Molecular Mechanisms of Alzheimer’s Disease Protection by the A673T Allele of Amyloid Precursor Protein

Results: A673T reduces BACE1 processing of APP by decreasing catalytic turnover, and reduces amyloid- β (A β ) 1-42 aggregation. Conclusion: APP by reducing A β production, and also by reducing aggregation. Significance: The biochemical nature of the A673T protective mutation provides insight into AD development. ABSTRACT Pathogenic mutations in the Amyloid precursor protein (APP) gene have been described as causing early onset familial Alzheimer’s disease (AD). We recently identified a rare APP variant encoding an alanine-to-threonine substitution at residue 673 (A673T) that confers protection against development of AD (1). The A673 residue lies within the β -secretase recognition sequence, and is part of the A β peptide cleavage product (position 2 of A β ). We previously demonstrated that the A673T substitution makes APP a less favorable substrate for cleavage by BACE1. In follow-up studies, we confirm that A673T APP shows reduced cleavage by BACE1 in transfected mouse primary neurons and in isogenic human iPSC-derived neurons. Using a biochemical approach, we show that the A673T substitution modulates the catalytic turnover rate (V max ) of APP by the BACE1 enzyme, without affecting the affinity (K m ) of the APP substrate for BACE1. We also show a reduced level of A β 1-42 aggregation with A2T A β peptides, an observation not conserved in A β 1-40 peptides. When combined in a ratio of 1:9 A β 1-42 :A β 1-40 to mimic physiologically relevant mixtures, A2T retains a trend toward slowed aggregation kinetics. Microglial uptake of the mutant A β 1-42 peptides correlated with their aggregation level. Cytotoxicity of the mutant A β peptides was not dramatically altered. Taken together, our findings demonstrate that A673T, a protective allele of APP, reproducibly reduces amyloidogenic processing of APP and also mildly decreases A β aggregation. These effects could together have an additive or even synergistic impact on risk of developing AD.

enzymes involved in the processing of APP, have been found in autosomal dominant familial cases of AD (2)(3)(4). More recently, PSEN1 mutations have also been identified in some sporadic cases of late onset AD (5). Numerous disease-associated mutations in the APP gene itself have also been identified (http://www.molgen.ua.ac.be/ADmutations). Many of these mutations cluster at or near the β or γ proteolytic sites, favoring either the overproduction of total amyloid-β (Aβ) (6)(7)(8) or an increased ratio of the pro-aggregating Aβ 1-42 species relative to Aβ 1-40 (9)(10)(11)(12). In other instances, mutations within the Aβ peptide promote an increased propensity for aggregation (7,13,14). Together, these genetic findings provide strong support for the amyloid hypothesis of AD, which postulates that an imbalance in the production and clearance of Aβ initiates a cascade of amyloid accumulation, neurotoxicity and neurodegeneration (15).
Identification of genetic variants that protect from AD can be equally informative in revealing the mechanisms underlying disease biology, and could guide therapeutic approaches to block disease development or progression. Recently, we reported a low-frequency variant in the APP gene (rs63750847-A) that confers significant protection against AD (1). This variant results in an alanine-to-threonine substitution at position 673 in the APP gene (A673T), adjacent to the β cleavage site. Our primary characterization of the A673T variant demonstrated that this mutation made APP a less favorable substrate for BACE1, resulting in reduced levels of Aβ production (1). However, the A673T substitution lies within the Aβ peptide, and thus could also alter the properties of the Aβ peptide itself. Here, we examine the effects of this variant on the kinetics of APP processing by BACE1, and additionally evaluate the impact of this substitution on Aβ aggregation and biological activity, including microglial uptake and cellular toxicity. Our results provide further support for reduced β-site cleavage of the A673T APP variant, and additionally reveal modest alterations in Aβ  peptide aggregation, as assessed using synthetic peptides.

EXPERIMENTAL PROCEDURES
Cellular APP Cleavage Assays: Human APP695 cDNA's encoding WT APP, APP A673T, or APP A673V were transfected into 293T cells maintained in DMEM + 10% FBS as described previously (1). After 4 days in vitro (DIV), cells were lysed using RIPA buffer with Complete Protease Inhibitor Cocktail (Roche, Mannheim, Germany) to assay for β-CTF's. The cell supernatant was also collected to confirm reduced Aβ production. Dissociated cortical neuron cultures (plated at 2.5 X 10 4 cells/well in a 96-well plate) were prepared from E16.5 C57BL/6J mice. Prior to plating, cells were Amaxa nucleofected (Lonza, Basel, Switzerland) with cDNA's encoding the same human APP695 constructs described above, or β-galactosidase (control). After 6 DIV, cells were incubated with 50µl/well fresh cell medium for 24 hours. The medium was collected and used for Aβ ELISA measurements (see below). For both cell types, the efficiency of transfection across different constructs was compared by lysing a separate well of cells with Ambion Cell to CT (Life Technologies, Carlsbad, CA) to assess APP RNA levels by qPCR. Human cortical neurons (catalog # DDP-NC-1.0) derived from control donor iPS cells (Clone ID 01279.107), the isogenic APP-A673V iPS cells (01279.A32), and the isogenic APP-A673T iPS cells (01279.A27) were ordered from Cellular Dynamics International Inc. (Madison, WI). Human iPSC-derived neurons were plated in 8well chamber slide (BD 354688) at a density of 1 X 10 5 cells/well. The culture slides were prepared with poly-D-lysine and laminin following CDI's user's guide. Half of the conditioned media was replaced with fresh warm media every two days. To confirm APP expression, cells were fixed with 4% paraformaldehyde/ 4% sucrose, permeabilized with 0.1% Triton-X100, and blocked in 2% BSA/PBS. Cells were stained with rabbit monoclonal anti-APP clone Y188 (Abcam, Cambridge, MA) and mouse anti-MAP2 clone 5F9 (Abcam, Cambridge, MA) in 2% BSA/PBS for 2 hrs. After washing, the secondary goat anti-mouse IgG AlexaFluor®488 antibody A10680 (Life Technologies, Grand Island, NY) and goat antirabbit IgG AlexaFluor®568-conjugated antibody A11036 (Life Technologies, Grand Island, NY) were applied for 30 min at 1:200 in 2% BSA/PBS. Media was harvested after 4 DIV for sAPP analysis, and after 14 DIV for Aβ measurements. Experiments were performed at least three times, with technical triplicates in each experiment.
Aβ and sAPP Immunoassays: Aβ x-40 and Aβ x-42 peptides were measured from transfected HEK 293 and primary neuron cell supernatants by sandwich ELISAs as described previously (1). Briefly, rabbit polyclonal antibody specific for the C terminus of Aβ 40 or Aβ 42 (Millipore, Bedford, MA) was coated onto plates, and biotinylated anti-Aβ monoclonal antibody 6E10 (Covance, Dedham, MA) was used for detection. The assays had lower limit of quantification values of 6.24 pg/ml for Aβ x-40 and 18.76 pg/ml for Aβ x-42 . For iPSCderived neurons, Aβ x-40 and Aβ x-42 peptides were measured from cell supernatants using a sandwich electro-chemiluminescence immunoassay (Meso Scale Discovery Cat #K15148E, Gaithersburg, MD, USA). Cell supernatants from all cell types were also assayed for sAPP cleavage products by using a sAPPα/sAPPβ multiplex kit (Meso Scale Discovery Cat #K15120E, Gaithersburg, MD). . Peptides in the reactions were then further diluted to 300 nM prior to the addition of equal volume of detection reagent (10 nM anti-Aβ 6E10-Eu, 75 nM SA-d2, 800 mM KF, 200 mM TRIS pH 8, 0.1% CHAPS, 0.1% BSA) and the mixture was incubated at ambient temperature for 1 hour. The plates were measured in an EnVision (Perkin Elmer) reader and TR-FRET ratio with 320 nm excitation, 615 nm donor emission, and 665 nm acceptor emission was calculated as (665em / 615em) * Z where Z = 5000.

TR-FRET cleavage of WT and
The reaction progress curves were determined by subtracting background signal at time 0 for each reaction and reaction rates (min -1 ) were calculated from linear fits using GraphPad Prism (Sunnyvale, CA).
FRET K m determination with short APP (668-675) peptides: Three FRET peptide substrates, containing WT APP amino acid sequence 668-675 (E(EDANS)EVKMDAEFK(Dabcyl)NH 2 ), A673T (E(EDANS)EVKMDTEFK(Dabcyl) NH 2 ) or A673V (E(EDANS)EVKMDVEFK(Dabcyl)NH 2 ) were custom synthesized by American Peptide Company (Sunnyvale, CA). Each peptide was conjugated with a fluorescent donor molecule EDANS at N-terminus and a quencher molecule Dabcyl at the C-terminus. The peptide substrates were titrated up to 100 µM in reaction buffer (50 mM Sodium Acetate pH 4.4, 0.1% CHAPS, 0.1% BSA) and the proteolytic reaction was carried out using 50 nM BACE1 at ambient temperature. The enzyme reactions were quenched at various time points up to 48 hours with the addition of equal volume stop buffer (200 mM TRIS pH 8, 0.1% CHAPS, 0.1% BSA). Fluorescence intensity at 490 nm with excitation at 340 nm was measured using Tecan M1000 Fluorescence Reader (Tecan, Mannedorf, Switzerland) and the total fluorescence intensity gain was determined by subtracting background fluorescence at time 0 from the fluorescence intensity of each reaction mixture at different time points. Reaction rates (min -1 ) were calculated from linear fits of the reaction progress curves and the Michaelis-Menton equation was applied to determine K m and V max using GraphPad Prism (Sunnyvale, CA).
IC50 determination of APP peptide A673 variants in FRET cleavage assay: FRET peptide substrate with the Swedish mutation (Rh-EVNLDAEFK-Quencher) was obtained from Invitrogen. Enzyme reactions were carried out with 50 nM BACE1, 150 nM FRET peptide substrate with Swedish mutation, and up to 100 µM biotinylated peptide in reaction buffer (50 mM Sodium Acetate pH 4.4, 0.1% CHAPS, 0.1% BSA) at ambient temperature.
Fluorescence intensity (530nm excitation, 590nm emission) was continuously monitored for 85 minutes on a Tecan M1000 Fluorescence Reader. Initial background signal at each peptide concentration was subtracted from each reaction time. Reaction rates (min -1 ) were calculated from linear fits and peptide IC 50 s were generated using a 4-parameter nonlinear fit in GraphPad Prism (Sunnyvale, CA).

Aggregation Kinetics as Measured by Thioflavin-T Fluorescence:
The synthetic peptides (AnaSpec, Fremont, CA) were disaggregated by dissolution in 1,1,1,3,3,3-Hexafluoro-2-propanol (HFIP), followed by evaporation under a stream of nitrogen. Immediately before analysis, the peptide film was thoroughly dissolved in 10 mM NaOH, neutralized in an equal volume of 100 mM sodium phosphate buffer, pH 7.4, and filtered with a 0.2 µm syringe-filter. UV absorbance of the solution at 280 nm was measured and the peptide concentration was determined using the respective theoretical extinction coefficient and adjusted to 1.0 mg/mL with 5mM NaOH, 50 mM sodium phosphate buffer, pH 7.4. The redissolved and adjusted peptide solution was transferred into a well of the 96-well plate containing an equal volume of 56 µM thioflavin-T (ThT) in 2 mM NaN 3 , 50 mM phosphate buffer, pH 7.4. All data were recorded with a Tecan Safire II reader and black, clear-bottom, sealed 96-well plates (Corning) in bottom-reading mode at 37°C. Kinetic reads were taken every 15 minutes (Aβ 1-40 ) or every 10 minutes (Aβ 1-42 and Aβ 1-42 /Aβ 1-40 mixtures). ThT was excited at 450 nm and the fluorescence emission signal was read at 490 nm. For each experiment, traces of the mean of three technical replicates were plotted using GraphPad Prism 5 and error bars of the final point indicate standard deviation.

Microglial
Aβ Uptake Assay and Immunocytochemistry: Primary microglia were prepared from P2 C57BL/6J mice at a density of 6x10 4 cells per well in 8-well poly-D-lysine coated chamber slides, and the Aβ uptake assay was performed 48 hours later. Aβ 1-40 and Aβ 1-42 peptides were custom synthesized by rPeptide (Bogart, GA) and prepared as oligomers according to Stine et al. (16) with the following modifications: peptides were resuspended in 1:10 v/v 1% NH 4 OH:HFIP to 20 µM and lyophilized overnight. Lyophilized peptides were resuspended in anhydrous dimethyl sulfoxide (DMSO) D2650 (Sigma, St. Louis, MO) to 5 mM and immediately diluted to 100 µM in ice-cold cell culture medium (phenol red-free Ham's F-12), vortexed for 30 sec, and incubated at 4°C for 24 hours. Peptides were aliquoted and stored at -80°C. 1 µM of Aβ 1-40 or Aβ 1-42 peptides were added to primary microglia cultures. After 90 min, microglia were washed 3 times with cold PBS, immediately fixed in 4% paraformaldehyde/ 4% sucrose. Cells were then blocked in 2% BSA/PBS and permeabilized in 0.1% Triton X-100 93426-250ml (Sigma, St. Louis, MO) for 20 min and stained with mouse anti-Aβ SIG-39320 (Covance, Dedham, MA) at 1:500 and rabbit anti-Iba1 019-19741 (Wako, Osaka, Japan) at 1:200 in 2% BSA/PBS for 2 hrs. After washing, the secondary goat anti-mouse IgG AlexaFluor®488 antibody A10680 (Life Technologies, Grand Island, NY) and goat antirabbit IgG AlexaFluor®568-conjugated antibody A11036 (Life Technologies, Grand Island, NY) were applied for 30 min at 1:200 in 2% BSA/PBS. Immunostained cells were analyzed on a Zeiss 200 M confocal microscope (Thornwood, NY) and processed with Photoshop CS5. To quantify intracellular Aβ uptake by microglia, the area corresponding to intracellular Aβ was delineated using Iba1 staining and the raw integrated signal intensity was determined per microglia. The quantification was performed using ImageJ software and values are represented in arbitrary units. 10-20 cells from each condition were quantified in each of three independent experiments, and data from the three experiments was then averaged.
Cellular Toxicity Assays: Cortical neurons were isolated from E16.5 C57BL/6J mice and plated at a density of 3.5x10 4 cells per well in 96-well poly-D-lysine coated plates (BD Biosciences, San Jose, CA). At 6 DIV, cells were treated with freshly prepared Aβ 1-40 or Aβ 1-42 WT, A2T, or A2V synthetic peptides (rPeptide, Bogart, GA) or vehicle. The peptides were initially resuspended in NH 4 OH:HFIP and lyophilized as described for microglia experiments. Lyophilized peptides were then dissolved in DMSO to generate a 5 mM stock, and immediately diluted in culture medium and incubated with cells. After 72 hours, cell viability was assessed using the Cell Titer Glo assay, following manufacturer's instructions (Promega, Madison, WI). Experiments were performed three times, with technical triplicates per construct in each experiment

The A673T variant reduces BACE1 cleavage of APP in primary neurons.
We previously reported that the A673T variant of APP decreases production of the BACE1 cleavage product sAPPβ and both of the amyloidogenic peptides Aβ x-40 and Aβ x-42 by ~40% when overexpressed in 293 HEK cells (1). To extend these findings, we looked at production of another APP BACE1 cleavage product, the β−carboxyterminal fragment (β−CTF), C99. We transfected 293 HEK cells with APP WT or the protective A673T variant. For comparison, we also examined two pathogenic APP variants, A673V and K670N/M671L (Swedish mutation), which have both been shown to increase Aβ and β−CTF production. We first examined β−CTF production by Western blot analysis of cell lysates using two antibodies, A8717 which detects both α-and β−CTF and 6E10 which detects only β−CTF (Fig.  1A). CTF's were readily detected in APPtransfected cells and absent in control GFPtransfected cells. As expected, higher levels of β−CTF were seen in cells expressing the pathogenic APP variants A673V or K670N/M671L; conversely, APP A673T reduced β-CTF levels. To quantify these changes, we used a β−CTF ELISA to analyze these cell lysates. A673T reduced β-CTF levels to ~25% that of WT APP levels (Fig. 1B). This reduction is consistent with the effects of A673T on sAPPβ and Aβ production as we previously reported (1).
Our findings thus far have focused on APP processing in the 293 HEK cell line. To confirm that A673T had similar effects on APP processing by BACE1 in neurons, we performed experiments in cultured primary mouse cortical neurons. Cultured neurons were transfected with cDNA encoding human APP WT, the protective A673T variant, the pathogenic APP A673V variant, or a control vector expressing β-gal. All constructs expressed APP at similar levels, as confirmed by qPCR (data not shown). We assessed the formation of APP cleavage products Aβ x-40 , Aβ x-42 , sAPPβ, and sAPPα in the media of transfected cells. Our assays were designed to specifically detect cleavage products from transfected human APP constructs and did not detect endogenous mouse APP cleavage products. Similar to what was observed in 293 HEK cells, we found that the A673T APP variant generated less Aβ x-40 and Aβ x-42 peptides as compared to WT APP (Fig. 1C, D). Aβ x-40 was reduced by 38 (±25) % and Aβ x-42 by 35 (±10) %. APP A673V increased Aβ x-40 and Aβ x-42 levels by ~2.5-fold. Immunoassay analysis for sAPPβ and sAPPα in cell supernatants (Fig.  1E), showed that the A673T APP variant reduced the ratio of sAPPβ relative to sAPPα when compared to WT APP by 49 (±5) %. Conversely, the pathogenic APP A673V variant robustly increased the sAPPβ/sAPPα ratio.
Overexpression of APP by transfection may result in non-physiological processing if overexpressed substrate overwhelms endogenous proteases. Thus we next set out to investigate APP processing of these variants when expressed at normal endogenous cellular levels. To do this, we used human induced pluripotent stem cell (iPSC) lines that have been engineered via TALEN-mediated SNP alteration to carry A673T or A673V specific changes in APP (17). The isogenic cell lines were differentiated into cortical neurons. The differentiated cells from the three isogenic lines appeared morphologically similar to each other, and uniformly expressed APP (Fig. 2A). Similar to transfected mouse neurons, we observed that the A673T APP iPSC human neurons generated less Aβ x-40 and Aβ x-42 peptides as compared to WT APP (Fig. 2B, C). Aβ x-40 was reduced by 33 (±7) % and Aβ x-42 by 36 (±12) %. In contrast, APP A673V increased both Aβ x-40 and Aβ x-42 levels by ~two-fold. The reduction in β-secretase cleavage products was confirmed by measuring the ratio of sAPPβ relative to sAPPα by immunoassay (Fig.  2D). The A673T APP variant reduced the sAPPβ/sAPPα ratio by 32 (±18) % compared to WT APP. Conversely, the pathogenic APP A673V variant increased the sAPPβ/sAPPα ratio ~fourfold.
The consistency of our results across transfected 293 HEK cells and primary neurons, and isogenic human iPSC-derived neurons confirms the strong influence of the residue at position 673 on BACE1 cleavage. The reproducible decreases in βsecretase cleavage of APP across these different cell types suggests that similar alterations in APP processing likely occur in neurons in the human brain.

A673T decreases the V max of APP processing by BACE1.
Previously, we demonstrated that the A673T mutation decreased catalytic cleavage of an APP substrate by BACE1 in a reconstituted in vitro enzyme assay (1). Here we investigated the mechanism by which the mutation at this residue modulated BACE1 enzymatic activity. First, utilizing TR-FRET detection against long (27 amino acid) APP peptide substrates, we demonstrated that the A673T variant displayed ~50% reduced cleavage activity relative to the WT substrate (Fig. 3A), consistent with our previous results determined by mass spectrometric detection. In comparison, the A673V mutation appeared to be a better substrate for BACE1 with an estimated 3-fold increase in cleavage rate relative to WT APP peptides at 5 µM substrate concentration. When the mutations at A673 were combined with the Swedish APP mutation (K670N/M671L, Swe), the same differential modulations of BACE1 activity by threonine and valine were observed (Fig. 3B), however with much more rapid kinetics as a result of the Swe mutations.
Utilizing a fluorogenic FRET substrate that spanned from P4 to P5' of the APP BACE1 cleavage sequence, the kinetic parameters of WT APP were compared with the A673T and A673V mutations.
The results showed that A673T decreased and A673V increased the V max , but neither altered the K m , suggesting that mutations at this residue have an impact on the catalytic turnover of the substrate instead of substrate binding affinity (Fig. 3C).
To confirm that the mutations at A673 do not modulate the binding affinity of peptide substrates to BACE1, we performed a substrate competition assay. We used the FRET peptide substrate containing the Swedish mutation (Swe) as a detection method for BACE1 activity, and titrated non-Swedish long peptides (WT, A673T and A673V) into the BACE1 reaction. At a reaction time of 1-2 hours, while the Swe FRET peptide substrate is efficiently cleaved by BACE1, the catalytic turnover of the unlabeled non-Swedish peptides by BACE1 is negligible. In other words, the non-Swedish peptides act as competitive binders to inhibit cleavage of the Swe FRET substrate in this assay setup, and their titration should yield IC 50 curves that reflect their respective binding affinities. The IC 50 values of WT, A673T and A673V peptides were comparable to each other (ranging between 26 to 33 µM), further supporting the hypothesis that the mutations at residue A673 of APP do not alter their binding affinity to the BACE1 active site (Fig.  3D).

Aggregation level of A2T Aβ variant peptides is reduced in the context of Aβ 1-42 , but not Aβ 1-40 .
Our results indicate that the APP A673T variant is a less favorable substrate for BACE1 cleavage. However, the A673T variant also results in a single amino acid substitution (A2T) in Aβ peptides, and thus could alter the properties of the Aβ peptide itself. Therefore, we compared the aggregation kinetics of synthetic WT, A2T and A2V Aβ peptides, as both Aβ 1-40 and Aβ 1-42 species, by thioflavin-T (ThT) incorporation and fluorescence.
As shown in Fig. 4, Aβ 1-40 peptides generally aggregate very slowly relative to Aβ 1-42 . However, the A2V substitution promotes a significant increase in both the aggregation kinetics and the extent of aggregation in Aβ 1-40 (Fig. 4A). This result is similar to observations reported by Di Fede et at. (7). By contrast, aggregation kinetics for WT and A2T Aβ 1-40 were indistinguishable from each other. When A2V Aβ 1-40 peptides are combined with WT Aβ   In vivo, Aβ peptides exist as a heterogeneous mixture. As an approximate representation of this heterogeneity, we looked at aggregation kinetics of a mixture of Aβ 1-42 and Aβ 1-40 in a 1:9 ratio, representative of what has been observed in human CSF (18). When mixed together in this manner, we observed aggregation kinetics intermediate between those with pure Aβ 1-40 or Aβ 1-42 peptides (Fig. 4C). The A2V peptide mixture again showed the most rapid kinetics. The A2T mixture showed slower kinetics than the WT mixed peptides in two of three experiments, and similar kinetics in the third. Mixing A2T with WT Aβ in this context results in an aggregation profile similar to WT Aβ alone.
Taken together, these results suggest that substitutions at position 2 in Aβ peptides can impact peptide aggregation properties. The most dramatic effect is with A2V substitution in the context of Aβ  . This mutation appears to convert Aβ 1-40 into a much more aggregation-prone molecule with Aβ 1-42 -like properties. However, A2V did not alter aggregation in the context of Aβ 1-42 , likely because this longer peptide is already very aggregation-prone. In contrast, A2T does not appear to alter Aβ 1-40 aggregation, but does reduce the aggregation level of Aβ 1-42 . When Aβ 1-42 and Aβ 1-40 are mixed together in a physiologically relevant ratio of 1:9, the A2T variant overall has slightly slower aggregation kinetics compared to WT. However, the resulting aggregation of A2T remains substantial, making it difficult to predict the impact of this mutation on pathological amyloid accumulation, particularly in heterozygous carriers.

Microglial uptake of A2T Aβ 1-42 synthetic peptides is reduced in comparison to WT Aβ peptides.
Many recent Alzheimer's genetic hits, including TREM2 and CD33, highlight the importance of microglial biology in Alzheimer's disease (19)(20)(21)(22). Microglial uptake of soluble Aβ is believed to be an important clearance mechanism in the brain (23). We wondered if the different Aβ variants we have examined might be differentially taken up by microglia. We used enriched primary microglial cultures to assess uptake of synthetic WT, A2T and A2V Aβ peptides. Again, we assessed both Aβ 1-40 and Aβ 1-42 peptides prepared as oligomers. Microglia were incubated with 1 µM Aβ oligomers for 90 minutes, after which cells were fixed, and internalized Aβ was visualized by immunocytochemistry.
As shown in Fig. 5A, Aβ could be visualized inside microglia under all conditions. However, when quantified (Fig. 5B), it was clear that WT Aβ 1-42 was internalized to a much greater extent than WT Aβ 1-40 . This likely reflects the higher level of aggregation with WT Aβ 1-42 compared to WT Aβ 1-40 , since aggregation stimulates microglial internalization of extracellular debris (including Aβ). A similar trend is seen when comparing the A2T peptides (Fig. 5A, middle panels); low levels of A2T Aβ 1-40 are internalized, while significantly higher levels of A2T Aβ 1-42 are taken up. However, the average Aβ fluorescence in cells treated with A2T Aβ 1-42 is approximately half that of cells treated with WT Aβ  . This correlates with the reduced overall aggregation of A2T Aβ 1-42 relative to WT (Fig. 4B). Interestingly the A2V peptides were internalized at high levels both as Aβ 1-40 and Aβ 1-42 species. This likely reflects the higher aggregation state of A2V Aβ 1-40 (see Fig. 4A), thereby resulting in behavior more like an Aβ 1-42 peptide.
In conclusion, our results suggest that microglial internalization of the different Aβ variants we assayed is highly correlated with the aggregation properties of that variant. There do not appear to be any further inherent differences between the ability of microglia to internalize the Aβ variants.

Cellular toxicity of A2T Aβ synthetic peptides does not differ from that of WT Aβ peptides.
Aβ peptides are believed to be toxic to neurons, and at high concentrations can induce cell death in vitro. To investigate the impact of the A2T substitution on Aβ toxicity, we treated neuronal cultures with synthetic WT, A2T, and A2V Aβ peptides, both in the context of Aβ 1-40 and Aβ  . Mixed primary cortical cultures were incubated with the different synthetic peptides ranging in concentration from 0.1-10 µM for 3 days, at which time cellular toxicity was assessed using a luminescent viability assay. For both Aβ 1-40 and Aβ 1-42 , only slight toxicity was observed after treatment with the WT and A2T peptides at concentrations up to 3 µM (Fig. 6). Notably, A2V peptides showed no toxicity at these concentrations. At 10 µM, all Aβ 1-40 and Aβ 1-42 peptides were somewhat toxic, reducing cell viability by ~20-40%, regardeless of the nature of the variant at position 2 of Aβ.
Comparing across the full concentration range, toxicity from A2T peptides was not significantly different from WT peptides. However, A2V peptides, both Aβ 1-40 and Aβ 1-42 , were overall less toxic than WT peptides across the concentration range, and this difference was significant (p < 0.05 for Aβ 1-40 and p < 0.001 for Aβ 1-42 , two-way ANOVA). However, the effect size was small. These results seem somewhat counterintuitive, as the pathogenic peptide might be expected to be more toxic. However, it has been proposed that low molecular weight oligomers are cytotoxic to cultured neurons. It is possible that fewer of these smaller oligomers are present in the rapidly aggregating A2V peptide preparations. Furthermore, an experimental system such as ours that requires application of high concentrations of synthetic peptides to cultured neurons in order to elicit any toxicity may not be a good model of the disease state (24). Nonetheless, these results suggest that the Aβ peptides produced by the A673T APP variant are no less or no more toxic than WT Aβ.

DISCUSSION
The results from our biochemical and cellular analyses of the A673T APP variant support the conclusion that a likely mechanism for protection is via reduced APP cleavage, leading to decreased Aβ production over the lifetime of an A673T carrier. We demonstrate that the reduction in amyloidogenic processing of APP can be replicated across transfected cell lines and primary neurons, as well as in isogenic human iPSCderived neurons that express APP at endogenous levels. Mechanistically, we show that A673T reduces the catalytic turnover of APP by BACE1, while an alternate residue, A673V, at the same site increases the catalytic turnover rate by BACE1. Neither residue alters the affinity of APP for BACE1. Therefore, the residue at the A673 site is a critical determinant of amyloidogenic processing of APP at both the enzyme and cellular level.
An alternative model, consistent with reduction in APP processing, is that protection is conferred by reduced production of β-CTFs and/or soluble APPβ. We observe that both β-CTFs and soluble APPβ are reduced in A673T expressing cells ( Fig.  1 and 2). This possibility is supported by reports that suggest a role for these APP products in mediating neuronal toxicity (25)(26)(27). Nevertheless, owing to the observation that mutations in APP center on the production and/or aggregation potential of Aβ (for example, the Dutch APP mutation E693Q causes Aβ to aggregate on blood vessels causing hereditary cerebral hemorrhage with amyloidosis (28)(29)(30)), combined with the paucity of mutations in APP directly in the cytoplasmic (β-CTFs) or N-terminus (soluble APPβ) region of APP that alter the risk for Alzheimer's disease, the most plausible hypothesis is that Aβ mediates disease.
Of note, we also observed a consistent reduction in aggregation level of A2T peptides compared to WT peptides for Aβ 1-42 , but not for Aβ  . When comparing a 1:9 mixture of Aβ 1-42 :Aβ 1-40 , the A2T peptides display slightly slower aggregation kinetics compared to WT peptides. An equimolar mixture of WT and A2T peptides prepared in a 1:9 ratio of Aβ 1-42 :Aβ 1-40 behaved identically to WT peptides (Fig. 4C). This in vitro experiment is our closest approximation of the Aβ species that might be present in an A673T heterozygous individual. As such, the reduced aggregation of the A2T Aβ 1-42 peptides likely translates to a small impact on amyloid deposition in vivo. Nevertheless, this may contribute in part to genetic protection by the A673T variant, in combination with the substantial reduction in Aβ generation.
Interestingly, the results of our microglial uptake assay correlated directly with the aggregation results. Since microglia are known to internalize aggregated proteins, this outcome is not surprising. These data confirm our findings that A2V increases aggregation in the context of Aβ 1-40 and A2T reduces aggregation in the context of Aβ 1-42 , functionally leading to increased and decreased microglial internalization, respectively. However, we failed to detect any other fundamental differences in how microglia recognize and engulf the variant peptides. Thus, microglial uptake of these different variants in vivo is likely to be driven purely by Aβ levels and extent of aggregation.
Our cell toxicity data suggests that the A673T variant does not confer protection against AD by generating an Aβ peptide that is inherently less toxic than WT Aβ. However, toxicity assays involving Aβ are highly artificial and must be interpreted with caution (24). As an illustration of the potential inconsistencies of these types of experiments, previous studies investigating the toxicity of the A673V variant suggested that A2V showed increased cellular toxicity (7); in contrast, we show that A2V may be slightly less toxic in vitro. Rather than draw spurious conclusions from these results, we simply suggest caution in overinterpreting Aβ-induced cellular toxicity results.
We previously reported that the A673T variant is equally protective in ApoE4 carriers and noncarriers (1); thus, the relative protective effect of the A673T variant is substantially more robust in ApoE4 carriers. Taking into consideration that ApoE4 carriers are considerably more likely to accumulate amyloid, and thus generate amyloid plaques and develop AD (31)(32)(33), the findings reported herein further support the hypothesis that ApoE4 is mediating its effects via Aβ. Specifically, the protective A673T APP variant that reduces amyloid production and may also reduce Aβ aggregation results in a determining factor for risk of developing AD, negating any ApoE genotype effect.
We investigated several mechanisms by which A673T may confer protection against AD. The most compelling mechanism supported by our studies is that this variant reduces BACE1mediated processing of APP. We observed this result reproducibly across multiple cell models. In addition, we found that A2T peptides are less prone to aggregation in the context of Aβ 1-42 , but not in the context of Aβ 1-40 . In terms of translating these findings to humans, our results predict that a heterozygote A673T carrier will have a significant, but relatively moderate, reduction in Aβ production (~20%). The rare nature of this mutation has made it difficult to examine endogenous Aβ levels in carriers, although attempts to do so are ongoing. In heterozygote A673T carriers, we predict that Aβ aggregation and amyloid formation will be similar or slightly reduced when compared to WT individuals. However, any resulting reduction in Aβ aggregation may work additively, or even synergistically, with reduced Aβ levels to lessen AD risk in carriers. While it is unclear how early Aβ reduction is needed to provide protection from AD, these mechanistic studies of human APP variants further validate efforts to reduce Aβ production and aggregation as therapeutic approaches to treat Alzheimer's disease.