A Novel Crosslinking Protocol Stabilizes Amyloid β Oligomers Capable of Inducing Alzheimer’s-Associated Pathologies

Abstract

Amyloid β oligomers (AβOs) accumulate early in Alzheimer’s disease (AD) and experimentally cause memory dysfunction and the major pathologies associated with AD, e.g., tau abnormalities, synapse loss, oxidative damage, and cognitive dysfunction. In order to develop the most effective AβO-targeting diagnostics and therapeutics, the AβO structures contributing to AD-associated toxicity must be elucidated. Here, we investigate the structural properties and pathogenic relevance of AβOs stabilized by the bifunctional crosslinker 1,5-difluoro-2,4-dinitrobenzene (DFDNB). We find that DFDNB stabilizes synthetic Aβ in a soluble oligomeric conformation. With DFDNB, solutions of Aβ that would otherwise convert to large aggregates instead yield solutions of stable AβOs, predominantly in the 50–300 kDa range, that are maintained for at least 12 days at 37°C. Structures were determined by biochemical and native top-down mass spectrometry analyses. Assayed in neuronal cultures and i.c.v.-injected mice, the DFDNB-stabilized AβOs were found to induce tau hyperphosphorylation, inhibit choline acetyltransferase, and provoke neuroinflammation. Most interestingly, DFDNB crosslinking was found to stabilize an AβO conformation particularly potent in inducing memory dysfunction in mice. Taken together, these data support the utility of DFDNB crosslinking as a tool for stabilizing pathogenic AβOs in structure-function studies.

Graphical Abstract.

The structures of amyloid beta oligomers (AβOs) most germane to the pathogenesis of Alzheimer’s disease (AD) are ill-defined. Here we demonstrate the utility of the crosslinker 1,5-difluoro-2,4-dinitrobenzene (DFDNB) as a tool for AβO stabilization in structure-function studies. We find that DFDNB-linked AβOs are detergent-insensitive and remain in a soluble conformation with time and with increased concentration. Importantly, DFDNB stabilization does not alter the ability of AβOs to invoke potent responses than in multiple AD-relevant pathogenic assays.

Introduction

Amyloid beta oligomers (AβOs) accumulate early in Alzheimer’s disease (AD) and experimentally cause memory dysfunction and major AD-associated pathologies, e.g., tau abnormalities, synapse loss, oxidative damage, and cognitive dysfunction (Selkoe & Hardy 2016, Cline et al. 2018). However, the AβO structures contributing to toxicity remain ill-defined. To most effectively target AβOs diagnostically and therapeutically, their toxic structures must be well understood (Benilova et al. 2012, Goure et al. 2014, Lesne 2013, Sengupta et al. 2016, Brody et al. 2017). The difficulty in characterizing AβO structures lies in their molecular heterogeneity and metastability (Teplow 2013). Consequently, a wide range of AβO sizes and conformations have been implicated in AD (Benilova et al. 2012, Sengupta et al. 2016, Brody et al. 2017).

New strategies are needed for structure-function studies of AD-relevant AβOs. One potential strategy is to create reliable, stable synthetic AβO preparations via covalent crosslinking. An early study utilized glutaraldehyde crosslinking to stabilize synthetic Aβ40 aggregates (Levine 1995). However, glutaraldehyde can polymerize into structures large enough to crosslink unassociated Aβ species. Photo-induced crosslinking (PICUP) has been applied successfully to structure-function studies of synthetic Aβ40 oligomers up to tetramers, but not to Aβ42 oligomers (Ono et al. 2009). Utilizing modified Aβ peptides enables PICUP stabilization of oligomers up to dodecamers (Hayden et al. 2017). As larger AβOs have been implicated in AD (e.g., (Noguchi et al. 2009, Upadhaya et al. 2012, Dohler et al. 2014, Mc Donald et al. 2015)), the current techniques may not be sufficient to study all AD-relevant species.

To stabilize larger AβOs, we have utilized the bifunctional crosslinking reagent 1,5-difluoro-2,4-dinitrobenzene (DFDNB). DFDNB reacts with amines (Lys, Arg, N-terminus) through nucleophilic aromatic substitution with its fluorine atoms to yield stable arylamine bonds. DFDNB also reacts reversibly with sulfhydryls (Cys), imidazoles (His), and phenolates (Tyr) (Green et al. 2001). The crosslinking arm of DFDNB is 3-5 Å, restricting crosslinking to Aβ species already associated. Preliminary indications show DFDNB is capable of stabilizing AβOs in the 100-200 kDa range (Grimm et al. 2007). Thus, DFDNB crosslinking is an attractive stabilization strategy for structure-function studies of AβOs larger than dodecamers.

Here, we apply DFDNB crosslinking to a commonly utilized metastable AβO preparation. The products are AβOs predominantly of 50-300 kDa that are detergent-insensitive and remain soluble with increased time and concentration. The DFDNB-stabilized AβOs induce tau hyperphosphorylation, inhibit choline acetyltransferase, and provoke neuroinflammation. Most interestingly, DFDNB crosslinking stabilizes an AβO conformation particularly potent in inducing memory dysfunction in mice. Altogether, these data support the utility of DFDNB crosslinking for stabilizing pathogenic AβOs, thereby facilitating structure-function analysis.

Materials and Methods

Aβ Antibodies:

mAbs NU2 and NU4 were obtained by vaccination of mice with synthetic AβOs, as previously (Lambert et al. 2007). NUsc1 was prepared free of phage from the Tomlinson I + J human scFv library, as previously (Sebollela et al. 2017). NU2, NU4, and NUsc1 will be provided upon reasonable request. ACU-193 (human mAb) was a gift from Acumen Pharmaceuticals (Krafft et al. 2013, Savage et al. 2014). 6E10 was purchased (BioLegend 803001; RRID:AB_2564653).

AβO preparation and DFDNB crosslinking:

AβOs were prepared as before, solubilizing Aβ(1-42) (California Peptide 641-15) at 0.03, 0.2 (Velasco et al. 2012) or 100 μM Aβ (Chromy et al. 2003) in Ham’s F12 medium (Caisson Laboratories HFL05) or sodium borate buffer (25 mM Borax, pH 8.5). Soluble AβOs were quantified via Coomassie Plus Protein Assay (ThermoFisher 23238) and used immediately or stored (−80°C).

Crosslinking with 1,5-difluoro-2,4-dinitrobenzene (DFDNB) was as described (Grimm et al. 2007). AβOs (100 μM/sodium borate buffer) were incubated for 10 min at room temperature (RT) with DFDNB (ThermoFisher 21525; reconstituted to 20 mM/DMSO) at a 5-fold molar ratio of DFDNB:Aβ, or various ratios, where indicated. DFDNB was quenched by dithiothreitol (DTT, Sigma 646563; reconstituted to 20 mM/sodium borate), in a molar equivalent to DFDNB, for ≥30 min/RT. Prior to i.c.v., free DFDNB and DTT were removed via 6x exchange into sodium borate by ultrafiltration (3 kDa MWCO, Amicon Ultra, Millipore UFC900396).

SDS-PAGE and Immunoblotting (Dot and Western):

Dot immunoblots: AβOs (≥200 μL) were vacuum-concentrated (Minifold I 96-well system, GE 09-927-322) onto pre-wetted nitrocellulose (0.45 μm). SDS-PAGE: AβOs diluted in Laemmli buffer (BioRad 1610737) were resolved on a 4-20% Tris-Glycine polyacrylamide gel (Novex, ThermoFisher XP04205) at 15 (Western) or 45 (Silver Stain; Silver Xpress, ThermoFisher LC6100) pmol/well with Precision Plus Kaleidoscope Protein Marker (BioRad 1610375). Immunoblots (nitrocellulose, 0.45 μm) were blocked for 1 h/RT with 5% (w/v) milk/TBS-T (20 mM Tris-HCl, 150 mM NaCl, 0.1% (v/v) Tween-20, pH 7.4), then probed with primary antibodies (1.5 μg/mL in block) for 90 min/RT (Western) or overnight/4°C (dot). Immunoblots were washed 3×10 min/TBS-T, incubated 1 h/RT with HRP-linked anti-mouse IgG (GE NA931, RRID:AB_772210; 1:20,000 in block), washed 3×10 min/TBS-T, then 3x briefly/ddH₂O. Signal was developed using half-strength SuperSignal West Femto Maximum Sensitivity Substrate (ThermoFisher 34095) and imaged on a Kodak Imaging Station. Blots were analyzed using Fiji, v2.0.0 (NIH; RRID:SCR_002285).

Molecular weight cutoff (MWCO) analysis:

AβOs (30 nM/sodium borate) were fractionated by 50, 100, or 300 kDa MWCO spin filters (Amicon Ultra, Millipore UFC5050, UFC5100, UFC8300), using manufacturer’s protocol. Filtrates and retentates were analyzed via dot immunoblotting (NU4).

FPLC-SEC:

A Superdex 200 Increase 10/300 SEC column (GE 28990944), attached to an Akta Explorer FPLC (GE discontinued), was pre-equilibrated with ≤3 column volumes of PBS at 4°C. Samples were filtered at 0.2 μm (Nanosep, Pall Z721980), injected at 100 μL (100 μM Aβ), 1 mL (30, 200 nM Aβ), or 2 mL (30 nM-6E10) and eluted at 0.75 mL/min. Fractions were collected in 1 mL (30 nM-6E10) or 250 μL (all other) intervals and the entire fraction immediately vacuum-concentrated onto nitrocellulose for dot immunoblot analysis.

Time-temperature stability of DFDNB-AβOs:

AβOs ± DFDNB (5-fold) were incubated at 4°C (4 weeks; solubility analysis) or 37°C (12 days; immunoreactivity analysis). Aliquots were removed periodically. Solubility analysis: aliquots were centrifuged at 14,000 rcf/10 min/4°C and supernatant quantified via Pierce Coomassie assay. Immunoreactivity analysis: aliquots were analyzed via dot immunoblotting (ACU-193). Immunoreactivity was quantified relative to AβO standards (100 μM/F12) using linear regression in Excel (Microsoft).

Atomic Force Microscopy (AFM):

AβOs (100 μM/sodium borate ± 5-fold DFDNB) were incubated for 7 days/4°C. AβOs (10 μg) diluted to 1 or 20 μM were incubated on freshly cleaved mica (Ted Pella 50) for 10 min/RT. To wash, mica was inverted and agitated over two ddH₂O drops for 5-10 sec each and air-dried. Samples were imaged dry with the Bruker ICON PT system in tapping mode. Several regions were scanned to examine sample uniformity.

Native Top-Down Mass Spectrometry (nTDMS):

AβO buffer was exchanged to 50 mM ammonium acetate (pH 8.5) with 3 kDa MWCO filters (Amicon® Ultra); final [Aβ] = 10-20 μM. nTDMS was performed with a customized ThermoFisher Q Exactive HF Mass Spectrometer with Extended Mass Range (EMR) (Belov et al. 2013). MS¹: AβOs were infused directly into a native electrospray ionization source at +2 kV with instrument in EMR MS mode. MS²: Aβ peptides were activated and ejected via isolation of complete AβO ion population using a quadrupole mass filter and collisional dissociation in the higher energy collision dissociation (HCD) cell (100-150 eV activation voltages). MS¹ and MS² data were analyzed using ThermoFisher Xcalibur 4.0 software. MS³: AβOs were activated at the source via CID energy (150 eV); released peptides were isolated using the quadrupole and fragmented. Additional vibrational activation of the ejected peptides via collisions in the HCD cell yielded backbone fragmentation products that were measured at isotopic resolution (120,000 resolving power). Data were analyzed using ProSight Lite Software (Fellers et al. 2015). Acquisition parameters: capillary temp, 330°C; FT resolution, 15,000; S-lens RF level, 50; scan-range, 500−15,000 m/z; desolvation, insource CID 1-195 eV; polarity, positive; microscans, 10; AGC target, 3×106; maximum injection time, 500; averaging, 0; source DC offset, 25 V; injection flatapole DC, 6 V; inter flatapole lens, 5 V; bent flatapole DC, 2 V, transfer multipole DC tune offset, 0 V; Ctrap entrance lens, 1.5-2.0 V; trapping gas pressure setting, 1-4. Spectra represent summation of ~100-500 scans.

ELISA:

AβOs (100 μL, 100-1.25 pmol/well in PBS) were incubated in high-absorption 96-well plates (MaxiSorp, ThermoFisher 44-2404-21) overnight/4°C. Plates were blocked with 3% BSA/PBS (2×10 min; 1×30 min) and primary antibodies (NUsc1, NU2, ACU-193) added at 16 μg/mL in 3% (w/v) BSA/PBS (block) and incubated for 2 h/RT. Plates were washed 3x briefly with 0.1% (v/v) Tween-20/PBS (wash) and incubated with HRP-conjugated antibodies (NUsc1: 6xHis, Proteintech HRP-66005; NU2: anti-mouse, RRID:AB_772210; ACU-193: Protein A, GE NA9120) 1:1000 in block for 45 min/RT. Following three brief washes, signal was developed using 3,3’,5,5’-Tetramethylbenzidine (Thermo Scientific 34028), then stopped by 1 N sulfuric acid. Absorbance was measured at 450 nm. Binding curves were fit with a 4-parameter nonlinear regression equation (y = D + ((A-D)/(1 + (x/C)^B); A = lower asymptote, B = Hill slope, C = EC50, D = upper asymptote) by the method of least squares in Excel (Microsoft).

Primary neuronal cultures and immunocytochemistry:

Embryonic rat hippocampal tissue (gestation day 18; BrainBits SDEHP) was dissociated with papain and trituration through a silanized pipette and cells were cultured in NbActiv4™ (BrainBits Nb4-500), using manufacturer’s instructions.

Binding and tau phosphorylation (pTau) assays:

Assays were carried out essentially as before (De Felice et al. 2008). Cells matured for 21 days in vitro (DIV) in glass 96-well plates were exposed to AβOs (500 nM) or vehicle for 6 h/37°C. Cells were fixed by adding an equal volume of 3.7% formaldehyde/PBS to the media for 10 min followed by complete replacement with 3.7% formaldehyde/PBS for 10 min. Cells were washed twice briefly then 3×5 min in PBS, then blocked with 10% normal goat serum (MP Biomedicals 191356)/PBS/1 h. Cells were labeled with phalloidin (ThermoFisher A12379; 5 units/mL), using manufacturer’s protocol, then incubated with NU4 (1 μg/mL in block) or anti-pTau antibodies against pT205, pT231, or pS396 (rabbit polyclonals; RRIDs:AB_261738, AB_261755, AB_261757; 1:500 in block) overnight/4°C. Cells were washed 3×5 min/PBS, incubated ≤3 h with Alexa Fluor-linked secondary antibodies (anti-mouse, Sigma A21422; anti-rabbit, Sigma A11034; 1:2000 in 1% normal goat serum/PBS), and washed. To visualize nuclei, Hoescht (ThermoFisher 3570; 2 μg/mL) was added immediately prior to imaging.

All conditions were represented by 4-5 wells on each plate, grouped in rows to enable use of multichannel pipets. Plates were imaged at 10 and 40x on the robotic ImageXpress Micro Confocal High Content Imaging System (Molecular Devices). A few wells/condition were scanned to set exposure and focus for entire plate. Without user intervention, the robotic system imaged a 5×5 grid/well, progressing serpentine through the plate.

To quantify pTau, images were thresholded to remove ~99% signal due to vehicle treatment in Fiji (Max Entropy); remaining pixel area was normalized by cell count (Hoescht staining). Images containing no cells, or cells clumped such that individual nuclei couldn’t be counted, were excluded. ≤20 cells/well were analyzed.

NKAα3 (α3 subunit of the NaK ATPase receptor) assay:

Cells matured for 16 DIV on glass coverslips coated with poly-D-lysine (Sigma P6407) were exposed to AβOs (200, 500 nM) or vehicle for 1 h/37°C. Cells were fixed and labeled as above with 0.1% (v/v) Triton X-100 in the block for membrane permeabilization (4 h/RT). Primary antibody was anti-Na/K ATPase-α3 (H-4, mouse mAb, RRID:AB_10848453; 4 μg/mL in permeabilization buffer) with an Alexa-linked secondary (Sigma A11029). Coverslips were mounted to glass slides with ProLong Diamond Antifade containing DAPI (ThermoFisher P36962), air-dried, and imaged using a Leica Spinning Disk Autofocus Confocal System with a CSU-X1 spinning disk head (Yokogawa Electric Corporation), a 63x Plan-Apo objective with 1.4 NA (Leica), and an Evolve 512 Delta EMCCD camera (Photometrics). Images were captured with Metamorph (Molecular Devices; RRID:SCR_002368) and deconvolved using AutoQuant X3’s (Media Cybernetics; RRID:SCR_002368) iterative, constrained 3D deconvolution method. Alternatively, images were captured on a Leica DM6B Widefield Fluorescent Microscope and 3D deconvoluted using Leica software. Images were thresholded to remove background; remaining pixel area quantified and particles analyzed in Fiji. Measurements were normalized to cell count. All imaging and analysis was performed blinded to treatments.

Choline acetyltransferase (ChAT) activity assay:

Avian retinal neurons were cultured from 9-day-old chick embryos (Tolomey hatchery, Rio de Janeiro, Brazil), as previously (Nunes-Tavares et al. 2012). Briefly, retinas were dissected, digested with 0.05% (v/v) trypsin, and dissociated by mechanical aspiration through a large bore pipette. Cells were cultured in 24-well plastic plates (TPP; Sigma Z707910) at 10⁶ cells/mL (0.5 mL/well) in minimum essential media (MEM; Gibco 11700-077) with 1% (v/v) fetal calf serum ± 2 mM GABA (Sigma A2129). After 4 DIV, cells were exposed to various concentrations of DFDNB-AβOs for 17 h/37°C. ChAT was measured by monitoring conversion of [³H]acetyl-CoA (PerkinElmer NET290050UC) into [³H]acetylcholine, as previously (Fonnum 1975). Results were expressed as percentage of vehicle treatment.

Animals:

Male B6SJLF1/J mice (Jackson Laboratories, RRID:IMSR_JAX:100012) were utilized at 4-6 months of age (30-50 g). Mice were kept under a 12/12 h light/dark cycle (7 AM/7 PM) at 22 ± 2°C. Mice had free access to food and water, including during behavioral experiments, and were housed at ≤5/cage (NexGen IVC, Allentown) with enriched environment and daily veterinarian assessment, according to NU’s standard procedures. Procedures complied with NIH’s Guide for the Care and Use of Laboratory Animals (NIH publication No. 80-23, 1996) and were approved by IACUC (protocol #IS00004010). Behavioral experiments were conducted between 12-6 PM.

AβO i.c.v. administration in mice:

I.c.v. injections and behavior testing were performed in 4 independent experiments of 13-21 mice each. Littermates were assigned arbitrarily to different injection groups, targeting 5-10 mice/group for statistical power (n = ((Z_α/2*σ)/E)² at α = 0.05; σ = 10.55 and E = 6.67 derived from pilot studies).

Mice were lightly anesthetized (2% isoflurane) during injection (~1 min). AβOs (1, 10 pmol in 3 μl) or vehicles were administered i.c.v. free-handed (Bicca et al. 2015). Separate needles were used for each vehicle, progressing from low-high AβO concentration to minimize carryover. No analgesics or anti-inflammatory agents were necessary. Mice were monitored constantly for recovery of consciousness and ambulation, then periodically for food-and-water intake until behavior analysis. Needle placement was confirmed by brain dissection after behavioral experiments (euthanization: CO₂ then decapitation). Mice showing needle misplacement (3 mice) or cerebral hemorrhage (2 mice) were excluded from analysis; final n = 5-7 mice/group.

Object Recognition (OR) Tasks:

Tasks were performed essentially as described (Bicca et al. 2015), to evaluate mouse ability to discriminate between familiar and new, or displaced, objects within an arena, measured by object exploration (sniffing, touching). The open-field testing arena was constructed of gray polyvinyl chloride at 21×21×12” (WxLxH), with a 5×5 square grid on floor and visual cue on wall. 24 h post-injection, mice underwent 6 min sessions of habituation and training, with 3 min between. All sessions were video recorded and analyzed by two researchers blind to experimental groups. During habituation and training, mice were screened for ability to move about the arena and explore the objects, two activities required for accurate memory assessment in subsequent testing sessions. Locomotive inclusion criteria (>100 grid crossings and >15 rearings; evaluated in habituation) were based on extensive previous experiments with the same mouse strain and arena; 3/65 mice did not meet this criterion. During training, mice were placed at the arena center with two objects, which were plastic and varied in shape, color, size and texture. Exploration inclusion criteria were low exploration (<3 sec total) or object preference (>50% of total time); 7 of remaining 62 mice did not meet this criterion. Supplementary Table 1 shows number of mice meeting each inclusion and exclusion criterion and the final sample size.

Supplementary Figure 7a depicts the OR Task timeline. Hippocampal-related memory function was assessed 24 h post-training by displacing one of the two training objects. Cortical-related memory function was assessed 24 h later by replacing the displaced object with a novel object. Hippocampal-related memory function was re-tested 31-38 days post-injection by displacing the novel object. Memory dysfunction was defined as exploration of the familiar object for >40% total time. Mice were arbitrarily assessed by cage. The arena and objects were cleaned between sessions with 20% (v/v) alcohol to minimize olfactory cues.

Analysis of pTau and inflammation in i.c.v.-injected mouse brains:

Following OR Tasks, mice were anesthetized with ketamine (150 mg/mL) and xylazine (20 mg/mL) in 0.9% (w/v) NaCl (0.1 mL cocktail/10 g body weight) and then transcardially perfused (0.9% (w/v) NaCl). Following euthanasia, brains were midline dissected, half fixed, half flash-frozen (lN₂), and stored (−80°C). Western immunoblotting: Frozen cortices were thawed on ice briefly, then homogenized manually in radio immunoprecipitation lysis buffer (RIPA). Soluble protein was extracted after two rounds of centrifugation (14,000 rcf/45 min/4°C), quantified by BCA (ThermoFisher 23225), and resolved by SDS-PAGE (30 μg/lane) as above, except with 1% (v/v) β-mercaptoethanol. Primary antibodies were diluted 1:1000 in block: anti-actin (Santa Cruz sc-47778), AT8 (RRID: AB_223647), and anti-GFAP (Santa Cruz sc-9065; 1:1000). Band density was quantified in Fiji and normalized to actin. Mean percentage of control were plotted ± SEM. Difference in group means was tested by 1-way ANOVA with repeated measures Bonferroni. Immunohistofluorescence: Hemibrains were fixed with 3.7% (v/v) formaldehyde/PBS (48 h/RT), 10% (w/v) sucrose/PBS (24 h/4° C), then 20% (w/v) sucrose/PBS (24 h/4° C). Free-floating slices (50 μm) were obtained via freezing microtome (Leica) and stored in 0.03% azide (w/v)/PBS at 4°C until staining. Slices were rinsed 2×10 min/TBS, incubated 5 min in 10% (v/v) methanol/3% (v/v) H₂O₂/TBS to inhibit endogenous peroxidases, and washed with TBS in 5-min intervals until bubbles ceased and slices sank. Slices were rinsed with 0.3% (v/v) Triton X-100/TBS (TBS-X) 3×10 min and blocked with 10% (v/v) normal goat serum/TBS-X for 20 min/RT. Slices were incubated with anti-GFAP-Cy3 (Sigma C9205; 1:1000 in block) or AT8 (1:500) followed by Alexa Fluor-linked secondary (ThermoFisher A11029). Primary incubations were overnight/4°C and secondary 2 h/RT, both with gentle orbital agitation. Slices were washed 5×10 min/TBS-X and mounted with ProLong Gold Antifade reagent containing DAPI and imaged on a Leica DM6B Widefield Fluorescent Microscope. Exposure time was set using secondary-only negative controls. Images were 3D deconvoluted using Leica software.

Statistical analyses:

Data normality was confirmed via the Shapiro-Wilk test and variance homogeneity via Levene’s test. For behavioral experiments, difference of exploration time for the familiar vs. displaced/novel object was tested by an unpaired Student’s t-test and between experimental groups by a two-way ANOVA with repeated measures Bonferroni. For in vitro experiments, difference of experimental condition vs. control was tested using one-way ANOVA with repeated measures Tukey’s or Bonferroni, as indicated. No outliers were removed; all data plotted as mean ± SEM. GraphPad Prism software (v5, RRID:SCR_002798) was used for statistical analysis (α = 0.05).

Results

Synthetic AβOs are metastable and show a size distribution that is concentration-dependent.

Many studies utilize use supra-pathophysiological synthetic Aβ concentrations, i.e., at or near 100 μM (Lambert et al. 1998, Benilova et al. 2012). However, soluble Aβ could be ≤0.1 nM in the AD brain, according to ISF and CSF measurements (Herukka et al. 2015, Georganopoulou et al. 2005, Savage et al. 2014). To investigate the impact of starting monomer concentration, AβOs were formed at various concentrations (100-0.03 μM) and analyzed via size exclusion chromatography (SEC) with dot immunoblotting detection. Figure 1a shows that AβO heterogeneity decreases with starting Aβ concentration, detected by the anti-Aβ antibody 6E10. At 100 μM, a broad distribution of Aβ species are detected (~10-700 kDa; ~2-8 nm Stokes radius, Supplementary Figure 1). Ultrafiltration analysis indicates that either the molecular weight of the less-retained species is over-estimated by SEC (perhaps due to shape) or these species are unstable (Yang et al. 2017), as 100% of this AβO preparation passes through a 300 kDa MWCO filter (Figure 1b, left). At 200 nM Aβ, the distribution narrows to an apparent ~30-200 kDa (~2.4-5.3 nm, Supplementary Figure 1). Strikingly, at 30 nM Aβ, only one peak is observed, at an apparent ~550 kDa. However, the raw immunoblot signal for this species was very low (Supplementary Figure 1).

Figure 1: Synthetic AβOs are metastable and show a size distribution that is concentration-dependent.

AβOs were prepared at 30 nM in sodium borate (red; square), 200 nM in F12 (green; diamond), or 100 μM in F12 (purple; triangle), fractionated by SEC, and analyzed via dot immunoblotting with anti-Aβ antibody 6E10 (A) or anti-AβO antibody NU2 (C). Signal was normalized to maximum intensity for each condition ± SEM (n = 3 independent experiments). B) AβOs prepared at 100 μM (left) or 30 nM (right) were fractionated using 50, 100, or 300 kDa MWCO filters and quantified with a modified-Bradford assay (100 μM Aβ; mean ± SEM, n = 2 technical replicates) or dot immunoblotting with anti-AβO antibody NU4 (30 nM Aβ), loading equal volumes of the filtrates and retentates. D) AβOs prepared at 30 nM were incubated at 4°C (blue, square) or 37°C (red, circle) and assessed daily for NU4 immunoreactivity via dot immunoblotting. Data plotted as mean percentage of immunoreactivity at day 0 ± SEM (n = 3 technical replicates).

To increase signal, we utilized MWCO fractionation instead of SEC, which divides the population into fewer fractions. Dot immunoblotting detection was via the anti-AβO antibody NU4, which has a higher AβO affinity than 6E10 (Lambert et al. 2007). Figure 1b (right) shows that while some NU4-reactive AβOs are retained by the 300 kDa MWCO filter, the majority are <300 kDa. We further analyzed the 30 nM population with SEC, this time using the anti-AβO antibody NU2, which detects AβO species not recognized by 6E10 (Lambert et al. 2007). Again, a single peak is observed, although at ~100 kDa (5.4 nm, Supplementary Figure 1), which is consistent with the NU4 MWCO data (Figure 1b, right).

One hypothesis for the discrepancy in the 6E10 and NU2 SEC analysis of the 30 nM population is that the less-retained, 6E10-reactive species are protofibrillar, their shape hindering column entry (6E10 is fibril-reactive). The more-retained, NU2-reactive species may be globular. Conformational instability also may contribute to SEC inconsistencies, evidenced by inability of the 30 nM preparation to maintain NU4 immunoreactivity over time (Figure 1d). The 100 μM and 200 nM populations also show different NU2- and 6E10-reactive species, but the same trend; i.e., AβO heterogeneity decreases with starting Aβ concentration.

DFDNB stabilizes Aβ in a soluble oligomeric conformation.

Due to the metastability of AβO preparations, we sought to stabilize AβOs with DFDNB crosslinking. Preliminary characterization of DFDNB-stabilized AβOs has been reported, but not their biological relevance (Grimm et al. 2007). Henceforth, we utilize an Aβ concentration of 100 μM to recapitulate this previous report.

First, DFDNB ability to stabilize AβO conformation was investigated. AβOs were incubated ± DFDNB at 4°C and assessed weekly for loss of solubility. As expected, AβOs progressed to insoluble aggregates in the absence of DFDNB, with 70% of the preparation becoming insoluble after 4 weeks (Figure 2a). By contrast, only 10% of DFDNB-linked AβOs became insoluble. Next, AβOs ± DFDNB were monitored for maintenance of AβO conformation at 37°C via the AβO-selective antibody ACU-193, which has negligible affinity for Aβ monomers or fibrils (Krafft et al. 2013, Savage et al. 2014). Consistent with AβOs formed at 30 nM (Figure 1d), 100% of Aβ lost ACU-193 reactivity after 12 days in the absence of DFDNB (Figure 2b). With DFNDB crosslinking, 100% maintained immunoreactivity. These analyses show that DFDNB locks AβOs into a soluble oligomeric conformation.

Figure 2: AβOs remain soluble over time if crosslinked with DFDNB.

A-B) AβOs were prepared at 100 μM peptide ± a 5-fold molar excess of DFDNB and incubated at 4°C for 4 weeks (A) or 37°C for 12 days (B). Aliquots were periodically removed and soluble protein quantified via a modified-Bradford assay (following high-speed centrifugation) (A) or immunoreactivity quantified with the AβO-specific antibody ACU-193 via dot immunoblotting (B). C) Increasing molar excesses of DFDNB were added to AβOs and resulting mixtures analyzed by Western immunoblotting using anti-AβO antibody NU4. D) DFNDB-Aβ mixtures were centrifuged at 14,000 rcf (10 min, 4°C) to remove insoluble aggregates and soluble AβOs quantified via modified-Bradford assay. Mean ± SEM plotted; n = 2 (A, D) or 3 (B) technical replicates.

Next, we investigated the DFDNB impact on SDS stability of AβOs. AβOs were incubated with increasing amounts of DFDNB and analyzed via Western immunoblotting using NU4. Without DFDNB, AβOs formed in sodium borate show only small differences in SDS-stable distribution from the standard F12 preparation (Supplementary Figure 2a). As the amount of DFDNB is increased relative to Aβ, the predominant populations stabilized are a NU4-reactive population at centered at ~80 kDa (Figure 2c), followed by a NU4-nonreactive species at ~160 kDa (Supplementary Figure 2b). Lower molecular weight, NU4-nonreactive species are present at high DFDNB concentrations, but in lesser abundance. To determine if insoluble aggregates were stabilized, or created, by DFDNB, we removed insoluble aggregates via high-speed centrifugation and quantified the remaining soluble protein. No significant amount of soluble protein was lost, even with a 100-fold molar excess of DFDNB (Figure 2d).

To further investigate this conformational stabilization, DFDNB-stabilized AβOs were incubated at 4°C for 1 week and then imaged via atomic force microscopy (AFM). We analyzed AβOs stabilized with a 5-fold excess of DFDNB, as this condition yielded the highest percentage of stabilized, NU4-reactive species. As expected, AβOs crosslinked with DFDNB appeared globular (Figure 3). Z-height analysis indicated that when the amount of AβOs deposited onto the mica is increased from 10 to 200 pmol, uncrosslinked AβOs increase 3-fold in size, but DFDNB-linked AβOs increase only 1.2-fold. These data further indicate that DFDNB locks AβOs into a soluble conformation not prone to further aggregation induced by time, temperature, or concentration.

Figure 3: AFM shows DFDNB crosslinking stabilizes AβO conformation against concentration-dependent aggregation.

AβOs were prepared in sodium borate ± DFDNB (5-fold excess), incubated for 7 days at 4°C, and absorbed onto mica at 10 (top) or 200 (bottom) pmol. Samples imaged dry by AFM in tapping mode. Image dimensions = 1.0×1.0 μm. Population z-heights are plotted (AβOs, black; AβOs + DFDNB, red; n = 386–499 measurements/sample). Several regions were scanned to examine sample uniformity.

Next, DFDNB stabilized AβOs were analyzed by native top-down mass spectrometry (nTDMS). nTDMS utilizes electrospray ionization at pH ~7 to analyze intact, non-covalent protein complexes (Belov et al. 2013, Skinner et al. 2016). Our workflow first analyzes the total mass of intact AβOs (MS¹), followed by collisional activation with nitrogen gas to eject non-covalently bound Aβ subunits (MS²), then further activation for backbone fragmentation of ejected Aβ (MS³). We analyzed AβOs crosslinked under conditions that maximize formation of the 80 kDa NU4-reactive species (5-fold DFNDB; “XL”) and the 160 kDa species (40-fold DFDNB; “HXL”). MS¹ analysis (Supplementary Figure 3) shows that increasing DFDNB concentration decreases AβO heterogeneity. The charge, and therefore mass, of these AβOs cannot be calculated precisely due to high spectral heterogeneity. However, using an equation for multiply charged species produced by electrospray ionization $(z=0.0778m^{1/2)}$ (Fernandez de la Mora 2000), the mass of UXL AβOs is estimated at 151-732 kDa, narrowing to 139-581 kDa with 5-fold DFDNB (XL), then to 97-581 kDa with 40-fold DFDNB (HXL). The median mass for the HXL population was 204 kDa ± 29 kDa (SEM; n = 3). MS² analysis demonstrates that Aβ peptides can be ejected from the UXL and XL populations, with DFDNB-linked Aβ ejected from the latter (Supplementary Figure 3, top-right). No Aβ peptides could be ejected from the HXL population, indicating complete covalent stabilization. MS³ analysis of the XL population localizes DFDNB attachment to Aβ residues Y10, H13, or H14 in AβO-ejected monomers ejected (Supplementary Figure 3, bottom). Both fluorine leaving groups are absent from DFDNB, although DFDNB is only attached to one residue, likely indicating that one of the DFDNB-Aβ bonds is cleaved in the spectrometer. Overall, nTDMS analysis substantiates the SDS-PAGE analysis in that increasing amounts of DFNDB progressively stabilize AβOs predominantly in the range of 50-300 kDa, while narrowing the apparent molecular weight distribution.

DFDNB crosslinking stabilizes biologically active AβO species.

Next, we investigated if DFDNB crosslinking disrupts the AD-associated pathobiological activity of AβOs. First, we tested for any DFDNB-induced changes in immunoreactivity to various AβO-selective antibodies shown immunoprotective in culture: NUsc1 (Sebollela et al. 2017), NU2, and ACU-193. Utilizing dose-response ELISA assays, we find that DFDNB (5-fold) does not significantly alter AβO immunoreactivity (Supplementary Figure 4). Statistically, the apparent AβO affinity, as measured by EC50, was increased by DFDNB stabilization for NUsc1 (p = 0.02; Student’s t-test). Considering NUsc1 targets a subset of NU2-reactive AβOs that are >50 kDa (Velasco et al. 2012, Sebollela et al. 2017), this small, but statistically significant, EC50 difference may indicate that DFDNB is stabilizing this NUsc1-reactive subset, especially considering the enrichment of AβOs within 50-300 kDa in this preparation. Additional testing is needed to validate this hypothesis.

Next, we tested whether DFDNB modified the already established abilities of AβOs to induce the following AD-associated cellular pathologies: synapse binding (Lacor et al. 2004), tau hyperphosphorylation (De Felice et al. 2008), re-distribution of the NaK ATPase receptor α3 subunit (NKAα3) (DiChiara et al. 2017), and inhibition of choline acetyltransferase (ChAT) (Nunes-Tavares et al. 2012). To test synapse binding, DFDNB-linked AβOs (5-fold DFDNB) were incubated with hippocampal cultures for 1 or 4 h. After treatment, cells were fluorescently probed for AβOs (NU4) and f-actin (phalloidin) to visualize dendrites. A low magnification overview shows that DFDNB-linked AβOs bind most hippocampal cells, but not all (Figure 4, top-right), as is typical for AβOs (Lacor et al. 2004). A higher magnification reveals the typical AβO punctate binding pattern (Figure 4, bottom-right) (Lacor et al. 2004, Viola et al. 2015). Unexpectedly, AβOs prepared in sodium borate without DFDNB showed little evidence of cell binding, even after 4 h (Figure 4, middle). Therefore, DFDNB stabilization in this buffer is crucial for AβO activity; partial stabilization in F12 is likely conferred by presence of metals (Sharma et al. 2013, Atwood et al. 2004).

Figure 4: AβOs stabilized by DFDNB retain high capacity to bind dendritic spines.

21 DIV primary hippocampal neurons were treated for 1 or 4 h with vehicle (left) or 500 nM AβOs without DFDNB (middle) or with DFDNB (5-fold excess; right). Neurons were visualized with a fluorescent antibody against phalloidin (f-actin; green) and AβOs (NU4; red) at 10x (top; scale = 100 μm) and 40x (bottom; scale = 25 μm). Images are representative of 100 images/condition.

To determine if DFDNB stabilization alters AβO ability to induce tau hyperphosphorylation, cells treated with AβOs ± DFDNB (5-fold excess) were probed for tau phosphorylated (pTau) by immunofluorescence. Three AD-associated epitopes - T205, T231, and S396 - were targeted for greater confidence in biological significance. Figure 5 shows that DFDNB stabilization did not alter AβO ability to induce hyperphosphorylation at these three epitopes, as AβOs ± DFNDB increased pTau relative to their respective vehicles (AβOs: p = 0.001, q = 9.49, df = 6; DFDNB-AβOs: p = 0.013, q = 5.98, df = 6). There was no statistically significant difference in the response of the two populations. Individual epitope data are shown in Supplementary Figure 5. Consistent with their decreased ability to bind neurons, AβOs formed in sodium borate without DFDNB showed little ability to induce tau hyperphosphorylation (not shown).

Figure 5: AβOs stabilized by DFDNB retain high capacity to induce tau hyperphosphorylation.

21 DIV primary hippocampal neurons were treated for 6 h with vehicle or 500 nM AβOs prepared ± DFDNB (5-fold excess). Tau phosphorylation was probed using a fluorescent antibody against pTau at T205, T231, or S396. Representative images of pTau[T205] are shown (scale = 25 μm). n = 3 independent experiments; each experiment comprises 4–5 quasi-independent cultures (wells of plate) probed for 1 of the 3 pTau epitopes, 25 images/culture, 1–30 cells/image. Fluorescence was normalized to cell count and respective vehicle treatment ± SEM. p < 0.05 (*) or 0.005 (**) in comparison to Veh (ANOVA and Tukey’s test).

Next, we tested if DFDNB-stabilization altered AβO ability to modify NKAα3 membrane distribution. Supplementary Figure 6a shows that DFDNB-AβO treatment trends toward increased NKAα3 clustering, indicated by increased particle size (p = 0.17, t = 2.09, df = 2) and decreased particle count (p = 0.32, t = 1.18, df = 3). At a lower concentration, DFDNB-AβO treatment trends toward decreased NKAα3 detection on neuronal processes (Supplementary Figure 6b; p = 0.05, q = 4.34, df = 6). This is more complete at dendritic sites farther from the cell body. Interestingly, the NKAα3 soma staining pattern in DFDNB-AβO-treated cells appears to show internal vesicular trafficking of NKAα3 (Supplementary Video 1). Although apparent trends are in harmony with properties of uncrosslinked AβOs (DiChiara et al. 2017, Zhao et al. 2010, Snyder et al. 2005, De Felice et al. 2009), the sample size must be expanded to confirm trends.

Lastly, we tested the effect of DFDNB stabilization on AβO ability to inhibit ChAT activity. Exposure to DFDNB-AβOs (0.25-2 μM) resulted in a dose-dependent decrease of ChAT activity in cultured avian retinal neurons, which could be blocked by pre-treatment with GABA (Figure 6). This is consistent with our previous observations with unstabilized AβOs (Nunes-Tavares et al. 2012, Louzada et al. 2004) and further supports our hypothesis that DFDNB is stabilizing pathogenically active AβO species.

Figure 6. AβOs stabilized by DFDNB induce ChAT inhibition in cultured neurons.

Primary avian neurons were cultured with (blue, circle) or without (red, triangle) 2 mM GABA and then treated for 17 h with various concentrations of AβOs stabilized with DFDNB (5-fold excess). ChAT was measured by monitoring conversion of [³H]acetyl-CoA into [³H]acetylcholine. Results are expressed as mean percentage of the control (vehicle-treated; 649 pmol ACh/mg/min) ± SEM (n = 3 independent experiments).

DFDNB-stabilized AβOs are potent instigators of cognitive decline and AD neuropathology in mice.

We sought to confirm that DFDNB crosslinking does not block AβO ability to induce AD-related pathologies in vivo. To this end, we injected mice i.c.v. with AβOs and evaluated the effect on memory function using Object Recognition Tasks (study timeline in Supplementary Figure 7a). Free DFDNB was removed from AβO preparations prior to injection. In these tasks, mice are introduced to two objects within a testing arena and then later tested for ability to recognize a novel (cortical-dependent memory) or displaced (hippocampal-dependent) object (Bicca et al. 2015, Thinus-Blanc et al. 1996). In this experimental paradigm, an i.c.v. injection of 10 pmol AβOs is sufficient to impair memory at 1-14 days post-injection (Figueiredo et al. 2013).

When evaluated 48 h post-injection by the hippocampal-dependent task, vehicle-injected mice were not cognitively impaired, i.e., they explored the displaced object more than the familiar (Figure 7). As expected, 10 pmol of uncrosslinked F12 AβOs did induce memory dysfunction, evidenced by equal exploration of the displaced and familiar objects, and 1 pmol of AβOs did not. DFDNB crosslinking at 5- (“XL”) or 40- (“HXL”) fold molar excess did not block AβO ability to impair memory function at a 10 pmol dose. In fact, XL AβOs also impair memory function at 1 pmol. Surprisingly, this was not true for the HXL preparation, suggesting that the epitope(s) required for pathogenicity may be obscured or absent. Results were consistent in the cortical-dependent task (Supplementary Figure 7b).

Figure 7: ICV injections of DFDNB-AβOs show potent instigation of memory deficits.

Male mice (4–6 months) were injected i.c.v. with 1 or 10 pmol of AβOs prepared in F12 without DFDNB (UXL F12) or sodium borate (SB) and cross-linked with DFDNB at a 5- (XL) or 40- (HXL) fold molar excess. As a negative control, mice were injected with AβO vehicles (VHC). Mice were then evaluated for ability learn and recall memories in Object Recognition Tasks. Prior to testing, mice explored two objects in a testing arena, denoted F1 and F2. Top: 48 h post-injection, mice were evaluated for ability to recognize object F1 (gray) when object F2 is displaced (D; orange). Bottom: 31–38 d post-injection, mice were tested for ability to recognize object F1 when a new object is placed in a new location (D; green). Mean percent exploration time for each object is plotted ± SEM (n = 5–7 mice/group). 50% exploration level is indicated by a dashed line. *p < 0.05 for difference in exploration time of familiar vs. displaced object (Student’s t-test). Red boxes indicate two mice injected with UXL F12 AβOs that recovered normal memory function, as defined by exploration ratio of 70/30% displaced/familiar object.

To evaluate the duration of memory dysfunction, mice were re-tested 31-38 days later in a hippocampal-dependent task. Again, the mice that had been injected with vehicles or 1 pmol of uncrosslinked F12 AβOs exhibited normal memory function. Mice were not re-trained in the task and consequently, their performance decreased from 80% exploration of the displaced object at the initial testing to 70%. Of mice injected with 10 pmol uncrosslinked AβOs, one-third recovered memory function (indicated by red boxes in Figure 7), performing equally to vehicle-injected mice. 100% of mice injected with XL (1, 10 pmol) or HXL (10 pmol) AβOs remained impaired. Surprisingly, mice injected with 1 pmol of HXL AβOs had become impaired since initial testing. These data demonstrate that DFDNB stabilizes AβOs capable of inducing memory dysfunction in mice, consequently increasing the potency of the response both in dose and duration of impairment, but that the time to onset changes with the degree of DFNDB modification.

To investigate mechanistic differences between F12, XL, and HXL AβOs in inducing memory dysfunction, we measured pTau and inflammation in the brains of the injected mice via Western immunoblotting and immunohistofluorescence. 24 h post-injection, pTau was increased in the brains of mice injected with either F12 or XL AβOs, compared to vehicle (Supplementary Figure 8a), with highest levels due to XL AβO injection. This observation is preliminary due to low sample size. By contrast, neuroinflammation, measured by the astrocyte marker GFAP, was increased only in XL AβO injected mice (Supplementary Figure 8b). 30 days post-injection, average pTau levels appear higher with injection of all AβO preparations (UXL, XL, HXL) compared to vehicle at both 1 and 10 pmol doses, except for 10 pmol XL AβOs (Supplementary Figure 9). The same was true for average GFAP levels, except for a 1 pmol injection of UXL F12 AβOs. Again, the sample size must be expanded to confirm trends.

Discussion

Here, the suitability of the crosslinker 1,5-difluoro-2,4-dinitrobenzene (DFDNB) for stabilizing AβO species for structure-function studies was examined extensively. We found that DFDNB not only prevented AβO dissociation in SDS, but locked AβOs into a soluble, neurologically-active conformation. Biochemical and nTDMS analyses indicated that DFDNB stabilized AβOs predominantly in the 50-300 kDa range. DFDNB did not block AβO ability to induce AD-associated pathologies in cell or animal models. Most interesting, DFDNB stabilized an AβO conformation particularly potent in inducing memory dysfunction.

We chose DFDNB for AβO stabilization due to preliminary indications that it stabilizes AβOs >50 kDa (Grimm et al. 2007), which have been demonstrated AD relevant (Gong et al. 2003, Lesne et al. 2006, Lacor et al. 2004, Lacor et al. 2007, Noguchi et al. 2009, Upadhaya et al. 2012, Velasco et al. 2012, Mc Donald et al. 2015). Current AβO stabilization protocols, PICUP and dityrosine crosslinking, are ineffective in this size range (Ono et al. 2009, Hayden et al. 2017, Williams et al. 2015, Zhang et al. 2017), limiting analysis.

Here we utilize SDS-PAGE and nTDMS to show DFDNB stabilizes AβOs in our preparation predominantly within 50-300 kDa. Analysis confirms predictions that DFDNB does not link Aβ peptides or oligomers that were not already associated prior to DFDNB addition. Biochemical, nTDMS, and AFM analysis revealed that DFDNB not only prevented AβO dissociation during analysis, but also locked AβOs into a soluble conformation, thus facilitating fibril-free analysis. The observation also supports the idea that Aβ aggregation does not follow a linear pathway from monomer to oligomer to fibril, but rather a series of parallel processes involving different intermediate conformations (Roychaudhuri et al. 2009). Additionally, it has been observed that AβOs <50 kDa tend to form insoluble aggregates more quickly than AβOs >50 kDa (Velasco et al. 2012). The present results are consistent with these past observations, as they show that under the conditions utilized, DFDNB-stabilized AβOs in the 50-300 kDa range are not prone to further aggregate into insoluble complexes.

We present nTDMS analysis as a first step toward higher-resolution AβO structural studies. nTDMS has been used elsewhere to analyze crosslinking efficiency and the impact of crosslinking on the conformation of intrinsically disordered proteins (Arlt et al. 2017). Indeed, the present nTDMS analysis effectively demonstrated the impact of DFDNB crosslinking on the covalent stability of AβOs. Precise masses could not be assigned to AβOs due to considerable spectral complexity resulting from variability in DFDNB attachment, Aβ number per oligomer, and salt adduction to Aβ peptides. Nevertheless, the mass estimations derived from the nTDMS data were consistent with the biochemical analyses and confirm the increased stability and decreased mass heterogeneity following DFDNB stabilization.

DFDNB stabilization did not block AβO ability to bind synapses in cultured neurons (Lacor et al. 2004, Lacor et al. 2007) and subsequently induce hyperphosphorylation of tau (De Felice et al. 2008) and inhibition of ChAT (Nunes-Tavares et al. 2012). Preliminarily, DFDNB also does not block AβO ability to induce clustering and internalization of NKAα3 on neuronal membranes (DiChiara et al. 2017). Furthermore, DFDNB-stabilized AβOs exhibited increased pathogenicity in vivo, specifically in their ability to induce memory dysfunction in mice via i.c.v. injection. DFDNB stabilization was found to increase the potency of the AβO response, both in dose and duration of memory dysfunction, indicating that DFDNB stabilizes a particularly toxic AβO species. Preliminary data indicate that DFDNB-stabilized AβOs induce greater levels of tau hyperphosphorylation and neuroinflammation in the injected mouse brain.

Interestingly, partial DFDNB stabilization caused sustained memory dysfunction in mice (“XL AβOs”), with more extensive stabilization (“HXL AβOs”) delaying memory dysfunction. These differential responses may result from the differential species distributions observed by SDS-PAGE and nTDMS. Indeed, the differential impact on memory function by different AβO species has been demonstrated in injected mice (Figueiredo et al. 2013) and rats (Reed et al. 2011). Furthermore, these reports are consistent with our observation that memory dysfunction following a single injection of unstabilized AβOs is transient. The increased duration of dysfunction following DFDNB-AβO injection could be due to impaired clearance/degradation in the brain and/or differential mechanisms of action.

The latter hypothesis was tested by measuring brain levels of pTau and inflammatory markers 30 days post-injection. Mice injected with HXL AβOs showed the highest cortical levels of pTau, although inflammation levels were equivalent to mice injected with other preparations. However, this analysis has not yet been conducted 24 h post-injection, which might shed further light on the delayed memory response to the HXL preparation. The NU4 epitope may be required for fast-acting memory dysfunction, since these “HXL AβOs” were not reactive with NU4 via Western immunoblotting, which is consistent with our previous reports that NU4 is immunoprotective in culture (Lambert et al. 2007) and its targets are highly relevant to memory function (Xiao et al. 2013, Knight et al. 2016).

We propose that DFDNB crosslinking of AβOs in vitro mimics what occurs in the AD brain. It is well recognized that synthetic and brain-derived AβOs are resistant to SDS breakdown (Benilova et al. 2012). Additionally, there is evidence that AβOs are endogenously crosslinked at DFDNB-reactive sites, e.g., at Y10 induced by metals/oxidative stress-induced (Smith et al. 2007) or at Q15, K16, and K28 via transglutaminase (Kawabe et al. 2017). Furthermore, levuglandins, which are amine-reactive modified prostaglandins similar in size to DFDNB, can crosslink Aβ peptides (Bi et al. 2016), accelerating oligomerization and increasing toxicity (Boutaud et al. 2006, Boutaud et al. 2002, Zagol-Ikapitte et al. 2005).

Based on the biochemical, structural, and functional analyses conducted here, we put forth DFDNB crosslinking as a new tool for AβO structure-function studies. We propose that when applied to more homogenous AβO populations, such as synthetic preparation at ≤30 nM or AD-derived AβOs (Gong et al. 2003), DFDNB crosslinking will enable more finely-tuned structure-function studies of soluble AβOs >50 kDa. Additional application of DFDNB to link AβOs to e.g., species-selective or therapeutic antibodies, may further decrease AβO heterogeneity to facilitate analysis. As DFDNB enhances the pathogenicity of synthetic AβOs preparations, it could be beneficial for improving applications such as AD models induced by AβO i.c.v. injection (Forny-Germano et al. 2014).

List of abbreviations:

AβO(s)

amyloid beta oligomer(s)

Aβ40 or Aβ(1-40)

amyloid beta 1-40 peptide

Aβ42 or Aβ(1-42)

amyloid beta 1-42 peptide

ACh

acetylcholine

AD

Alzheimer’s disease

AFM

atomic force microscopy

ChAT

choline acetyltransferase

DIV

days in vitro

DFDNB

1,5-difluoro-2,4-dinitrobenzene

DTT

dithiothreitol

EMR

extended mass range

FPLC

fast performance liquid chromatography

GABA

γ-aminobutyric acid

GFAP

Glial fibrillary acidic protein

HCD

higher energy collision dissociation

HXL

AβOs crosslinked with a 40-fold molar excess of DFDNB

i.c.v.

intracerebral ventricular

mAb

monoclonal antibody

MS

mass spectrometry

MWCO

molecular weight cutoff

NKAα3

α3 subunit of the NaK ATPase receptor

nTDMS

native top-down mass spectrometry

PICUP

photo-induced crosslinking of unmodified proteins

pTau

phosphorylated tau

RRID

Research Resource Identifier

RT

room temperature

scFv

single chain variable fragment

SEC

size exclusion chromatography

Veh

vehicle

XL

AβOs crosslinked with a 5-fold molar excess of DFDNB

UXL

uncrosslinked AβOs