Mechanism of IAPP amyloid fibril formation involves anintermediate with a transient β-sheetLauren E. Buchanana, Emily B. Dunkelbergera, Huong Q. Trana, Pin-Nan Chengb, Chi-Cheng Chiuc,d, Ping Caoe,Daniel P. Raleighe, Juan J. de Pabloc,d, James S. Nowickb, and Martin T. Zannia,1

aDepartment of Chemistry, University of Wisconsin–Madison, Madison, WI 53706-1396; bDepartment of Chemistry, University of California, Irvine, CA92697-2025; cInstitute for Molecular Engineering, The University of Chicago, Chicago, IL 60637; dArgonne National Laboratory, Lemont, IL 60439;and eDepartment of Chemistry, State University of New York, Stony Brook, NY 11794-3400

Edited by Robert Tycko, National Institutes of Health, Bethesda, MD, and accepted by the Editorial Board October 15, 2013 (received for reviewAugust 1, 2013)

Amyloid formation is implicated in more than 20 human diseases, yetthe mechanism by which fibrils form is not well understood. We use2D infrared spectroscopy and isotope labeling to monitor the kineticsof fibril formation by human islet amyloid polypeptide (hIAPP oramylin) that is associated with type 2 diabetes. We find that anoligomeric intermediate forms during the lag phase with parallelβ-sheet structure in a region that is ultimately a partially disorderedloop in the fibril. We confirm the presence of this intermediate,using a set of homologous macrocyclic peptides designed to rec-ognize β-sheets. Mutations and molecular dynamics simulationsindicate that the intermediate is on pathway. Disrupting the olig-omeric β-sheet to form the partially disordered loop of the fibrilscreates a free energy barrier that is the origin of the lag phaseduring aggregation. These results help rationalize a wide range ofprevious fragment and mutation studies including mutations inother species that prevent the formation of amyloid plaques.

inhibitors | aggregation pathway | vibrational coupling

The misfolding of proteins into β–sheet-rich amyloid fibrils isassociated with the pathology of more than 20 human dis-

eases, including Alzheimer’s, Parkinson, and Huntington dis-eases (1). Amyloid plaques are formed by masses of fibrils, butgrowing evidence suggests that the toxic species may be prefi-brillar intermediates (2, 3). As a result, there is much interest inunderstanding the mechanism by which these proteins formfibrils and identifying intermediates in the aggregation pathway.However, obtaining structural information about intermediatespecies is difficult due to their transient nature. Solid-state NMR(ssNMR) and X-ray crystallography provide high-resolution struc-tures of fibrils (4, 5) and optical techniques can track structuralchanges in real time (6, 7), but few techniques have both thestructural and the temporal resolution to extract specific structuraldetails about intermediates. Fragments have been trapped in in-triguing oligomeric structures that may represent intermediatestates (5, 8) and transient secondary structures are known to existfrom circular dichroism measurements and other experiments (9–11), but for full-length proteins it has been difficult to identify thespecific residues that contribute to the secondary structure and thusunderstand their role in the aggregation mechanism. In this paper,we use 2D infrared (2D IR) spectroscopy and isotope labeling tomonitor the structural evolution of the full-length human isletamyloid polypeptide (hIAPP or amylin), a 37-residue peptide im-plicated in type 2 diabetes. We observe the formation of a struc-tured prefibrillar intermediate in a region that has long been knownto influence aggregation, but that does not form well-ordered cross-β structure in the amyloid fibril. Its presence provides uniquestructural insights into the mechanism of amyloid aggregation andhelps unify many seemingly inconsistent prior studies.Many studies on hIAPP have focused on the region spanning

residues 20–29 because it plays a critical role in fibril formation.Amyloid fibrils are found only in humans, cats, and some primates(12) even though the sequence for all mammals is highly con-

served except for residues 20–29 (Fig. S1A). Early work revealeda correlation between the ability to form amyloid in vitro and thesequence in the 20–29 region (13, 14). The fragment hIAPP20–29(SNNFGAILSS) readily forms amyloid fibrils, as do smallerfragments such as hIAPP23–27 (13–17) and hIAPP21–27; the latterhas been characterized with X-ray crystallography (5). Moleculardynamics simulations corroborate the assembly of such frag-ments into β-sheet structures (18–20). Although other segmentsalso form amyloid fibrils (21, 22), the simplest explanation forthe correlation between the sequence and propensity for fibrilformation is that the region near 20–29 forms β-sheets that arethe “amyloidogenic core” of the fibrils (14, 15, 17, 23). Thishypothesis is also consistent with many single-point mutationswithin residues 20–29 of hIAPP that can decrease or preventfibril formation (23–25). However, this long-standing hypothesiswas recently discredited by ssNMR, which showed that the ma-jority of the residues attributed to the amyloidogenic core did notform β-sheets in the fibril, but instead created a partially disor-dered loop (Fig. 1 A and B) (4). A loop should be able to ac-commodate many types of mutations, making it unclear why thisregion has such a dramatic impact on aggregation. The resultspresented here reveal an intermediate with parallel β-sheetstructure spanning residues 23–27 that is ultimately disrupted toform the loop of the fibril. The location of this transient β-sheetexplains the importance of the FGAIL region in the aggregationof hIAPP and provides a structural explanation for the lag phasethat is a unifying feature of all amyloid proteins.

Significance

There is an enormous interest in the mechanism by whichproteins misfold and aggregate into amyloid fibrils. Amyloidhas been implicated in many human diseases, but the mecha-nism of aggregation is not understood. Intermediates havebeen postulated to play an important role in the process, butthere have been very few direct measurements that providespecific structural details. The use of isotope labeling and 2DIR methods has allowed the characterization of a criticalintermediate generated during amyloid formation by isletamyloid polypeptide, the peptide responsible for amyloidformation in type 2 diabetes. Identification of this intermediateprovides a structural explanation for the lag phase and mayexplain why some species develop amyloid deposits of hIAPPwhile others do not.

Author contributions: L.E.B., E.B.D., D.P.R., J.J.d.P., J.S.N., and M.T.Z. designed research;L.E.B., E.B.D., H.Q.T., P.-N.C., C.-C.C., and P.C. performed research; L.E.B., E.B.D., H.Q.T.,and C.-C.C. analyzed data; and L.E.B. and M.T.Z. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. R.T. is a guest editor invited by the EditorialBoard.1To whom correspondence should be addressed. E-mail: zanni@chem.wisc.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1314481110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1314481110 PNAS Early Edition | 1 of 6

CHEM

ISTR

Y

ResultsThe FGAIL Region Forms β-Sheets in a Lag-Phase Oligomer. Usingrapid-scan 2D IR spectroscopy, we follow the structural changesof hIAPP aggregation by monitoring the frequency of the amideI band created by backbone carbonyl stretches: Random coilstructures absorb near 1,645 cm−1, whereas amyloid β-sheetsabsorb near 1,620 cm−1. The time course of amyloid aggregation,obtained by monitoring the β-sheet mode at 1,620 cm−1, followsa sigmoidal curve consisting of a lag phase, a growth phase, andan equilibrated phase in which the complete fibril structure hasformed (for example, Fig. 2). This growth curve is similar to thatobtained using thioflavin-T fluorescence assays (Fig. S2). Residue-level structural resolution is achieved by isotope labeling a singlebackbone carbonyl with 13C = 18O, which spectroscopically re-solves the labeled residue from the “bulk” unlabeled modes. Ina random coil conformation, isotope-labeled residues typicallyabsorb at 1,595 cm−1. Within an in-register parallel β-sheetconformation, the labeled residues form a coupled linear chainalong the length of the sheet, shifting the isotope mode fre-quency to between 1,570 cm−1 and 1,588 cm−1, depending on theprecise coupling strength and structural order of the labeledresidue. Isotope dilution experiments, in which only 25% of thepeptides are labeled, provide an independent test of β-sheetstructure by eliminating the effects of coupling and its charac-teristic spectral shift without changing the structure (26, 27). Thislabeling strategy has previously been used to study amyloidstructure, aggregation kinetics, and inhibitor binding (26, 28, 29).Samples of 0.5 mM hIAPP labeled with 13C = 18O at residues

V17, F23, G24, A25, L27, and G33 were studied (Fig. 1 A and B).Shown in Fig. 1 C and D are 2D IR spectra and diagonal slices ofthe V17-labeled peptide obtained during the lag phase and the

equilibrated phase, respectively. In Fig. 1C, the pair of broadpeaks centered around 1,645 cm−1 is the signature of a disor-dered peptide. The V17 isotope label appears at 1,595 cm−1 andis very weak and broad, indicating it is also in a random coilconfirmation. The spectrum of the final fibrils (Fig. 1D) displaysa pair of sharp peaks at 1,620 cm−1, which are characteristicfeatures of amyloid β-sheets formed by the unlabeled residues.Likewise, we know that the V17 residue is incorporated intoa parallel β-sheet in the fibril because it exhibits a peak at1,578 cm−1 (Fig. 1D, box and arrow), which we verify with iso-tope dilution (Fig. S3A). Thus, V17 has gone from a disorderedto a β-sheet structure upon fibril formation, as expected becauseV17 resides in a well-ordered region of the N-terminal β-sheetaccording to the ssNMR structure (4, 30).Spectra obtained for the F23-labeled peptide are markedly

different. The lag-phase spectrum (Fig. 1E) has the broad peaksat 1,645 cm−1 expected for a random coil, but the F23 labelexhibits a sharp peak at 1,587 cm−1 (box and arrow), indicative ofparallel β-sheet formation. β-sheet structure is confirmed by di-lution experiments (Fig. S3B). Thus, during the lag phase, mostresidues are disordered whereas F23 participates in a parallelβ-sheet. In the equilibrated fibril, the label absorbs at 1,593 cm-1

and is weak and broad (Fig. 1F), consistent with the ssNMRmodel (4) in which F23 resides within the partially ordered loopof the fibril where the coupling is overwhelmed by the structuraldisorder. The evolution of a sharp peak to a weak and broadfeature, indicating a local structural change from a parallel β-sheetto a more disordered structure, is the opposite of what was ob-served for V17.Monomeric hIAPP cannot produce the coupling necessary to

generate the peak at 1,587 cm−1; only the presence of an aggregated

Fig. 1. Structure, 2D IR data, and TEM of isotope-labeled hIAPP. (A and B) Sequence (A) and ssNMRmodel (B) (4) of hIAPP fibrils. β-Sheets are markedwith black arrows and isotope-labeled positions withbeads. (C–F) Two-dimensional IR spectra and diagonalintensity slices of lag-phase and equilibrated-phaseV17 (C and D) and F23 (E and F ), respectively. (G)Kinetics of the unlabeled β-sheet (1,620 cm−1, blue)and isotope-labeled modes (1,587 cm−1, green) forF23. (H–J) Lag-phase spectra and slices for G24 (H),A25 (I), and L27 (J). Coupled isotope-labeled modesare highlighted with a white box and a red arrow.(K) Electron micrograph of lag-phase hIAPP. (Scalebar, 200 nm.)

2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1314481110 Buchanan et al.

species with parallel β-sheets at F23 will create such a feature.This conclusion is supported by an analysis of the kinetics. Fig.1G shows the intensity of this peak monitored continuously asa function of time throughout aggregation relative to the in-tensity of the unlabeled β-sheet peak at 1,620 cm−1. The F23peak is present at the earliest times measureable, increases toa maximum during the lag phase, and then disappears at thesame rate as the amyloid fibrils form. This rise and fall is theclassic kinetic signature for an intermediate.Similar to F23, samples labeled at G24, A25, and L27 all ex-

hibit a coupled isotope peak during the lag phase (Fig. 1 H–J,boxes and arrows), indicating that the lag-phase intermediate hasparallel β-sheet structure extending at least over residues 23–27(FGAIL). Peak frequencies for both lag-phase and fibril spectraare summarized in Table S1 and additional lag-phase spectra aregiven in Fig. S4. F23 provides the clearest kinetic signature forthe lag-phase β-sheet because F23 is completely uncoupled in thepartially disordered loop and so does not appear as a definedpeak in the fibril. The other residues remain coupled in themature fibrils (Fig. S5 A–C) but are shifted to lower frequenciesthan in the oligomer, reflecting the change in structure from theoligomer to the fibril. Spectral overlap prevents a kinetic analysisof G24 and L27, but the A25 kinetics reveal that the spectralsignature of the oligomer peaks at about the same time for A25as for F23 (Fig. S6), which is consistent with the FGAIL in-termediate forming in a single-step transition.We do not know whether this β-sheet extends past the FGAIL

sequence to include the flanking asparagine and serine residuesbecause these amino acids are difficult to isotope label, but itdoes not extend as far as V17 or G33, neither of which is coupledduring the lag phase (Fig. 1C and Fig. S4B); but V17 and G33

are coupled in the mature fibrils as expected (Fig.1D and Fig.S5D). It is possible that the early β-sheets are limited to residues23–27, which would be consistent with early fragment studiesshowing that hIAPP23–27 is the smallest fragment of hIAPP ca-pable of forming amyloid fibrils (15). However, the hIAPP22–27 frag-ment aggregates 40 times faster than hIAPP23–27 and hIAPP20–29aggregates faster still, suggesting that the oligomeric β-sheets mightextend beyond our labeled region to include the serines (15). None-theless, we know that the β-sheets must be small because, liketypical random coil peptides, the unlabeled region of the spectrahas only a minimal peak at 1,620 cm−1 (and does not extend toV17 or G33). Furthermore, transmission electron microscopy(TEM) images collected during the lag phase do not containfibrils (Fig. 1K and Fig. S7A), in contrast to the images observedafter aggregation is complete (Fig. S7B). Spherical deposits havebeen observed in other amyloid-forming systems and are oftenattributed to oligomeric species (31, 32).

The Use of Rationally Designed Macrocycles Confirms the Presence ofthe FGAIL Intermediate. To provide an independent test of ourmodel, we studied the aggregation of hIAPP in the presence ofrationally designed macrocyclic β-sheet peptides that are ho-mologous to segments of hIAPP. The macrocycles comprise twoanti-parallel β-strands connected by two δ-linked ornithines thatmimic β-turns (Fig. 2A). One strand contains a heptapeptidesequence from hIAPP and the other strand contains an un-natural amino acid, Hao, which constrains the macrocycle to aβ-sheet structure with four internal hydrogen bonds, but cannotform external hydrogen bonds. Thus, the heptapeptide strand ofthe macrocycle is designed to bind to amyloid β-sheets witha matching sequence, whereas the other strand prevents furtherelongation. Similar macrocycles have been used previously tostudy the aggregation of a τ-derived hexapeptide, amyloid-β, β2-microglobulin, and truncated α-synuclein (33, 34).Five macrocycles were synthesized with heptapeptide sequences

that span the length of hIAPP (Fig. 2B). Two macrocycles have se-quences that overlap the N-terminal β-sheet of the fibrils (Mac11–17and Mac15–21), two overlap the C-terminal sheet (Mac26–32 andMac31–37), and one targets the FGAIL region under investigation(Mac21–27). The sequences of the macrocycles are given in Fig.S8A. The macrocycles were mixed at equimolar concentrationswith monomeric hIAPP and 2D IR spectra were collected con-tinuously over the course of 5 h. Shown in Fig. 2C are kinetictraces plotting the intensity of the unlabeled β-sheet mode at∼1,620 cm−1. The macrocycles do not have any spectral featuresin this range (Fig. S8 B–F) and so do not contribute significantlyto the intensity measurements. The four macrocycles that over-lap the fibril β-sheets accelerate hIAPP aggregation, reducing thelag time by 50% or more. The macrocycle that targets the FGAILregion (Mac21–27) is markedly different. It slows amyloid forma-tion, extending the lag time by 150%. TEM of equilibrated sam-ples confirms the formation of amyloid fibrils (Fig. S7 C and D).These results provide further evidence for an oligomeric in-

termediate. It is well established that the ends of sonicatedamyloid fibrils will seed fibril growth by providing a β-strand towhich hIAPP monomers template to form fibrils. Because thesemacrocycles are homologous with the β-sheets of the fibrils, ac-celerated kinetics are consistent with these macrocycles actingas seeds and templating formation of the N- and C-terminalβ-sheets of the fibrils. In contrast, Mac21–27 does not seed fibrilformation because its target region is disordered in the fibril.Instead, Mac21–27 inhibits fibril formation by stabilizing inter-molecular interactions that must ultimately be disrupted to formthe final fibril structure. This mode of inhibition is consistentwith it recognizing the FGAIL β-sheet observed in the 2D IRexperiments. The fact that Mac21–27 is the only macrocycle thatinhibits fibril growth is further evidence for formation of an in-termediate containing β-structure in the FGAIL region.

Fig. 2. Designed macrocycles and their effect on hIAPP aggregation. (A)General design of macrocyclic β-sheet peptides, including the “recognition”β-strand, the “blocking” β-strand (with the Hao unit in red), and the δ-linkedornithine turns (in blue). (B) Sequence of hIAPP labeled by its secondarystructure. Sequences targeted by the recognition strand of the macrocycle areunderlined. (C) Aggregation kinetics of hIAPP fibrils with and without mac-rocycles, monitored at 1,620 cm−1 and normalized to the equilibrated intensity.

Buchanan et al. PNAS Early Edition | 3 of 6

CHEM

ISTR

Y

The FGAIL Intermediate Is Predicted by Molecular Dynamics Simulation.To gain insight into the thermodynamics of hIAPP self-assembly,we use a bias-exchange metadynamics technique to calculate thefree energy landscape for two hIAPP monomers (35, 36). Wedescribe the surface with two order parameters, Q8–16 and Q27–35,where the subscripts refer to the hIAPP sequence. These order pa-rameters are calculated by comparison with the N- and C-terminalβ-sheets of the ssNMR fibril structure (4) and are designed tomeasure how closely the simulated structure matches the fibril (SIMaterials and Methods). The resulting free energy landscape (Fig.3) contains several intermediates labeled with Roman numerals.For each minimum, representative structures are presented withthe FGAIL sequence colored in red. The map illustrates twopossible minimum free energy pathways for the dimerizationprocess: I → II → III → V and I → II → VI → V. The first in-termediate encountered along either pathway is intermediate II,which contains a parallel β-sheet from residues 23–29, corre-sponding to the FGAILSS segment. Thus, the free energy land-scape predicts that, as observed in experiments, the FGAIL regionforms parallel β-sheets.This landscape provides a physical rationale for the existence

of the experimentally observed FGAIL intermediate. To prog-ress from the intermediate (II) to the final structure (V), theFGAILSS β-sheet must break into a partially disordered loop.The loss of the FGAILSS structure is reflected in the high freeenergies of the intermediates III and VI, but the folded dimerwith fully formed N- and C-terminal β-sheets (V) exhibits thelowest free energy. Thus, there is a free energy barrier associatedwith the breaking of the intermediate FGAILSS β-sheets into thepartially disordered loop of the fibril, in agreement with experiment.

The metadynamics approach provides information on thermo-dynamics, not kinetics, and so experimental timescales are notrelevant. For a dimer, the free energy minima all lie within ∼1.4kJ/mol and thus all species will be populated at room tempera-ture; however, we expect that intermediate II and the fibril willbe further stabilized in simulations with additional peptides be-cause free energy scales with the number of strands for smallβ-sheets (18).

Mutation Suggests That the FGAIL Oligomer Is on Pathway. Thequestion remains as to whether these oligomers are on or offpathway. Kinetics experiments cannot distinguish between thesetwo possibilities, nor can the macrocycle data because stabilizingthe intermediate will slow fibril formation in either case. For thisreason, we turn to the proline mutation I26P. Proline disruptsβ-sheets and so I26P will prevent at least a portion of the FGAILintermediate from forming. Destabilizing an on-pathway specieswill slow or prevent fibril formation (Fig. 4A), whereas destabi-lizing an off-pathway species should either not affect fibril for-mation or enhance it by closing a competing pathway (Fig. 4B).We find that the aggregation of I26P is >10 times slower thanthat of native hIAPP (Fig. 4C and Fig. S9 A–C), which is con-sistent with an on-pathway intermediate. To verify this conclu-sion, we performed two control experiments. First we confirmedthat the altered kinetics are due to disruption of the intermediaterather than the fibril β-sheets by adding Mac31–37. It seeds I26Pfibril formation as it does with wild-type hIAPP, thus provingI26P can readily form fibrils and the proline mutation is notsimply incompatible with the fibril structure. Second, we addedMac21–27. Rather than inhibiting aggregation as it did with wild-type hIAPP, the macrocycle accelerates the aggregation of I26P.The reversed role is consistent with Mac21–27 partially counter-acting the effect of the proline and helping to template the for-mation of the on-pathway hIAPP23–27 β-sheets. As a result, weconclude that the oligomeric species identified in this paper islikely an on-pathway intermediate in the fibril formation pathway.

Discussion and ConclusionsThese experiments and simulations show that hIAPP fibril for-mation is a multistep process and that the hydrophobic sequenceFGAIL plays a critical role in the aggregation mechanism underthese conditions. It drives aggregation by forming a parallel β-sheetduring the lag phase, but the FGAIL β-sheet is transient becausethe N- and C-terminal β-sheets of the fibril are ultimately favoredby thermodynamics, for reasons that we discuss below. Our ex-periments and free energy calculations suggest that the lag phaseduring hIAPP aggregation under these conditions arises froma free energy barrier caused by breaking the intermolecularFGAIL β-sheets into the partially disordered loop of the fibril,thereby creating a free energy surface for amyloid formation thatlooks qualitatively like that shown in Fig. 5. The barrier may bevery large, depending on the number of peptides contributing tothe oligomeric β-sheet that must be disrupted. Once the turn isformed, aggregation into amyloid fibrils proceeds quickly. Mac-rocycles that match the fibril β-sheets circumvent this barrierwhereas Mac21–27 stabilizes the intermediate, increasing thebarrier and thus lengthening the lag phase.The data unambiguously indicate that the β-sheets have par-

allel, in-register strands across the FGAIL region. We cannotdetermine the size of these oligomers because both coupling andstructural flexibility determine β-sheet frequencies, but the ob-served frequencies (1,585–1,588 cm−1) are higher than thoseobserved in the fibril and consistent with residues lying at thefrayed edges of fibril β-sheets (30), implying a fair amount ofstructural disorder or smaller couplings than in typical flatβ-sheets. Aromatic stacking interactions and hydrophobic col-lapse, often considered to be driving forces for fibril formation(37, 38), may drive the formation of the oligomers, although the

Fig. 3. Conformational free energy map for the dimerization of hIAPP interms of order parameters Q8–16 and Q27–35. Contour lines are drawn every2.5 kJ/mol. Representative dimer conformations for the seven local freeenergy minima are labeled I–VII, respectively, and the minimum free energyvalue in kJ/mol with respect to structure I is listed. Residues 23–27 arehighlighted with red and the N-terminal disulfide bridge is represented witha stick model. Two minimum free energy paths of dimerization are high-lighted by a white dashed arrow.

4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1314481110 Buchanan et al.

amino acid composition of a peptide may not determine its ag-gregation propensity as much as the specific sequence (39).Thefibrillation propensity of hIAPP, as calculated in the ZipperDBdatabase compiled by Eisenberg and coworkers (39), predicts theFGAIL region as the most amyloidogenic region in the poly-peptide (Fig. S1B). Crystal structures of short fragments ofamyloidogenic proteins have highlighted the importance of stericzippers in which the side chains of stacked β-sheets interdigitateto convey stability (40). Likewise, hIAPP fibrils have interdig-itated side chains (4). Interestingly, crystal structures of theNNFGAIL fragment do not form a steric zipper, but instead atight and dry interface formed by the main chains (5). Thus, wepostulate that the amyloidogenic nature of the FGAIL sequencedrives the formation of intermediate β-sheets, which might existas stacks of β-sheets, but the fibril is thermodynamically favorablebecause it has more extensive β-sheets and interdigitated side chains.hIAPP can form at least two structures of fibrils, known as poly-

morphs (4). hIAPP fibrils prepared according to our methodshave a backbone structure that closely matches that of the knownpolymorphs (29, 30), which differ in the packing of the side chainsbut not in the β-sheet secondary structure of the backbone. Be-cause our 2D IR experiments do not probe side-chain packing,such polymorphs are not necessarily apparent in the spectra eventhough they are likely to be present. For the same reason, theremay be multiple structures of the FGAIL intermediate to whichwe are insensitive, each leading to a different fibril polymorph. Ifso, the barriers associated with each polymorph must be similarbecause only a single sigmoidal rise is observed in the kinetic traces.The intermediate appears to explain why mutations within and

near the FGAIL region have a large impact on fibril formation.For example, rats, hamsters, and rabbits have multiple prolinemutations within the 20–29 region that prevent the formation ofthe intermediate β-sheet, impeding or altogether inhibiting fibrilformation (12). Nonnatural mutations, such as the inclusion ofN-methylated amino acids (25) and the I26P mutant used in thiswork, further show that altering the FGAIL sequence dramaticallyaffects the aggregation of hIAPP. In fact, peptide inhibitors thatslow down or prevent hIAPP fibrillization often have mutationsnear residues 20–29 (23, 41, 42). Of course, mutations outside this

region can play an important role by destabilizing the fibril itself(21, 22) and the aggregation pathway may be different in vivo or atdifferent concentrations (43, 44). These results suggest that a strat-egy for preventing fibril formation and cytotoxic species could beto design inhibitors that disrupt the ordering of the FGAILintermediate rather than targeting the fully formed fibrils.In addition to helping explain the differences in aggregation

propensity between species, the FGAIL intermediate also rec-onciles many of the original fragment studies (from which it washypothesized that residues 20–29 formed the core β-sheets of thefibril) with the fibril structure (in which these residues are mostlydisordered). We propose that the fragments highlight the im-portance of the intermediate in the aggregation mechanismwhereas the full sequence is necessary to generate the actualfibril structure. This intermediate also sheds light on how structureinfluences amyloid kinetics. A common feature of all amyloid-forming proteins is the lag phase. Recent efforts to model am-yloid formation with mass action expressions have highlightedthe importance of primary and secondary nucleation (45–47). Thesemodels were generated from experiments on other proteins andunder different conditions, but so far have not invoked stableintermediates or provided a structural mechanism that accountsfor a lag phase. For hIAPP, the barrier created by the FGAILintermediate impedes the proteins from adopting their final foldand so slows down the structural kinetics to create the lag phase.The barrier explains our original work on hIAPP kinetics inwhich we found that the N- and C-terminal β-sheets formed laterthan the loop (29), which we illustrate by the subsequent steps inFig. 5. It is not known whether other amyloid-forming proteinshave stable on-pathway intermediates, but fragments that formfibrils have been found in many other proteins, which may serveas a guide for uncovering other intermediate species.

Materials and MethodsA full description of methods is given in SI Materials and Methods. Briefly,hIAPP (48) and macrocyclic peptides (34) were synthesized and purifiedaccording to published methods. hIAPP was dissolved in deuteratedhexafluoroisopropanol to fully disaggregate the peptides. Aggregation wasinitiated by reconstituting lyophilized samples in deuterated buffer (20 mMTris, pH = 7.4). The final total peptide concentration was 0.5 mM for purehIAPP samples, 1 mM for 75% isotope-diluted experiments, and 1 mM for

Fig. 4. Effect of I26P mutation on hIAPP aggregation. (A and B) Schematicsshowing the effects of proline mutation I26P on the aggregation of hIAPP, ifthe identified oligomers are (A) on pathway or (B) off pathway. (C) Aggre-gation kinetics of I26P fibrils, monitored at 1,620 cm−1, alone (red) and withmacrocycle Mac31–37 (cyan) or Mac21–27 (green).

Fig. 5. Schematic energy landscape for hIAPP aggregation with schematicstructures. Residues 23–27 are highlighted in red. Mac21–27 binds to theFGAIL intermediate, thereby stabilizing the intermediate (solid cyan) andincreasing the free energy barrier, whereas the other four macrocycles cir-cumvent the barrier by seeding fibril growth (green arrow). Proline muta-tions within the FGAIL sequence, such as I26P, inhibit aggregation bydestabilizing the intermediate (dashed black line), which can still be inducedto form by the addition of Mac21–27 (dashed cyan line).The subsequent stepsare taken from our previous work on hIAPP aggregation (29).

Buchanan et al. PNAS Early Edition | 5 of 6

CHEM

ISTR

Y

samples containing equimolar hIAPP and macrocycle. Experiments with I26Pwere conducted at double the concentration of wild-type samples due to itssignificantly slowed kinetics. Two-dimensional IR spectra were collected aspreviously described (49). Bias-exchanged metadynamics simulations wereconducted as described previously (36), using the ssNMR structure (4) asa reference.

ACKNOWLEDGMENTS. We thank Randall Massey at the University ofWisconsinMedical School ElectronMicroscope Facility for helpwith TEM. Supportfor this researchwas provided by theNational Science Foundation (NSF) GraduateResearch Fellowship ProgramGrantDGE-0718123 (to L.E.B.), National Institutes ofHealth (NIH) Grant DK79895 (to E.B.D., H.Q.T., and M.T.Z.), NSF Grant CHE-1112188 (to P.-N.C. and J.S.N.), NIH Grant 1R01DK088184 (to C.-C.C. and J.J.d.P.),and NIH Grant GM078114 (to P.C. and D.P.R.).

1. Chiti F, Dobson CM (2006) Protein misfolding, functional amyloid, and human disease.Annu Rev Biochem 75(1):333–366.

2. Butler AE, Janson J, Soeller WC, Butler PC (2003) Increased β-cell apoptosis preventsadaptive increase in β-cell mass in mouse model of type 2 diabetes: Evidence for roleof islet amyloid formation rather than direct action of amyloid. Diabetes 52(9):2304–2314.

3. Kayed R, et al. (2003) Common structure of soluble amyloid oligomers implies com-mon mechanism of pathogenesis. Science 300(5618):486–489.

4. Luca S, Yau W-M, Leapman R, Tycko R (2007) Peptide conformation and supramo-lecular organization in amylin fibrils: Constraints from solid-state NMR. Biochemistry46(47):13505–13522.

5. Wiltzius JJW, et al. (2008) Atomic structure of the cross-β spine of islet amyloidpolypeptide (amylin). Protein Sci 17(9):1467–1474.

6. Ban T, Hamada D, Hasegawa K, Naiki H, Goto Y (2003) Direct observation of amyloidfibril growth monitored by thioflavin T fluorescence. J Biol Chem 278(19):16462–16465.

7. Measey TJ, Schweitzer-Stenner R (2011) Vibrational circular dichroism as a probe offibrillogenesis: The origin of the anomalous intensity enhancement of amyloid-likefibrils. J Am Chem Soc 133(4):1066–1076.

8. Laganowsky A, et al. (2012) Atomic view of a toxic amyloid small oligomer. Science335(6073):1228–1231.

9. Suzuki Y, et al. (2013) Resolution of oligomeric species during the aggregation of Aβ1-40 using (19)F NMR. Biochemistry 52(11):1903–1912.

10. Stroud JC, Liu C, Teng PK, Eisenberg D (2012) Toxic fibrillar oligomers of amyloid-βhave cross-β structure. Proc Natl Acad Sci USA 109(20):7717–7722.

11. Kirkitadze MD, Condron MM, Teplow DB (2001) Identification and characterization ofkey kinetic intermediates in amyloid β-protein fibrillogenesis. J Mol Biol 312(5):1103–1119.

12. Cao P, et al. (2013) Islet amyloid: From fundamental biophysics to mechanisms ofcytotoxicity. FEBS Lett 587(8):1106–1118.

13. Betsholtz C, et al. (1989) Sequence divergence in a specific region of islet amyloidpolypeptide (IAPP) explains differences in islet amyloid formation between species.FEBS Lett 251(1-2):261–264.

14. Westermark P, Engström U, Johnson KH, Westermark GT, Betsholtz C (1990) Isletamyloid polypeptide: Pinpointing amino acid residues linked to amyloid fibril for-mation. Proc Natl Acad Sci USA 87(13):5036–5040.

15. Tenidis K, et al. (2000) Identification of a penta- and hexapeptide of islet amyloidpolypeptide (IAPP) with amyloidogenic and cytotoxic properties. J Mol Biol 295(4):1055–1071.

16. Ashburn TT, Auger M, Lansbury PT (1992) The structural basis of pancreatic amyloidformation: Isotope-edited spectroscopy in the solid state. J Am Chem Soc 114(2):790–791.

17. Glenner GG, Eanes ED, Wiley CA (1988) Amyloid fibrils formed from a segment of thepancreatic islet amyloid protein. Biochem Biophys Res Commun 155(2):608–614.

18. Wu C, Lei H, Duan Y (2004) Formation of partially ordered oligomers of amyloido-genic hexapeptide (NFGAIL) in aqueous solution observed in molecular dynamicssimulations. Biophys J 87(5):3000–3009.

19. Mo Y, Lu Y, Wei G, Derreumaux P (2009) Structural diversity of the soluble trimers ofthe human amylin(20-29) peptide revealed by molecular dynamics simulations.J Chem Phys 130(12):125101.

20. Rivera E, Straub J, Thirumalai D (2009) Sequence and crowding effects in the ag-gregation of a 10-residue fragment derived from islet amyloid polypeptide. Biophys J96(11):4552–4560.

21. Jaikaran ETAS, et al. (2001) Identification of a novel human islet amyloid polypeptideβ-sheet domain and factors influencing fibrillogenesis. J Mol Biol 308(3):515–525.

22. Nilsson MR, Raleigh DP (1999) Analysis of amylin cleavage products provides newinsights into the amyloidogenic region of human amylin. J Mol Biol 294(5):1375–1385.

23. Porat Y, Mazor Y, Efrat S, Gazit E (2004) Inhibition of islet amyloid polypeptide fibrilformation: A potential role for heteroaromatic interactions. Biochemistry 43(45):14454–14462.

24. Abedini A, Meng F, Raleigh DP (2007) A single-point mutation converts the highlyamyloidogenic human islet amyloid polypeptide into a potent fibrillization inhibitor.J Am Chem Soc 129(37):11300–11301.

25. Yan L-M, Tatarek-Nossol M, Velkova A, Kazantzis A, Kapurniotu A (2006) Design ofa mimic of nonamyloidogenic and bioactive human islet amyloid polypeptide (IAPP)as nanomolar affinity inhibitor of IAPP cytotoxic fibrillogenesis. Proc Natl Acad SciUSA 103(7):2046–2051.

26. Middleton CT, et al. (2012) Two-dimensional infrared spectroscopy reveals the com-plex behaviour of an amyloid fibril inhibitor. Nat Chem 4(5):355–360.

27. Kim YS, Liu L, Axelsen PH, Hochstrasser RM (2008) Two-dimensional infrared spectraof isotopically diluted amyloid fibrils from Abeta40. Proc Natl Acad Sci USA 105(22):7720–7725.

28. Kim YS, Liu L, Axelsen PH, Hochstrasser RM (2009) 2D IR provides evidence for mobilewater molecules in β-amyloid fibrils. Proc Natl Acad Sci USA 106(42):17751–17756.

29. Shim S-H, et al. (2009) Two-dimensional IR spectroscopy and isotope labeling definesthe pathway of amyloid formation with residue-specific resolution. Proc Natl Acad SciUSA 106(16):6614–6619.

30. Wang L, et al. (2011) 2DIR spectroscopy of human amylin fibrils reflects stable β-sheetstructure. J Am Chem Soc 133(40):16062–16071.

31. Hoshi M, et al. (2003) Spherical aggregates of β-amyloid (amylospheroid) show highneurotoxicity and activate tau protein kinase I/glycogen synthase kinase-3β. Proc NatlAcad Sci USA 100(11):6370–6375.

32. Papanikolopoulou K, et al. (2008) Formation of amyloid fibrils in vitro by humangammaD-crystallin and its isolated domains. Mol Vis 14:81–89.

33. Zheng J, et al. (2011) Macrocyclic β-sheet peptides that inhibit the aggregation ofa tau-protein-derived hexapeptide. J Am Chem Soc 133(9):3144–3157.

34. Cheng P-N, Liu C, Zhao M, Eisenberg D, Nowick JS (2012) Amyloid β-sheet mimics thatantagonize protein aggregation and reduce amyloid toxicity. Nat Chem 4(11):927–933.

35. Laio A, Parrinello M (2002) Escaping free-energy minima. Proc Natl Acad Sci USA99(20):12562–12566.

36. Piana S, Laio A (2007) A bias-exchange approach to protein folding. J Phys Chem B111(17):4553–4559.

37. Marshall KE, et al. (2011) Hydrophobic, aromatic, and electrostatic interactions playa central role in amyloid fibril formation and stability. Biochemistry 50(12):2061–2071.

38. Pawar AP, et al. (2005) Prediction of “aggregation-prone” and “aggregation-sus-ceptible” regions in proteins associated with neurodegenerative diseases. J Mol Biol350(2):379–392.

39. Goldschmidt L, Teng PK, Riek R, Eisenberg D (2010) Identifying the amylome, proteinscapable of forming amyloid-like fibrils. Proc Natl Acad Sci USA 107(8):3487–3492.

40. Sawaya MR, et al. (2007) Atomic structures of amyloid cross-β spines reveal variedsteric zippers. Nature 447(7143):453–457.

41. Sellin D, Yan L-M, Kapurniotu A, Winter R (2010) Suppression of IAPP fibrillation atanionic lipid membranes via IAPP-derived amyloid inhibitors and insulin. BiophysChem 150(1-3):73–79.

42. Meng F, Raleigh DP, Abedini A (2010) Combination of kinetically selected inhibitors intrans leads to highly effective inhibition of amyloid formation. J Am Chem Soc132(41):14340–14342.

43. Brender JR, Salamekh S, Ramamoorthy A (2012) Membrane disruption and earlyevents in the aggregation of the diabetes related peptide IAPP from a molecularperspective. Acc Chem Res 45(3):454–462.

44. Williamson JA, Loria JP, Miranker AD (2009) Helix stabilization precedes aqueous andbilayer-catalyzed fiber formation in islet amyloid polypeptide. J Mol Biol 393(2):383–396.

45. Xue W-F, Homans SW, Radford SE (2008) Systematic analysis of nucleation-dependentpolymerization reveals new insights into the mechanism of amyloid self-assembly.Proc Natl Acad Sci USA 105(26):8926–8931.

46. Knowles TPJ, et al. (2009) An analytical solution to the kinetics of breakable filamentassembly. Science 326(5959):1533–1537.

47. Cohen SIA, et al. (2013) Proliferation of amyloid-β42 aggregates occurs througha secondary nucleation mechanism. Proc Natl Acad Sci USA 110(24):9758–9763.

48. Marek P, Woys AM, Sutton K, Zanni MT, Raleigh DP (2010) Efficient microwave-as-sisted synthesis of human islet amyloid polypeptide designed to facilitate the specificincorporation of labeled amino acids. Org Lett 12(21):4848–4851.

49. Middleton CT, Woys AM, Mukherjee SS, Zanni MT (2010) Residue-specific structuralkinetics of proteins through the union of isotope labeling, mid-IR pulse shaping, andcoherent 2D IR spectroscopy. Methods 52(1):12–22.

6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1314481110 Buchanan et al.

Supporting InformationBuchanan et al. 10.1073/pnas.1314481110SI Materials and MethodsHuman Islet Amyloid Polypeptide Synthesis and Purification. Humanislet amyloid polypeptide (hIAPP) samples were synthesized on aCEM Liberty1 automated microwave peptide synthesizer, usingFmoc solid-phase peptide synthesis and standard trifluoroacetic acid(TFA) cleavage and deprotection protocols, as previously described(1). Peptide amide linker-polyethylene glycol-polystyrene (PAL-PEG-PS) resin was used to produce an amidated C terminus. Ala-nine, glycine, phenylalanine, leucine, and valine were 13C18O isotopelabeledat thebackbone carbonyl, starting from13CC1-labeledFmoc-protected amino acids (Cambridge Isotope Laboratories) and 18Owater in a procedure also previously reported (2). Crude peptide wasdissolved overnight in dimethyl sulfoxide (DMSO) to form the di-sulfide bond between Cys-2 and Cys-7. Peptides were purified fromDMSO by reversed-phase HPLC with a Vydac C18 preparativecolumnandusingHClas the counterion, as describedelsewhere (3).

Macrocycle Synthesis and Purification.The synthesis and purificationof macrocyclic peptides is outlined in detail elsewhere (4). Afterpurification, the macrocycles were lyophilized from 1 mM DClmultiple times until residual TFA had been fully removed (5).

Sample Preparation. Lyophilized hIAPP was dissolved to a finalconcentration of 1 mM in deuterated hexafluoroisopropanol(d-HFIP); after 8 h, the peptides were fully deuterated and anyfibrils were disaggregated. Lyophilizedmacrocycles were dissolvedto afinal concentration of 1mM inD2O. Sampleswerepreparedbydrying aliquots of stock solution under vacuum and reconstitutingthem in 2 μL of 20 mM deuterated Tris buffer solution (pH ∼ 7.4)to initiate aggregation. The final total peptide concentration was0.5 mM for pure hIAPP samples and 1 mM for samples containingequimolar hIAPP and macrocycle. For isotope dilution experi-ments, unlabeled hIAPP is mixed with the labeled peptide in a 3:1ratio with a total peptide concentration of 1 mM. As I26P hIAPPhas significantly slower kinetics than wild-type hIAPP (Fig. 4), thefinal total peptide concentrations were 1 mM for pure I26P hIAPPand 2 mM for samples of equimolar I26P hIAPP and macrocycle.Samples were placed between CaF2 windows separated by a 56-μMTeflon spacer and kept in a purge box under dry air to preventhydrogen exchange from ambient water vapor. The samples re-mained at room temperature (∼22 °C) throughout aggregation.

Two-Dimensional Infrared Measurements. The methods for collect-ing and processing 2D infrared (IR) spectra are described in detailelsewhere (2). Briefly, 800-nm pulses were generated by a home-built Ti:sapphire oscillator and amplified in a neodymium-dopedyttrium lithium fluoride (Nd:YLF)-pumped regenerative ampli-fier. The pulses were sent to a home-built optical parametric am-plifier (OPA) with difference frequency generation to produce60-fs mid-IR pulses. The 6-μm light was split into pump and probebeams and the pump pulse was passed through a Ge acousto-opticmodulator (AOM) pulse shaper (6, 7) before both beams werefocused onto the sample. The signal was detected with a mercurycadmium telluride (MCT) array. For kinetic experiments, therewas a delay of ∼10 min between initiating aggregation and mea-suring the first 2D IR spectrum, after which spectra were collectedcontinuously until the aggregation was complete and the samplehad equilibrated. As hIAPP typically has a lag phase of 45 min to1 h, lag-phase spectra were collected 15 min after aggregation wasinitiated and averaged for 10 min. Steady-state fibril spectra werecollected for either 5 h (for wild-type hIAPP experiments) or 13 h(for I26P hIAPP experiments) and averaged for 10 min.

Transmission Electron Microscopy. Transmission electron micros-copy (TEM) was performed at the University of WisconsinMedical School Electron Microscope Facility. Samples were pre-pared as for spectroscopy experiments, although sample volumeswere scaled up to 40 μL. A 2-μL aliquot was removed to monitoraggregation via 2D IR and 2-μL aliquots were flash frozen at var-ious points during aggregation to allow correlation of 2D IR spectrawith TEM images. Aliquots were thawed immediately before ob-taining TEM images and stained using the negative stain Nano-W(methylamine tungstate) (Nanoprobes) in a two-step staining pro-cedure. The samples were placed on a pioloform-coated 300-meshthin-bar Cu grid and viewed on a Philips CM120 electron micro-scope at 80 kV. Images were documented with an SIS (OlympusSoft Imaging Systems) MegaView III digital camera.

Thioflavin-T Binding Kinetics. We monitored thioflavin-T binding ki-netics as described previously (8), with an excitation wavelength of450 nm and an emission wavelength of 485 nm. Samples were pre-pared under the same conditions as for the 2D IR experiments, butwith 16 mM thioflavin-T added to the buffer solution. A 1.0-cm cu-vettewasusedand the samplewasnot stirred.Data collectionwasnotrepeated, because the fluorescence experiment requires a largeamount of sample at the concentrations used in 2D IR experiments.

Metadynamics Simulations of the Dimerization Free Energy Landscape.To calculate the dimerization free energy of hIAPP, we performedbias-exchange metadynamics (BE-Meta) simulations (9). BE-Metahas been used extensively to study the folding of various proteinsand the nucleation of short amyloid peptides (10–13). BE-Metacombines metadynamics and replica-exchange techniques, usingseveral replicas performing metadynamics in parallel at the sametemperature. Each replica is biased with a time-dependent potentialto help the system escape local free energetic minima and enhancesampling. The bias added is a sum of Gaussian potentials (14),

V iGðξiðxÞ; tÞ=w

Xt′∈½nτG�<t

exp

−ðξiðxÞ− ξi

�x�t′��2

2σ2i

!;

whereV iGðξiðxÞ; tÞ is the total bias applied to replica i at time t, ξi(x) is

a collective variable (CV) used to guide the simulation in replica i andis defined on the system coordinates x,w and σi are the height and thewidth of the Gaussians, and τG is the Gaussian deposition rate.A trial of system conformations between replicas i and j is

attempted every τex and is accepted with probability (9)

Pij =min�1; exp

nhV iGðξiðxiÞ; tÞ+Vj

G

�ξj�xj�; t�

−ViG

�ξi�xj�; t�−Vj

G

�ξjðxiÞ; t

�i.kBT

o�:

Here Pij is the probability of accepting an exchange at time t,ξmðxnÞ is the mth CV value of the replica n, and kB and T are theBoltzmann constant and the temperature, respectively.In this work, we used the following five CVs to enhance the

conformational sampling of the hIAPP dimer:

CV1: The Q parameter for residues 8–16, Qres 8−16, is defined as

Qres  8−16 =Dexp�−�rNMRij − rij

�.9�E

ij;

where rNMRij and rij are the pairwise distances (in angstroms) of the

backbone atoms for residues 8–16 in the reference NMR structure

Buchanan et al. www.pnas.org/cgi/content/short/1314481110 1 of 8

and in a given structure, respectively (15). Parameter Q measureshow a conformation deviates from the reference fibril structurederived from solid-state NMR (16). This CV characterizes theN-terminal β-sheet of the U-shaped dimer.

CV2: Q parameter for residues 27–35, Qres  27−35 : This CV hasthe same definition as CV1 but is applied to residues 27–35.Therefore, it characterizes the C-terminal β-sheet of the U-shaped dimer.

CV3: para− βrmsd for residues 8–16 counts the number of pairsof three-residue segments in the backbone between residues 8and 16 for two monomers that form the ideal parallel β-sheet.para− βrmsd is defined as (17)

para− βrmsd =Xδ

n½RMSD�

with

n½RMSD�= 1− ðRMSD=0:8Þ81− ðRMSD=0:8Þ12;

where RMSD measures the root mean-square deviations, in ang-stroms, of the N, Cα, C, O, and Cβ atom positions for every pairof 3 + 3 residue blocks from those in an ideal parallel β-sheet.Whereas CV1, Qres 8−16, characterizes only the parallel β-sheetfound in the NMR structure, this CV samples all of the possibleparallel β-sheets formed between two monomers at the N terminus.

CV4: para− βrmsd for residues 27–35: This CV has the samedefinition as CV3, but is defined on residues 27–35. Therefore,it characterizes all of the possible parallel β-sheets formed atthe C terminus of two monomers.

CV5: para− βrmsd for residues 20–29 has the same definition asCV3 and CV4, but is defined on residues 20–29, the centralfragment. This CV characterizes the parallel β-sheet formed atthe central fragments and assists in capturing the possibleintermediates during the dimerization process.

Segments of residues 8–16 and 27–35 used in CV1–CV4 werechosen on the basis of the N-terminal and C-terminal parallelβ-sheets of the hIAPP mature fibril identified via solid-stateNMR measurement by Luca et al. (15) and our previous 2D-IRstudy (16, 18). Other than the above five CVs, i.e., five replicas,we also introduced a neutral replica in which no biasing poten-tials were applied. This replica promotes sampling of the un-biased canonical distribution (9). The Gaussian height w used inthis work was 2.0 kJ/mol, the width σ was 0.02 for CV1 and CV2and 0.2 for CV3–CV5, and the deposition time τG was 5 ps.Exchange of configurations was attempted every 50 ps.After a certain amount of time, tfill, most of the free energy

minima of the system in each replica were filled, and VG in each

replica reached a stationary state and fluctuated around an av-erage profile. The free energy F along the CV, ξi, in the ithreplica can be approximated with the time average of Vi

G (19),

FðξiÞ≈ −V iG = −

1tall − tfill

Ztalltfill

dtV iGðξi; tÞ;

where tall is the total simulation time. The filling time tfill in thisstudy was taken to be 200 ns, after monitoring the convergenceof free energy profile in each replica following the protocol byMarinelli et al. (10) and Marinelli and coworkers (20). The totalsimulation time tall for each replica was 400 ns. The free energymap along any selected CV is evaluated by combining the con-formations sampled in all replicas, using the weighted histogramanalysis method (WHAM) (10, 20, 21),

FðξÞ= − kBT ln

0B@ Pk

i niξPk

j exph�

f j −V jGðξÞ

�.kBT

i1CA;

where k denotes the number of replicas, niξ is the number of

times the replica i has the CV value ξ, V jG is the average biasing

potential acting on ξ in the replica j, and f j is a constant de-termined through the WHAM iterative procedure.The U-shaped dimer conformation extracted from the NMR

structure was used as the initial configuration in each replica (15).The C terminus of each peptide was capped with an NH2 group,and a disulfide bridge was formed between the Cys2 and Cys7residues. The amino acid side chains were ionized on the basis oftheir pKa values at pH 7, and the resulting dimer had a netpositive charge of +6. The corresponding chloride ions wereadded to the simulation box to neutralize the system. A total of26,596 water molecules were used to solvate the system in a cubicbox of box length of 94.1 Å. Proteins were modeled using theGROMOS96 53a6 force field with simple point-charge (SPC)water (22–24). This combination has been shown to providegood agreement with experimental IR spectra in our previousstudies of rat IAPP and hIAPP monomers and hIAPP fibrils (18,25–27). Van der Waals and short-range electrostatic interactionswere cut off at 9 Å; long-range electrostatic interactions weretreated with a particle-mesh Ewald sum (28, 29). All bonds wereconstrained at their equilibrium lengths, using the LINCS algo-rithm (30), and a 2-fs integration time step was used for all simu-lations. Simulations were performed in the NVT ensemble. Thetemperature was controlled at 298 K, using the Nose–Hooverthermostat with a relaxation time of 0.2 ps (31–33). BE-Metasimulations were carried out using the GROMACS 4.5 molecularsimulation package and the PLUMED plug-in (20, 34, 35).

1. Marek P, Woys AM, Sutton K, Zanni MT, Raleigh DP (2010) Efficient microwave-assisted synthesis of human islet amyloid polypeptide designed to facilitate thespecific incorporation of labeled amino acids. Org Lett 12(21):4848–4851.

2. Middleton CT, Woys AM, Mukherjee SS, Zanni MT (2010) Residue-specific structuralkinetics of proteins through the union of isotope labeling, mid-IR pulse shaping, andcoherent 2D IR spectroscopy. Methods 52(1):12–22.

3. Abedini A, Raleigh DP (2005) Incorporation of pseudoproline derivatives allows thefacile synthesis of human IAPP, a highly amyloidogenic and aggregation-pronepolypeptide. Org Lett 7(4):693–696.

4. Cheng P-N, Liu C, Zhao M, Eisenberg D, Nowick JS (2012) Amyloid β-sheet mimics thatantagonize protein aggregation and reduce amyloid toxicity. Nat Chem 4(11):927–933.

5. Andrushchenko VV, Vogel HJ, Prenner EJ (2007) Optimization of the hydrochloric acidconcentration used for trifluoroacetate removal from synthetic peptides. J Pept Sci13(1):37–43.

6. Shim S-H, Strasfeld DB, Fulmer EC, Zanni MT (2006) Femtosecond pulse shapingdirectly in the mid-IR using acousto-optic modulation. Opt Lett 31(6):838–840.

7. Shim S-H, Strasfeld DB, Zanni MT (2006) Generation and characterization ofphase and amplitude shaped femtosecond mid-IR pulses. Opt Express 14(26):13120–13130.

8. Cao P, Meng F, Abedini A, Raleigh DP (2010) The ability of rodent islet amyloidpolypeptide to inhibit amyloid formation by human islet amyloid polypeptide hasimportant implications for the mechanism of amyloid formation and the design ofinhibitors. Biochemistry 49(5):872–881.

9. Piana S, Laio A (2007) A bias-exchange approach to protein folding. J Phys Chem B 111(17):4553–4559.

10. Marinelli F, Pietrucci F, Laio A, Piana S (2009) A kinetic model of trp-cage folding frommultiple biased molecular dynamics simulations. PLoS Comput Biol 5(8):e1000452.

11. Cossio P, Marinelli F, Laio A, Pietrucci F (2010) Optimizing the performance of bias-exchange metadynamics: Folding a 48-residue LysM domain using a coarse-grainedmodel. J Phys Chem B 114(9):3259–3265.

12. Baftizadeh F, Biarnes X, Pietrucci F, Affinito F, Laio A (2012) Multidimensional view ofamyloid fibril nucleation in atomistic detail. J Am Chem Soc 134(8):3886–3894.

Buchanan et al. www.pnas.org/cgi/content/short/1314481110 2 of 8

13. Baftizadeh F, Pietrucci F, Biarnés X, Laio A (2013) Nucleation process of a fibrilprecursor in the C-terminal segment of amyloid-β. Phys Rev Lett 110(16):168103.

14. Laio A, Parrinello M (2002) Escaping free-energy minima. Proc Natl Acad Sci USA 99(20):12562–12566.

15. Luca S, Yau W-M, Leapman R, Tycko R (2007) Peptide conformation and supramolecularorganization in amylin fibrils: Constraints from solid-state NMR. Biochemistry 46(47):13505–13522.

16. Takada S, Luthey-Schulten Z (1999) Folding dynamics with nonadditive forces: Asimulation study of a designed helical protein and a random heteropolymer. J ChemPhys 110(23):11616.

17. Pietrucci F, Laio A (2009) A collective variable for the efficient exploration of proteinbeta-sheet structures: Application to SH3 and GB1. J Chem Theory Comput 5(9):2197–2201.

18. Wang L, et al. (2011) 2DIR spectroscopy of human amylin fibrils reflects stable β-sheetstructure. J Am Chem Soc 133(40):16062–16071.

19. Laio A, Gervasio FL (2008) Metadynamics: A method to simulate rare events andreconstruct the free energy in biophysics, chemistry and material science. Rep ProgPhys 71(12):126601.

20. Biarnés X, Pietrucci F, Marinelli F, Laio A (2012) METAGUI. A VMD interface for analyzingmetadynamics andmolecular dynamics simulations. Comput Phys Commun 183(1):203–211.

21. Kumar S, Rosenberg JM, Bouzida D, Swendsen RH, Kollman PA (1992) THE weightedhistogram analysis method for free-energy calculations on biomolecules. I. Themethod. J Comput Chem 13(8):1011–1021.

22. Oostenbrink C, Villa A, Mark AE, van Gunsteren WF (2004) A biomolecular force fieldbased on the free enthalpy of hydration and solvation: The GROMOS force-fieldparameter sets 53A5 and 53A6. J Comput Chem 25(13):1656–1676.

23. Berendsen HJC, Postma JPM, van Gunsteren WF, Hermans J (1981) IntermolecularForces, ed Pullman B (Reidel, Dordrecht, The Netherlands).

24. Mark P, Nilsson L (2001) Structure and dynamics of the TIP3P, SPC, and SPC/E watermodels at 298 K. J Phys Chem A 105(43):9954–9960.

25. Middleton CT, et al. (2012) Two-dimensional infrared spectroscopy reveals thecomplex behaviour of an amyloid fibril inhibitor. Nat Chem 4(5):355–360.

26. Reddy AS, et al. (2010) Solution structures of rat amylin peptide: Simulation, theory,and experiment. Biophys J 98(3):443–451.

27. Reddy AS, et al. (2010) Stable and metastable states of human amylin in solution.Biophys J 99(7):2208–2216.

28. Darden T, York D, Pedersen L (1993) Particle mesh Ewald: An N·log(N) method forEwald sums in large systems. J Chem Phys 98(12):10089–10092.

29. Essmann U, et al. (1995) A smooth particle mesh Ewald method. J Chem Phys 103(19):8577–8593.

30. Hess B, Bekker H, Berendsen HJC, Fraaije JGEM (1997) LINCS: A linear constraint solverfor molecular simulations. J Comput Chem 18(12):1463–1472.

31. Nosé S (1984) A unified formulation of the constant temperature molecular dynamicsmethods. J Chem Phys 81(1):511–519.

32. Nosé S (2002) A molecular dynamics method for simulations in the canonicalensemble. Mol Phys 100(1):191–198.

33. Hoover WG (1985) Canonical dynamics: Equilibrium phase-space distributions. PhysRev A 31(3):1695–1697.

34. Hess B, Kutzner C, van der Spoel D, Lindahl E (2008) GROMACS 4: Algorithms forhighly efficient, load-balanced, and scalable molecular simulation. J Chem TheoryComput 4(3):435–447.

35. Bonomi M, et al. (2009) PLUMED: A portable plugin for free-energy calculations withmolecular dynamics. Comput Phys Commun 180(10):1961–1972.

Fig. S1. Sequence and fibrillation propensity of IAPP. (A) Sequences of IAPP from various species. Residues that differ from the human IAPP sequence arehighlighted in red. Primate and cat IAPP has been shown to form islet amyloid, whereas IAPP from rodents, cows, and dogs does not (1). (B) Propensity profilegraph for human IAPP, obtained from the ZipperDB database of protein fibrillation propensities established by David Eisenberg and coworkers (methodsdescribed in ref. 2). The predictions are based on hexapeptide sequences, as they demonstrated that specific sequence is more indicative of fibrillation pro-pensity than simple residue composition. The histogram bars represent the Rosetta energy of a hexapeptide starting at the indicated position of the proteinsequence. Segments with energies below the indicated threshold of −23 kcal/mol are predicted to form fibrils. As shown in the plot, the FGAIL region andsurrounding residues compose the most amyloidogenic sequence of hIAPP.

1. Cao P, et al. (2013) Islet amyloid: From fundamental biophysics to mechanisms of cytotoxicity. FEBS Lett 587(8):1106–1118.2. Goldschmidt L, Teng PK, Riek R, Eisenberg D (2010) Identifying the amylome, proteins capable of forming amyloid-like fibrils. Proc Natl Acad Sci USA 107(8):3487–3492.

Buchanan et al. www.pnas.org/cgi/content/short/1314481110 3 of 8

Fig. S2. hIAPP amyloid fibril formation monitored by thioflavin-T fluorescence. Fluorescence was measured at 22 °C, in 20 mM Tris, pH 7.4. The observedaggregation time course is similar to that obtained by 2D IR spectroscopy when monitoring the intensity of the unlabeled β-sheet mode at 1,620 cm−1.

Fig. S3. Two-dimensional IR spectra and diagonal slices of isotope-diluted hIAPP. Isotope dilution experiments provide an independent test of β-sheetstructure by eliminating the effects of coupling without changing the structure (1, 2). Unlabeled hIAPP is mixed 3:1 with isotope-labeled hIAPP while dis-aggregated. Spectra and diagonal slices are shown for (A) equilibrated V17 fibrils, (B) lag-phase F23 oligomers, and (C) lag-phase A25 oligomers. Diagonal slices(Fig. 1) of pure isotope-labeled hIAPP are plotted in light gray for comparison. In each case, the diluted label is blue-shifted to around 1,594 cm−1 and issignificantly broadened compared with the undiluted label. The magnitudes of the peak shifts upon dilution are noted in red.

1. Middleton CT, et al. (2012) Two-dimensional infrared spectroscopy reveals the complex behaviour of an amyloid fibril inhibitor. Nat Chem 4(5):355–360.2. Kim YS, Liu L, Axelsen PH, Hochstrasser RM (2008) Two-dimensional infrared spectra of isotopically diluted amyloid fibrils from Abeta40. Proc Natl Acad Sci USA 105(22):7720–7725.

Buchanan et al. www.pnas.org/cgi/content/short/1314481110 4 of 8

Fig. S4. Additional 2D IR spectra and diagonal slices of lag-phase isotope-labeled hIAPP. (A) V17 and (B) G33 display weak and broad isotope-labeled modes,centered around 1,594 cm−1, indicating these residues are uncoupled during the lag phase. (C) F23, (D) G24, (E) A25, and (F) L27 all exhibit coupled isotope-labeledmodes, which are highlighted with a white box in the spectra and a red arrow in the slices. The frequencies of isotope-labeled modes are given in Table S1.

Fig. S5. Two-dimensional IR spectra and diagonal slices of equilibrated isotope-labeled hIAPP fibrils. (A) G24, (B) A25, (C) L27, and (D) G33 all exhibit coupledisotope-labeled modes, which are highlighted with a white box in all spectra and a red arrow in the diagonal slices. The frequencies of the isotope-labeledmodes are given in Table S1.

Buchanan et al. www.pnas.org/cgi/content/short/1314481110 5 of 8

Fig. S6. Kinetic traces for A25 hIAPP and comparison with F23 hIAPP. (A) Intensity traces for the unlabeled β-sheet, 1,585-cm−1 isotope, and 1,570-cm−1 isotopemodes during the aggregation of A25 hIAPP. The traces were normalized to their values after 5 h of aggregation, when the unlabeled β-sheet peak had fullyequilibrated. The 1,570-cm−1 trace matches that of the fibril β-sheets (1,620 cm−1), confirming that the 1,570-cm−1 peak results from the formation of amyloidfibrils. Much like F23, the 1,585 cm−1 of A25 is present at the earliest times measurable, increases during the lag phase, and then begins to decrease. However,unlike F23, this peak does not fully go away, suggesting that the 1,585-cm−1 peak represents both a lag-phase oligomer and a fibril polymorph. To extract thekinetic curve of the oligomeric species, the unlabeled β-sheet curve was subtracted from the 1,585-cm−1 isotope curve. The resulting curve (purple) followsa very similar time course to that of the oligomer species in F23. We note that this type of subtraction was not necessary for F23 as the isotope-labeled modeappears at different frequencies for the oligomer and the fibril. (B) Comparison of the oligomer and fibril kinetic traces for F23 and A25 (extracted from Fig. 1Gand Fig. S5A, respectively). To allow comparison of individual samples, the kinetic traces for each sample were normalized and scaled by their respective t50times, defined as the time to half-maximum of the unlabeled β-sheet mode (1). The oligomer traces show very similar time evolutions, suggesting that thestructural transformation of the oligomer is likely occurring simultaneously across multiple residues.

1. Shim S-H, et al. (2009) Two-dimensional IR spectroscopy and isotope labeling defines the pathway of amyloid formation with residue-specific resolution. Proc Natl Acad Sci USA 106(16):6614–6619.

Fig. S7. TEMs of hIAPP aggregates. (A and B) TEMs of pure hIAPP collected (A) during the lag phase and (B) after equilibration. (C and D) TEMs of hIAPP fibrilsformed in the presence of (C) seeding macrocycle Mac26–32 and (D) inhibiting macrocycle Mac21–27. Fibers are clearly present in the mixed samples, inagreement with 2D IR data, whereas the large spheres are likely deposited macrocycles (1). (Scale bars, 200 nm.)

1. Khakshoor O, et al. (2010) X-ray crystallographic structure of an artificial β-sheet dimer. J Am Chem Soc 132(33):11622–11628.

Buchanan et al. www.pnas.org/cgi/content/short/1314481110 6 of 8

Fig. S8. Sequence and 2D IR data for macrocycles. (A) Sequence of residues R1–R7 of the recognition strand and residues R8–R11 of the blocking strand for thefive macrocycles. (B–F) Two-dimensional IR spectra and diagonal slices of (B) Mac11–17, (C) Mac15–21, (D) Mac21–27, (E) Mac26–32, and (F) Mac31–37. The spectra allshow a large peak centered between 1,636 cm−1 and 1,640 cm−1 due to the antiparallel β-strands of the macrocycle and a higher-frequency shoulder near 1,670cm−1 caused by the β-turns.

Buchanan et al. www.pnas.org/cgi/content/short/1314481110 7 of 8

Fig. S9. Spectra and diagonal slices of I26P with and without macrocycles. (A–C) Shown is 1 mM I26P alone after (A) 20 min, (B) 13 h, and (C) 50 h of ag-gregation. Over the course of aggregation, the unlabeled amide I mode shifts from 1,647 cm−1 to 1,619 cm−1, indicating a structural transition from randomcoil to amyloid β-sheets. (D and E) Spectra of 1 mM I26P mixed with equimolar Mac31–37 obtained after (D) 20 min and (E) 13 h of aggregation confirm thatMac31–37 accelerates amyloid formation. (F and G) Spectra of 1 mM I26P mixed with equimolar Mac21–27 obtained after (F) 20 min and (G) 13 h of aggregationconfirm that Mac21–27 accelerates amyloid formation.

Table S1. Frequencies of the isotope-labeled modes of hIAPP

Label Oligomer frequency, cm−1 Fiber frequency, cm−1

V17 1,595, broad 1,578F23 1,588 1,593, broadG24 1,587 1,585A25 1,585 1,570/1,585L27 1,586 1,578G33 1,593, broad 1,583

Frequencies are given for the isotope-labeled mode during the lag phase(oligomers) and the equilibrated phase (fibrils) for each of the labeled hIAPPsamples.

Buchanan et al. www.pnas.org/cgi/content/short/1314481110 8 of 8