Despite amyloid plaques, consisting of insoluble, aggregated amyloid-β peptides, being a defining feature of Alzheimer’s disease, their significance has been challenged due to controversial findings regarding the correlation of cognitive impairment in Alzheimer’s disease with plaque load. The amyloid cascade hypothesis defines soluble amyloid-β oligomers, consisting of multiple amyloid-β monomers, as precursors of insoluble amyloid-β plaques. Dissecting the biological effects of single amyloid-β oligomers, for example of amyloid-β dimers, an abundant amyloid-β oligomer associated with clinical progression of Alzheimer’s disease, has been difficult due to the inability to control the kinetics of amyloid-β multimerization. For investigating the biological effects of amyloid-β dimers, we stabilized amyloid-β dimers by an intermolecular disulphide bridge via a cysteine mutation in the amyloid-β peptide (Aβ-S8C) of the amyloid precursor protein. This construct was expressed as a recombinant protein in cells and in a novel transgenic mouse, termed tgDimer mouse. This mouse formed constant levels of highly synaptotoxic soluble amyloid-β dimers, but not monomers, amyloid-β plaques or insoluble amyloid-β during its lifespan. Accordingly, neither signs of neuroinflammation, tau hyperphosphorylation or cell death were observed. Nevertheless, these tgDimer mice did exhibit deficits in hippocampal long-term potentiation and age-related impairments in learning and memory, similar to what was observed in classical Alzheimer’s disease mouse models. Although the amyloid-β dimers were unable to initiate the formation of insoluble amyloid-β aggregates in tgDimer mice, after crossbreeding tgDimer mice with the CRND8 mouse, an amyloid-β plaque generating mouse model, Aβ-S8C dimers were sequestered into amyloid-β plaques, suggesting that amyloid-β plaques incorporate neurotoxic amyloid-β dimers that by themselves are unable to self-assemble. Our results suggest that within the fine interplay between different amyloid-β species, amyloid-β dimer neurotoxic signalling, in the absence of amyloid-β plaque pathology, may be involved in causing early deficits in synaptic plasticity, learning and memory that accompany Alzheimer’s disease.
Alzheimer’s disease is the most prevalent neurodegenerative disease in humans. Amyloid-β-containing amyloid plaques and neurofibrillary tangles consisting of the microtubule associated protein tau (encoded by MAPT ) in post-mortem neuropathological examinations, in conjunction with the clinical diagnosis of dementia, have been the three defining features for Alzheimer’s disease for a century ( Alzheimer, 1906 ).
While neurofibrillary tangles and amyloid-β plaque pathology are widely used for staging of Alzheimer’s disease ( Braak and Braak, 1991 ; Thal et al. , 2002 ), neurofibrillary tangles have been demonstrated to correlate better with cognitive deficits of patients with Alzheimer’s disease than amyloid-β plaques ( Nelson et al. , 2012 ). Similarly, in vivo neuroimaging technology with Pittsburgh compound B showed a lacking accuracy in identifying patients with mild cognitive impairment who would later develop Alzheimer’s disease dementia ( Zhang et al. , 2014 ).
Yet, the amyloid cascade theory ( Hardy and Selkoe, 2002 ), founded on the solid genetic evidence of familial Alzheimer’s disease, states that aberrant amyloid-β accumulation is the initial step of a cascade of events that leads to tau hyperphosphorylation, neuroinflammation and, ultimately, massive neuronal degeneration. The presence of amyloid-β oligomers, soluble multimers of amyloid-β monomers, rather than the occurrence of insoluble fibrillar amyloid-β that only poorly correlated with cognitive impairment ( Terry et al. , 1991 ; Dickson et al. , 1995 ) is likely at the beginning of the amyloid-β cascade. Moreover, low- n oligomers like dimers are reasonably well correlating to synaptic loss and Alzheimer’s disease ( Klyubin et al. , 2008 ; Shankar et al. , 2008 ; Mc Donald et al. , 2010 ). Although unequivocal data on abundance and significance of amyloid-β oligomers in vivo are still limited, compelling in vitro studies have led some investigators to conclude that amyloid-β dimers, which have been detected to be highly abundant in brains of Alzheimer’s disease mouse models and patients with Alzheimer’s disease, are sufficient to account for neurotoxicity and initiating the Alzheimer’s disease-typical cascade ( Kawarabayashi et al. , 2004 ; Shankar et al. , 2008 ; Hashimoto et al. , 2012 ; Mc Donald et al. , 2010 , 2015 ).
So far, the investigation of amyloid-β dimers and their elicited neuropathological, biochemical and behavioural effects in vivo , has been impossible owing to the equilibrium that exists between the different multimer and conformer pools of amyloid-β. The introduction of cysteines to generate new covalent intra- or intermolecular disulphide bonds has been successfully applied in numerous studies to increase the stability of proteins or to lock proteins in a specific conformation ( Matsumura et al. , 1989 ; Shimaoka et al. , 2002 ), including the generation of synthetic amyloid-β dimers ( Schmechel et al. , 2003 ; O’Nuallain et al. , 2010 ; Yamaguchi et al. , 2010 ). These synthetic amyloid-β dimers spontaneously formed protofibrillar structures at micromolar concentrations ( Hu et al. , 2008 ; Yamaguchi et al. , 2010 ), of which one, the Aβ-S26C protofibril also turned out to be synaptotoxic ( O’Nuallain et al. , 2010 ). Unfortunately, it is almost impossible to model single amyloid-β oligomer species with synthetic peptides in cell-free in vitro systems because of: (i) low folding reliability during bulk refolding; (ii) competition between specific oligomer assembly and unspecific aggregation; (iii) a steady state equilibrium between different amyloid-β multimers; and (iv) the usually unphysiologically high concentrations of amyloid-β in cell-free systems that do not reflect cellular amyloid-β assembly ( Selkoe, 2008 ).
Soluble naturally secreted amyloid-β oligomers have been generated in a cell culture model, termed 7PA2, which expresses APP with the familial Indiana mutation APPV717F initially believed to secrete high amounts of amyloid-β dimers and trimers ( Podlisny et al. , 1995 ). Recently, however, the nature of these cell-derived amyloid-β species has been questioned, as they may predominantly represent neurotoxic N-terminal elongated amyloid-β monomers, although low amounts of oligomers still have been detected ( Welzel et al. , 2014 ). Previously, we demonstrated that stabilizing amyloid-β dimers by an intermolecular disulphide bridge within the N-terminus of amyloid-β (Aβ-S8C) led to the natural secretion of nanomolar amounts of synaptotoxic and stable amyloid-β dimers in a cell culture model, without influencing the trafficking or processing of APP by cellular secretases ( Muller-Schiffmann et al. , 2011 ).
In the present study, we investigated the selective role of amyloid-β dimers and their signalling on brain pathology and behaviour in vivo , by expressing the Aβ-S8C containing APP transgene in a mouse, termed tgDimer mouse. We used the same expression vector that led to the generation of the well characterized APP23 Alzheimer’s disease mouse model, and used the same mouse strain (C57BL/6) ( Sturchler-Pierrat et al. , 1997 ). In tgDimer mice, we detected high amounts of amyloid-β dimers, but no amyloid-β monomers in brain tissue of these mice. We demonstrate that amyloid-β dimers are unable to initiate the formation of insoluble amyloid-β or amyloid plaques, and can do so only when amyloid-β plaques are present from another source. Thus, the tgDimer mice for the first time allowed the dissection of specific effects mediated by the unique conformer of amyloid-β dimers in the absence of amyloid-β plaques. We observed no induction of tau hyperphosphorylation and neuroinflammation, but learning and memory deficits as well as impairment of synaptic plasticity suggesting that initial stages of cognitive decline in Alzheimer’s disease are independent of amyloid-β plaque pathology.
Materials and methods
The IgG2b monoclonal antibody IC16 ( Muller-Schiffmann et al. , 2010 ) recognizes the N-terminus of amyloid-β (amino acids 2–8) and was obtained from supernatants of IC16 hybridoma cells, that were cultured in CELLine two compartment bioreactors (Integra Biosciences). APP was visualized by the polyclonal CT15 antibody, which recognizes the C-terminal 15 amino acids of APP ( Sisodia et al. , 1993 ). 4G8 recognizes residues 17–24 of amyloid-β and was obtained from Covance. 6F/3D, which is specific for amino acids 9–14 from amyloid-β (M0872) and a GFAP-specific antibody (Z0334) were purchased from Dako, an anti-Iba1 antibody (AP08912PU-N) was bought from Acris Antibodies and the total tau antibody HT7 (MN1000), as well as the anti-phospho-PHF-tau antibodies AT8 (MN1020) and AT180 (MN1040), from Thermo Scientific. In addition we used phospho-tau antibodies pS396 and pS404 from Life Technologies. Anti-actin (A2066) and anti-tubulin (T9026) antibodies were purchased from Sigma. The following secondary antibodies were used: goat anti-mouse-peroxidase, goat anti-rabbit-peroxidase (both from Thermo Scientific), goat anti-mouse IRDye 680RD and 800CW as well as goat anti-rabbit IRDye 680RD and 800CW (all from LI-COR). It is important to note that all antibodies that were used here for detection of amyloid-β are also capable of recognizing APP.
The Aβ-S8C mutation was introduced into the pTSC21-APP751swe vector ( Sturchler-Pierrat et al. , 1997 ) by in vitro mutagenesis using the following oligonucleotides: Aβ-S8C-5’-3’ gcagaattccgacatgactgcggatatgaagttcatcatcaaaaattggtg; Aβ-S8C 3’-5’ caccaatttttgatgatgaacttcatatccgcagtcatgtcggaattctgc to yield the final pTSC21-APP751swe-S679C construct. After validation of the complete vector sequence the construct was used for microinjection into C57BL/6N embryos. The gene doses of three founder lines were quantified by quantitative polymerase chain reaction and the founder line exhibiting the highest dose was bred to homozygosity and termed tgDimer. For the seeding experiment tgDimer mice were crossed with CRND8 transgenic mice. CRND8 mice express APP with the familial Swedish and Indiana mutations and had been extensively characterized before ( Chishti et al. , 2001 ). In the experiments described here offspring from the F1 generation were used that were heterozygous for both APP expression vectors: APP carrying the Swedish and Aβ-S8C mutation (from tgDimer) and APP carrying the Swedish and Indiana mutation (from CRND8). Brains from differently aged APP23 mice were used as control and obtained from Novartis.
Enzyme-linked immunosorbent assay
Amyloid-β 40 and amlyloid-β 42 were extracted from brain homogenates with 5 M guanidine/50 mM Tris HCl, pH 8.0 and quantified using the human/rat beta Amyloid-40 or -42 ELISA kit from Wako according to the manufacturer’s recommendations. The kits used the BNT77 antibody [epitope amyloid-β (11–28)] for capturing and BA27 for detection of amyloid-β 40 or BC05 for detection of amlyloid-β 42 . The amyloid-β concentrations were calculated relative to the monomer concentration of the standard and are indicated as pmol/g (wet weight).
For quantification of total amyloid-β within the size exclusion chromatography fractions, 1 µg of IC16 in 100 µl phosphate-buffered saline (PBS) was immobilized per well of a 96-well Nunc ® MaxiSorp ® microtitre plate overnight. After washing the plate with PBS, the wells were blocked with PBS/1% bovine serum albumin (BSA) for 2 h at room temperature. Standards [synthetic amyloid-β 40 (Sigma-Aldrich)] and samples were diluted in PBS/0.05% Tween 20 (PBST) + 1% BSA. After washing the wells with PBS, these dilutions were applied to the coated wells and incubated with agitation overnight at 4°C. The next day, the wells were washed four times with PBST and incubated with 50 µl of biotin labelled 4G8 (1:2500 in PBST/1% BSA) for 2 h at room temperature. The wells were again washed four times with PBST and bound 4G8-Biotin was detected with horseradish peroxidase-coupled streptavidin (Thermo Fisher Scientific) diluted 1:10 000 in PBST/1% BSA. After 1 h of incubation at room temperature and four washing steps with PBST bound enzyme activity was measured with 100 µl of OptEIA ™ substrate solution (BD Biosciences). The enzyme activity was stopped with 100 µl of 25% H 2 SO 4 and absorbance was read at 450 nm. Molar concentrations of Aβ-S8C dimers were calculated after quantifying amyloid-β relative to the monomer concentration of the standards used.
For visualization of APP, Iba1, GFAP, and tau, 30 µg of 10% whole brain homogenates in 100 mM Tris pH 7.5, 140 mM NaCl, 3 mM KCl (TBS) containing Complete protease inhibitor cocktail (Roche) were separated on a NuPAGE ® 4-12% Bis-Tris Gel (Life technologies), using NuPAGE ® sample buffer either with addition of 2% (v/v) of β-mercaptoethanol or without, and blotted to a 0.2 µm nitrocellulose membrane. The membranes were blocked with 5% skimmed milk and incubated with either primary antibodies against APP (CT15: 1:3500), tau (MN1000, MN1020, MN1040, pS396 or pS404, each 1:1000), Iba1 (1:2000) or GFAP (1:2000) and actin (1:5000) or tubulin (1:5000). After incubation with appropriate secondary antibodies, signals were quantified either by densitrometric analysis using ImageJ (NIH) or by applying the Odyssey infrared imaging system (LI-COR).
Immunoprecipitation of amyloid-β
Whole mouse brain hemispheres were homogenized to 10% w/v in TBS containing Complete protease inhibitor cocktail (Roche). Amyloid-β was immunoprecipitated from the samples by incubation with IC16-NHS-sepharose overnight at 4°C ( Muller-Schiffmann et al. , 2011 ). After centrifugation, the beads were washed twice with PBS and amyloid-β was eluted with sodium dodecyl sulphate (SDS) sample buffer with or without reducing additives and boiled for 5 min. SDS-polyacrylamide gel electrophoresis (PAGE) and western blot were performed as described previously ( Podlisny et al. , 1995 ). Conditioned medium from amyloid-β secreting 7PA2 cells ( Podlisny et al. , 1995 ) served as control.
Four-step ultracentrifugation fractionation
Fractionation by 4-step ultracentrifugation was performed as described previously ( Kawarabayashi et al. , 2001 ). Briefly, 300 µl of the homogenates were centrifuged at 100 000 g at for 1 h at 4°C. The supernatants were harvested and the pellets were resuspended in 300 µl TBS/1% Triton ™ X-100 by sonication. After centrifugation at 100 000 g at 4°C for 1 h the supernatants were taken and the precipitates dissolved in 300 µl TBS/2% SDS. Following another round of centrifugation at 100 000 g at room temperature, the supernatants were harvested and the precipitates were finally dissolved in 300 µl of 70% formic acid before being centrifuged again for 1 h at 100 000 g at room temperature. The first three supernatants were diluted 20-fold in TBS and the formic acid fraction was neutralized in 20 volumes of 1 M Tris solution. Amyloid-β was immunoprecipitated by IC16-NHS-sepharose and visualized by western blot as described above.
Amyloid-β (6F/3D), astrocytes (GFAP) and microglia (Iba1) staining was performed automatically in a TechMate instrument (Dako) on 5 µm sagittal brain sections. For amyloid-β staining, slices were pre-treated at room temperature for 3 min with 85% formic acid and then blocked with 3% H 2 O 2 and a pre-blocking solution (Zytomed) for 5 min and 10 min at room temperature, respectively. Afterwards, sections were incubated for 30 min with the primary 6F/3D antibody (1:100) at room temperature, followed by a 20 min post-block incubation step (Zytomed) and subsequent incubation with a peroxidase-coupled polymer (Zytomed) for 30 min. Sections were visualized via incubation with 3,3’-diaminobenzidine (K5001, Dako) for 2 × 5 min. To highlight astrocytes, sections were pretreated with H 2 O 2 and a pre-blocking solution as described above, followed by incubation with anti-GFAP antibody (1:4000) for 30 min at room temperature. Visualization was conducted by incubation with peroxidase polymer and 3,3’ diaminobenzidene. Microglia were stained by pretreating sections with boiling citrate buffer (30 min, room temperature, pH 6.0), followed by permeabilization with 1% Triton for 20 min at room temperature. Then, sections were pre-blocked and incubated with Iba1-specific antibody (1:100) for 2 h at room temperature. Afterwards, sections were washed and incubated with anti-goat antibody (1:500, E0466, Dako) for 30 min at room temperature, followed by incubation with peroxidase polymer and 3,3’ diaminobenzidine. Haematoxylin (1:25) served for counterstaining.
Animal experiments were performed in accordance with the German Animal Protection Law and were authorized by local authorities (LANUV NRW, Germany).
Seven-month-old male transgenic tgDimer mice ( n = 12) and wild-type C57BL/6N control mice ( n = 9) (weight 28–36 g) were bred and obtained from the Central Animal Laboratory at the University Hospital Essen. They were individually housed in translucent plastic cages (36.5-cm long, 20.7-cm wide, 14.0-cm high) under a reversed 12:12 h light-dark cycle (light off at 07:00 a.m.) and temperature controlled conditions (20 ± 2°C) with food and water administered ad libitum. They were allowed to adapt to the housing conditions for 10 days before the behavioural testing. Their health status (food and drinking behaviour, coat condition, bodily orifices) was monitored daily.
The water maze consisted of a circular black pool 120 cm in diameter and 35 cm in height filled with 16.8 cm of water (20 ± 2°C). The water contained coffee whitener to enhance contrast for animal observation. The maze was divided into four equally sized virtual quadrants. The escape platform was a white circular platform (10.5 cm in diameter), located in the centre of one of these quadrants and submerged 2 cm below the water surface. Diffuse illumination by four LED lights (two focused to the cues, and two covered by a red transparent cover) around the maze provided a light density of ∼19 lx at the upper edge, and 15 lx above the water surface. A video camera mounted above the pool was linked to a computer-based tracking (Ethovision XT 8, Noldus). Extra-maze cues were provided by the LED lights and different shaped figures placed on the wall.
At the beginning of each trial the animal was placed into the pool, facing the wall at one of four possible starting points (north, south, east, west) along the perimeter of the maze. The order of start positions was randomized. One habituation trial of 90 s duration was run without a platform to assess swimming ability and motor control. This was followed by 9 days of acquisition by training the mice on the hidden platform place-learning task (four trials per day, 90 s intertrial interval). An acquisition trial was terminated when the animal escaped onto the hidden platform with all four paws or after 90 s had elapsed. In the latter case the animal was guided onto the platform by the experimenter. It was allowed to stay on the platform for 30 s and then removed from the maze and dried under a red-light heating lamp between the trials. The acquisition trials were followed by a probe trial without a platform to assess retrieval of the previously learned platform. On the probe trial, the mice were removed from the pool after 90 s. Next, they were tested in the visible platform version (cued task) to check for general sensory-motor abilities ( Morris et al. , 1982 ). This platform was cued via a 10.5 cm high cylinder, painted with black and white stripes. Four trials were averaged for each training day. In the probe trial, the time spent within the previously reinforced platform quadrant was assessed.
The data on acquisition and swimming speed in the water maze (the means of four trials per day) were subjected to repeated measures three-way ANOVAs with the factors Day, Genotype and Age. When appropriate this was followed by post hoc t -test to assess group differences. For the cued version of the water maze task, a t -test was carried out for between-group comparisons. The probe trial (as a measure of retention) assessed the time spent searching for the platform in the four quadrants of the maze. For this purpose the time engaged in thigmotactic swimming within the outer ring (width: 8 cm) along the maze wall was excluded. A two-way ANOVA with the factors Genotype and Age was followed by post hoc t -tests to compare the time spent in the former platform quadrant with time spent in the three non-reinforced quadrants, pooled. The P -values given are two-tailed and were considered significant if α ≤ 0.05. The software IBM SPSS Statistics 20.0 was used for all analyses.
Electrophysiological study on hippocampal slices
Experiments were performed on 400-µm thick hippocampal slices prepared from control and transgenic mice as described previously ( Chepkova et al. , 2012 ). Slices were maintained in a submersion type recording chamber at 32°C and perfused at a flow rate of 2–2.5 ml/min with artificial CSF containing 125 mM NaCl, 1.8 mM KCl, 1.2 mM KH 2 PO 4 , 1.2 mM MgCl 2 , 2.4 mM CaCl 2 , 26 mM NaHCO 3 , 10 mM D-glucose and saturated with 95%O 2 / 5%CO 2 gas mixture. A glass recording electrode filled with artificial CSF was positioned in the stratum radiatum of area CA1 of the hippocampus to record field excitatory postsynaptic potentials (EPSPs) evoked by stimulation of the Schaffer collateral-commissural system. Bipolar stimulating electrodes were placed in the stratum radiatum at the CA2–CA1 border. After the initial testing of stimulus–response relationships, stimulus intensity was adjusted to induce field EPSPs with the amplitude of 30–50% of its value threshold for generation of population spikes. The standard experimental protocol included 20–30 min control recording with test stimuli applied every 30 s followed by high frequency stimulation (HFS) (two 1-s trains at 100 Hz and double intensity) and 90 min post-HFS recording. Signals were amplified, digitized at 10 kHz, and recorded on a PC using pClamp software (Molecular Devices). Ten consecutive responses (5 min recordings) were averaged off-line and the averaged field EPSP slopes were measured by straight line fitting. All measurements were normalized to the average slope value through the control period (baseline) and expressed in per cent of baseline. The magnitude of field EPSP potentiation was determined by averaging the first (initial potentiation) and the last (long-term potentiation, LTP) 20 min periods of post-HFS recordings. The data are presented as mean ± standard error of the mean (SEM). Statistical analysis of the data (one- and two-way ANOVA, t -test, chi-square and Fisher’s exact tests) was performed by the GraphPad Prism 5 for Windows (GraphPad Software, San Diego, California, USA).
Isolation of Aβ-S8C dimers
Cell-secreted Aβ-S8C dimers were obtained from the supernatants of confluent CHO-APP-Aβ-S8C cells that were induced with 1 µg/ml doxycycline ( Muller-Schiffmann et al. , 2011 ). The conditioned medium was harvested after 24 h and Aβ-S8C was immunoprecipitated with IC16 monoclonal antibody, which had been covalently coupled to NHS-sepharose (GE Healthcare). After washing with PBS, Aβ-S8C was eluted with 100 mM triethanolamine and neutralized with 100 mM Tris HCl, pH 8.0. In this manner, ∼95% of Aβ-S8C was precipitated from the conditioned medium. Synthetic Aβ 42 -S8C was obtained from JPT and dissolved to 1 mg/ml in hexafluoroisopropanol (HFIP). Ten micrograms of Aβ 42 -S8C were evaporated, resolved in dimethyl sulphoxide (DMSO, final concentration: 100 µM) for 30 min at room temperature with agitation and oxidized to dimers in DMEM/F12 medium without FCS and phenol red (Gibco, life technologies) for 30 min at room temperature (final concentration: 2 µM). The cellular secreted and synthetic Aβ-S8C samples were then subjected to size exclusion chromatography using a HiLoad 16/60 Superdex 75 prep grade column (GE Healthcare) and eluted at 1 ml/min with 10 mM ammonium acetate pH 8.5. Calibration of the column was performed using Dextran blue (2000 kDa) and albumin (66 kDa; both from Sigma) as well as dimeric Aβ-1-16-S8C-GB1 (18.8 kDa) and monomeric Aβ-1-16-GB1 (9.4 kDa) that have been described in Dornieden et al. (2013) and were chosen to allow optimal comparison of the separated amyloid-β species. Fractions of 1 ml were collected. To prevent loss of the dimers by attaching to the surface of the vial, every second fraction was collected in vials, which already contained 500 µl of Neurobasal ® medium supplemented with B27 (Gibco, Life Technologies), leading to a final concentration of ammonium acetate of 6.67 mM. Neighbouring fractions were eluted without medium and lyophilized fractions of the cellular secreted Aβ-S8C or 10 µl of the fractions obtained from the synthetic Aβ 42 -S8C were analysed by Tricine-SDS-PAGE and western blot as described previously ( Podlisny et al. , 1995 ). Fractions containing solely dimeric Aβ-S8C were pooled and quantified by ELISA. All amyloid-β samples were snap-frozen in liquid nitrogen and stored at −80°C prior to biological assessment.
Whole cell patch clamp
Cultures of primary cortical neurons were prepared from C57BL/6 J mouse foetuses at embryonic Day 18 (E18) as described previously ( Mohrmann et al. , 2003 ). Neurons were grown in Neurobasal ® A medium (Gibco, Life Technologies) supplemented with 2% B27, 100 U/ml penicillin, 100 µg/ml streptomycin, and GlutaMAX ™ (Gibco, Life Technologies) on glass coverslips coated with 1 mg/ml poly- l -ornithine. After 10 days in vitro the neurons were incubated with varying amounts of purified Aβ-S8C dimers with equivalent volumes of 33% Neurobasal A/B27, 6.67 mM ammonium acetate being used as a control. AMPA receptor-mediated miniature EPSCs were recorded from cells after 4 days of incubation. Using standard patch-clamp procedures, whole-cell recoding mode was established by monitoring capacitance changes using an oscilloscope. AMPA miniature EPSCs were isolated by addition of tetrododoxin to block action potentials, and by addition of gabazine and by the presence of Mg 2+ (holding potential: −60 mV) to block GABAA receptors and NMDA receptors, respectively ( Muller-Schiffmann et al. , 2011 ).
Expression of high amounts of soluble dimeric amyloid-β in tgDimer mice
To investigate the isolated effects of amyloid-β dimers in vivo , we generated a transgenic mouse line, termed tgDimer mouse. In this mouse, the transgene for the full-length human APP751 includes the familial Swedish mutation (APP-K670N/M671L) for inducing a higher rate of amyloid-β processing ( Cai et al. , 1993 ) and the amyloid-β dimer-stabilizing mutation (APP-S679C), here, for reasons of clarity termed Aβ-S8C. In the tgDimer mouse, transgenic mutant APP under the control of the neuron-specific Thy1 promoter was ∼7-fold overexpressed in relation to endogenous APP of age-matched wild-type mice ( Fig. 1 A). Whereas processing of APP, as well as β- and α-secretase activity was not changed within tgDimer mice in comparison to APP23 mice ( Fig. 1 B), production of the APP intracellular domain from active membrane preparations of brain homogenates from tgDimer mice was reduced ( Supplementary Fig. 1 ). We also observed significant amounts of dimeric APP and dimeric APP C-terminal fragments in addition to the monomeric forms in brain homogenates from tgDimer mice under non-reducing conditions ( Fig. 1 B). Interestingly, the tgDimer mice generated very high amounts of disulphide stabilized amyloid-β dimers that could be dissociated to monomers when applying reducing buffers ( Supplementary Fig. 2 ), but not monomers or other SDS-stable low- n oligomers (trimers or tetramers) ( Fig. 1 C). Amyloid-β secreted from the 7PA2 cell line was initially used as a positive control for low- n oligomeric amyloid-β. However, the corresponding western blot signals of amyloid-β multimers are actually now thought to define N-terminal elongated amyloid-β monomer species, thus explaining the slightly different migration speed of the S8C-dimer and the 7PA2-derived amyloid-β species ( Welzel et al. , 2014 ). We did not observe analogous signals to these elongated amyloid-β monomer species in tgDimer mice ( Fig. 1 C) suggesting that there are no N-terminally elongated amyloid-β species of significant amount generated in tgDimer mice.
Absence of insoluble amyloid-β aggregates and amyloid-β plaque pathology in tgDimer mice
Amyloid-β 40 and amyloid-β 42 levels in whole extracts of brain homogenates from tgDimer mice were quantified by ELISA in which the affinity of the capturing antibody towards amyloid-β was only slightly influenced by the S8C mutation, most likely due to a sterically limited binding of the antibodies to their epitope that resided in near proximity to the disulphide bridge of the dimer ( Supplementary Fig. 3 ). The levels of amlyloid-β 40 and amlyloid-β 42 in the brain remained constant throughout the lifetime (as assayed at 3–24 months) of tgDimer mice ( Fig. 2 A) and were similar to those detected in young APP23 mice that had not yet deposited plaques ( Maia et al. , 2013 ; Morales-Corraliza et al. , 2013 ). Also, the amlyloid-β 42 /amlyloid-β total ratio was not significantly changed throughout the lifespan, similar to wild-type mice ( Supplementary Fig. 4 ), indicating no accumulation of amyloid-β and especially amyloid-β 42 , in contrast to other Alzheimer’s disease animal models ( Sturchler-Pierrat et al. , 1997 ; Chishti et al. , 2001 ).
Consistent with the constant amyloid-β levels, no insoluble amyloid-β or amyloid-β plaques were detected in aged tgDimer mice after performing a sequence of biochemical extraction procedures ( Fig. 2 B and Supplementary Fig. 5 ) or immunohistochemistry ( Fig. 2 C). Of note, similar to reports of many other publications (for review see Wirths and Bayer, 2012 ) we observed intracellular immunoreactivity against the amyloid-β epitope in neocortex and hippocampus between 12 and 24 months of age ( Fig. 2 C). However, intraneuronal amyloid-β has been demonstrated to be common in human brain ( Blair et al. , 2014 ) and not to be a predictor of brain amyloidosis ( Wegiel et al. , 2007 ). Accordingly, after biochemical purification of brain extracts, in the tgDimer mouse, irrespective of their age and despite a high APP expression levels, we never saw amyloid-β immunoreactivity in the formic acid fraction of the tgDimer mouse that would correspond to amyloid-β deposited as amyloid ( Roher et al. , 1993 ) ( Fig. 2 B). This contrasts with the results of the fractionation of brain homogenates of CRND8 mice ( Fig. 2 B) that display severe and early amyloid-β plaque pathology, here used as a technical positive control ( Chishti et al. , 2001 ) and for example to another mouse model where no plaques, but insoluble amyloid-β was detected ( Tomiyama et al. , 2010 ). Moreover, in contrast to tgDimer mice, free TBS-soluble amyloid-β was completely lacking in brain homogenates from CRND8 control mice, further demonstrating the high solubility of Aβ-S8C dimers. Of note, it has been shown that both in young mice of Alzheimer’s disease models or wild-type mice, high amounts of amyloid-β are detected in the SDS-fraction ( Kawarabayashi et al. , 2001 ; Hong et al. , 2011 ) indicating that Aβ-S8C dimers do not exhibit different solubility patterns.
Absence of endogenous tau-hyperphosphorylation and neuroinflammation in brains of tgDimer mice
The high amount of soluble amyloid-β dimers in tgDimer mice allowed the differentiation of neuropathological consequences triggered by either insoluble amyloid-β species (fibrils and plaques) or soluble amyloid-β dimers. We did not detect a significant change in phosphorylation of endogenous tau in biochemical analysis of brain homogenates ( Fig. 3 A and Supplementary Fig. 6A ), nor in immunohistochemical staining (data not shown). We also found no signs of neuronal death and neuroinflammation with astrogliosis or microgliosis in the tgDimer mouse ( Fig. 3 B, C and Supplementary Fig. 6B ), supporting the notion that neuroinflammatory responses and neuronal death are linked to the presence of amyloid-β species other than amyloid-β dimers, for example to those existing in amyloid-β plaques ( Akiyama et al. , 2000 ).
Cognitive impairment in tgDimer mice
In Alzheimer’s disease mouse models that develop amyloid-β plaques, learning and memory deficits have been demonstrated as evidence for aberrant amyloid-β pathology leading to cognitive deficits ( Chen et al. , 2000 ; Chishti et al. , 2001 ). The fact that synaptic and behavioural deficits have also been observed in the absence of ( Chen et al. , 2012 ), or prior to the appearance of plaques ( Knobloch et al. , 2007 ; Hamilton et al. , 2010 ; Skaper, 2012 ), suggests that amyloid-β plaque toxicity may not be the critical or sole mechanism that underlies such deficits.
We tested the tgDimer mice in the Morris water maze, a standard behavioural task used to assess hippocampal dysfunction-related deficits in spatial learning and memory in Alzheimer’s disease models ( Morris et al. , 1982 ; Stewart et al. , 2011 ). We observed a deficit in learning to escape from the water maze in tgDimer mice that were tested at both 7 and 12 months of age ( Fig. 4 A). The results of three-way ANOVA revealed significant main effects of day (acquisition) [ F (8,240) = 22.14 P < 0.001] and genotype [ F (1,30) = 28.00, P < 0.001], a trend for a main effect of genotype interacting with age [ F (1,30) = 4.06, P = 0.053], and a significant main effect of age interacting with day (acquisition) [ F (8,240) = 3.17, P = 0.002]. Post hoc t -tests revealed significant differences in time to find the platform between adult tgDimer and C57BL/6N mice on Day 5 ( P = 0.002), Day 6 ( P = 0.004), Day 7 ( P = 0.029) and Day 8 ( P = 0.024) as well as between aged tgDimer and C57BL/6N mice on Day 3 ( P = 0.001), Day 7 ( P = 0.006), Day 8 ( P = 0.007) and Day 9 ( P = 0.006), indicating superior learning in the C57BL/6N mice. Unlike the C57BL/6N, at 12 months the transgenic mice showed virtually no savings from the learning level achieved at 7 months and did not exhibit a learning curve over the 9 days (36 trials) of testing ( Fig. 4 A). This poor performance of the tgDimer mice at 12 months retest could have hypothetically been influenced by the worse level achieved at the end of testing at 7 months, providing a retest advantage to the C57BL/6N mice. However, such an advantage cannot account for their failure to exhibit any improvement over the 36 trials. Additional analysis of the slope of the performance over trials confirmed that at 12 months of age the tgDimer mice did not display a significant improvement over trials (by paired sample t -test between first and last acquisition day P > 0.05) whereas they did so at 7 months ( P < 0.001). The C57BL/6N mice displayed significant learning curves (slope) at both 7 and 12 months of age ( P < 0.001 and P = 0.016, respectively). Thus, only the tgDimer mice displayed an ageing-related progression in the severity of cognitive decline, as indicated by failure to exhibit a trial-dependent improvement in performance at 12 months of age.
The tgDimer mice also failed to show significant savings (retention/memory) for the location of the hidden platform in the probe trial at both ages ( Fig. 4 B). Unlike the C57BL/6N controls, they did not spend a significant amount of time in the former safe platform quadrant Q3, indicating poor retention. The two-way ANOVA for the results at 12 months revealed a significant main effect of genotype and quadrants for the aged 12-month-old mice [ F (1,17) = 4.68, P = 0.045 and F (1,17) = 9.52, P = 0.007, respectively], as well as for the adult 7-month-old mice [genotype, F (1,19) = 4.79, P = 0.041, and quadrant, F (1,19) = 14.36, P = 0.001, respectively]. Post hoc paired sample t -test between Q3 versus pooled non-reinforced quadrants showed that only the C57BL/6N animals spent significantly more time in Q3 than in the pooled non-reinforced quadrants (C57BL/6N adult P = 0.001; aged P = 0.029). In fact, neither the adult nor aged tgDimer animals spent a significant amount of time in the former safe quadrant Q3.
These behavioural changes indicate deficient learning and memory in the tgDimer mouse. These deficits are unlikely to be a result of motoric or sensory deficits, as there were no significant differences between C57BL/6N and tgDimer mice in swimming speed and in success in escaping onto the platform when it was visible; also, neither speed of swimming nor escape onto a visible platform showed an age-dependent decline ( Fig. 4 A and C). These results demonstrate that hippocampus-related cognitive deficits occur in an animal that expresses amyloid-β dimers in the absence of plaques in the adult mouse and that these impairments are not compensated with age. Instead, some evidence indicates that they progress over age in this preparation.
tgDimer mice show impaired hippocampal synaptic plasticity
Synaptic plasticity is hypothesized to be a neurophysiological substrate for learning and memory. Hippocampal LTP has been found to be impaired in different transgenic Alzheimer’s disease models ( Marchetti and Marie, 2011 ; Spires-Jones and Knafo, 2012 ) and been shown to be correlated with the degree of impairment in learning of the Morris water maze in the old rodent ( Schulz et al. , 2002 ). We compared the characteristics of basal neurotransmission and plasticity in the Schaffer collateral-CA1 synaptic system of hippocampal slices from age-matched wild-type C57BL/6N and tgDimer mice. Analysis of input-output relationships using two-way ANOVA showed no significant difference in basal neurotransmission between wild-type and tgDimer mice at the ages of 6–8 months (adult) and 15–17 months (aged) ( Fig. 5 A). In the majority of hippocampal slices prepared from wild-type and tgDimer mice of the two ages, HFS of the synaptic input resulted in potentiation of evoked field EPSPs that either persisted (sustained LTP) or gradually decayed (transient LTP) towards the end of the 90-min observation period. There was no significant difference between genotypes and ages in the occurrence of sustained or transient LTP ( P > 0.05 Fisher exact test, Table 1 ). The time course of the HFS-induced changes in CA1 postsynaptic responses ( Fig. 5 B) shows a significant deficit of synaptic plasticity in the tgDimer hippocampus: slices from adult and aged wild-type mice maintained the potentiation at the level of 150 ± 8% and 145 ± 5% of baseline, respectively, whereas in slices from age-matched tgDimer mice the potentiation decreased to 126 ± 7% and 117 ± 3% of baseline, respectively (the difference is significant at P = 0.023 for adult and P < 0.0001 for aged mice, unpaired t -test). Comparative analysis of characteristics of sustained and transient LTP showed that the decreased amount of potentiation in tgDimer mice is largely due to the lower magnitude of sustained LTP ( Table 1 ). Besides, tgDimer mice showed more pronounced age-related decrease in the maintenance of sustained LTP: in slices from aged mice the amount of potentiation decreased by 16% (not significant) in wild-type and by 25% in tgDimer ( P = 0.0002, unpaired t -test) compared to the initial level. The lower levels of the initial potentiation in the tgDimer slices ( Fig. 5 B and Table 1 ) and the more pronounced decline of potentiation with time suggest an impairment of both induction and maintenance mechanisms of hippocampal LTP in tgDimer mice.
|Mouse/slices tested ( n )|| Sustained LTP || Transient LTP |
|Adult wild-type||201 ± 7%||202 ± 9%||169 ± 6%||87 ± 3%|
|n = 29||n = 13||n = 10|
|Adult tgDimer,||173 ± 4% **||171 ± 7%*||178 ± 9%||91 ± 5%|
|n = 23||n = 11||n = 9|
|Aged wild-type||174 ± 7% ##||158 ± 5% ###||165 ± 12%||98 ± 4%|
|n = 23||n = 16||n = 4|
|Aged tgDimer||161 ± 6%||136 ± 3% ***,###||173 ± 11%||92 ± 4%|
|n = 24||n = 14||n = 5|
|Mouse/slices tested ( n )|| Sustained LTP || Transient LTP |
|Adult wild-type||201 ± 7%||202 ± 9%||169 ± 6%||87 ± 3%|
|n = 29||n = 13||n = 10|
|Adult tgDimer,||173 ± 4% **||171 ± 7%*||178 ± 9%||91 ± 5%|
|n = 23||n = 11||n = 9|
|Aged wild-type||174 ± 7% ##||158 ± 5% ###||165 ± 12%||98 ± 4%|
|n = 23||n = 16||n = 4|
|Aged tgDimer||161 ± 6%||136 ± 3% ***,###||173 ± 11%||92 ± 4%|
|n = 24||n = 14||n = 5|
The table presents the average magnitudes of field EPSP slope potentiation in per cent of baseline (mean ± SEM) calculated for the periods 5–20 min (initial) and 75–90 min (late) after two 1-s trains of HFS delivered to the Schaffer collateral-commissural input. The terms ‘adult’ and ‘aged’ refer to male mice ( n = 4 in each group) at the age of 6–8 months and 15–17 months, respectively. Sustained LTP was determined as a persistent enhancement of field EPSP slope observed for 90 min post-HFS. Transient LTP was characterized by a gradual decline of potentiated field EPSP to the baseline within 60–90 min post-HFS. According to the chi-square test the difference in the HFS outcome (sustained, transient or no LTP) in slices from wild-type and tgDimer mice is not significant. Asterisks mark significantly lower magnitudes of sustained LTP in tgDimer hippocampus compared to the age-matched wild-type: * P = 0.0201; ** P = 0.0014; *** P = 0.0004, Student’s t -test. Hash signs mark significant age-related decrease in the potentiation magnitude in the hippocampus of both wild-type and tgDimer mice: ##P = 0.0057; ###P < 0.0001, Student’s t -test.
Purified Aβ-S8C dimers are synaptotoxic in the picomolar range
Amyloid-β-induced impairment of synaptic functions has been attributed to the removal of AMPA receptors from synapses ( Kamenetz et al. , 2003 ; Hsieh et al. , 2006 ). To validate the role of Aβ-S8C dimers in the impairment of LTP in the tgDimer mice we measured the decrease of AMPA receptor-mediated miniature EPSCs in primary mouse cortical neurons. We applied size exclusion chromatography to purify either naturally secreted Aβ-S8C dimers that were immunoprecipitated from conditioned medium of Aβ-S8C dimer-secreting cells, or oxidized synthetic Aβ 42 -S8C dimers ( Supplementary Fig. 7 ). Although naturally secreted Aβ-S8C dimers consist mainly of amyloid-β 40 and thus differed in their C-termini from the synthetic Aβ 42 -S8C dimers, both amyloid-β dimer preparations mediated synaptotoxicity in the pM range and showed a significant decrease in both miniature EPSC frequency and amplitude ( Fig. 6 ). Consistent with previous reports on different biological activities of synthetic versus secreted amyloid-β species ( Jin et al. , 2011 ; Reed et al. , 2011 ), naturally secreted Aβ-S8C dimers of the same amount were∼30-fold more synaptotoxic than the fully synthetic counterpart.
Aβ-S8C dimers can be sequestered by amyloid-β plaques
It has been hypothesized that insoluble amyloid-β plaques are an immediate cellular defence against amyloid-β oligomer toxicity ( Cheng et al. , 2007 ; Treusch et al. , 2009 ). To determine whether our paradigmatic Aβ-S8C dimers are able to integrate into existing amyloid-β plaques, we performed a genetic seeding experiment where we crossed tgDimer mice into CRND8 mice that develop amyloid-β plaques starting at 3 months of age ( Chishti et al. , 2001 ). These double-transgenic mice generate both Aβ-S8C dimers as well as aggregation-prone wild-type human amyloid-β, but they did not develop differently from either parent and did not show obvious clinical symptoms at the age of 6 months. In this double transgenic mouse line, Aβ-S8C dimers were detected in the insoluble formic acid fraction, co-purifying with insoluble wild-type amyloid-β and, thus were indeed pulled down along with genuinely insoluble material. A control fractionation of brain homogenate of tgDimer mouse that was spiked with CRND8 brain homogenate immediately before fractionation was devoid of dimeric Aβ-S8C in the formic acid fraction, demonstrating that Aβ-S8C dimers were not artificially co-purified by insoluble wild-type amyloid-β ( Fig. 7 ). We did not detect any Triton ™ X-100-soluble Aβ-S8C dimers in the double transgenic mice, indicating an efficient sequestration of soluble Aβ-S8C dimers into the amyloid-β plaques. This finding demonstrates that Aβ-S8C dimers can principally be incorporated into amyloid-β plaques but cannot itself initiate plaque formation in the absence of insoluble amyloid-β seeds.
Our results show that the tgDimer mouse, which expresses a single, distinct amyloid-β-conformation, the amyloid-β dimer, exhibits impairments in learning and memory, as well as in neuroplasticity, all this in the absence of insoluble amyloid-β or amyloid-β plaque deposition. Behavioural deficits and changes in neuroplasticity have also been reported prior to the appearance of plaques in other Alzheimer’s disease mouse models, indicating that toxicity mediated by insoluble amyloid-β may not be the critical or sole mechanism that underlies such deficits in Alzheimer’s disease ( Knobloch et al. , 2007 ; Hamilton et al. , 2010 ; Chen et al. , 2012 ; Skaper, 2012 ). Our results therefore suggest that Aβ-S8C dimers may play a causal role in the cognitive deficits preceding the onset of significant amyloid-β plaque accumulation.
In terms of the level and cellular specificity of transgene expression, the tgDimer mouse is comparable to the well characterized APP23 Alzheimer’s disease mouse model ( Sturchler-Pierrat et al. , 1997 ), as both of these models use the C57BL/6 background to express APP751 including the Swedish mutation driven by the neuron-specific Thy1 promoter. Both models yield a ∼7-fold overexpression of APP and generate comparable levels of soluble amyloid-β. Despite these similarities, the APP23 mouse displays a spectrum of Alzheimer’s disease-like neuropathology including amyloid-β plaque deposition at ∼6 months of age, followed by massive gliosis and incipient hyperphosphorylation of tau. In contrast, all of these symptoms, which are comparable to later stages of Alzheimer’s disease in humans, are absent in tgDimer mice. The mutation introduced to stabilize the amyloid-β dimer neutralizes the amyloid-β plaque promoting effect of the Swedish mutation. This remarkable difference allows us to put major neuropathological hallmarks of Alzheimer’s disease pathology into a causal relationship: our results suggest that neuroinflammation and tau hyperphosphorylation are likely to be triggered by soluble or insoluble higher structured amyloid-β assemblies, which are not displayed by our mouse model, rather than amyloid-β dimers. A potential caveat could be that endogenous mouse tau in tgDimer mice is not a target of dimeric human amyloid-β whereas human tau could well be, as tau pathology, like tau tangles, has never been observed in an single transgenic APP-overexpressing mouse model for Alzheimer’s disease with endogenous tau ( Morrissette et al. , 2009 ).
The high amount of dimeric Aβ-S8C in the brain homogenates of tgDimer mice was surprising. In accordance with our cell culture model of Aβ-S8C ( Muller-Schiffmann et al. , 2011 ) and another work expressing APP with a cysteine mutant at position 28 of amyloid-β in vitro ( Scheuermann et al. , 2001 ), we observed dimerization already at the level of APP. Formation of APP dimers early after protein biogenesis suggests that Aβ-S8C dimers may be cleaved from APP as a preformed dimer thus, explaining the high amounts of dimeric β-C-terminal fragment and Aβ-S8C dimers measured in the absence of monomers. Early dimerization of APP in our model may also prevent non-specific oxidation with other cellular factors via the free cysteine. This is also evidenced by the clear absence of any non-specific western blot signals under non-reducing conditions in comparison to APP23 ( Fig. 1 B). The dimers purified from the Aβ-S8C cell culture supernatant were highly synaptotoxic, therefore likely accounting for the effects observed in our LTP and behavioural analysis. However, in tgDimer mice, as in every other Alzheimer’s disease model that overexpresses APP, we cannot completely exclude potential effects due to a high abundance of APP or its derivatives, of which some, like the β-C-terminal fragment, APP intracellular domain and caspase cleaved APP fragments, also have been attributed to contribute to toxicity ( Yankner et al. , 1989 ; Kinoshita et al. , 2002 ; Lu et al. , 2003 ).
The synaptotoxic potential of the stabilized amyloid-β dimers is obvious in the tgDimer mice through their exhibition of age-related impairments of learning and memory ( Fig. 4 ), which may be associated with a synaptic plasticity deficit. Alterations in hippocampal LTP ranging from no change to a dramatic impairment were reported in previous studies on different mouse models of Alzheimer’s disease ( Marchetti and Marie, 2011 ; Spires-Jones and Knafo, 2012 ). The neurotoxic effects of the Aβ-S8C dimers not only in the hippocampus, but also in the cortex and ascending monoaminergic systems responsible for cortical arousal, could be a source of the progressive neural damage resulting in age-related behavioural/neural deficits. This reflects the situation in APP23 mice, in which age-dependent cognitive decline has been observed already in mice younger than 6 months, before the onset of plaque pathology ( Van Dam et al. , 2003 ). We propose that the state of pathology in the tgDimer mouse resembles in many key aspects, an early stage of Alzheimer’s disease or mild cognitive impairment, at which point the classical Alzheimer’s disease neuropathology has not yet developed and synaptotoxic amyloid-β low- n oligomers are biochemically (see Fig. 2 B, TBS fraction), but not yet microscopically detectable ( Selkoe, 2008 ).
Purified Aβ-S8C dimers, whether naturally secreted or synthetic, were extremely synaptotoxic in the picomolar range ( Fig. 6 ), clearly confirming that they had been folded into a bioactive conformation and thus reached at least a similar level of activity to synthetic Aβ-S26C dimer preparations published before ( Hu et al. , 2008 ; Shankar et al. , 2008 ; O’Nuallain et al. , 2010 ; O’Malley et al. , 2014 ). However, toxicity of S26C dimers has been reported to be dependent on formation of non-fibrillar, Thioflavin T-positive aggregates that were generated in concentrations >2.5 µM ( O’Nuallain et al. , 2010 ). Size exclusion chromatography isolation of Aβ-S8C dimers yielded concentrations in the low nanomolar range, far below the reported critical concentration for Aβ-S26C dimers. Thus, it is unlikely that fibrillar aggregates of Aβ-S8C dimers are the underlying toxic entity. However, this does not exclude the possible existence of soluble meta-stable higher structured oligomers that were formed by building blocks of dimeric amyloid-β.
Of note, synthetic Aβ 42 -S8C dimers were 30-fold less neurotoxic when directly compared to equimolar amounts of the cell-derived secreted Aβ-S8C dimers. This has been observed before with wild-type amyloid-β oligomers ( Jin et al. , 2011 ; Reed et al. , 2011 ) and most likely reflects N- or C-terminal modifications within amyloid-β that are dependent on eukaryotic cellular factors and are, therefore, absent in synthetic amyloid-β preparations ( Saido et al. , 1996 ). Although we were not able to detect them, small amounts of N-terminal elongated synaptotoxic amyloid-β monomers ( Welzel et al. , 2014 ) may have co-migrated and co-eluted from size exclusion chromatography together with the Aβ-S8C dimers and contributed to the synaptotoxic effect. The high synaptotoxic potency of both preparations suggest that the main toxic effects of the amyloid-β dimers are not determined by the C-terminus of the amyloid-β species that formed the dimer, since in contrast to the Aβ 42 -S8C dimer preparations cellular secreted amyloid-β consisted mainly of amyloid-β 40 ( Supplementary Fig. 8 ). Interestingly, the lowest amounts of amyloid-β dimer preparations were sufficient to affect the amplitude of AMPA-mediated miniature EPSCs, which indicates changes in postsynaptic functions, such as AMPA receptor synaptic density ( Lisman et al. , 2007 ). AMPA receptors, which are important for hippocampal synaptic plasticity, have previously been shown to be affected by soluble naturally-secreted amyloid-β oligomers ( Shankar et al. , 2008 ). Furthermore, endocytosis of synaptic AMPA receptors by amyloid-β has been linked to synaptic depression and dendritic spine loss ( Hsieh et al. , 2006 ). Amyloid-β dimers may therefore represent a trigger for this effect.
The strong synaptotoxicity of the Aβ-S8C dimers is likely a direct consequence of the conformational transition state-stabilizing disulphide bond within the N-terminus of amyloid-β. This domain of amyloid-β plays a crucial role in oligomer/fibril formation and execution of toxic functions, as demonstrated by the fact that both post-translational modifications and naturally occurring mutations within the N-terminal domain increase oligomerization propensity and toxicity ( Schlenzig et al. , 2009 ; Ono et al. , 2010 ; Kumar et al. , 2012 ). Antibodies that target the N-terminus, but not those targeting the C-terminus, of amyloid-β were shown to reduce amyloid-β plaque load, but also to avert synaptotoxicity mediated by synthetic amyloid-β oligomers ( Bard et al. , 2000 ; Zago et al. , 2012 ). Recently, an intermolecular β-sheet structure has been attributed to the N-termini of amyloid-β oligomers, but not fibrils, and has been shown to be critical in the impairment of LTP ( Pan et al. , 2011 ; Haupt et al. , 2012 ). This structural feature of synaptotoxic amyloid-β oligomers may be mimicked by the covalent stabilization caused by the N-terminal cysteine of Aβ-S8C dimers. Therefore, Aβ-S8C dimers may particularly be interesting for screening of drugs targeting the synaptotoxic N-terminal domain of oligomers in vitro and in vivo.
The mutation of serine to cysteine at codon 8 of amyloid-β was selected purposefully after extensive molecular modelling with the aim of conserving a structure resembling that of the wild-type dimer as closely as possible, while also stabilizing its half life time dramatically ( Supplementary Fig. 9 ; and Horn and Sticht, 2010 ; Muller-Schiffmann et al. , 2011 ). Introduction of disulphide-bonded cysteine 8 into the experimentally determined structure of the mature fibril ( Lu et al. , 2013 ) revealed only very minor steric clashes thus offering no satisfactory explanation for the complete absence of fibrils. An alternative explanation would be that the S8C mutation might inhibit fibrillation itself. There exists considerable experimental ( Stravalaci et al. , 2011 ; Lam et al. , 2013 ) and computational ( Nguyen et al. , 2007 ; Rojas et al. , 2010 ) evidence that fibril growth occurs via the addition of flexible monomers to an amyloid-β seed via a two-step ‘dock-lock’ process. Here, in a rapid first step the monomer docks to a preformed oligomer/fibril seed. During the slower lock phase the monomer undergoes conformational changes by forming a β-strand that is in registry with the oligomer/fibril. This mechanism cannot occur in the tgDimer mice due to the absence of monomeric amyloid-β and fibrillar seeds ( Figs 1 C and 7). Moreover, molecular modelling indicates that seed nucleation itself is disfavoured in tgDimer mice, as the disulphide-bonded dimer adopts a rather rigid conformation that is distinct from both the flexible monomer and the U-shaped oligomer ( Horn and Sticht, 2010 ; Muller-Schiffmann et al. , 2011 ). Thus, nucleation into higher order oligomer seeds would require the simultaneous docking of two Aβ-S8C dimers that, at this time, adopt an elongation competent conformation, which is highly unlikely due to the low physiological concentrations of the amyloid-β dimers ( Supplementary Fig. 10A ). Consequently, the Aβ-S8C dimer cannot efficiently generate plaque seeds, providing a plausible explanation for the absence of fibrillar oligomers and plaques in tgDimer mice.
Our findings demonstrate the difference between being able to form the initial building block (or seed) for plaques themselves and being able to be incorporated into such existing building blocks. Our modelling data demonstrate that Aβ-S8C dimers can be stabilized in elongation competent fibrillar structures, without reducing their stability ( Supplementary Fig. 10B ). Specifically, our genetic seeding experiment confirmed that an environment that allows amyloid-β plaque formation (the CRND8 mouse) and, therefore, provides amyloid-β seeds, can facilitate the sequestration of toxic amyloid-β dimers into insoluble amyloid assemblies, despite their ability to seed plaques alone ( Fig. 7 ). This is in line with findings showing that amyloid-β dimers were more abundant in insoluble than in soluble fractions of brain homogenates of patients with Alzheimer’s disease ( McLean et al. , 1999 ). Therefore, our data feed to recent suggestions that sequestration and accumulation of oligomers into amyloid inclusion bodies and extracellular plaques may at least initially serve protective functions in neurodegenerative diseases by shifting the equilibrium away from the toxic oligomers ( Cohen et al. , 2006 ; Cheng et al. , 2007 ; Treusch et al. , 2009 ).
So far, our Aβ-S8C dimer model is the only in vitro and in vivo model available that reliably produces high amounts of exclusively soluble amyloid-β oligomers (i.e. specifically dimers). Our work suggests that amyloid-β dimers in the absence of plaques may play a causal role in the learning and memory deficits and the LTP changes commonly described in plaque-developing Alzheimer’s disease models and sometimes seen prior to the appearance of plaques. Moreover, amyloid-β dimers do not form seeds. However, we show that they can be incorporated into amyloid-β plaques. Our data thus strongly support the significance of amyloid-β dimers as a valid target for therapeutic intervention as previously proposed ( Hefti et al. , 2013 ).
Funding for this project from the following sources is acknowledged: grants from the Bundesministerium für Bildung und Forschung (BMBF-Kompetenznetz Degenerative Demenzen: KNDD-rpAD 01GI1010A), Stiftung für Altersforschung HHUD, and EU-FP7 PRIORITY to C.K., a grant from the Volkswagenstiftung (I/ 82649) to C.K. and H.S., a grant from the Forschungskommission of the University of Düsseldorf Medical School to A.M.S (54/2013) and O.A.S., a Deutscher Akademischer Austauschdienst (DAAD) grant to L.A.-H., and Heisenberg Fellowship SO 1032/5-1 65 and grant SO 1032/2-5 from the Deutsche Forschungsgemeinschaft to M.A.d.S.S.
Supplementary material is available at Brain online.
excitatory postsynaptic current
excitatory postsynaptic potential
high frequency stimulation
N -methyl D-aspartate