“Error is modern while truth is ancient” (Ramón y Cajal, 1928 [168])

When Ramón y Cajal pointed out “One also might imagine that amnesia, a paucity of thought associations, retardation, and dementia could result when synapses between neurons are weakened as a result of a more or less pathological condition, that is, when processes atrophy and no longer form contacts, when cortical mnemonic or association areas suffer partial disorganization” [167], he was probably the first to realize that dementia results from a dysfunction of synaptic contacts.

In the late nineteenth century, some scientists began to speculate about how neuronal extension and retraction, even in the adult brain, could relate to behavior [45, 49, 50, 108, 118–121, 165, 166, 210, 215, 220]. One particularly attractive idea was that axons and dendrites of different cells could grow closer to each other at frequently used junctions to enhance communication. This sort of action, some scientists thought, could account for learning and perhaps even the association of higher ideas. Rabl-Rückhard [165] and others [49] advanced the idea that certain conscious activities may be interpreted in terms of neuronal "ameboid" movements that could account for learning and memory. Forgetting, on the contrary, was believed to involve a “spreading of the gap.” Disuse, diseases of the nervous system, and aging were believed to be the factors that could cause processes to pull away from each other at the intercellular junction. The idea that intercellular junctional growth could account for memory was first anticipated in 1872 by Alexander Bain [14]. With this idea, however, Bain was much ahead of his time, and contemporaries did not give his theory serious consideration. In the 1890s, Ramón y Cajal similarly believed it probable that information could be stored by modifying interneuronal connections and that mental excercise leads to greater growth of neuronal collaterals in the stimulated regions of the brain [166, 167]. He also proposed that the proliferation of neural connections in the cerebral cortex may correlate with intelligence. In the Croonian Lecture given in England in 1894, he surmised that “brain jogging” (“gymnastique cérébrale”) may increase the growth of certain dendrites and the branching of axon collaterals [166]. Intelligence, he argued, can be achieved economically, without adding more neurons and without demanding more space, by increasing dendrite and axon branching. Only a few years later, in 1897, Charles Scott Sherrington coined the term "synapse" to describe the junction between nerve cells.

Synapse loss is the major neurobiological substrate of cognitive dysfunction in Alzheimer’s disease

More than 100 years after Ramón y Cajal’s original suggestion, his ideas on the neurobiological substrate of amnesia and dementia have basically been proven [212]. Data obtained by electron microscopy [40, 42–44, 182–184], immunocytochemical and biochemical analyses on synaptic marker proteins in AD biopsies and autopsies [47, 59, 83, 140, 211] indicate that synaptic loss in the hippocampus and neocortex is an early event [81, 135] and the major structural correlate of cognitive dysfunction [19, 27, 42–44, 63, 66, 72, 80, 83, 106, 130, 134, 158, 228].

Quantitative ultrastructural studies performed on temporal and frontal cortical biopsies within 2 to 4 years after the clinical onset of the disease have revealed a 25% to 35% decrease in the numerical density of synapses and a 15% to 35% loss in the number of synapses per neuron [40]. Synaptic loss might even be more pronounced in the hippocampus where it amounts to 44% to 55% [183].

From all cortical areas analyzed, the hippocampus appears to be the most severely affected by the loss of synaptic proteins, while the occipital cortex is affected least [22, 25, 26, 33, 38, 39, 56, 59, 79–81, 83, 106, 113, 117, 135, 139, 140, 157, 196, 206, 207, 211]. In other words, there appear to be regional differences within the cerebral cortex with respect to the severity of synaptic marker loss that basically match the pattern of neurofibrillary degeneration as outlined in the Braaks’ staging [24]. Further, the highly correlative relationship between loss of synaptic markers and neurofibrillary tangle counts in the same brain region [79, 83, 117, 132, 180, 221] also supports this link between tangle formation and synaptic dysfunction. More detailed studies on the relationship between synaptophysin mRNA and tangle formation at the level of individual neurons directly proved a significant reduction of synaptophysin mRNA in neurons affected by tangle formation [29, 30].

On the contrary, results of studies on plaques are less consistent. While a few studies [106, 180, 207] described a link between higher plaques counts and lower synaptic protein measures, other studies failed to establish such a relationship [79, 80, 83, 117, 141].

Most studies analyzing the link between premortem severity of cognitive impairment and synaptic pathology in AD agree on the correlation between lower synaptic protein levels and some aspect of cognitive dysfunction [22, 79, 106, 131, 132, 133, 149, 180, 181, 206, 207, 211, 221], although in several studies, correlations between synaptic proteins and cognitive functions are region-specific.

The reduction in synaptic vesicle proteins in AD is likely related to the clinical symptoms of dementia, given their function in vesicle trafficking, docking and fusion to the synaptic membrane and neurotranmitter exocytosis. Synaptic pathology in AD is reflected by a loss of all major components of small synaptic vesicles that accommodate classical neurotransmitters and large dense core vesicles that store most peptides, together with a loss of molecular components of pre- and postsynaptic compartments accompanied by extensive pathological changes of the synapse [38, 84, 95, 105–107, 109, 128, 170, 194, 195, 206, 207, 221, 233] (Table 1).

Table 1 Alterations of synaptic proteins in AD
Full size table

Although degeneration of subcortical input might contribute to cortical synapse loss [7], most of the synaptic loss in the neocortex might derive from loss of cortico-cortical associational fibers [86, 110, 136, 137, 150].

Synaptic failure is an early event and accelerates slowly in a process of dynamic reorganization

Synapse formation and stabilization in the nervous system are dynamic processes and so is the synaptic degeneration in AD. AD is a slowly progressing disorder apparently preceded by a clinically silent period of several years or even decades. Similarly, synaptic degeneration might be a slow process progressing from an initially reversible functionally responsive stage of down-regulation of synaptic function to stages irreversibly associated by marked synapse loss [169].

Disturbances of synaptic integrity can be detected already in patients with mild cognitive impairment (MCI), a prodromal state of AD [133, 184], suggesting that synaptic degeneration occurs very early on in the process of AD. Ultrastructural stereological investigations on rapid postmortem autopsy samples revealed an 18% synapse loss in the hippocampal CA1 region of MCI patients that progressed to a 55% synapse loss in mild AD [183]. Although with progression of the disease there is a steady decline in synaptic population, eventually accompanied by the loss of about 10-20% of cortical neurons [94, 130], recent studies suggest a biphasic process with an initial rise of synaptic markers in early stages of the disease, suggesting synaptic reorganization. An initial transient rise, for example, was observed for synaptophysin and other presynaptic proteins in cortical association areas at Braak stage III, i.e., prior to neurofibrillary pathology [153]. Other studies indicated an increase in drebrin in the frontal cortex of patients with MCI followed by a 40% to 60% decrease in severe AD [36], an increase in the expression of postsynaptic density protein PSD-95 in AD brain [109], as well as an elevation in glutamatergic presynaptic bouton density in midfrontal gyrus of MCI patients [16]. Functional magnetic resonance imaging (fMRI) on MCI patients similarly suggests that there might be a phase of paradoxically increased activation early in the course of prodromal AD [200].

Dynamic synaptic reorganization during the process of degeneration is further supported by observations of an increase in synaptic size that accompanies synaptic loss in various cortical regions in AD and might reflect the attempt of a functional compensatory increase of synaptic efficacy [18, 44, 182]. As a result, the total synaptic area per unit tissue volume is initially preserved, but decreases thereafter, paralleled by cognitive decline [44, 138, 211]. Synaptic compensatory mechanisms that in normal aging [15, 19, 20, 28, 55] succeed in preserving considerable cognitive function are, thus, disrupted in AD. At least in early stages of the disease, processes of compensation and decompensation might be present at the time in the same region as well as in different brain regions, which might to some extent obscure direct linear relationships between brain pathology and functional measures.

Findings of a deregulation of proteins involved in structural plasticity of axons and dendrites [1, 51, 77, 93, 116, 146] as well as results of computational studies [76, 85] indicate a failure of local neuronal regulatory mechanisms of synaptic plasticity and make a primary disturbance of synapse turnover very likely. This assumption is further supported by alterations in the composition [20, 67, 102, 126, 205] and fluidity of membranes [53, 238] as well as by direct morphological evidence of a disturbed axonal and dendritic remodeling (for review, see [4, 5]).

Recent studies applying functional magnetic resonance imaging (fMRI) provide further evidence of disrupted organization of functional brain connectivity in AD. AD patients have significantly lower regional connectivity and show disrupted global functional brain network organization when compared to healthy controls [199, 224]. Reduced functional connectivity of the hippocampus can already be obtained in MCI patients compared to normal aging control subjects [13, 92]. In AD patients, cognitive decline is associated with disrupted functional connectivity in the entire brain [204].

Mechanisms of synaptic failure in Alzheimer’s disease, potential roles of Aβ, APP and tau

Synaptic loss is currently the best neurobiological correlate of cognitive deficits in AD. In addition to the synapse loss due to the death of neurons, there is evidence that still living neurons lose their synapses in AD [34]. Further, synaptic function is impaired in living neurons as demonstrated by decrements in transcripts related to synaptic vesicle trafficking [34]. The question, thus, has been addressed to what extent the established amyloid and tau pathology might contribute to synaptic dysfunction in AD.

Disruption of synaptic function by Aβ oligomers

Although Aβ might show some deleterious cellular effects, brain amyloid load does not correlate well with synaptic loss, neuronal death or cognitive dysfunction ([211], see above). Defects in synaptic transmission, furthermore, occur in AD well before the formation of amyloid plaques and neurofibrillary tangles [4, 5, 223]. It might, thus, be argued that soluble molecular species that are generated at very early stages of the disease and that only at more advanced stages are deposited in an aggregated form could be involved in synaptic failure. It has, thus, been suggested that soluble assembly states of Aβ peptides can cause cognitive problems by disrupting synaptic function in the absence of significant neurodegeneration. There is evidence that soluble oligomers of Aβ can selectively impair synaptic plasticity mechanisms necessary for memory processing [35, 71, 176]. Both cell-derived and synthetic soluble Aβ oligomers can disrupt hippocampal long-term potentiation in slices and in vivo, and can also impair the memory of complex learned behavior in rodents and decrease dendritic spine density in organotypic hippocampal slice cultures [48, 98, 99, 104, 156, 190, 192, 222]. Intracerebroventricular injections of soluble synthetic Aβ1-40-dimers rapidly inhibit the plasticity of excitatory synaptic transmission at doses of 10-42 pmol comparable to natural Aβ [87]. Other studies, however, have suggested opposite effects with picomolar levels of Aβ1-40 playing a neurotrophic role in cell cultures [161, 231]. Also, low picomolar concentrations of preparations containing both Aβ1-42 monomers and oligomers have been shown to be able to induce a marked increase in hippocampal long-term potentiation and produce a pronounced enhancement of both reference and contextual fear memory [164]. Current findings on the effects of soluble Aβ oligomers on synaptic function are, thus, not entirely conclusive. Still, a recent study applying array tomography, a technique that combines ultrathin sectioning of tissue with immunofluorescence, to a mouse model of AD provides compelling evidence that senile plaques are a potential reservoir of synaptotoxic oligomeric Aβ [100].

Synaptotrophic function of human wild-type APP and its failure in FAD-mutated APP

Disturbances of synaptic function and plasticity have also been observed in transgenic mice overexpressing FAD-mutated APP [32, 54, 62, 65]. APP is a type I transmembrane protein that belongs to a conserved family including Apl-1 in Caenorhabditis elegans [37], APPL in Drosophila [129, 173] and APP [209], APP-like protein 1 (APLP1) [226] and APLP2 [197, 227] in mammals. Within the brain, APP can be detected in synaptic membranes [97] and has been shown to localize to postsynaptic densities, axons and dendrites [187, 193]. APP undergoes fast axonal transport [101, 178, 229] and is targeted to synaptic sites [187, 193]. Studies on overexpression or knockout of APP and its homologues indicate a critical role in neuronal survival, neurite outgrowth, synaptogenesis and synaptic plasticity [41, 115, 148, 152, 191].

The proteolytic processing of APP has been investigated extensively, and numerous studies have focused on elevations in the Aβ-peptide, which is generated during normal metabolism of APP [70] and forms the major component of plaques, a major hallmark of the disease [142]. While mechanisms of APP processing giving rise to Aβ have received the most attention, the physiological function of APP and potential sequelae of impaired function are less well understood.

Several genetic knockout models for APP and its homologues have been generated to gain insight into its function [78, 219, 237]. Triple knockouts for APP and its homologues are lethal and show cortical dysplasia [82]. Studies on the Drosophila APP homologue APPL support a potential role in axonal arborization, axonal transport and synapse formation both during development and in the mature nervous system [69, 111, 122, 214]. APPL does not contain an Aβ-peptide sequence, suggesting that the conserved physiological function of APP does not involve Aβ. Mice deficient in APP show abnormalities in expression of synaptic markers and in axonal and dendritic arborization, together with neurological and behavioral dysfunction and impaired long-term potentiation [41, 124, 159, 188, 237]. During rat brain development, expression of APP peaks in the second postnatal week, the time of synapse formation. After development is completed, high expression levels of APP persist in the adult olfactory bulb, where continuous synaptogenesis occurs in the adult animal [114, 179], and its expression level increases in animals reared in enriched environments [89]. This all suggests an involvement of APP in the process of cell differentiation and the establishment and plastic maintenance of synaptic contacts [68].

Evidence for synaptotrophic effects of APP were, in addition, obtained by a variety of transgenic approaches generating different lines of mice overexpressing different hAPP isoforms under control of different promotors [131, 151, 171, 214]. It had not been analyzed previously, however, whether this synaptotrophic effect is maintained in mutated forms of APP that show linkage to familial forms of AD (FAD) and thus have been identified as one potential cause of the disease [64].

In a recent study, we analyzed whether familiar AD (FAD)-linked mutations of APP might impair synaptotrophic function, potentially contributing to synaptic deficiencies seen in AD. We therefore performed a quantitative electron microscopy study on synapse number in the cerebral cortex of well-characterized expression-matched transgenic mouse lines expressing either wild-type human (h)APP or FAD-mutated hAPP [152], using unbiased stereological methods. We could obtain clear evidence for a synaptotrophic effect in mice overexpressing wild-type hAPP demonstrated by an increase in synaptic number [189]. This effect was abolished when FAD-mutated APPsw,Ind was expressed instead of wild-type APP. In agreement with previous reports on reduced synaptic vesicle number in APP/APLP2 knockout mice [225, 230], we further observed increased density of synaptic vesicles in mice overexpressing wild-type hAPP. Again, this effect was abolished when FAD-mutated hAPP was expressed instead of wild-type hAPP.

Taken together, this strongly indicates a failure in synaptotrophic function of FAD-mutated forms of APP, which likely contributes to synaptic pathology in AD. Thus, at least in familiar cases of AD, not only “too much Aβ” but also “too little functional intact APP” might be of potential pathogenetic significance for synaptic loss. It remains to be determined whether a similar mechanism might be relevant also for sporadic cases of AD.

Studies on APP fragment function have demonstrated that both sAPPalpha and the APP intracellular domain may have neurotrophic properties and enhance synaptic plasticity and memory [2, 57, 91, 111, 123, 143–145, 151, 171, 198, 213]. Intracerebral administration of sAPP significantly increases synaptogenesis, reduces neuronal injury and improves functional motor outcome following brain injury in rats [17, 213], an effect that likely depends on a conserved motif in the C-terminus [111]. On the other hand, activity-dependent cleavage of APP by BACE enhances short-term and long-term synaptic plasticity. This effect correlates with elevated levels of the APP intracellular domain, which has been implicated in the regulation of gene transcription and calcium signaling [123].

Recent evidence suggests that APP and neurotrophic factors such as NGF and BDNF use similar intracellular pathways to control neuronal plasticity [3]. The low-affinity neurotrophin receptor p75 shares similarities in processing with APP. After initial cleavage by alpha secretase, p75 is cleaved by gamma secretase [234] in an event identical to the cleavage of APP. Both NGF and BDNF enhance APP promoter activity in a process that involves activation of the ras-MAP-kinase pathway [21, 174, 175, 177, 216]. As we showed previously, the ras-MAP-kinase pathway is induced already at very early stages of AD prior to any noticeable pathological alterations, such as plaque or tangle formation [60, 61]. It might, thus, be suggested to regard the activation of this signaling pathway as an, apparently ineffective, attempt to compensate for the strongly attenuated synaptotrophic effects of APP. As the ras-MAP-kinase pathway also mediates mitogenic effects eventually resulting in activation of the cell cycle (see below), a scenario can be envisaged where a disturbed synaptotrophic action of APP triggers an aberrant activation of intracellular signaling cascades that eventually induce cell cycle activation and subsequent cell death [12, 155, 217].

Synaptic disconnection in a model of cerebral hypometabolism is associated with PHF-like phosphorylation of tau

Recently, we have demonstrated that hypometabolism during topor in hibernating animals is associated with a PHF-like pattern of phosphorylation of the microtubule-asssociated protein tau, a process believed to be critically involved in the mechanism of neurofibrillary degeneration in AD [11, 74, 201, 202]. Furthermore, the stage of torpor in hibernating animals shows significant analogies to the pathophysiological condition of AD with respect to an altered synaptic connectivity [11, 125, 162, 163, 203, 218] and the impairment of cognitive function [147, 201].

Depression of the metabolic state of neurons during torpor in hibernating mammals leads to greatly reduced electroencephalographic activity [58, 103]. As activity is a measure of use, and neuronal connections remain functional through regular use, this decrease negatively affects the maintenance of neuronal connections [96]. The hibernation cycle thus represents a physiological model that allows the study of sequelae of reduced neuronal connectivity. Synaptic regression during torpor and subsequent reinnervation in phases of arousal have been particularly well characterized for mossy fibers terminating on CA3 hippocampal pyramidal neurons [11, 90, 162, 163]. Stages of synaptic disconnection are associated with the formation of PHF-like phosphorylated tau in CA3 pyramidal cells, which lose their afferentation. This PHF-like tau phosphorylation is quickly and fully reversible during arousal when mossy fibers re-connect to pyramidal neurons. These findings implicate a critical link between a dysfunction and/or loss of synaptic afferentation and PHF-like phosphorylation of tau.

A direct link between hyperphosphorylated tau and synaptic pathology is also supported by recent reports on an accumulation of abnormally phosphorylated tau species within synaptic terminals in AD brains and APP Swedish mutant transgenic mice [154, 208].

Synaptic plasticity and cell cycle activation in neurons are alternative effector pathways

To form synaptic connections and to continuously re-shape them in a process of ongoing structural adaptation, neurons must permanently withdraw from the cell cycle. That means synaptic plasticity can only occur at the expense of the ability to proliferate. In the “Dr. Jekyll and Mr. Hyde concept,” we have formulated the hypothesis that differentiated neurons after having withdrawn from the cell cycle are able to use molecular mechanisms primarily developed to control proliferation alternatively to control synaptic plasticity [5]. The existence of these alternative effector pathways within a neuron might put it at risk of erroneously converting signals derived from plastic synaptic changes into the program of cell cycle activation that subsequently leads to cell death. The molecular mechanisms involved in cell cycle activation might, thus, link synaptic plasticity to cell death [5, 6].

Up-regulation of a variety of molecules critically involved in the activation and progression of the cell cycle, indicating a cell cycle re-entry of neurons, occurs at early phases of neurodegeneration in AD [12, 155, 217] (Fig. 1). This cell-cycle re-activation most likely is a down-stream effect of aberrantly activated mitogenic signalling pathways [10, 60, 61]. The p21ras-MAP-kinase pathway, a mitogenic pathway that in cycling cells controls proliferation, in the adult nervous system regulates neuronal plasticity of differentiated neurons [9]. These observations provide direct evidence that depending on the cellular context, cell cycle activation and plasticity might involve identical molecular pathways. In AD, these pathways are upregulated very early during the course of the disease. This activation can be found, for example, in the frontal isocortex, as early as Braak stage I-II, i.e., prior to any other noticeable sign of pathology [10, 60, 61].

Fig. 1
figure 1

According to a hypothesis we proposed several years ago [5], neurons might have evolutionarily acquired the ability to use molecular mechanisms primarily developed to control proliferation alternatively to control synaptic plasticity. Cell cycle regulation and control of synaptic plasticity might, thus, be alternative effector pathways of “canonical” cell cycle regulating molecules. During neuronal differentiation a switch might take place from the control of the cell cycle to the control of synaptic plasticity. This switch might be reversed during degeneration, an event that might be critical for cell death

Full size image

Cell cycle regulators might serve non-canonical functions in differentiated neurons: linking synaptic plasticity to cell cycle

Recent studies indicate that, contrary to classical beliefs, molecules known to be involved in activation and progression of the cell cycle are not entirely repressed in differentiated neurons in the adult nervous system where they might serve in alternative “non-canonical functions” such as regulation of neuronal and synaptic plasticity [185, 186]. A functional link between cell cycle regulation and synaptic changes potentially requires a signaling mechanism between synaptic terminals and gene regulation. Such a mechanism has indeed been identified in Drosophila, where Latheo might serve as an information shuttle between the nerve terminal and the nucleus participating in synapse-to-nucleus sigaling [52]. Latheo is present in the cytoplasm of postmitotic neurons and is also abundant in boutons of presynaptic terminals at the Drosophila neuromuscular junction, far from the nuclei [172]. It regulates both evoked transmission amplitude and activity-dependent forms of synaptic facilitation and potentiation [172] and has been implicated in learning [23, 160, 172]. Latheo is a homologue of ORC3, a component of the origin recognition complex (ORC), a critical “guard” of DNA replication that controls initiation of DNA replication and prevents re-replication during the cell cycle [88]. These data show that Latheo might play a dual neuronal role: a nuclear role in DNA replication/transcription and a role in synaptic plasticity. A similar role was shown for ORC subunits in the mammalian brain. ORC3 and ORC5 loss of function phenotypes in hippocampal pyramidal neurons induced by siRNA in mice neuronal cultures revealed a regulation of dendrite and spine development [88]. These data directly support our hypothesis that neurons have evolutionarily acquired the ability to use molecular mechanisms primarily developed to control proliferation alternatively to control synaptic plasticity [5]. More recently, we could provide evidence for an involvement of ORC units in AD pathology [8]. In AD, ORC units show pathological alterations of their subcellular compartmentation, basically reflected by a close association with neurofibrillar tau pathology in the form of neurofibrillary tangles, neuropil threads and plaque-associated dystrophic neurites. This abnormal compartmentation might segregate these regulatory elements from their physiological function in regulating plasticity and gene silencing and will, thus, potentially result in de-repression of genes triggering an apopototic phenotype.

Taken together, synaptic degeneration is the major neurobiological substrate of cognitive dysfunction in AD. Synaptic failure occurs very early in the course of the disease and progresses slowly in a process of dynamic reorganisation. Although the cause for this failure is still unknown, recent evidence indicates a link between plastic synaptic changes and control of differentiation and cell-cycle-repression within a neuron. We have thus put forward the hypothesis [5] that molecular mechanisms are shared between control of synaptic plasticity and control of the cell cycle, and as a consequence of this link, attempts to compensate for synapse loss may activate the cell cycle, which finally may lead to cell death. It will thus be the challenge for future therapeutic approaches to lock the neurons in a differentiated stage but still in a highly plastic phenotype.