Direct Observation of Heterogeneous Amyloid Fibril Growth Kinetics via Two-Color Super-Resolution Microscopy

The self-assembly of normally soluble proteins into fibrillar amyloid structures is associated with a range of neurodegenerative disorders, such as Parkinson’s and Alzheimer’s diseases. In the present study, we show that specific events in the kinetics of the complex, multistep aggregation process of one such protein, α-synuclein, whose aggregation is a characteristic hallmark of Parkinson’s disease, can be followed at the molecular level using optical super-resolution microscopy. We have explored in particular the elongation of preformed α-synuclein fibrils; using two-color single-molecule localization microscopy we are able to provide conclusive evidence that the elongation proceeds from both ends of the fibril seeds. Furthermore, the technique reveals a large heterogeneity in the growth rates of individual fibrils; some fibrils exhibit no detectable growth, whereas others extend to more than ten times their original length within hours. These large variations in the growth kinetics can be attributed to fibril structural polymorphism. Our technique offers new capabilities in the study of amyloid growth dynamics at the molecular level and is readily translated to the study of the self-assembly of other nanostructures.

at labeling ratio 1:20, labeled with Alexa Fluor ® 647 vs. unlabeled monomeric protein. We thus optimized the dye labeling density and found the 1:20 ratio as an optimal one in order to further reduce any potential interference of the labeling with the aggregation process and to improve the localization precision (see section Super-resolution fluorescence microscopy below). This yielded optimum precision at low to moderate laser excitation powers. Finally, the incubation at 37 • C took place in an Eppendorf tube in 20 mM PB and samples at different time points were removed for subsequent microscopy imaging, in the absence of shaking or mixing.
Super-resolution fluorescence microscopy by single molecule localization and dSTORM imaging. Single and two-color super-resolution imaging was performed with a dSTORM microscopy setup, based on a Nikon Eclipse TE 300 inverted widefield microscope and a 100x, 1.49 NA TIRF objective lens (Nikon UK Ltd.). For dSTORM imaging, samples were placed on a glass or quartz coverslip glued to the bottom of an imaging chamber. Photoswitching buffer solution was added which consisted of 100 mM mercaptoethylamine (MEA) in phosphate buffered saline (PBS, pH 7.4), together with a glucose-enzyme oxygen scavenger (40 mg/ml glucose, 50 µg/ml glucose oxidase, 1 µg/ml catalase). The chamber was filled to the top and sealed with a glass coverslip to minimize entrance of oxygen. For the two-color imaging, laser illumination was at 640 nm (Toptica Photonics AG, Graefelfing, Germany) for excitation of the Alexa 647 dye (red channel) and at 561 nm (Oxxius SLIM-561) for excitation of the Alexa 568 dye (green channel). They were collimated and combined by dichroic mirrors and a beam expanding telescope. The laser beams were subsequently focused onto the back focal plane of the objective. A 405 nm (Mitsubishi S3 Electronics Corp., Tokyo, Japan) laser was used as reactivation source. In order to separate the individual emissions from the two channels (red and green), the fluorescence light in the detection path went through a dichroic filter (Semrock multi-edge filter Di01-R405/488/561/635-25x36 followed by a FF01-446/523/600/677-25 filter, Semrock, Rochester NY, USA) and was subsequently filtered further using band-pass filters (Semrock BP-607/35-25 and BP-642/35-25 for the green and the red channel respectively), before being projected onto a low-noise, highly sensitive electronmultiplying CCD camera (Ixon DV887 ECS-BV, Andor). The excitation intensity was 2 kW/cm 2 for the Toptica laser and 5 kW/cm 2 for the Oxxius laser. The reactivation laser was only turned on when the number of active fluorophores in the field of view was reduced and no spatial drift of the sample was observed during the acquisition time of the two channels. Imaging was performed in TIRF illumination in all cases at the exact centre of an area consisting of 64x64 camera pixels, corresponding to an area on the sample of 10x10 µm 2 . Typically, 10.000-20.000 fluorescence frames with 10-12 msec exposure time were recorded; the exposure time was matched with the "on" state of the fluorescent dyes. From each image stack a reconstructed dSTORM image was generated by using in-house developed software 3 based on MATLAB (The MathWork Inc., Natick, USA). The 1:20 labeling ratio (labeled vs. unlabeled protein) was found to be optimum for dSTORM imaging.
High labeling density (e.g. 1:1, 1:5, etc) led to frequent mislocalizations or overlapping and thus asymmetric point spread functions (PSFs) which were rejected by the localization algorithm.
Super-resolution algorithm for localization and visualization. To obtain super-resolved images of the samples, the raw image data was processed using the Open Source rainSTORM localization microscopy software. 3 To corroborate the robustness of this method, it was confirmed that similar results were obtained using the rapidSTORM software package. 4 The rainSTORM analysis and visualization was as follows: fluorophore positions were determined by applying a "sparse segmentation and least squares Gaussian fitting" algorithm to the data. 5 Imprecise localizations were excluded by a "quality-control" process: localizations were rejected if their localization precision, estimated using the Thompson formula, 6 was worse than a user-defined threshold and if S4 they corresponded to asymmetric, or to too wide or too elliptical PSFs, corresponding to out of focus emission. The accepted localizations were visualized using an Adaptive Gaussian Rendering method. The average localization precision achieved in our imaging was 12 nm.
Image and Histogram Analysis. For the two-color localization microscopy images, chromatic aberration between the red and green channels was evaluated using the method described in Ref. 7 , and found to be small compared with the obtained resolution. Data and two-color co-localization image analysis was performed using our software based on MATLAB 3 to quantify corresponding localization microscopy parameters such as localization precision and standard routine analysis with ImageJ (NIH, Bethesda, Maryland, USA). Cross-talk between the green and the red channel was estimated to be 16% and was subtracted from the images.

Supplementary Figures
The presence of the dye at residue N122 does not affect the elongation of α-synuclein from preformed seeds. The labeled protein α-synuclein N122C follows the same aggregation kinetics as the unlabeled protein, as illustrated by the superposition of the two normalized kinetic traces describing the fibril elongation of the unlabelled and labelled protein in Suppl. Figure S1. Suppl. Figure S2 shows the time course study with AFM of fibril growth for the same time points as the ones used in the dSTORM experiment. Samples were left to dry before imaging. Fragmentation of fibrils can thus often occur, and also the formation of fibril clusters results in difficulty in length measurements. Additionally, AFM imaging does not allow to distinguish between the seed and the attached monomeric/elongated part of each fibril and thus it is not possible to determine the individual fibril elongation rates.
Seeds catalyse the aggregation process and no end-to-end association is observed in vitro.
We incubated monomeric protein labeled with AF647 at 37 • C for 24 h in the absence of seeds, in quiescent conditions and at the same concentration as the one used for the seeded assay. Suppl. we investigated whether the seed fibrils could, if incubated under similar conditions as the ones mentioned in the previous experiments, give rise to longer fibrils. This would suggest that seed to seed association and further aggregation is possible. We therefore incubated α-synuclein seeds S7 labeled with AF568 at the same concentration as in the seeded assay for 24 h at 37 • C. Suppl. Figure S3c) shows the fluorescence image and S3d) shows the corresponding dSTORM image of the same area. In this experiment we confirm that the sample consisting of seeds only is stable over time and no further aggregation occurs. This observation suggests that α-synuclein seed fibrils in the absence of monomeric protein are not able to grow further. S8