Many studies support the view that accumulation of the β-amyloid peptide (Aβ) is central to Alzheimer's disease pathology 1 . Synthetic Aβ is toxic in both neuronal cell lines 2 3 and primary neurons 4 , and Alzheimer's-like pathology has been observed in a range of transgenic models that accumulate Aβ 5 6 7 8 . Although numerous studies have implicated oligomeric Aβ as the toxic species 9 10 11 , the structure of the key toxic Aβ species is unresolved, as is the toxic mechanism. As it has not been possible to obtain atomic structures for Aβ by X-ray crystallography, there is significant disagreement as to whether the toxic form of Aβ involves an α-helical or β-sheet structure.

Synthetic Aβ preparations can convincingly form pores in synthetic membranes, leading to the proposal that in vivo Aβ toxicity results specifically from direct membrane damage 12 . Interestingly, the hydrophobic C-terminal region of Aβ (corresponding to the transmembrane portion of the amyloid precursor protein, APP) contains a Gly-XXX-Gly-XXX-Gly motif, termed a "glycine zipper" by Kim et al 13 . These researchers pointed out that this motif is present in the transmembrane domains of a number of bacterial channel proteins, and structural modeling of these channel proteins suggests that the glycine zipper motif can drive the packing of transmembrane α-helices. A schematic model showing the potential glycine zipper motif interface between α-helical regions of two Aβ peptides is shown in Figure 1. To support the idea that the glycine zipper motif of Aβ drives the formation of membrane pores, Kim et al demonstrated that Gly-to-Leu substitutions in this motif (particularly the G37L substitution) could block Aβ pore formation in synthetic membranes and reduce Aβ toxicity in cultured Neuro 2a neuroblastoma cells. However, these studies did not directly demonstrate that the G37L substitution disrupted the formation of Aβ oligomers, or that oligomer disruption was the basis for the reduced toxicity in cell culture.

Hung et al 14 compared the toxicity of synthetic Aβ 1-42 containing G25L, G29L, G33L or G37L substitutions (generally termed GSL peptides) to wild type Aβ in primary mouse cortical neurons. These substitutions were all found to be less toxic than wild type Aβ, with the G33L and G37L substitutions having the greatest reduction in toxicity. The reduced toxicity of these GSL peptides was correlated with reduced formation of smaller Aβ oligomers (dimers, trimers, tetramers, and pentamers) in vitro. These researchers also found that the GSL peptides had increased rates of fibril formation using an in vitro Thioflavin T assay. Harmeier et al 15 examined the role of G29 and G33 in Aβ 1-42 toxicity and oligomerization, and found that substitutions at G33 (G33A or G33I) dramatically reduced Aβ toxicity in neuroblastoma cells and biased in vitro oligomerization towards higher molecular weight forms. These researchers also demonstrated that unlike wild type Aβ, G33-substituted Aβ did not inhibit hippocampal LTP or disrupt eye formation in a Drosophila transgenic Aβ expression model.

In theory, the relevance of the glycine zipper to Aβ toxicity could be explored in transgenic mouse models expressing Amyloid Precursor Protein (APP) that contains substitutions in the glycine zipper motif. However, this motif has been demonstrated to play a role in the dimerization of APP, and substitutions in the motif alter APP proteolytic processing and Aβ production 16 17 , thus confounding any interpretations of this hypothetical model. We have therefore employed a transgenic C. elegans model 18 that directly expresses Aβ42 to assay effects of glycine zipper substitutions, particularly G37L, on toxicity and amyloid formation, neither of which has been previously tested in vivo. To address the important issue of whether the correlations between the effects of glycine zipper substitutions on oligomerization and toxicity are causally linked, we used multiple cell models to ask whether G37L substitutions are actually anti-toxic, a prediction of the toxic oligomer model. Finally, we used second site substitutions in Aβ G37L to test rigorously structural predictions of the glycine zipper model in vivo.

Although the β-amyloid peptide is toxic in a wide range of model systems, the specific toxic form of Aβ has not been convincingly identified, nor is it clear if the same form of Aβ is toxic in different model systems. Numerous studies have pointed to soluble, oligomeric Aβ as being the key toxic species, but there is no general agreement about the subunit composition or quaternary structure of this toxic oligomer. Defining the toxic Aβ species is difficult because while the oligomeric nature of Aβ can be assayed before it is added to cells (e.g., size exclusion profiling of ADDLs prepared from synthetic Aβ) or after Aβ is expressed (e.g., immunoblotting protein preparations from transgenic Aβ-expressing animals), capturing the toxic species "in action" has not been feasible. Direct biochemical approaches may also be problematic if the key Aβ toxic species is unstable and/or a minor component of total Aβ. To avoid these problems we have assayed the toxicity, in vitro and in vivo, of a set of Aβ variants designed to test a specific model of the toxic Aβ species. Our results demonstrate a critical role for C-terminal glycine residues in the formation of toxic Aβ oligomers, and thereby serve to significantly constrain structural models for these toxic oligomers.

We find that Aβ G37L is less toxic than Aβ wild type both in vivo and in primary neuronal culture. Our results are complementary to those of Harmeier et al 15 , who showed that G33A or G33I substitutions also reduce Aβ toxicity in vitro and in vivo. (These authors did not report testing G37 substitutions.) These combined studies strongly argue that it is the disruption of the glycine zipper itself, rather than the alteration of any specific residue, that reduces Aβ toxicity.

How does Aβ assemble into the toxic oligomeric species? A large fraction of Aβ is membrane-associated both in the C. elegans model described here and in the AD brain (G. McColl and A. Bush, personal communication). We speculate that like a number of vertebrate antimicrobial peptides, Aβ exists in a β-sheet or random coil configuration when soluble extracellularly or in the cytoplasm, but converts to an α-helical structure upon membrane association. This conversion has been demonstrated by biophysical methods for plasticins 30 31 and bombinins 32 , anti-microbial peptides that contain Gly-XXX-Gly-XXX-Gly motifs. Interestingly, Aβ has also been shown to have anti-microbial activity 33 . While x-ray crystal structures have not been determined for small anti-microbial peptides, the first atomic structure for an α-helical pore-forming toxin (ClyA) has recently been described 34 . We note that this study reveals a β-sheet "tongue" present in the soluble form of ClyA that converts to α-helical structure upon membrane insertion. This tongue region also contains a glycine zipper motif (...G¹⁸⁰AAAGVVAG¹⁸⁸...), and a glycine-to-valine substitution at position 180 has been shown to compromise hemolytic activity of this toxin 35 . Thus, although ClyA is significantly larger than Aβ, these structural data are consistent with our model of α-helix formation by the glycine zipper region of Aβ occurring upon membrane association.

Our observation that AβG37L does not form detectable amyloid in vivo differs from the results of Hung et al 14 , who found that Aβ G37L had enhanced fibrilization in vitro. These results can be reconciled if there are significant differences in the amyloid seeding step 36 under in vivo and in vitro conditions. It has been proposed that the formation of α-helical structure by membrane-associated Aβ drives the formation of oligomers and therefore a local increase in Aβ concentration, and this localized Aβ concentration catalyzes β-sheet formation and amyloid formation 37 . α-helical intermediates have been observed preceding amyloid formation by Aβ 38 . We suggest that in the C. elegans model, glycine zipper-driven oligomerization at membranes is required to reach the critical local Aβ concentration to initiate amyloid formation, and this oligomerization and subsequent amyloid formation is blocked by the G37L substitution. However, other interactions (perhaps influenced by overall peptide hydrophobicity) may drive the formation of initial amyloid seeds in vitro, and thus the glycine-substituted Aβ peptides have faster amyloid formation rates under these conditions.

While Aβ G37L appears anti-toxic in both Neuro-2a cells and primary hippocampal neurons, we did observe different co-incubation requirement for these two model systems: inhibition of wild type Aβ toxicity required immediate co-oligomerization with Aβ G37L in the Neuro-2a model but not in the primary neuronal culture model. One possible explanation for these differences is that Aβ oligomers show highly preferential binding to synaptic sites in primary neurons 25 , suggesting there may be a limiting number of Aβ binding sites in these neurons. Consequently, unoligomerized Aβ G37L may not only counter the toxic oligomerization of wild type Aβ, but oligomerized Aβ G37L may also compete with wild type oligomers for binding to relevant sites on the neuronal plasma membrane. Competition for binding sites may not be relevant to Aβ toxicity in the undifferentiated Neuro-2a cells used in our study.

Our demonstration of anti-toxic effects of Aβ G37L in cell culture and primary neurons clearly supports the oligomer model for Aβ toxicity, although we cannot readily infer the subunit composition of these oligomers from our experiments. We have observed variable patterns of presumed Aβ oligomeric bands in strains expressing different Aβ variants, but we have been unable to associate a specific oligomeric band with toxicity. A caveat to this approach to interpreting immunoblots is that we do not know to what degree they capture the true distribution of Aβ oligomers in vivo. It has been reported that different biophysical forms of Aβ actually show similar oligomer banding patterns after SDS-PAGE 39 .

Even though Kim et al have demonstrated that the G37L substitution blocks Aβ channel formation in synthetic membranes, and we have rigorously demonstrated that Aβ G37L has reduced toxicity in multiple models, at present we have no direct evidence that the reduced toxicity of Aβ G37L is due to effects on membrane conductance per se. We also note that Aβ need not form pores in order to disrupt membrane function; instead it could result from Aβ altering the local membrane environment such that the conductance of endogenous ion channels (e.g., the NMDA receptor) is altered. An appealing component of the membrane disruption hypothesis is that it explains why Aβ is toxic when both outside cells (e.g., hippocampal neurons exposed to ADDLs) and inside cells (e.g., intracellular accumulation in C. elegans): in either case, Aβ has access to the plasma membrane. In theory, intracellular Aβ could also damage organelle membranes, such as the ER or mitochondrial membranes. Aβ accumulation has often been associated with mitochondrial dysfunction 40 41 42 , and we have observed mitochondria with abnormal ultrastructure specifically in C. elegans muscles accumulating Aβ (Figure 2). Given that we did not observe abnormal mitochondria in transgenic worms expressing Aβ G37L, we speculate that expression of wild type Aβ in this model may lead to disruption of mitochondrial membrane function.

Introduction of second site mutations into Aβ G37L resulted in enhanced toxicity. This compensatory interaction is most striking in the N27G G37L variant, which is clearly more toxic than either single substitution (see Additional file 1 Figure S2). However, interpretation of the enhanced toxicity in Aβ I31G G37L and Aβ M35G G37L relative to Aβ G37L is complicated by the observation that the single Aβ I31G and Aβ M35G strains produce reduced levels of Aβ relative to the control Aβ wild type strain but show comparable paralysis rates, suggesting these single substitutions may be more toxic than wild type Aβ. If this is the case, we cannot exclude the possibility that these second site mutations increase the toxicity of Aβ G37L by an additive rather than compensatory mechanism. If the correct interpretation is that these second site mutations all increase toxicity by restoring helix packing in the glycine zipper region, our results imply that there is no single packing arrangement involved in the formation of toxic Aβ oligomers, as all three of the tested second site substitutions enhanced the toxicity of Aβ G37L. An alternative explanation for the enhanced toxicity of the double mutant Aβ is that it is the total number of glycine residues in the C-terminal region of Aβ that is the key determinant of toxicity. Although we cannot exclude this possibility, it is difficult to develop a molecular model that would account for this explanation.

We did undertake multiple attempts to generate transgenic worms with a leucine substitution at the single glycine residue in Aβ that lies outside the glycine zipper region (G9), which we predicted would not reduce toxicity. Unfortunately, all of the chromosomally-integrated transgenic lines recovered in this series of experiments did not produce detectable Aβ, an experimental failure that did not occur for any of the other glycine substitutions we attempted. It has also been more difficult to recover transmitting lines expressing wild type Aβ, presumably due to selection against lines expressing this toxic peptide (even under non-induced conditions). Thus, while we do not have data for the G9L control, we speculate that this variant may be at least as toxic as wild type Aβ.

Although Aβ G37L appears to be significantly less toxic than wild type Aβ when expressed in C. elegans, it does not appear to be non-toxic, as transgenic worms expressing Aβ G37L still ultimately become paralyzed. A possible explanation for this observation is that there may be two components to Aβ toxicity in this C. elegans model: a specific toxicity due to activity of Aβ oligomers, and a more general toxicity due to accumulation of misfolded protein. There is strong evidence that accumulation of misfolded proteins in C. elegans muscle can lead to a disruption of cellular protein handling (proteostasis) 43 44 , and expression of an aggregation-prone polyglutamine::YFP fusion protein in C. elegans muscle leads to reduced motility 45 . Aβ G37L and the other variants examined in this study all make immunoreactive cytoplasmic inclusions when expressed in C. elegans muscle, consistent with our transgenic inducible system generally producing Aβ levels that exceed cellular protein handling capacity.

Our studies predict that compounds that specifically interfere with Aβ glycine zipper formation will block Aβ toxicity, and therefore would be candidate therapeutics for Alzheimer's disease. Computational methods exist to design peptides that specifically disrupt helix-helix interactions in protein transmembrane regions 46 47 ; in theory this approach can be used to target the Aβ glycine zipper. Extending this approach to develop practical therapeutic compounds for Alzheimer's disease will likely require identification of brain-accessible small molecules that specifically inhibit Aβ glycine zipper interactions. Our studies also suggest the possibility that there may exist in the human population alleles of APP that encode substitutions in the glycine zipper regions that protect against Alzheimer's disease.

Aβ: β-amyloid peptide; N2a: Neuro2a; MTT: (3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide): SEM: Standard Error of the Mean.

The authors declare that they have no competing interests.

C. elegans experiments were performed by VF, VD, CMR, PG, and CDL. Neuro 2a experiments were performed by GHS and ADDL-treated hippocampal were analyzed by PL. (oligomer binding and drebrin expression) and ND, EYF and MAS (tau phosphorylation). Adenovirus transfection of cortical neurons was performed by JM GHS, PL, MAS, JM and CDL prepared the manuscript. All authors have read and approved the final manuscript.