Shaping Up Amyloid Toxicity: Does It Compute?
Aggregated proteins lie at the heart of many neurodegenerative diseases, but researchers have struggled to pin down the precise conformations that trigger pathology. For amyloid-β (Aβ), the Alzheimer disease protein, it appears that small, structured oligomers are the evil players, but how they get their shape, and why they are so toxic, remain unsolved mysteries.
A pair of recent papers features in silico, in vitro, and in vivo approaches to address those questions. In one report, Christopher Dobson, Damian Crowther, David Lomas, and colleagues at the University of Cambridge in England provide evidence that the tendency of Aβ peptides to aggregate into small, oligomeric protofibrils, as predicted from primary amino acid sequence, predicts pathogenicity in a fly model of Alzheimer disease. The investigators calculated the propensity of different mutated forms of Aβ42 to form oligomers using a novel algorithm, and backed up their calculations by direct measurements in the test tube. The paper, out October 30 in PLoS Biology, raises the possibility that researchers might be able to assess the potential toxicity of any protein based on its amino acid sequence.
Looking at how oligomers come to be, an earlier publication from other Cambridge researchers Giorgio Favrin, Michele Vendruscolo, and again Dobson suggests that the generation of protofibrils occurs via a two-step process and is heavily dependent on the hydrophobicity of the starting peptide. These scientists used computer modeling to simulate the aggregation and folding of two different peptides derived from Aβ42. They found that the first stage of oligomerization occurs when peptides cluster through hydrophobic interactions. Then, the aggregated peptides reorganize into β-sheet-containing amyloid structures. This second step re-exposes hydrophobic residues to solvent, an event that may explain the toxicity of the oligomers, the authors speculate. The work was published in the September issue of PLoS Computational Biology.
In recent years, it has been recognized that the ability to aggregate and form amyloid fibrils does not depend on a given protein’s primary sequence, but may be a common property of proteins with widely varying amino acid sequence. Not all proteins form toxic oligomers, however, and it has been difficult to figure out precisely what determines toxicity in vivo. In the first paper, lead author Leila Luheshi and coworkers use an in-vivo measure of Aβ toxicity, the Drosophila model of AD. In this model, expression of Aβ42 peptide in the fly nervous system leads to early death and motor problems due to neuronal toxicity (Crowther et al., 2005). Using a computer algorithm to predict the propensity of peptides to aggregate based on their amino acid sequence (Chiti et al., 2003), the investigators assigned aggregation scores to all of the 798 possible point mutants of Aβ42. They then chose 15 of these with varying aggregation tendencies, expressed them in Drosophila, and assessed toxicity by changes in lifespan and motor activity.
From this analysis, the scientists found that the predicted tendency of peptides to aggregate correlated with shorter lifespan and greater decrements in motor ability. The correlation with aggregation was confirmed by measuring aggregation in vitro for five of the mutant peptides. For example, the F20E mutant had a low aggregation score and displayed minimal neurotoxicity. In vitro, the protein aggregated slowly compared to Aβ42, and in vivo, it formed no deposits in the fly brain under the same conditions where Aβ42 produced abundant accumulation. The researchers speculate that because the F20E peptide aggregates slowly, natural clearance mechanisms have a chance to rid the brain of it before it forms deposits.
Some peptides did not fit the pattern, however. One particular double mutation (I31E/E22G) stood out because its primary sequence predicted it would aggregate as avidly as the highly toxic E22G Aβ42, yet the double mutant was not toxic in the flies. In vitro, the two peptides aggregated at similar rates, and flies showed similar levels of deposition, suggesting that the computer model accurately predicted aggregation, but for some reason aggregation did not predict toxicity in this case.
This stimulated the researchers to look at another parameter, which was the tendency of peptides to form protofibrils. These small but highly organized oligomeric precursors to amyloid fibrils are the leading candidates for causing Aβ neurotoxicity (see ARF related news story and ARF news story). The investigators developed a new algorithm to calculate the tendency of mutant peptides to form protofibrils, based on experiments with other protofibril-forming peptides (Pawar et al., 2005). The scores from this calculation correlated even better with toxicity than the aggregation scores. In a separate experiment, the researchers report that analysis of the E22G multimers by electron microscopy revealed many protofibrils, while the I31E/E22G protein produced only fibrils, further strengthening the association between protofibrils and toxicity.
“These results provide compelling evidence that, despite the presence within the cell of multiple regulator mechanisms such as molecular chaperones and degradation systems, it is the intrinsic, sequence-dependent propensity of the Aβ42 peptide to form protofibril aggregates that is the primary determinant of its pathological behavior in living systems,” the authors write. Their fly system provides a way to measure the relationship between neuronal dysfunction in a complex organism and the fundamental factors that determine whether Aβ42 peptides will form protofibrils. The same principles may apply to other aggregation diseases as well, they conclude.
In the second paper, the Cambridge researchers take an in-silico approach to ask how protofibrils form, and whether the process reveals anything about their toxicity. Because of their amorphous and transient nature, the structure of early-stage aggregates of Aβ has been hard to determine. To get a virtual view of the process, first author Mookyung Cheon, who is also affiliated with the Pusan National University in Pusan, Korea, carried out a computer simulation of the folding and aggregation of two Aβ fragments, covering residues 16-22 and 25-35.
According to the simulations, the formation of oligomers followed a two-step pathway that depended on the degree of hydrophobicity of the peptide. The first step, which Cheon readily detected with the more hydrophobic 16-22 fragment, involved the aggregation of peptides into large disordered molten oligomers, driven by hydrophobic interactions. Then, over time, these amorphous blobs with greasy centers reorganized into β-sheet structures by forming interchain hydrogen bonds. If a peptide is not very hydrophobic, it may bypass the first step, as the authors observed with the 25-35 peptide, which proceeded straight to the hydrogen-bond interactions.
As a net result of the formation of the β-sheet, hydrophobic residues become exposed to the solvent. Therein could lie the toxicity of oligomers, the authors speculate, as the sticky protofibrils might interact with any number of proteins in cells. Further, Cheon and colleagues found that the extent of hydrophobic surface exposed depends on the size of Aβ(17-21) oligomers. Because smaller assemblies have weaker total hydrophobic interactions, they expose a larger fraction of their hydrophobic residues. As the oligomers grow larger, the number of exposed hydrophobic residues decreases because of decreasing surface-to-volume ratio and a reduction in the total number of oligomers. This model leads to the prediction that there will be an optimal aggregate size where toxicity is greatest, a situation borne out by experimental results (see ARF related news story and Baglioni et al., 2006).
This scenario of oligomer formation pits the tendency of peptides to make hydrophobic interactions against the propensity to form ordered, hydrogen-bonded structures. In this competition, the hydrophobicity of the peptide sequence balances with the generic, sequence-independent tendency to form interchain hydrogen bonds. It must be noted that the simulations were carried out with Aβ fragments, and thus may not reflect the behavior of the entire protein. However, the fragments used are known to be important to lend Aβ its toxicity. It remains to be seen if the process outlined with these small peptides might supply a common mechanism for the toxicity of oligomers derived from a variety of proteins, as occurs in other neurodegenerative diseases.—Pat McCaffrey
Brigham & Women's Hospital
Homing In on the Molecular Determinants of Aβ Assembly and Toxicity: Lessons from the Computer, Test Tube, and Fly
The paper by Luheshi et al. describes the direct comparison of in silico, in vitro, and in vivo analysis of Aβ aggregation. It builds on many years of computational modeling by Vendruscolo and Dobson and animal modelling by Crowthers and Lomas, and it strongly suggests that protofibrillar assemblies of Aβ are the primary neurotoxic species both in their fly models and perhaps in brains of patients suffering with Alzheimer disease.
Using a previously described algorithm, the authors estimated intrinsic aggregation propensities of all 419 possible single point mutations of Aβ1-42 and all 379 single point mutations of the more toxic Aβ1-42E22G. Based on these data, they selected 15 mutations predicted to produce peptides with a broad range of aggregation propensities. The authors then generated flies transgenic for each of the 15 mutant peptides, plus flies transgenic for wild-type Aβ40 and Aβ42. In each case four to six independent lines for each of the 17 Aβ sequences were generated, i.e., a total of 68-102 different individual transgenic fly lines. After completing this gigantic task, the authors then set about characterizing the transgenic flies.
Certain mutations, such as E22G, caused a dramatic reduction in both the life span and locomotor ability of flies, whereas other mutations had the opposite effect. Strikingly, comparison of intrinsic aggregation propensity versus longevity or locomotor activity revealed a strong correlation. This demonstrates a direct link between aggregation and decreased locomotion/longevity.
However, the correlation was not perfect, and one particular mutation, I31E/E22G, did not fit this trend. In vitro analysis of the fibril-forming propensity of the I31E/E22G confirmed that this peptide had a similar aggregation propensity as did the E22G peptide. Moreover, when the authors examined brain from flies expressing I31E/E22G and E22G, they found highly similar levels of amyloid deposits. So if I31E/E22G flies had abundant amyloid deposits, why did they not exhibit a phenotype similar to E22G flies?
The suggested answer came from careful immunohistochemical analysis, which indicated that the E22G flies had not only profound Aβ deposition, but also substantial vacuolation, whereas vacuoles were absent in brains of I31E/E22G flies. Since both the I31E/E22G and E22G flies had abundant amyloid deposits, this led the authors to speculate that something other than amyloid fibrils was precipitating vacuole formation. Mindful of the burgeoning evidence that non-fibrillar soluble forms of Aβ play an important role in cognitive impairment, the authors revised their approach and developed a second algorithm designed to predict the relative propensity of proteins to form protofibrils (PFs). Comparison of predicted propensity of different Aβ mutations to form PF with the relative change in longevity or locomotor ability yielded a substantially improved correlation between that predicted in silico and that observed in flies.
This is an impressive piece of work, but like all leaps forward it raises more questions than it answers. Specifically, while the second algorithm was designed to predict PF forming propensity, it is not clear what the authors actually define as PF. Also unclear is whether the algorithm can predict the formation of structures other than PFs; for instance, what if the algorithm predicts the formation of low-n oligomers? Whatever the answer, it will be extremely interesting to isolate Aβ species from brains of the different mutant flies and attempt to identify the Aβ assembly form that precipitates their impaired locomotion and untimely death.View all comments by Dominic Walsh
Hydrophobic Residues Exposed to Solvent: A Cause of Oligomer Toxicity?
Assembly of proteins into toxic soluble oligomers and highly ordered fibrils is believed to be critical to amyloidogenic diseases associated with protein misfolding and aberrant aggregation. Thus, understanding these processes at atomic resolution has become the center of many computational studies. Computational approaches need to be simplified to enable studies of processes starting from separated protein molecules into ordered aggregates. Cheon et al. employ constant temperature, Monte Carlo simulations and an implicit water protein model which incorporates all atoms but reduces the degrees of freedom to Ramachandran and side-chain torsional angles only. Using a thus simplified computational approach, Cheon et al. study early stages of oligomer and fibril formation of two amyloid-β protein (Aβ) fragments, Aβ(16-22) and Aβ(25-35).
Cheon et al. demonstrate that the process of aggregation into amorphous versus ordered species is determined by a competition between the hydrophobicity of the primary structure and the tendency of amino acids to form arrays of hydrogen bonds. The two fragments, Aβ(16-22) and Aβ(25-35), differ by the degree of their overall hydrophobicity, with Aβ(16-22) being significantly more hydrophobic than Aβ(25-35). Consequently, Cheon et al. make an important observation that while formation of disordered oligomers, primarily driven by hydrophobic collapse, is significantly stronger in Aβ(16-22), Aβ(25-35) proceeds to form ordered fibril-like aggregates with no significant amount of hydrophobically collapsed oligomers. This conclusion nicely complements a more general work on assembly of polyalanine molecules done in the Hall group [1-4] as well as more particular studies of full-length Aβ assembly done in our group [5-8].
During the formation of oligomers, hydrophobic residues are buried on the inside, away from the solvent. Later on, when hydrogen bonds start forming, eventually yielding an ordered fibrillar structure, the hydrophobic residues are forced to get more exposed to the solvent. When the fibrillar structure grows from the initial seed, the ratio of surface to volume steadily decreases, decreasing the total solvent exposure of hydrophobic residues. Thus, the intermediate-size oligomers with some β-strand structure are the ones with a maximal solvent exposure of hydrophobic residues. Based on this observation and on the fact that oligomers of intermediate sizes are typically associated with the highest degree of toxic function, Cheon et al. suggest a relationship between the degree of solvent exposure of hydrophobic residues in an assembly and the cytotoxicity of the assembly. This is a very appealing general hypothesis that needs to be tested in future in-vitro and in-vivo studies.
Nguyen HD, Hall CK. Molecular dynamics simulations of spontaneous fibril formation by random-coil peptides. Proc Natl Acad Sci U S A. 2004 Nov 16;101(46):16180-5. PubMed.
Nguyen HD, Hall CK. Phase diagrams describing fibrillization by polyalanine peptides. Biophys J. 2004 Dec;87(6):4122-34. PubMed.
Nguyen HD, Hall CK. Kinetics of fibril formation by polyalanine peptides. J Biol Chem. 2005 Mar 11;280(10):9074-82. PubMed.
Nguyen HD, Hall CK. Spontaneous fibril formation by polyalanines; discontinuous molecular dynamics simulations. J Am Chem Soc. 2006 Feb 15;128(6):1890-901. PubMed.
Peng S, Ding F, Urbanc B, Buldyrev SV, Cruz L, Stanley HE, Dokholyan NV. Discrete molecular dynamics simulations of peptide aggregation. Phys Rev E Stat Nonlin Soft Matter Phys. 2004 Apr;69(4 Pt 1):041908. PubMed.
Urbanc B, Cruz L, Ding F, Sammond D, Khare S, Buldyrev SV, Stanley HE, Dokholyan NV. Molecular dynamics simulation of amyloid beta dimer formation. Biophys J. 2004 Oct;87(4):2310-21. PubMed.
Urbanc B, Cruz L, Yun S, Buldyrev SV, Bitan G, Teplow DB, Stanley HE. In silico study of amyloid beta-protein folding and oligomerization. Proc Natl Acad Sci U S A. 2004 Dec 14;101(50):17345-50. PubMed.
Yun S, Urbanc B, Cruz L, Bitan G, Teplow DB, Stanley HE. Role of electrostatic interactions in amyloid beta-protein (A beta) oligomer formation: a discrete molecular dynamics study. Biophys J. 2007 Jun 1;92(11):4064-77. PubMed.View all comments by Brigita Urbanc
Aggregation Propensity of Aβ Predicts Longevity of Fly Model of AD
AD pathology is strongly associated with initial stages of amyloid-β protein (Aβ) aggregation. At different stages of Aβ assembly, Aβ oligomers, protofibrils, and fibrils are observed which differ in structure as well as toxic function. In particular, earlier assemblies, oligomers, are known to be toxic to cells in cell cultures and in a transgenic mouse model. While Aβ oligomers seem to be involved in cell death, it is a lot less clear how Aβ oligomers mediate their toxic function. There is an ongoing debate on intra- versus extracellular assembly processes that are associated with toxicity. There are numerous studies indicating strong and potentially disruptive interactions between Aβ oligomers and lipid bilayers, some suggesting that Aβ assemblies form ion channels in the cell membrane, thereby inducing abnormal calcium transport and consequently cell death. It is quite possible that there is more than one potentially toxic pathway of Aβ assembly.
Given the complexity of Aβ oligomer-mediated toxicity in humans and in transgenic mouse models, it is thus quite surprising to encounter a study that shows a clear correlation between the protein's primary structure, which determines its aggregation propensity, and in vivo consequences of such aggregation. In an extensive in vivo-->in silico study, Leila Luheshi and colleagues link the aggregation propensity of full-length Aβ to the neuronal dysfunction in a Drosophila model of AD. Luheshi et al. used single point mutations of Aβ42 wild-type (WT) and its Arctic mutant E22G in combination with a previously reported algorithm to calculate intrinsic aggregation propensities based only on the amino acid sequence. They selected 17 mutational variants out of 798 total, and expressed them throughout the central nervous system of fruit flies. The longevity and locomotor ability of multiple lines of flies for each variant were compared to Aβ42-WT and Aβ42-E22G. Luheshi et al. found a statistically significant correlation between the propensity of a variant to aggregate and its effect on the longevity as well as locomotor ability.
There were a few exceptions, most notably Aβ42-I31E/E22G, where there was no correlation between the predicted aggregation propensity (which was found to be similar to that of Aβ42-E22G) and its effect on longevity and locomotor ability. Further examination of the Aβ42-I31E/E22G showed that while this variant has a high propensity to aggregate into fibrils (similar to Aβ42-E22G), it does not create vacuoles in the brain of flies, which translates into a lack of neurodegeneration. When Luheshi et al. adjusted their computational approach to calculation of propensity to form protofibrils instead of fibrils, they found a significantly stronger correlation between the protofibril formation propensity and locomotor activity/longevity.
As Dominic Walsh noted before me, the definition of a protofibril that Luheshi et al. use in their redesigned computational approach is not clearly explained, but that is crucial to a deeper understanding of processes leading to various longevities of different fly variants. It is not necessarily clear what the cause of death of the flies was. Was a reduced/enhanced longevity in all variants related to increased/decreased neuronal loss? Did Aβ soluble oligomers form at all or is Aβ oligomerization inhibited in this animal model? In a naturally occurring human mutation, i.e., the E22G Arctic, protofibril formation is enhanced . However, recent work by Cheng et al. in Lennart Mucke's lab demonstrated in transgenic mice that overexpress Aβ with the Arctic mutation a significantly higher propensity to form fibrils compared to the wild-type, but reduced functional deficits and reduced levels of deficit-causing Aβ*56 oligomers .
We have to be cautious when extrapolating results from one species to another, in particular because Aβ is very sensitive to relatively small changes in the environment, such as pH, and has a high propensity to interact with other proteins. The present findings by Luheshi et al. provide important insights into aberrant Aβ aggregation and its deleterious effects. They also raise a series of questions that will hopefully be addressed in future studies.
Nilsberth C, Westlind-Danielsson A, Eckman CB, Condron MM, Axelman K, Forsell C, Stenh C, Luthman J, Teplow DB, Younkin SG, Näslund J, Lannfelt L. The 'Arctic' APP mutation (E693G) causes Alzheimer's disease by enhanced Abeta protofibril formation. Nat Neurosci. 2001 Sep;4(9):887-93. PubMed.
Cheng IH, Scearce-Levie K, Legleiter J, Palop JJ, Gerstein H, Bien-Ly N, Puoliväli J, Lesné S, Ashe KH, Muchowski PJ, Mucke L. Accelerating amyloid-beta fibrillization reduces oligomer levels and functional deficits in Alzheimer disease mouse models. J Biol Chem. 2007 Aug 17;282(33):23818-28. PubMed.View all comments by Brigita Urbanc
David Geffen School of Medicine at UCLA
On Computers, Flies, and Alzheimer Disease
Two recently published papers address the fundamental question of how amyloid proteins form neurotoxic assemblies (see Luheshi et al., 2007 and Cheon et al., 2007). Pat McCaffrey has written an informative and insightful news report that summarizes their key findings and implications. The work reported extends efforts by the ”Cambridge group” (broadly defined, and including those in Firenze, Italy; Busan, Korea; and Jülich, Germany) to explore ”generic” protein folding pathways and their biological consequences. In these latest publications, the group extends the idea of generic protein structures to generic toxicity, meaning that protein assemblies that share structural features also share toxic activity. Importantly, algorithms have been developed that allow prediction of assembly state and neurotoxicity from protein primary structure.
The technical rigor of the two studies is excellent. Thus, within the contexts of the experimental systems employed, namely in silico and in vivo (in Drosophila), one may have great confidence in the results. However, now comes the more philosophical and difficult question of meaning. Specifically, how do these results contribute to our understanding of diseases of protein folding?
In this brief discussion, I consider this question and raise a number of others for consideration by the reader. My goal in playing the metaphorical ”Devil's advocate” is to stimulate scientific discourse.
”Meaning” is a nebulous and malleable term for which a definition invariably depends upon the system of evaluation one employs. The goal of the studies of Luheshi et al. and Cheon et al. is to answer the questions of “... the molecular basis of amyloid formation and the nature of the toxic species.” In this context, it is reasonable to ask whether simulating the self-association of Aβ(16-22) or Aβ(25-35) has any relevance (meaning) to Alzheimer disease (AD) or any other disease. Why? Neither peptide is found in vivo. Historically, the former has been a favorite of theorists (including this writer), as its size makes it amenable to in silico study and it forms fibrils in vitro. The latter has been studied since 1990, when the suggestion was made that it was homologous to the tachykinin family of neuropeptides (Yankner et al., 1990). However, the homology relationship was tenuous (as are many when sequence length is so short), authentic tachykinin peptides had no trophic or toxic effects on neurons, and significant evidence supporting the tachykinin connection has not emerged in the subsequent 17 years. Thus, without compelling biological precedent, one must ask what study of these peptides can reveal. For example, are these peptides proxies for holo-Aβ? Clearly, the answer must be ”no,” as the critical determinant of peptide pathogenicity lies at the Aβ C-terminus in the form of the Ile-Ala dipeptide.
Why are people studying what may be irrelevant peptides, and why is such irrelevance not recognized? An answer may come from what, until recently, has been one of the most controversial and contentious fields of modern biology, i.e., prions. The prion theory postulates that the causative agent of a variety of neurodegenerative diseases in animals and humans is composed entirely of protein (no nucleic acid). In the last three decades, the status of this ”protein only” hypothesis in the scientific community has moved from heresy to orthodoxy. However, questions about the scientific appropriateness of this changing perspective have led some, including Laura Manuelidis, to suggest that a re-examination is warranted of ”the objectivity of science and whether it is a myth vanished.” Manuelidis opines that the acceptance of the theory reflects "the peculiarly American sport of betting on popular momentum” (Manuelidis, 2000). A more apropos metaphor, considering that one prion disease is bovine spongiform encephalopathy (“Mad Cow” disease), might have been that of “following the herd.”
Much research on AD could be subject to the same type of criticism. Consider the example of what may be called the "generic” herd. This herd believes that amyloid structure is "generic” because many (most? all?) proteins form amyloids with some common structural organization. Although amyloids, by definition, do share a number of biophysical and spectroscopic features, great structural diversity may be found in the assemblies formed by classical and non-classical amyloid proteins and peptides (e.g., see Sawaya et al., 2007). Importantly, no generic structure outside of the cross-β core of the amyloid fibril has been shown to exist, for obvious reasons. Regions outside of the core, which can be quite extensive in protein, as opposed to peptide amyloids, are likely to influence the biological behavior of the assemblies significantly.
Now, Cheon and colleagues suggest that amyloid formation involves a second generic process, a two-step mechanism of “collapse” of monomers and their subsequent rearrangement into amyloid fibrils. This idea appears to invoke known processes of globular protein folding in the context of amyloid formation, specifically the classical idea of hydrophobic collapse into a molten globule followed by proper arrangement of secondary structure elements to form the native tertiary structure. The idea that some peptides bypass this two-step pathway if they can immediately form hydrogen bonds in their eventual cross-β organization is quite interesting. However, although plausible for short, disordered peptides of the sort studied here, what happens in the common case of natively folded proteins forming amyloid? Here, and as the authors themselves suggest implicitly, factors other than the intrinsic properties of the protein monomer likely moderate amyloid assembly. This increased complexity requires me to question the value of this suggestion of generic mechanisms. Scientists, especially medicinal chemists, need targets. Does a “generic amyloid target” exist? Could a single compound directed at such a target be of value in the treatment of the greater than two score amyloid diseases defined thus far?
Maybe a generic target does exist. In Luheshi et al., studies of the effects of expression of human Aβ42 in Drosophila suggest that protofibril formation correlates with neuronal dysfunction and neurodegeneration. In addition, in a kind of Anfinsen redux (Anfinsen, 1973), an algorithm has been created to predict from primary structure alone the propensity of a protein to form toxic protofibrils. My question: Does the experimental assessment of Aβ-induced locomotor and longevity effects in flies, and its correlation with the toxicity metric, have any relevance to the consideration of Aβ-induced disease in humans? Granted, the same question is sometimes raised as gratuitous criticism of work in a variety of non-human animal models, and it is an easy concern to raise, but that does not diminish the significance of the question.
In closing, it may appear to some that the answers to the questions I have asked are implicit in the construction of the questions themselves. This certainly was not my intention. From a purely academic perspective, I found the publications rigorous, enjoyable to read, and quite thought-provoking. It is the provocation aspect of the experience that operates here, particularly with respect to establishing the meaning of the results and their impact on our shared efforts to understand and treat diseases of aberrant protein folding and assembly.
Luheshi LM, Tartaglia GG, Brorsson AC, Pawar AP, Watson IE, Chiti F, Vendruscolo M, Lomas DA, Dobson CM, Crowther DC. Systematic in vivo analysis of the intrinsic determinants of amyloid Beta pathogenicity. PLoS Biol. 2007 Oct 30;5(11):e290. PubMed.
Cheon M, Chang I, Mohanty S, Luheshi LM, Dobson CM, Vendruscolo M, Favrin G. Structural reorganisation and potential toxicity of oligomeric species formed during the assembly of amyloid fibrils. PLoS Comput Biol. 2007 Sep;3(9):1727-38. PubMed.
Yankner BA, Duffy LK, Kirschner DA. Neurotrophic and neurotoxic effects of amyloid beta protein: reversal by tachykinin neuropeptides. Science. 1990 Oct 12;250(4978):279-82. PubMed.
Sawaya MR, Sambashivan S, Nelson R, Ivanova MI, Sievers SA, Apostol MI, Thompson MJ, Balbirnie M, Wiltzius JJ, McFarlane HT, Madsen AØ, Riekel C, Eisenberg D. Atomic structures of amyloid cross-beta spines reveal varied steric zippers. Nature. 2007 May 24;447(7143):453-7. PubMed.
Anfinsen CB. Principles that govern the folding of protein chains. Science. 1973 Jul 20;181(4096):223-30. PubMed.