A structural model of toxic amyloid oligomers involved in type 2 diabetes

Significance

Protein oligomers formed from hIAPP are a likely cytotoxic species that drive β-cell loss in type 2 diabetes. We report an atomic structural model of an hIAPP oligomer using a combination of multidimensional IR and NMR spectroscopy. The structure has an N-terminal helical bundle, a kink, and an intermolecular β-sheet spanning FGAILS. This structure helps explain why some species develop type 2 diabetes and not others, as well as providing a structural rationale for a familial mutation that causes early-onset diabetes in humans.

Abstract

Amyloid oligomers of the human islet amyloid polypeptide (hIAPP) are a likely cytotoxic species driving β-cell death in type 2 diabetes, but their transient nature has precluded atomic-level structural characterization. We obtained a high-resolution structure of a physiologically relevant hIAPP oligomer. Using 2D IR spectroscopy, we identified three substitutions that slowed aggregation sufficiently for comprehensive 2D/3D NMR analysis while retaining the key wild-type structural features and cytotoxicity. The structural model reveals a dimeric assembly with N-terminal helices and a kink that facilitates an intermolecular β-sheet. The β-sheet spans the famous FGAILS portion of the sequence, helping to explain species-specific diabetes susceptibility and the origin of early-onset familial mutations. The integrated 2D IR/NMR strategy provides a unique approach to obtaining high-resolution structures of amyloid oligomers.

Type 2 diabetes (T2D) is one of the most prevalent diseases (1) in the United States with ~40% of the population in a diabetic or prediabetic state (2, 3). The disease starts as insulin resistance but in later stages leads to loss of pancreatic β-cell mass and/or loss of insulin secretory capacity (4, 5). A key driver of β-cell death is the aggregation of human islet amyloid polypeptide (hIAPP or amylin), a 37 amino acid hormone (Fig. 1) that regulates appetite and gastric emptying via receptors in the brain (6, 7) but also forms amyloid plaques in pancreatic islets (8). Interestingly, the mature fibrils have little to no cytotoxicity; hIAPP is most toxic in the lag phase, when it exists as soluble oligomers (9–14). hIAPP oligomers are thought to interfere with receptor-mediated processes, cause inflammation, and permeabilize the cell membrane (7, 15). Newly reported monoclonal antibodies that selectively bind hIAPP oligomers and prefibrillar aggregates protect β-cells and improve glucose control in transgenic mice, suggesting that hIAPP oligomers may be a target for therapeutic development (16–18). Fibrils are toxic in some amyloid diseases (19, 20). For hIAPP, oligomers are a likely contributor.

Fig. 1.

Schematic aggregation mechanism and IAPP sequences. (Top) Oligomeric hIAPP is an intermediate to amyloid fibrils. By destabilizing fibrils, aggregation is slowed, leaving intact the oligomer population. (Bottom) Sequence of hIAPP, 3A-hIAPP, and rat IAPP. Sequence 20 to 29 (red box) that correlates with in vitro aggregation, plaque formation, and diabetes. The FGAILS residues (bold) form a parallel β-sheet in the oligomer structure of hIAPP, as marked. Sequence variations to hIAPP are colored.

The monomeric and fibrillar structures of hIAPP are known with atomic resolution (21–26), but little structural information exists about the oligomers because their transient nature, low concentrations, and heterogeneous size-distributions have prevented their atomic characterization with standard structural biology techniques. Using 2D IR spectroscopy and isotope labeling, we monitored hIAPP aggregation and identified a transient oligomeric species with a parallel β-sheet across residues 23 to 27 (FGAIL, Fig. 1) (27–29). This oligomer is an intermediate in the pathway toward the fibrillar state and is kinetically stabilized by the structural rearrangement required to form plaques. Interestingly, the β-sheet falls within a portion of the hIAPP sequence that is well known to correlate with aggregation and T2D in other mammals (Fig. 1, red box). Indeed, the 20 to 29 IAPP sequence dictates whether a species is prone to developing T2D or not (30–32). For example, rodent IAPP, which has three prolines in the 20 to 29 region (Fig. 1) that prevent β-sheet formation, does not aggregate in vitro and is not cytotoxic (7, 33).

Given the critical nature of the 20 to 29 sequence, we hypothesized that mutations outside this region could slow or prevent aggregation without impacting the key structural features of the oligomer and its cytotoxicity (Fig. 1), thereby permitting the application of high-resolution structural methods. In this article, we present results for an hIAPP oligomer with alanine substitutions at Leu12, Phe15, and Asn31 (Fig. 1). These three mutations slowed aggregation from hours to days, maintained hIAPP toxicity, and reproduced the 2D IR spectral signatures of the parallel β-sheet in the 20 to 29 region. The extended lag time of 3A-hIAPP enabled solution 2D/3D NMR spectroscopy that generated >200 structural constraints, containing both intra- and interprotein NOEs. Using these constraints, a dimeric atomic structural model was derived. It has N-terminal α-helices followed by a kink that aligns the C-terminal residues to permit a small intermolecular β-sheet across the physiologically critical FGAILS residues. We discuss the structure with regard to prior biophysical studies, its significance for mammalian sequence homology, and implications for a familial mutation that causes early-onset T2D.

Three Substitutions Extend the Lag Time

To assess the impact of alanine mutations at Leu12, Phe15, and Asn31 on hIAPP fibrillization, morphology, and cytotoxicity, we performed a variety of assays, which are shown in Fig. 2. 2D/3D NMR data collection typically requires >12 h of signal averaging at 500 μM to generate a dataset with sufficient signal-to-noise for chemical shift assignments and NOE cross peaks. Thus, the first diagnostic is the timescale of fibril formation, which we monitor using Thioflavin-T (ThT) fluorescence (Fig. 2A). At pH 5.3 and 7.4, hIAPP aggregates within t₅₀ = 3 h (defined as half-rise) while 3A-hIAPP takes 23 to 26 h at 500 µM and can be twice as long under gentle conditions (SI Appendix, Fig. S6).

Fig. 2.

Biophysical and biochemical assays comparing hIAPP to 3A-hIAPP. (A) ThT fluorescence experiments of 500 µM hIAPP (blue) and 3A-hIAPP (yellow) at pH 7.4 and 5.3 showing the extended lag phase for 3A-hIAPP. (B) TEM images at 0 and 24 h, 500 µM hIAPP and 3A-hIAPP for pH 7.4 and 5.3 under gentle conditions. (C) Cross-linking gel at pH 7.4 and 5.3 at 500 µM. (D) Concentration-dependent cytotoxicity in INS-1 cells. (E) Cell viability and TEM at 5 h and pH 7.4 and 5.3 for 150 µM hIAPP, 3A-hIAPP and control media when hIAPP is fibrillar but 3A-hIAPP is still oligomeric. Additional data, controls, and replicates are given in SI Appendix.

Transmission electron microscopy (TEM) images of hIAPP and 3A-hIAPP were compared at 5 min and 24 h (Fig. 2B, gentle conditions). At 5 min, no fibrils are observed for either polypeptide. At 24 h, 3A-hIAPP still lacks fibrils, whereas hIAPP has formed long amyloid fibrils. Thus, the ThT and TEM data indicate that 3A-hIAPP has a sufficiently long lag time at 500 µM for 2D/3D NMR analysis. Our conditions were chosen because in islets, hIAPP is stored at a pH of 5.0 to 6.0 and at concentrations of 1 to 4 mM (34, 35).

We used cross-linking (PICUP) to compare the size of the oligomeric assemblies in the lag phase of hIAPP vs. 3A-hIAPP (36). The cross-linked peptides were analyzed by SDS-PAGE and visualized with silver staining. Shown in Fig. 2C are the results at pH 5.3 and 7.4 (see SI Appendix, Fig. S5 for the full gel with additional controls). Protein bands corresponding to monomers through hexamers are observed. hIAPP and 3A-hIAPP have very similar distributions indicating that the 3A mutations do not perturb the degree of oligomerization.

To test cytotoxicity, two sets of cell viability assays were performed using INS-1 832/13 rat insulinoma cells, an insulin-producing cell line (37, 38). In the first assay (Fig. 2D), the cells are dosed with hIAPP or 3A-hIAPP 5 min after aggregation was initiated. At this time-point, the toxicity curves are very similar with an LC50 (the concentration at which 50% of the cells are alive) of 50 ± 16 and 63 ± 12 µM for hIAPP and 3A-hIAPP, respectively (SI Appendix, Table 2). Thus, 3A-hIAPP has nearly the same extracellular toxicity as hIAPP during the lag phase prior to aggregation. The second assay tested for toxicity at 5 h, when ThT (Fig. 2A) and TEM (Fig. 2E) indicate that 3A-hIAPP is still prefibrillar, whereas hIAPP is aggregated. At 5 h, 150 μM 3A-hIAPP is very toxic while hIAPP is less so, and similar results are obtained at both pH values (Fig. 2E). Thus, we conclude that 3A-hIAPP retains its toxicity during the lag phase, owing to long-lived oligomers, while hIAPP loses its toxicity upon aggregation to amyloid fibrils.

3A-hIAPP Oligomers Retain Native Structure

To ascertain if the 3A mutations impact the structure of the oligomer in the critical 20 to 29 region, we used rapid scan 2D IR spectroscopy paired with isotope labeling (Fig. 3) (27, 28, 39). The backbone carbonyls of residues 23-26 (FGAI) are labeled with ¹³C¹⁸O, which shifts their vibrational frequencies without altering the protein structure, spectroscopically resolving them from the remaining amino acids. Aggregation was initiated in the same manner as above and 2D IR spectra were continuously recorded. Representative 2D IR spectra of the isotope-labeled region for lag phase and fully aggregated state are shown for hIAPP (Fig. 3 A and B) and 3A-hIAPP (Fig. 3 D and E) measured under identical conditions.

Fig. 3.

2D IR spectra of ¹³C¹⁸O FGAI-labeled hIAPP and 3A-hIAPP at 500 µM. (A) hIAPP aggregation in the lag phase showing monomers and oligomers. (B) hIAPP aggregation in the saturation phase showing presence of only fibrils and (C) Contour plot made from normalized diagonal slices of the isotope region (A and B) to emphasize frequency shifts (SI Appendix, Table S1). Superimposed is the aggregation curve (white) derived from the intensity of the unlabeled β-sheet feature at 1,620 cm⁻¹ that falls outside the isotope region. (D–F) Identical experiments for 3A-hIAPP. During the lag phase, only oligomers are present and at t₅₀, spectroscopic signatures of fibrils begin to appear. (A, B, D, and E) Red is positive and blue negative intensity. (C and F) White is maximum positive intensity. This figure is reproduced in SI Appendix, Fig. S20 with color bars.

To illustrate the kinetics, we plot diagonal slices through the 2D IR spectrum as a function of time. Each slice is normalized to its maximum to highlight the changes in frequency (Fig. 3 C and F and SI Appendix, Figs. S1, S4, and S20). Overlayed on the slices is the intensity of the unlabeled β-sheet feature at 1620 cm⁻¹, which measures the formation of amyloid fibrils in a similar manner to ThT. For hIAPP, the 0 h spectrum exhibits a mixture of monomers and oligomers that span 1565 to 1585 cm⁻¹, which we assign from similar experiments performed in the past (Fig. 3A) (28). At the end of the lag time (t₅₀ = 3 h), they convert to fibrils which absorb at 1,550 ± 4 cm⁻¹. For 3A-hIAPP, only the oligomer peak at 1570 cm⁻¹ is observed during the lag time (Fig. 3C), indicating that the alanine mutations stabilize the oligomers. The fibrils grow in over many hours (t₅₀ = 30 h) (Fig. 3F). Interestingly, oligomers still exist in the presence of fibrils, albeit with decreasing concentrations over time (see SI Appendix, Fig. S2 for unnormalized diagonal slices), which is most likely a sign that the alanine mutations destabilize the fibrils. Thus, the 3A mutations both stabilize the oligomers and destabilize the fibrils as intended. Replicates of 2D IR data are shown in SI Appendix, Fig. S3.

In summary, 3A-hIAPP recapitulates the toxicity, TEM, and cross-linking assays of hIAPP, despite delaying aggregation for many hours at concentrations of hundreds of micromolar. Moreover, the mutations do not impact the structure of the oligomers across the critical residues 23 to 26 (FGAI). With these controls in place, we turn to detailed structural analysis using multidimensional NMR.

NMR Constraints and a Structural Model

Next, we synthesized 3A-hIAPP with ¹³C¹⁵N labels (see Fig. 4A for labeling pattern) and collected NMR spectra between pH 4.0 and 7.4, at either ambient temperature (Fig. 4 B–D) or 5 °C (See SI Appendix, Table S3 for complete list of experiments). At 5 °C, data could be collected for 3 to 5 d. From the pH-dependent spectra, we conclude that 3A-hIAPP is monomeric at pH 4.0 and oligomeric at higher pH and that the structure of the oligomers is different from that of the monomers, based on progressively broader linewidths and pH-dependent chemical shifts (Fig. 4 B–D and SI Appendix, Fig. S14). Consequently, we used a series of 2D and 3D NMR spectra at pH 4.0 to assign peaks (SI Appendix, Fig. S8 and S9), including all the unlabeled amino acids. We then transferred the assignments to the spectra of oligomeric structures at higher pH using the pH titration data and 3D-HSQC-TOCSY experiments to confirm side-chain identity (SI Appendix, Figs. S10 and S11).

Fig. 4.

NMR methodology used for the characterization of 3A-hIAPP. (A) Position of ¹³C¹⁵N-labeled amino acids in the hIAPP sequence. (B) HSQC spectra of 3A-hIAPP at pH 4.0. (C) HSQC spectra of 3A-hIAPP at pH 5.3. (D) HSQC spectra of 3A-hIAPP at pH 7.4. (E) Plot showing the temperature coefficients of ¹H chemical shifts as a function of residue number, measured within the specified temperature range. The dashed line marks the typical cutoff ppb value for hydrogen-bonded residues. Box representing the residues involved in hydrogen bonding. (F) To selectively probe intermolecular NOEs, mixed samples containing 50% labeled and 50% unlabeled 3A-hIAPP were used. The strip plots display residues exhibiting intermolecular NOEs between the labeled and unlabeled peptides, indicated by arrows. Red: 3D-HSQC-NOESY spectrum of 100% labeled peptide. Green: Double half-filtered 3D-HSQC-NOESY spectrum of a 1:1 mixture of labeled and unlabeled peptides.

We identified backbone hydrogen bonding of the oligomer via temperature dependence of proton and carbon chemical shifts of the amides (Fig. 4E and SI Appendix, Fig. S12) at pH 5.3. The amide protons of residues 4 and 12 to 17 exhibit coefficients above –4 ppb/K at temperatures below 303 K and a markedly reduced temperature dependence above 303 K, consistent with intramolecular hydrogen bonds. The amide carbonyl (CO) chemical shifts (SI Appendix, Fig. S13) of residues 5 and 12 to 15 have similar temperature dependence and consistently lie above the other residues, also consistent with continuous hydrogen bonding. Thus, most of the labeled amide carbonyl groups at the N-terminus form intramolecular hydrogen bonds, such as via formation of an N-terminus helix, which we further confirm below.

To obtain constraints for structural modeling, we performed ¹⁵N-edited 3D NOESY experiments. Representative spectra from five different planes in the region of FGAI are shown in Fig. 4F, highlighting distinct NOE cross-peaks between the amide proton (HN) of residue i and the alpha proton (Hα) of residue i–1. Additional datasets are provided in SI Appendix, Fig. S15. In total, more than 200 NOE cross-peaks were identified including 18 long-range NOEs (Fig. 5A and SI Appendix, Tables S4–S7). To determine if the long-range NOEs arise from intra- or intermolecular contacts within the oligomer, we prepared a 1:1 mixture of labeled and unlabeled 3A-hIAPP and performed a double half-filtered NOESY experiment that selectively captures NOEs between protons on labeled and unlabeled peptides (SI Appendix, Fig. S15). Intermolecular NOEs (Fig. 4F, green) were observed between residues Ala12, Ala13, and Ala15 in the N-terminal region as well as between residues Phe23 and Gly24 within the FGAIL region previously implicated in β-sheet formation by 2D IR studies (SI Appendix, Table S4). Diffusion experiments were not size-selective as the oligomers and monomers are in the fast exchange limit. Thus, the isotope dilution results indicate that the oligomers are at least dimeric in nature and that the monomer–monomer interface involves both the N-terminal (helix) region and the FGAIL β-sheet-forming region.

Fig. 5.

NMR NOE constraints and structural model of hIAPP. (A) NMR constraints (solid red lines) are superimposed on the initial oligomer geometry obtained from NIH-XPLOR. (B) Oligomer structure obtained after MD simulations using the constraints shown in (A). The three alanine substitutions at positions 12, 15, and 31 (yellow) are shown. Location of S20 (green) is the site of the S20G hereditary mutation. Atomic structure of FGAILS (box) shows packing of side chains in this critical region.

Since some residues were not isotope labeled, we built a structural model via a combination of NMR constraints and molecular dynamics simulations (SI Appendix). Initial structures were generated using Xplor-NIH software and NMR-derived constraints (Fig. 5A and SI Appendix, Fig. S16). A structure with a parallel β-sheet consistent with the 2D IR experiments was used as the starting point for replica exchange molecular dynamics with the constraints incorporated into the force field (SI Appendix, Figs. S17 and S18). Shown in Fig. 5B is the most stable conformation found after 1 µs of simulations. It has a free energy of −14.1 kcal/mol relative to the monomer. The structure has an N-terminal dimer of helices. One helix spans residues 4 to 21 while the other extends from 5 to 17, which includes the disulfide bond. Following each helix, there is a kink which allows the two N-terminal regions of the peptides to align and form a parallel β-sheet from residues 23 to 28. Free energy calculations indicate that the helix packing contributes −10.8 kcal/mol and the FGAILS region contributes −4.8 kcal/mol to the overall free energy. The NMR data indicate that the structure is highly dynamic and so is likely sampling a distribution of conformations with this general fold. Indeed, structures with disordered helices also agree with the experimental data, albeit with slightly higher free energies (SI Appendix). Regardless, all stable conformations contained an intermolecular FGAILS β-sheet. Since 3A-hIAPP provided chemical shift assignments, it was then possible to measure NMR spectra of FGAI-labeled hIAPP, confirming that 3A-hIAPP adopts the native-fold (SI Appendix, Fig. S19) as was concluded using 2D IR spectroscopy. This moll.del is a structure obtained for oligomeric hIAPP.

Summary and Implications for Disease

Our model establishes three key structural features of this hIAPP oligomer: 1) a helical bundle formed by two helices of slightly different lengths at the N-terminus (residues 4 to 21 or 5 to 17), 2) a kink (residues 18 to 21), and 3) an intermolecular parallel β-sheet that spans the critical FGAILS region (residues 23 to 28). This model helps interpret many prior biophysical studies. N-terminal α-helices have long been suspected of playing a role in the aggregation mechanism of hIAPP (40, 41). In monomeric studies of hIAPP, α-helices have been detected from residues 8 to 17 (25) that extend to residue 28 in micelles or when bound to negatively charged bilayers (42). Notably, these observations led to the hypothesis that the oligomer forms an N-terminus helical bundle (42, 43), which is consistent with our structure. A kink between residues 18 to 22 was previously identified in monomeric hIAPP using NMR (42). We now show that a kink or loop is necessary to align the C-termini of the two polypeptides, thereby permitting the FGAILS β-sheet. Indeed, the kink seems to explain how the hereditary S20G mutation can lead to early-onset T2D. In vitro, S20G aggregates faster than hIAPP (22, 44), is more cytotoxic in extracellular assays (44, 45), and islets from transgenic mouse models expressing S20G are more impaired than humanized mouse controls (46). Glycine substitutions at turns usually stabilize structure because glycine can adopt a wide range of dihedral angles (47, 48). Thus, we hypothesize that the S20G mutation leads to a more flexible kink (Fig. 5B, residue S20 shown in green), which stabilizes the oligomer by making it easier to transition from the N-terminus helical bundle to the parallel C-terminus strands. Stabilizing the oligomer would increase its concentration, consistent with the increased toxicity and faster aggregation.

This structural model also helps explain the long-established correlation between 20 and 29 sequences and T2D susceptibility across mammals (39–41). It was originally postulated that mutations in this region disrupt β-sheets of the amyloid fibrils. That could still be true for some fibril polymorphs, but they will certainly destabilize the oligomer by preventing formation of the FGAILS β-sheet. For instance, of the six amino acids in the rat IAPP sequence that differ from human IAPP, (Fig. 1), four are within the FGAILS region and one flanks it. The A25P substitution places a proline in the middle of FGAILS, which would prevent β-sheet formation due to steric constraints. Thus, we believe that mammalian sequences long known to prevent amyloid plaque formation are doing so by preventing oligomer formation and, consequently, the cytotoxic effects of the oligomers. Indeed, we have recently reported that the beta-sheet at G24 is conserved across several species that develop type 2 diabetes and not in others (49).

Indeed, it was the correlation of disease pathology with mutations within the 20 to 29 sequence that led us to postulate that mutations outside this region could trap oligomers without altering their physiological toxicity. Alanine was chosen so as not to disrupt helices that might exist in the oligomers while also destabilizing the hydrophobic packing of the fibrils. We suspect that there are many mutations that might be used to prolong aggregation with minimal perturbations to the oligomer and that by studying additional variations, the rigor of our structural model could be evaluated. We also note that neither the 2D IR nor the NMR experiments presented here are size-selected, due to fast exchange on the NMR timescale. Oligomers at least as large as hexamers exist, according to the cross-linking data (Fig. 2C). We consider our dimer model the minimal oligomer structure that satisfies the experimental data, but larger oligomers with a similar fold would also be consistent with our results.

We note that the oligomer structure is unlike any known structures of hIAPP fibrils (21–26). Thus, a large structural rearrangement is needed for the oligomer to adopt the structure of the fibril upon aggregation. We previously found that this structural transition between oligomer and fibril created a free energy transition state, enabling a stable population of oligomers that far exceeds concentrations predicted by standard nucleation theory (28). Thus, the oligomers are not simply seeds that template monomers (50), but represent a distinct structural species.

The correlation between sequence, aggregation, and propensity for a species to develop T2D has been known for decades (30–32). Indeed, antibody blocking therapies have been tested based on this correlation (16–18). The structural model reported here, developed from physiological observations and tested against the native protein, provides a structural basis to interpret these long-held empirical correlations. Our approach, using 2D IR spectroscopy, biochemical assays, and 2D/3D NMR, might also be used to determine the oligomer structure of other mammalian IAPP or the familial S20G mutation. A rigorous connection between oligomer structure, toxicity, and physiological response will help a therapeutic approach to combating T2D.

Materials and Methods

Sample Preparation.

hIAPP and 3A-hIAPP were synthesized with FMOC microwave-assisted solid phase peptide synthesis and peptide amide linker-polyethylene glycol-polystyrene (PALPEG-PS) resin for an amidated C-terminus. FMOC-protected ¹³C¹⁵N isotopically labeled amino acids (¹³C 99%, ¹⁵N 99%) were obtained from Cambridge Isotope Laboratories. FMOC-¹³C amino acids (99%, CIL) underwent ¹⁸O exchange using ¹⁸O water and anhydrous 1,4 dioxanes containing 4 M HCl. Synthesized peptides were cleaved from the resin and side chain protecting groups were simultaneously removed using an acidic cleavage cocktail comprised of trifluoroacetic acid: thioanisole: ethanedithiol: anisole (90:5:4:1, v/v/v/v) at 37 °C for 30 min using a microwave assisted cleavage system (CEM). Cleaved peptides were then crudely purified by washing three times with diethyl ether. The Cys2-Cys7 disulfide bond was formed by adding DMSO to peptides for 24 h followed by a 50% dilution with 20% acetic acid for an additional 12 h. hIAPP was then purified by two rounds of reverse phase HPLC on a C18 column (XSelect CSH, Waters) with a gradient of 95% Buffer A consisting of 0.045% HCl in water (v/v) and 5% Buffer B consisting of 80% acetonitrile, 20% water, and 0.045% HCl (v/v/v), changing at 1% per minute until 45% Buffer B was reached. After purification round 1, the samples are collected and lyophilized, resuspended in 20% hexafluoroisopropanol (HFIP) and 80% water, and repurified via reverse phase HPLC using the same gradient described above. Sample identities were confirmed by MALDI-TOF Mass Spectrometry (Bruker Ultraflex™). The ¹⁸O exchange efficiency of ¹³C amino acids was analyzed using electrospray ionization (ESI) mass spectrometry. All ¹³C¹⁸O Fmoc-amino acids had >91% labeling efficiency. Peptide samples were quantified using 280 nm absorbance on a NanoDrop™ (Thermo Fisher), calculated with an extinction coefficient of 1,490 M⁻¹ cm⁻¹, dissolved in hexafluoroisopropanol (HFIP) to fully disaggregate the peptides, and stored as a lyophilized powder at −20 °C.

Thioflavin-T Assay.

ThT fluorescence was measured using a Perkin-Elmer Envision plate reader with a 430 nm excitation filter (20 nm bandwidth) and 492 nm emission filter (20 nm bandwidth). Aggregation assays were initiated by resuspending lyophilized protein samples in 50 mM Tris or NaAc, respectively, with 10 µM ThT and immediately transferring to a 96-well flat-bottom plate (Corning 96 Well Half-Area Microplate). A glass cover was attached to the top of the plate to prevent evaporation. Fluorescence measurements were taken at 25 °C every 5 min. Final conditions were 100 μM hIAPP, 10 µM ThT, 50 mM Tris, or 50 mM NaAc for a final volume of 90 μL per well. Results were fit to a Boltzmann sigmoidal with 2 to 3 replicates per condition. The concentration of hIAPP was lowered to 100 μM relative to other experiments due to large amounts of sample required. When collecting data from multiple wells, the plate reader moves between the wells every 5 min. When a single well is measured the plate reader does not move. In the latter situation, which we call “gentle,” the aggregation time can be up to twice as long. Not all experiments could be done a single well at a time.

TEM.

2 μL aliquots of 500 μM hIAPP in either 50 mM Tris (pH 7.4) or 50 mM NaAc (pH 5.3) at the indicated time points were negatively stained with methylamine tungstate (Nanoprobes Nano-W™) and blotted onto a carbon-coated copper grid. TEM images were taken at the UW-Madison School of Medicine and Public Health Electron Microscope Facility.

Cross-Linking Gel.

500 μM of peptide was incubated in 50 mM Tris (pH 7.45) or 50 mM NaAc (pH 5.3) for 30 min The peptide was then diluted with more buffer and Tris(2,2-bipyridyl)dichlororuthenium(II) hexahydrate (Ru(II)) and ammonium persulfate (APS) were added to the peptide giving a final ratio of 70:3.5:1 APS: Ru(II):hIAPP. The sample was irradiated with a 150 W bulb for 10 s and quenched with a 5% β-mercaptoethanol solution. 2X SDS (Invitrogen) was added, samples were heated at 85 °C for 3 min and centrifuged at 14,000 rpm for 4 min. Samples were then loaded into a 10% Tris-Tricine gel (Invitrogen) and ran at 30 V for 30 min and then ramped up to 90 V for 2 h. The gel was then silver stained (Thermo SilverXpress™) and imaged.

Cell Viability Assay.

Rat insulinoma (INS-1 832/13) cells were grown in a sterile 96-well, clear bottom plate in RPMI 1490 media supplemented with 1 mM sodium pyruvate, 2 mM L-glutamine, 10% fetal bovine serum (FBS), 50 µM β-mercaptoethanol, 10 mM HEPES, and 50 mM sodium bicarbonate. The cells were plated at 50,000 cells/well and incubated for 24 h to insure they were healthy. Cell media was replaced with 50% volume of a 2× RPMI media. For assays with varying peptide concentrations, the peptides were dissolved at a 2× concentration in water and immediately added to the cells yielding 1× peptide and RPMI concentrations. Peptide concentrations were measured again by UV–Vis spectroscopy to verify concentrations. Another cell viability assay was run with hIAPP concentrations of 300 μM that were incubated in 50 mM Tris (pH 7.4) or 50 mM NaAc (pH 5.3) for 5 h (SI Appendix, Fig. S7). These peptide samples were then added to cells after 5 h as mentioned prior with a final concentration of 150 μM hIAPP and 1× RPMI media per well. The peptides were incubated with the cells for 24 h after which 10 μL of alamarBlue was added. After 2 h, fluorescence of the wells was measured (excitation 545 nm; emission 590 nm) with a plate reader, and the data were fit to a Boltzmann sigmoidal relative to media control.

Two-Dimensional Infrared Spectroscopy.

The 1,030 nm, 0.4 mJ output from a Yb:KGW amplifier (Carbide, Light conversion) served as an input to the twin OPA (Orpheus, Light conversion). The spatiotemporal overlap of pump and signal generates mid-Infrared (mid-IR) pulses ~75 fs centered at 6,200 nm (Lyra, Light conversion). The repetition rate was set to 10 kHz to avoid thermal heating and transient grating effects (51). Mid-IR pulses then enter a home-built 2D IR spectrometer, where the mid-IR pulse is split into pump and probe pulses using a 75/25 ZnSe beam splitter. The amplitude and phase of 75% pump pulses are regulated through a pulse shaper set in a 4f geometry. The pulse shaper uses principles of acousto-optic modulation (AOM) where a radio frequency (RF) waveform is transduced through a Ge crystal. This RF waveform was generated using an arbitrary waveform generator (AWG) (Signatec) and amplified (Isomet inc.) before it reaches the transducers. To collect a 2D IR spectrum, pump pair of different phases are generated, and coherence time (t₁ delay) is scanned up to 3 ps. A four-frame phase cycling scheme that removes the pump–probe scatter and transient absorption background was used. To further improve the quality of spectra, perpendicular polarization scheme <0°, 0°, 90°, 90°> was used to collect all data. Pump and probe were focused on the sample using 90° parabolic mirrors. Probe beam is dispersed on a nitrogen cooled mercury cadmium telluride (MCT) detector (Infrared systems) using a 75 g/mm grating setting the resolution to 4 cm⁻¹. The analog to digital conversion was done using JackHammer (PhaseTech). To prepare 2D IR samples, 500 µM of lyophilized peptide was reconstituted with 50 mM Tris (pH 7.4) prepared in D₂O and placed in the 2D IR sample cell comprising of two CaF₂ windows separated by a spacer thickness of 75 µm. All the diagonal slices are plotted through the peak maxima.

NMR Experiments.

Lyophilized 3A-hIAPP was dissolved into NMR buffer (50 mM deuterated acetic acid, pH 5.3, or 50 mM TRIS pH 7.4, 2 to 5 mM TRIS pH 4.0) containing 10% D₂O for locking. All the samples were prepared at 500 µM at 4 °C. 50% mixed samples were prepared by mixing the unlabeled and labeled peptide in HFIP and lyophilized. All the NMR experiments were performed at a given temperature; all the samples were equilibrated at least 30 min after the required temperature is reached. Experiments were performed on 700, 800, and 900 MHz spectrometers equipped with cryoprobe. Backbone and side chain assignments were done using triple resonance experiments. Side chain assignments are done using 3D-HCCH-TOCSY. All the spectra were processed using NMRpipe and analyzed using NMRFAM-sparky. Reported resonance assignments are available through the BMRB accession number 53307, and the NMR data are publicly available on the network for advanced NMR (NAN) Data Browser under ARK ID “c16be2f687-3f23-4986-9181-d1fd71cffc56.1.”

Structure Calculation.

The 3A-hIAPP dimer used for free-energy calculations with well-tempered metadynamics was obtained through a series of molecular dynamics (MD) simulations, including several enhanced sampling methods as described in SI Appendix. First, 10 dimer structures were generated from the NMR data using the Xplor-NIH package version 3.6.6. Each of the 10 initial structures was subjected to a multistep simulation protocol schematically illustrated in SI Appendix, Fig. S17. First, a 300 ns MD simulation was performed during which all NMR constraints were applied. These constraints were then gradually released in successive 300 ns intervals, sequentially removing interpeptide, interresidue, and helical restraints. This step is followed by an unrestrained 1 µs simulation. The structures that did not exhibit stability were discarded. The initial set of 10 dimer structures was reduced to 8 structurally stable dimers. For these eight selected structures, the NMR constraints were reapplied, and the simulation protocol was repeated. Stable structures that satisfied the most NMR constraints were selected and subjected to enhanced sampling simulation (more details can be found in SI Appendix). Additionally, a few intermediate conformations that were likely forming stable structures were extracted as well. Finally, enhanced sampling methods identified the most stable conformation that best satisfied the NMR constraints. This optimal dimer structure was subsequently used for free-energy calculations. Detailed methodology for the structure calculations is provided in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

NMR data have been deposited in Network for Advanced NMR (c16be2f687-3f23-4986-9181-d1fd71cffc56.1) (52).