Mitochondrial Ca2+ Overload Underlies Aβ Oligomers Neurotoxicity Providing an Unexpected Mechanism of Neuroprotection by NSAIDs

Dysregulation of intracellular Ca2+ homeostasis may underlie amyloid β peptide (Aβ) toxicity in Alzheimer's Disease (AD) but the mechanism is unknown. In search for this mechanism we found that Aβ1–42 oligomers, the assembly state correlating best with cognitive decline in AD, but not Aβ fibrils, induce a massive entry of Ca2+ in neurons and promote mitochondrial Ca2+ overload as shown by bioluminescence imaging of targeted aequorin in individual neurons. Aβ oligomers induce also mitochondrial permeability transition, cytochrome c release, apoptosis and cell death. Mitochondrial depolarization prevents mitochondrial Ca2+ overload, cytochrome c release and cell death. In addition, we found that a series of non-steroidal anti-inflammatory drugs (NSAIDs) including salicylate, sulindac sulfide, indomethacin, ibuprofen and R-flurbiprofen depolarize mitochondria and inhibit mitochondrial Ca2+ overload, cytochrome c release and cell death induced by Aβ oligomers. Our results indicate that i) mitochondrial Ca2+ overload underlies the neurotoxicity induced by Aβ oligomers and ii) inhibition of mitochondrial Ca2+ overload provides a novel mechanism of neuroprotection by NSAIDs against Aβ oligomers and AD.


Introduction
Alzheimer's Disease (AD) is a devastating neurodegenerative disorder due to a massive neuron dysfunction and loss related to development of senile plaques that are made of amyloid b peptide (Ab), a cleavage product of the amyloid precursor protein. Early affected areas in the brain are the cortex and hippocampus. However, neuropathological studies show frequent and varied cerebellar changes in the late stages of the disease [1]. In fact, the cerebellum has been shown to be a unique organ in terms of AD manifestations because it is virtually free of neurofibrillary pathology but there is an strong correlation between cerebellar atrophy -with large cell death in the granular layer-and duration and stage of AD [1]. Although many in vitro studies have been carried out using cortical and hippocampal neurons, cerebellar granule cells have been also used for studies of Ab neurotoxicity [2][3][4]. Several mechanisms have been envisioned. First, the ''inflammatory hypothesis'' proposes that Ab may promote a damaging inflammation reaction. This view is supported among other evidence by the neuroprotection afforded by NSAIDs [5]. Second, Ab promotes mitochondrial dysfunction and apoptosis and this toxicity contributes to AD [6] but the mechanism is unclear. Ab may associate with mitochondrial membranes in mutant mice and patients with AD and mitochondria from mutant mice show lower levels of oxygen consumption and reduced respiratory complex-associated enzymatic activity suggesting that mitochondria-bound Ab may impact on mitochondrial activity [7][8][9]. Finally, AD has been also related to a general dyshomeostasis of intracellular Ca 2+ , a key second messenger involved in multiple neuronal functions. This view is supported by reports on dysregulation of intracellular Ca 2+ promoted by Ab and mutant presenilins [10]. Ab may promote Ca 2+ entry into neurons but results are controversial [11,12]. Part of the controversy may relate to the fact that Ab toxicity depends on its assembly state that varies from monomers to small, soluble oligomers and fibrils [13]. Small assemblies (oligomers) of unmodified Ab are becoming the proximate neurotoxin in AD [13,14], but most studies used fibrils. Intracellular Ca 2+ levels are important for AD since overexpression of calbindin28k, an endogenous Ca 2+ buffer, prevents neuron death in AD models [15]. However the link between putative changes in intracellular Ca 2+ and cell damage is unknown. A rise in mitochondrial Ca 2+ concentration ([Ca 2+ ] mit ) might contribute to neurotoxicity but monitoring [Ca 2+ ] mit in individual neurons has been challenging. We have addressed the effects of Ab assembly state on Ca 2+ influx and mitochondrial Ca 2+ uptake using photon counting imaging of neurons expressing targeted aequorin [16]. We found that only oligomers, but not fibrils, increased cytosolic and mitochondrial Ca 2+ concentrations. Accordingly we asked for the role of mitochondrial Ca 2+ uptake on neurotoxicity induced by Ab oligomers. Finally, we tested whether NSAIDs may protect against Ab toxicity acting on subcellular Ca 2+ fluxes. For these studies we have used mainly cerebellar granule cells although some experiments have been also carried out in cortical and hippocampal neurons.

Ab oligomers but not fibrils induce entry of Ca 2+ into neurons
We have used the protocol reported by Klein [17] to prepare oligomers and fibrils from Ab 1-42 obtained from a commercial source (Bachem). Since the standard protocol of preparation includes a precipitation step in which some protein sample is lost, we hydrolyzed an aliquot of the final solution of both oligomers and fibrils in order to carry out an amino acid analysis. This procedure allowed us to obtain the real concentration of these compounds in solution. In the second place we characterized the quaternary structure (dimers, trimers, tetramers, etc.) of both the oligomers and fibrils using non-denaturing SDS-PAGE (pseudonative gels). We were unable to stain the Aß 1-42 peptides using Coomassie blue even when 2 mg of peptide were loaded per lane (data not shown). However, using silver staining we were able to determine the presence of high-molecular mass species in the SDS-PAGE. We rationalized that despite the fact that the samples were not boiled, a significant population of protein-protein interactions might be lost in the presence of SDS. However, were able to clearly identify monomers, dimers and tetramers in our preparation of oligomers ( Figure S1A). This migration pattern of Aß  oligomers in SDS-PAGE gels is well characterized [18]. When the preparation of fibrils was analyzed by SDS-GEL and silver staining, we could unambiguously identify the presence of monomers, dimers, trimers, tetramers and some larger oligomerization species in the gel. In addition, a certain amount of large molecular weight fibrils appeared at the top, incapable of entering the separating gel ( Figure S1B). Similar results were obtained when the gel was transferred to a nitrocellulose membrane and the distribution of high-molecular mass species was determined by Western-blot using a monoclonal antibody raised against Aß  (data not shown). In the third place, electron microscopy was used in order to characterize our Aß 1-42 fibrils. Negative staining using uranyl acetate undoubtedly showed the presence of large fibrils in solution ( Figure S1C). Most of these fibrils were similar in width and with a length that usually varied between 200 and 800 nm.

Mitochondrial Ca 2+ overload contributes to Ab-induced apoptosis
We asked next for the contribution of mitochondrial Ca 2+ overload to the apoptosis induced by Ab 1-42 oligomers and Ab [25][26][27][28][29][30][31][32][33][34][35] . For this end we have studied whether inhibition of mitochondrial Ca 2+ uptake prevents Ab oligomers-induced apoptosis or not. Ca 2+ uptake by mitochondria depends exponentially on DY, a huge driving force of 2180 mV built upon the respiratory chain. We have reported recently that a small mitochondrial depolarization is enough to prevent largely mitochondrial Ca 2+ uptake [29,30]. Accordingly, we studied the effects of a small mitochondrial depolarization using low concentrations of the mitochondrial uncoupler FCCP. To assess the effects of this treatment on DY we used TMRM, a cationic dye that accumulates in mitochondria according to DY and is considered one of the most sensitive DY probes available [31]. Confocal images of cerebellar granule cells co-stained with mitotracker green and TMRM show the mitochondrial localization of the probe (Fig. 5A). Addition of FCCP decreases TMRM fluorescence in a dose-dependent manner consistent with a mitochondrial depolarization (Fig. 5B). At 100 nM, FCCP decreases TMRM fluorescence by about 25% relative to the total fluorescence decrease induced by 10 mM FCCP. According to our previous report [30], we estimate that such a decrease in TMRM fluorescence induced by 100 nM FCCP corresponds to a loss of DY of about 10-20 mV; whereas at 10 mM, FCCP decreases fully TMRM fluorescence consistently with a collapse of DY (Fig. 5C). Similar results were found in GT1 neural cells (data not shown). Strong plasma membrane depolarization with high K + medium (50 mM) did not affect TMRM fluorescence excluding the possibility that TMRM is reporting changes in the plasma membrane potential (data not shown). Next, we studied the effects of such an small mitochondrial depolarization on mitochondrial Ca 2+ uptake. We found that FCCP, at 100 nM, prevents the increase in [Ca 2+ ] mit induced by Ab 1-42 oligomers (Fig. 5D). Specifically, 100 nM FCCP inhibits by 7864% the [Ca 2+ ] mit increase induced by Ab 1-42 oligomers (n = 21 cells, 3 experiments). FCCP also inhibited the rise in [Ca 2+ ] mit induced by Ab [25][26][27][28][29][30][31][32][33][34][35] ( Figure S3A). Specifically, 100 nM FCCP inhibits by 8265% the [Ca 2+ ] mit increase induced by Ab 25-35 (17 cells studied, 3 experiments). This effect is not due to inhibition of Ca 2+ entry through the plasma membrane since 100 nM FCCP does not prevent at all the increase in [Ca 2+ ] cyt induced by Ab 1-42 oligomers (Fig. 5E) or Ab 25-35 ( Figure S3B).
It has been reported that mitochondrial Ca 2+ overload during excitotoxicity may promote reactive oxygen species (ROS) production and contribute to neuron cell damage [32]. Accordingly, we also studied the effects of Ab 1-42 oligomers on ROS production in cerebellar granule cells. Incubation of cerebellar granule cells with 500 nM Ab 1-42 oligomers for 4 h elicited a clear increase in number of neurons strongly stained by the ROS probe CM-H2DCFDA. This effect was absent in cells incubated previously with 100 nM FCCP added here to prevent the Ab 1- Figure S4). These results suggest that mitochondrial Ca 2+ overload may contribute to ROS production induced by Ab 1-42 oligomers.
These results indicate that NSAIDs, at low mM concentrations, inhibit specifically mitochondrial Ca 2+ uptake without preventing entry of Ca 2+ through the plasma membrane and this effect is most likely mediated by a partial mitochondrial depolarization. We recently adapted an algorithm to convert TMR fluorescence values into estimated DY expressed in mV [30]. According to that algorithm, we estimate that the loss of DY induced by the low FCCP and NSAIDs concentrations used here is only 10-20 mV. This very small mitochondrial depolarization should not compromise cell metabolism as suggested by the lack of effects of these treatments on cell ATP levels.

Discussion
We show here that Ab 1-42 oligomers, the assembly state that correlates best with brain damage and cognitive deficits in AD [13,14], but not Ab fibrils, promote massive entry of Ca 2+ into GT1 neural cells and cerebellar granule neurons but not glia. This view is supported by the finding that cells responding to Ab 1-42 oligomers also respond to NMDA and by immunocytochemical identification of responsive cells. Ab 1-42 oligomers, but not fibrils, also induced large increases in [Ca 2+ ] cyt in cortical and hippocampal neurons. These results agree with those recently reported [34] showing that Ab 1-42 oligomers but not monomers or fibrils increased [Ca 2+ ] cyt in a human neuroblastoma cell line. The results suggest that the mechanism of neurotoxicity by fibrils and oligomers may be different as previously proposed [21]. The effects of Ab 1-42 oligomers are quite well reproduced by the toxic fragment Ab 25-35 although at 40-fold larger concentrations. The increases in [Ca 2+ ] cyt induced by Ab 1-42 oligomers and the toxic fragment Ab [25][26][27][28][29][30][31][32][33][34][35] can be attributed to enhanced entry of extracellular Ca 2+ through the plasma membrane rather than release from intracellular Ca 2+ stores as they were entirely prevented by removal of extracellular Ca 2+ . The route for this enhanced Ca 2+ influx is not solved yet but candidate mechanisms may include plasma membrane permeabilization [34], formation of the so-called amyloid channels [35,36] and/or activation of NMDA receptors [37]. Neither Ab [25][26][27][28][29][30][31][32][33][34][35] or Ab 1-42 oligomers induced parallel decreases in fura2 fluorescence excited at 340 and 380 nm (data not shown) suggesting that the rises in [Ca 2+ ] cyt are not due to membrane permeabilization. It has been shown previously that the increases in [Ca 2+ ] cyt induced by Ab species can be prevented by the NMDA receptor open channel blocker memantine [37] and specific amyloid channel blockers [36] suggesting that both NMDA receptors and amyloid channels might be involved. In any case, the massive entry of Ca 2+ induced by Ab 1-42 oligomers promotes mitochondrial Ca 2+ overload as shown directly by bioluminescence imaging of neurons expressing a low-affinity aequorin targeted to mitochondria. This probe was originally developed to monitor the high [Ca 2+ ] inside the endoplasmic reticulum that reach the mM range in resting conditions [24]. Using this probe, we and others have shown previously that [Ca 2+ ] mit may reach several hundred mM upon stimulation of voltage-gated Ca 2+ entry and Ca 2+ release from intracellular stores [16,23,24]. However, the above mentioned increases in [Ca 2+ ] mit were transient and restricted to a pool of mitochondria close to sites of Ca 2+ entry or release [23,24]. The recent introduction of low-affinity, targeted chameleons [38] has confirmed that [Ca 2+ ] mit may reach values above 200 mM, the sensitivity limit of these probes, stressing that [Ca 2+ ] mit levels actually rise substantially and the requirement of low-affinity probes for actual measurements of [Ca 2+ ] mit , particularly when a mitochondrial Ca 2+ overload is to be measured. Using the lowaffinity, mitochondria-targeted aequorin, we find that Ab 25-35 and Ab 1-42 oligomers, but not fibrils, induce a massive mitochondrial Ca 2+ overload that reaches values close to the mM level. As aequorin is consumed by the high Ca 2+ level achieved, the results suggest that, at variance with stimulation with high K + , most mitochondria take up Ca 2+ when cells are stimulated by Ab 1-42 oligomers. As mitochondrial Ca 2+ uptake through the mitochondrial Ca 2+ uniporter requires high [Ca 2+ ] cyt levels, these results suggest that Ab 1-42 oligomers promote a massive entry of Ca 2+ , large and sustained enough to activate the mitochondrial Ca 2+ uniporter of most mitochondria.
It has been reported that mitochondrial Ca 2+ overload may promote mPTP opening and apoptotic cell death [26,39]. Our results indicate that mitochondrial Ca 2+ overload contributes to the apoptotic cell death induced by Ab oligomers. This view is supported by the findings that Ab 1-42 oligomers promote i)a mitochondrial Ca 2+ overload in the whole mitochondrial population, ii)mitochondrial calcein quenching in a cyclosporin sensitive manner, iii)release of cytochrome c, iv)apoptosis as determined by TUNEL assay and v)cell death, again in a cyclosporin A-sensitive manner. Furthermore, inhibition of mitochondrial Ca 2+ uptake by low concentrations of FCCP inhibit both cytochrome c release and cell death without preventing Ab 1-42 oligomers-induced increases in [Ca 2+ ] cyt or decreasing cell ATP levels. Finally, Ab 1-42 oligomers induce ROS production and this effect is prevented by low concentrations of FCCP. Taken together, our results suggest that the large and sustained entry of Ca 2+ induced by Ab 1-42 oligomers, an effect mimicked by larger concentrations of the toxic fragment Ab [25][26][27][28][29][30][31][32][33][34][35] , activate the mitochondrial Ca 2+ uniporter of most mitochondria leading to a mitochondrial Ca 2+ overload. This effect may promote mPTP opening by itself or, cooperate with the excess of ROS production promoted by the own mitochondrial Ca 2+ overload. Finally, mPTP opening allows release of pro-apoptotic factors including cytochrome c leading to apoptosis and cell death (see proposed model in Fig. 9). This view resembles the mechanism of excitotoxicity reported for glutamate. In fact, glutamate-induced neuron death requires mitochondrial Ca 2+ uptake [40]. In addition, low concentrations of FCCP have been reported to prevent mitochondrial Ca 2+ uptake and cell death induced by NMDA [41]. Finally, glutamate induces also ROS production and this effect is prevented by mitochondrial uncoupling and blockers of the mitochondrial Ca 2+ uniporter [32]. Whereas this mechanism may contribute largely to cell death induced by Ab oligomers, our results do not exclude that additional mechanisms might contribute to this toxicity. For example, it has been reported that intracellular Ab species may also interact with mitochondria in AD mouse models and affected AD brains [7][8][9] and this interaction may promote apoptosis and cell death. It remains to be established whether mitochondrial alterations induced by mitochondria-bound Ab species cooperate with mitochondrial Ca 2+ overload to promote cell death.
The above findings indicate that mitochondrial Ca 2+ overload contributes to the neurotoxicity induced by Ab 1-42 oligomers. Accordingly, any strategy intended to prevent specifically mitochondrial Ca 2+ uptake could potentially protect neurons against Ab 1-42 oligomers toxicity. Among the most promising agents for neuroprotection against AD are a series of classic NSAIDs [5,42,43]. Recent evidence indicates that neuroprotection afforded by NSAIDs is due to a mechanism other than their antiinflammatory activity. This view is based in that some, but not all NSAIDs show neuroprotection and that structural analogs of classic NSAIDs lacking anti-inflammatory activity like, for instance, R-flurbiprofen, offer protection as well [43]. Some NSAIDs including R-flurbiprofen have been reported to target and inhibit c-secretase activity at rather large concentrations (100 mM) leading to a lower Ab burden [44,45]. We show here that, at very low concentrations (1 mM), NSAIDs depolarize mitochondria to the same extent than low concentrations of FCCP, inhibit mitochondrial Ca 2+ overload induced by Ab oligomers without preventing Ca 2+ entry through the plasma membrane, prevent ROS production induced by Ab 1-42 oligomers and inhibit Ab oligomers-induced cytochrome c release and cell death. Accordingly we propose a novel mechanism of neuroprotection against soluble Ab 1-42 species by non-specific NSAIDs based on the primary inhibition of mitochondrial Ca 2+ overload and prevention of the ensuing mPTP opening an downstream steps to cell death (Fig. 9). Notice that, at these low concentrations, NSAIDs have little or no effect on c-secretase activity whereas fitting the concentration range achieved in brains by human therapeutic dose [45]. We must stress that these results have been obtained mainly in cerebellar granule cells, a brain area generally considered not affected earlier in AD. However, as mentioned above, neuropathological studies have shown frequent and varied cerebellar changes in the late stages of the disease [1]. Nevertheless, we show that Ab 1-42 oligomers also induced large increases in [Ca 2+ ] cyt and [Ca 2+ ] mit in cortical neurons as well as rises in [Ca 2+ ] cyt in hippocampal neurons.
The mechanism of inhibition of mitochondrial Ca 2+ uptake by NSAIDs is most likely mediated by mitochondrial depolarization. This view is supported by i)the chemical structure of NSAIDs resembling mitochondrial uncouplers, ii)the release of Ca 2+ induced by salicylate in Ca 2+ -overloaded cells and iii)direct TMRM fluorescence measurements showing NSAID-induced mitochondrial depolarization. In addition, uncoupling activity is considered a common characteristic of anti-inflammatory agents with an ionizable group [46]. It may seem surprising that a small mitochondrial depolarization be enough to inhibit largely mitochondrial Ca 2+ uptake. However, several facts support this view. First the Ca 2+ channel associated to the mitochondrial Ca 2+ uniporter is inwardly rectifying, making it especially effective for Ca 2+ uptake into energized mitochondria [47]. In this scenario, a mild mitochondrial depolarization will decrease largely the Ca 2+ current through the mitochondrial Ca 2+ uniporter. Second, the Nernst's equilibrium for Ca 2+ across mitochondria predicts that a 50% fall in DY reduces the free mitochondrial [Ca 2+ ] that can be reached at equilibrium by 1,000 fold [23]. This prediction is based on the huge DY built in respiring mitochondria and the fact that Ca 2+ is a divalent cation present at extremely low concentrations (100 nM) in resting mitochondria. Thus, it is thermodynamically possible that even small drops in DY could influence dramatically the [Ca 2+ ] mit increase achieved during cell stimulation. In support of this view, we have shown recently that small drops in DY of a few tens of mVs are enough to prevent largely mitochondrial Ca 2+ uptake [30].
Interestingly, other AD-related factors seemingly independent of Ab species may contribute as well to mitochondrial Ca 2+ overload and, perhaps, to neurotoxicity. For example, in familial AD, loss of function presenilin 1 mutants show exaggerated Ca 2+ release from intracellular stores due to either defective Ca 2+ leak from the ER or increased activity of Ca 2+ release channels at ER [48-50, but see also 51 for alternative results]. The close coupling between ER and mitochondria [52] may favor mitochondrial Ca 2+ overload in patients carrying these mutations. Further research is required to support this view. If that were the case, the possible PS1 mutationmediated mitochondrial Ca 2+ overload and ensuing neuron damage should be limited by NSAIDs. An even more important factor, at least from the point of view of the potential number of patients affected, could be the reported depletion of endogenous Ca 2+ buffers, particularly calbindinD28k, in selected brain regions during aging and sporadic AD [53,54]. A diminished cytosolic Ca 2+ buffer capacity may place mitochondria at risk of Ca 2+ overload during normal neural activity, a process that should be ameliorated by NSAIDs. Therefore, our results point to a pivotal role of mitochondrial Ca 2+ overload in Ab 1-42 oligomers toxicity and AD. Interestingly, the mitochondrial Ca 2+ overloads likely induced by AD related processes such as Ab oligomers (shown here), excitotoxicity [40], excess Ca 2+ release from ER [49] and loss of endogenous Ca 2+ buffers [53,54] could be all ameliorated by low NSAID concentrations, regardless of the source of Ca 2+ excess being either intracellular or extracellular.

Materials
Wistar rats were obtained from the Valladolid University animal facility. Fura2/AM, TMRM, DAPI, CM-H2DCFDA, n coelenterazine, the cytochrome c antibody, Alexa 488 a mouse IgG and Alexa F594 a rabbit IgG were purchased from Invitrogen (Barcelona, Spain). DMEM (ref. 41966-029), fetal bovine serum, horse serum, neurobasal medium, B27, penicillin and streptomycin are from Gibco (Barcelona, Spain). Papain solution is from Worthington (Lakewood, NJ). The kit for TUNEL assay is from Roche Diagnostics (Penzberg, Germany). Ab peptides were purchased from Bachem AG (Bubendorf, Switzerland). The mouse anti b-tubulin III is from Covance (Princeton, USA) and the rabbit anti-GFAP is from Sigma (Madrid, Spain). The mGA plasmid was kindly donated by P. Brulet (Gif-sur-Yvette, France). Hypothalamic GT1 neural cells were provided by R. Weiner (San Francisco, USA). Other reagents and chemicals were obtained either from Sigma or Merck. Figure 9. A model of Ab-induced toxicity and neuroprotection by NSAIDs based on mitochondrial Ca 2+ . Ab 1-42 oligomers and the toxic fragment Ab 25-35 induce a large entry of Ca 2+ through the plasma membrane likely mediated by formation of amyloid channels and/or NMDA receptors. This entry promotes in a sequential manner mitochondrial Ca 2+ overload, ROS production, mPTP opening, cytochrome c release, apoptosis and cell death. Other factors related to AD may favor mitochondrial Ca 2+ overload including exaggerated IP 3induced release of Ca 2+ from the ER in loss of function PS1 mutants related to familial AD and/or decreased abundance of endogenous Ca 2+ buffers as calbindinD28k during aging or sporadic AD. NSAIDs, at low concentrations (1 mM), depolarize partially mitochondria and inhibit mitochondrial Ca 2+ overload, thus preventing cytochrome c release and apoptosis induced by Ab oligomers. doi:10.1371/journal.pone.0002718.g009

Cell Cultures
Cerebellar granule cells were obtained from 5-day old Wistar rat pups killed by dislocation followed by decapitation as reported previously [55]. Cortical neurons were obtained from P0 Wistar rat pups following the protocol reported by Murphy et al. [56]. Hippocampal neurons were prepared from P0 Wistar rat pups as reported by Brewer et al. [57] with the modifications introduced by Perez-Otano et al. [58]. Briefly, after brain removal, meninges were discarded and the hippocampus was separated from cortex. Hippocampal tissue was cut in small pieces, transferred to papain solution and incubated at 37u C for 30 minutes with occasional gentle shaking. Tissue pieces were washed with neurobasal medium and dissociated into a single cell suspension. Hippocampal cells were plated onto collagen-coated, 12 mm diameter glass coverslips at 40610 3 cells/dish, and grown in Neurobasal Medium supplemented with B27 and 10% horse serum. Cells were cultured for 7-10 days before experiments. Cerebellar granule cells and cortical neurons were plated on poly-L-lysine coated, 12 mm diameter glass coverslips and cultured in high-glucose, low K + , Dulbeccos modified Eagles medium (DMEM, ref. 41966-029. Gibco, Spain) plus 10% fetal bovine serum, 5% horse serum, 100 u ml 21 penicillin and 100 mg ml 21 streptomycin for 2 days. Then the culture medium was replaced by Satos medium [59] plus 5% horse serum to avoid excessive proliferation of glia and cultured for 2-4 (cerebellar) or 3-5 (cortical) days before experiments. GT1 neural cells were grown in DMEM with 10% fetal bovine serum, 5% horse serum, 100 u ml 21 penicillin and 100 mg ml 21 streptomycin.

Preparation of Ab oligomers and fibrils
Ab 1-42 oligomers and fibrils were prepared as reported previously by Klein and Dahlgren et al. [17,20]. Briefly, Ab 1-42 was initially dissolved to 1 mM in hexafluoroisopropanol and separated into aliquots in sterile microcentrifuge tubes. Hexafluoroisopropanol was removed under vacuum in a speed vac., and the peptide film was stored desiccated at 220uC. For the aggregation protocol, the peptide was first resuspended in dry dimethyl sulfoxide to a concentration of 5 mM and treated differently for preparation of oligomers and fibrils. For preparation of oligomers, Hams F-12 was added to bring the peptide to a final concentration of 100 mM and incubated at 4uC for 24 h. The preparation was then centrifuged at 14,0006g for 10 min at 4uC to remove insoluble aggregates (protofibrils and fibrils) and the supernatants containing soluble Ab 1-42 oligomers were transferred to clean tubes and stored at 4uC. For preparation of fibrils, the peptide resuspended in dimethyl sulfoxide at 5 mM concentration was diluted in 10 mM HCl to bring the peptide to a final concentration of 100 mM and incubated at 37uC for 24 h. Actual concentrations of both oligomers and fibrils were measured by amino acid analysis. Ab [25][26][27][28][29][30][31][32][33][34][35] was solved in PBS at a concentration of 1 mM, sonicated 3 times and stored at 220uC. For experiments, Ab [25][26][27][28][29][30][31][32][33][34][35] was solved in medium at a final concentration of 20 mM.

Amino Acid analysis
Samples of both Aß 1-42 oligomers and fibrils in glass tubes were extensively vacuum dried in a speed-vac. Hydrolysis was performed using a 5.9 HCl solution containing 0.1% phenol. Norleucine was added to the hydrolytic solution as an internal standard. The tubes were sealed under vacuum and the hydrolysis was performed during 24 h at 110uC. Subsequently, the samples were vacuum dried. Finally, the samples were injected into a Biochrom 30 amino acid analyzer. Both composition and concentration data was obtained from the chromatogram.

PAGE-SDS and silver staining
A standard 17% PAGE-SDS was prepared and 10 ml samples of either Ab 1-42 oligomers or fibrils at a 40 mM concentration were incubated with 1X loading buffer. The samples were loaded in the gel without boiling and the gel was run at constant amperage of 20 mA. The gels were subsequently fixed with a freshly prepared solution of 50% methanol, 12% acetic acid, 0.02% paraformaldehyde for at least 1h. Then, the gel was washed three times with 50% ethanol. The next step consisted on an incubation of the gel for 1 min in 0.2 mg/ml NaS 2 O 3 . Subsequently, the gel was washed three times with miliQ water for 20 seconds. Then, the gel was incubated for 20 minutes in 2 mg/ml AgNO 3 in 0.025% formaldehyde. After two 20-second washes with miliQ water, color was allowed to develop using a solution consisting of 60 mg/ml Na 2 CO 3 , 0.02% formaldehyde, 0.004 mg/ml NaS 2 O 3 . After the bands appear, the gel was washed twice with miliQ water and the reaction was stopped using a 50% methanol, 12% acetic acid solution.

Electron microscopy
A 10 ml sample of Aß 1-42 fibrils (40 mM) was applied to a 200 mesh Formvar-coated copper grid and incubated for 15 minutes at room temperature. The sample was then wicked off with filter paper, washed briefly by placing the grid face down on a droplet of water, and stained by transferring the grid face down on a droplet of 2% uranyl acetate for 2 minutes before wicking off the solution and air drying. Grids were visualized in a JEOL transmission electron microscope.

Fluorescence imaging of [Ca 2+ ] cyt and in situ immunofluorescence
Coverslips containing cells were incubated in standard medium containing (in mM) NaCl 145, KCl 5, CaCl 2 1, MgCl 2 1, glucose 10 and Hepes 10 (pH, 7.42) and loaded with 4 mM fura2/AM or fura4F/AM for 60 min at room temperature. Then coverslips were placed on the heated stage of an inverted microscope (Nikon Diaphot), perfused continuously with the same prewarmed standard medium containing and epi-illuminated alternately at 340 and 380 nm. Light emitted above 520 nm was recorded with a Magical Image Processor (Applied Imaging). Pixel by pixel ratios of consecutive frames were captured and [Ca 2+ ] cyt was estimated from these ratios as previously reported [16]. For differential identification of neurons and glia, the single cell content of btubulin III and glial fibrillary acidic protein (GFAP) were assessed by indirect inmunofluorescence in the same cells used for calcium imaging as reported previously [55]. Briefly, after calcium imaging, cells were fixed with p-formaldehyde and incubated with anti b tubulin III (1:400) and anti GFAP (1:200) for 1 h at 37uC. Then, cells were washed and incubated with 1:100 labeled anti IgG antibodies. Nuclei were stained by incubation with DAPI 0,2 mg/ml for 5 min.

Bioluminescence imaging of [Ca 2+ ] mit
Cells were transfected with the mGA plasmid using a Nucleofector II H device and the VPG-1003 transfection kit (Amaxa Biosystems, Cologne, Germany). The mGA probe contain a mutated, low affinity aequorin targeted to mitochondria and a GFP sequence to help select transfected neurons [25]. After 24 h, cells were incubated for 1 h with 1 mM n coelenterazine, washed and placed into a perfusion chamber thermostated to 37 u C under a Zeiss Axiovert S100 TV microscope and perfused at 5-10 ml/ min with test solutions based on the standard perfusing solution described above prewarmed at 37uC. At the end of each experiment, cells were permeabilized with 0.1 mM digitonin in 10 mM CaCl 2 to release all the residual aequorin counts. Bioluminescence images were taken with a Hamamatsu VIM photon counting camera handled with an Argus-20 image processor. Photonic emissions were integrated for 10 s periods. Photons/cell in each image were quantified using the Hamamatsu Aquacosmos software. Total counts per cell ranged between 2?10 3 and 2?10 5 and noise was (mean6SD) 161 counts per second (c.p.s.) per typical cell area (2000 pixels). Data were first quantified as rates of photoluminescence emission/total c.p.s remaining at each time (% of remaining counts) and divided by the integration period (L/L TOTAL in s 21 ). Emission values of less than 4 c.p.s were not used for calculations. Calibrations in terms of [Ca 2+ ] mit were performed using as reported previously [23]. In some experiments, cells were permeabilized with digitonin 20 mM in ''intracellular'' medium of the following composition 130 mM KCl, 10 mM NaCl, 1 mM MgCl 2 , 1 mM K 3 PO 4 , 0,2 mM EGTA, 1 mM ATP, 20 mM ADP, 2 mM succinate, 20 mM HEPES/KOH, pH, 6.8. Then, the cells were incubated with the same medium containing 200 nM Ca 2+ (buffered with EGTA) with or without NSAIDs for 5 min. Then, perfusion was switched to ''intracellular'' medium containing 5 mM Ca 2+ (with or without NSAID). Further details have been reported previously [16,23,25].
Mitochondrial permeability transition pore (mPTP) mPTP opening was assessed directly by the calcein/cobalt method [28]. Cells were co-loaded with calcein-AM 1 mM and CoCl 2 1 mM for 30 min at 37uC and subjected to conventional fluorescence imaging or confocal microcopy. Fluorescence traces from individual cells were normalized relative to the value before addition of test solutions and averaged. Background fluorescence corresponding to regions of interest devoid of cells was subtracted.
In some experiments, cells were incubated with cyclosporin A 1 mM for 15 min before recordings.

ROS Formation
ROS formation was evaluated in live neurons using CM-H2DCFDA as reported by De Felice et al. [37]. Cerebellar granule cell cultures were incubated for 4 h at 37uC with vehicle or 500 nM Ab 1-42 in the absence or in the presence of FCCP 100 nM and R-flurbiprofen 1 mM. ROS formation was assessed using 2 mM CM-H2DCFDA with 40 min of probe loading. Then neurons were superfused for 5 min with prewarmed (37uC) PBS. Fluorescence in cells was immediately visualized using the Zeiss S100 TV inverted microscope, a FITC filter set, an OrcaER digital camera from Hamamatsu and the Hamamatsu Aquacosmos software.

Mitochondrial potential (DY)
The effects of treatments on DY were estimated by fluorescence imaging in cells loaded with the DY sensitive probe TMRM as reported previously [29][30][31]. Briefly, cells were loaded with TMRM (10 nM) for 30 min at room temperature, placed on the perfusion chamber of a Zeiss Axiovert S100 TV inverted microscope and superfused continuously with the prewarmed (37uC) standard medium described above. Fluorescence images were taken at 10 s intervals with a Hamamatsu VIM photon counting camera handled with an Argus-20 image processor. Traces from individual cells were normalized relative to the value before the addition of either vehicle or treatment and averaged. Background fluorescence -after collapse of the mitochondrial potential induced by 10 mM FCCP-was subtracted. Location of TMRM staining and fluorescence intensity ratios of TMRM in mitochondrial and cytosolic areas was tested by confocal microcopy in cells stained with both TMRM and mitotracker green.

Cytochrome c release
Cytochrome c release from mitochondria was tested by immunofluorescence and conventional or confocal microscopy. Cells were treated under the various experimental conditions for 72 h and fixed. Location of cytochrome c was tested by indirect immunofluorescence. In conventional fluorescence, nuclei were identified by DAPI staining. Confocal images were obtained using a BIO-RAD laser scanning system (Radiance 2100) coupled to a Nikon eclipse TE2100U, inverted microscope. For quantification of cytochrome c release by confocal microscopy, the relative abundance (%) of cells showing diffuse vs. punctate immunofluorescence was calculated [60].

Cell death and apoptosis
Cells were plated in wells at about 5610 4 cells/ml and treated with test solutions for 72 h. Cell death was estimated in the same samples by staining with fluorescein diacetate (FDA, 50 mg/ml, 3 min) and propidium iodide (PI, 20 mg/ml, 30 s) and assessed by fluorescence microscopy using a Nikon Eclipse 80i microscope coupled to a DM 1200C digital camera using a 20x objective. For determination of apoptotic cells at the single cell level, cells were plated at about 5610 4 cells/ml and incubated with test solutions for 72 h. Apoptotic cells were revealed by the terminal deoxynucleotidyl transferase-mediated dUTP nick-end labeling (TUNEL) method by fluorescence imaging and a cell death detection kit following the protocol provided by the manufacturer.

Cell ATP levels
Cerebellar granule cells were plated in 4-well plates and cultured with test solutions for 72 h. Then, cells were washed twice with PBS at 37uC and 1 ml of boiling 20 mM Tris, pH, 7.75 and 4 mM EDTA was added. After 2 min, samples were centrifuged for 4 min at 10.000 g. ATP was measured later from the supernatant by the luciferin-luciferase assay using a Cairn photon counting device (Cairn Research, UK) and a standard curve prepared using pure ATP over a 10 25 to 10 29 M concentration range.

Statistics
When only two means were compared, Student's t test was used. For more than two groups, statistical significance of the data was assessed by ANOVA and compared using Bonferroni's multiple comparison test. Differences were considered significant at p,0,05.  Figure S4 Ab oligomers induce ROS formation that is prevented by FCCP and R-flurbiprofen. Cerebellar granule cells were incubated for 4 h with vehicle, Ab 1-42 oligomers (500 nM) and oligomers plus FCCP 100 nM or R-flurbiprofen. Then, ROS production was imaged using the ROS-sensitive probe CM-H2DCFDA. Ab 1-42 oligomers induced an increase in fluorescence compared to the vehicle that was prevented by FCCP and Rflurbiprofen. The Pictures are representative of 5-9 microscopic fields in at least 3 independent experiments for each condition. Addition of FCCP or R-flurbiprofen alone produced similar results than vehicle (data not shown).