AD is distinguished histopathologically from other dementias by abundant extraneuronal deposits of amyloid β-protein (Aβ). Numerous reports describe neuronal alterations induced by supraphysiological concentrations of synthetic Aβ peptides, by Aβ species secreted by cultured cells, or by complex mixtures of Aβ assembly forms in the brains of APP transgenic mice1–5
. While these findings demonstrate that Aβ can alter synapse physiology in experimental models, the nature of the pathogenic species in the human brain and direct demonstration of its neurobiological effects are unresolved.
Aqueously soluble (Tris-buffered saline (TBS)), detergent-soluble (TBS+1% Triton) and “insoluble” (5M GuHCl) extracts were prepared by sequential centrifugation of brain homogenates from humans with various neuropathologically confirmed dementias (Supp. Table 1a
). Sensitive immunoprecipitation/Western blotting (IP/WB)5,6
revealed Aβ monomers and lithium dodecylsulfate (LDS)-stable dimers and trimers in all three extracts of the frontal and temporal cortices of AD subjects and an adult with Down’s syndrome and AD (). Cortical extracts from some non-AD subjects showed modest levels of Aβ in the insoluble (GuHCl) extracts (,) but little or none in the soluble (TBS) extracts () compared to the AD cases. Notably, a subject with AD histopathology but no clinical AD (pathological AD, P-AD) showed Aβ in the insoluble but not the soluble fraction. While Aβ was detectable in all three sequential extracts, we chose to characterize the physiologic effects of the TBS-soluble fraction because AD dementia correlates strongly with soluble Aβ levels7–9
. Indeed, the profile of our extracts suggested that levels of TBS-soluble Aβ correlated best with the clinical AD state (). Moreover, we wished to focus on the earliest Aβ assemblies: soluble oligomers that form initially from monomers.
Figure 1 Monomeric and oligomeric Aβ is detected in brain extracts of humans with clinically and neuropathologically typical late-onset AD. IP/WB analysis (see Methods) was performed on supernatants of the soluble (a, TBS), membrane-associated (b, TBS-TX100) (more ...)
We first asked whether soluble Aβ from AD cortex (; Supp. Table 1b
) alters long-term potentiation (LTP) in mouse hippocampus. TBS extracts from control (Con TBS) or AD (AD TBS) cortex did not alter basal synaptic transmission or paired-pulse ratio (Supp. Figs. 1a,b
), indicating that neurotransmitter release probability was unaffected10
. Slices exposed to TBS vehicle (Veh) or Con TBS for 20 minutes exhibited robust LTP induction following high-frequency stimulation (HFS) (152.9 ± 9.1% and 144.2 ± 7.1% of baseline fEPSP slope, respectively) (). In contrast, AD TBS inhibited LTP (111.3 ± 3.9%, P
<0.05) (). Immunodepleting AD TBS with an Aβ antiserum (R1282) prevented the LTP inhibition (), indicating that Aβ was necessary. The effect of AD TBS on LTP was strongly dose-dependent (Supp. Fig. 1d
). Importantly, TBS extracts prepared identically from FTD or DLB cortices did not significantly alter LTP (137.0 ± 5.3% and 148.1 ± 6.1%, respectively) (; Supp. Fig. 1e
). Additional brain extracts from two control and three AD subjects fully replicated the above findings ().
Figure 2 Soluble Aβ extracted from AD brain alters hippocampal synapse physiology and learned behavior. (a) IP/WB of the TBS brain extracts used to study LTP. Clinical and neuropathological diagnoses for each of these seven cases are provided in Supp. (more ...)
Long-term depression (LTD) of hippocampal synapses is induced by repetitive subthreshold stimulation11
. Standard protocols for LTD induction in adult rodent hippocampus require delivery of 600–900 pulses at low frequency12,13
. Accordingly, 300 pulses at 1 Hz failed to induce LTD in the presence of vehicle or Con TBS (). However, AD TBS facilitated LTD induction by this weak stimulus (74.7 ± 4.8% of baseline for AD TBS vs. 101.9 ± 5.6% for Con TBS, P
<0.05) (). LTD induced with AD TBS was NMDAR-independent, as the NMDAR antagonist AP-V did not block this effect (68.1 ± 4.3%) (). However, both MCPG, a Group I/II mGluR antagonist (94.8 ± 2.4%, P
<0.05), and SIB1757, an mGluR5 antagonist (101.1 ± 6.9%, P
<0.05), prevented LTD facilitation by AD TBS (). Whereas mGluR activation was necessary for the LTD facilitation by soluble Aβ, SIB1757 did not prevent AD TBS-mediated LTP inhibition (Supp. Fig. 1f
). This finding is consistent with earlier data that Aβ can influence synaptic plasticity through various receptors, including NMDAR, mGluR and nicotinic acetylcholine receptors14–17
Passive administration of monoclonal Aβ antibodies has entered human testing. We found that the ability of a co-administered Aβ antibody to block the above LTD facilitation correlated with its ability to IP soluble Aβ from AD TBS (Supp. Fig. 2
). Antibodies to the free N-terminus of Aβ (3D6; 82E1) almost completely precipitated soluble Aβ from AD TBS and also prevented the LTD facilitation (98.4 ± 3.0%), whereas antibodies to the Aβ C-termini (2G3, 21F12) weakly precipitated Aβ and did not block the LTD effect (72.1 ± 4.9%) (Supp. Fig. 2a,b
). Aβ mid-region antibodies IP’d a fraction of the Aβ species in AD TBS and only partially blocked the LTD effect (Supp. Figs. 2a, c
). Similarly, N-terminal but not C-terminal antibodies neutralized the LTP deficit (Supp. Fig. 2d
To assess the effects of soluble AD cortical extracts directly on memory function, rats were trained on a step-through passive avoidance task18
. At 0, 3 or 6 hr post-training, AD TBS or R1282-immunodepleted AD TBS (AD TBS-ID) (Supp. Fig. 3a
) was microinjected into the lateral ventricle. AD TBS administered 3 hr post-training significantly impaired the animals’ recall of the learned behavior 48 hr later (). The latency to enter the dark chamber, where the rat had received a shock during training, was significantly shorter for animals injected with AD TBS than with AD TBS-ID. Notably, AD TBS injected at 0 or 6 hr after training did not significantly alter the escape latency (Supp. Fig. 3
). The 3 hr post-training time point at which AD TBS significantly impaired recall is consistent with the temporal pattern of transcriptional regulation of synapse remodelling following passive avoidance training19
Decreased synapse density is the strongest neuropathological correlate of the degree of dementia in AD20
. To determine whether soluble Aβ in AD brain contributes directly to synapse loss, we quantified dendritic spine density in GFP-transfected pyramidal cells in organotypic rat hippocampal slices21
. To properly reconstitute brain extracts in slice culture medium, TBS extracts underwent non-denaturing size exclusion chromatography (SEC). Pyramidal neurons in slices cultured for 10 days with plain medium (sham) or medium reconstituted with lyophilized SEC fractions of Con TBS (Con TBS-SEC) displayed similar spine densities (0.79 ± 0.02 and 0.86 ± 0.03 spines/μm, n
=6/890 and 5/628 cells/spines, respectively). In contrast, slice medium reconstituted with SEC fractions from AD TBS (AD TBS-SEC) caused a 47% decrease in spine density vs. Con TBS-SEC (0.46 ± 0.03 spines/μm, P
=6/517) (; Supp. Fig. 4
). MCPG did not prevent the loss of spines with AD TBS-SEC treatment (0.45 ± 0.03 spines/μm; n
=5/337) (). CPP, an NMDAR antagonist, did not alter spine density when applied alone but prevented the decrease observed with AD TBS-SEC (0.73 ± 0.03 and 0.84 ± 0.03, respectively; n
=5/619 and 5/748; P
<0.05 for AD TBS-SEC alone vs. AD TBS-SEC with CPP) (). These findings support prior evidence that NMDAR activation is necessary for Aβ-mediated spine loss16,17
We next asked which soluble Aβ species present in AD brain mediated these effects on synapse physiology. Two lines of evidence indicated that the Aβ-immunoreactive species migrating at 8 kDa on LDS-PAGE gels were true Aβ dimers. First, mass spectrometry of the 4 and 8 kDa bands IP’d from the GuHCl extract of AD cortex confirmed that each contained tryptic peptides of human Aβ (Supp. Fig. 5b–e
). Second, IP of this extract with an Aβ40-specific antibody (2G3) and WB with an Aβ42-specific antibody (21F12) revealed an Aβ40/42
heterodimer migrating at 8 kDa (Supp. Fig. 5a
). Performing this co-IP using 21F12 for both IP and WB yielded a much stronger dimer signal, indicating that most of the 8 kDa species are Aβ42/42
homodimers (Supp. Fig. 5a
Having confirmed that the 8 kDa bands detected by WB in AD brain samples ( and ) are bona fide
Aβ dimers, we used non-denaturing SEC to separate the various Aβ species in AD TBS and characterized their respective effects on LTP. Most of the Aβ in AD TBS eluted in the void volume (>60 kDa, based on co-eluting linear polydextran standards22
), but this higher MW complex dissociated into Aβ monomers and dimers when denatured by LDS-PAGE (, fractions 3–4). The SEC profile also showed dimers eluting at ~8–16 kDa (fractions 7–8) and monomers at ~3–6 kDa (fractions 10–11). Taken together, these results indicate that in AD cortex, soluble Aβ exists in various assemblies, with the smallest native oligomer being a dimer.
Figure 3 Soluble dimers are the smallest Aβ assembly form in human brain to acutely perturb synapse physiology. (a) TBS extracts of AD (top panel) and control (bottom panel) brains were subjected to non-denaturing SEC. SEC fractions were lyophilized and (more ...)
To establish which soluble Aβ species were responsible for the impaired synaptic plasticity, SEC fractions of AD TBS containing either higher MW complexes (AD SEC 4), native Aβ dimers (AD SEC 8) or monomers (AD SEC 10) were each tested separately. Only AD SEC 8 significantly inhibited LTP (107.0 ± 2.2%; P
<0.05 vs. Con TBS), whereas AD SEC 4, AD SEC 10, and identically prepared fractions from Con TBS were all inactive (). Notably, AD SEC 4 fraction contained the highest concentration of Aβ (), suggesting that the specific activity of this higher MW Aβ assembly is very low. To achieve a purer preparation of Aβ dimers, AD TBS was IP’d with 3D6, eluted with denaturing LDS buffer and subjected to SEC (IP-SEC) (). Most of the soluble Aβ now eluted at the size of dimers (fractions 7–8) rather than in the void volume (Supp. Fig. 6
), suggesting that elution with LDS disrupts non-covalent interactions among the higher MW Aβ assemblies. LTP was significantly inhibited by IP-SEC fractions 7–8, containing Aβ dimers (106.6 ± 2.4%, P
<0.05), but not by fractions 10–11 containing monomers nor by any IP-SEC fractions from Con TBS ().
Although these SEC experiments show that soluble Aβ dimers inhibit LTP, it remained possible that a small molecule from human brain was bound to the Aβ dimers and was required to impair LTP. To address this possibility, we generated a synthetic Aβ40 peptide in which serine 26 was mutated to cysteine (Aβ40-S26C). An Aβ dimer was observed upon oxidation (), and this inhibited LTP nearly 20-fold more potently than did wild-type synthetic Aβ40 (). This pure, synthetic dimer cannot contain any other factors present in AD TBS, establishing that Aβ dimers alone are sufficient to perturb synapse physiology.
Previous studies suggested that unlike soluble Aβ levels, amyloid plaque burden correlates poorly with AD severity7,9,20
. We asked whether insoluble amyloid cores isolated from AD cortex can inhibit hippocampal LTP. To isolate these detergent-resistant foci of fibrillar Aβ from neuritic plaques23–25
, we homogenized TBS-insoluble pellets of plaque-rich AD cortex in 2% SDS24
. IP/WB of supernatants after washing in SDS buffer showed that no additional Aβ was liberated by SDS or TBS from this preparation (). Congo red staining of the residual pellet revealed intact amyloid cores displaying characteristic birefringence (). Although resistant to disruption by many solvents, AD amyloid cores are efficiently solubilized by formic acid23,24
. This treatment released Aβ dimers and monomers from the washed cores (). When this formic acid extract of the AD core prep was applied to hippocampal slices, LTP was inhibited (116.2 ± 4.6%, P
<0.05 vs. formic acid vehicle; ). Formic acid extracts of identically prepared fractions from control brain allowed normal LTP (). Thus, amyloid cores contain Aβ dimers that can impair synaptic plasticity. In contrast, addition of intact cores () to the ACSF perfusate did not affect LTP (139.4 ± 7.6%, ). Therefore, in physiologic buffer (ACSF), amyloid cores do not acutely release soluble Aβ dimers to alter synaptic plasticity. IP/WB analyses revealed that Aβ dimers also were not released from amyloid cores incubated in physiological buffers at 37°C for 24 hr (Supp. Fig. 7
), suggesting that highly insoluble Aβ aggregates such as amyloid plaque cores represent dimer-rich structures that do not readily dissociate.
Figure 4 Insoluble amyloid cores contain Aβ dimers with synaptotoxic potential but are not readily released. (a) IP/WB of sequential extracts of the TBS-insoluble pellet prepared from 100 mg of a plaque-rich AD brain (case AD 5 from ) (see Methods). (more ...)
Here, we show that soluble Aβ isolated directly from AD brains potently and consistently induces several AD-like phenotypes in normal adult rodents: it decreases dendritic spine density, inhibits LTP and facilitates LTD in hippocampus, and interferes with the memory of a learned behavior. We used non-denaturing gel filtration coupled with IP/WB and subsequent immunodepletion or neutralization with epitope-specific Aβ antibodies to ascribe the pathogenic effects to soluble Aβ oligomers, principally dimers.
Our findings support the emerging concept that the effects of Aβ in AD center initially on subtly altered synapse function. Neither Aβ monomers nor insoluble amyloid plaque cores significantly altered synaptic plasticity. This does not mean that insoluble amyloid plaques have no pathogenic role; their invariant accumulation may signify that they serve as relatively inert reservoirs of small bioactive oligomers, and they may disassemble more readily in the presence of lipids26
. That plaque cores may release locally active Aβ species in vivo
is suggested by a penumbra of synapse loss around cores in APP transgenic mice27
Our examination of soluble dimers obtained from AD brain is partially consistent with findings using synthetic2
Aβ oligomers. However, there are unresolved differences regarding the precise biochemical nature of the synaptotoxic species found in these various systems. For example, we did not detect ( and Supp. Fig. 6
) a soluble, SDS-stable dodecamer of Aβ in human cortical extracts such as the Aβ*56 species observed in brain extracts from certain APP transgenic mice29
. Soluble Aβ complexes from AD cortex eluted in the void volume (>70 kDa) upon non-denaturing SEC, but these dissociated into dimers and monomers upon LDS-PAGE. Some AD and aged control CSF samples that contain soluble Aβ dimers were recently shown to impair LTP30
, a finding consistent with our data. However, the invariant detection of dimers in the soluble fraction of AD cortex and their multiple synaptic effects strongly suggest that cortical dimers contribute directly to synapse dysfunction in AD patients, whereas any additional effects of CSF dimers in the minority of AD subjects who have them30
remain to be determined.
Mechanistically, we show that soluble Aβ dimers from AD cortex induce their effects by perturbing glutamatergic synaptic transmission. Although we find that mGluRs are required for the induction of LTD while NMDARs are needed for spine loss, these receptors are unlikely to be the sole effector targets of soluble Aβ oligomers. Aβ extracted from human brain can now serve as the most pathophysiologically relevant material for further pathway analysis and for preclinical validation of agents designed to neutralize Aβ aggregates. Our findings fulfill an essential requirement for establishing disease causation in AD.