A droplet reactor on a super-hydrophobic surface allows control and characterization of amyloid fibril growth

Methods to produce protein amyloid fibrils, in vitro, and in situ structure characterization, are of primary importance in biology, medicine, and pharmacology. We first demonstrated the droplet on a super-hydrophobic substrate as the reactor to produce protein amyloid fibrils with real-time monitoring of the growth process by using combined light-sheet microscopy and thermal imaging. The molecular structures were characterized by Raman spectroscopy, X-ray diffraction and X-ray scattering. We demonstrated that the convective flow induced by the temperature gradient of the sample is the main driving force in the growth of well-ordered protein fibrils. Particular attention was devoted to PHF6 peptide and full-length Tau441 protein to form amyloid fibrils. By a combined experimental with the molecular dynamics simulations, the conformational polymorphism of these amyloid fibrils were characterized. The study provided a feasible procedure to optimize the amyloid fibrils formation and characterizations of other types of proteins in future studies.

To study the flow field driven protein aggregation in real-time, a confined and steady flow is 2 required. Although several mechanical stirring based systems had been reported, none of them 3 showed the ability to do real-time investigations. 1-3 On the other hand, according to our previous 4 research of droplet on SHS, the inherent convective flow in evaporating droplet should be a highly 5 potential candidate as fiber aggregation driving force. 4-6 Here we produced a confined and steady 6 convective flow driven by temperature gradient, T (i.e., a water droplet with given size sitting on 7 a heated substrate, T = T1-T0, T1 is the temperature of hot substrate, T0 is ambient temperature). 8 For a evaporating droplet, its temperature distribution is complicated and difficult to control 9 since it relies on both the contact angle and the ratio of substrate and liquid thermal 10 conductivities. 7 However, consider a water droplet with given size sitting on a heated substrate, 11 its temperature gradient can be expressed as following: 12 where Ts is the substrate temperature and Ta is ambient temperature. 14 At this condition, the temperature gradient will drive two convective flows with converse 15 direction, capillary force based Marangoni convective flow and buoyancy force based natural 16 convective flow. 17 The Marangoni convective flow and its intensity can be described as in Eq. 2: 18 where Ma is the Marangoni number,  is surface tension, T is temperature, T is temperature 20 gradient within the droplet,  is kinematic viscosity of the liquid,  is the thermal diffusivity, and a 21 is the diameter of the droplet. 8 22 Meanwhile, natural convection driven by the buoyancy force is characterized with Grashof 23 number (Gr), which is given by: 24 ( ) 3 3 2 2 1 2 r g a g a G T T T g is the gravitational acceleration,  is the thermal expansion coefficient, a is the characteristic 2 length,  is the kinematic viscosity, T is temperature gradient within the droplet. And according 3 to previous research, buoyancy force is negligible when Gr  2400. 8 4 In this research, for a water droplet with given volume (6.0 L), the only factor affecting Gr is Therefore, for our specific system, when the T  24 K, the buoyancy effect is negligible. 9 The gradient T could be controlled by a hot stage and measured directly with thermal 10 camera (Supplementary Figure 4). Due to the heating transfer from hot stage to droplet, the 11 temperature distribution in the droplet from bottom to the top is not linear at the initial state 12 (Supplementary Figure 4a, the first 100 millisecond once droplet touched the hot surface), 13 however, after 10 s, the heating transfer in droplet tend to steady state and the temperature 14 distribution in droplet from bottom to top is close to linear (Supplementary Figure 4b). In addition, 15 due to the heating transfer from droplet surface to ambient air, the measured T in droplet is 16 about 2 K lower than the setting point.
In the Air phase, ideal gas law was used to estimate the density  and the momentum equation 4 for solving the velocity field (V), laminar flow of Newtonian fluid can be written as: 5 For Liquid droplet, implementation of buoyancy-induced convection, the momentum equation can 7 be written as: 8 Where the liquid viscosity is assumed constant and the density is a function of the temperature 10 (T) using the Boussinesq approximation. 11 In the liquid, the energy balance to take into account the cooling by evaporation, and can be 13 written as: 14 Where Cp is a specific heat capacity, k is a conductivity of the liquid, in the equation (9) the last 16 right side term is a heat released for evaporation, this is a function of latent heat of vaporization 17 Lv and sensible heat Cp(T-T0) is a volumetric mass flow of water from the droplet to the air phase. 18 For the Air phase the energy source term, was neglected. Since the time scale of volume change 19 by evaporation is significantly larger than the thermal and fluid field time scale an assumption of 20 quasi-steady state is acceptable In the Liquid-Air boundary conditions, two principals transport processes were taken into 3 account, heat transfer by convection and stress boundary conditions by Marangoni effect. The 4 Marangoni effect is used to describe the natural convection due to the changes in surface tension 5 by temperature, on the other hand, the Solid-Liquid boundary condition was defined as a constant 6 temperature, the same of the solid surface. The outer of the Air phase was defined as semi-7 spherical geometry with homogenous temperature and set as pressure outlet. 8 ANSYS®, Release 19.1 (Fluent solver) is used to solve the governing equations, using the 9 pressure-based finite volume scheme, the pressure-velocity coupling is solved using the PISO 10 algorithm with PRESTO spatial discretization for pressure and Second Order Upwind for 11 Momentum, Density and Energy equations. The mesh is showed in detail in Supplementary 12  At first, the convective flow was simulated with general Marangoni condition, i.e., surface 7 tension is decreasing with increased temperature. The convective flow in droplet with three 8 different given temperature gradients, T = 5 K, T = 10 K and T = 15 K, were simulated 9 (Supplementary Figure 6). strongly affected by the temperature gradient, i.e., T, and the velocities increase with an 20 increasing T from 0 K to 15 K in agreement with the Marangoni theorem, these profiles are not 1 strictly symmetric due to the disturbance from crosswind airflow. 11 Hence, a continuous 2 convective flow field which can drive the particles in droplet and without contact to the substrate 3 is highly imperative. 4 in the droplet. It is reasonable due to the dissipation of energy is radial from the hot zone to the 2 borders, and the flow movement is going in this direction. However, at low differences in 3 temperature, the magnitude of the up area is increased. 4 In addition, the fluid field in real is affected by Buoyancy and Marangoni effect. To evaluate 5 the effect that these parameters have in the internal fluid of the droplet, five study cases were 6 carried out. In each one of them, a specific phenomenon was studied. So, Buoyancy, Buoyancy simulation corresponds to the non-physical case because it is not possible to decouple these 10 effects, and was made to study the magnitude of the Marangoni effect in the field of fluid, in both 11 cases it is evident that the Marangoni effect preserves the fluid patterns inside the droplet and is 12 mainly responsible for the vortex formation. 13 The deviation of the experimental results is the product of the non-symmetric behavior because 14 the surrounding air is not in stagnation condition. The Marangoni effect was used to describe slip 15 condition in the wall by the effect of the surface tension with the temperature. The internal behavior 16 of the experiments with ∆T equal to 5 K, and 10 K does not show second recirculation area; due 17 to the low surface temperature gradient, the variation of the surface tension is not significant, and 18 the Marangoni effect is negligible. When the temperature gradient becomes higher than 10 K, the 19 surface tension gradient in the air-water surface becomes more important. Therefore, the stress 20 tensor at the wall is higher, and the velocity in the top area is lower than in the bottom, which 21 allows the double recirculation. 22

Supplementary Note 4. Samples preparation and characterization 1
Lysozyme amyloid fibrils were produced starting from a solution of HEWL powder (Sigma) 2 (10 mg/ml) in MilliQ water according to previous reports. 14,15 Acidification of the solution to pH 2.0 3 was obtained by adding 9% (v/v) of HCl (1N) in the solution. The mix was then immersed in a 4 water bath at 60C for 120 h until the formation of amyloid fibrils, as verified by AFM 5 (Supplementary Figure 10). Briefly, the previously formed amyloid fibrils solution was diluted 5 6 times with MilliQ water of which a 30 l drop was spotted on a freshly cleaved mica sheet. The 7 sample was then rinsed several times and gently dried with an N2 flux. AFM was then performed 8 in JPK Nanowizard III by using a XSC-Al-BS probe (MikroMasch, GmbH), run in tapping mode at 9 the resonance frequency of about 120 KHz. In addition, full length Tau  cofactors, such as e. g. heparin. 27 The most used techniques for the determination of intrinsically 25 disordered protein (IDP) structure, are X-ray crystallography 28 , NMR 28,29 and Small-angle Scattering (SAS, both X-ray and neutron) 30 . However, despite the great progress of these 1 techniques, the sample preparation and the accuracy of the results for structure determination 2 are still a bottleneck. An available method to predict the portions of the molecule involved in the 3 amyloid formation, is to test the aggregation propensity of short peptides by crystallographic 4 methods. One of these synthetic peptides, the so called PHF6, with sequence 306 VQIVYK 311 5 (PHF6 in R3 portion of Tau441, PDB file: 2ON9) is widely recognized to be involved in Tau  6 aggregation 31-33 . In this work, both PHF6 and full length Tau441 were used to test our method of 7 protein fiber and IDP formation and to provide an experimental benchmark of the relevant 8 parameters that regulated the responsible physical phenomena. 9 In the case of PHF6 and Tau441, we wanted to verify if at time zero, right before the 10 deposition on SHS, amyloid fibrils were already formed. In Supplementary Figure 10b  The SHS with micro-holes were used for protein self-aggregation and suspension from a 2 sessile droplet. After drying at same condition, even the substrates alone (with no deposited 3 proteins) were characterized by X-ray diffractometer with 2D detector (D8 Venture, Bruker, 4 USA).To minimize the diffraction signal from substrate, the SHS was glued to stand on a tip 5 vertically and orthogonal to X-ray source (Cu target, 50 kV) perpendicularly (Supplementary 6 Supplementary Figure 12). The diffraction patterns were acquired by a detector at 40 mm 7 distance with 600 s exposure time.   Due to our final interest in Tau protein structure we analyzed in our study the fiber structure 1 of PHF6 and Tau441 after formation on our SHS samples. As in the lysozyme case, we performed 2 the PHF6 Raman measurements at different heights along the protein hairpin and analyzed their 3 composition in terms of secondary structure (Supplementary Figure 14). In Supplementary 4 -helix appears. In all these cases the presence of disordered structure detected was not 13 significant. If we compare these results with the ones on the Lysozyme hair-pins, we can notice 14 that PHF6 appears to have a major component of -sheet structure even at extreme conditions 15 (HP-tip). This can be explained by noticing that PHF6 is a short peptide and therefore is 16 predictable that its structure will maintain a certain degree of ordered structure even if snapped-17 off by the breaking of the capillary during drop evaporation. 18

Supplementary Note 9. Fibrils formation from Tau441 protein 1
In the Raman spectra, a strong peak at 1670 cm -1 is evidencing the β-sheet structure 2 formation in the suspended fibrils, the data is shown in main text Figure 7. 3

Scattering (WAXS) Measurements 2
Small Angle X-ray Scattering (SAXS) and Wide Angle X-ray Scattering (WAXS) data were 3 collected at the X-ray MicroImaging Laboratory (XMI-Lab) which is equipped with a super bright 4 synchrotron class table top X-ray micro-source (Cu Ka,  = 0.15405 nm, 2475W, rotating anode) 5 coupled, by a multilayer focusing optics (Confocal Max-Flux; CMF 15-105), to a SAXS/WAXS 6 three pinhole Rigaku SMAX-3000 camera 36 (Supplementary Figure 14a) Figure  4 7e shows two major peaks, marked as "1" and "2", plus other faint additional peaks. These two 5 major peaks were fitted by Gaussian function (Main text Figure 7f) to determine the peak position 6 of 0.51 and 0.64 Å -1 , which means that the lattice planes distances are of 12.30.5 ("1") and 7 9.80.5 Å ("2"), respectively. The same procedure of data reduction was repeated also for Main 8 text Figure 7g, resulting in the profile in Main text Figure 7h. Here, apart for peak "2", which is 9 slightly moved with respect to the same peak marked as "2'" in Main text Figure 7e, additional 10 peaks were measured. The fit of the most intense of all, marked as peak "5", is reported in Main  11 text Figure 7i. The peaks and corresponding distances measured for Tau441 are shown in 12 Supplementary Figure 14i. 13 Therefore, with the conjoint analysis with Raman, XRD and WAXS, we can confirm that these 14 suspended fibrils are highly aligned β-sheet amyloid fibrils. In addition, we suggest that these 15 fibrils have long range anisotropic order along the fiber axis. 16

Supplementary Note 11. Molecular dynamics (MD) calculation 1
Supplementary Note 11.1. Molecular dynamics protocol 2 Molecular dynamics simulation is a powerful method to understand proteins providing an 3 atomistic view of molecular interactions and dynamics. 37-42 We applied molecular dynamics (MD) 4 in explicit waters on PHF6 and Tau441 fibril models in order to have detailed insights of the fibril 5 structural properties. Our trajectory analyses corroborate the stability profiles, experimentally 6 observed, of the fibrils. 7 Molecular dynamics simulations of PHF6 (50ns) and Tau441 (two morphologies for 30ns 8 each) were performed applying the same protocol. In details, the systems were solvated in an 9 octahedron box using the TIP3P water models 43 with a 1.2 nm distance to the border of the 10 molecule using GROMACS (5.1.2) and parametrized with Charm27 force field. 44 al. with X-ray diffraction. 32 In details, several consecutive modules of the 2ON9 pdb coordinates were firstly acetylated at the N-terminus and then replicated along both longitudinal and 1 orthogonal fibril axes, adapting a previous modelling protocol. 47 it is a building block to understand the structure of the more complex case of Tau441. We created 17 a PHF6 fibril composed by 144 peptides in pre-defined β-sheet conformation using the 18 coordinates of the pdb X-ray structure which shows highly ordered cross β PHF6 architecture 19 (method section for details, Supplementary Figure 15). The obtained model represents an 20 elongated/a grown fibril state of PHF6 oligomer solved by Sawaya which is the repeat unit of 1 PHF6 aggregates, 33  of the ending frame of PHF6 Molecular Dynamics. Only the peptide heavy atoms of the first sheet 10 are displayed in sticks and colored by type (oxygen: red, Nitrogen: blue, carbon: with). Red arrows 11 represent the βsecondary structure. c, PHF6 MD2bis, center of mass distances of the C atoms 12 of each Valine residue versus the facing Isoleucine within each dimer along the PHF6 simulation. 13 The top panel report the mean values of all the 144 distances averaged along the simulation. d, 14 six PHF6 dimer peptides forming a fibril. For clarity, the peptide backbone heavy atoms of the first 1 layer are displayed in sticks and colored by type (oxygen:red, Nitrogen: blue, carbon: with), in one 2 PHF6 dimer all heavy atoms are displayed to show the face-to-face arrangement of steric zipper 3 arrangement. Red dashes represent the distances between the C atoms of facing Valine and 4 Isoleucine residues 5 6 The Cα-root mean square deviation (RMSD) profiles respect to the starting model suggest 7 that the overall system is stable till the end of the trajectory, with a slightly higher level of 8 rearrangements affecting the terminal regions (layers 1-2 and 11-12). Indeed, the core fibril 9 (layers 3 to 10) show RMSD mean values spanning from 0.24 to 0.31 nm, but those of the first 10 and the last layers reach mean values of 0.45 and 0.46, respectively (Supplementary Table 4   Molecular dynamic simulations were performed starting from the Cryo-EM atomic models of 2 paired helical (p-helical) and straight Tau filaments (s-filaments), recently purified and 3 characterized from AD brain by Fitzpatrick (Cryo-EM). 50 These Tau441 morphologies are both 4 composed by residues 306-378 of Tau protein, which represent the seed regions of Tau  5 aggregation. The residues arrange into combined cross-β and β-helix structures to form a double 6 helical stack of C-shaped subunits. The p-helical and s-filaments supermolecular inter-7 protofilament packings differ in the lateral contacts between two fibril subunits each composed by 8 5 layers of Tau441 core regions (Supplementary Figure 19) Our simulations suggest a higher variability of inter-molecular with the respect to the intra-4 molecular distances (standard deviations in Supplementary Table 10 and Supplementary  5   Table 11, respectively), likely explaining the smaller pick density at 12.5 Å of XRD measurement. 6 Indeed, these connections are responsible of supermolecular architecture which represents a 7 second step of aggregation, weaker than the intra-molecular interactions. MD trajectories suggest 8 that the s-filaments proto-filament subunit interface confers greater global stability than that in p-9 helical arrangement. The simulations could clarify the pick at 12.5 Å of XRD measurement 10 registered only at high temperature, suggesting that the laminar convective flow induces the 11 ordered molecular assembling, and that the fiber growth formed at high temperature preferentially 12 arrange in the more stable s-filament morphology. 13 Our findings reveal that in the p-helical filaments some structural elements are less stable 14 than the s-filaments morphologies. The s-filaments proto-filament subunit interface confers 15 greater stability than that in p-helical arrangement, but for both systems no major fibril 16 perturbations occur during our simulations and the global fibril compactness is maintained as 17 shown in Supplementary Figure 19.