Fibril formation and ordering of disordered FUS LC driven by hydrophobic interactions

Biomolecular condensates, protein-rich and dynamic membrane-less organelles, play critical roles in a range of subcellular processes, including membrane trafficking and transcriptional regulation. However, aberrant phase transitions of intrinsically disordered proteins in biomolecular condensates can lead to the formation of irreversible fibrils and aggregates that are linked to neurodegenerative diseases. Despite the implications, the interactions underlying such transitions remain obscure. Here we investigate the role of hydrophobic interactions by studying the low-complexity domain of the disordered ‘fused in sarcoma’ (FUS) protein at the air/water interface. Using surface-specific microscopic and spectroscopic techniques, we find that a hydrophobic interface drives fibril formation and molecular ordering of FUS, resulting in solid-like film formation. This phase transition occurs at 600-fold lower FUS concentration than required for the canonical FUS low-complexity liquid droplet formation in bulk. These observations highlight the importance of hydrophobic effects for protein phase separation and suggest that interfacial properties drive distinct protein phase-separated structures.


S1
Supporting Information (SI)   Figure S1. Experimental scheme to study lipid/protein interactions: the trough is filled with the PBS buffer, the probing needle of the SP sensor calibrated at the air/water interface is inserted into the subphase, lipid monolayer used as a model biomembrane is spread at the air/buffer interface, SFG is typically measured before and after addition of a protein solution into the subphase.

II. Humidity: Experimental conditions of SFG and SP measurements.
In the experimental enclosure, we flush with nitrogen continuously and measure the humidity with the temperature/humidity sensor "EBI 20-TH1 Standard-Temperatur-/Feuchtedatenlogger".
According to the technical information, the systematic error is (depending on the relative humidity (RH) value, in brackets): ±3% RH (30% RH ... 90% RH), ±5% RH (10% RH ... 30% RH ). As we see, below 10% RH, the value is actually in the range of 0-10% RH. However, the more important factor for us is not the exact humidity value, but rather the ability to provide experimental conditions to sufficiently reduce water IR absorption bands that would "contaminate" an SFG spectrum. The water bending mode, in particular, would complicate the interpretation of the amide I region (see Figure S2). The nitrogen flushes out water vapor and thus allows us to suppress water absorption bands. In this sense, the amplitude of water bands acts like an in situ humidity sensor. In our experimental setup, we almost completely remove water bands, which puts an upper bound on the RH of <5%. Nitrogen flushing allows to get rid of water absorption bands, as clear from blue curve, from which we infer an RH less than 5%.
As an additional demonstration of the concept, we present an experiment where we record nonresonant SFG signal from z-cut quartz while varying humidity in the sample box. At each humidity value, we record both SFG signal and background (by blocking the IR beam), subtract background from the signal (and thus get the background-corrected signal). Then for different RH from the sensor, the background corrected spectrum at ~0% humidity (completely purged) is subtracted from the background-corrected spectrum recorded from each RH. We perform the experiment in the free OH region (peak centered at ~3750 cm -1 ), since the water band structure is stronger and easier to interpret compared to OH bending. In this way, we measure the water band intensity at each RH value and plot the number of counts at the main peak centered at ~3760 cm -1 versus the RH value ( Figure S3). Note that we added (0;0) point as a natural "boundary" condition (no water vapor = no bands). A linear correlation between the water band amplitude and RH measured by humidity sensor is evident from the graph.
Since we flush the sample box with nitrogen in such a way that no water bands can be observed in the spectrum, the RH is <5% can be empirically determined with a good level of precision, although the systematic error of the sensor not reported in that range.

III. CAPS molecule in pH 11: Control experiments on interfacial behavior.
The CAPS buffer at pH 11 was used to stabilize FUS LC molecules in the monomer state as outlined in by Burke et al. 1 To the best of our knowledge, there have not been any studies reporting the interfacial behavior of CAPS molecule. Thus, we have performed additional surface-specific experiments for the conditions relevant to our work.
In Figure S4, we present SP measurements. The SP is first calibrated to the air/PBS buffer interface (at t~0s), and then the SP measurement is started. At time t~100s, the trough with PBS buffer solution is carefully drained, and the SP probing needle becomes free (i.e. out of the sample). As the definition of SP is = 0 − , the surface tension = 0~7 2.5 mN/m for the needle out of the sample, exactly what we observe. At t~180s, we carefully fill the trough with 20 mM CAPS pH 11 buffer to immerse the end of needle into the liquid, as marked with an arrow in Figure S4. For our measurement, we use a height sensor (Keyence LK-G82 Lasersensor, precision ±1 μm) to keep the sample height constant, thus eliminating the effect of a different depth of needle immersion after the sample exchange. We observe a ~ 1.5 mM/m change in surface pressure when changing the PBS buffer to CAPS. Therefore, we conclude that the minor population of CAPS molecules is present at the interface and change the SP only slightly (in the range of ~1.5-2 mN/m) compared to PBS. In addition to SP measurements through complete buffer exchange, we doped CAPS into a PBS subphase at precisely the same dilution, the same as was done when introducing monomeric FUS LC in the SFG experiments. The SP measurement was started at t~0s (see Figure S5), and at times marked with black arrows, 100 μL of CAPS buffer solution (20 mM in miliQ, pH 11) was added with a glass syringe into the sample trough (no protein was added), so that the final CAPS      We record kinetic SFG spectra until no changes in spectral shape and intensity are observed (in average, this is reached in ~2 hours after FUS LC injection into the PBS subphase). Both kinetic as well as steady-state SFG spectra present an amide I band centered at ~1670 cm -1 and does not contain any noticeable feature centered at ~1620 cm -1 . Because of the kinetic nature of the film formation, it is challenging to determine whether the final state we measure is the true thermodynamic equilibrium state or, rather, a long-lived metastable state. We perform measurements of liquid samples with an open surface, so the effect of evaporation is unavoidable, and this limits the experiment time, especially in comparison with bulk studies of closed samples, e.g. in Eppendorf tubes or 96-well plates. Nevertheless, we can definitively say that on timescales up to ~3.5 hours, we see no additional peaks (including the resonance at 1620 cm -1 ), but rather only the slight growth of the already-present peak in the SFG response (see Figure S11).
In fact, from our AFM experiments (reported in the main text in Figure 1) and from FTIR data ( Figure S16) we can claim: (1) for AFM measurements, the film was allowed to develop at the airwater interface on over a few hours, was transferred to the solid substrate, and fibrils were observed; (2) a comparison of FTIR measurements of the film on the liquid surface with the transferred film revealed a very similar infrared response for the on-liquid vs. on-solid films ( Figure   S16). Thus, from the statement (2) follows: the two films, i.e. on-liquid and on-solid, are structurally very similar. Finally, from the statements (1)&(2) follows: fibrils were present already on the liquid surface.

XI. SFG spectra for FUS LC at the air/PBS buffer interface at different FUS LC bulk concentrations.
We tested the effect of protein bulk concentration on the adsorption and organization of FUS LC at the air/PBS interface. Various final concentrations of FUS LC in PBS were tested, namely 1.5, 3, and 6 μM. No major differences in SP were observed ( Figure S13a), all three curves are fluctuating, and only for 6 μM the SP is slightly higher (by ~3 mN/m). Figure S13b shows that the SFG C=O signal is the same for all concentrations while the amide I intensity is different. The spectral shape observed for 6 μM FUS LC is different from that for 1.5 and 3 μM. Fitting is needed to disentangle the contributions from different folding motifs at each protein concentration. Fitting of the amide I mode contribution for the SFG spectra presented in Figure S13b was performed in the frequency region 1615-1705 cm -1 . The results are presented in Table S1. For each considered FUS LC concentration, the ratio of the integrated SFG intensity of the peak at ~1671 cm -1 (assigned to different β-folded structures) [2][3][4][5][6][7][8] to that of the peak at ~1656 cm -1 (assigned to the S13 random-coil and/or α-helix structures) 9 was calculated based on fitting results and is shown in the inset in Figure S13b. As evident from the inset, the intensity of the β-contribution decreases upon increasing the protein concentration. Since protein fibrils are commonly densely packed and enriched with β-folded structures, 10 we assume that the observed trend for the different protein bulk concentration arises from the different the orientation of FUS LC molecules folded into βconformation at the interface. (b) SFG spectra in the amide I and carbonyl stretching region after equilibration (t~5000s) for (blue) 1.5 μM, (green) 3 μM, and (red) 6 μM, respectively. Inset: the ratio ISFG(β)/ISFG(random-coil and/or α-helix) as obtained from the spectral fitting of the experimental data plotted against the FUS LC concentration in PBS.

S14
(c)-(e) Resonant contributions (see Table S1) in the amide I region presented in graph (b): α-helix/random coil (gray) and β-folded (cyan). Table S1. Results of the fitting for SFG spectra presented in Figure S13b. Two FUS LC stock solutions were prepared at different concentrations, namely 75 μM and 37.5 μM so that 200 μL of the former or 400 μL of the latter were injected in the PBS buffer to reach a final FUS LC concentration of 3 μM. The obtained SP and SFG data are shown in Figure S14a and S14b, respectively. In Figure S14a no major differences in SP are observed, and the final SP values are similar. Figure S14b shows the intensity of the amide I signal is different, but not its spectral shape suggesting that the protein conformation is similar. Taking into account the similar SP values (and thus the similar amount of the FUS LC protein adsorbed at the interface), we speculate that the difference in the amide I signal intensity (but not shape) could arise from the different protein backbone orientation. The indistinguishable carbonyl stretching mode responses for both experiments suggests that the protein side-chain conformation is also very similar.

XIII. FTIR measurements of the FUS LC film.
We employed FTIR technique, which is sensitive to vibrational modes of molecules and their structure, but not to their orientation. We performed FTIR experiments, both from the on-liquid film, and the film transferred onto a solid substrate. To perform these measurements, we first determined the fraction of protein organized in the film vs. left in solution bulk. We found that, following the film formation, only ~5% of protein is left in bulk, while 95% is contained in the formed protein film (by checking the absorbance and recalculating for the protein concentration). The high partitioning of the protein into the film enables a straightforward interpretation of these (in principle, surface-insensitive) FTIR measurements, due to the limited contribution from the molecules in bulk. Because FTIR is usually performed for bulk liquids (or pellets) or solid samples in the ATR mode, we employed a special unit for the FTIR measurement of a protein film at the air/water interface in reflection geometry (see photo in Figure S15). To correct for fluctuating CO 2 and H 2 O absorption in the background without applying nitrogen purging, which would otherwise perturb the liquid surface, polarization modulation (PM) using a photoelastic modulator (PEM) was essential to achieve the necessary signal-to-noise ratio in the infrared reflection absorption measurements (IRRAS). 11,12 This experimental design thus allowed for the in situ FTIR measurement of an on-liquid protein film.
As shown below (see Figure S16), the resulting PM-IRRAS amide I spectrum (dark blue line) is fully consistent with the SFG results (violet line), as shown below. In addition to PM-IRRAS data at the air/water interface, we also measured FTIR of the film on a solid substrate (blue line). For supported-film measurements, we used the same sample preparation as for the AFM measurements of the FUS LC film on a silicon wafer (shown in Figure 1 of the main text). However, due to the presence of water absorption bands, the amide band was obscured by the OH bending of water. Therefore, we deposited the film onto conventional CaF 2 windows, which allowed us to detect protein bands with standard transmission FTIR (see Figure S16). Importantly, these new FTIR data show that the vibrational amide I response of the fibrillar film on a solid substrate is very similar to those of the film on buffer solution, both of which are highly similar to our original SFG data. Specifically, we find that the PM-IRRAS spectra from the on-liquid film and that deposited onto the solid substrate both show amide I and amide II bands centered at similar frequencies, confirming that fibrillar species are present already on the liquid interface.
Center frequencies of amide I and amide II bands are ~1655 cm -1 and ~1540 cm -1 , respectively.

XIV. SFG signal in amide II spectral region.
In the study on human islet amyloid polypeptide (hIAPP), Tan et al. have revealed a correlation between the appearance of amide I (in fact, at ~1624 cm -1 ) and amide II SFG bands, reporting on oligomer/fibril peptide organization. 5 The center frequency of the amide I peak for FUS LC is more blue-shifted compared to that observed for hIAPP. Despite that, hypothesizing that the amide II signal can also potentially be used for FUS LC as a reporter of fibrillar structures, we also investigated the amide II SFG spectra for FUS LC. As shown in Figure S17, we observe a very weak amide II (almost insignificant compared to amide I) signal, but its signal is too low for reliable analysis. Note that, due to interference between the different signals, the very weak amide II peak first appears as a slightly negative feature around 1580 cm -1 , before developing into a positive peak around 1560 cm -1 . Figure S17 shows three different sets of experiments recording the SFG spectra for the FUS LC film formed at the air/PBS buffer interface, with a 1.5 μM FUS LC initial concentration in solution bulk upon injection. By performing time-dependent measurements, we find that the amide II response develops and remains in the spectrum over time, which excludes that the amide II response is caused by intermediate states. Resonant contributions (see Table S2) are presented as follows: (purple) CH2 symm, (cyan) CH3 symm, (green) CH3 FR, (light green) CH3 asymm, (red) and (orange) OH stretch contributions. For simplicity, the aromatic CH peak centered at ~3025 cm -1 is not presented, due to low contribution. (b) The corresponding surface pressure curve.
After FUS LC is added in the PBS buffer, a substantial increase in the OH-stretching signal intensity is observed ( Figure S19a). Likely, the observed increase is due to an increased ordering of water molecules at the air/PBS buffer interface due to ordered proteins present at the interface.

S23
The fitting of the complete CH-and OH-stretching region was performed. Each resonant contribution was fitted with a Lorentzian. Parameters obtained from the fitting are presented in Table S2. The signs selected for the Im( (2) ) of the contributions are as follows (according to HD-SFG, Figure 3b in the main text): CH 2 (symm)<0, CH 3 (symm)<0, CH 3 (FR)<0, CH 3 (antisymm)>0, CH(aromatic)>0, OH(3200 cm -1 )>0, OH(3400 cm -1 )>0. Table S2. Results of the fitting for SFG spectra presented in Figure S19a. Based on the HD-SFG data, we propose that the presence of FUS LC (which adsorbs, orders, and folds at the air/PBS buffer interface) aligns hydrogen-bonded interfacial water (with hydrogens pointing towards the air phase). We note that when we compare the homodyne SFG result ( Figure   S19a) of a separately performed experiment with the heterodyne SFG result (Figure 3b), it is evident that in homodyne SFG data the difference in the OH-stretching signal intensity (with vs.

PBS+1.5 μM FUS LC
without FUS LC) is bigger than that in heterodyne SFG data. This can, for the most part, be explained by the homodyne SFG signal being proportional to the square of the second-order S24 susceptibility. Figure 3b shows a two-fold increase in the heterodyne signal strength for water in the presence of FUS -this leads to a four-fold increase in SFG intensity, as observed in Figure   S19a. Secondly, the experiments were performed under slightly different experimental conditions.
For the homodyne SFG, during the entire experiment, the SP needle was inserted into the trough (stainless steel trough, D=4.5 cm), and the trough was rotated. For the heterodyne SFG, during the experiment, there was no SP needle in the trough (pure Teflon trough, D=2.9 cm), and the trough was not rotated during the measurement to maintain the phase stability.

XVII. Control experiments: HD-SFG of FUS LC in CAPS pH 11, and HD-SFG of 0.8 mM CAPS in PBS.
We measured HD-SFG spectra for the pristine CAPS buffer (20 mM pH 11) and that in the presence of FUS LC (1.5 μM concentration in bulk). The corresponding spectra for the imaginary part of the complex second-order nonlinear response (2) (which contains information on vibrational resonances and is free from the non-resonant background) are presented in Figure   S20. Figure S20. Imaginary part of the second-order nonlinear susceptibility for (red) the pristine CAPS buffer (20 mM pH 11) and that after addition of FUS LC into the subphase and measured ~20 mins (black) and ~2 hours (green) since the protein injection.

S25
The spectrum of the pristine CAPS pH 11 buffer shows a positive signal in the OH-stretch region (reporting, on average, H-up oriented hydrogen-bonded OH moieties), and a minor contribution from CH (consistent with a minor, slow, but steady increase in SP and amide I SFG signal, see

Figures 1 and 3a in the main text, respectively). The H-up orientation of water can be understood
by noting that, since the pH of the CAPS solution 0.6 pH units above the pK a (pH 11, pK a =10.4), ~ half of the CAPS molecules will be deprotonated and negatively charged. The interfacial water molecules will interact with the negatively charged sulfonate group resulting in the H-up orientation of interfacial water molecules. A substantial red shift of the OH stretch down to 3200 cm -1 can arise from the (3) (bulk) contribution in addition to interfacial (2) response. We note that while the orientation of the water molecules is similarly H-up for pure CAPS buffer, this spectrum in In this case, the interfacial water signal is identical to that in the pure PBS since there is very little CAPS in the solution to adsorb to the surface (see Figure S20). This is entirely consistent with Figure S5 showing almost no change in SP with 0.8 mM CAPS in a PBS subphase. We also recorded HD-SFG spectrum with the sample trough rotation switched on to exclude possible effect of the laser pushing molecules out of focus area. The spectra of pure PBS and 0.8 mM CAPS + PBS (with and without rotation) are identical. Thus we confirm that the signal presented in Figure 3b is indeed the effect of FUS LC (and not arising from CAPS surface propensity). air/PBS buffer interface. (2) response measured with the heterodyne SFG spectroscopy for FUS LC at the air/PBS buffer interface and presented in Figure 3b of the main text was modelled according to the following formula: 15 and are amplitude and phase of the non-resonant contribution; (2) and ( where corresponds to the SFG, Vis or IR beam, is the wavelength and is the incident angle of the -th beam. For our case the considered salt concentrations are as follows: 1 = 0.137 (only NaCl from the PBS buffer was taken into account here), 2 = 0.313 , 3 = 0.665 . Further, we construct the differences between the imaginary part (Im) of the (2) ( ) response measured at different salt concentrations, and : For simplicity, we denote the real and imaginary parts of (3) ( ) as follows and write the equations for the salt concentrations 1 , 2 , and 3 : The system above is Solving the above presented system of linear equations with relation to and , one obtains the following expressions: Inserting experimental data into these expressions, we get ( ) and ( ) , i.e the real and imaginary parts of (3) ( ). Further, we plot the imaginary part of the entire (3) contribution, i.e.
� (at highest salt concentration 3 as presented in Figure S22, red curve) versus frequency together with the experimental data, (Red) imaginary part of the (3) contribution calculated for 3 = 665 mM. Figure S22 clearly shows that the main contribution in the measured (2) consists mainly of the (2) term and not the (3) term. A much larger (2) contribution than the (3) contribution indicates that the SFG signal probed in the hydrogen-bonded OH-stretching region originates from the molecular ordering of OH moieties in the topmost layers at the air/PBS buffer interface.

XIX. SP data for PBS→CAPS→PBS buffer exchange experiment.
SP measurements were performed with a height sensor (Keyence LK-G82 Lasersensor) to correct for possible evaporation from the liquid surface (and for changes of the surface height during buffer exchange experiment). As stated in the main text, the FUS film was first formed at the air/PBS buffer solution interface (we further recall this part of the experiment as "PBS"). Once the film formation equilibrated, as was determined by a stabilized SP, the PBS buffer subphase was exchanged to CAPS in the following way: without perturbing the formed film, the CAPS buffer solution was added into and an equal volume of the subphase was simultaneously removed from the trough. This cycle was repeated 5X, which showed a subphase pH (checked with a pH meter) consistent with that of the exchange buffer (pH 11 for CAPS). Once the PBS→CAPS buffer was exchanged, the film was again allowed to equilibrate. This experiment is referred to as "PBS →CAPS". In experiments where the buffer was again echanged to PBS, "PBS→CAPS→PBS", we performed the buffer exchange from CAPS back to PBS as described and the pH was verified.
Note that during the buffer exchange, no additional protein was added into the subphase. See Figure S23 for the SP data obtained.

XX. Analysis of SP fluctuations.
In addition to the SP average values, we quantified the amplitude of SP fluctuations detected during each stage of buffer exchange experiments, "PBS", "PBS→CAPS", "PBS→CAPS→PBS" (see explanation of notations in Figure S23). Zooming-in to the raw surface pressure data presented in Figure S23 reveals a periodicity in signal, which is shown in Figures S24a, S25a, and  Figures S24c, S25c, and S26c.
Taking the FFT of ACF reveals the dominant frequency centered at 0.08 Hz which corresponds to a period of T~12.5 s. As these expeiments are performed with a rotation trough, this periodicity corresponds to the rotational motion of ~5 rpm of the trough. The reason for such periodic SP curve fluctuation is solely technical: upon the trough rotation, the surface of the liquid in the trough is not flat, but has some curvature due to the centrifugal force. At the same time, if the trough is positioned not at the exact center of the rotation stage (whose center determines the rotation axis), the SP needle inserted into the trough then makes a circle at the surface with the center displaced with repect to the center of the trough. Since the liquid surface has some curvature, the depth of the needle insertion is a periodic function of time with a period equal to the time needed for one rotation of the trough, which accounts for the presence of the periodic signal we observe in SP curve. Since periodicity originates due to the technical reason, we filter it out using an infinite impulse response (IIR) notch filter with the center frequency corresponding to 0.08 Hz. The clean, filtered SP signal is presented in Figures S24d, S25d, and S26d, respectively. For the analysis and estimation of the SP fluctuation amplitude, three non-overlapping regions at each stage of the experiment were chosen so that the number of points in each region was N=2000 which corresponded to a 500s time interval. For each region, the signal was filtered according to the procedure described above and the standard deviation was estimated. The mean of those standard deviation values and the error were calculated (taking Student coefficient for α=0.9), and the data were presented in Figure S27.    Mobile fraction data is extracted from N = 56 FRAP unique curves. Bars are mean, and the error bars are standard deviation. Asterisks show statistical significance, P < 0.05 using a Two-way ANOVA followed by a Student t-tests between different groups. I, II, III denote three experimental stages namely PBS, PBS→CAPS, PBS→CAPS→PBS, respectively. Individual data points are shown as dots.

XXII. FTIR experiments on FUS LC film upon buffer exchanges.
In complement to SFG, we present additional FTIR experiments which are insensitive to molecular orientation. For these measurements, we used the solid-supported films measurements to avoid the challenges of buffer exchange in the PM-IRRAS setup. Having established that film transfer to the CaF 2 windows did not fundamentally add or remove vibrations compared to the on-liquid film, this allowed us to make the buffer exchanges and examine the vibrational spectra of the film with FTIR thereafter (3 spectra, buffer exchange data, see Figure S29 below). Surprisingly, after the PBS→CAPS buffer exchange and film deposition, along with the peak centered at ~1670 cm -1 , we observe a distinct peak at ~1620 cm -1 (purple curve, Figure S29). To check whether this peak originates from the protein, we performed a control experiment: we "deposit" and measure FTIR of the subphase after the PBS→CAPS buffer exchange but without FUS LC present (light magenta curve, Figure S29). The results show that the response at ~1620 cm -1 is not from the protein, but from the CAPS molecule present in the buffer. In fact, this peak is also observed for the film deposited on a substrate after the PBS→CAPS→PBS buffer exchange (dark blue curve). The experiments thus show that the FTIR spectrum at ~1620 cm -1 is "contaminated" by this vibrational mode of the CAPS molecule, which complicates the interpretation of spectra and addressing the question if there is any protein contribution present at this frequency. The FTIR spectrum of CAPS molecule (see https://spectrabase.com/spectrum/AQFVQlNGKyn) confirms that this resonance originates from CAPS. However, at the same time, CAPS does not have any significant response at this frequency in Raman (see https://spectrabase.com/spectrum/9ruifMdyFQF), so that in the SFG measurements (requiring both IR-and Raman-activity), this CAPS mode does not appear.
Summarizing our findings, the FTIR data thus support the SFG data, in that neither of the techniques contains an indication of the distinct amide I response at ~1620 cm -1 , for the original film formed at the PBS subphase. For the film after the PBS→CAPS→PBS buffer exchange, a weak SFG signal is detected at ~1620 cm -1 . Using FTIR, we tried to clarify the origin of this contribution. We have, however, revealed that CAPS contributes to FTIR spectrum, and this fact complicates the interpretation of CAPS vs. possible protein contribution at that frequency. At the same time, based on our data along with the available Raman spectrum, we conclude that, while being FTIR-active, CAPS cannot contribute to the SFG response due to the lack of Raman active modes in the amide I region. The combination of these experiments allows us to conclude that the SFG response at 1620 cm -1 evident after the PBS→CAPS→PBS exchange indeed originates from FUS LC. The amide SFG spectra of the original PBS-film and the one formed after buffer exchange with CAPS and returning to PBS show that the film is structurally different at the molecular scale, with the 1620 cm -1 shoulder indicating a more classical amyloid behavior for the latter.