [146–148], providing a potential mechanistic explanation for the appearance of nuclear phospho-tau in various tauopathies. Importantly, besides in the nucleus [149–152], pro-inflammatory cytokines, which as mentioned play key roles in axon/synapse pruning, have been found to also induce extensive DNA damage in the mitochondria [151, 153–156]. Aβ oligomers, the proposed activity-dependent agent mediating axon/synapse elimination during competition, also directly target a significant number of mitochondrial regulators, many of which are linked to the production of reactive oxygen species [157–162]. In addition, many components of the molecular machinery known for their function in nuclear DNA repair also localize to the mitochondria and perform key roles in maintaining the integrity of the mitochondrial genome [163–166]. Thus, these findings suggest that, besides nuclear DNA repair, APP/AICD may also play a role in mitochondrial DNA repair, a process that may potentially be more relevant to activity-dependent axon/synapse competition, since mitochondrial DNA may be the main target of the reactive oxygen species produced in this process. Indeed, studies have reported the localization of APP cleavage pathway components and products in the mitochondria. However, the interpretation of these results has been complicated by factors including the frequent use of gene overexpression in these experiments (reviewed in [167]). Nonetheless, recent studies show that mitochondrial DNA damage and release into the cytoplasm and the extracellular environment is closely associated with aging and the development of neuroinflammation and cognitive function decline in humans [168–171]. Alzheimer’s disease fibroblasts have also been found to show increased levels of cytoplasmic mitochondrial DNA [172]. This suggests that both nuclear and mitochondrial DNA damage may play a role in the development of Alzheimer’s disease and may be potential targets for clinical intervention [173, 174].

Relationship between roles played by oligomeric Aβ and ApoE in physiological synapse elimination and in Alzheimer’s disease.

In axon/synapse competition, competing axons/synapses must not only protect themselves but also eliminate competitors. Strong evidence indicates that neurotoxic Aβ oligomers induced by high levels of neural circuit activity play a large role in effecting competitor elimination in this process [1]. Among the evidence, it has been found that PirB, a high affinity Aβ oligomer receptor, and related signaling pathway play a crucial role in ocular dominance plasticity and mutations in this pathway severely disrupt cortical synapse reorganization induced by monocular deprivation [175–178]. Besides PirB, recent studies have also identified several other high affinity Aβ oligomer receptors and show that they play similar roles in homeostatic and competitive plasticity across the nervous system [179–198]. Furthermore, besides specific receptors, studies show that Aβ oligomers also mediate competitor elimination through general mechanisms that target unique membrane lipid changes induced by synapse competition [25–31, 34–36, 38–41]. In addition, unlike the protective role played by astrocytic ApoE, recent studies show that neuronal ApoE may instead promote competitor elimination during axon/synapse competition [199–214]. In this section, I will discuss these new findings and highlight how they may shed new light on the widespread loss of synapses as well as the selective vulnerability to neurodegeneration of ApoE expressing neurons in Alzheimer’s disease.

Aβ oligomer receptors and the targeting of different classes of dendritic spines during synaptic plasticity and in Alzheimer’s disease

Dendritic spines are the most common postsynaptic component of excitatory synapses in the brain and are targets for activity-dependent competitive elimination during axon/synapse competition. Spine elimination is typically preceded by synaptic depression, which involves at least two different mechanisms, metabotropic glutamate receptor (mGluR)- and NMDA-type glutamate receptor (NMDAR)-dependent long-term depression (LTD) [215, 216]. Spines come in different shapes and sizes. Large spines, frequently of a mushroom shape, typically contain stacks of endoplasmic reticulum (ER) known as the spine apparatus, while small spines lack the spine apparatus [217–221]. Interestingly, recent studies show that only large spines appear to be capable of undergoing mGluR-dependent LTD [222, 223] and require the activation of both mGluR- and NMDAR-dependent LTD to undergo size shrinkage [224] (but see also [189]), while small spines can undergo size shrinkage upon induction of NMDAR-dependent LTD alone.

Importantly, independent studies show that different high affinity Aβ oligomer receptors appear to differentially regulate mGluR- and NMDAR-dependent signaling and LTD at excitatory synapses [179–198]. This raises a possibility that Aβ oligomers may target and regulate the function of different types of dendritic spines through different combinations of Aβ oligomer receptors (Fig. 3). For example, Aβ oligomers have been found to bind with high affinity to the cellular prion protein (Prp^c) [179], which is highly enriched at excitatory postsynaptic sites and forms a complex with mGluR5. Upon Aβ oligomer binding, cellular prion protein traps mGluR5 in the plasma membrane, leading to spine loss through activation of the nonreceptor tyrosine kinase Fyn [180–182]. In vivo, increased interactions between mGluR5 and Fyn have been observed in the barrel cortex following whisker trimming, a well-known model of competitive plasticity [183], suggesting that this regulatory mechanism takes place under physiological conditions. Thus, the cellular prion protein may mediate Aβ oligomer effects on a select group of dendritic spines (potentially large spines since only they appear capable of undergoing mGluR-LTD) through activating mGluR-LTD or related pathways [184, 185]. On the other hand, while the direct binding of Aβ oligomers to NMDARs has not been observed, Aβ oligomers have been found to bind to several NMDAR associated surface receptors at excitatory synapses. Aβ oligomers bind with high affinity to EphB2 [186], a tyrosine kinase receptor that plays key roles in the synaptic localization and activity of NMDARs [225–228]. Aβ oligomers also bind to and affect the function of several planar cell polarity (PCP) proteins such as Frizzled and Celsr3 [187, 188] that play similarly critical roles in the formation and function of excitatory synapses including the NMDARs [229–232]. Furthermore, Aβ oligomers also inhibit NMDAR-dependent long-term potentiation (LTP) through the cellular prion protein and mGluR5 [185]. Most important, while the potential effects of spine size, if any, remain unclear, large numbers of studies consistently show that NMDAR activity is required for both Aβ oligomer-induced synaptic depression and Aβ oligomer-induced shrinkage at spines, possibly through both its ionotropic and metabotropic activities [189–198]. These results therefore support the possibility that during synapse competition, Aβ oligomers may potentially target and eliminate large spines through both mGluR-linked (e.g., the cellular prion protein) and NMDAR-linked receptors but target small spines through NMDAR-linked receptors alone. This not only provides further evidence supportive of the role of Aβ in synaptic competition but is also consistent with observations that the prion protein forms a complex with Aβ and mediates the toxicity of soluble Aβ aggregates in the Alzheimer’s disease brain [233–235] and the levels of NMDARs are severely perturbed [236–239].

Figure 3. Mechanisms by which Aβ oligomers may potentially target different classes of dendritic spines.

Studies (see text) suggest that only large but not small spines can undergo and require mGluR-dependent LTD for shrinkage, while both large and small spines require NMDAR-dependent LTD for shrinkage. Thus, a potential possibility is that Aβ oligomers may target and induce the shrinkage of different classes of spines through activating different combinations of Aβ oligomer receptors at these spines. Figure created with BioRender.com.

Aβ interaction with phosphatidylserine and the targeting of axons/synapses under physiological and diseased conditions

Phosphatidylserine (PS), a phospholipid that normally resides in the inner leaflet of the plasma membrane but is exposed on the outer leaflet upon induction of apoptosis, is a well-known “eat-me” signal that promotes cell debris clearance during programmed cell death [240, 241]. Recent studies show that this signal is conserved and plays a critical role in activity-dependent axon/synapse competition [242–245]. Notably, phosphatidylserine exposure on pre- and post-synaptic membranes has been found to be developmentally regulated and coincide with the period of activity-dependent synapse pruning in vivo, and blockade of its interaction with phosphatidylserine receptors on microglia by multiple approaches all results in compromised microglial engulfment and elimination of supernumerary synapses during normal development [242–245]. Aβ oligomers have potent membrane disrupting activities mediated by mechanisms including pore formation [246–248], lipid extraction [249, 250], and partial insertion as well as carpet-like mechanisms [251–254]. These mechanisms may in part mediate the pruning of weak axons/synapses in activity-dependent competition [1]. As such, the timing and the kinetics of Aβ oligomer formation are likely key factors in axon/synapse competition. Indeed, studies show that Aβ oligomer formation is regulated by many factors including a complex interaction with plasma membrane components [25–30]. For example, the different phospholipids of the plasma membrane that reside in the inner or outer leaflet appear to impact Aβ oligomerization very differently [25, 26]. Phosphatidylserine-like lipids that are only exposed in the outer leaflet upon induction of processes such as apoptosis accelerate, while phosphatidylcholine-like lipids that normally populate the outer leaflet slow down the aggregation of Aβ peptide fragments [25, 26]. Likely as a result, the extent of cell surface exposure of phosphatidylserine has been found to determine the susceptibility of cells to Aβ toxicity while phosphatidylcholine, which is normally found in the outer leaflet, renders protection for these cells [27–29]. Recent studies have further shown that when incorporated into the outer leaflet of a lipid bilayer, phosphatidylserine, but not phosphatidylcholine, can trigger rapid Aβ oligomerization even at subnanomolar concentrations and as a result induce extensive membrane pore formation [30]. These unique effects of phosphatidylserine suggest a potential self-amplifying mechanism during axon/synapse competition in which phosphatidylserine-induced Aβ oligomerization may possibly act to increase the strength or duration of the initial “eat-me” signal produced by phosphatidylserine exposure and as such facilitate the specific tagging and elimination of weak axons/synapses (Fig. 4).

Figure 4. Mechanisms by which phosphatidylserine (PS) may promote Aβ oligomerization, amplify the “eat-me” signal, and tag synapses for phagocytosis.

PS on the outer leaflet of the plasma membrane, exposed initially by competitive interaction between synapses, may potentially catalyze the local oligomerization of Aβ at the tagged specific synapse and thus amplify the initial signal. PS and Aβ oligomers have both been shown to directly and indirectly activate several microglial receptors that promote phagocytosis and thus likely synapse pruning. Together, it appears as if they act in a concerted manner to promote synapse pruning. Figure created with BioRender.com.

In support of above interpretation, Aβ oligomers have been found to enhance the surface exposure of phosphatidylserine and phosphatidylserine-directed engulfment of spine synapses [31], suggesting a mutually reinforcing interaction between phosphatidylserine and Aβ oligomers. Several receptors known to bind to phosphatidylserine and promote cell debris phagocytosis and clearance have been found to also bind to and be activated by Aβ oligomers [32–41]. For example, the complement cascade and receptors are well known to play a prominent role in activity-dependent axon/synapse competition [255–257], by recognizing as well as being activated by phosphatidylserine in this process [32, 33]. Importantly, large numbers of studies show that Aβ oligomers also directly and independently activate both the classical and the alternative complement pathway [34–36]. Thus, phosphatidylserine and Aβ oligomers may act in a concerted manner in the activation of the complement-dependent clearance pathways. Similarly, the TAM receptor tyrosine kinases Tyro3, Axl, and Mer, which play key roles in synapse competition in the retinogeniculate system [258], are also known to be activated by phosphatidylserine through bridging proteins [37]. Recent studies have shown that microglia may also employ the TAM receptors to detect and engulf Aβ aggregates associated with amyloid plaques [38]. Furthermore, the microglia specific Trem2 receptor, the activation of which promotes microglial phagocytosis [39, 107, 108, 111, 112], is known to bind with significant affinity and be activated by phosphatidylserine [39, 40]. Independent studies show that it can also bind directly to Aβ oligomers with nanomolar affinity and this interaction is required for both Aβ regulation of downstream signaling activity and Aβ degradation by microglia [41]. Altogether, these results strongly suggest that, besides activation of specific surface receptors, the regulation of Aβ oligomer formation by membrane-exposed phosphatidylserine may also promote activity-dependent axon/synapse competition by amplifying the “eat-me” signal and enhancing the specific tagging and elimination of weak axons/synapses. Importantly, large numbers of studies show that the dysregulation of this mechanism likely plays a significant role in Alzheimer’s disease [259–265]. For example, studies show that caspase 3, a key effector caspase in apoptosis, inactivates flippases such as ATP11A and ATP11C and activates scramblases such as Xkr8, both of which promote the cell surface exposure of phosphatidylserine [259–261]. In the Alzheimer’s disease brain, the levels of active caspase 3 have been found to be elevated especially in the postsynaptic density [262–265], potentially contributing to the extensive loss of synapses.

Regulation of MHCI and related competition genes by neuronal ApoE and the selective vulnerability of ApoE expressing neurons in Alzheimer’s disease

In previous sections, I discuss evidence indicating that astrocytic ApoE acts as a neuroprotective agent in axon/synapse competition by, among others, upregulating APP gene transcription and directly and indirectly suppressing the assembly of toxic Aβ species. However, new studies indicate that this is unlikely the sole function of ApoE. Indeed, ApoE, especially ApoE proteins of neuronal origin, may be part of a regulatory network that normally promotes the expression and the activity of a specific set of genes directly involved in axon/synapse competition [199–214] (Fig. 5). ApoE is most highly expressed in astrocytes in the normal brain. However, it is also expressed in a select group of neurons in a manner dependent on neural activity [73, 74]. The level of ApoE expression in these neurons not only strongly correlates, on a cell-by-cell basis, with that of immune response genes in these same neurons but also regulates the expression of these immune response genes [199]. Many of these ApoE-regulated genes in turn play key roles in axon/synapse competition. For example, neuronal ApoE regulates the expression of MHCI, the class I major histocompatibility complex and a molecular partner of PirB that play pivotal roles in ocular dominance plasticity [177, 178, 200, 201]. Neuronal ApoE also regulates the expression of several other genes such as Tap2 that interact and cooperate with MHCI at the molecular level [202]. It also regulates the expression of C1qa, a component of the complement cascade with widespread roles in activity-dependent synapse elimination [255, 256]. Moreover, it regulates the neuronal expression of interleukin 4, a cytokine that potently modulates brain microglial pathways and neuronal network activity [203, 204]. Thus, neuronal ApoE appears to play a pivotal role in coordinating the expression of genes involved in axon/synapse competition and as such promoting competitive plasticity in the normal brain.

Figure 5. Autocrine and paracrine ApoE may together regulate a gene network that promotes axon/synapse competition.

Studies (see text for detailed discussion) indicate that neuronal ApoE promotes, in an autocrine manner, the expression in the same neurons of large numbers of genes, including MHCI, Tap2, and C1qa, that directly participate in axon/synapse competition. Astrocyte-derived ApoE has also been shown to increase APP transcription and Aβ secretion by neurons. In addition, as shown in Fig. 1, ApoE also appears to directly interact with Aβ. Thus, ApoE appears to be a key part of the gene network that regulates axon/synapse competition. Figure created with BioRender.com.

Importantly, the regulation by ApoE of this group of immune response genes with demonstrated or likely roles in axon/synapse competition appears to be linked to the selective vulnerability of these expressing neurons in Alzheimer’s disease. For example, functional reduction of MHCI has been found to ameliorate tau pathology in both ApoE4-expressing primary neurons and pathological tau-expressing mouse hippocampi [199]. Interestingly, previous studies have shown that Stat1, a transcriptional factor that mediates JAK2 driven elimination of inactive synapses during activity-dependent competition [205], binds to the enhancer and regulates the transcription of ApoE gene [206]. Similarly, the neuronal expression of RORβ, another marker for selectively vulnerable neurons in Alzheimer’s disease [207], is not only required for the normal formation of the barrel cortex in a well-known activity-dependent competitive process, and sufficient to induce barrel-like neuronal clusters when overexpressed [208, 209], but it is also regulated by ApoE4 [210]. Furthermore, Stat1/2 also bind to the promoter and promote the expression of RORβ gene [211]. Conversely, strong Stat1 activity has been linked to suppressed expression of MEF2C, a key transcriptional factor that confers resilience to neurodegeneration in Alzheimer’s disease and protect neurons when overexpressed [212–214]. Thus, together with the results showing the regulation of APP transcription and Aβ secretion by ApoE, these findings suggest that Aβ, ApoE, MHCI, Stat1, RORβ, and other genes may normally form a regulatory network that promotes neural activity-dependent axon/synapse competition under physiological conditions. Perturbation of the activity of this network may upset the delicate balance in its neuroprotective and neurotoxic activities and lead to the development of Alzheimer’s disease. This may in part explain the selective vulnerability of ApoE and RORβ expressing neurons in Alzheimer’s disease since the expression of these genes in these neurons suggest that they may normally be engaged in competitive plasticity under physiological conditions and thus especially vulnerable to perturbations in the process.

Relationship between Aβ, homeostatic and competitive plasticity, and sleep

An intriguing observation of competitive synaptic plasticity was made over twenty years ago when it was found that sleep robustly enhances the effects of competitive plasticity such as those observed following monocular deprivation [266]. Studies since have shown that rapid eye movement (REM) sleep, a sleep stage when dendritic spine elimination is especially prevalent in the brain [267, 268], appears to play an important role in ocular dominance plasticity and related processes [269–272]. Aβ levels in the brain are regulated by the sleep-wake cycle, being dramatically up-regulated upon sleep deprivation [42]. As mentioned, a large body of evidence also suggests that Aβ oligomers that form in the brain as a result of high-level neural activity may directly mediate activity-dependent axon/synapse competition [1]. These findings thus raise the question of whether there are mechanistic links between Aβ, sleep, and activity-dependent synapse competition. Recent studies show that this may indeed be the case, potentially explaining the well-known role of sleep perturbation in Alzheimer’s disease [42–54].

First, strong evidence indicates that Aβ oligomers may directly induce sleep [43–50]. Aβ levels in the human brain are generally low in the morning, but gradually increase during the day before peaking at night [42]. Aβ oligomerization has been shown to be likely regulated by a concentration-dependent nucleation mechanism [273, 274]. As the day goes on and the levels of Aβ in the brain rise, the levels of Aβ oligomers in the brain are likely to also rise. As mentioned, one of the high-affinity Aβ oligomer receptors in the brain is the cellular prion protein, which activates mGluR5 and Fyn and induces synaptic depression and spine loss upon Aβ oligomer binding [179–182]. Interestingly, recent studies show that in both zebrafish and mice, Aβ oligomers also act through the cellular prion protein to regulate sleep [43, 44]. Exogenous application of long Aβ oligomers, for example, has been found to dampen brain neuronal activity, increase sleep, and reduce waking activity in zebrafish, an effect prevented by prion gene mutation [43]. The increased sleep is linked to increases in the number of sleep bouts but not the lengthening of individual bouts, indicating that long Aβ oligomers boost sleep initiation. Conversely, inhibiting the activities of mGluR5 and Fyn both blocks the sleep-inducing effects of long Aβ oligomers [43]. Prior to this, studies have also shown that loss-of-function mutations in the cellular prion gene result in altered circadian activity rhythms and sleep in mice [45, 46]. Knockout of mGluR5 similarly results in severe dysregulation of sleep-wake homeostasis, including a lack of recovery sleep after sleep deprivation [47]. Furthermore, in humans, increased levels of mGluR5 after sleep loss have been found to tightly correlate with biomarkers of elevated sleep need [48]. Thus, these results all suggest that there are strong molecular links between Aβ and sleep. Indeed, independent studies show that the function of cellular prion proteins is required for the nighttime increases in melatonin [49], a sleep-promoting hormone that plays key roles in sleep cycle regulation [275, 276]. Evidence also suggests that activity of the cellular prion protein may increase the degradation of norepinephrine [50], a hormone that normally promotes wakefulness [277–279]. Activity of the glymphatic system normally increases during sleep and promotes the clearance of metabolic waste from the brain [280, 281]. Interestingly, signaling by mGluR5 as well as by adenosine, the latter being a critical sleep-promoting metabolite [282–284], have both been found to promote astrocyte expression of Aquaporin 4 [285, 286], a water channel with crucial roles in the function of the glymphatic system [287, 288]. Thus, these findings indicate that Aβ, a candidate molecular agent for mediating competitive synaptic plasticity, may potentially also directly regulate sleep, a brain state with crucial roles in competitive plasticity.

Besides sleep induction, research on the mechanisms of synaptic scaling-down, a form of homeostatic plasticity, during sleep have also revealed molecular links between Aβ oligomers and synaptic changes in sleep and provided further support for a role of Aβ and sleep in homeostatic and competitive synaptic plasticity [51–53] (Fig. 6). Sleep plays important roles in many aspects of brain function [289, 290]. Among the hypotheses proposed on the function of sleep, the synaptic homeostasis hypothesis (SHY) proposes that a key function of sleep is to restore synaptic homeostasis in the brain after a net increase in synaptic strength following a day of wakefulness [291]. Indeed, three-dimensional electron microscopy (EM) studies show that the average axon-spine interface (ASI) in the mouse motor and sensory cortices decreases by ~18% after sleep and the decrease is proportional to ASI size, indicating synaptic scaling [292–294](but see also [295]). The scaling appears to be selective, sparing synapses that are large and lack recycling endosomes (spines with and without a spine apparatus, however, both appear to undergo scaling) [292]. Interestingly, independent studies show that mGluR5, the molecular partner of the cellular prion protein that mediates the induction of sleep by Aβ oligomers [43, 44, 47, 48], also plays a crucial role in synaptic scaling-down during sleep [51, 52]. mGluR5 and related mGluR1 have been found to trigger synaptic scaling-down in sleep by activating ligand-independent signaling through a mechanism dependent on Homer1a, a postsynaptic cytoskeletal protein induced by sleep loss [53]. Specifically, these studies find that Homer1a levels in the postsynaptic density are increased by the sleep-promoting adenosine but decreased by the wake-promoting norepinephrine [51]. As a result, during sleep, high levels of Homer1a accumulate at the postsynaptic density and, through altering postsynaptic organization, activate mGluR1/5 signaling, which then promotes the removal of surface AMPA receptors and the weakening of these synapses [51]. Indeed, mutations in Homer1a and inhibition of mGluR1/5 activity have both been found to block sleep cycle-associated changes in synaptic AMPA receptors [51]. These results thus implicate a critical role played by Homer1a-activated mGluR1/5 signaling in synaptic scaling-down during sleep. Importantly, both Homer1a and mGluR5 have also been found to be required for the establishment of the contralateral bias as well as the maintenance of ocular dominance in the mouse cortex [54]. This supports the interpretation that sleep-associated synaptic scaling-down may be physiologically linked to and play a key role in competitive plasticity, a process also associated with and regulated by sleep [269–272].

Figure 6. Mechanisms by which Aβ oligomers may potentially regulate the synaptic specificity of Homer1a targeting during homeostatic synaptic scaling-down in sleep.

Synaptic targeting of Homer1a has been shown to play a key role in homeostatic down-scaling in sleep. Significant evidence (see text) suggests that ERK cascade activity may regulate the synapse specificity of Homer1a targeting. As shown in Fig. 3, Aβ oligomers can activate both mGluR5 and NMDAR signaling, which have also both been shown to activate ERK. Thus, a potential determinant for the synapse specificity of Homer1a-medaietd down-scaling may be the local concentration of Aβ oligomers near a synapse and the degree and combination of Aβ oligomer receptors they may activate. Figure created with BioRender.com.

Furthermore, studies suggest that the regulation of synaptic scaling-down during sleep, especially the specificity of the synapses that are being targeted, may be linked to the activation of Aβ oligomer receptors at these synapses [180, 296–306]. This potentially parallels the differential targeting of large and small spines by Aβ oligomers, through the different combination of receptors, as discussed in the previous section. For example, in parallel to the morphological scaling of synapses during sleep, synaptic targeting of Homer1a has also been found to be selective [296–298], in a process dependent on the activation of the ERK serine/threonine kinase [296, 297, 299]. Interestingly, the neuronal ERK cascade has been found to be activated by cellular prion protein signaling through the Fyn receptor tyrosine kinase [300–303], both of which are in turn activated by Aβ oligomer binding to the prion protein [180]. In addition, NMDARs, which are required for and mediate Aβ oligomer-induced synaptic depression and spine shrinkage at both large and small spines, have also been found to regulate the activity of the ERK cascade in neurons [304–306]. Thus, these results suggest that, besides regulating sleep, Aβ oligomers may possibly also regulate the specificity of synaptic scaling-down during sleep through regulating ERK and Homer1a activity at specific synapses. As such, Aβ appears to play a central role in orchestrating and coordinating many aspects of synaptic reorganization and homeostatic and competitive plasticity during sleep. This may explain the prominent role that sleep disturbances play in the development of Alzheimer’s disease.

Concluding remarks

Aβ and ApoE are two of the biggest players in Alzheimer’s disease pathology, but they also play physiological roles in the healthy brain in processes including activity-dependent homeostatic and competitive plasticity. In this review, I have summarized a large body of research showing that the roles played by Aβ and ApoE in Alzheimer’s disease can be unified with those they play in physiological processes and that it is the perturbation of these physiological processes that underlies the development of Alzheimer’s disease. For example, being part of a normal mechanism that regulates activity-dependent homeostatic and competitive plasticity in the brain, the formation of Aβ oligomers likely, under physiological conditions, promotes the removal of supernumerary synapses, through both activating different combinations of receptors at different dendritic spines and amplifying general cell surface “eat-me” signals. In Alzheimer’s disease, however, perturbation of these processes leads to excess Aβ oligomer formation. This may not only provoke an intense neuroinflammatory response but also lead to extensive dendritic spine pruning and loss, both common pathologies observed in the Alzheimer’s disease brain. Similarly, the ApoE protein likely normally regulates the kinetics of Aβ aggregation and turnover as well as glial lipid metabolism (astrocytic ApoE) as well as promotes the expression of genes involved in the physiological process of competitive synaptic plasticity, by forming a gene regulatory network with Aβ and other molecules (neuronal ApoE). In Alzheimer’s disease, however, perturbation of these ApoE regulated processes not only contributes to excess Aβ oligomer formation but also underlies the selective vulnerability of ApoE expressing neurons in the diseased brain. Furthermore, Aβ oligomers not only induce sleep but may also orchestrate the processes of synaptic scaling-down and spine pruning, plasticity processes linked to this brain state. Thus, this body of research in aggregate provides compelling evidence that Alzheimer’s disease is a disease of competitive synaptic plasticity gone wrong, a perspective that may promote advance in this area.