Interaction of amyloid-β (Aβ) oligomers with neurexin 2α and neuroligin 1 mediates synapse damage and memory loss in mice

Brain accumulation of the amyloid-β protein (Aβ) and synapse loss are neuropathological hallmarks of Alzheimer disease (AD). Aβ oligomers (AβOs) are synaptotoxins that build up in the brains of patients and are thought to contribute to memory impairment in AD. Thus, identification of novel synaptic components that are targeted by AβOs may contribute to the elucidation of disease-relevant mechanisms. Trans-synaptic interactions between neurexins (Nrxs) and neuroligins (NLs) are essential for synapse structure, stability, and function, and reduced NL levels have been associated recently with AD. Here we investigated whether the interaction of AβOs with Nrxs or NLs mediates synapse damage and cognitive impairment in AD models. We found that AβOs interact with different isoforms of Nrx and NL, including Nrx2α and NL1. Anti-Nrx2α and anti-NL1 antibodies reduced AβO binding to hippocampal neurons and prevented AβO-induced neuronal oxidative stress and synapse loss. Anti-Nrx2α and anti-NL1 antibodies further blocked memory impairment induced by AβOs in mice. The results indicate that Nrx2α and NL1 are targets of AβOs and that prevention of this interaction reduces the deleterious impact of AβOs on synapses and cognition. Identification of Nrx2α and NL1 as synaptic components that interact with AβOs may pave the way for development of novel approaches aimed at halting synapse failure and cognitive loss in AD.

Synapse failure is a prominent feature of Alzheimer pathogenesis (1)(2)(3)(4), and considerable evidence indicates that soluble oligomers of the amyloid-␤ protein (A␤Os) 4 are responsible for synapse dysfunction and elimination in AD models (3)(4)(5)(6)(7). A␤Os target synapses (8,9) and interact with a number of synaptic membrane proteins putatively assembled into a supramolecular receptor complex (5). Nonetheless, it appears likely that additional components of such an A␤O receptor complex at synapses remain to be identified. Determination of the role played by each of those components in synaptotoxicity and cognitive impairment triggered by A␤Os may provide novel information about mechanisms germane to AD.
As part of a strategy to identify novel A␤O-interacting proteins at synapses, we have screened a phage display library to detect peptides that bind A␤ and are homologous to synaptic membrane proteins. Using this approach, we identified a heptapeptide (IGTVDRS) that exhibits significant sequence homology to an amino acid sequence motif contained within the NL-interacting domain in Nrxs. This suggests that Nrxs could represent novel synaptic components targeted by A␤Os. Consistent with this hypothesis, we report that A␤Os bind different isoforms of Nrx and its trans-synaptic partner, NL. Anti-Nrx2␣ and anti-NL1 antibodies reduced A␤O binding and prevented A␤O-induced oxidative stress and synaptotoxicity in cultured hippocampal neurons. Significantly, anti-Nrx2␣ and anti-NL1 protected against novel object recognition (NOR) memory impairment triggered by A␤Os in mice, and anti-Nrx2␣ further protected against oligomer-induced object location (OL) memory impairment. These findings identify Nrx2␣ and NL1 as synaptic components targeted by A␤Os and indicate that interfering with interactions between A␤Os, Nrx2␣, and NL1 decreases the impact of A␤Os on synapses and cognition. The results may illuminate novel approaches to fight cognitive decline in AD.

A␤Os interact with different isoforms of Nrx and NL
We initially screened a commercial phage display peptide library (see "Experimental Procedures") to identify A␤-binding peptides presenting sequence homology to synaptic membrane proteins. A systematic search of protein databases in NCBI using BLAST revealed that one of the A␤-binding peptides thus identified (corresponding to the amino acid sequence IGT-VDRS, henceforth termed IG peptide) presented significant sequence similarity to a sequence motif present in both Nrx␣ and Nrx␤ (Fig. 1A). Interestingly, this motif is contained within the NL1-binding domain in Nrxs (Fig. 1, B and C) (22).
We next aimed to biochemically validate the interaction between A␤Os and Nrx. Using a plate binding assay (see "Experimental Procedures"), we first found that A␤Os bind to the IG peptide (Fig. 1, D and E). Importantly, A␤Os bound to purified recombinant full-length Nrx1␣ and Nrx2␣ (but not to BSA, used as a negative control) (Fig. 1, F and G).
Consistent with recent reports (10,11), A␤Os bound purified recombinant NL1 (Fig. 1F). We further found that A␤Os robustly bound NL2 but not NL3 (Fig. 1G). These results suggest that different isoforms of Nrx␣ and NL can be synaptic targets of A␤Os. A, amino acid sequence alignment reveals significant homology between the IGTVDRS peptide, identified by phage display (see "Results") and different isoforms of human Nrxs; identical residues are shown in red, and conservative amino acid replacements are shown in blue. B and C, ribbon diagrams of the structures of Nrx1␣ (B) and the Nrx1␤ LNS6 domain (C, left) that interacts with NL1 (C, right). The Nrx amino acid sequence motif homologous to the IGTVDRS peptide is highlighted in red (B and C). Nrx residues involved in interaction with NL1 are shown in blue (C), and NL1 residues involved in interaction with Nrx are shown in green (C). The epitope recognized by the anti-Nrx2␣ antibody used in this study appears in yellow in B. The figures were generated using Visual Molecular Dynamics (University of Illinois), and the atomic coordinates for Nrx1␣ and the Nrx1␤/NL1 complex (PDB codes 3POY and 3BIW, respectively). D, binding assay (see "Experimental Procedures") between plate-immobilized IG peptide (2.4 or 24 M, as indicated) or BSA (150 M, used as a negative control) and A␤Os added in soluble form to the wells (0.5, 1 or 2 M, as indicated). The detected signal was the immunoreactivity toward the anti-A␤ oligomer NU4 antibody (30). E, optical density quantification of the binding assay in D. Error bars represent mean Ϯ S.D. of three wells for each experimental condition. a.u., arbitrary units. F, plate binding assay between immobilized purified recombinant Nrx1␣ (60 M) or NL1 (150 M) and A␤Os (0.9 M) added in soluble form to the wells. The figure shows representative experiments from two to four independent experiments each. G, plate binding assay between immobilized purified recombinant Nrx2␣ (3 nmol), NL1, NL2, or NL3 (150 M) and A␤Os (0.9 M) added in soluble form to the wells. Veh, vehicle. H, ligand capture assay (see "Experimental Procedures") between plate-immobilized A␤Os and proteins present in human cortical homogenate. Ligand detection was performed using anti-NL1, anti-Nrx2␣, or anti-GABA A R5 antibodies. The figure illustrates a representative experiment from two independent experiments that yielded similar results. I, plate binding assay between immobilized purified recombinant Nrx2␣ (60 M) and NL1 (150 M), with the latter added in soluble form to the wells. Binding of soluble NL1 to immobilized Nrx2␣ was evaluated, after appropriate washing, by immunoblotting (IB) using anti-NL1 to detect the complex.
We then sought to investigate whether A␤Os interact with human NL1 and Nrx2␣. Using a ligand capture assay (see "Experimental Procedures"), we found that A␤Os interact with both NL1 and Nrx2␣, present in human adult brain tissue homogenates, but not with GABA A receptor subunit 5 (GABA A R5), mostly expressed at inhibitory synapses), supporting the notion that A␤Os target these proteins in the human brain (Fig.  1H). We further confirmed that recombinant NL1 and Nrx2␣ directly interacted with each other in our plate binding assay (Fig.  1I).

The IG peptide, anti-Nrx2␣, and anti-NL1 reduce A␤O binding to hippocampal neurons
To determine whether the amino acid sequence motif homologous to Nrx that was identified by phage display was relevant for neuronal targeting by A␤Os, we tested the ability of the IG peptide to act as a scavenger for A␤Os. A␤Os were preincubated in solution with the IG peptide at different molar ratios before addition to neuronal cultures. Hippocampal neurons exposed to A␤Os incubated previously at a 5:1 A␤:IG peptide molar ratio showed a 60% decrease in dendritic A␤O binding (detected as immunoreactivity toward the NU4 monoclonal anti-A␤Os antibody (23)) compared with neurons exposed to A␤Os alone (Fig. 2, A-D). A molar ratio of 1:1 A␤:IG peptide yielded similar results (data not shown).
Because we identified Nrx2␣ and NL1 as proteins that interact with A␤Os, we next hypothesized that Nrx2␣ and NL1 could mediate A␤O binding to cultured hippocampal neurons. We thus asked whether blocking Nrx2␣ or NL1 with specific antibodies would attenuate A␤O binding to neuronal den- Figure 2. Anti-Nrx2␣, anti-NL1, and the IG peptide attenuate A␤O binding to hippocampal neurons. A-C, representative images of A␤O binding (NU4 immunoreactivity) in hippocampal cultures exposed to vehicle (A), A␤Os (B, 500 nM), or A␤Os previously incubated overnight with the IG peptide at a 5:1 A␤:IG molar ratio (C). Scale bar ϭ 10 m. D, integrated A␤O immunofluorescence levels (NU4 immunoreactivity, vehicle (Veh) and A␤Os, n ϭ 5 experiments with independent neuronal cultures and A␤O preparations; A␤:IG complex, n ϭ 4 independent experiments paired with vehicle-and A␤O-treated cultures). E, antibodies against Nrx2␣ and NL1 do not immunoreact with membrane-immobilized A␤Os. A␤Os (1.5 pmol) were spotted (in triplicate) onto nitrocellulose, and membrane strips were incubated separately with 1 g/ml polyclonal anti-A␤Os (LDN1), 5 g/ml anti-NL1, 8 g/ml anti-Nrx2␣, or 10 g/ml anti-cyclophilin B (Ciclo, irrelevant IgG was used as a negative control) and developed by chemiluminescence. F-I, representative images of A␤O binding (NU4 immunoreactivity) in hippocampal cultures exposed for 3 h to vehicle (F), A␤Os (G, 500 nM), A␤Os ϩ anti-Nrx2␣ (H), or A␤Os ϩ anti-NL1 (I). Antibodies were added to the cultures 30 min prior to A␤Os. Scale bar ϭ 10 m. J, integrated A␤O immunofluorescence levels (NU4 immunoreactivity). Anti-Nrx2␣ results are from five experiments with independent neuronal cultures and A␤O preparations (8 g/ml anti-Nrx2␣, n ϭ 3; 16 g/ml anti-Nrx2␣, n ϭ 2). Anti-NL1 (5 g/ml) results are from three experiments with independent neuronal cultures and A␤O preparations. K-N, representative images of A␤O binding (LDN1 immunoreactivity) in hippocampal cultures exposed for 3 h to vehicle (K), A␤Os (L, 500 nM), A␤Os ϩ anti-Nrx2␣ (M), or A␤Os ϩ anti-NL1 (N). Antibodies were added to the cultures 30 min prior to A␤Os. Scale bar ϭ 10 m. O, integrated A␤O immunofluorescence levels (LDN1 immunoreactivity) in a representative experiment from two experiments that yielded similar results. Error bars correspond to means Ϯ S.D. from three replicates. P-S, representative images of A␤O binding (detected by Alexa-streptavidin binding to biotinylated A␤Os) in hippocampal cultures exposed for 3 h to vehicle (P), A␤Os (Q, 500 nM), A␤Os ϩ anti-Nrx2␣ (R), or A␤Os ϩ anti-NL1 (S). Antibodies were added to the cultures 30 min prior to A␤Os. Scale bar ϭ 10 m. T, integrated Alexa-streptavidin fluorescence levels in a representative experiment from two experiments that yielded similar results. Error bars correspond to means Ϯ S.D. from three replicates. In all experiments, 20 -30 images (from two to three coverslips) were acquired and analyzed per experimental condition per independent culture. Symbols correspond to mean values from each independent experiment. ***, p Ͻ 0.001; ****, p Ͻ 0.0001, one-way ANOVA followed by Holm-Sidak post test.
drites. For this, we used an anti-Nrx2␣ antibody recognizing an epitope distal to the Nrx-NL interaction domain (Fig. 1C), thus minimizing the chance that the antibody itself might interfere with trans-synaptic Nrx-NL1 interactions. In addition, the anti-NL1 antibody we used recognizes a region distal to the NL1 esterase domain known to bind Nrxs. To exclude the possibility that the antibodies against synaptic proteins could directly recognize A␤Os and thus mask their detection, we initially performed a control plate binding assay and found that neither anti-NL1 nor anti-Nrx2␣ bound to synthetic A␤Os (Fig. 2E). Hippocampal neuronal cultures were then incubated for 30 min with anti-Nrx2␣ or anti-NL1 antibodies and subsequently exposed to A␤Os (500 nM for 3 h). Interestingly, both antibodies caused significant decreases in dendritic binding of A␤Os, as detected by the immunoreactivities of both NU4 (23) and LDN1 (24) antibodies ( Fig. 2, F-J and K-O, respectively).
Detection of neuronal binding of A␤Os using biotin-conjugated A␤Os and Alexa-conjugated streptavidin yielded similar results ( Fig. 2, P-T). Collectively, these observations suggest that both Nrx2␣ and NL1 are involved in dendritic binding of A␤Os in hippocampal neurons.

Anti-Nrx2␣ and anti-NL1 prevent A␤O-induced neuronal oxidative stress and loss of dendritic spines and synapses
A␤Os trigger neuronal oxidative stress (25)(26)(27), as confirmed here with cultured hippocampal neurons (Fig. 3). When neuronal cultures were pretreated with antibodies against Nrx2␣ and NL1, A␤Os failed to induce excessive generation of reactive oxygen species (ROS) (Fig. 3).
A␤Os have been shown to cause dendritic spine elimination and synapse loss in vitro (8, 28 -30), in vivo (29,31), and in ex vivo human brain slices (32). We hypothesized that interactions of A␤Os with Nrx/NL1 could mediate spine/synapse loss in cultured hippocampal neurons. To test this hypothesis, we investigated whether anti-Nrx2␣ or anti-NL1 antibodies protected synapses from the toxic impact of A␤Os.
We initially assessed dendritic spine density in hippocampal cultures by phalloidin labeling. In line with previous studies Further, pretreatment of hippocampal cultures with anti-NL1 blocked A␤O-induced loss of synaptophysin/PSD-95 co-localized puncta, a readout of synapse density in cultured neurons (33) (Fig. 5, A-D). Similarly, anti-Nrx2␣ prevented synapse loss induced by A␤Os in hippocampal neurons, as measured by synaptotagmin/PSD-95 co-localization (Fig. 3

, E-H).
Collectively, the results indicate that interaction of A␤Os with NL1 and Nrx2␣ contributes to A␤O-induced neuronal oxidative stress and synapse loss.

Anti-Nrx2␣ and anti-NL1 prevent A␤O-induced memory impairment in mice
Last, we asked whether antibodies targeting Nrx2␣/NL1 could prevent memory impairment induced by intracerebroventricular (i.c.v.) infusion of A␤Os in mice. Initial analysis in an open field arena showed that the exploratory behavior and locomotor activity of mice were not affected by treatments in any of the experimental groups (data not shown). In line with our previous reports (29,31,34), mice infused i.c.v. with a single dose of 10 pmol A␤Os showed impaired NOR memory 24 h post-infusion (Fig. 6, A-C). Interestingly, i.c.v. infusion of anti-Nrx2␣ or anti-NL1 (30 min prior to A␤Os) prevented NOR memory impairment induced by A␤Os (Fig. 6). Control experiments showed that an unrelated IgG (anti-GABA A R5) had no effect on cognitive impairment induced by A␤Os (Fig. 6B).
We further investigated whether anti-Nrx2␣ and anti-NL1 could prevent A␤O-induced memory impairment in the OL task (Fig. 6, D and E), which assesses a contextual/location type of memory that is more dependent on the hippocampus than NOR (35,36). We found that, although A␤O-injected mice failed to identify the displaced object, mice injected with anti-Nrx2␣ had preserved OL memory (Fig. 6E). There was also a trend toward prevention against cognitive impairment in mice injected with anti-NL1, but this was not statistically significant (Fig. 6E).

Discussion
Synapse loss is a cardinal feature of AD and is considered the best neuropathological correlate of memory impairment (1,2). A large body of evidence indicates that synapses are targeted by A␤Os and become dysfunctional in AD (reviewed in Refs. 3, 5-7, 37). Thus, approaches aimed to identify specific synaptic components targeted by A␤ hold significant potential to illuminate novel therapeutic strategies in AD. Using phage display of a peptide library, we identified an A␤-binding heptapeptide that was highly homologous to a sequence motif present in the NL-binding site in Nrxs and represents a potential novel A␤Obinding site at synapses.
Nrxs and NLs are cell adhesion molecules that promote synapse stabilization and plasticity, learning, and memory (15,18). Dysfunction in Nrx-NL interactions and related signaling pathways has been implicated in a variety of neurological disorders (reviewed in Ref. 18), including AD. Indeed, a loss-of-function mutation in NL1 has been associated with increased risk of AD (13), and epigenetic suppression of NL1 has been reported to mediate A␤-induced neurotoxicity (14).
Here we report that A␤Os bind Nrx1␣ and Nrx2␣ as well as NL1 and NL2. Our findings further indicate that A␤Os interact with Nrx through its NL-binding domain. This raises the possibility that A␤Os could compromise the Nrx-NL interaction, thus contributing to synapse destabilization and failure. This is supported by our findings that antibodies targeting Nrx2␣ and NL1 prevented A␤O binding and oligomer-induced spine/synapse loss in hippocampal neurons, establishing these proteins as mediators of A␤⌷ interactions at synaptic terminals.
In the experiments employing the IG peptide, we modified the original sequence of the peptide identified by phage display to make it identical to the homologous amino acid sequence motif present in Nrx␣. Two modifications were included: replacing a positively charged arginine residue with a neutral isoleucine residue at position 6 of the sequence and introducing an additional isoleucine residue at the C terminus of the pep-   ) and PSD-95 (red) immunofluorescence in hippocampal neurons exposed for 4 h to vehicle (E), A␤Os (F, 500 nM), and A␤Os ϩ 8 g/ml anti-Nrx2␣ (G). H, quantification of co-localized synaptotagmin/PSD-95 puncta. When used, anti-Nrx2␣ and anti-NL1 were added to cultures 30 min prior to A␤Os. Ten to fifteen fields were imaged and analyzed per experimental condition from four (for anti-NL1) or three (for anti-Nrx2␣) experiments with independent cultures and A␤O preparations. Symbols correspond to mean values from each independent experiment. *, p Ͻ 0.05; **, p Ͻ 0.01, one-way ANOVA followed by Holm-Sidak post test. Scale bars in A-C and E-G indicate 10 m.
tide. The former amino acid replacement (arginine to isoleucine) changes the overall charge of the peptide from neutral to negative at physiological pH values, which might have had an impact on the interaction of the IG peptide with A␤. However, our in vitro binding results revealed that this was not the case, suggesting that the interaction between the IG peptide and A␤ was not much affected by changes in peptide electrostatics. Introduction of the C-terminal isoleucine residue in the IG peptide, mimicking the amino acid sequence in Nrx␣, may have been beneficial in terms of the establishment of non-polar interactions with A␤.
Although NLs and Nrxs were originally described as structural proteins, they have been linked to mechanisms that control long-term potentiation and memory consolidation (38 -40). Nrx-NL interactions and trans-synaptic signaling regulate synapse stability and NMDAR function in the postsynaptic membrane (41,42). Mounting evidence suggests that NMDARdependent calcium influx is germane to A␤O neurotoxicity (3,26,27,43,44). A particular outcome of excessive calcium influx into neurons is an abnormal increase in ROS (27,45). Indeed, brain oxidative stress is a hallmark of AD pathology (46 -48). It seems possible, therefore, that interaction between A␤Os and Nrx-NL mediates neuronal dysfunction that is, at least in part, related to NMDAR-mediated elevation in ROS. In addition, our findings support the notion that interaction between A␤Os and Nrx-NL contributes to the loss of dendritic spines and synapses promoted by A␤Os.
Reduced levels or function of NL1, as demonstrated in AD models (14), could further underlie A␤O-induced deficits in long-term potentiation (49 -51). It is thus possible that A␤Os target NL1 and Nrx2␣ to cause synaptic dysfunction early in disease progression. Prolonged A␤O action may further trigger signaling pathways that down-regulate NL1 and exacerbate synapse defects.
A␤Os have been reported to interact with several proteins present at synaptic membranes, and significant efforts have been directed at identifying the complete set of synaptic targets of oligomers (3,5). Our findings contribute to the identification of new pieces in the puzzle of A␤O-interacting proteins by showing that Nrx2␣ and NL1 are mediators of synaptotoxicity initiated by oligomers. Recent reports have suggested that NL1 could be an A␤-interacting protein at synapses (10,11). Our results are in agreement with the notion that A␤Os target excitatory synapses in hippocampal neurons (9,26,52,53), as NL1 is mostly expressed in excitatory terminals (12). A number of molecules present at excitatory synapses have been described to mediate A␤O neurotoxicity, and we now provide evidence supporting the notion that interference with trans-synaptic interactions between Nrx and NL by A␤Os could contribute to synaptic and memory impairments in AD.
Intriguingly, and in contrast with previous findings (10), our results revealed a robust interaction between A␤Os and NL2, which is mostly expressed at inhibitory synapses. Nonetheless, given the predominance of excitatory terminals as A␤O targets (8), it remains to be determined whether A␤O-NL2 interactions are as relevant as A␤O-NL1 interactions to explain synapse failure and memory outcomes in AD.
Our finding that A␤Os bind to Nrx to promote synaptotoxicity in cultured neurons extends the impact of A␤Os to an essential presynaptic element. In addition to postsynaptic defects and synapse loss, it seems plausible that this interaction contributes to the presynaptic dysfunction induced by A␤Os (3). We have demonstrated previously that a single i.c.v. injection of A␤Os (10 pmol) causes memory impairment in NOR and contextual fear conditioning in mice (29,31,34,54). Our results demonstrated that anti-Nrx2␣ prevented A␤O-induced impairment in both object recognition (NOR) and contextual/ spatial (OL) forms of memory. On the other hand, anti-NL1 prevented A␤O-induced memory deficits in the NOR paradigm but not in the OL test. NOR memory appears to rely largely on frontal and perirhinal cortical circuits, whereas OL is predominantly dependent on the hippocampus (35). Our findings thus suggest an important role for A␤O-Nrx interactions in impairing both recognition and contextual memory processes that are dependent on hippocampal and cortical circuits. In contrast, the selective protective effect of anti-NL1 on NOR memory suggests that A␤O-NL interactions may play a more relevant role in disrupting object recognition than contextual/ location memory.
We further attempted to determine the protective actions of anti-Nrx2␣ and anti-NL1 in the Morris water maze, an established hippocampal spatial memory paradigm (55), as well as in a modified water maze protocol, the so-called memory flexibility test, originally developed to detect age-related memory deficits in a transgenic mouse model of AD (56). However, we found that a single i.c.v. injection of A␤Os had no impact on spatial memory in either task (data not shown). This may be related to the fact that water maze paradigms employ multiple training sessions (performed during several days) in an aversive environment. This could lead to stronger memory formation, not likely to be affected by a single infusion of a low dose of oligomers. Conversely, NOR and OL are grounded by innate curiosity and voluntary exploratory behavior toward novelty after a single training session in a non-aversive environment. Memories established in a single event are perhaps more vulnerable than ones established over multiple training sessions (57), which may explain why NOR and OL are more sensitive to A␤O-induced memory impairment.
Our findings that Nrx2␣ and NL1 present in human cortical homogenates bind A␤Os and that antibodies against Nrx2␣ and NL1 protected mice from cognitive impairment triggered by A␤Os suggest that the impact of A␤Os on Nrxs and NL1 could have potential implications for translational approaches in humans. However, further studies are required to determine the specific mechanisms underlying the preservation of cognitive function by anti-Nrx2␣ and anti-NL1. Antibodies targeting sites distal from the Nrx/NL-interacting domain in each protein (as done here) may sterically hinder the interaction of A␤Os with Nrxs/NLs, leading to preservation of normal synaptic signaling and function.
In conclusion, we provide biochemical and cell biology evidence that A␤Os bind presynaptic Nrx at or near the site of Nrx-NL1 interaction and that this instigates synapse damage and neuronal toxicity. Significantly, our findings indicate that A␤O-binding to Nrx2␣ or NL1 mediates the impact of A␤Os on synapses and memory in mice. Taken together, the results identify the trans-synaptic partners Nrx and NL as novel A␤O targets and suggest that interfering with A␤O binding to Nrx-NL could constitute a novel approach toward muchneeded AD therapeutics.

Ethics statement
All procedures were in accordance with the Principles of Laboratory Animal Care (National Institutes of Health) and performed in certified facilities under protocols approved by the Institution Animal Care and Use Committee (IACUC) of the Federal University of Rio de Janeiro (protocols IBQM 022 and IBQM 044). Experiments involving human cortical tissue were approved by the Committee for Research Ethics of the Clementino Fraga Filho University Hospital of the Federal University of Rio de Janeiro (protocol 0069.0.197.000-05). Donors gave written informed consent for the use of brain tissue that would otherwise have been discarded.

A␤ preparation and characterization for the biopanning experiment
A␤ 1-40 was diluted to 0.1 g/ml in 50% trifluorethanol/PBS, and 50-l aliquots were dried in the wells of a 96-well plate at 37°C under constant shaking for 16 h. Size-exclusion chromatography-HPLC (using the intrinsic fluorescence emission of the single tyrosine residue in A␤) showed that A␤ 1-40 solubilized in 50% trifluoroethanol/PBS solution displayed no sign of aggregation or oligomerization for at least 24 h (data not shown).

A␤O preparation and characterization
A␤Os were prepared weekly from synthetic A␤ 1-42 following the original procedure described by Lambert et al. (58) and were routinely characterized by SEC-HPLC as described previously (24,27,31,32). For biotinylated A␤O preparations, a 1:4 ratio of biotinylated:non-biotinylated A␤ 1-42 was used. Oligomers were maintained at 4°C at all times and used within 48 h of preparation.

Biopanning against A␤
Screening of a phage display peptide library for A␤-binding peptides was performed as described previously (60 -62). Briefly, wells coated previously with non-aggregated A␤ 1-40 (prepared as described above) were blocked in 0.5% bovine serum albumin in PBS for 1 h, and 10 l of a commercially available phage display library (PhD-C7C, New England Biolabs, Ipswich, MA) was added to each well. The procedures were carried out according to the instructions of the manufacturer. Among 24 different phages bound to A␤, one expressed the amino acid sequence IGTVDRS displayed on the phage filamentous protein. To identify candidate proteins to which this amino acid sequence might belong, the sequence was compared with protein sequences deposited in several data banks (available at NCBI) using BLAST.

In vitro plate binding assay
The IGTVDISI peptide (referred to as IG peptide, comprising a slight amino acid sequence modification relative to the original IGTVDRS sequence identified by phage display to correspond exactly to the amino acid sequence present in Nrx␣), purified recombinant Nrx1␣, Nrx2␣, NL1, NL2, and NL3 were freshly resuspended in PBS and added separately to the wells of a 96-well plate (0.5 g/well for Nrxs, 1 g/well for NLs), and allowed to dry overnight at 37°C. A␤Os or vehicle was then added to each well and incubated overnight at 4°C under gentle agitation. Wells were washed with PBS, blocked for 2 h with 1% BSA in PBS, washed six more times in PBS, and incubated for 1 h with 0.3 g/ml anti-A␤O NU4 monoclonal antibody (23). Wells were incubated with HRP-conjugated secondary antibody for 1 h, developed with SuperSignal Femto substrate, and imaged on photographic film. Control wells were coated with BSA. Luminescence values for NU4 immunoreactivity were quantified using NIH ImageJ software (63).

Immunocytochemistry
Hippocampal cultures were fixed with 4% paraformaldehyde containing 4% sucrose for 10 min and blocked for 1 h with 10% normal goat serum in PBS. For A␤O binding assays in hippocampal neurons, cells were incubated overnight at 4°C with 1 g/ml monoclonal NU4 antibody (23) or 1 g/ml polyclonal LDN1 antibody (24) in 10% NGS under non-permeabilizing conditions. After washing, Alexa-labeled anti-mouse or antirabbit IgG (1:2000 in 1% NGS in PBS) was added.
For biotinylated A␤O assays, cells were incubated with Alexa 566-conjugated streptavidin (1:2000 in 10% NGS). After 2-h incubation at room temperature, cells were extensively washed in PBS and mounted in ProLong Antifade Gold with DAPI. Images were obtained from 10 -15 randomly chosen fields per experimental condition, from two to three coverslips under each condition, using a ϫ63 objective in a Zeiss Axioplan microscope. Integrated fluorescence intensity was measured using ImageJ software. For synapse density assays, we used a slightly modified protocol that included cell permeabilization with 0.1% Triton X-100 for 5 min prior to antibody incubation. Mouse anti-synaptophysin (1:500), mouse anti-synaptotagmin (1:1000), or rabbit anti-PSD-95 (1:1000) primary antibodies were incubated for 2 h at room temperature. Alexa Fluor 488conjugated goat anti-rabbit (1:300) and Alexa Fluor 546-conjugated goat anti-mouse (1:1000) secondary antibodies were incubated for 1 h at room temperature, and cells were imaged. To determine co-localization of synaptic proteins, neurons separated by a distance of at least two cell body diameters from neighboring cells were selected. Green and red channel images were merged, and co-localization was quantified using the Puncta Analyzer plug-in in ImageJ as described previously (29,33,64). In each experiment, at least 30 -40 neurons were analyzed per experimental condition (with two to three coverslips per condition).

Phalloidin labeling
For dendritic spine labeling, cells were fixed in freshly prepared 4% formaldehyde solution for 15 min, permeabilized in 0.1% Triton X-100 for 5 min, blocked in BSA 3% for 60 min and labeled with Alexa Fluor 594-phalloidin for 2 h, according to the instructions of the manufacturer, as described previously (28). Coverslips were DAPI-counterstained and mounted on ProLong Antifade Gold reagent for imaging.

ROS
ROS formation was evaluated in living neurons using CM-H 2 DCFDA as described previously (25)(26)(27). Hippocampal cultures were incubated for 4 h at 37°C with vehicle or 500 nM A␤Os in the absence or presence of anti-NL1 or anti-Nrx2␣ antibodies. When present, antibodies were added to the medium 30 min before A␤Os. 2 M CM-H 2 DCFDA was loaded during the last 40 min of incubation with A␤Os. Neurons were rinsed three times in warm PBS containing 2% glucose and immediately imaged on a Nikon Eclipse TE300 inverted microscope. Analysis of DCF fluorescence was carried out using ImageJ software. Twelve images were acquired under each experimental condition, carried out in triplicate (ϳ300 cells analyzed per condition). Three and four independent experiments (with different hippocampal cultures and A␤O preparations) were performed using anti-Nrx2␣ and anti-NL1, respectively.

Ligand capture assay in human cortical tissue
Healthy cortical tissue was obtained from patients with drugresistant temporal lobe epilepsy subjected to surgical interventions for removal of epileptic foci, as described previously (32). Tissue fragments were immediately frozen in liquid nitrogen until the assay. Tissue was homogenized in radioimmune precipitation assay buffer (5 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1% Nonidet P-40, 1% sodium deoxycholate, 0.1% SDS, 5 mM EDTA, and 1% Triton X-100) containing protease and phosphatase inhibitor cocktails (Thermo-Pierce), and total protein concentration was determined by BCA. 500 g of total protein was added to A␤O-coated wells (1 g A␤O/well) and incubated overnight at 4°C. BSA (1%) was used as a control. Wells were washed ten times in PBS and incubated with antibodies against Nrx2␣, NL1, GABA A R5 (negative control), or PrPc (cellular prion protein, positive control) at 1 g/ml overnight at 4°C. This was followed by incubation with appropriate secondary antibodies for 1 h at room temperature. Wells were developed with SuperSignal Femto substrate and imaged on photographic film.

Intracerebroventricular infusions in mice
We used 2-to 3-month-old male Swiss mice obtained from an animal facility at the Federal University of Rio de Janeiro. Intracerebroventricular administration of A␤Os (10 pmol), anti-NL1 (750 ng), or anti-Nrx2␣ (1200 ng) was performed (a single dose delivered in 1.5 l per animal) as described previously (29,31,34). Briefly, animals were anesthetized using 2.5% isoflurane through a vaporizer system and gently restrained only during the infusion procedure (ϳ5 min). A 2.5-mm-long needle was inserted in the left hemisphere according to the following coordinates: 1 mm to the left of the midline point equidistant from each eye and 1 mm posterior to a line drawn through the anterior base of the eye. When used, anti-NL1 or anti-Nrx2␣ was infused 30 min prior to A␤Os.

NOR test
NOR was used to evaluate short-term declarative memory in mice. In this test, mice are exposed to a known object (presented previously during a training phase) and a novel object (distinct in shape, color, and size and presented simultaneously with the known object during a test phase). Mice with preserved memory spend more time exploring the novel object than the familiar (previously presented) object, therefore indicating they remember the object being presented before. NOR was carried out in an open field arena measuring 30 ϫ 30 ϫ 45 cm. Before training, each animal was subjected to a 5-min-long habituation session. Training consisted of a 5-min-long session during which animals were exposed to two identical objects. The amount of time mice spent exploring each object was recorded. Two hours later, one of the objects was replaced by a novel object, and the animals were again placed in the arena for the test session. The time spent exploring the familiar and novel objects was recorded. Results were expressed as percentage of time exploring each object during the test session and analyzed using one-sample Student's t test, comparing the mean exploration time for each object with the chance value of 50%. Animals that recognize the familiar object as such explore the novel object Ͼ50% of the total time.

OL test
OL was used to evaluate short-term spatial recognition memory in mice (57). In this task, mice are presented to two identical objects in a training phase. In the test session, one of the objects is moved to a different position inside the arena. Mice with preserved memory are expected to spend more time exploring the displaced object than the object in the original position. OL was carried out in an open field arena measuring 30 ϫ 30 ϫ 45 cm. Initially, animals were subjected to a habituation session of 30 min. Twenty-four hours later, animals were exposed to two identical objects side-by-side for 5 min (training session). Two hours later, animals were exposed to the same pair of objects for 5 min, but one of the objects had been relocated to the opposite corner of the arena (test session). The time spent exploring each object in both sessions was recorded. The results were expressed as the percentage of time spent exploring each object during the test session and analyzed using onesample Student's t test, comparing the mean exploration time for each object with the chance value of 50%. Animals that recognize the relocated object as such explore it for more than 50% of the total time.

Statistics
Statistical analysis was carried out using Student's t test when two different experimental conditions were compared or by one-way ANOVA followed by appropriate post-hoc test when more than two conditions were compared.