QIAD assay for quantitating a compound’s efficacy in elimination of toxic Aβ oligomers

Strong evidence exists for a central role of amyloid β-protein (Aβ) oligomers in the pathogenesis of Alzheimer’s disease. We have developed a fast, reliable and robust in vitro assay, termed QIAD, to quantify the effect of any compound on the Aβ aggregate size distribution. Applying QIAD, we studied the effect of homotaurine, scyllo-inositol, EGCG, the benzofuran derivative KMS88009, ZAβ3W, the D-enantiomeric peptide D3 and its tandem version D3D3 on Aβ aggregation. The predictive power of the assay for in vivo efficacy is demonstrated by comparing the oligomer elimination efficiency of D3 and D3D3 with their treatment effects in animal models of Alzheimer´s disease.


Results and Discussion
The principle of the QIAD assay. The QIAD assay consists of the following steps ( Fig. 1): i) preparation of Aβ (1-42) assemblies containing the target aggregate species by Aβ (1-42) pre-incubation; ii) incubation with and without the agent of interest; iii) separation of Aβ (1-42) assemblies by DGC and subsequent fractionation; and iv) determination of the total Aβ (1-42) amount in each fraction, e.g. by integrating the Aβ (1-42) absorption signal during RP-HPLC analysis. The essential requirement for the quantitative analysis of different Aβ  aggregates is total disassembly of all different Aβ (1-42) assemblies. We were able to achieve this by the harsh conditions during the RP-HPLC runs (presence of acetonitrile and column heating to 80 °C) converting all Aβ assemblies into a single species with the same retention time. Thus, the integrated peak area of this peak correlates with total Aβ (1-42) amount (Fig. 2a). The obtained results indicate that under these conditions the integrated peak areas were independent from Aβ (1-42) incubation time and thus independent from its aggregation state. Prior to RP-HPLC DGC allows matrix-free separation and fractionation of different Aβ (1-42) assemblies according to their sedimentation coefficients, which are dependent on particle size and shape 4,5 . Alternatively the Aβ (1-42) aggregate size distribution could have been studied by a combination of SEC (Size Exclusion Chromatography) and MALS (Multi-Angle Light Scattering) detection. Such approach seems to be more convenient since it would take less time for a single sample and the direct outcome would be a distribution of molecular masses instead of sedimentation coefficients. Nevertheless, DGC was superior to SEC/MALS with regard to the recovery rates of Aβ . Possible interactions of Aβ  with the column matrix and the necessity of either an online filtration or centrifugation step prior to chromatography might be responsible for material losses. Aβ (1-42) fibrils, huge aggregates or huge complexes (Aβ /ligand) could not pass through in the AD research field prevalent SEC columns with a void volume of 75 or 200 kDa. For DGC the whole sample can be loaded onto the gradient regardless of the aggregation state of the sample. Additionally, the fact, that for each sample a fresh gradient is prepared, while SEC columns have to be reused, contributes clearly to the very good reproducibility of the analysis. We used steps (i) to (iii) previously to investigate the effect of agent candidates on size distribution of Aβ (1-42) aggregates formed in vitro, although only in a qualitative or semi-quantitative way 6 . Here, for the first time, we add the fully quantitative analysis of a compound's impact on the Aβ (1-42) aggregate size distribution and call it QIAD assay (Fig. 2b,c). The principle of the QIAD assay can easily be transferred to measure the quantitative interference on aggregates consisting of aggregation-prone peptides or proteins other than Aβ . Thus, to avoid any confusion with QIAD variants yet to come, we have termed the Aβ -specific QIAD assay "Aβ -QIAD".
The method abstains from monomeric initial conditions, instead it starts from a reproducibly adjustable assembly state, where Aβ (1-42) oligomers are enriched by a pre-incubation step. While Aβ recovery rates are rarely demonstrated 7 in the AD research field, the herein described procedure reliably yielded total Aβ (1-42) recovery rates close to 100%, i.e., 95.6 ± 7.9% for control DGC runs (Aβ without agent) and 89.2 ± 16.1% for DGC runs in the presence of D3 or D3D3. Moreover, the method described here allows the quantification of yields for any DGC-separated Aβ (1-42) fraction within the sample. For example, the average amount of the Aβ (1-42) oligomers (fractions 4 to 6) in eleven independent preparations was 31.7 ± 3.9%.
Characterisation of Aβ oligomers. Aβ (1-42) assemblies located in fractions 4 to 6 were specifically sensitive to our agents D3 6,8-10 and D3D3. Therefore, these species were further characterized in detail. According to the calibration of the density gradient with a set of globular proteins with known s-values, the Aβ (1-42) assemblies in fractions 4 to 6 had s-values of about 7 S (Supplementary Figure 1a). Atomic force microscopy (AFM) analysis of fraction 4 to 6 revealed assemblies in spherically shaped particles with heights of 4.7 nm and diameters corrected for the tip dimensions of 8.7 nm ( Fig. 3a and Supplementary Figure 2). Modelled as oblate spheroids with an axial ratio of ~2, these Aβ (1-42) oligomers have a molecular weight of about 104 kDa, which corresponds to about 23 monomeric units. This value is in good accordance with the estimated s-value of 7 S. Circular dichroism (CD) measurements of the oligomers revealed a spectrum with β -sheet characteristics (Fig. 3b). However, since the Aβ (1-42) particles were negative for Thioflavin T (ThT) fluorescence staining (Supplementary Figure 3), we conclude that the secondary structure of the oligomers differs from regular amyloid cross-β fibrils, which typically exhibit strong ThT fluorescence. The DGC derived Aβ (1-42) oligomers exhibited high cytotoxicity to differentiated human SH-SY5Y neuroblastoma cells [11][12][13] (Fig. 3c), thus meeting a further criterion of toxic Aβ (1-42) oligomers involved in AD pathogenesis.
Aβ oligomer elimination by D3 and D3D3. Here, we demonstrate the power of the Aβ -QIAD assay by quantitatively comparing the outcome of the assay for two therapeutically interesting agents. The first agent is a mirror image phage display 8,14 selected D-enantiomeric peptide, termed "D3", which inhibits the formation of regular Aβ (1-42) fibrils, removes Aβ (1-42) oligomers and reduces Aβ (1-42) cytotoxicity in vitro. In vivo, D3 binds to amyloid plaques 10 , is able to reduce plaque load, decrease inflammation and enhance cognition in a transgenic AD mouse model even after oral application 6,9 . The second agent   is a head-to-tail tandem version of D3, termed "D3D3", that is expected to have enhanced efficacy due to increased avidity. As shown in Fig. 4a, both agents yielded significant reduction of Aβ (1-42) oligomers in DGC fractions 4 to 6 in comparison to agent-free controls. D3 (32 μ g/ml, 20 μ M) and D3D3 (32 μ g/ml, 10 μ M) reduced Aβ (1-42) oligomers by 50% and 96%, respectively. The oligomer elimination efficiency is defined as the percentage of reduction of Aβ (1-42) contents in fractions 4 to 6 in presence of agent as compared to absence of agent. D3D3 proved to be significantly more efficient than D3 ( Fig. 4a and Supplementary Figure 4). Notably, monomeric Aβ (1-42) located in DGC fractions 1 and 2 were negligibly affected by D3 and D3D3. The net reduction of Aβ (1-42) oligomers by D3 and D3D3 was balanced by an increase of high-molecular weight aggregates located in DGC fractions 9 to 12 or in fractions 12 to 15, respectively ( Fig. 4a and Supplementary Figure 4). Increase of Aβ (1-42) content in DGC fractions 9 to 12 or in fractions 12 to 15 does not mean that formation of pre-fibrils or fibrils is induced by D3 and D3D3. The Aβ (1-42) species in fractions 9 to 15 have completely different morphologies depending on presence or absence of D3 or D3D3. Thus, D3 and D3D3 are active in vivo and obviously both have the same mechanism of action, which is elimination of Aβ (1-42) oligomers of the same size as shown in vitro in the QIAD assay, with D3D3 being more efficient than D3, and with D3D3 yielding larger co-aggregates than D3. D3-Aβ -co-aggregates were shown previously to be non-toxic, non-amyloidogenic, amorphous and ThT negative 6 .

Correlation of the in vitro and in vivo results.
To verify the hypothesis that increased efficiency in Aβ oligomer elimination of an agent correlates with in vivo potency of the agent, therapeutic effects of D3 and D3D3 were investigated in the transgenic mouse model TBA2.1, which is expressing human N-terminally truncated and pyroglutamated Aβ 15 . Thus, the analysis of protein levels in brains of TBA2.1 mice showed substantial amounts of both Aβ and pE3-Aβ 15 . Three groups of TBA2.1 mice were treated either with phosphate buffered saline, D3 or D3D3 (5.1 ± 0.1 mg per kg body weight per day, i.p., for 4 weeks using minipumps). Their phenotypes before and after the treatment were assessed using SHIRPA tests 16 . Importantly, D3D3-treated mice did not show significant worsening of their phenotypes in the SHIRPA test, whereas the phenotypes of D3-treated mice and saline-treated mice worsened significantly, during the four weeks duration of the experiment (Fig. 4b). Thus, D3D3 was more efficient than D3, exactly as predicted from the Aβ -QIAD assay results of both compounds. Despite this nice correlation, the observed in vivo effects of D3 and D3D3 treatment are not necessarily exclusively due to the Aβ oligomer elimination activity of both compounds. One may well think of other potential activities, e.g. interference with described Aβ oligomer receptors like cellular prion protein [17][18][19] , N-methyl-D-aspartate receptor or paired immunoglobulin-like receptor B 20 , overall inflammation reduction 9 , or interference with the postulated Aβ oligomer induced cell membrane permeation 1,21,22 .
After having compared efficacies of D3 and D3D3 in the TBA2.1 model, we wanted to further verify the in vivo efficacy of D3D3, and thus, treated the transgenic AD mouse model Tg-SwDI 23

Further compounds tested by QIAD.
To demonstrate the usefulness of the QIAD assay, we applied it to a number of agents, which were previously reported to influence Aβ (1-42) assembly distribution and/ or did show efficacy in cognition improvement of transgenic AD animals, like homotaurine (Alzhemed), scyllo-inositol, and epigallocatechin gallate (EGCG). EGCG clearly converted Aβ (1-42) monomers into larger particles ( Fig. 5a and Supplementary Figure 7), exactly as had been reported 24 . Homotaurine and scyllo-inositol did not have any significant effect on the Aβ (1-42) aggregate size distribution even at 2 mM concentrations (Fig. 5a). This is in agreement with recent work reporting that scyllo-inositol does not have any effect on Aβ toxicity and Aβ oligomer formation 25 . Also, for homotaurine negative results on Aβ aggregation were reported and discussed in relation to failed clinical trials 26 .
Recently, 6-methoxy-2-(4-dimethylaminostyryl) benzofuran (KMS88009) (Supplementary Figure 8) was reported to reduce oligomeric Aβ species and to reverse cognitive deficits in AD transgenic mice 27,28 . Because KMS88009 was not water soluble, but needed DMSO and Tween20 for solvation, it was necessary to adapt the QIAD assay conditions to contain 1.6% DMSO and 0.2% Tween20 in the solvent (final concentrations). Figure 5b shows the Aβ (1-42) aggregate size distributions with and without KMS88009. This clearly shows that QIAD assay conditions can be adapted to experimental needs, easily. Similar to EGCG, KMS88009 reduces the amount of Aβ (1-42) monomers in DGC fractions 1 and 2 in favour of oligomer sized particles in fractions 5 and 6 (Fig. 5b). This, however, does not necessarily mean that KMS88009 generates toxic Aβ (1-42) oligomers. The increase of Aβ amounts in fractions 4 to 6 is not the opposite of the reduction of Aβ amounts in fraction 4 to 6, because the newly generated particles are most likely not consisting of pure Aβ (1-42), but rather are co-assemblies of Aβ (1-42) and KMS88009.
Another example compound, ZAβ 3W, which is an Affibody protein that prevents Aβ toxicity by sequestration of monomeric Aβ 29,30 , induced reduction of Aβ (1-42) content in fraction 1 and an increase  Note that the assay with KMS88009 contained 1.6% DMSO and 0.2% Tween20 (final concentrations). Therefore, the control without agent was done in presence of 1.6% DMSO and 0.2% Tween20 (red bars in (b)).
Scientific RepoRts | 5:13222 | DOi: 10.1038/srep13222 content in DGC fractions by RP-HPLC allows quantitative comparison between various drug candidates. Although Aβ oligomer elimination is not necessarily the only mechanism of compounds that show in vivo efficacy in AD animal models, Aβ oligomer elimination is -from the current point of view -the most promising option for causative AD treatment and therapy. The observed relation between Aβ oligomer elimination of D3 and D3D3 and their in vivo results strengthens the role of Aβ oligomers in AD pathology and the role of D3 and D3D3 as potential causative agents for AD treatment. Furthermore, the assay will help to avoid testing of ineffective compounds in animals and time consuming in vivo assays or even clinical trials. The QIAD assay is adjustable for any aggregation-prone peptide or protein. 4. All the fractions were analyzed with respect to their Aβ (1-42) content by RP-HPLC (see below for a detailed description).

Methods
As a control for each density centrifugation run, a sample without the added agent was used. Statistical evaluation of experimental results from eleven independent control density gradient centrifugation analyses demonstrated the good reproducibility of the procedure (Fig. 4a, black bars).
Please note: All ligands (inhibitors), but ZAβ 3W, used in this work are small molecules when compared with Aβ (1-42). Therefore, even 1:1 stochiometric complexes will not increase the molecular weight of Aβ (1-42) oligomers enough to be detected as significant difference in the density gradient centrifugation.
Analytical RP-HPLC. Quantifications of Aβ (1-42) present in DGC fractions were performed by RP-HPLC on a Zorbax SB-300 C8 column (5 μ m, 4.8 × 250 mm, Agilent, Waldbronn, Germany) connected to an Agilent 1260 Infinity system. Denaturation of Aβ (1-42) assemblies and separation of Aβ (1-42) from other compounds, especially from the density gradient forming iodixanol, was achieved using 30% (v/v) acetonitrile, 0.1% (v/v) trifluoroacetic acid in H 2 O as the mobile phase, an elevated column temperature of 80 °C and a flow rate of 1 ml/min. Applied sample volumes were 20 μ l. Eluting substances were detected by their UV absorbance at 215 nm. Data recording and peak area integration was achieved by the program package ChemStation (Agilent, Waldbronn, Germany). Calibration of the column was achieved with Aβ (1-42) solutions of known concentrations (0 to 25 μ M Aβ (1-42)) and the resulting linear equation from a plot of peak area vs. Aβ (1-42) concentration allowed the calculation of molar Aβ (1-42) concentrations (from corresponding peak areas) (Supplementary Figure 9).
The quantitative analysis of Aβ (1-42) amounts in every fraction allows us to determine the recovery rate R: In the case the calculated recovery rates were higher than 100%, the Aβ (1-42) contents of all fractions were divided by the effective recovery rate. Limiting for detection of Aβ (1-42) was the sensitivity of the UV detector connected to the HPLC. In our case the sensitivity limit was 20 nM Aβ . In principle, the sensitivity can easily be increased for example by coupling of RP-HPLC with mass spectrometry, but there was no need to do that for the experiments described in the manuscript. Thus, the QIAD could be easily performed with nano-or even picomolar Aβ concentrations.

Animals.
For study #1 23 four months old female tg/tg TBA2.1 (insertion of precursor of N-terminally modified pyroglutamate-Aβ in C57BL/6 x DBA1 background) and for study #2 19 four months old female Tg-SwDI mice (human APP with Swedish K670N/M671L, Dutch E693Q and Iowa D694N mutations on a C57BL/6 background) were used. Group sizes were decided on the basis of data from previous studies 6 . Inclusion and exclusion criteria were determined in advance. Because of the known heterogeneity of phenotypes between different sexes, only female mice with an age between 3, 5 and 4 months were included in the study. Exclusion criteria were defined as weight loss of ≥ 15%, surgical site infection and very serious general conditions in addition to the normal phenotype of Tg-SwDI or TBA2.1 mice. All animals fulfilled the inclusion criteria and no animal was excluded after enrolment in the study.
In study #2, treatment took place with the D3D3 peptide (n = 10) and control group (n = 9, saline). Animals were divided into two groups based on similar average body weight. The applied peptide amount was 1 mg/pump, 0.9 mg of unconjugated peptide and 0.1 mg peptide conjugated with fluorescein.
The Alzet minipumps were filled with the appropriate solutions and implanted into the peritoneal cavity.
Behavior and functional assessment. In study #1, selected tests from the primary screening of the SHIRPA test battery 16 were used to assess levels of spinocerebellar function, muscle and lower motor neuron functions, sensory function, neuropsychiatric function and autonomic function of the TBA2.1 mice. The following tests were used: abnormal body carriage, alertness, abnormal gait, startle response, loss of righting reflex, touch response, pinna reflex, cornea reflex, forelimb placing reflex, hanging behavior and pain response. An arena of 55 cm × 19.5 cm × 33 cm (L × H × W) was used for individual observation and analysis. The scoring values were from 0 (similar to wild type) to 3 (extremely changed compared to wild type). Before starting the treatment, mice were stratified into treatment groups using the aforementioned tests and two days before the end of the treatment they were tested again. Experimenters were blind to group allocation. Additionally, the motor balance and motor coordination of TBA2.1 mice were determined on the rotarod for 2 days before start and end of treatment using a published protocol 15 . In the morning of the first day mice were trained to stay on the rod for at least 60 s at 10 rpm. The three following test sessions, one in the afternoon and two at the second day, started with habituation for 30 min before the mice ran three times on the rotarod with an accelerated beam from 4 till 40 rpm in 5 min. After mice falling off the rod latency time was recorded.
During the last week of treatment in study #2, animals were tested in four different behavioral tests to assess cognition and to monitor side effects. Experimenters were not informed about group allocation. First the open field test was conducted. The maze consisted of an arena of 42 cm × 42 cm with clear Plexiglas sides (20 cm high). The animal was put into the arena and observed for 4 min, with a camera driven tracker system, Ethovision 8.5 (Noldus, Wageningen, The Netherlands). The arena was subdivided into two areas, the "open" center and the area by the wall. The system recorded the position of the animal in the arena at 5 frames/second and the data were analyzed regarding time spent in each area (center vs. wall), similarly speed and distance were recorded. For disinfection the apparatus was wiped down with chlorhexidine and 70% ethanol and allowed to airdry.
Next, the zero maze test was conducted. The maze consisted of a circular arena with a diameter of 65 cm and four areas of equal size, two without walls, and two with walls of nontransparent material. The animal was put in the arena, and observed for 4 min, with a camera driven tracking system, Ethovision 8.5. The time spent in each area (open vs. closed) was recorded, similarly speed and distance were also recorded.
Subsequently the object recognition test (ORT) was carried out in a maze consisting of a rectangular polycarbonate box, with partitions separating the box into three chambers. The partitions had openings that allowed the animal to move freely from one chamber to another. The animal was monitored by the Ethovision 8.5 tracking system 8.5. On day one two objects were placed on each side of the box. The mouse was placed in the box and allowed to move freely throughout the apparatus for a 10-minute test session. After 24 h, a new object replaced one of the "old" objects and the mouse was put in the box and allowed to move freely throughout the box over a 4-minute test session. The time spent with each object and the transitions between the objects were recorded. The apparatus was wiped down with chlorhexidine followed by ethanol and water and dried with paper towels for each mouse tested.
Finally, the mice were tested for 5 days in a Morris water maze (MWM). The maze consisted of a blue circular tank of clear water (23 ± 1 °C). The mice were placed in the water at the edge of the pool and allowed to swim in order to locate a hidden, but fixed escape platform, using extra maze cues. On day 1, the mice were placed in the pool and allowed to swim freely for 60 s or until the hidden platform was found; each animal was tested for four trials per day. A maximum swim time per trial of 60 s was allowed; if the animal did not locate the platform in that time, it was placed upon it by the experimenter and left there for 10 s. The intertrial interval was 120 s. Each start position (east, north, south and west) was used equally in a pseudo random order and the animals were always placed in the water facing the wall. The platform was placed in the middle of one of the quadrants of the pool (approximately 30 cm from the side of the pool). The mouse's task throughout the experiment was to find the platform. The animal was monitored by Ethovision 7.1.

Statistics.
Data were averaged and represented as mean ± standard error of mean (SEM). Statview, Version 5.0.1 served for all calculations, p > 0.05 was considered not significant ("n.s. ").