Wild-type sTREM2 blocks Aβ aggregation and neurotoxicity, while the Alzheimer’s R47H mutant does the opposite

Missense mutations (e.g. R47H) of the microglial receptor TREM2 increase risk of Alzheimer’s disease (AD), and the soluble ectodomain of wild-type TREM2 (sTREM2) appears to protect in vivo, but the underlying mechanisms are unclear. We show that Aβ oligomers bind to TREM2, inducing shedding of sTREM2. Wild-type sTREM2 inhibits Aβ oligomerization, fibrillization and neurotoxicity, and disaggregates preformed Aβ oligomers and protofibrils. In contrast, the R47H AD-risk variant of sTREM2 is less able to bind and disaggregate oligomeric Aβ, but rather promotes Aβ protofibril formation and neurotoxicity. Thus, in addition to mediating phagocytosis, wild-type TREM2 may protect against amyloid pathology by Aβ-induced release of sTREM2 that blocks Aβ aggregation and neurotoxicity; while R47H sTREM2 promotes Aβ aggregation into neurotoxic forms, which may explain why the R47H variant gene increases AD risk several fold.


Introduction
A prominent neuropathological feature of Alzheimer's disease (AD) is the presence of extracellular deposits of the amyloid β-peptide in amyloid plaques, surrounded by activated microglia [1][2][3] . The importance of this microglial response to the pathogenesis of AD has been highlighted by the recent discovery of sequence variants in multiple genes expressed in microglia that alter risk for AD. Prominent amongst these microglial AD-risk genes is the "triggering receptor expressed on myeloid cells 2" (TREM2) [4][5][6] , of which there are several missense mutations in the ectodomain, including R47H, associated with increased risk for AD [4][5][6] .
The biological mechanisms underlying this association remain unclear.
Full length TREM2 is expressed on the plasma membrane of microglia, where it can be cleaved by one or more metalloproteases to produce i) a membrane-bound C-terminal fragment (CTF); and ii) an N-terminal fragment consisting of the soluble ectodomain of TREM2 (sTREM2), which is released into the extracellular space [7][8][9] . sTREM2 has been thought of as a non-functional, degradation product of TREM2, and used as a biomarker of microglial activation [10][11][12] . However, several recent observations suggest the possibility that sTREM2 per se may play a role protecting against AD by interacting with Ab. First, oligomeric Ab binds to TREM2 and to sTREM2-Fc fusion protein [13][14][15] , suggesting that sTREM2 might bind Ab and potentially affect its aggregation state. Second, injection or expression of sTREM2 in the hippocampus of 5xFAD mice reduces both amyloid plaque load and memory deficits 16 , indicating that sTREM2 inhibits amyloid pathology somehow. Third, in transgenic mouse models overproducing Ab, the knockout of TREM2 expression accelerates amyloid plaque seeding 17 and the plaques have increased protofibrillar halos and hotspots 3,16,18 , indicating that either TREM2 or sTREM2 inhibit plaque formation. Fourth, in the earliest pre-symptomatic stages of AD, at the time of Ab deposition, sTREM2 levels in cerebrospinal fluid (CSF) are lower than in healthy controls 10,19 , consistent with sTREM2 being an endogenous inhibitor of Ab deposition.
However, the CSF Aβ levels rise in the early symptomatic stages of AD, and then decline again at later stages of AD 11,12 . Fifth, mild cognitive impairment (MCI) and AD patients with higher sTREM2 levels in CSF have slower brain atrophy, cognitive decline and clinical decline 20,21 , consistent with sTREM2 inhibiting AD progression. Sixth, healthy controls and MCI patients with higher sTREM2 levels in CSF have slower progression of amyloid and tau deposition 22 , consistent with sTREM2 inhibiting Ab aggregation and subsequent tau pathology.
All of these in vivo findings are compatible with the hypothesis that sTREM2 might protect against AD, potentially by impacting Ab aggregation. To explore this hypothesis and the mechanisms involved, we investigated the interaction of Aβ with sTREM2 in vitro. We report that soluble Aβ oligomers bind TREM2 receptor on microglia and induce shedding of sTREM2. Next, we show that sTREM2 can bind and disaggregate Aβ oligomers, block Aβ fibrillization and reduce Aβ neurotoxicity. These activities are attenuated in the R47H TREM2 holoprotein and R47H sTREM2. Moreover, the R47H sTREM2 promotes formation of morphologically distinct Aβ protofibrils. These data indicate additional mechanisms by which TREM2 protects against Alzheimer's disease, and a previously unrecognised mechanism by which the R47H mutant increases risk.

Aβ oligomers bind TREM2 holoprotein and induce TREM2 ectodomain shedding
To confirm that TREM2 can act as a cell surface receptor for Aβ oligomers, we treated mouse primary microglia or TREM2-transfected HeLa cells with Aβ42 oligomers that have been characterized both by electron microscopy and by oligomer/fibril specific antibodies (Supplementary Fig. 1; note only Aβ42 was used in this work and will be referred to as Aβ henceforth). We then immunoprecipitated TREM2 from lysates of these cells, and found that Aβ co-immunoprecipitated with endogenous TREM2 on primary microglia from wild-type mice, but not on microglia from TREM2 -/knockout mice ( Supplementary Fig. 2i). The Aβ binding was prevented by TREM2-blocking antibody ( Supplementary Fig. 2ii). Oligomeric Aβ bound to TREM2 but not TREM1 ( Supplementary Fig. 2iii). Monomeric Aβ co-immunoprecipitated with TREM2 less efficiently than oligomeric Aβ (Supplementary Fig. 2iv and v, p = 0.03). The TREM2: Aβ oligomer interaction is at least partially specific because neither of two other neurodegeneration-associated oligomeric proteins (oligomeric α-synuclein or oligomeric Tau) bound to TREM2 even in HeLa cells overexpressing TREM2 and DAP12 ( Supplementary Fig.   3). These results, confirm and extend the work of other groups showing that oligomeric Aβ binds TREM2 [13][14][15] .
We next tested the consequences of Aβ oligomers binding to TREM2 on primary microglia. Aβ oligomers induced release of sTREM2 into the medium of primary microglia from wild-type mice, but not microglia from mice engineered to express R47H TREM2 ( Supplementary Fig. 4i, p<0.05). To study this effect in more detail, we stably expressed TREM2 (together with DAP12) in HEK293 cells. Treatment of these cells with Aβ oligomers resulted in dose-dependent release of sTREM2 into the medium and the accumulation of   one-way ANOVA with Tukey's post-hoc multiple comparisons test. iv) Example field of singlemolecule TIRF imaging of mixture of Aβ oligomers (green) and wild-type TREM2 ectodomain (red), where co-localised spots appear yellow. Scale bar: 1 micron. Magnified image of three sections of field at right. v) Proportion of monomeric or oligomeric Aβ colocalized with wildtype sTREM2. vi) Proportion of Aβ oligomers colocalized with wild-type or R47H TREM2 ectodomain. For v) & vi), error bars = SEM; ****p<0.0001, n=3 independent preparations, each analysed in 9 fields each; two tailed t test of significance.  To study reversal of oligomerization, pre-assembled Aβ oligomers were diluted to 2 µM and mixed with wild-type or R47H sTREM2 (at molar ratios of sTREM2: total Aβ of: 1:5 and 1:1) and incubated for 30 minutes at 37°C. The Aβ assemblies were then assessed using the same methods as above. This experiment revealed that wild-type sTREM2 disaggregated pre-formed    prevented by 1 µM sTREM2 (Fig. 3iv). In contrast, at the same low dose (20nM), R47H sTREM2 had minimal effect on Ab fibrillization (Fig. 3v).
To test whether sTREM2 could disaggregate fibrils of Aβ, small fibrils were preassembled and diluted to 2 µM (monomer equivalent) and mixed with wild-type or R47H  treated with R47H sTREM2 at molar ratios of 1:1 and TEM imaged after 30min. ix) Quantification of the area of Aβ fibrils confirmed that WT sTREM2 decrease Aβ fibrils, and R47H sTREM2 had no effect. Error bars represent SD. Statistical analysis was performed using one-way ANOVA followed by Bonferroni's multiple comparison test (n=3-8, *p<0.05 vs preformed Aβ fibril).

Wild-type sTREM2 inhibits, while R47H sTREM2 increases Aβ oligomer neurotoxicity
Because sTREM2 bound Aβ oligomers and inhibited Aβ oligomerisation and fibrillization, we next tested whether sTREM2 affected the neurotoxicity of Aβ. Aβ neurotoxicity is thought to be mediated by 2 different mechanisms -namely by permeabilization of neuronal membranes 28 , and by microglial activation 28,29 .
Consequently, we initially used nanosized phospholipid membrane vesicles containing a calcium-sensitive fluorophore to explore the ability of Aβ to permeabilize membranes. Prior work has shown that Aβ oligomers (but not monomers or fibrils) can permeabilize these membranes 28 . We found that wild-type sTREM2 and R47H sTREM2 proteins themselves had no effect on membrane permeabilization when added in the absence of Aβ ( Supplementary Fig. 13i). By contrast, Aβ oligomerized for 6 hours in the absence of sTREM2 induced permeabilization of the vesicles. However, Aβ oligomerized in the presence of wild-type sTREM2 (at molar ratios of sTREM2: total Aβ of 1:10) induced significantly less permeabilization (Figure 4i). Wild-type sTREM2 also inhibited the permeabilization induced by pre-formed Aβ oligomers (at a 1:1 molar ratio) (Fig. 4ii). The simplest explanation for this inhibition of permeabilization is our previous finding that wild-type sTREM2 inhibits Aβ oligomerization and disaggregates Aβ oligomers at these concentrations. However, we do not discount the possibility that binding of wild-type sTREM2 to Aβ oligomers might also reduce their toxicity.
When these experiments were repeated using R47H sTREM2 (at a molar ratio of sTREM2: total Aβ of 1:10) we found that Aβ oligomers formed in the presence of R47H sTREM2 also induced less permeabilization than Aβ oligomerized alone, but allowed more permeabilization than Aβ oligomerized with wild-type sTREM2 (Fig. 4i). In contrast to wild-type sTREM2, R47H sTREM2 did not significantly inhibit permeabilization induced by pre-formed Aβ oligomers ( Figure 4ii). These results are consistent with wild-type sTREM2 reducing Aβ oligomer abundance, but R47H sTREM2 not reducing Aβ oligomer abundance (Fig. 2).
We next tested whether sTREM2 could block the neurotoxicity induced by Aβ in neuronal-glial co-cultures. We have previously shown that 250 nM Aβ induces slow, progressive neuronal loss mediated by microglia in this co-culture system 29 . We found that 250 nM Aβ induced neuronal loss over three days, and this Aβ-induced neuronal loss was substantially reduced by co-treatment with 25 nM wild-type sTREM2 (Fig. 4iii & iv). In contrast, co-treatment of the cultures with 25 nM R47H sTREM2 increased the neuronal loss above the level induced by 250 nM Aβ alone (Fig. 4iv). Wild-type and R47H sTREM2 proteins by themselves had no significant effect on neuronal loss ( Supplementary Fig. 13ii). As membrane permeabilization is thought to be mediated by smaller Aβ oligomers, while microglial activation is thought to be mediated by larger Aβ aggregates 28 , our finding that R47H sTREM2 increased Aβ-induced neuronal loss (Fig. 3iv), which is known to be microglia-mediated 29 , is consistent with R47H sTREM2 increasing larger Aβ aggregates (Fig. 2). Overall, our results indicate that wild-type sTREM2 reduces Aβ neurotoxicity, while R47H TREM2 increases Aβ neurotoxicity.   : iii), and the number of apoptotic, necrotic and healthy neurons were quantified (mean data: iv). Loss is the decrease in neuronal density relative to vehicle treated cultures. Error bars = SEM; **=p<0.01, ***=p<0.001, ****=p<0.0001; n=4 independent experiments on separate cell cultures. Statistical analysis was by one-way ANOVA and Tukey's post hoc test.

Discussion
We have shown that Aβ oligomers, but not Aβ monomers or fibrils, bind to microglial TREM2 and induce sTREM2 release, and this released sTREM2 also preferentially binds to Aβ oligomers, consistent with previous reports [13][14][15] . Crucially however, we also found that wild-type sTREM2 inhibits the formation of fibrils and larger Aβ oligomers and disaggregates protofibrils and larger Aβ oligomers into the smallest Aβ oligomers. In contrast, R47H sTREM2 promoted the formation of Aβ protofibrils, indicating a gain of function by this mutation. These effects may provide a partial explanation of how a single copy of the TREM2 R47H mutant is associated with increased risk for AD, by increasing production of neurotoxic forms of Aβ.
One potential caveat is the protein concentrations used in our experiments. In order to observe Aβ aggregation over a reasonable experimental timeframe, the lowest concentration of Aβ used here was 100 nM, whereas levels found in CSF are about 0.1 nM 10,20,30 . However, the Aβ oligomer concentrations near amyloid plaques was estimated as 700 nM 31 , and plagues are likely be a more relevant location for Aβ aggregation in AD than CSF. sTREM2 concentrations in lumbar CSF average about 0.2 nM 10,20,30 ; however again, it is likely that sTREM2 levels are higher near amyloid plaques surrounded by activated microglia, which express high levels of TREM2 3 . Note also that the observed effects of sTREM2 on Aβ are rapid: 20 nM sTREM2 dissolved 100 nM Aβ oligomers within 30 mins in vitro (Supplementary Fig 11). Lower concentrations are likely to have the same disaggregating effect over a slower time course, which may be more relevant to the slow time course of AD.
We did not investigate the molecular mechanisms by which sTREM2 affected Aβ aggregation, but our finding that wildtype sTREM2 stabilises the smallest observable oligomers (Supplementary Figure 9 & 11), suggests the possibility that sTREM2 binds preferentially to these (or undetectable oligomers such as dimers), thereby potentially blocking further growth of aggregates, and disaggregating larger Aβ aggregates into smaller, potentially non-toxic forms.
Precedent for this is the finding that α-synuclein-specific single-domain antibodies (nanobodies) bind α-synuclein stable oligomers and convert them into less stable oligomers with reduced toxicity 32 . However, clearly more research is required to determine the mechanisms for sTREM2 and Aβ.
In summary, our experimental data reveal how Aβ oligomer binding to TREM2 may mediate direct (via activation of intracellular TREM2 dependent signalling pathways) and indirect protective mechanisms (via effects on Aβ oligomer assembly and toxicity). Our studies are broadly congruent with previous research showing that knockout of TREM2 expression in APP mice, resulted in accelerated amyloid plaque seeding 17 , with increased protofibrillar halos and hotspots around these plaques 3,16,18 . Our studies provide a potential explanation of the clinical observation of slower rates of cognitive and clinical decline in patients with MCI or AD who have higher levels of sTREM2 in the CSF 20,21 , and slower rates of amyloid deposition in healthy controls and MCI patients with higher sTREM2 levels in CSF 22 . They also provide a potential explanation for the beneficial effects of infusing or expressing sTREM2 into the brain of mouse models of AD 16 . Additional biophysical studies will be required to identify the key Aβ species dynamically interacting with wild-type sTREM2 to prevent neurotoxicity, and the Aβ species stabilised by sTREM2 R47H to increase neurotoxicity. This knowledge could be exploited for the design of small brain-penetrant molecular mimics of sTREM2 as potential therapeutics for AD.

Antibodies
The following antibodies were used: human TREM2 (#AF1828, 1:100 for immunocytochemistry (ICC); 1:1000 for western blot; R&D); mouse TREM2 (#AF1729, 1:500 for TREM2 deficient mice were obtained from Dr. J. Gommerman, and were constructed by targeted homologous recombination 33 , which removed Exon 1 and 2, which include the start codon and the major extracellular IgG domain. In contrast to the recently reported Velocigene construct, the direction of the Hygromycin cassette was "reversed". Crucially, in agreement with two other models, but in contrast to the Velocigene construct, RT-PCR analyses confirmed specific loss of TREM2 expression without the perturbation of TREM1L expression observed in the Velocigene construct 33 .

Primary microglial cell culture
Primary cultures of mixed glial cells and pure microglia were prepared from the cerebral cortex of P0-3 day old C57BL/6 TREM2 +/+ or TREM2 -/mice as described 33 . After removal of olfactory bulbs, cerebral hemispheres were cut into ~1mm pieces, vortexed for 1 minute, filtered through a 40 µm cell strainer and plated in 80 cm 2 tissue culture flasks. Mixed glia were cultured until microglia appeared (10-18 days after plating). Microglia were isolated by shaking flasks.
Purified microglia were replated for analysis. On average, each experiment used microglia from 30-35 pups. All pups used in this study (wild type; R47H, and TREM2KO) were maintained on a C57BL/6 background.

Mixed neuronal-glial co-cultures and treatments
Mixed neuronal-glial co-cultures were prepared from cerebella of 5-7 day-old rats 33 , cultured for 7 days, then treated with either: vehicle, 250 nM monomeric amyloid beta 1-42 (Ab), 250 nM Ab + 25 nM wild-type sTREM2 or 250 nM Ab + 25 nM R47H sTREM2 for 3 days. Three days later, neuronal death and loss was quantified by staining the live cultures with propidium iodide for necrosis, with isolectin IB4 for microglia, and with Hoechst 33342 for nuclei, which is used to distinguish healthy (uncondensed) from apoptotic (condensed) nuclei, and to distinguish astrocytes (large bean-shaped nuclei) from neurons and microglia (small, round nuclei). The number of apoptotic, necrotic and healthy neurons, astrocytes and microglia are then counted on photos of multiple set fields, as previously described 29,33 ,

Amyloid β-peptide
Human Aβ42 (used in TREM2 cleavage), HiLyte Fluor 488-labelled Aβ42 (used in Single molecule TIRF imaging), HiLyte Fluor 647-labelled Aβ42 (used in IP and dot blot) and Biotin-Aβ42 (used in BLI) were purchased from AnaSpec and Bachem; Aβ42 oligomers were prepared as endotoxin-free preparations 34 , and validated by antibodies, gels and electron microscopy (Supplimentary Figure 1A). Prior work has established that these fluorescent tags do not significantly impact Aβ assembly 35 .

Recombinant ectodomain expression and purification
cDNAs encoding residues 19-143 of WT and R47H TREM2 were ordered as linear DNA strings (Life Technologies) and inserted into pHLSec. The plasmids were transfected into Freestyle 293-F cells (Life Technologies) grown in suspension in FreeStyle293 medium. Cells were transfected at a density of 2.5 x 10 6 / ml with a DNA concentration of 3 µg/ml and polyethyleneimine (Linear PEI 25 kDa molecular weight, PolySciences Inc) at 9 µg/ml. On day 6 post-transfection, conditioned medium was harvested and TREM2 protein was purified using IMAC (Histrap excel, GE Healthcare) followed by gel filtration chromatography in storage buffer -20 mM HEPES, 200 mM NaCl, pH 7.0. To maintain a low endotoxin level: endotoxin-free chemicals and plasticware were used; all chromatography media, columns and concentrators (Vivaspin Turbo) were soaked for > 12 hours in 1 M NaOH. Proteins in storage buffer were concentrated to between 5-10 mg/ml for use and tested for endotoxin with the EndoZyme recombinant Factor C assay (Hyglos GmbH). Endotoxin levels in all assays were maintained at < 0.1 EU/ml.

Co-immunoprecipitation and western blot
Primary microglia were rinsed with ice-cold DMEM without phenol red and incubated with 100 nM HiLyte Fluor 647-labelled Aβ at 4°C for 1 hour. For antibody blocking assays, before Aβ incubation, microglia cultures were pre-incubated with 25 µg/ml monoclonal anti-TREM2 antibody or rat IgG1 control for 20 minutes Cells were collected by scraping, washed with ice cold 1xPBS and centrifuged. Primary microglia were lysed with 1% CHAPSO buffer ( In vitro co-precipitation of TREM2 or TREM1 with Aβ oligomers was performed by mixing recombinant Fc-tagged TREM2 or TREM1 ectodomain (residues 19-171; 100 ng/ml, R&D systems), 100 nM HiLyte Fluor 647-labelled Aβ oligomers and Protein G Dynabeads in 0.5% fatty acid free BSA PBS-Tween buffer. The mixture was incubated at room temperature for 1 hour and then washed, eluted and analysed the same as described above.

Dot blot
Aβ monomers or oligomers (1 μg/dot) prepared as described above and anti-TREM2 antibody (0.3 μg/dot) were spotted onto a nitrocellulose membrane. Membrane strips were blocked with 3% fatty acid free BSA (Sigma Aldrich) and then incubated with 100 nM recombinant His-tagged WT/R47H TREM2 ECD diluted in blocking buffer at 4°C for 1 hour.
Bound TREM2 ECDs were probed with HRP-conjugated anti-His tag antibody and detected with Odyssey FC imaging system.  iv) Single-molecule TIRF Imaging. The co-localization experiment was designed to investigate binding of Aβ oligomers with TREM2. The Aβ 6 hour aggregation contained a mixture of monomeric and oligomeric species hence illumination was performed for significant time periods to photo-bleach the monomeric species. Automated TIRF co-localization experiments were performed with the use of a custom script (BeanShell, micromanager). Each data set consisted of a 3 X 3 grid of 9 images at different areas of the coverslide, distances between images was 350 µm as previously described 20 . Images were recorded at 33 ms exposure, beginning with 800 frames excitation with 641 nm followed by 800 frames excitation 488 nm excitation in the same field of view. v) Co-localization Analysis. Co-localization data was analyzed with a bespoke ImageJ macro.

Bio-Layer Interferometory i) Preparation of Aβ oligomers for Bio-Layer Interferometry (BLI) studies. Synthetic human
The 488 nm illumination channel contained a mixture of monomeric and oligomeric Aβ species, it was determined that the majority of monomeric species were photo-bleached by 40 frames of 488 nm illumination. Therefore, only frames after this point were considered in this analysis. The 488 nm and 641 nm illumination channels were compressed in time to create two single frame images representing the average pixel intensities. Following which, points of intensity representing HiLyte™ Fluor 488 Aβ or AF647-TREM2 above a background threshold were located, counted and binary images of these maxima were created. The two images were then summed to identify co-localized points. Chance co-incident spots were excluded by 90° rotation of the binary image representing AF647-TREM2 points and summation with the HiLyte™ Fluor 488 Aβ image. Chance coincident spots were subtracted from the actual coincidence value and percentage coincidence was calculated with the equation below: . × 100

TREM2 cleavage assays
Primary microglia from wild and TREM2 mutant mice were isolated from mixed glia

Membrane permeabilization assay
Lipid vesicles were prepared as previously described 38

Resource Availability
The published article includes all data generated or analyzed during this study.

Declaration of Interests
The authors have no conflicts of interests to declare.   . iv) Aβ oligomers bind to TREM2 significantly better than monomers. HeLa cells transfected with DAP12-FLAG plus TREM2-FLAG were co-incubated with 100 nM freshly prepared monomeric or oligomeric HiLyte Fluor 647-labelled Aβ at 4°C for 1 hour. Prior to incubation, the concentration of monomer/oligomer were adjusted and tested according to fluorescence dot exposure to ensure total Aβ concentrations were the same in the working solutions of monomeric and oligomeric Aβ preparations. Cell membrane lysates were then immunoprecipitated with anti-hTREM2 antibody. The IP products were western blotted and probed with the indicated antibody. v) Aβ oligomers bind to TREM2 significantly better than monomers. Quantification of iv) with n = 4 replicates in 2 independent experiments; p = 0.03 Mann-Whitney U test). Error bars = SEM.

TREM2
38-  Primary microglia were incubated with 100 nM oligomeric α-synuclein or Tau protein at 4°C for 1 hour. Cell membrane lysates were then immunoprecipitated with human TREM2 antibody or control IgG. The IP products were probed with the indicated antibody.   Table 1. WT sTREM2 decreases Ab oligomer assembly, whereas R47H sTREM2 increases Ab oligomer disassembly. i) Aβ was oligomerized in the presence of absence of WT or R47H sTREM2 (sTREM2 1:50 or 1:100 molar ratio to Ab). After by transmission electron microscopy imaging (representative images in Fig 2), the numbers of Aβ aggregates (oligomers) were counted. ii) Aβ was oligomerized, and then incubated in the presence of absence of WT or R47H sTREM2 (sTREM2 1:5 or 1:1 molar ratio to Ab). After imaging (representative images in Fig 2), the numbers of Aβ aggregates (oligomers) were counted. Error bars represent SD. Statistical analysis was performed using one-way ANOVA followed by Bonferroni's multiple comparison test (n=3, *p<0.05, **p<0.01 vs Aβ oligomer alone). ii) 21   Error bars represent SD. Statistical analysis was performed using one-way ANOVA followed by Bonferroni's multiple comparison test (n=4, *p<0.05, ***p<0.001 Aβ oligomer alone). iii) Size distribution of the aggregates on from three independent TEM images.