Abstract

INTRODUCTION

Alzheimer's disease is associated with the accumulation of pathogenic amyloid-β (Aβ) assemblies in the brain and results in the progressive dismantling of synapses, neuronal circuits and networks. In vitro and in vivo studies have shown that Aβ oligomers reduce glutamatergic synaptic transmission strength and plasticity^{1, 2, 3}. Subsequent studies provided evidence that neuronal activity regulates Aβ production^{4, 5} and that elevated Aβ attenuates excitatory synaptic transmission by decreasing the number of surface AMPA receptors (AMPARs) and NMDA receptors (NMDARs), associated with a collapse of glutamatergic dendritic spines^{4, 6, 7}.

More recently, we and others have begun to investigate the effects of Aβ at the level of neuronal circuits and wider networks. These studies have yielded unexpected results. In mice transgenic for human amyloid precursor protein (hAPP), pathologically elevated levels of Aβ promote the formation of pathogenic Aβ oligomers and cause wide fluctuations in the neuronal expression of synaptic activity–regulated genes⁸, as well as epileptiform activity and nonconvulsive seizures⁸. It also increases the proportion of abnormally hypoactive or hyperactive neurons in cortical circuits⁹.

In humans with Alzheimer's disease, increased Aβ is also associated with complex derangements of neuronal activity. For example, hypometabolic regions in the parietal cortex show aberrant increases in neuronal activity during memory encoding^{10, 11}, and individuals from many pedigrees with early-onset autosomal dominant Alzheimer's disease have epileptic activity¹². Thus, synaptic depression and aberrant patterns of neuronal network activity seem to coexist not only in mouse models, but also in the human condition. This intriguing scenario suggests that Aβ, and possibly other Alzheimer's disease–related factors, have a central role in controlling neuronal activity at specific types of synapses, wider neuronal networks, or both.

Here we discuss recent findings indicating that Aβ is part of an activity-regulated mechanism that controls synaptic excitatory activity, in which Aβ induces presynaptic facilitation at low concentrations and postsynaptic depression at high concentrations. Aβ also causes GABAergic dysfunction, which may contribute to the development of aberrant synchrony in neural networks and disruption of cognitive functions.

Investigating Aβ in Alzheimer's disease

It is likely that diverse factors contribute to the pathogenesis of late-onset Alzheimer's disease^{13, 14, 15}. Among them, Aβ and the genetic risk factor apolipoprotein (apo)-E4 stand out on the basis of overwhelming genetic evidence and strong experimental data^{16, 17, 18, 19}. Mutations in genes encoding hAPP, presenilin 1 (PS1) or presenilin 2 (PS2) that cause early-onset autosomal dominant familial Alzheimer's disease (FAD) increase the production of Aβ ending at amino acid 42 (Aβ42), the Aβ42/Aβ40 ratio or Aβ aggregation. Although not identical, early-onset FAD and sporadic late-onset Alzheimer's disease show wide clinical and pathological overlap²⁰ and thus may share common Aβ-dependent mechanisms of cognitive dysfunction.

To study this complex human disease, researchers have used many in vitro and in vivo models. Increased levels of Aβ in these experimental models are commonly achieved by overexpression of FAD-mutant forms of hAPP (with or without overexpression of PS1) in neurons; by exogenous application of synthetic, purified, or naturally secreted Aβ; or by inhibiting Aβ-degrading enzymes. Much of what we know about the effects of Aβ on neural function has been gleaned from the analysis of such models, particularly transgenic mice, acute brain slice preparations and cultures of neural cells or tissues. Human Aβ can exist in diverse assembly states, including monomers, dimers, trimers, tetramers, dodecamers, higher-order oligomers and protofibrils, as well as mature fibrils, which can form microscopically visible amyloid plaques in brain tissues²¹.

Much evidence suggests that Aβ oligomers are more potent than Aβ fibrils and amyloid deposits in eliciting abnormalities in synaptic functions and neural network activity^{7, 22, 23, 24, 25, 26, 27, 28, 29}. Therefore, many recent studies focusing on functional Aβ effects have used oligomers of human Aβ prepared from synthetic Aβ peptides³⁰, isolated from transfected cell lines²⁴ or purified from brains affected by Alzheimer's disease²⁵. Most of these preparations are devoid of Aβ fibrils and contain Aβ monomers as well as diverse Aβ oligomers that may be in dynamic equilibrium with one another. Different studies have only rarely used strictly comparable Aβ preparations. Nevertheless, the results obtained with different Aβ oligomer preparations have yielded rather consistent results. For example, although synthetic Aβ oligomer preparations are less potent than Aβ oligomers isolated from the supernatant of transfected cell cultures, their effects on acute hippocampal slices are qualitatively similar^{30, 31}. Therefore, we will only comment on the specifics of Aβ preparations when they stand out as unusual or unique. Notwithstanding this approach, we do consider the better standardization of Aβ preparations across studies an important objective.

The diversity of Aβ assemblies in most preparations of human Aβ42 makes it difficult to interpret the precise contribution of molar concentrations of Aβ to the observed physiological alterations. Most studies have used Aβ oligomer preparations whose Aβ content is equivalent to that of a high picomolar, nanomolar or low micromolar concentration of monomeric Aβ. The question then arises of what concentration of Aβ monomers or specific oligomers should be considered normal or physiological and what concentration abnormally low, high or pathological. The answer is unknown and will depend, in part, on whether the Aβ is located within or around the synaptic cleft, near a specific receptor, or within a particular intracellular compartment. Because of these issues, we will simply consider physiological ranges of Aβ to be those found in any given compartment in healthy young adults without cognitive impairments. Any increase in an Aβ species above this level that is associated with functional impairments might be abnormally 'high' or 'elevated'. As illustrated by the marked adverse vascular effects of prolonged subtle elevations in diastolic blood pressure, even small changes in critical variables can have profound biological consequences in the long run.

Modulation of synaptic transmission by Aβ

Synaptic loss is one of the pathological hallmarks of Alzheimer's disease and the best correlate of cognitive decline^{32, 33}, suggesting that it is a critical event in the pathophysiology of the disease. In vivo and in vitro studies have demonstrated that high levels of Aβ reduce glutamatergic synaptic transmission and cause synaptic loss^{1, 3, 4, 34}.

Notably, the production of Aβ and its secretion into the extracellular space are tightly regulated by neuronal activity in vitro⁴ and in vivo⁵. Increased neuronal activity enhances Aβ production, and blocking neuronal activity has the opposite effect⁴. This synaptic regulation of Aβ production is mediated, at least in part, by clathrin-dependent endocytosis of surface APP at presynaptic terminals, endosomal proteolytic cleavage of APP, and Aβ release at synaptic terminals⁵. In addition, pathogenic Aβ species can also be released from dendrites³⁵. This tight neuronal activity–dependent regulation of Aβ secretion has been observed during pathological events, such as epileptiform activity induced by electrical stimulation⁵, as well as during normal physiological processes, such as the sleep–wake cycle³⁶. It is also supported by the earlier development of amyloid plaques in patients with epilepsy³⁷. These findings support the notion that APP and Aβ are part of a feedback loop that controls neuronal excitability⁴. In this model (Fig. 1), Aβ production is enhanced by action potential–dependent synaptic activity, leading to increased Aβ at synapses and reduction of excitatory transmission postsynaptically. Pathological elevation of Aβ would be expected to put this negative feedback regulator into overdrive, suppressing excitatory synaptic activity at the postsynaptic level.

However, a recent study suggests that Aβ also acts as a positive regulator at the presynaptic level (Fig. 1a,b). In this study, relatively small increases in endogenous Aβ abundance (~1.5-fold), induced by inhibition of extracellular Aβ degradation in otherwise unmanipulated wild-type neurons, enhanced the release probability of synaptic vesicles and increased neuronal activity in neuronal culture³⁸. Enhanced extracellular Aβ increased spontaneous excitatory postsynaptic currents without significantly altering inhibitory currents. All these effects were exclusively presynaptic and dependent on firing rates, with less facilitation seen in neurons with higher firing rates. Thus, small increases of Aβ may facilitate presynaptic glutamatergic release in neurons with low activity but not in neurons with high activity.

Consistent with this finding, application of low (picomolar range) concentrations of Aβ markedly potentiates synaptic transmission, whereas higher concentrations (low nanomolar range) of Aβ cause the expected synaptic depression³⁹. The potentiating effect of Aβ does not affect postsynaptic NMDAR and AMPAR currents but is dependent on α7-nicotinic acetylcholine receptor (nAChR) activation, suggesting a presynaptic mechanism mediated by build-up of Ca²⁺ in presynaptic terminals (Fig. 1b). Thus, Aβ may directly act on presynaptic α7-nAChR⁴⁰ and be part of a positive feedback loop that increases presynaptic Ca²⁺ and Aβ secretion. Indeed, blocking nAChRs or removing α7-nAChRs decreases Aβ secretion and blocks Aβ-induced facilitation³⁵. Of particular importance, Aβ-induced presynaptic facilitation depends on an optimal Aβ concentration, with higher or lower concentrations impairing synaptic transmission³⁸. A positive modulatory effect of Aβ on synaptic transmission is further supported indirectly by the finding that an abnormally low Aβ level in mice deficient in APP (ref. 41), PS1 (ref. 42) or BACE1 (ref. 43) is associated with synaptic transmission deficits.

Overall, the above data suggest a bell-shaped relationship between extracellular Aβ and synaptic transmission in which intermediate levels of Aβ potentiate presynaptic terminals, low levels reduce presynaptic efficacy and high levels depress postsynaptic transmission (Fig. 1a).

Elevated Aβ impairs synaptic transmission

Excitatory synaptic transmission is tightly regulated by the number of active NMDARs and AMPARs at the synapse. NMDAR activation has a central role, as it can induce either long-term potentiation (LTP) or long-term depression (LTD), depending on the extent of the resultant intracellular calcium ([Ca²⁺]_i) rise in the dendritic spines and the downstream activation of specific intracellular cascades⁴⁴. Activation of synaptic NMDARs and large increases in [Ca²⁺]_i are required for LTP, whereas internalization of synaptic NMDARs, activation of perisynaptic NMDARs and lower increases in [Ca²⁺]_i are necessary for LTD. LTP induction promotes recruitment of AMPARs and growth of dendritic spines, whereas LTD induces spine shrinkage and synaptic loss⁴⁴.

Pathologically elevated Aβ may indirectly cause a partial block of NMDARs and shift the activation of NMDAR-dependent signaling cascades toward pathways involved in the induction of LTD and synaptic loss^{4, 6, 7}. This model is consistent with the fact that Aβ impairs LTP^{3, 29} and enhances LTD^{6, 31, 45} (Fig. 2). Although the mechanisms underlying Aβ-induced LTD have not yet been fully elucidated, they may involve receptor internalization^{6, 46} or desensitization⁴⁷ and subsequent collapse of dendritic spines^{6, 46}. Aβ-dependent effects on synaptic function may be mediated by activation of α7-nAChR⁴⁶, perisynaptic activation of NMDARs^{7, 31} and downstream effects on calcineurin–STEP–cofilin, p38 MAPK and GSK-3β signaling pathways^{7, 30, 31, 48}.

Recent findings suggest that pathologically elevated Aβ blocks neuronal glutamate uptake at synapses, leading to increased glutamate at the synaptic cleft³¹. A rise in glutamate would initially activate synaptic NMDARs, which might be followed by desensitization of the receptors and, ultimately, synaptic depression. A second effect of increased glutamate would be a spillover and activation of extra- or perisynaptic NR2B-enriched NMDARs, which have a key role in LTD induction⁴⁷. The activation of perisynaptic metabotropic glutamate receptors (mGluRs) may also be involved in the facilitation of LTD by Aβ^{6, 31} (Fig. 2a,b). Thus, Aβ-induced synaptic depression may result from an initial increase in synaptic activation of NMDARs by glutamate, followed by synaptic NMDAR desensitization, NMDAR and AMPAR internalization, and activation of perisynaptic NMDARs and mGluRs. Aβ-induced LTD-like processes may underlie Aβ-induced LTP deficits, as blocking LTD-related signaling cascades, such as those mediated by mGluR or p38 MAPK, prevents Aβ-dependent inhibition of LTP³⁰ (Fig. 2c).

Elevated Aβ destabilizes neural network activity

Physiological levels of Aβ may facilitate neuronal activity by presynaptic potentiation. This positive feedback loop is unlikely to escalate into aberrant network activity under normal circumstances, as increased neuronal activity would further enhance Aβ production, triggering negative postsynaptic regulation of excitatory synaptic transmission (Fig. 1). Dysregulation of Aβ in Alzheimer's disease could override these activity-dependent synaptic mechanisms, leading to synaptic failure and cognitive decline. Indeed, even small chronic increases in Aβ ultimately lead to synaptic depression³⁸. Pathologically elevated Aβ may also affect cognitive performance by inducing abnormal patterns of neuronal activity and compensatory responses at the level of neuronal circuits and networks⁴⁹. For the purposes of this discussion, we define neuronal circuits as smaller assemblies of interconnected neurons within a specific brain region and neuronal networks as larger assemblies of interconnected circuits involving different brain regions.

Our working model of Aβ-induced cognitive dysfunction proposes that high Aβ leads to aberrant excitatory network activity and compensatory inhibitory responses involving learning and memory circuits, and that both alterations contribute to cognitive decline^{8, 50}. Hippocampal compensatory responses may include calbindin depletion, GABAergic sprouting and ectopic expression of inhibitory neuropeptides^{8, 51, 52}.

Although the effects of Aβ on specific hippocampal glutamatergic synapses have been studied extensively, fewer investigations have focused on the effects of Aβ on neuronal circuits and more complex neuronal networks. Neuronal circuits are assembled through very large numbers of synaptic interactions between excitatory, inhibitory and neuromodulatory cells (Fig. 3a). The overall effect of Aβ probably depends critically on the abundance of Aβ at each synapse, the intrinsic vulnerability of each synaptic type, the circuit architecture and the engagement of 'nonphysiological' targets by high levels of pathogenic Aβ assemblies. It is possible that Aβ affects excitatory and inhibitory synapses differentially, which could produce complex imbalances in circuit and network activity.

Several recent reports in Alzheimer's disease–related mouse models suggest that pathologically elevated Aβ destabilizes neuronal activity at the circuit and network levels. We demonstrated by electroencephalogram (EEG) recording from cortical and hippocampal networks in hAPP transgenic mice that elevation of Aβ elicits epileptiform activity, including spikes and sharp waves⁸ (Fig. 3b). hAPP mice also have intermittent unprovoked seizures involving diverse regions of the neocortex and hippocampus that are not accompanied by tonic or clonic motor activity. These results demonstrate that chronic exposure to pathologically elevated Aβ is sufficient to elicit epileptic activity in vivo, a conclusion that is also supported by findings obtained in other hAPP lines^{53, 54, 55}.

These aberrant patterns of neuronal activity are associated with wide fluctuations in the neuronal expression of synaptic activity–regulated gene products, such as Arc and Fos, in the dentate gyrus⁸ (Fig. 3c). Consistent with these findings, in vivo calcium imaging of cortical circuits shows that hAPP and PS1 doubly transgenic (hAPP/PS1) mice have a greater proportion of hyperactive and hypoactive neurons than nontransgenic controls⁹ (Fig. 3d). Notably, these Alzheimer's disease–related mouse models have reduced glutamatergic excitatory currents and synaptic loss, suggesting that high Aβ leads to aberrant patterns of neuronal activity by enhancing synchrony among the remaining glutamatergic synapses rather than by increasing excitatory synaptic activity per se.

The processes described above would be expected to diminish the amount of time neural networks spend in activity patterns that promote normal cognitive functions (Fig. 3c). In this context, it is noteworthy that hippocampal alterations in synaptic activity-regulated proteins are tightly associated with learning and memory deficits in independent hAPP transgenic lines^{26, 51, 56}. Moreover, experimental manipulations that prevent seizure activity and compensatory responses in hAPP mice also prevent cognitive deficits in these models⁵⁷, suggesting that Aβ-induced aberrant network synchronization could contribute to cognitive impairments in Alzheimer's disease.

Although the incidence of seizures in individuals with late-onset Alzheimer's disease is clearly higher than that in age-matched undemented controls^{12, 58}, frank convulsive seizures are rare and only affect 5% to 20% of patients with Alzheimer's disease. In contrast, individuals from many pedigrees with autosomal dominant early-onset Alzheimer's disease show generalized convulsive seizures and myoclonic activity^{12, 59, 60, 61}. Seizures are part of the natural history of Alzheimer's disease associated with any one of over 30 different PS1 mutations⁵⁹ and have been observed in 31% of FAD patients with PS2 mutations⁶², 56% of patients with APP duplications⁶¹, ~83% of pedigrees with very early-onset Alzheimer's disease (<40 years)⁶⁰, and 84% of Down syndrome patients who develop Alzheimer's disease⁶³. The incidence of nonconvulsive epileptiform activity in early- or late-onset Alzheimer's disease is unknown.

Radiological studies have also provided evidence for abnormal network activity in Alzheimer's disease. Hypometabolism visualized by positron-emission tomography or single-photon-emission computed tomography and atrophy visualized by magnetic resonance imaging (MRI) are particularly prominent in posterior components of the 'default network'^{20, 64} (Fig. 4a). These alterations may reflect overall decreases in neuronal and synaptic activity but could also result from intermittent excesses in excitatory neuronal activity, which are often associated with decreased rather than increased cerebral metabolism⁶⁵. Consistent with the latter possibility, functional MRI (fMRI) studies have revealed aberrant increases in default network activity during memory encoding in subjects with Alzheimer's disease¹¹ (Fig. 4b).

Aberrant network synchronization

Abnormalities in synaptic activity could cause network instability and promote synchrony, which in turn can lead to epileptiform activity. It is also likely that Aβ-induced synaptic depression affects distinct types of synapses, neurons and brain regions differentially, which could further enhance imbalances and instability. Emerging evidence suggests that GABAergic dysfunction may be key in the pathogenesis of network dysfunction in Alzheimer's disease (Fig. 5).

An important clue came from the finding that hyperactive neurons in cortical circuits of hAPP/PS1 mice are associated with decreased GABAergic inhibition rather than increased glutamatergic transmission, suggesting impairments in GABAergic function⁹. Mouse models of apoE4, the main genetic risk factor for Alzheimer's disease, show prominent GABAergic dysfunction and impaired GABA release in the hippocampus⁶⁶. Another interesting report indicates that APP expression regulates GABAergic function by altering L-type calcium channels expressed by GABAergic neurons⁶⁷. Specifically, APP removal increases L-type calcium channel numbers and calcium currents in GABAergic interneurons in vivo, thereby enhancing GABAergic function and GABAergic plasticity. These effects are reversed by overexpression of APP. Thus, APP or Aβ may regulate ion channels that are critically involved in cellular excitability.

hAPP mice and humans with Alzheimer's disease have high levels of metenkephalin in the hippocampus and entorhinal cortex, which could block µ-opioid receptors on inhibitory interneurons and thereby disinhibit neuronal networks⁵². Pharmacological blockade of these receptors improves the memories of hAPP mice in the water maze⁵².

In addition to acting through neuronal mechanisms, Aβ may alter non-neuronal functions important for network stability. Spontaneous or neuron-induced rises in astroglial [Ca²⁺]_i can release glutamate and activate extrasynaptic NR2B subunit–containing NMDARs on neurons, promoting neuronal excitability and synchronized firing in hippocampal pyramidal neurons⁶⁸. Glutamate released from astrocytes can also act presynaptically on mGluRs, increasing the probability of transmitter release in glutamatergic terminals⁶⁹. Of note, astrocytes from hAPP/PS1 mice have synchronous hyperactivity in [Ca²⁺]_i transients across long distances that is uncoupled from neuronal activity⁷⁰. In addition, activated microglia in Alzheimer's disease generate inflammatory mediators, such as cytokines, that can enhance neuronal excitability⁷¹. Thus, diverse mechanisms could contribute to network dysfunction in Alzheimer's disease.

Unresolved issues and next steps

Despite advances in our understanding of Aβ-induced neuronal dysfunction, several issues remain to be addressed more conclusively.

ACKNOWLEDGEMENTS

REFERENCES