Amyloid-β Protein Oligomer at Low Nanomolar Concentrations Activates Microglia and Induces Microglial Neurotoxicity*

Neuroinflammation and associated neuronal dysfunction mediated by activated microglia play an important role in the pathogenesis of Alzheimer disease (AD). Microglia are activated by aggregated forms of amyloid-β protein (Aβ), usually demonstrated in vitro by stimulating microglia with micromolar concentrations of fibrillar Aβ, a major component of amyloid plaques in AD brains. Here we report that amyloid-β oligomer (AβO), at 5–50 nm, induces a unique pattern of microglia activation that requires the activity of the scavenger receptor A and the Ca2+-activated potassium channel KCa3.1. AβO treatment induced an activated morphological and biochemical profile of microglia, including activation of p38 MAPK and nuclear factor κB. Interestingly, although increasing nitric oxide (NO) production, AβO did not increase several proinflammatory mediators commonly induced by lipopolyliposacharides or fibrillar Aβ, suggesting that AβO stimulates both common and divergent pathways of microglia activation. AβO at low nanomolar concentrations, although not neurotoxic, induced indirect, microglia-mediated damage to neurons in dissociated cultures and in organotypic hippocampal slices. The indirect neurotoxicity was prevented by (i) doxycycline, an inhibitor of microglia activation; (ii) TRAM-34, a selective KCa3.1 blocker; and (iii) two inhibitors of inducible NO synthase, indicating that KCa3.1 activity and excessive NO release are required for AβO-induced microglial neurotoxicity. Our results suggest that AβO, generally considered a neurotoxin, may more potently cause neuronal damage indirectly by activating microglia in AD.

pathological hallmarks of AD. A␤ aggregates, especially those of the 42-residue A␤42, are potent toxins to neurons. In addition, A␤ aggregates activate microglia, which clear A␤, but in the process release inflammatory mediators that cause indirect neurotoxicity (1). A commonly held view is that microglia react to fA␤ deposited in amyloid plaques, which form a nidus for microglia-mediated neuronal damage. However, because the number of amyloid plaques correlates poorly with the degree of dementia (2,3), it follows that plaque-associated microglia may not be a major player in cognitive impairment. Emerging evidence including two recent in vivo positron emission tomography studies indicates that microglia may respond to stimuli other than plaque fA␤ (4,5), suggesting that alternative ways to activate microglia must occur in AD.
In addition to fibrillar forms, A␤ exists in various smaller assemblies in AD brains, which may mediate diverse toxic effects at different stages of the disease. Although these smaller aggregates, variously named as oligomers (A␤O), protofibrils, amyloid pores, or AD diffusible ligands, have been considered transient or metastable intermediates in fibril formation (6), some of them may not be obligate intermediates in the fibril formation pathway and can be stable (7,8). Importantly, recent in vitro and in vivo studies have revealed that the build-up of soluble A␤O may be an early and central event in the pathogenesis of AD (9 -12). The strong and rapidly disruptive effect of A␤O on synaptic plasticity and neuronal integrity is hypothesized to cause memory problems in AD and is generally attributed to their direct neuro-or synaptotoxicity (3). However, one plausible but less studied possibility is that A␤O activates microglia and causes indirect, microglia-mediated neuro-and synaptotoxicity. However, A␤O, at micromolar concentrations, appeared no more potent than fA␤ in activating microglia (13)(14)(15)(16)(17). In contrast, A␤O is much more potent than fA␤ in causing synaptic disruption and neuronal death (10,11,18,19).
To examine microglia activators that are pathologically relevant to the development of AD, we investigated the ability of low levels of A␤O to activate microglia. Here we report a previously unrecognized A␤O effect in activating microglia at 5-50 nM, concentrations that are usually not sufficient to cause direct neurotoxicity. This effect follows a bell-shaped dose-response curve and tapers off at concentrations above 100 nM. Importantly, we show that A␤O-activated microglia release NO to cause neuronal damage, and this mode of microglia activation and neurotoxicity is dependent on microglial KCa3.1. Therefore, our data suggest a significant contribution of A␤O to microglial neurotoxicity in AD, which may even be more potent than its direct neurotoxicity. This mode of neurotoxicity could be alleviated by inhibitors of KCa3.1 channel activity and NO production.
Preparation of A␤ Solutions-A␤O solutions as well as the unaggregated and fibrillary A␤ were prepared as described (21). Our preparation of A␤O follows the procedure described by Lambert et al. (9) except that the A␤42 peptide was diluted with Opti-MEM culture medium instead of the F12 medium originally described, before incubation at 4°C for 24 h to generate oligomers. This preparation of A␤O has been extensively characterized in our laboratory (21,22). To ensure consistency of quality, a random sample from each batch was chosen and imaged using electron microscopy and atomic force microscopy to characterize the size and shape of the aggregates (21)(22)(23). The biological activities of each batch were confirmed by determining for A␤O the neurotoxic activity, synaptic binding activity, and ability to rapidly induce exocytosis of MTT formazan, as described previously (21,23).
Soluble A␤O from Human Brain Tissue-We obtained hippocampal tissues from three AD subjects and two cognitively and pathologically normal control subjects from the University of California Davis Alzheimer's Disease Center. The use of these samples has been approved by the University of California Davis Institutional Review Board Administration and was reported before (21). All of the subjects had comparable postmortem intervals averaging 5.5 h. Soluble extracts from brain tissues were prepared as described (19,21,24). Molecular weight fractionation of oligomeric species was performed using Centricon YM-100 and YM-10 concentrators (Millipore, Bedford, MA). The relative abundance of A␤O in the resulting solutions was determined by Western blots using the 6E10 antibody and dot blots using the A11 antibody (25). Although the AD samples contained various amounts of A␤O, the two control samples showed no detectable A␤O on Western blots and almost background levels on dot blots.
To ensure accurate measurement of the total quantity of A␤ in the extract, we used two A␤ ELISA kits purchased from IBL America (Minneapolis, MN) and Wako Chemicals USA (Richmond, VA), respectively. The procedures were conducted according to manufacturer's protocols.
Primary Cultures-Primary microglia cultures derived from newborn C57BL/6J mice were prepared from mixed glia cul-tures (1-2 weeks in vitro) with the "shaking off" method as described previously (26). Cultures were Ն99% pure for microglia as demonstrated by anti-IBA1 immunostaining. To obtain conditioned medium (CM) for treating neurons, microglia were first cultured in 24-well culture plates at a density of 1.3 ϫ 10 5 cells/cm 2 for 24 h in DMEM with 10% fetal bovine serum (DMEM10). The cultures were washed extensively and changed to the neurobasal medium with B27 supplement (NB/B27; Invitrogen) without serum and cultured for another 24 h. The NB/B27-based microglia CMs were collected, briefly centrifuged, and used immediately or frozen for future uses.
Hippocampal neuronal cultures were prepared from newborn wild type C57BL/6J mice according to the method of Xiang et al. (27). The neurons were cultured in NB/B27 at a density of 2.5 ϫ 10 5 cells/well in 12-well plates or 8 ϫ 10 5 cells/well in 6-well plates for at least 14 days before they were treated with microglia CM.
Hippocampal slice cultures (400 m thick) were prepared from 7-day-old C57BL/6J mice as described previously (26) and cultured for 10 days in vitro before use. Neuronal damage in the slices was monitored by PI uptake following Bernardino et al. (28). PI itself is not toxic to neurons (28). Briefly, hippocampal slices were pretreated with or without doxycycline (20 M), TRAM-34 (1 M), 1400W (5 M), or L-NIL (100 M) for 1 h and then treated with medium containing PI (2 M) and A␤O of indicated concentration, with or without the above inhibitor compound. After 24 h, the slices were observed under a Nikon Eclipse E600 microscope, and the red (630 nm) fluorescence emitted by PI taken up by damaged cells was photographed by a digital camera (SPOT RTke, SPOT Diagnostics, Sterling Heights, MI) with fixed exposure time.
5Ј-Bromodeoxyuridine (BrdUrd) Incorporation Assay-Microglia were plated onto 48-well culture plates at a density of 1 ϫ 10 5 cells/well in DMEM10 and incubated for 24 h. The cells were washed with serum-free Opti-MEM medium three times and treated with the indicated concentrations of A␤O in Opti-⌴⌭⌴. After 5 h of incubation, 10 M BrdUrd was added and allowed to be incorporated into DNA of proliferating cells during an additional 16 h of incubation. The cells with positive BrdUrd incorporation were determined by an immunocytochemical stain for BrdUrd using an anti-BrdUrd antibody conjugated with Alexa594 (Invitrogen) following the manufacturer's protocol (Chemicon) and counted.
NFB Assay-The detection of cells with active NFB was performed according to a previously published method (29). Briefly, microglia were plated as described above and treated with activators for 2 h. The cells were fixed and stained with anti-NFBp65 antibody (1:250; Chemicon) and with the nuclear dye DAPI.
Assays for Neuronal Viability-Hippocampal neurons were prepared as described above and were plated onto 96-well plate at a density of 6 ϫ 10 4 cells/well and cultured for 14 days. CM from microglia cultures was added onto neurons with indicated dilutions, and the cultures were incubated for 24 h. Neuronal viability was evaluated by the MTT assay and the lactate dehydrogenase release assay as described previously (22,26).
For quantification of the PSD95-immunoreactive puncta along the dendrites, photomicrographs of PSD95 immunostained cultures were randomly taken from each culture condition. The images were transformed to 8-bit gray scale and analyzed with the Image J program. The number of puncta in each photomicrograph was counted and normalized by dendritic length. The photography and analysis were conducted in an investigator-blinded manner.
Western Blot Analysis-To obtain lysates, cells or tissues were washed with ice-cold PBS and incubated with a buffer containing 50 mmol/liter Tris-HCl, pH 7.4, 150 mmol/liter NaCl, 2% SDS, proteinase inhibitor mixture (Sigma), and phosphatase inhibitor mixture (Sigma). The lysates were briefly sonicated and cleared by centrifugation at 50,000 rpm for 10 min. Equivalent amounts of protein were analyzed by Tris-HCl gel electrophoresis. The proteins were transferred to polyvinylidene difluoride membranes and probed with antibodies. Visualization was performed using ECL (Amersham Biosciences).
The following primary antibodies (dilutions) were used: Assay for NO-Conditioned medium collected from microglia cultures (1 ϫ 10 5 /0.75 cm 2 ) and hippocampal slices treated with A␤O for 24 h were analyzed by a nitric oxide quantization kit according to the protocol of the manufacturer (Active Motif, Carlsbad, CA). The data were normalized to the amount of total protein.
Patch Clamp Experiments-Microglia derived from mixed glia cultures (1-2 weeks in vitro) were washed, attached to poly-L-lysine-coated coverslips, and then studied in the whole cell mode of the patch clamp technique with an EPC-10 HEKA amplifier. The pipette solution contained 145 mM K ϩ aspartate, 2 mM MgCl 2, 10 mM HEPES, 10 mM K 2 EGTA, and 8.5 mM CaCl 2 (1 M free Ca 2ϩ ) , pH 7.2, 290 mOsm. To reduce chloride "leak" currents, we used a Na ϩ aspartate external solution containing 160 mM Na ϩ aspartate, 4.5 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 5 mM HEPES, pH 7.4, 300 mOsm. K ϩ currents were elicited with voltage ramps from Ϫ120 to 40 mV of 200-ms duration applied every 10 s. Whole cell KCa3.1 conductances were calculated from the slope of the TRAM-34-sensitive K Ca current between Ϫ80 and Ϫ75 mV, where KCa3.1 currents are not "contaminated" by Kv1.3 (which activates at voltages above Ϫ40 mV) or inward rectifier K ϩ currents (which activate a voltages more negative than Ϫ80 mV). Cell capacitance, a direct measurement of cell surface area, was continuously monitored during recordings. KCa3.1 current density was determined by dividing the TRAM-34-sensitive slope conductance by the cell capacitance.
Statistical Analysis-Statistical analyses were performed using SigmaPlot 11 software (Systat Software, Inc). Analysis of variance or repeated-measures analysis of variance was used to compare quantitative values from cultures across groups. Tukey's studentized range test was used to adjust for multiple comparisons in post hoc pairwise tests.

A␤O at Low Nanomolar Concentrations Stimulates Microglia into a Distinct Activation
Phenotype-Our A␤O preparations affected viability of cultured hippocampal neurons at concentrations above 100 nM, consistent with previously published results (22). Interestingly, the same preparations induced proliferation of cultured microglia, starting at 5 nM (22.5 ng/ml) and maximizing at ϳ50 nM (225 ng/ml) (Fig. 1, A and B). At 24 and 48 h post-stimulation, cell counts in cultures treated with 20 nM A␤O were 184 Ϯ 8.59% (mean Ϯ S.E., n ϭ 6, p Ͻ 0.001; Fig. 1D) and 330 Ϯ 21.8% (n ϭ 3, p Ͻ 0.001; Fig. 1B) of the values of solvent-treated controls, respectively. The identity of the proliferating cells was confirmed by their immunoreactivity to anti-ionized calciumbinding adaptor molecule 1 (IBA-1), a microglia-specific marker (Fig. 1A). The mitogenic effect was confirmed by BrdUrd incorporation. Proliferation measured by both cell counting and BrdUrd followed a bell-shaped curve (Fig. 1, B and C), which is very similar to the previously reported chemotactic activity of soluble A␤ for macrophages/microglia that also maximized at low nanomolar concentrations (30). The mitogenic effect was blocked by a neutralizing oligomerspecific antibody A11 (25) and by CR, an amyloid ligand that neutralizes the activity of A␤O (21) (Fig. 1D). Because both A11 and CR recognize conformational features but not the primary sequence of A␤, our results indicate that the conformation of A␤O was responsible for the mitogenic effect. The unaggregated A␤1-42 showed a mild although statistically not significant induction of proliferation (Fig. 1B), attributable to the small amount of A␤O invariably formed in aqueous solution because this effect was also blocked by A11 and CR. Fibrillar A␤ and preparations of reverse sequence A␤ peptide (A␤42-1) up to 500 nM showed no effect (supplemental Fig.  S1). These results indicate that at concentrations lower than those required for direct neurotoxicity, A␤O is mitogenic for microglia.
To determine whether A␤O derived from human brain would also promote microglia proliferation, we obtained soluble extracts from hippocampi of AD and age-matched nondemented control subjects. Using Centricon concentrators with molecular mass cut-off between 10 and 100 kDa, we prepared fractions containing A␤O but excluding unaggregated A␤ and large aggregates. This preparation of human A␤O has been well characterized in previous studies (19,21,24,31). It contains bioactive A␤O indistinguishable from synthetic A␤O in terms of structure and selective attachment to nerve cell sur-FIGURE 1. Synthetic and human-derived A␤O stimulates microglia proliferation at subneurotoxic concentrations. A, microglia were treated with A␤O of indicated A␤ concentrations for 48 h. All of the cells in the microglia cultures were stained with Calcein AM. The identity of cells was confirmed by immunostaining with IBA-1, a microglia marker. Cont, control. B, the number of cells in each condition was determined. A␤O treatment caused microglia proliferation in a dose-dependent manner, and the effect tapered off at 100 nM. A␤ monomer caused a mild but statistically not significant increase of proliferation. The data presented are the means Ϯ S.E. (n ϭ 3). *, p Ͻ 0.05; **, p Ͻ 0.001 compared with the 0 nM solvent treatment controls. C, mitogenic assay based on BrdUrd incorporation. A␤O treatment caused a dose-dependent incorporation of BrdUrd, following a bell-shaped curve (n ϭ 3). *, p Ͻ 0.05; **, p Ͻ 0.001 compared with the 0 nM solvent treatment controls. D, microglia were treated for 24 h with 20 nM A␤O or mock-treated with solvent, in the presence of A11 (50 nM), or CR (100 nM), and the cell numbers were determined (n ϭ 5). *, p Ͻ 0.001; #, p Ͻ 0.05. As a control for the specificity of A11, a rabbit polyclonal antiserum specific for A␤40 (anti-A␤40) was also tested and showed no difference from the control. E, soluble extracts with a 10 -100-kDa molecular mass cut-off were obtained from AD and control hippocampi. Microglia were treated with these extracts diluted into the medium at indicated percentile (v/v), and the numbers of cells were determined after 24 h. Shown are data from using a representative pair of AD (black bars) and control (white bars) extracts. The actual A␤42 concentrations (Conc., in nM) for the indicated dilutions of the AD extract are indicated on the x axis (n ϭ 3). *, p Ͻ 0.05 for control versus AD in each concentration between the pair of extracts; **, p Ͻ 0.001 for 10% AD versus 10% AD ϩ A11 and for 10% AD versus 10% AD ϩ CR.
faces (21,24). It contains A␤O species recognizable by A11 (21) and a main species at 56 kDa, which is perhaps equivalent to the A␤42 12-mer oligomers (dubbed A␤*56) shown to correlate with cognitive deficits in Tg2576 mice (10). We determined the quantities of A␤ in these extracts by sandwich ELISA. Because ELISA may underestimate the total quantity of A␤ when in oligomeric forms (32), we first deaggregated A␤ in the extract with guanidine HCl as described previously (33). In addition, we used two different commercial ELISA kits and obtained consistent A␤ concentrations. Surprisingly, soluble extracts containing A␤O from all three AD brains consistently stimulated microglia proliferation at sub-nanomolar to low nanomolar concentrations (ϳ0.11-1.24 nM A␤42) ( Fig. 1E and supplemental Fig. S2) and thus were ϳ50fold more potent than the synthetic A␤O. Parallel treatments of hippocampal neurons with soluble AD brain extracts of the same or higher concentrations did not show any effect on neuronal viability (supplemental Fig. S3). The equivalent fractions from two age-matched control individuals contained little A␤O and showed no microglia mitogenic effect. The addition of A11 or CR blocked the mitogenic effect (Fig. 1E), confirming that this effect was mediated by A␤O. These results again show that A␤O is a potent mitogen for microglia at concentrations that do not cause direct neurotoxicity.
Microglia after synthetic or AD patient brain-derived A␤O treatment showed a morphology resembling those activated by LPS or macrophage colony stimulatory factor ( Fig. 2A). Biochemical and immunocytochemical evidence also supports that A␤O treatment induces an activated phenotype. A␤Otreated microglia showed strong up-regulation of MAC-1 (CD11b) and SRA, two markers of microglia activation ( Fig.  2A) (34), and increased levels of phosphorylated, therefore activated, p38 MAPK, although the total levels of total p38 MAPK protein were not changed (Fig. 2, B and C). A␤O treatment further induced a 6-fold increase of cells showing immunoreactivities of active nuclear factor B (NFB, n ϭ 6, p Ͻ 0.001; Fig. 2D) (29). This set of cellular responses is qualitatively similar to those induced by LPS activation through the CD14/TLR4 co-receptors, including the stimulation of pathways that are dependent on p38 MAPK and NFB signaling (26).
Microglia activation is often accompanied by increased release of NO, synthesized by the inducible NO synthase (iNOS) (35). We found that A␤O treatment significantly increased NO generation as evaluated by measuring the concentration of nitrite, its stable metabolite, released into the medium (36). After normalization to the total amount of cellular protein, the data indicate an ϳ80% increase of NO release/cell; therefore this increase cannot be explained solely by A␤O-induced cell proliferation (Fig. 2E). Parallel treatment with A␤ monomer or fibril failed to show any increase in NO (data not shown). Besides NO, however, our initial investigation of several commonly studied neuroinflammatory mediators did not show any A␤O-induced increases above basal, sometimes undetectable, levels. These mediators include prostaglandin E2, glutamate, and proinflammatory cytokines such as TNF-␣, IL-1␤, and IL-6. Parallel microglia cultures treated with LPS showed substantially increased releases of all of these mediators (supplemental Fig. S4). These results suggest that A␤O induces a "milder" microglia activation that is significantly different from LPS-induced activation.
A␤O-stimulated Microglia Activation Depends on SRA and Can Be Blocked by Doxycycline-To identify potential targets for pharmacological blockade of A␤O-induced microglia activation, we tested several known inhibitors of microglia activation. It was previously shown that scavenger receptors, in particular, SRA, participate in the binding of fA␤ (37,38). Previously we showed that fA␤ and A␤O share cell surfacebinding conformations that can be recognized by amyloid dyes, such as CR (21). Because A␤O treatment increased microglial expression of SRA ( Fig. 2A), it is possible that SRA may also recognize the amyloid conformation of A␤O and mediate at least some of the effects of A␤O. Both the general scavenger receptor inhibitor polyinosinic acid and two neutralizing SRA antibodies, E-20 and 2F8 (37), blocked the A␤O-induced microglia activation (Fig. 3), suggesting that SRA mediates the interaction of A␤O with microglia.
Doxycycline, a second generation tetracycline, has been shown to provide neuroprotection in various models of neuronal injuries by inhibiting microglia activation (39,40). The reported mechanisms include inhibitions of iNOS, p38 MAPK, and NFB (41). We found that doxycycline was able to inhibit A␤O-induced microglia activation, including NFB activation and NO generation (Figs. 3 and 2, D and E, and supplemental Fig. S5). Doxycycline also blocked microglia activation induced by human brain A␤O (supplemental Fig.  S6). Interestingly, doxycycline suppressed the initial (earlier than 30 min) but not the later phase of p38 MAPK phosphorylation following stimulation by A␤O (Fig. 2, B and C).
Regulation of A␤O-stimulated Microglia Activation by KCa3.1-By modulating Ca 2ϩ signaling, the calcium-activated potassium channel KCa3.1 has been reported to be involved in microglia proliferation, oxidative burst, NO production, and microglia-mediated neuronal killing in rat microglia (42)(43)(44). To determine whether our mouse microglia also expressed functional KCa3.1 channels, we studied their K ϩ channel expression in the whole cell configuration of the patch clamp technique (45). Similar to previous electrophysiological studies of mouse and rat microglia cultured under various conditions (46 -49), we observed K Ca , K V , and K ir currents. The K V currents exhibited the typical biophysical properties of Kv1.3 and were not studied in detail here. We likewise did not further characterize the K ir currents, which were observed in only a subset of cells. Instead we focused on the K Ca currents, which became visible when microglia were dialyzed through the patch pipette with 1 M of free Ca 2ϩ (Fig. 4A). These currents were clearly not carried by the large conductance K Ca channel KCa1.1 (also named BK or maxi-K) because they were voltage-independent as determined by ramp pulses ranging from Ϫ100 to ϩ100 mV. The currents further proved insensitive to 100 nM of the KCa2 channel inhibitor apamin (Fig. 4B). However, most of the voltage-independent current component visible during application of voltage ramps from Ϫ120 to ϩ40 mV (Fig. 4A) was sensitive to the selective KCa3.1 blocker TRAM-34 (20). The current remaining after perfusion of 1 M TRAM-34 probably consti- The activation state of p38 MAPK was evaluated by Western blot using an antibody for its phosphorylated epitope. An antibody for p38 MAPK was used to quantify the total p38 MAPK level. The activation of p38 MAPK is represented by the band intensity of phosph-p38 MAPK (p-p38) normalized to that of total p38 MAPK (t-p38). C, quantification from three independent experiments. #, p Ͻ 0.001 compared with the 0-min control; *, p Ͻ 0.05; **, p Ͻ 0.001 compared with the A␤O group in the same time point. There was no significant difference between the 0-min control versus A␤O ϩ TRAM-34 at any time point. There was no significant difference between A␤O versus A␤O ϩ doxycycline after 30 min. D, microglia after the indicated treatment for 2 h were immunostained with an antibody for p65 of NFB to mark cells with NFB activation. Numbers of p65-immunoreactive cells/200 DAPI-labeled cells were determined. Treatment with A␤O and LPS increased the number of cells with activated NFB (n ϭ 6). *, p Ͻ 0.001 compared with control. The A␤O-induced activation was prevented by 20 M doxycycline and 1 M TRAM-34. **, p Ͻ 0.001 compared with the A␤O group. LPS-induced activation was prevented by doxycycline. #, p Ͻ 0.001 compared with the LPS group but not significantly by TRAM-34. E, nitric oxide (i.e. nitrite) production by microglia after 24 h of indicated treatment was measured in the conditioned medium and normalized by the amount of total cellular protein in each culture. Treatment with A␤O and LPS increased microglia NO production (n ϭ 4). *, p Ͻ 0.001 compared with control. The A␤O-and LPS-induced increase was prevented by 20 M doxycycline and 1 M TRAM-34 (n ϭ 4). **, p Ͻ 0.001 compared with the A␤O group. Doxycycline or TRAM-34 alone did not affect NO production. tutes a combination of Kv1.3, K ir , and another uncharacterized K current. On average, mouse microglia exhibited a KCa3.1 current density of 0.026 pA/pF (Fig. 4A).
Using TRAM-34 as a pharmacological tool, we probed the role of KCa3.1 in A␤O-induced microglia activation and neu-rotoxicity. TRAM-34 alone did not affect microglia viability, as judged by cell count and NO generation (Figs. 2E and 3). TRAM-34 blocked A␤O-induced microglia proliferation, p38 MAPK phosphorylation, NFB activation, and NO generation (Figs. 3 and 2, D-F, and supplemental Fig. S5), suggesting that the A␤O effect requires Ca 2ϩ signals maintained by KCa3.1regulated K ϩ efflux. TRAM-34 also blocked microglia activation induced by human brain A␤O (supplemental Fig. S6). Interestingly, TRAM-34 reduced A␤O-but not LPS-induced NFB activation (Fig. 2E), consistent with a previous study (44) and further indicating the divergent signaling pathways stimulated by A␤O and LPS.
A␤O Cause Indirect Neuronal Damage via Microglia-Although A␤O induced microglia activation at concentrations lower than those required for direct neurotoxicity, we asked whether this activation would result in indirect neuronal damage. To test this possibility, we activated cultured microglia using subneurotoxic concentrations of A␤O and transferred the microglia-conditioned medium (A␤O-CM) to cultured hippocampal neurons. As negative controls, we treated neurons with CM derived from microglia mock-treated with solvent (Con-CM) and medium with A␤O not conditioned by microglia. As positive controls, we treated neurons with CM derived from microglia treated with LPS (26). Fig. 5 shows that A␤O-CM significantly reduced neuronal viability in a dose-dependent manner. Con-CM also caused mild neuro-  toxicity at high concentrations, but the toxicity was much less than that exerted by A␤O-CM. In contrast, the equivalent amount of A␤O directly added to the medium without being conditioned by microglia did not reduce neuronal viability. Independent experiments using the lactate dehydrogenase release assay for cell death showed similar results (supplemental Fig. S7). This indirect neurotoxic effect of A␤O is still significant even after normalization by microglia cell number, which increased ϳ1.8-fold after A␤O treatment. The A␤O-CM-mediated neurotoxicity was not blocked by adding A11 or CR to the neuron cultures (data not shown), further supporting that it was not due to residual A␤O in CM.
Because dendritic and synaptic damage likely play a more important role than neuronal loss in the early stages of AD, we asked whether A␤O could enhance the ability of microglia to damage dendrites and synapses. We found that neurons treated with A␤O-CM showed much more robust signs of dendritic damage than those treated with Con-CM, as demonstrated by immunofluorescent stains with two dendritic markers: acetylated tubulin (Ac-TN) and microtubule associated protein 2 (MAP2) (Fig. 6, A and B). Dendrites of A␤O-CMtreated neurons were thinner and shorter, with significantly stunted arborization and frequently fragmented or "beaded" appearance (50), best illustrated by sparsely plated neurons (Fig. 6B). Treatments with A␤O-CM also substantially reduced the number of PSD95-immunoreactive puncta, which represent the PSD of excitatory synapses (Fig. 6, C and D). Western blots of neuronal extracts revealed that treatments with A␤O-CM resulted in reduced levels of Ac-TN, MAP2, PSD95, and another scaffolding protein in the postsynaptic compartment called glutamate receptor-interacting protein 1 (GRIP1) (51) but not the presynaptic marker protein synaptophysin ( Fig. 6E; for quantification, see supplemental Fig. S8).
In contrast, adding the low nanomolar amount of A␤O directly to the hippocampal neuron cultures did not cause a significant reduction of postsynaptic proteins (Fig. 6, A-C).
To confirm that the indirect neurotoxicity was a consequence of microglia activation by A␤O, we tested whether blocking A␤O-induced microglia activation by doxycycline or TRAM-34 also blocks the indirect neurotoxicity. CM from A␤O-treated microglia co-incubated with doxycycline or TRAM-34 did not alter the dendritic morphology (Fig. 6, A  and B) and did not cause significant decreases in levels of Ac-TN, MAP2, PSD95, and GRIP1 (Fig. 6, D and E). The lack of KCa3.1 expression in neurons (45,52,53) makes it unlikely that the above neuroprotective effect was due to a direct effect of residual TRAM-34 in transferred CM on neurons. Taken together, these results indicate that A␤O at subneurotoxic concentrations damage dendrites and synapses by an indirect, microglia-mediated mechanism.
A␤O Induces Neurotoxicity via Activating Microglia in Organotypic Hippocampal Slices-We further tested the A␤O effect on microglia using organotypic hippocampal slice cultures, which would better reflect the conditions in the brain in terms of microglial density and their interaction with astroglia and neurons. Treating hippocampal slices with 5-50 nM A␤O caused a significant increase in cells immunoreactive for SRA and CD11b, markers for activated microglia (Fig. 7A). A␤O treatment resulted in a transformation of microglia into an activated morphology with retracted processes and enlarged cell bodies (Fig. 7A). Microglia activation was blocked by cotreatment with A11 antibody, confirming that this effect was due to A␤O (Fig. 7A). The CM after the treatment contained increased levels of NO (Fig. 7B). A␤O treatment did not increase the release of glutamate, prostaglandin E2, TNF-␣, IL-1␤, or IL-6 (supplemental Fig. S9), consistent with our findings using dissociated microglia cultures. Parallel treatments with unaggregated and fibrillar A␤ did not induce microglia activation.
The uptake of the fluorescent dye propidium iodide (PI) is frequently used to monitor neuronal damage in slice cultures (28,54). A␤O treatment resulted in significantly increased uptake of PI by neurons of all hippocampal subfields, a pattern approximating the pattern of microglia activation, although the distribution of activated microglia was broader, extending outside the neuronal domains (Fig. 7C). By costaining the slices used for PI uptake assay for the neuronal marker NeuN, we further showed substantial NeuN immunoreactivity in nuclei that had taken up PI, indicating neuronal damage ( Fig. 8D and supplemental Fig. S11). Western blotting demonstrated that A␤O treatment also resulted in significant damage to post-synaptic elements, as demonstrated by reduced levels of Ac-TN, MAP2, PSD95, and GRIP1 but not the pre-synaptic marker synpatophysin ( Fig. 7D; for quantification, see supplemental Fig. S10).
To determine whether the neurotoxic effects of low nanomolar A␤O were indeed mediated by activation of microglia, we again tested doxycycline and TRAM-34 for their ability to block toxicity to neurons. A␤O-induced microglia activation (Fig. 7C), NO generation (Fig. 7B), PI uptake (Fig. 7C), and reduction of dendritic and post-synaptic proteins (Fig. 7D) were substantially ameliorated by co-treatment with either inhibitor. These effects were unlikely to be caused by direct actions of the compounds on neurons because neurons do not express KCa3.1 as discussed above, and doxycycline was shown to provide neuroprotection by regulating microglia but not via neuronal mechanisms (55). Although a direct antiamyloidogenic property of doxycyline was previously shown (56), our results did not support its direct effect on A␤O because we could not demonstrate any effect of doxycycline on the morphology, size, or neurotoxic activities of A␤O (data not shown). Therefore, our results indicate that in the more physiological hippocampal slice cultures, low nanomolar A␤O causes indirect neurotoxicity by activating microglia.
We also found that human A␤O in AD brain extracts induced neurotoxicity in hippocampal slices, which was substantially alleviated by co-treatment with CR, doxycycline, and TRAM-34 (supplemental Fig. S11). This result indicates that, similar to synthetic A␤O, AD A␤O at low concentrations causes indirect neurotoxicity by activating microglia.

NO Is the Major Mediator of A␤O-induced Microglial
Neurotoxicity-Because NO is the major inflammatory mediator we have found so far released by A␤O-stimulated microglia, we tested whether NO mediates neurotoxicity. We used two selective inhibitors of iNOS, L-NIL (57) and 1400W (58), and found that both compounds significantly blocked the increased NO release induced by A␤O (Fig. 8A). When A␤Otreated microglia were concurrently treated with iNOS inhibi-FIGURE 7. A␤O activated microglia and caused indirect, microglia-mediated neurotoxicity in hippocampal slices. The treatment consisted of 20 nM A␤O for 24 h. A, hippocampal slices were treated as indicated and stained with Hoechst to outline the slices and with anti-CD11b (green) and anti-SRA (red) to evaluate microglia activation. A␤O treatment caused increased staining of CD11b and SRA. A magnified image from an outlined SRA-immunoreactive area is shown on the far right to demonstrate the activated morphology of microglia. B, NO production was measured as nitrite level in the conditioned medium as described in Fig. 2E and normalized to the amount of total protein of the slice. Treatment with A␤O and LPS increased NO production from hippocampal slices (n ϭ 4). #, p Ͻ 0.001 compared with the control group. The A␤O-induced increase was prevented by 20 M doxycycline and 1 M TRAM-34 (n ϭ 4). *, p Ͻ 0.05 compared with the A␤O group. Doxycycline or TRAM-34 alone did not affect NO production. C, paired consecutive slices received the same indicated treatment. One slice was then used for PI uptake study for neuronal damage, and the other for CD11b staining for microglia activation. There was low background PI uptake and CD11b staining in the control slices. A␤O significantly enhanced PI uptake and CD11b staining, which were ameliorated by doxycycline and TRAM-34. The locations of hippocampal subfields CA1, CA3, and dentate gyrus (DG) are indicated. D, Western blot analysis of lysates from hippocampal slices with indicated treatment, analyzed by antibodies to dendritic proteins Ac-TN and MAP2, postsynaptic proteins PSD95 and GRIP1, and the presynaptic protein synaptophysin. tors, their CM were no longer neurotoxic (Fig. 8B). When applied to cultured hippocampal slices 1 h prior to A␤O application, both iNOS inhibitors reduced NO release (Fig. 8C) and inhibited neuronal damage as indicated by reduced PI uptake (Fig. 8D).

DISCUSSION
Our data show that A␤O is able to activate microglia at concentrations at least 10-fold lower than those used to induce direct neurotoxicity. Low nanomolar A␤O activates microglia to release soluble neurotoxic factors and thus indirectly damages the integrity of neurons and synapses. A␤O stimulates a unique neuroinflammatory pattern with increased NO generation but without the production of a regular panel of inflammatory mediators such as prostaglandin E2, glutamate, and the cytokines TNF-␣, IL-1␤, and IL-6. These observations were reproduced using the more physiological hippocampal slices in addition to dissociated microglia cultures. Neurotoxicity was ameliorated by a well studied inhibi-tor of microglia activation, doxycycline, the KCa3.1 blocker TRAM-34, and two iNOS inhibitors, indicating that KCa3.1 activity and excessive NO release are required for A␤O-induced microglial neurotoxicity.
We also found that soluble A␤O extracted from AD hippocampi were ϳ50 times more potent than synthetic A␤O in activating microglia, further suggesting a role of A␤O in activating microglia. The reason for the higher potency of brainderived A␤O is not known but is possibly due to the presence of co-fractionated co-stimulators or in vivo modifications of brain-derived A␤ that are not present in synthetic A␤ peptides.
The great majority of prior studies have focused on fA␤ because of the observed abundance of activated microglia in and around amyloid plaques (1,59). However, the ability of fA␤ to activate microglia is generally low or absent when fA␤ is the only stimulant; activation requires micromolar concentrations (2-100 M) of fA␤ and enhancement by co-stimulators such as ␥-interferon, lipopolysaccharides, advanced gly- FIGURE 8. NO mediates A␤O-induced microglial neurotoxicity. A, NO production was quantified as described for Fig. 2E. The A␤O-induced increase was prevented by 5 M 1400W and 100 M L-NIL (n ϭ 3). *, p Ͻ 0.01 compared with the A␤O group. B, A␤O-induced microglial neurotoxicity was evaluated as described for Fig. 5. Control or A␤O-treated microglia cultures were at the same time treated with vehicle, 1400W, or L-NIL for 24 h. Hippocampal neuron cultures were then treated with 25% CM from each microglial culture. Neuronal viability was determined after 24 h. Both 1400W and L-NIL blocked the A␤O-induced microglial neurotoxicity (n ϭ 3). *, p Ͻ 0.001 compared with the A␤O-CM/vehicle group. C, NO released by hippocampal slices was measured as described in Fig. 7B. Both 1400W and L-NIL blocked A␤O-induced NO production (n ϭ 3). *, p Ͻ 0.001 compared with the A␤O group. D, hippocampal slices were treated with the indicated conditions, and the PI uptake assay was performed as described for Fig. 7C. The same slices were then immunostained with NeuN that marked neuronal nuclei (NeuN, green; PI uptake, red). A magnified image from an outlined area in the CA1 region of the A␤Otreated hippocampal slice is shown on the right upper panel to demonstrate the substantial colocalization of NeuN immunoreactivity and PI fluorescence. Both 1400W and L-NIL blocked A␤O-induced neuronal damage as indicated by significantly reduced PI uptake. cation end products, complement factors, and macrophage colony stimulatory factor (35, 60 -65). A few reports showed that unaggregated forms of A␤ at micromolar concentrations were able to activate microglia (14,65). Tan et al. (66) showed that freshly solubilized, nonaggregated A␤42 at 500 nM activated microglia in the presence of CD40 ligand. However, it is unlikely that any soluble A␤42 preparations are devoid of oligomeric A␤ at high nanomolar concentrations. A few groups have investigated the activation of microglia by variably prepared A␤O and found that A␤O, in concentrations ranging from 2 to 50 M, was equally or less potent than fA␤ in activating microglia (13)(14)(15)(16)(17). In contrast to previous studies, we used well defined A␤O at concentrations ϳ1,000 times lower and were able to show that the potency of A␤O was significantly higher than that of fA␤ or unaggregatd A␤ in activating microglia. Because this activity has a bell-shaped concentration-response curve, which maximizes at 20 -50 nM A␤, it was not recognized in prior studies using micromolar or high nanomolar A␤. The reason for the bell-shaped response is not known. Interestingly, a study by Schubert and co-workers (30), conducted prior to our knowledge of A␤ oligomerization, described a similar bell-shaped chemotactic effect of low nanomolar soluble A␤ on microglia/macrophages. In retrospect, their preparation of soluble A␤, without prior solubilization by hexafluoroisopropanol, likely contained A␤O. In contrast, when its effect on neurons was evaluated, A␤O showed a nearly linear dose response in causing neuronal death, up to 10 M (9, 18, 22). The linear dose-response pattern of neuronal A␤O potency argues against the possibility that A␤O self-aggregates into higher order and therefore less bioactive assemblies when its concentration is above 50 nM.
Soluble nonfibrillar A␤ assemblies have been classified according to their size, structure, and immunological response to conformation-specific antibodies (8). The activities in our A␤O preparations and in AD brain lysates reported here can be attributed to the prefibrillar oligomer according to the Glabe classification, because they were immunoneutralized by the conformation-specific antiserum A11 (8). In AD brains, prefibrillar oligomers and other forms of oligomers were found to accumulate focally as clusters of immunoreactive deposits (67)(68)(69). It is highly likely that focally A␤O can reach the low nanomolar concentrations required for activating microglia in the course of AD development when A␤O progressively accumulates (3). In contrast to the relatively permanent nature of plaque-associated inflammation that is prominently detectable histologically, the microglial response to soluble A␤O may be transient in nature and resolve after successful A␤ clearance by microglia. Alternatively, A␤O-induced focal inflammation may enhance neuronal stress and further increase A␤ deposition, engendering a vicious cycle that is conducive to plaque formation (70,71). In both scenarios, neurotoxic mediators would be generated to cause indirect neuronal damage and synaptic disruption. This notion of A␤O-induced microglia activation is compatible with two recent positron emission tomography studies concluding that stimuli other than plaque fA␤ activate microglia in AD patients (4,5). Notably, within the sensitivity of positron emis-sion tomography detection, Okello et al. (5) were able to detect microglia activation independent of plaque formation in patients with mild cognitive impairment, a prodrome that with a high likelihood precedes AD. The non-fA␤ stimuli may thus include A␤O. Because levels of soluble A␤ correlate best, whereas numbers of amyloid plaque correlate poorly with the degree of clinical dementia (2,72,73), it follows that A␤Oactivated microglia might play a more significant role than plaque-associated microglia in the cognitive decline of the patient.
A␤O was also found concentrated at synaptic sites (74), where it binds postsynaptic density complexes with a high affinity (19,21,75). This provides strong evidence for direct A␤O toxicity to post-synaptic components, a possible physical basis of synaptic dysfunction in AD. Our results, however, showed that the detrimental effects of A␤O upon synapses were ameliorated by inhibitors of microglia activation and therefore support an alternative, microglia-mediated mechanism of synaptic dysfunction. We showed that A␤O at low nanomolar concentrations activated microglia and caused reduced levels of critical dendritic and postsynaptic proteins in both dissociated neuronal cultures and hippocampal slices. We also found a pattern of preferential postsysynaptic damage mediated by A␤O-activated microglia; this pattern is similar to those found in Tg2576 transgenic mice (76) and in human AD cortex (77), suggesting a pathological relevance.
Recently it was found that microglia processes make regular direct contact with synapses and that prolonged contact increases the turnover of synapses (78). Therefore we hypothesize that synaptic A␤O might attract microglia by its potent chemotactic effect (30) and promote a synapse-centered neuroinflammatory reaction to damages synapses. Supporting this notion, a previous study showed that the inhibition of NMDA receptor-dependent long term potentiation by soluble A␤ (500 nM, which might contain oligomers) can be prevented by minocycline, a microglia activation inhibitor in the same class as doxycycline, and iNOS inhibition to reduce NO production from microglia (79). Interestingly, NO is the only A␤O-promoted mediator we have so far uncovered and mediates A␤O-induced microglial neurotoxicity.
One limitation of our experimental paradigm is that although cultures derived from early postnatal brains are currently the standard tool for functional and molecular studies of microglia, the microglia may differ from their adult counterparts in vivo. One alternative approach is to isolate microglia from adult brains without culturing, which may generate microglia samples closer in phenotype to adult microglia in vivo (80,81). However, this approach has the drawbacks of low yields, limited applications, and potential cell sortinginduced activation. Therefore, based on the current findings, future in vivo experiments such as those utilizing the KCa3.1 knock-out mice and "A␤ oligomer" mice (82) are needed to fully understand the significance of A␤O-induced microglia activation in AD.
Microglia are promising cellular targets for controlling harmful CNS inflammation in a wide variety of neurodegenerative disorders (83). The unique mechanism utilized by A␤O to activate microglia as well as its potential impact on the development of AD provides an excellent possibility for novel therapeutic approaches. One example we show here is TRAM-34, a small molecule that selectively blocks the intermediate conductance calcium-activated potassium channel KCa3.1 (20,84,85). Our data provide the first evidence that a specific K ϩ channel regulates A␤-induced microglia activation and neurotoxicity. KCa3.1 regulates Ca 2ϩ signaling by maintaining a negative membrane potential through K ϩ efflux, thus facilitating Ca 2ϩ entry through a calcium-release activated Ca 2ϩ channel, a channel responsible for the storeoperated Ca 2ϩ entry required for microglia activation (44,86). The anti-inflammatory and neuroprotective properties of KCa3.1 blockers have been shown in models of traumatic brain injury (87), multiple sclerosis (88), and retinal ganglion cell degeneration after optic nerve transection (44). In addition, TRAM-34, although inhibiting microglia-mediated neurotoxicity, does not affect the beneficial activities of microglia such as migration and phagocytosis (44). Our results suggest that specific KCa3.1 blockers, targeting microglia selectively because of the microglia-restrictive cellular expression of KCa3.1 in the CNS, might provide a novel anti-inflammatory strategy for AD.