Proteins that form amyloid fibrils often break a rule of protein folding; the amino acid sequence does not necessarily encode for a single, unique structure. In the case of the amyloid beta peptide (Aβ), infamous for its association with Alzheimer's disease, a single 42 amino acid peptide can fold and assemble in many different ways, and these folded peptides can be found stacked on top of each other in the form of amyloid fibrils. Thanks to developments in technology, particularly with respect to cryogenic electron microscopy (cryoEM) over the last 10 years or so, we have learned a lot about the structures of amyloid fibrils at near-atomic levels of detail. Something that keeps coming up is that the same sequence can often generate many different fibril folds. However, despite this, when amyloid fibrils are extracted from the brains of patients with varying diseases associated with the formation of such fibrils, it is often reported that there is a relationship between the structure of the fibril and the disease, and this is most commonly observed in both tauopathies and synucleinopathies. In the case of Aβ, fibrils from patients with Alzheimer's have been found to fold more heterogeneously, but these fibrils are also often different to the fibrils formed in vitro. These results have led to a key hypothesis, that there are disease specific fibril strains, meaning a disease relevant fibril structure rather than any of the many other fibril structures that could have been made from the same protein sequence. If we can develop an understanding of how these specific strains come about in vivo then we will unlock vital information for our fundamental understanding of the diseases that they are involved in.
In previous work we established that we can very effectively map out the landscape of different amyloid fibril structures using atomic force microscopy. This works by making a variety of measurements from a three-dimensional reconstruction of the surface envelope of the individual fibrils, a feature which is extremely sensitive to variations in fibril structure.
So, we then set out to understand how the landscape of Aβ fibril structures changes when the fibrils are grown in different ways. Well, where do you start with that? There are many ways to grow a fibril, many of which have nothing to do with what happens in a patient’s brain. Do we start then, with complicated ideas about extracellular environments and biological relevance? How about with the presence of potential chemical modifiers? Good ideas for the future maybe but what we decided to do for this work was to have a look at how other labs have been growing fibrils. There is after all many decades worth of research into the formation of Aβ fibrils and the consequences of such formation. Where better to start characterising the landscapes of fibril structures than by characterising the fibrils that researchers have been growing in their labs for over a decade? To do that, we hunted through the literature for popular papers involving fibril assembly. This became a more prominent task for many of us once the covid pandemic hit and we looked for tasks that didn’t require a lab whilst the lab itself was shut down. We ended up with a list made up of the 20 most highly cited publications from each of the years between 2000 and 2020, over 400 papers in total. From these papers we extracted a list of “buffers” following the rule that monomeric Aβ must have been converted into fibrils in that buffer in the paper. Expectedly, this generated a lot of varied information that we needed to “boil down”. The buffering agents that were used immediately became an obvious source of variation with most researchers choosing one of HEPES, Tris or phosphate. We followed suit, growing fibrils in one of HEPES, Tris or phosphate, all at pH 7.4 and repeated the phosphate growth at pH 8 noting that this exact condition was used in many experiments where the aim was to study the mechanisms of fibril assembly.
Having made our 4 sets of fibrils we set out to examine them, much the same way we had previously established, we deposited them onto mica, a flat surface amenable to AFM imaging and collected large datasets. From each image, we extracted information about all of the fibrils on the image that had a non-overlapping segment and an identifiable twist within that segment with no visible breakages. More simply, we used every fibril that we could tell by eye was just one fibril and not multiple overlapping fibrils. For each of these individual fibrils, we reconstructed their surface envelopes, plotted some of the key features such as twist and thickness together to get a visual representation of the landscape of fibrils, and used a hierarchical clustering analysis to generate a “family tree” quantifying the differences and similarities of the over 400 fibrils in our total dataset. Finally, we made comparisons between the fibrils in our dataset and the fibrils structures that have been solved by cryoEM to near-atomic resolution.
What we found was that subtle changes such as changing just the buffer salt can change the landscape of fibrils formed. Many fibril structures did overlap between samples, but the landscape could shift. Some examples of this are when fibrils were grown at pH 8 rather than 7.4 there was an increase in the number of thicker fibrils, and when grown in Tris rather than the other buffers, the fibrils twisted more. Additionally, the fibrils were most homogeneous when grown in the HEPES buffer although all samples displayed some heterogeneity.
When compared to the cryoEM structures we noticed something striking. During our analysis of these data, a manuscript was published detailing the high-resolution structural maps of ex vivo Aβ fibrils from the brains of patients with Alzheimer’s disease. Unexpectedly, when comparing the surface envelope to each of the fibrils in our sample we found a small number of fibrils that were an extremely close match! These fibrils matched more closely compared to the ex vivo structures than any of the other cryoEM structures to any of the fibrils in our sample. These rare matching fibrils make up a total of about 3% of the total in vitro sample. Them being such a minority percentage makes it difficult to extract them for further study but 3% is a much larger number than we were expecting (in that 3% >> 0%). So maybe then the sequence does encode a favourable structure, and the years of accumulation of fibrils in a patient’s brain end up biasing formation towards exclusively that one. Or maybe just 3% of the de novo (grown from monomers only) fibrils is enough to eventually seed a dominant polymorph? Ultimately these are the answers we seek to find out and our approach to structural analysis of the whole fibril population by looking at the individual fibrils by AFM will make it possible to do so. What we already found here is that we can manipulate the populations of fibril structures by making subtle changes to how they grow and that these samples likely always contain a small population of disease relevant structures.