Introduction
Alzheimer’s disease (AD) is the most prevalent cause of dementia and currently no effective treatment exists. Multiple strands of evidence suggest that amyloid precursor protein (APP) and its proteolytic fragment, amyloid β-protein (Aβ), play a crucial role in the pathogenesis of AD [
62]. APP is a single-pass transmembrane protein enriched at synapses [
19]. The highly conserved APP gene is located on chromosome 21 and overexpression of APP in Down’s syndrome (trisomy 21) causes accumulation of amyloid plaques early in life [
21]. Through sequential enzymatic cleavage by β and γ-secretases, full-length APP is processed to yield amyloid beta (Aβ) as well as other fragments. Accumulation of fibrillar Aβ leads to formation of senile plaques, the typical neuropathological hallmark of AD. Soluble oligomeric Aβ, in contrast, is thought to mediate synapse dysfunction and loss, which strongly correlate with cognitive decline in AD [
20,
32]. The amyloid hypothesis takes the imbalance between Aβ production and clearance as the primary cause of AD [
20]. Based on this hypothesis and the discovery of familial AD mutations that facilitate Aβ production, transgenic mouse models overexpressing mutant APP and/or presenilins (PS), which form part of the γ-secretase complex, have been created to recapitulate AD pathology.
Among the APP transgenic mouse models, APP23 and APPswe/PS1deltaE9 (deltaE9) mice have been extensively used for exploring AD-related pathology and drug development [
63]. To recapitulate the pathogenesis of human AD, APP23 mouse model overexpresses human APP with the Swedish mutation under the murine Thy1 promoter [
57], while deltaE9 mice express APP with the Swedish mutation controlled by mouse prion protein promoter elements together with mutant human PS1 lacking exon 9, which is associated with familial AD [
29,
52]. Although these two transgenic mouse models display neuronal loss, cholinergic deficit, cognitive impairments, amyloid plaques and neuroinflammation in old age, the onsets of amyloid plaque formation and cognitive decline between them are very different in early adulthood [
5,
8,
9,
30,
38,
56]. Aβ deposits are not observed in APP23 mice younger than 6 months, but age-matched deltaE9 mice have already developed plaques [
28]. Despite the slower progress of amyloid plaque formation, APP23 mice show faster cognitive decline than deltaE9 mice. APP23 mice begin to develop cognitive deficits at 3 months, while deltaE9 mice do not have typical impaired memory until 1 year of age [
60,
61]. Uncovering and understanding the discrepancies between them are important for the utility of particular animal models to deepen our knowledge of synaptic failure in AD.
Using in vivo two-photon imaging of cortical layer V pyramidal neurons, we found reduced dendritic spine density in 4–5-month-old APP23 mice. In age-matched deltaE9 mice, loss of dendritic spines was only observed in close proximity to plaques. Furthermore, chronic in vivo imaging revealed that spine loss in AD transgenic mouse models was the consequence of decreased spine formation. Also, morphologies of dendritic spines in APP23 and deltaE9 mice were altered differently. Immunostaining showed accumulated intracellular APP in APP23 mice. The amount of intracellular APP was negatively correlated with spine density and morphology. These results suggest that spine abnormalities in young adult APP23 and deltaE9 mice might be caused by intracellular APP and extracellular Aβ deposits, respectively.
Discussion
Extracellular Aβ is accepted to be in the center of AD pathogenesis due to its neurotoxicity that disrupts multiple physiological processes [
53]. Guided by the amyloid hypothesis, AD mouse models have been created to recapitulate the cognitive impairments seen in AD patients. These mouse models typically express human APP with or without PS1 with familial AD mutations, which both cause familial forms of AD. Although most of the mouse models develop typical amyloid plaques and cognitive deficits with age, the pathophysiology in young transgenic mice, reflecting preclinical forms of AD, is less well understood [
63]. APP23 mice display cognitive impairments before plaque formation, while deltaE9 mice develop abundant plaques before the decline of cognitive performance. The underlying mechanisms of these discrepancies are still not clear.
The major correlate of cognitive impairment is synapse loss, which is closely associated with spine loss as excitatory glutamatergic synapses normally reside at dendritic spines in the mammalian brain [
43]. In addition to absolute spine density, the dynamic turnover of spines, termed structural plasticity, is also involved with learning and memory: the formation and elimination of dendritic spines rewire neural circuits by establishing or abolishing connections in the brain during learning experiences [
15]. Thus, it is plausible to examine alterations of dendritic spines as readout for structural correlate of cognitive decline in AD transgenic mouse models.
In this study, we found that 4–5-month-old APP23 mice displayed reduced spine density of cortical layer V pyramidal neurons. In deltaE9 mice, spine loss was only evident on dendrites that were located close to plaques. We found similar results in the APPswe/PS1L166P mouse model [
48], which accumulates plaques faster than the deltaE9 model: here, spines were lost only in the vicinity (<50 µm) of plaques, while spines were not altered distant (>50 µm) to plaques or before plaques had appeared [
3]. These results suggest that spine loss mediated by fibrillar amyloid plaques occurs only in the immediate vicinity of extracellular Aβ deposits in deltaE9 and APPswe/PS1L166P mice.
The decreased spine densities observed in APP23 and deltaE9 mice were caused by reduced spine formation as revealed by chronic repetitive in vivo two-photon imaging. Interestingly, we found two different patterns of spine morphological alterations in these two transgenic mouse models. In APP23 mice, the spine length was reduced and the relative proportion of stubby spines was increased. In deltaE9 mice, in dendrites close to plaques, the findings were identical. In contrast, the dendrites that were far away from plaques in deltaE9 mice showed decreased spine head width and elevated thin spine fraction. With amyloid plaque growth in deltaE9 mice, dendrites, that were originally located 50–80 µm away from plaques, became closer to plaques and started to lose spines. This effect was accompanied with an increase in the fraction of stubby spines. In APP23 mice, APP accumulated intracellularly. A higher content of APP was inversely correlated with spine density. Furthermore, an increased fraction of mushroom spines and decreased fraction of stubby spines were observed in neurons, which contained higher levels of intracellular APP. In summary, our data suggested that different pathological mechanisms, intracellular APP and extracellular amyloid plaques, might lead to spine abnormalities in young adult APP23 and deltaE9 mice, respectively.
Dendritic spines are the small bulbous postsynaptic elements of the majority of excitatory synapses and serve as the basic units for learning and memory [
22]. Loss of dendritic spines is the major correlate of cognitive impairment in human AD [
59]. In agreement with the spine loss described before, APP23 mice younger than 6 months show memory impairments in multiple cognitive tests, including Morris-type water maze test, Y-maze test, Barnes-maze test and novel-object recognition test [
12,
25,
31,
60]. On the other hand, the performance of deltaE9 mice at the same age is normal in most cognitive tests (T-maze test, Y-maze test, Morris-type water maze test, novel taste neophobia test, response acquisition test, Barnes-maze spatial memory task with hidden-target strategies), with the exception of impairments observed in Barnes-maze spatial memory task with cued-target strategies and modified radial-arm water maze test [
18,
34,
45,
49,
61]. The specific spatial learning deficit described in young deltaE9 mice may depend on dendritic spine shape, rather than a reduced spine number, considering that spine loss is only observed on dendrites that are localized very close to amyloid plaques, which just start to emerge in 4–5-month-old deltaE9 mice [
6,
16,
28]. With aging, Aβ deposits grow in size. Amyloid plaques mice are abundant in hippocampus and cortex of 1-year-old deltaE9 mice. At this age general axon degeneration and synapse loss are observed, along with impaired cognitive performance [
18,
46]. Thus, loss of synapses coincides with decline in cognitive performance in these models.
Indeed, there is convincing evidence that not only the absolute spine number contributes to cognitive performance. In fact, dendritic spine size and shape are known to affect various functional parameters relevant for cognition, including spine motility, neurotransmitter receptor numbers and organelle abundance [
33,
51]. Growing evidence shows that morphological changes of dendritic spines are associated with long-term synaptic plasticity (LTP) [
68]. LTP increases spine head volume while shortening and widening spine neck [
67]. This morphological plasticity allows generating changes in electrical properties of dendritic spines, which serve as isolated electrical compartments. For instance, it has been shown that shorter spine necks lead to larger depolarization while longer necks generate smaller somatic potentials [
1]. It is believed that different types of memories need to obey different electrophysiological rules, and thus require morphological diversities of spines [
51]. Along with changes in spine density, distinct alterations of spine morphology in APP23 mice and deltaE9 mice might also result in the different cognitive impairments described before [
12,
18,
25,
31,
34,
45,
49,
60,
61]. Layer V pyramidal neurons in the somatosensory cortex are involved in motor learning [
14,
64‐
66] and the formation of new dendritic spines correlates with the performance after learning [
66]. While most behavioral tests focus on hippocampus-dependent memory tasks, the resulting behavior results from a complex interplay of various brain regions, in which somatosensory cortex neurons may play crucial roles. Thus, the alterations of dendritic spines which we found may well reflect part of the behavioral phenotype observed in these mice. Yet, the susceptibility of spines to the various toxic insults due to the overexpression of APP and its cleavage products may differ between brain regions, between different functional locations within a neuron (e.g. between apical and basal dendrites) or with the age of the experimental animals. Therefore, the relation of dendritic spine loss in layer V pyramidal neurons to cognitive dysfunction is not certain.
Compared to APP23 mice, deltaE9 mice harbor an additional transgene of a familial AD mutation in PS1 with a deletion of exon 9, accelerating the cleavage of APP and thereby Aβ formation. In consistence with previous studies [
10,
16], extracellular amyloid plaques have developed in 4–5-month-old deltaE9 mice but not APP23 mice. Being the abnormal protein aggregates that characterize human AD, Aβ deposits are one of the biomarkers for AD neuropathologic assessment [
26]. Aβ production and aggregation might initiate serial molecular cascades, thus lead to clinical AD [
20]. This amyloid cascade hypothesis seems to be feasible in early-onset AD, which is known to be caused by mutations of genes that increase Aβ accumulation [
27]. However, as early-onset AD only accounts for a few percent of AD cases and the correlation between cognitive decline and Aβ deposits is weak [
2,
17], alternative explanations for the pathogenesis of AD have emerged [
36,
40].
In contrast to age-matched deltaE9 mice, only a minor soluble Aβ burden was found in the brains of young APP23 mice [
10,
37,
61]. Overexpressed APP in APP23 mice is predominantly localized intracellularly and the mechanisms of this aberrant accumulation and its relevance in sporadic AD need to be further investigated. Interestingly, a number of studies have reported increased amount of APP mRNA in AD patients [
39,
41,
47], indicating that up-regulated transcriptional activity of APP may also contribute to AD pathophysiology. Moreover, accumulated APP has been found in dystrophic neuritis of AD [
11,
54]. It is therefore tempting to speculate that intraneuronal accumulation of APP and/or its cleavage products including Aβ in AD may also contribute to synaptic damage [
44,
58]. Indeed, an extra copy of the APP gene can cause neuronal dysfunction and symptoms similar to those seen in AD [
42]. APP gene triplication in Down’s syndrome and APP locus duplication in rare families lead to clinical AD-like pathology in adults and result in early-onset dementia [
21,
50]. The neurotoxicity of APP is largely thought to be caused by its proteolytic fragments. Besides Aβ, other proteolytic APP fragments, such as C83, C99 and APP intracellular domain, could also be involved in AD pathogenesis [
55]. By regulating gene expression, these derivatives may give rise to neuronal degeneration [
35,
55]. Additionally, through the direct interaction between APP and
N-methyl-
d-aspartate receptors (NMDARs), overexpressed APP up-regulates the expression of NMDARs and thus may contribute to neuronal toxicity by disrupting synaptic homeostasis [
23].
To conclude, despite the fact that APP23 and deltaE9 mice show similar cognitive impairments and neuropathology in advanced age, our data clearly show different dendritic spine abnormalities in these two transgenic mouse models in young adulthood. Our findings imply that synaptic failure in these mouse models may be caused by different mechanisms in an age-dependent manner. Since the mechanisms underlying the development of sporadic AD are still uncertain, this study has significant implications for the analysis of distinct AD transgenic mouse models during preclinical drug evaluation for treatment of early-stage AD.