Discriminating early stage Aβ42 monomerstructures using chirality-induced 2DIRspectroscopy in a simulation studyWei Zhuanga,1, Nikolaos G. Sgourakisb, Zhenyu Lic, Angel E. Garciab, and Shaul Mukameld,1

aState Key Lab of Molecular Reaction Dynamics, Dalian Institute of Chemical Physics, Dalian, Liaoning, China, 116023; bCenter for Biotechnology andInterdisciplinary Studies, Rensselaer Polytechnic Institute Troy, NY 12180-3590; cHefei National Laboratory for Physical Science at Microscal, University ofScience and Technology of China, Hefei, Anhui, China, 230026; and dDepartment of Chemistry, University of California, Irvine, CA 92697-2025

Edited* by Robin M. Hochstrasser, University of Pennsylvania, Philadelphia, PA, and approved July 16, 2010 (received for review February 19, 2010)

Elucidating the structural features of the Aβ monomer, the peptideconstituent of amyloid fibrils found in Alzheimer’s disease, canenable a direct characterization of aggregation pathways. Recentstudies support the view that the ensemble of Aβ42 monomers isa mixture of diverse ordered and disordered conformational spe-cies, which can be classified according to the formation of a char-acteristic β-hairpin conformation in a certain region. Despite thedisparity in the structural features of these species, commonly usedspectroscopic techniques such as NMR may not directly trace theconformational dynamics in the ensemble due to the limited timeresolution and the lack of well-resolved spectral features for differ-ent comformers. Herewe use molecular dynamics simulations com-bined with simulations of two-dimensional IR (2DIR) spectra toinvestigate the structure of these species, their interchange ki-netics, and their spectral features. We show that while the discrimi-nation efficiency of the ordinary, nonchiral 2DIR signal is limiteddueto its intrinsic dependence on common order parameters that aredominated by the generally unstructured part of the sequence, sig-nals with carefully designed chirality-sensitive pulse configurationshave the high resolution required for differentiating the variousmonomer structures. Our combined simulation studies indicatethe power of the chirality-induced (CI) 2DIR technique in investigat-ing early events in Aβ42 aggregation and open the possibility fortheir use as a novel experimental tool.

amyloid ∣ REMD ∣ nonlinear spectroscopy

The aggregation of amyloidogenic peptides into insoluble amy-loid fibrils has been identified as the pathological hallmark

of more than 20 diseases including Alzheimer’s disease (AD),which is characterized by the progressive deterioration in cogni-tive function and behavioral symptoms related to alterations inneural and glial function in certain regions of the brain. Amyloidpeptides are products of the proteolytic cleavage of a multido-main integral membrane protein, Amyloid-β Precursor Protein(APP) (1, 2). The cleavage can happen at different transmem-brane sites of APP that result in different lengths of the Aβproduct, which varies from 38 to 43 residues. The 42-residue de-rivative, Aβ42, has a much stronger tendency to form fibrils invitro. It is widely believed that in aqueous solution the Aβ peptideis generally unstructured. Its fibrils, however, have dominating βsheet structure, as shown in two recently reported models of bothAβ40 and 42 (3, 4). Moreover, Aβ peptides may also form keyoligomeric species with neurotoxic properties. Little is knownabout the structure of the oligomeric species and the mechanismof transitions among monomers, oligomers, and fibrils. Under-standing this process is essential for designing therapeutics thattarget amyloid formation at an early stage of the disease (5).Characterization of the structural features of the monomers isthus of particular interest because they may provide a startingpoint for modeling the aggregation pathways. Different monomerconformations can provide the seed for pathologic oligomericspecies as well as intermediates en route to fibril formation.

Several spectroscopic and computational studies have focusedon structural features of the Alzheimer’s peptides Aβ(1–40) andAβ(1–42) and various fragments. (6–19). Recent evidence sup-ports the view that, at the monomeric level, the ensemble ofAβ is a mixture of multiple, nearly isoenergenic conformationalspecies, in agreement with the energy landscape view of proteindynamics proposed by Frauenfelder and coworkers (22). In a re-cent combined molecular dynamics (MD) simulation and NMRstudy, Sgourakis and coworkers generated, through theMD simu-lation, the monomeric forms of Aβ40 and 42 in aqueous solution(20), which nicely reproduced the NMR J-coupling data. Theythen investigated the structural diversity of the structural ensem-ble of Aβ42. This study showed that in a significant population ofthe sampled structures of Aβ42 (28%), the C terminus had a fairlywell defined β-hairpin structure involving short strands at resi-dues 31–34 and 38–41. It was further proposed that this structuralfeature may rigidify the peptide C terminus. This is consistentwith two independent NMR NOE and relaxation studies, whichboth reported a more rigid C terminus for Aβ42 relative to theless fibrilogenic isoform Aβ40 (10, 19). Based on this experimen-tal evidence, Yan and Wang have proposed that this formation ofhairpin structure might reduce the C-terminal flexibility of theAβ42 and be responsible for the higher propensity of this peptideto form amyloids (19). Furthermore, a previously published 3Dprofile study has indicated that this region has the lowest energyin the fibril state (21). Taken together, both experiments and com-putations support the presence of a more structured C terminusfor Aβ42, thus providing a plausible structural basis for the highamyloidogenic potential of this toxic Aβ isoform.

Whether and how the completely disordered conformations ofthe monomer turn into the relatively ordered is of considerableinterest because they represent the earliest step along the aggre-gation pathway (18). In order to explore the possible nucleationpathways, suitable experimental tools will be needed that canprobe the structures with adequate temporal, spectral, and spatialresolution. The conformational changes in peptides of this sizeare expected to start to occur on the submillisecond to millise-cond time scale. While the agreement between the experimental(NMR) and the theoretical (MD) J-coupling data justifies thecorrectness of the MD-generated structural ensemble, thusstrongly supporting the existence of multiple monomeric forms,an experimental tool with direct spectral features distinguishingthese forms will still be needed for monitoring the variation ofdifferent components. Also, an ideal technique should possessa higher temporal resolution to help generate a more completepicture of the early aggregation events.

Author contributions: W.Z., A.E.G., and S.M. designed research; W.Z., N.G.S., and Z.L.performed research; W.Z. and N.G.S. analyzed data; and W.Z. and N.G.S. wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor.1To whom correspondence may be addressed. E-mail: wzhuang@dicp.ac.cn or smukamel@uci.edu.

www.pnas.org/cgi/doi/10.1073/pnas.1002131107 PNAS ∣ September 7, 2010 ∣ vol. 107 ∣ no. 36 ∣ 15687–15692

CHEM

ISTR

YBIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Apr

il 25

, 202

0

Two-dimensional IR (2DIR) spectroscopy with 20–100 fs laserpulses is a potentially useful tool to investigate ultrafast kineticsin real time (30–32). In 2D optical experiments, three ultrashortlaser pulses interact with the molecular system to generate acoherent nonlinear signal, which depends parametrically on threeconsecutive time delays t1, t2, and t3. Two-dimensional correlationplots are obtained by a double Fourier transform of the signalwith respect to t1 and t3, holding t2 fixed (33). By spreading thecongested transitions of linear absorption in two dimensions,these techniques significantly enhance the spectral resolution.The cross peaks (off diagonal peaks in 2D frequency correlationplots) carry signatures of intermode couplings, and their spectralline shapes probe solvent and conformational fluctuations. Opti-cal 2D techniques have been successfully used toward the study offast peptide folding kinetics (32, 34), hydrogen bonding dynamics(35), ultrafast chemical exchange (30, 31), and energy transport inphotosynthetic antennae (36). Very recently, there has been con-siderable interest in the 2DIR field in studying the structure andformation kinetics of various kinds of amyloid aggregates (12–15).

In this paper, we have used enhanced-sampling MD simulationprotocols described in ref. 20 and MD simulations to investigatethe interchange time scale between the ordered and disorderedstructures. Due to the similar highly unstructured nature of thetwo conformers, common order parameters such as backbonedihedral angles can not discriminate between them, thus render-ing traditional 2DIR techniques insensitive to their structuraldifferences. We thus propose to improve 2DIR’s differentiationbetween the two conformers by making use of chirality. We de-monstrate that the two conformer ensembles have very differentstructural chirality characteristics, and the chirality-induced (CI)2DIR signals (37, 38) provide direct and clean spectral features todistinguish between them and thus have the capacity to monitorthe structural variation during the early aggregation.

MD Simulations of Interchange Kinetics Between the TwoConformersUsing clustering analysis, we have previously identified distincttypes of conformational species in an ensemble of Aβ42 thatwas generated by means of replica-exchange MD (REMD) andfurther validated by three bond J-coupling data from NMR(20). The ensemble of conformations can be adequately classifiedinto two populations based on patterns of secondary structuralelements: The first population (A) contains conformations thatdisplay a characteristic β5-hairpin motif toward the C terminusof the sequence, while the second (B) is devoid of any elementsof regular secondary structure other than transient turns. Repre-sentative structures of these two groups are shown inFig. 1 I and II.In the present study, we have sampled fluctuations starting fromrepresentative conformations of the two conformational speciesusing 10 ns trajectories of NPT molecular dynamics at room tem-perature (300 K). This resulted in the generation of two structuralensembles of 10,000 conformations each, which were used as astarting point for the calculation of the IR spectra.

We are guided by the original REMD trajectories to obtainan estimate for the interconversion time scale between the twoensemble populations. The distribution of atomistic root meansquare deviations from the representative conformation withthe β-hairpin motif of Fig. 1I, which was used to assign the room-temperature REMD ensemble into the two populations, is shownin Fig. 1III. We obtain a bimodal distribution between the orderedand the disordered species. According to this analysis, a rmsdthreshold of 6 Å is sufficient in discriminating between the twopopulations and was further used to identify exchange events.The population ratio of the two species is (A∶B) 28∶72. Analysisof interconversions between the two populations along room-temperature segments of the REMD simulation trajectoriesshows that the exchange between the two conformers occurs atthe subnanosecond time scale. This is shown in Fig. 1IV. We find

that approximately 60% of the transitions occur within 20 ps andmore than 80% in less than 100 ps. In the next section, we comparethe ordinary (nonchiral) 2DIR signals of the two conformers.

Similarity of the Nonchiral 2DIR Features for the TwoConformers and Its Structural BasisThe amide-I band of xxxx photon echo signal (imaginary part) ofboth conformers in the Amide I range are shown in Fig. 2 to-gether with their absorption spectra. Here all optical pulsesare polarized along X direction. This is a nonchiral technique.The 2D signals do not reveal additional information comparedwith the absorption because the cross peaks are covered bythe broad diagonal peaks. For the unlabeled samples, both con-formers have a single peak. The maximum of A’s peak (black inthe absorption signal) is about 20 cm−1 red-shifted comparedwith B’s (red in the absorption signal) and slightly (10 cm−1)broader. It also has a slightly asymmetric feature with a maximumat approximately 1;630 cm−1, consistent with the typical featureof β-hairpin in the literature. B has an almost perfect gaussianshape. The A and B signals are very similar, as demonstratedby the ensemble-averaged absorption signal of 28%A þ 72%B.For samples with residues 31–34 and 38–41, the hairpin regionin conformer A, labeled with 13C18O, we observed an isotopeband at approximately 1;575 cm−1 for A and approximately1;595 cm−1 for B. These two peaks thus have the same splittingas in the unlabeled samples. At the same time, the main “randomcoil” band of A becomes symmetric and gaussian-like as B, andits shift from the main band of B is approximately 5 cm−1. Thissuggests that A and B have very similar nonchiral vibrational sig-nal features, especially in the random coil segments. Fig. 3 Uppershows the decomposition of the various normal modes in the un-labeled sample of A (NMD) into the two structural motifs β-hair-

Fig. 1. I and II) Representative conformations of the two ensemble popula-tions of Aβ (1–42) used to model the locally structured “A” and disordered“B” states, as extracted from the replica exchange molecular dynamics(REMD) trajectories. Atoms of residues that participate in the C-terminalβ-hairpin are explicitly shown. The first and last residues of each conforma-tion are numbered for clarity. Graphics were generated by mean of thevisualization suite pymol (www.pymol.org). (III) Distribution of atomistic(Cα) rmsd from the representative conformation of the B (locally ordered)state, as calculated for all REMD conformations sampled at low temperature.All conformations were sampled over a range of temperatures from 276 K to305 K. A bimodal distribution with a gap at 6 Å indicates the existence oftwo populations with distinct structural characteristics. (IV) Distribution oftransition times (first passage time) between the A (locally ordered) and B(disordered) states, as observed in the low-temperature segments of theREMD trajectories. All events happen over a range of temperatures from276 K to 305 K. The majority (>80%) of events take place in the sub-100 ps time scale, which makes our dataset suitable for analysis with the2DIR simulation method.

15688 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1002131107 Zhuang et al.

Dow

nloa

ded

by g

uest

on

Apr

il 25

, 202

0

pin (red) and coil (green). Modes 1;630 cm−1 have the most sig-nificant β strand content, which causes the characteristic β strandfeature peak in the absorption signal, labeled with a red arrow inFig. 2III, while the shoulder labeled with a green arrow mainlycontains the coil feature. Fig. 3 Lower shows the same analysisfor the labeled sample, in which the β-strand components areshifted to below 1;630 cm−1, which gave the peak at approxi-mately 1;575 cm−1 in Fig. 3VI, labeled with a red arrow. Nowin the major band, the peak labeled with a green arrow becomesdominant. This explains why an apparent significant red shift

(15 cm−1) in the absorption signal is caused by a labeling a rela-tively small portion of the structure ( only 20% of the structure ishairpin).

Introducing Chirality into the 2D PictureStructural chirality allows for novel spectroscopic probes. Typicalmolecules’ response to the optical field is adequately described bythe dipole approximation because they are smaller than the op-tical wavelength. Here the field is uniform across the molecule.Taking the spatial variation of the field into account yields newcontributions to the response stemming from the variation of thephase of the optical field at different points within the molecule.These contributions are typically 100 to 1,000 times weakerthan the leading dipole contributions. (This is due to the ratioof protein size and the optical wavelength). However, by choosingcertain polarization configurations of the laser fields, the highermultipole corrections generate distinct background-free signals.These signals are finite only in chiral systems and vanish in race-mates and nonchiral molecules.

Circular dichroism (CD) is an important example of such sig-nals in linear spectroscopy (39–41), broadly applied for probingthe folding states of proteins and their conformational stability.We recently theoretically proposed a chirality-induced 2DIR(CI-2DIR) technique (37), in which the structural sensitivitycan be greatly enhanced by a judicious choice of polarization con-figurations aimed at chiral structures. These CI techniques thusconstitute a natural extension of CD to nonlinear spectroscopy.We next apply this technique to calculate the CI-2DIR spectra ofthe A and B conformational species of Aβ42.

In Fig. 4, the simulated 2D CI (XXXY) signals (I, II) as well asthe (linear) circular dichroism signal (III) of both conformers arepresented. The signals of the two conformers show significantdifferences: Along the diagonal line, A’s CI-2DIR signal has astrong pair of (positive plus negative) lobes around 1;630 cm−1

and a weak peak around 1;670 cm−1 regime, while B has a strongpair of lobes around 1;670 cm−1 and a weak pair around1;630 cm−1. Similar to the CD signals (III), these different fea-tures originate from the system structural chirality and can thusdifferentiate between the two conformer ensembles. However,these partially overlapping diagonal features do not offer a cleanspectral window for our observation of fast exchange kinetics.

2D techniques significantly enhance the spectral resolutionby spreading the congested transitions of linear absorption intwo dimensions. The cross peaks carry signatures of intermodecouplings, and their spectral line shapes probe solvent and con-

Fig. 2. (Column A) 2D photon echo signal of conformer A (imaginary part)with XXXX polarizations configuration. The upper panel shows the signal forthe unlabled samples, while the lower panel is for samples with residues31–34 and 38–41 isotopically labeled by 13C18O. (Column B) Same signalsfor conformer B. (Right Column) The upper panel gives the absorption signalof the unlabeled samples. A (black), B (Red), and Aþ B (green). The signal ofA is red-shifted compared with B. The lower panel gives the absorption signalof samples with both 31–34 and 38–41 labeled by 13C18O. There is a ∼20 cm−1

shift between the isotope peak of A and B.

Fig. 3. The NMD diagram for conformer A. The upper panel shows the NMDfor the unlabeled samples, while the lower panel is for samples with residues31–34 and 38–41 isotopically labeled by 13C18O. The β-strand is shown in redand coil in green, respectively.

Fig. 4. (I and II) 2D photon echo signal of A and B with XXXY polarization.(III) Circular dichroism of A (black) and B (red). (IV and V) Similar 2D signalsas I and II, with both 31–34 and 38–41 labeled by 13C18O.

Zhuang et al. PNAS ∣ September 7, 2010 ∣ vol. 107 ∣ no. 36 ∣ 15689

CHEM

ISTR

YBIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Apr

il 25

, 202

0

formational fluctuations. One feature of particular interest ispeak Φ at (−1;610 cm−1, 1;640 cm−1) (marked by a red arrow),which is absent in the B conformer. Fig. 4 V–VI depicts the simu-lated CI-2DIR signals of both conformers with the residues 31–34and 38-–41 (hairpin region in A) labeled (with 13C18O). The Φpeak remains at the same location.

These results demonstrate the resolving power of CI-2DIRon the Aβ monomer system. To analyze these signals, we definethe following function ηðω1;ω3Þ of the two conformer ensemblesA and B:

ηðω1;ω3Þ ¼

8><>:

−jAðω1;ω3Þj Bðω1 ;ω3ÞAðω1;ω3Þ < 5%; Aðω1;ω3Þ

MAX > 10%

jBðω1;ω3Þj Aðω1;ω3ÞBðω1 ;ω3Þ < 5%; Bðω1;ω3Þ

MAX > 10%

0 else

[1]

Here Aðω1;ω3Þ and Bðω1;ω3Þ are the 2D signals of A and Bat the ðω1;ω3Þ point, and MAX is the maximum of the absolutevalue of signals. Negative (positive) values of ηðω1;ω3Þ thusrepresent a significant signal with dominating contribution fromA (B) ensembles. Blue regions Fig. 5II have dominant contribu-tion from A. The diagonal features, similar to the CD signals, giveclearly different signals for the two conformers. However, thefeatures are highly congested in the 1;600 cm−1 ∼ 1;700 cm−1 fre-quency range. All of the diagonal features partially overlap. Thismakes it difficult to find an adequate window to monitor the in-terchanging kinetics between the conformers and the follow-upoligomerization. In contrast, the cross peak Φ, as denoted bythe red arrow, is the well-resolved and separated peak that hasa dominant A contribution. This gives a good window for mon-itoring the interchanging kinetics of the Aβ monomers.

Furthermore, the similar gaussian shape of A and B conven-tional 2DIR spectra originates from the similar randomly coiledstructure in both conformers. As suggested by the combinedMD/NMR study (20), the two main conformers are generallyunstructured and may not be distinguished by commonly usedorder parameters. This is contrasted with the case of a stable,folded protein in which the RMSD to the native state can be usedquite efficiently to describe the unfolding/folding equilibrium.Fig. 6 I and II depicts the Ramachandran plots of the two con-formers. The similarity of two plots indicates that the differencesin the structural features of the two monomers cannot be attrib-uted to local features of the backbone, implying that featuresother than secondary structure alone may be crucial in shapingthe energy landscape of this dynamic peptide system. We furtherdemonstrate this trend in the Ramachandran plots of the randomcoil part of the sequence (residues 1–30 in Fig. 6 III and IV) thatcorresponds to a region with no observed elements of cooperativesecondary structure for both of the conformational subensem-bles, A and B. We observe two very similar patterns, which indi-cates that even in the presence of a systematic conformationaldifference between the random coil part of the two conformers,routine order parameters such as backbone dihedral angles willnot be able to differentiate it, as illustrated in the Fig. 6.

For a system such as Aβ that consists of a multitude of diverseconformational species, the absence of a typical order parameterlimits the characterization of the transitions between unstruc-tured conformations and conformations with some elements ofsecondary structure that could serve as the nucleation sites forthe assembly of larger oligomers, en route to amyloid formation.On the other hand, the CI-2DIR signals are different for the en-tire structure of the two Aβ conformers due to their differences instructural chirality. In the signals with the hairpin region (in A)labeled, we see that the Φ peak remains at the same location. Φis therefore related to the chirality of the entire random coilsegment and reflects the intrinsic, global structural differencebetween A and B. To quantify this difference in chirality, we de-fine a chirality factor (CF):

CFðm;nÞ ¼ jΨ emΨenRmn · ðμm × μnÞj [2]

Here,m, n are the indices of local amide I modes,Ψem is the eigenstate e wavefunction given as a superposition of localized amideI states.Rmn is the displacement vector between the locations oftwo transition dipoles μm and μn. Sum over all the eigenstates atone frequency for SðeÞ ¼ ∑m;nΨemΨenRmn · ðμm × μnÞ will givethe circular dichroism signal (neglecting the magnetic dipole con-tribution). Thus the chirality factor reveals the contribution of aspecific pair of chromphoresm and n to the chirality-induced CDsignal. Fig. 5I shows the CFðm;nÞ for A (upper triangle) and B(lower triangle). The spectral regime between m ¼ ð31;41Þ andn ¼ ð31;41Þ shows significant differences between A and B. Thisis attributed to the difference of structural chirality betweenthe hairpin-like (in A) and randomly coiled (in B) structures.Interestingly, the CF of the unstructured part of the sequence(residues 1–30) also shows clear differences. This implies thatstructural features that do not manifest themselves as elementsof regular secondary structure, such as clusters of locally interact-ing atoms, could play an important role in shaping the energylandscape of the system. Also, the random coil parts of thetwo conformers, which are indistinguishable with ordinary orderparameters, show very different chirality-related structural prop-

Fig. 5. (I) The plot of chirality factor of A and B, as defined in Eq. 2. (II) 2Dplot of ηðω1;ω3Þ, as defined in Eq. 1. Negative values are colored blue, positiveare colored red.

Fig. 6. (I and II) The Ramachandran plots of two conformers. The basins thatcorrespond to extended, polyproline type 2 and helical conformations can beobserved. Contours are based on a log scale of counts from the simulationtrajectories. Glycine residues at positions 9, 25, and 29 have been omittedfrom this analysis. (III and IV) Same Ramachandran plots of the unlabeled(“random coil” ) part of the sequence (residues 1–30).

15690 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1002131107 Zhuang et al.

Dow

nloa

ded

by g

uest

on

Apr

il 25

, 202

0

erties. The chirality factor carries information about the distribu-tion of the localized amide I transition dipoles and shows a dif-ferent pattern for the two conformers. Thus it can be consideredas a chirality-related spectroscopic order parameter that monitorsthe structural transition between the two conformers.

DiscussionMolecular dynamics simulations validated by NMR experimentaldata suggested that the Aβ42 monomer system can be describedas a diverse ensemble of conformations with various degrees ofordered structure (20). The present analysis based on the simula-tion suggests a sub-100 ps interchange time scale between theordered and disordered conformations of the Aβ42 peptide.We propose a previously undescribed optical spectroscopic meth-od, chirality-induced 2DIR spectroscopy (CI-2DIR), to character-ize the ensemble of Aβ42 and investigate its thermodynamic andkinetic properties. In addition to the spectral resolution requiredto discriminate between the two major conformational species,CI-2DIR offers ps time scale temporal resolution and can thusmonitor the transition from disordered to ordered conformations.

Vibrational spectroscopies offers unique advatages in the struc-tural characterization of peptides and proteins. Vibrational tran-sitions are sensitive to the peptide local structure and are ideal fordistinguishing between various secondary structural motifs andmonitoring the effects of changing environments through hydro-gen bonding and dielectric effects. The 1;600–1;700 cm−1 amid Iband that originates from the stretchingmotion of the C ¼ Opep-tide bond (coupled to in-phase N-H bending and C-H stretching)is particularly suitable for this purpose because it has a strong tran-sition dipole moment and is spectrally well-separated from othervibrational modes. The up to 20 cm−1 variation of the amid I fre-quency with secondary structure and conformation is widely usedas a marker in polypeptide and protein structure determination.

Compared with the linear IR signals, 2DIR significantly en-hances the spectral resolution by spreading the congested transi-tions of linear absorption in two dimensions. The cross peakscarry signatures of intermode couplings, and their spectral lineshapes probe solvent and conformational fluctuations. However,for larger peptides and proteins 2DIR still suffers from an insuf-ficient spectral resolution. The cross peaks are congested andoverlap with the strong diagonal features. The number of inde-pendent structural unknowns is much larger than the number ofindependent measured quantities.

There are efforts to extend chiral-sensitive techniques intononlinear regimes (45). Chirality-related techniques such asCI-2DIR show much richer features than their chirality-insensi-tive counterparts and greatly improve the spectral resolution. Forextended systems such as Aβ, we estimate the chirality-related tothe global (e.g., helical) arrangement of the residues to be thedominating effect and neglect the local chiral characteristics ofthe individual residues in the following argument: The nonchiralnonlinear response to three pulses depends on the orientationallyaveraged product of four dipoles hμν4mμν3n μν2k μν1l i whereas its CIcomponent depends on products of the form hrν5mnμ

ν4mμ

ν3n μ

ν2k μ

ν1l i.

Here h⋯i denotes orientational averaging of molecules with re-spect to the lab frame. μm is the mth cartesian component (ν ¼ x,y, z) of transition dipole of the mth chromophore, and rνmn is thedistance between themth and nth chromophores. Nonchiral tech-niques depend on the structure only implicitly through its effecton the frequencies and transition dipoles that affect peak posi-tions and intensities. The explicit coordinate (rmn) dependenceof the chiral response amplifies the cross peaks, thus resultingin a higher sensitivity to fine details of the structure.

The indistinguishable distribution patterns of the ordinary(chirality-insensitive) order parameters imply that the normalnonchiral signals of the two conformer ensembles might be intrin-sically indistinguishable regardless of the spectral resolution. Onthe other hand, there exist significant chirality-related differencesbetween the conformer structures, demonstrated by the chiralityfactor discussed above. CI-2DIR spectroscopy shows a high dis-crimination power for the Aβ monomer system. We identifieda well-resolved cross peak Φ associated with the dominatingconformer A contribution. This feature can thus be used as awindow for investigating the monomer conformational kinetics.This could reveal useful information about the seeding, oligomer-ization, and fibrilization processes. CI-2DIR has strong potentialin investigating early Aβ misfolding events. A systematic charac-terization of the Aβ conformer moieties based on a statisticalanalysis of all interactions in the low-temperature ensemble willbe an interesting topic for future research.

MethodsReplica-exchange MD simulations were performed using a variety of forcefields and validated by NMR data as described in ref. 20. This study has shownthat, at the limit of extensive sampling, the ensemble derived using the OPLS/AA force field (23) is consistent with the three-bond JHNHα data. We have usedthe previously described MD trajectories to extract representative conforma-tions of the “ordered” and “disordered” populations, identified using theclustering algorithms described in ref. 20 as well as to investigate theexchange time scale between the two species Fig. 1III. To perform this task,we have first ranked the transitions according to the time difference be-tween their observation in the REMD trajectory segments. Events occurringduring temperature exchanges larger than 30 K have been omitted from thisanalysis because they correspond to much lower probability transitions atphysiological temperature. (The average temperature spacing betweenthe eight replicas sampling at low T was 3.6 K.) In addition to the existingsimulation dataset, we performed constant-temperature (NPT) simulationsin explicit solvent to adequately sample fluctuations within the free energybasin of each representative conformation. The OPLS/AA force field (23) andTIP3P water model (24) were used in these calculations. To build thesesystems, we started from the protein coordinates extracted from the REMDsimulations, which we solvated by adding 3,670 water molecules. The totalnumber of solvent molecules was equal for both peptide conformations. Thesolvated systems were equilibrated at constant temperature (300 K) andpressure (1 atm) for 100 ps, using an integration time step of 1 fs. Finally,molecular dynamics simulations were performed starting from the equili-brated systems for 10 ns each. At this stage, the implementation of the LINCS(25) and SETTLE (26) algorithms to restrain high-frequency bond vibrationsallowed the use of a larer integration timestep (2 fs). All simulations wereperformed using the GROMACS simulation package (28). The Nose–Hoover(27) method was used to couple the system to a heat bath, while the methodby Berendsen was used to control pressure (29).

The simulation protocol of the 2DIR spectra is described in (42), thefluctuating exciton hamiltonian for these snapshots uses an electrostaticmap (33, 42), which had been proved to be able to reproduce 2DIR signalsfor peptide amide I band quantitatively comparable to the experimental data(37, 42–44). The truncated NEE technique with the truncation coefficient η ¼0.3 (42) was used to simulate the signal. All transitions have a Lorentzian line-shape with 4.5 cm−1 (43) full-width at half maximum. 13C18O isotope labelingwas simulated by 65 cm−1 red-shift of the local model frequency (43).

The NMD was calculated by summing the squares of its expansion coeffi-cients for all of the local modes belonging to a given structural motif. Thecomponents were then binned for all eigenmodes in the frequency window(�5 cm−1) to obtain the NMD figures.

ACKNOWLEDGMENTS. W.Z. gratefully thanks the support of the DICP 100Talents Support Grant of 2010 and 2010 QingNian Grant of NSFC. S.M. grate-fully acknowledges the support of National Institutes of Health GrantGM59230 and National Science Foundation Grant CHE-0745892. Z.L. is sup-ported by the Key Program (20933006) of NSFC. A.E.G. acknowledges supportfrom National Science Foundation Grants MCB0543769 and DMR0117792.

1. Kang J, et al. (1987) The precursor of Alzheimer’s disease amyloid A4 protein resem-bles a cell-surface receptor. Nature 325(6106):733–6.

2. Hardy J, Selkoe DJ (2002) Medicine—The amyloid hypothesis of Alzheimer’sdisease: Progress and problems on the road to therapeutics. Science 297(5580):353–6.

3. Petkova AT, et al. (2002) A structural model for Alzheimer’s beta-amyloid fibrilsbased on experimental constraints from solid state NMR. Proc Natl Acad Sci USA99(26):16742–7.

4. Luhrs T, et al. (2005) 3D structure of Alzheimer’s amyloid-beta(1–42) fibrils. Proc NatlAcad Sci USA 102(48):17342–7.

Zhuang et al. PNAS ∣ September 7, 2010 ∣ vol. 107 ∣ no. 36 ∣ 15691

CHEM

ISTR

YBIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Apr

il 25

, 202

0

5. Soto C (2003) Unfolding the role of protein misfolding in neurodegenerative diseases.Nat Rev Neurosci 4(1):49–60.

6. Zhang S, et al. (2000) The Alzheimer’s peptide a beta adopts a collapsed coil structurein water. J Struct Biol 130(2–3):130–41.

7. Baumketner A, et al. (2006) Amyloid beta-protein monomer structure: A computa-tional and experimental study. Protein Sci 15:420–8.

8. Massi F, Peng JW, Lee JP, Straub JE (2001) Simulation study of the structure anddynamics of the Alzheimer’s amyloid peptide congener in solution. Biophys J80(1):31–44.

9. Hou L, et al. (2004) Solution NMR studies of the Abeta(1–40) and Abeta(1–42) peptidesestablish that the met35 oxidation state affects the mechanism of amyloid formation.J Am Chem Soc 126(7):1992–2005.

10. Riek R, Guntert P, Dobeli H,Wipf B,Wuthrich K (2001) NMR studies in aqueous solutionfail to identify significant conformational differences between the monomeric formsof two alzheimer peptides with widely different plaque-competence, a beta(1–40)(ox)and a beta(1–42)(ox). Eur J Biochem 268(22):5930–6.

11. Barrow CJ, Yasuda A, Kenny PT, Zagorski MG (1992) Solution conformations andaggregational properties of synthetic amyloid beta-peptides of Alzheimer’s disease.Analysis of circular dichroism spectra. J Mol Biol 225(4):1075–93.

12. Shim SH, 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.

13. Kim YS, Liu L, Axelsen PH, Hochstrasser RM (2008) Two-dimensional infrared spectraof isotopically diluted amyloid fibrils from A beta 40. Proc Natl Acad Sci USA 105(22):7720–7725.

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

15. Zhuang W, Abrarnavicius D, Voronine DV, Mukarmel S (2007) Simulation of two-dimensional infrared spectroscopy of amyloid fibrils. Proc Nat Acad Sci USA 104(36):14233–14236.

16. Danielsson J, Jarvet J, Damberg P, Graslund A (2005) The alzheimer beta-peptideshows temperature-dependent transitions between left-handed 3-helix, beta-strandand random coil secondary structures. Febs J 272(15):3938–49.

17. Schweitzer-Stenner R, Measey T, Hagarman A, Eker F, Griebenow K (2006) Salmon cal-citonin and amyloid beta: Two peptides with amyloidogenic capacity adopt differentconformational manifolds in their unfolded states. Biochemistry 45(9):2810–9.

18. Bitan G, et al. (2003) Amyloid beta -protein (abeta) assembly: Abeta 40 and abeta 42oligomerize through distinct pathways. Proc Natl Acad Sci USA 100(1):330–5.

19. Yan Y, Wang C (2006) Abeta42 is more rigid than Abeta40 at the c terminus: Implica-tions for abeta aggregation and toxicity. J Mol Biol 364(5):853–862.

20. Sgourakis NG, Yan YL, McCallum SA, Wang CY, Garcia AE (2007) The Alzheimer’speptides Aβ40 and 42 adopt distinct conformations in water: A combined MD/NMRstudy. J Mol Biol 368:1448–1457.

21. Thompson MJ, et al. (2006) The 3D profile method for identifying fibril-formingsegments of proteins. Proc Natl Acad Sci USA 103(11):4074–8.

22. Frauenfelder H, Sligar SG, Wolynes PG (1991) The energy landscape and motions ofproteins. Science 254:1598–1603.

23. Kaminski GA, Friesner RA, Tirado-Rives J, Jorgensen WL (2001) Evaluation and repar-ametrization of the OPLS-AA force field for proteins via comparison with accuratequantum chemical calculations on peptides. Phys Chem B 105(28):6474–6487.

24. Jorgensen WL, Chandrasekhar J, Madura JD, Impey RW, Klein ML (1983) Comparisonof simple potential functions for simulating liquid water. J Chem Phys 79(2):926–935.

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

26. Miyamoto S, Kollman PA (1992) SETTLE—An analytical version of the SHAKE andRATTLE algorithm for rigid water models. J Comput Chem 13(8):952–962.

27. Nose S (1991) Constant temperature molecular-dynamics methods. Prog Theor PhysSupp 103:1–46.

28. Van der Spoel D, et al. (2005) Gromacs: Fast, flexible, and free. J Comput Chem26(16):1701–1718.

29. Berendsen HJC, Postma HJC, van Gunsteren JPM, Dinola A, Haak JR (1984) Molecular-dynamics with coupling to an external bath. J Chem Phys 81(8):3684–90.

30. Zheng JR, Kwak KW, Xie J, Fayer MD (2006) Ultrafast carbon–carbon single-bondrotational isomerization in room-temperature solution. Science 313:1951–1955.

31. Kim YS, Hochstrasser RM (2005) Chemical exchange 2D IR of hydrogen-bond makingand breaking. Proc Natl Acad Sci USA 102:11185–11190.

32. Kolano C, Helbing J, Kozinski M, Sander W, Hamm P (2006) Watching hydrogen-bonddynamics in a beta-turn by transient two-dimensional infrared spectroscopy. Nature444:469–472.

33. Zhuang W, Hayashi T, Mukamel S (2009) Coherent multidimensional vibrationalspectroscopy of biomolecules: Concepts, simulations, and challenges. Angew ChemInt Edit 48:3750–3781.

34. Ganim Z, et al. (2008) Amide I two-dimensional infrared spectroscopy of proteins.Accounts Chem Res 41:432–441.

35. Cowan ML, et al. (2005) Ultrafast memory loss and energy redistribution in the hydro-gen bond network of liquid H2O. Nature 434:199–202.

36. Brixner T, et al. (2005) Two-dimensional spectroscopy of electronic couplings in photo-synthesis. Nature 434:625–628.

37. Zhuang W, Abramavicius D, Mukamel S (2006) Two-dimensional vibrational opticalprobes for peptide fast folding investigation. Proc Natl Acad Sci USA 103:18934–18938.

38. Abramavicius D, ZhuangW,Mukamel S (2006) Probingmolecular chirality via excitonicnonlinear response. J Phys B-At Mol Opt 39:5051–5066.

39. Berova R, Woody RW, Nakanishi K (2000) Circular Dichroism. Principles and Applica-tions (Wiley, New York), 2nd Ed.

40. Nafie LA (1997) Infrared and Raman vibrational optical activity: Theoretical andexperimental aspects. Annu Rev Phys Chem 48:357–386.

41. Barron L (2004)Molecular Light Scattering and Optical Activity (Cambridge Univ Press,London), 2nd edition.

42. Zhuang W, Abramavicius D, Hayashi T, Mukamel S (2006) Simulation protocols for co-herent femtosecond vibrational spectra of peptides. J Phys Chem B 108:18034–18045.

43. Wang JP, Zhuang W, Mukamel S, Hochstrasser R (2008) Two-dimensional infraredspectroscopy as a probe of the solvent electrostatic field for a twelve residue peptide.J Phys Chem B 112:5930–5937.

44. Sengupta N, et al. (2009) Sensitivity of 2D IR spectra to peptide helicity: Aconcerted experimental and simulation study of an octapeptide. J Phys Chem B113:12037–12049.

45. Feng W, Xu Y, Guo Y, Liu S, Wang H (2009) Quantitative surface chirality detectionwith sum frequency generation vibrational spectroscopy: Twin polarization angleapproach. Chin J Chem Phys 22(6):592–600.

15692 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1002131107 Zhuang et al.

Dow

nloa

ded

by g

uest

on

Apr

il 25

, 202

0