Infrared nanospectroscopy characterization of oligomeric and fibrillar aggregates during amyloid formation

Amyloids are insoluble protein fibrillar aggregates. The importance of characterizing their aggregation has steadily increased because of their link to human diseases and material science applications. In particular, misfolding and aggregation of the Josephin domain of ataxin-3 is implicated in spinocerebellar ataxia-3. Infrared nanospectroscopy, simultaneously exploiting atomic force microscopy and infrared spectroscopy, can characterize at the nanoscale the conformational rearrangements of proteins during their aggregation. Here we demonstrate that we can individually characterize the oligomeric and fibrillar species formed along the amyloid aggregation. We describe their secondary structure, monitoring at the nanoscale an α-to-β transition, and couple these studies with an independent measurement of the evolution of their intrinsic stiffness. These results suggest that the aggregation of Josephin proceeds from the monomer state to the formation of spheroidal intermediates with a native structure. Only successively, these intermediates evolve into misfolded aggregates and into the final fibrils.


Comparison of Josephin Aggregation on ZnSe and Mica Substrates
In order to compare the aggregation process on the ZnSe prism ( Fig. 1 and Supplementary Fig.   1) with the previous experiments in literature on charged substrates, 1 we performed conventional AFM measurements in air of the sample deposited on positively functionalized mica ( Supplementary Fig. 2).
The AFM measurements on the mica substrate were conducted on the same solution and in parallel with the infrared nanospectroscopy experiments on the ZnSe prism. We investigated the process at the same time points: 0, 2 and 7 days. At 0 days, before incubation, we could observe only monomers and spheroidal oligomers (height ≈30 nm). At 2 days, the images showed interacting oligomers losing their spherical shape and high resolution measurements evidenced flexible and beaded prefibrillar structures. These results were in agreement with what shown by Masino et al. 1 Furthermore, the presence of interacting oligomers losing their spheroidal shape is well corresponding with the measurements on the ZnSe prism ( Fig. 1-2). Finally, at 7 days, we could not observe stiff fibrillar bundles and aggregates as in the case of the sample deposited on the ZnSe; only low concentrated fibrillar species on mica were present. This is likely caused by the hydrophilicity of the mica surface that did not allow the stable deposition of hydrophobic fibrillar aggregates and bundles. Probably because of a weak sample-surface interaction, these structures do not attach firmly to the mica substrate and that they were rinsed away in the last step of sample's preparation (rinsing by water and flushing by nitrogen).
The morphological differences between structures observed on positively functionalized mica ( Supplementary Fig. 2) and the structures on the ZnSe prism can be explained by a differential

NanoIR principle and sensitivity
The combination of AFM and IR spectroscopy is based on a photothermal induced resonance effect (PTIR). Briefly, if a pulse of IR light at a given wavelength is absorbed by a sample, the local rise in temperature leads to local photo-thermal expansion. This expansion excites the mechanical resonances of the AFM cantilever in contact with the sample. The AFM detection of this temporary dilatation of the scanned region allows reconstructing, together with the acquisition of conventional morphology imaging, the IR-absorption map of the sample and its chemical spectra with a lateral resolution defined in principle by the dimensions of the AFM tip. 7,8 The critical thickness (AFM average height) of the sample needed to acquire IR spectra is S12 in the order of 50-100 nm and is determined by the ultimate instrument sensitivity to the measurement of the photo-thermal expansion. 8 By acquiring IR maps at different wavenumbers and spectra at specific locations, we could retrieve the spectroscopic properties of native oligomers as small as 50 nm in average height (Supplementary Fig. 3-4). This is one of the main advantages of nanoIR compared to conventional IR techniques, which have spatial resolution limited by the smallest achievable IR spot size (~20 m). 9 NanoIR also requires very small quantities of biological specimens (less than picograms or sub-femtomoles are sufficient). 15 This is a major advantage when working with biological materials whose production is expensive and time costly.
The resolution of the instrument is clearly showed in the Supplementary Fig. 4 Finally, it is worth to note that sensitivity of the instrument in the spectra region within 1610-1550 cm -1 is limited by the extremely low laser power in this region (Supplementary Fig.   5), which causes low signal-to-noise ratio. This is clear in the Supplementary Fig. 4, where the signal on the aggregate is clearly distinguishable from the one on the substrate in the entire spectral range, except in the amide II band. For this reason, in this region, the acquisition of chemical properties of objects with thickness close to the sensitivity limit (average height of 50 S13 nm) is affected by high noise. Thus, spectra of oligomers with average height close to 50 nm and amide II band between about 1590-1560 had extremely low signal to noise ratio. While, in the case of fibrillar structures, the lengths of several micrometres, average heights as big as 200 nm and a shift at lower wavenumbers, where laser power is more intense, allowed to increase the signal to noise ratio in the amide II bands.

Secondary Structure Estimation of Josephin before incubation
We estimated the secondary structure content of Josephin before incubation at 37 ºC by deconvolution of the amide I band of the spectrum of the uniform protein aggregates before incubation (Fig. 3f). The analysis was performed by XPSPEAK and a -squared of at least 0.001 was reached for the statistical significance of the fit. Although the shape of the band is affected by the high noise between 1610 cm -1 and 1550 cm -1 , we could estimate a secondary structure content of 33% -helix, 29% random coil, 23% -sheet and 15% -turn.
The shoulder of the spectrum around 1715-1725 cm -1 was fitted by considering the IR absorption peaks of the side chain vibrations of aspartic and glutamic acids. 10 The addition of this band in the fit is easily explicable considering that these amino acids are relatively abundant in the Josephin structure (~15% of total composition, ~30% more abundant than in the average protein composition). Moreover, these residues have stronger absorbance coefficients than the average side chains absorption and they are not superimposed to any other secondary structure component of amide I band.

Principal Component Analysis
Unprocessed spectra show pronounced baseline effects. To avoid the possible influence of these baselines on the results and also to present in details the spectral differences related to S14 the conformational transition of amyloids, PCA was also performed on the second derivatives of the collected data (Supplementary Fig. 7-8). In this case, we considered the opposite situation in the interpretation of the correlation between spectra clustering and Loading Plots ( Supplementary Fig. 8a-b): the maximum (minimum) of loading is related to the position of a particular vibrational motion, which is typical of the spectra clustered on the negative (positive) values of corresponding Principal Component (PC) on the Scores Plot.
Three clusters of three groups of spectra collected from i) native oligomers, ii) misfolded oligomers and iii) fibrils are well delineated on the three dimensional (PC-1 vs PC-2 vs PC-3) Scores Plot (Supplementary Fig. 8a). Two additional Scores Plots (Supplementary Fig. 8c,d) are shown to visualize the separation along each PC. However, the boundary between a cluster of spectra collected from native oligomers and misfolded oligomers is not clearly evident. This suggests that their conformation is more similar in comparison of the conformation of fibrils.
The Loadings Plot (Supplementary Fig. 8b) indicates that PC-1 (33 % of total variance) is dominated by amide I, which is the most significant marker of amyloid conformational changes.
The PC-1 is negatively correlated with an antiparallel -sheet/-turn band at 1710 cm -1 . This indicates that fibrils (oppositely correlated with a positive PC-1), contain higher levels of antiparallel -sheet/-turn conformation in comparison to native oligomers (negative PC-1). In amide I, the PC-1 is positively correlated to the infrared motions corresponding to native sheets at 1628 cm -1 , which confirms the presence of this conformation in native oligomers. The PC-1 indicates also a shift of the COO-band from ~1435 cm -1 (fibrils) to 1410 cm -1 (native oligomers). This band dominates the PC-2, showing also a strong positive correlation at 1435 cm -1 for fibrils and, partially, for misfolded oligomers (negative PC-2), while for native and misfolded oligomers there is a negative correlation. This is another clear indication that the peak S15 at 1440 cm -1 is typical of amyloid fibrils. The separation along PC-2 can also be explained by the negative loading for PC-2 in the spectral region that corresponds to amide III at 1309 cm -1 (oligomers) and positive at 1290 cm -1 (fibrils). PC-3 (17 % of total variance) is dominated by the spectral features attributed to -helix at 1658 cm -1 , which is oppositely correlated to all spectra of native oligomers, confirming their high content of this secondary structure, and about 50 % of misfolded oligomers spectra.