Bovine PI3K-SH3 domain was expressed in Escherichia coli BL21 (DE3) as an N-terminal glutathione S-transferase (GST)-fusion followed by a thrombin protease cleavage site using pGEX-4T (GenScript Biotech, Netherlands) as vector plasmid. The 86 amino acid sequence of the purified bovine PI3K-SH3 domain used in this study is GS MSAEGYQYRA LYDYKKEREE DIDLHLGDIL TVNKGSLVAL GFSDGQEAKP EEIGWLNGYN ETTGERGDFP GTYVEYIGRK KISP with the N-terminal Gly-Ser residues remaining as overhang from the thrombin cleavage.

BL21 cells were grown in M9 medium (for [U-¹³C, ¹⁵N] labeling with 3.2 g U-¹³C-glucose and 2 g ¹⁵N-ammonium chloride per l culture medium as the sole carbon and nitrogen sources) at 37°C for ∼4.5 h until an OD₆₀₀ of 0.6 was reached. Protein expression was induced by the addition of 0.5 mM Isopropyl β-d-1-thiogalactopyranoside (IPTG). The cells were then grown overnight at 28°C, harvested by centrifugation (5.000xg, 15 min) and resuspended in 5–8 mL per g cell mass of lysis buffer (50 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) and 100 mM NaCl, 0.3 mM phenylmethylsulfonyl fluoride (PMSF) pH 7.6 containing 5 mg/mL lysozyme). After sonication with a Bandelin Sonopuls sonicator (Bandelin) using a VS 70T probe (60% amplitude, 3 × 5 min, 3 s “on,” 5 s “off,” ice cooling) and ultracentrifugation in an Optima XPN-80 ultracentrifuge (Beckman) equipped with a 70Ti rotor at 4°C and 42.000 rpm for 1 h, the supernatant was loaded on two serially connected 5 mL Protino GST/4B columns (Macherey-Nagel, Düren, Germany) equilibrated with 50 mM HEPES-NaOH buffer, 100 mM NaCl, pH 7.6, and 3 mM Dithiothreitol (DTT). The protein was eluted with 20 mM reduced glutathione in 50 mM HEPES, 100 mM NaCl, 3 mM DTT, and pH 7.6. The fractions containing the eluted GST-PI3K-SH3 fusion protein were cleaved using 5–7.5 NIH units of thrombin (SERVA Electrophoresis GmbH, Heidelberg, Germany) per mg protein for 2 days at 4°C under mild shaking conditions. Then, the solution was loaded on a Superdex 75 HiLoad 16/600 size exclusion column (GE Healthcare Europe GmbH, Freiburg, Germany), equilibrated, and run with 5 mM ammonium acetate (pH 7.7) as a volatile buffer system, leading to the separation of the cleaved GST from PI3K-SH3. To ensure complete removal of GST traces, fractions containing PI3K-SH3 were pooled and loaded again on two serially connected 5 mL Protino GST/4B columns (Macherey-Nagel, Düren, Germany) equilibrated with 5 mM ammonium acetate. The flow-through fractions containing purified bovine [U-¹³C, ¹⁵N] PI3K-SH3 were pooled, aliquoted, and freeze-dried for further usage. Analysis by SDS-PAGE and NMR suggest bovine PI3K-SH3 purities above 97%, as neither method detected impurities.

The production of bovine PI3K-SH3 fibril seeds was essentially carried out according to Röder et al. (2019). In brief, 100 nmol lyophilized non-isotopically labeled PI3K-SH3 was resuspended in 500 µL 10 mM glycine-hydrochloride pH 2.5 buffer, leading to a final concentration of 200 µM PI3K-SH3. The glycine-hydrochloride buffer comprised 10 mM glycine and 5.66 mM hydrochloric acid, resulting in a final pH 2.5. The solution was shaken in an Eppendorf tube at 1400 rpm at 42°C for 24 h. A ∼5 μL sample was withdrawn for Atomic Force Microscopy (AFM) measurements (see below). The remaining 495 μL were transferred into a 15 mL Falcon tube and then diluted to approximately 100 μM with another 500 μL 10 mM glycine-hydrochloride pH 2.5 buffer. The solution was then sonicated under ice-cooling for 15 s (1 s “on,” 2 s “off,” 10% amplitude) with a Bandelin Sonopuls sonifier using an M72 probe to disrupt the fibrils into smaller species called seeds.

100 nmol lyophilized bovine PI3K-SH3 was resuspended in 390 μL 10 mM glycine-hydrochloride buffer (pH 2.5, 8% D₂O). The sample was divided into three parts (one for solution NMR, one for HR-MAS, and one for MAS solid-state NMR experiments). A total volume of 10 μL unlabeled PI3K-SH3 fibril seeds (100 μM stock) was added proportionally to each PI3K-SH3 sample, corresponding to a 1% (mol/mol monomer equivalents) seed mass. The final PI3K-SH3 protein concentration was 250 μM. The “dead time” (the time between the addition of seeds but before NMR spectra were recorded) was measured after adding the PI3K-SH3 fibril seeds.

All NMR experiments described in the following were conducted at 298 K (25°C). The temperature at every spectrometer was calibrated using either Methanol (solution NMR probes, HR-MAS probe) or DSS (solid-state NMR probe). Details on all NMR experiments are listed in Supplementary Tables S1, S2.

After the NMR experimental series, all PI3K-SH3 samples were investigated using Atomic Force Microscopy (AFM). The aggregated solutions of HR-MAS and solid-state NMR rotors and the NMR tube were transferred to 2 mL Eppendorf tubes. For AFM samples of the fibril seeds, the solution was diluted in 10 mM glycine-hydrochloride, pH 2.5 buffer to a protein concentration of 0.5 μM; for HR-MAS and solid-state NMR AFM samples to 50 μM, and the liquid NMR sample to 5 μM. A 5–10 μL volume of the solutions was pipetted on a mica surface. After 10 min of incubation, the mica was washed extensively with Milli-Q water and dried under a nitrogen gas flush. The AFM micrographs were recorded in ScanAsyst mode using the peak-force tapping mechanism on a Bruker Multimode 8 (Billerica, Massachusetts, United States) using OMCL-AC160TS cantilevers (Shinjuku, Tokyo, Japan) at 1024 × 1024 pixel resolution. The images were processed with Gwyddion 2.61 (Nečas and Klapetek, 2012).

Far-UV-CD spectroscopy of bovine PI3K-SH3 samples was performed on a JASCO J-1500 CD spectropolarimeter (Jasco, Gross-Umstadt, Germany) using 1 mm path length quartz cuvettes (Hellma, Müllheim, Germany). Spectra were recorded at 20°C after dissolving lyophilized PI3K-SH3 to a protein concentration of 20 µM in 10 mM glycine-hydrochloride buffer, pH 2.5, or in 20 mM Na phosphate buffer, pH 6.8, with instrument settings as follows: step size 1 nm, scan speed 100 nm min⁻¹, bandwidth 2 nm, DIT 4. Far-UV-CD spectroscopy data of bovine PI3K-SH3 fibrils were recorded at a protein concentration of 20 µM (monomer equivalents) in 10 mM glycine-hydrochloride buffer, pH 2.5, as described above. The signal-to-noise ratio was improved by accumulation of 6 scans per sample. The mean residue ellipticity, MRE or θ m r w in deg·cm²·dmol⁻¹ was calculated from the equation θ mrw = θ obs × MRW / c × d × 10 , with θ obs , observed ellipticity (in millidegrees); c , concentration (in g/l); d , cell path length (in cm); M R W (mean residue weight), molecular weight divided by number of peptide bonds.

For the fibril elongation assay, which was monitored through the amyloid-specific dye Thioflavin-T (ThT), fibril growth experiments were conducted in triplicates (using three separate batches). Each batch/sample contained 250 μM PI3K-SH3, 0.04% NaN₃, and 1% seed material in 10 mM glycine-hydrochloride buffer, pH 2.5, using the same experimental conditions as the NMR measurements. The sample was incubated under quiescent conditions at 25°C for 80 h. Due to the limitations of ThT fluorescence measurements at low pH (Becker et al., 2023), every 8 h a 10 µL aliquot of the reaction mixture was withdrawn and diluted 1:10 in 20 mM sodium acetate buffer (pH 5.0), 50 mM NaCl, containing 20 μM ThT. Subsequently, the diluted samples were transferred to a 96-well half-area, non-binding surface, polystyrene plate (Corning, No. 3881, black with clear bottom), and the plate was sealed with clear sealing tape (232701, Thermo Scientific). The plate was then loaded into a FLUOstar Omega plate reader (BMG Labtech, Ortenberg, Germany). Fluorescence measurements were performed with an excitation wavelength of 448 nm and emission measurements at 482 nm at 25°C. Measurements were repeated every 20 s in cycles 1 to 500 to measure the sample for a total of 2.7 h for each data point of the original aggregation assay. For analysis, the plateau values of the ThT fluorescence curve between 2.2 and 2.7 h were averaged and plotted against the time.

Bovine PI3K-SH3 loses its stable globular fold below pH 4 and aggregates into fibrils, as found previously by Dobson and coworkers (Zurdo et al., 2001). Here, we studied the PI3K-SH3 aggregation at low pH and under seeding conditions using three different NMR techniques (solution, HR-MAS, solid-state) and, for comparison, using a Thioflavin T assay. As a starting point, we dissolved lyophilized bovine PI3K-SH3 at pH 2.5 in 10 mM glycine-hydrochloride pH 2.5 buffer. We divided the sample into three aliquots of 250 µM final concentration (PI3K-SH3 monomer) for the three NMR experiments (solution, HR-MAS, solid-state) at a sample temperature of 25°C. To initiate immediate fibrillation, fibril seeds were added to each sample at a concentration of 1% molarity relative to the PI3K-SH3 monomer concentration in the solution. Initially, before fibrillation started, all PI3K-SH3 monomers were in solution and fully visible in the solution-state NMR spectra. To monitor the presence of the monomeric PI3K-SH3 species, we recorded a time series of consecutive two-dimensional (2D) solution NMR ¹H-¹⁵N HSQC spectra of approximately one-hour duration each. Figure 1A shows 2D solution NMR ¹H-¹⁵N HSQC spectra of unfolded monomeric bovine PI3K-SH3 at three exemplary time points after adding fibril seeds (1 h, 51 h, and 101 h). The narrow dispersion in the ¹H dimension of the most intense peaks indicates that the major population of monomeric PI3K-SH3 is disordered under these conditions. However, a minor population of PI3K-SH3 with β-sheet conformation, similar to the globular conformation, is still present, visible as the less intense peaks with ¹H chemical shifts ranging between 6 and 10 ppm. A direct comparison of ¹H-¹⁵N HSQC spectra of PI3K-SH3 at pH 2.5 and PI3K-SH3 at pH 6.8 (globular conformation), see Supplementary Figure S1, underlines the presence of a minor folded species, similar to the globular protein at pH 6.8 and a major unfolded population. From the first increment of the solution NMR 2D spectra, shown in Figure 1C, we estimated the population of the minor folded species to approx. 15%–20%, based on intensities associated with the folded β-sheet region (9–10 ppm) compared to the central region between 7.5 and 8.5 ppm, which is dominated by intensities of the disordered conformation. Supplementary Figure S1 shows a direct comparison (overlay) of the 2D solution NMR spectra of SH3 at low and neutral pH. Supplementary Figure S2 shows the circular dichroism (CD) spectrum of PIK3-SH3 at pH 2.5, underlining that at pH 2.5 and 25°C, SH3 is mainly disordered. A CD spectrum of SH3 fibrils at the same pH of globular folded SH3 at neutral pH is compared. CD spectra agree with data shown in (Zurdo et al., 2001).

During the time course of aggregation, a series of solution NMR ¹H-¹⁵N HSQC spectra were recorded. Spectra recorded at later time points show lower signal intensities of monomeric PI3K-SH3. Spectral intensities declined exponentially. The decay rate for the disordered and the minor folded conformation appears similar, as the inspection and comparison of the one-dimensional (1D) spectra of the first increments of the HSQC spectra indicates, see Figure 1C. Also, the intensities of representative peaks in the 2D solution NMR spectra decay with similar rate constants (Supplementary Figure S3). This suggests a dynamic equilibrium between the major unfolded population and the minor folded population, which must be slow on the NMR timescale (slower than about 10 ms, as two separate conformations are visible) but fast on the aggregation time scale (as all resonances decay with similar rate constants, in the order of hours by inspection of spectral intensities, see Figure 1 and Supplementary Figure S3). Intensities are decreasing because of increasing monomer aggregates during the experiment. The aggregated PI3K-SH3 species is not visible to solution NMR because of the high molecular weight of the aggregated species and the associated slow overall tumbling correlation time.

The emergence of the aggregated species can be followed by solid-state NMR, which does not face any overall weight limitation, as overall tumbling is absent in the solid state. Therefore, in parallel to the solution NMR time series, we recorded one-dimensional ¹³C-detected solid-state NMR spectra on the second aliquot of the P13K-SH3 sample (250 µM with 1% seeds added). Two types of magnetization transfer were used for the solid-state NMR spectra: In the first spectrum, a dipolar cross-polarization (CP) based transfer was employed, which is sensitive to the more rigid part of the sample, namely the aggregated species. Second, a scalar coupling-based INEPT (insensitive nuclei enhanced polarization transfer) was used, which is sensitive to the more flexible species, which is the unbound monomer in solution. Therefore, the INEPT-based 1D spectra connect to the solution-state NMR spectra on the monomeric species. Figures 2A, B show representative spectra of both time series, which were recorded inter-leaved. The CP-spectra (Figure 2B) show increasing intensities indicating the emergence of the aggregated species. This corresponds to a decay of the monomeric species, as visible in the INEPT-based spectra (Figure 2A).

To connect the observations by solution-state NMR and solid-state NMR, we also recorded a time series of interleaved ¹H-¹⁵N HSQC spectra and ¹³C-detected INEPT spectra on the third aliquot of the sample, using an HR-MAS probe. This probe can record solution NMR-like spectra using proton-detection under MAS conditions, as in solid-state NMR. The influence of the magic angle spinning on the aggregation kinetics and the resulting product has not been investigated in detail yet. Figure 1B shows the time series of ¹H-¹⁵N HSQC spectra recorded on bovine PI3K-SH3 under MAS conditions, with MAS frequencies of 6 kHz, recorded using an HR-MAS probe. Comparison of Figures 1A, B reveals that the initial PI3K-SH3 monomer spectra appear similar under both conditions, the quiescent condition during solution NMR experiments and under agitated conditions, namely centrifugal forces experienced during MAS conditions. However, the intensity decay rate appears slightly faster for the HR-MAS experiments than for the solution NMR experiments under quiescent conditions. This is also visible in the 1D spectra corresponding to the first increments of the respective HSQC spectra (Figure 1C) and will be discussed later. ¹H-¹⁵N HSQC spectra, or their first increments, respectively, report on the protein backbone. A similar trend is observed for the HR-MAS ¹³C-detected INEPT spectra (Figure 2C), which are sensitive to protein sidechains. Consistent with the solid-state NMR spectra (Figure 2A), we observe that the concentration of the monomeric species decreases exponentially (see below) with time (Figure 2C).

To analyze the aggregation kinetics extracted from the various NMR spectra more quantitatively, we integrated and normalized spectral intensities of the 1D spectra (or first increments of the 2D spectra), as described in the Materials and Methods section. The normalized integrated intensities as a function of time (after the start of the aggregation experiment) are shown in Figure 3A. Interestingly, in the case of the solid-state NMR CP- vs. INEPT-based 1D spectra, the sum of the two integrated intensities gives an almost constant value (Figure 3B). (We interpret the observed slight decrease of the sum of integrated intensities over several days as a slight detuning of the probe over time.) Thus, the aggregation appears as a two-state process, with a relatively direct transition from the monomeric to the fibrillar state, without a high population of an invisible oligomeric on-pathway intermediate. (Loss to an invisible off-pathway (“dead end”) oligomeric species that does not give rise to any CP-detected solid-state NMR signal, however, cannot entirely be excluded.) Mono-exponential decay curves (or the asymptotic growth curves, respectively) were fitted to the integrated intensities using Eq. 1 in the Materials and Methods section. The obtained half-life times, τ, are shown in Table 1. Supplementary Figure S4 gives details on the fits. Overall, the lifetimes measured using the different NMR techniques (solution-state, HR-MAS, solid-state) show good qualitative agreement. Under quiescent solution NMR conditions, the monomeric species seems to have a slightly longer lifetime than under MAS solid-state NMR conditions (37.4 ± 0.6 h vs. 30.2 ± 1.1 h), representing agitated conditions.

For comparison, a ThT assay was recorded (Figure 3C), monitoring the increase of the fibrillar state. The observed increase in fibrillar structure is overall in good qualitative agreement with the solid-state NMR CP-based experiment, sensitive to the emergence of aggregated species (squared brown data points shown in Figure 3A). A comparison of the timeline of the CP-based solid-state NMR experiment and the ThT assay suggests that most CP signals originate from fibrillar SH3.

Integrated intensities of the first increment HR-MAS ¹H-¹⁵N HSQC spectra, sensitive to the amide proton backbone signals of the monomer, decrease simultaneously to ¹H-¹³C INEPT spectra, sensitive to the side chain signals, with almost identical lifetimes (25.0 ± 1.2 h vs. 24.0 ± 0.6 h). This suggests that both backbone or side-chain signals report unanimously on the decrease of the monomeric species. Under HR-MAS conditions, the decay of the monomeric species appeared slightly faster than under solid-state NMR conditions (τ = 24.0 ± 0.6 h vs. 30.2 ± 1.1 h). Thus, under both MAS conditions, the lifetime of the monomeric species appears slightly shorter than under quiescent solution NMR conditions (37.4 ± 0.6 h). We rationalize the slightly faster decrease of the monomeric species and, consequently, somewhat more rapid aggregation kinetics observed under agitated conditions in the MAS rotor by the influence of centrifugal forces on the aggregation process. Centrifugal forces will increase the local mass density close to the rotor wall. They may, therefore, slightly increase the likelihood of primary nucleation events, thus slightly speeding up the aggregation process.

To assess the reproducibility of the experimental setup, we repeated all experiments under identical conditions (Supplementary Figures S5, S6). While the results are in qualitative agreement, we observe differences when fitting the lifetimes of the monomeric species (from the INEPT spectra) and, in particular, for the build-up time of aggregated species (from the CP-based spectra), as described in Supplementary Table S3. Here, the observed differences in the lifetime of the monomeric species appear larger than the variation between the different NMR techniques (solution, solid-state, HR-MAS). Notably, in the second set of experiments, build-up times of aggregation as well as lifetimes of monomers in all experiments appear to be shortened to 50%–60% of the values in the original experiment, suggesting that this deviation is caused by an (unintended) variation of the starting conditions (e.g., slightly local differences in seed concentrations within the rotor or the NMR tube, respectively). This variation seems to have a larger influence than the NMR technique, making it difficult to judge or rationalize differences between the different NMR techniques fully. The way of seeding, or the seed itself, seems to have a major influence on the obtained aggregation kinetics. Also, we observe differences in the “appearance” time of the aggregated species extracted from the CP-based spectra. There will be a size distribution of aggregates during the aggregation process (see below). More stable and rigid aggregates like fibrils will contribute more strongly to the CP signal than smaller, more flexible species. Therefore, a slight variation of kinetics in the emergence of fibrils of different sizes will impact the growth of the CP signal over time.

To check the morphology of the bovine PI3K-SH3 fibril seeds before and after the NMR time series, we recorded atomic force microscopy (AFM) images: First of the fibril seeds (Figure 4A) and second of all NMR samples, once the recording of the fibrillation time series was finished. Samples were taken from all sample containers of the three different NMR methods employed: solution NMR (Figure 4B), HR-MAS (Figure 4C), and solid-state NMR (Figure 4D). The consistency of all samples was solution-like and not gel-like. The AFM images show that elongated amyloid fibrils with several µm length were formed during all measurements, substantially longer than the sub-µm short fibrils used for seeding (Figure 4A). For the sample obtained from the quiescent solution NMR conditions (Figure 4B), a wealth of fibrils is visible in the AFM images. The observed fibril density is less for the HR-MAS (Figure 4C) and the solid-state NMR sample (Figure 4D), both recorded under MAS conditions. At the same time, the fibrils in the solution NMR sample show a homogenous distribution; fibrils in the samples obtained after the MAS experiments are distributed more inhomogeneous. However, the sample was obtained from the center of the rotor. At the same time, the highest local density of fibrils will be found close to the rotor wall due to the centrifugal forces present during the MAS sample rotation. In addition to separated PI3K-SH3 fibrils, a few large aggregates with unresolvable structures can be observed in the AFM images of the experiments involving a MAS rotor and sample rotation (solid-state NMR and HR-MAS). The faster aggregation kinetics observed in the second NMR time series results in shorter but also more fibrils (Supplementary Figure S7).

The observance of substantially elongated fibrils suggests that fibril elongation is the dominant mechanism for fibril formation. This is also in line with the observed exponential decay of the monomer concentration, considering that the monomer concentration decreases upon fibril formation and the monomer supply is not unlimited. Following Occam’s razor, we fitted the observed decay on the monomer concentration using a mono-exponential decay function (Supplementary Figure S4). If the decay rate of the monomer concentration, d d t c m o n o m e r , solely depends on the available monomer concentration (and the decay rate constant α), a simple differential equation can describe this d d t c m o n o m e r = − α   c m o n o m e r . The solution of that differential equation is a mono-exponential decay function c m o n o m e r = c 0 * exp ⁡ ⁡ − α * t , with the initial monomer concentration c(0) and the decay constant α = 1 τ where τ is the fitted half-life time. This is consistent with the fits described in Supplementary Figure S4, where the fitted intensity corresponds to the monomer concentration (see also Materials and Methods).