Microscopic insights into the protein-stabilizing effectof trimethylamine N-oxide (TMAO)Jianqiang Maa,b, Ileana M. Pazosa,b, and Feng Gaia,b,1

aDepartment of Chemistry, University of Pennsylvania, Philadelphia, PA 19104; and bUltrafast Optical Processes Laboratory, Philadelphia, PA 19104

Edited by William A. Eaton, National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda, MD, and approvedApril 24, 2014 (received for review February 20, 2014)

Although it is widely known that trimethylamine N-oxide (TMAO),an osmolyte used by nature, stabilizes the folded state of proteins,the underlying mechanism of action is not entirely understood. Togain further insight into this important biological phenomenon,we use the C≡N stretching vibration of an unnatural amino acid,p-cyano-phenylalanine, to directly probe how TMAO affects thehydration and conformational dynamics of a model peptide and asmall protein. By assessing how the lineshape and spectral diffu-sion properties of this vibration change with cosolvent conditions,we are able to show that TMAO achieves its protein-stabilizingability through the combination of (at least) two mechanisms: (i)It decreases the hydrogen bonding ability of water and hence thestability of the unfolded state, and (ii) it acts as a molecular crowder,as suggested by a recent computational study, that can increase thestability of the folded state via the excluded volume effect.

infrared | crowding | 2D IR | linear response function

Nature employs a variety of small organic molecules to copewith osmotic stress. Trimethylamine N-oxide (TMAO) is

one such naturally occurring osmolyte that protects intracellularcomponents against disruptive stress conditions (1). In particu-lar, previous studies have shown that TMAO is able to enhanceprotein stability and to counteract the denaturing effect of urea(2, 3). TMAO (Fig. 1, Inset) adopts a skewed tetrahedral struc-ture with a charged oxygen capable of accepting hydrogen bonds(H bonds) and three hydrophobic (methyl) groups. This am-phiphilic structural arrangement makes TMAO a rather specialcosolvent, because it can form H bonds with water, self-associatein a manner similar to surfactants, and show preferential inter-actions with or exclusion (4–12) from certain protein functionalgroups (13–23). Indeed, these molecular properties of TMAOhave been used, either individually or in combination, to ratio-nalize its biological activities. For example, a prevailing viewabout TMAO is that its osmotic and protective role is causedby the molecule’s tendency to be preferentially depleted fromprotein surfaces, as suggested by physicochemical measurements(24–28). This thermodynamic picture, as described by Bolen andcoworkers (29), implies that the protein is preferentially hy-drated. Although this notion is consistent with TMAO’s beinga protecting osmolyte, to the best of our knowledge no experi-mental studies have been carried out to directly examine theeffect of TMAO on protein hydration dynamics, an aspect thatis also important to protein function. Herein, we use two-dimensional infrared (2D IR) spectroscopy and a site-specific IRprobe to explore this critical issue and gain insight toward achievinga microscopic understanding of the molecular mechanism of theprotecting action of TMAO.2D IR spectroscopy is capable of assessing the frequency–

frequency correlation function (FFCF) of a given IR probe,which reports on the underlying dynamics of events that lead tofluctuations of its vibrational frequency (30). For an IR probethat is able to interact with water via H bonding, measurement ofits FFCF can provide, sometimes in a site-specific manner, de-tailed information about the hydration dynamics of the proteinmolecule of interest. For example, this approach has been used

to identify the existence of mobile water molecules inside Aβ40amyloid fibrils (31, 32) and to interrogate the water-assisteddrug-binding mechanism of HIV-1 reverse transcriptase (33),among many other applications (34–39). In the current study, wecapitalize on the established sensitivity of the nitrile stretchingvibration (C≡N) to local hydration and electrostatic environment(40) and use the unnatural amino acid p-cyano-phenyalanine(PheCN) as a local IR reporter. The advantages of using the C≡Nstretching vibration of PheCN are that (i) it is located in a spec-trally uncongested region (2,000−2,400 cm−1), where water hasa relatively low absorbance; (ii) it has a reasonably large ex-tinction coefficient; and (iii) it is, in most cases, a simple tran-sition that is decoupled from other vibrational modes of themolecule (41).Recently, we have shown that upon addition of TMAO, the

C≡N stretching vibration of PheCN in a short peptide shifts toa lower frequency compared with that obtained in pure water(42). As shown (Fig. 1), the peptide studied in the current case,Gly-PheCN-Gly (hereafter referred to as GFCNG), exhibits thesame trend. Taken together, these results suggest that TMAOweakens the strength of the H bond formed between the nitrilegroup and water molecules. Whereas this finding is consistentwith the ability of TMAO to increase protein stability, a micro-scopic picture regarding how TMAO alters protein–water inter-actions has yet to be established. To provide new insights into themolecular mechanism of the protecting action of TMAO, hereinwe measure the FFCF of the nitrile probe in two model systems,the tripeptide GFCNG and the villin headpiece subdomain (HP35)with PheCN mutations under different cosolvent conditions. Inparticular, we are able to extract both the homogeneous andthe inhomogeneous contributions to the total lineshape of theC≡N stretching band and, perhaps more importantly, to provide

Significance

The ability of trimethylamine N-oxide (TMAO) to counteractosmotic stress along with the deleterious effects of denatur-ants such as urea is fascinating and has inspired many studies.To further our understanding of how TMAO acts to stabilizefolded proteins, we carry out an infrared experiment designedto probe the microscopic details of this action explicitly fromthe perspective of the solute molecule. Our results reveal thatthe protein-stabilizing effect of TMAO originates from twocontributions: One is entropic and the other is enthalpic innature. Thus, this study provides not only microscopic detailsunderlying the stabilizing action of TMAO, but also a methodthat can be used to study the stability-perturbing effect ofother cosolvents.

Author contributions: F.G. designed research; J.M. and I.M.P. performed research; J.M.and I.M.P. analyzed data; and J.M., I.M.P., and F.G. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. E-mail: gai@sas.upenn.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1403224111/-/DCSupplemental.

8476–8481 | PNAS | June 10, 2014 | vol. 111 | no. 23 www.pnas.org/cgi/doi/10.1073/pnas.1403224111

Dow

nloa

ded

by g

uest

on

Mar

ch 2

6, 2

020

a dynamic view, from the perspective of the PheCN sidechain, ofhow a given solution condition (i.e., in the presence of TMAO,urea, or both) affects the hydration dynamics of the peptide inquestion (43). Our results suggest that TMAOmolecules producea local protein environment that strengthens backbone–backboneH bonds and also reduces the conformational entropy of the system.In other words, TMAO increases protein stability by acting as anano-crowder, as proposed by Thirumalai, Straub and coworkersbased on computer simulations (44).

ResultsConsistent with our previous study, TMAO affects the C≡Nstretching band of GFCNG by shifting its peak to a lower fre-quency and broadening its width (42). Although these changesin lineshape indicate a change in the local electrostatic andH-bonding environment of the C≡N probe, and thus containinformation about the molecular effect of TMAO, a quanti-tative assessment of the factors underlying these changes, basedon linear IR measurements alone, is not feasible. Thus, herein wecombine data from linear and nonlinear IR measurements anduse response function theory to fully characterize the lineshape ofthe C≡N stretching vibrations obtained under different solventconditions. This analysis allows us to dissect the homogeneousand inhomogeneous contributions to the linear IR spectrum ofinterest, which provides insights into the mechanism of actionof TMAO.

2D IR Photon Echo Spectra of GFCNG. As indicated (Fig. 2), theabsorptive 2D IR spectra of the nitrile group of GFCNG, col-lected at different waiting times (T) and solution conditions,clearly show that the 2D lineshapes and the spectral diffusiondynamics of the C≡N stretching vibration depend on the cosol-vent. For example, at T = 500 fs, the 2D IR lineshape of thisvibration in water becomes more circular than that in 6 M TMAO,suggesting that adding TMAO significantly changes contribu-tions from the homogenous part, as well as the inhomogenouspart, to the total lineshape of the vibrational transition. Toprovide a more quantitative assessment of how a particularcosolvent affects the microscopic environment of the nitrileprobe and hence its vibrational transition, below we examinehow the key spectral and dynamic properties of this vibratorchange with solution condition.First, we compare the vibrational lifetimes (T1) of the nitrile

probe obtained under different solution conditions. As shown(Table 1), the T1 of the C≡N stretching vibration, consistent with

previous studies (45, 46), is insensitive to solvent and, thus, is nota good reporter of its local environment. This finding is consis-tent with the study of Cho and coworkers (47), which showedthat the vibrational relaxation of a nitrile moiety attached toa large molecule in water is dominated by intramolecular path-ways. In addition, the measured T1 value (∼4 ps) indicates that inthis case the lifetime-broadening contribution (∼1.3 cm−1) to thespectral width (10−15 cm−1) is small.Second, we compare the effects of cosolvent on the spectral

diffusion of this probe, using the center line slope (CLS) methoddeveloped by Fayer and coworkers (48). A center line is definedas the line that connects the maxima of the peaks of a series ofcuts through a particular 2D IR spectrum that are parallel to theωτ frequency axis. It has been shown that, within the short timeapproximation, the waiting-time (i.e., T) dependence of the CLSdirectly reports on the normalized FFCF of the vibrational probein question (48). As indicated (Fig. 3), the CLS (T) of the nitrileprobe, obtained under all solution conditions, can be described bythe following equation:

CLS ðTÞ= A · e−t=τ +B; [1]

where A, B, and τ are constants. The exponential term in Eq. 1 isoften referred to as the Kubo term, which, in the current case,

Fig. 1. The C≡N stretch bands of GFCNG obtained under different solventconditions, as indicated. The solid lines are fits of these spectra to the linearresponse function using Eqs. 2–4 in the text and the resulting fittingparameters are listed in Table 1. The structures of urea and TMAO are shownas insets.

Fig. 2. Representative absorptive 2D IR photon echo spectra of GFCNGobtained under different solvent conditions and at different waiting times(T), as indicated.

Ma et al. PNAS | June 10, 2014 | vol. 111 | no. 23 | 8477

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Mar

ch 2

6, 2

020

most likely reports on the dynamics of H-bond formation andbreaking between the nitrile group and water, which cause thevibrational frequencies to fluctuate (49–51). As shown (Table 2),the decay time constant of this Kubo term determined forGFCNG in water is about 2.2 ps, which is in good agreementwith previously reported values for the nitrile stretching vibra-tion in aqueous solutions (30, 37, 46, 52). In addition, theoffset term B obtained in water is practically zero. However,adding a cosolvent, either TMAO or urea, results in a measurabledecrease in τ and an appreciable increase in B and, interestingly,TMAO is more effective than urea at increasing this offset. Fayerand coworkers (48) have shown, assuming that the system followsBloch dynamics, that this offset term is related to the total in-homogeneous component of the lineshape. Thus, these resultssuggest that TMAO can effectively slow down or even suppresssome motions that lead to local environmental changes of thenitrile probe and, hence, its spectral diffusion.To provide a more quantitative assessment of the cosolvent-in-

duced changes in the spectral and dynamic properties of the nitrilestretching vibration, we calculate the homogeneous and in-homogeneous contributions to the total lineshape, by analyzing thelinear IR spectrum using the linear response function (30). Specif-ically, we fit the IR spectrum, SIR(ω), to the following function (49):

SIRðωÞ∝ℜZ t

0

eiðω−ω0Þt · e−gðtÞ · e−t=2T1dt; [2]

where T1 is the vibrational lifetime, ω0 is the center frequency,and g(t) is the vibrational lineshape function, defined as

gðtÞ=Z t

0

Zτ′

0

dτ′dτ″Cðτ″Þ; [3]

where C(t) represents the full FFCF of the IR probe and is de-scribed by the following equation (53):

CðtÞ= 2δðtÞTp2

+Δ2s +Δ2 · e−t=τ; [4]

where the first term represents the motional narrowing compo-nent determined by the pure dephasing time (T2*), the secondterm represents a static or slowly varying component, and thethird term is the aforementioned Kubo term. As shown (Fig. 1),all of the linear IR spectra can be reasonably well described byEq. 2 and the resultant fitting parameters (i.e., T2*, Δs, and Δ) arelisted in Table 1. In particular, the parameters obtained for waterare in good agreement with those previously reported for thisprobe (37). With the knowledge of T2*, we further calculated thepure dephasing linewidth (Γ) for each case using the relationshipof Γ = 1/πT2*.In the current case, we have neglected the contribution from

Trot, which represents the rotational relaxation time of the probe(54) and is expected to be much longer than T1 or T2*. As in-dicated (Table 1), the spectral parameters produced from theabove analysis clearly show that the pure dephasing linewidth (Γ)of the C≡N stretching vibration decreases in the following order:H2O > 6 M urea > 1 M TMAO > TMAO/urea (3 M/3 M) >6 M TMAO, whereas the total inhomogeneous component (i.e.,ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiΔ2 +Δ2

s

q) follows the exact opposite trend [i.e., H2O < 6M urea <

1 M TMAO < TMAO/urea (3 M/3 M) < 6 M TMAO]. Moreover,the contribution from the Kubo term (i.e., Δ2) decreases uponadding cosolvents to the system, which, consistent with the aboveCLS analysis, indicates that cosolvents weaken the H bonding be-tween the C≡N probe and water molecules.As shown in Table 1, the most striking differences for the C≡N

stretching vibration under different solution conditions are in thevalues of the static term ðΔ2

s Þ. The origin of this static term,

Table 1. Spectroscopic parameters and lifetimes (T1) of the C≡N stretching vibration of GFCNGin different solvent conditions

Cosolvent condition ω0, cm−1 FWHM, cm−1 T1, ps Γ, cm−1 Δ, cm−1 Δs, cm

−1

PheCN* 2,236.3 12.2 4.5 5.8 ± 0.1 2.0 ± 0.1 —

H2O 2,236.3 10.4 4.2 ± 0.1 5.8 ± 0.1 2.1 ± 0.2 06 M urea 2,234.7 11.2 4.0 ± 0.2 5.2 ± 0.2 1.8 ± 0.1 1.0 ± 0.21 M TMAO 2,236.0 10.6 4.1 ± 0.2 5.1 ± 0.2 1.7 ± 0.1 1.2 ± 0.2TMAO/urea (3 M/3 M) 2,234.2 13.3 3.8 ± 0.2 5.0 ± 0.2 2.0 ± 0.2 2.2 ± 0.16 M TMAO 2,232.3 13.4 3.9 ± 0.2 4.8 ± 0.2 2.0 ± 0.2 2.9 ± 0.3

*The PheCN parameters are taken from ref. 37, which were measured in water. The center frequency (ω0) andwidth (FWHM) were determined by fitting the linear IR spectra to a Gaussian function. The T1 values weredetermined by fitting the isotropic transient grating signals to an exponential function. Other lineshape param-eters were obtained by fitting the C≡N stretching bands in Fig. 1 to Eqs. 2–4.

Fig. 3. CLS versus T plots of GFCNG obtained under different solvent con-ditions, as indicated. The solid lines are fits of these data to Eq. 1 with thosefitting parameters listed in Table 2.

Table 2. Parameters obtained from fitting the CLS curves in Fig.3 to Eq. 1

Cosolvent condition A τ, ps B

H2O 0.26 ± 0.02 2.17 ± 0.43 0.01 ± 0.026 M urea 0.20 ± 0.01 1.44 ± 0.30 0.15 ± 0.301 M TMAO 0.22 ± 0.02 1.30 ± 0.43 0.18 ± 0.02TMAO/urea (3 M/3 M) 0.31 ± 0.02 1.85 ± 0.39 0.24 ± 0.026 M TMAO 0.18 ± 0.02 1.46 ± 0.43 0.51 ± 0.02

8478 | www.pnas.org/cgi/doi/10.1073/pnas.1403224111 Ma et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

6, 2

020

for large molecules such as proteins, is generally attributed tomotions that can affect the local electrostatic environment of theIR probe but occur on a timescale that is much slower than thevibrational lifetime or the effective time window of the experi-ment, such as backbone conformational changes (55–57). In-terestingly, in the current case, the static component obtained inwater is nearly zero, indicating that during the time window ofthe experiment the nitrile probe can exhaustively sample, sta-tistically speaking, all possible environments. This is consistentwith the notion that the spectral diffusion of the nitrile probe inpure water is dominated by water H-bonding dynamics, whichoccur on the timescale of 1−3 ps (58). Furthermore, and perhapsmore interestingly, adding TMAO to the peptide solution leadsto a significant increase in the static component, which indicatesthat TMAO not only broadens the frequency distribution ofthe C≡N stretching vibration (Fig. 1), but also “freezes” a certainnumber of motions leading to vibrational frequency fluctuations.As shown in Table 1, although urea exhibits a similar effect, thenet influence, in comparison with that of TMAO at the sameconcentration, is much smaller. Garcia and coworkers (59) haveshown that the osmotic pressure of TMAO exhibits a positivedeviation from its ideal value, whereas that of urea shows anegative deviation. Thus, a more rigorous comparison betweenthe effects of TMAO and urea would require measurementson TMAO and urea samples that have identical osmoticpressures.

Linear IR and 2D IR Spectra of HP35 Mutants. To confirm that theconclusions reached above are not limited to short peptides, wealso performed similar experiments on two PheCN mutants ofa well-studied helical protein, HP35 (60, 61). The first variantcorresponds to a Phe76/PheCN mutation at the C terminus of theprotein (referred to as HP35-P76), wherein the PheCN residue isexposed to solvent. Because several studies have suggested thatprotecting osmolytes, such as TMAO, could directly interact witharomatic sidechains (14), the second mutant we chose to studycontains a Phe58/PheCN mutation at the core of the protein(referred to as HP35-P58). Because the PheCN residue in HP35-P58 is, to a large extent, buried in a hydrophobic environment (37,53, 61), it is unlikely that TMAO will show any direct interactionswith the nitrile probe. Thus, results obtained from both mutantswill allow us to gain a more complete assessment of the effect ofTMAO on protein stability. Unfortunately, we found that HP35readily aggregates even in 1 M TMAO, which prevented us frommaking a direct evaluation of the effect of TMAO on the C≡Nstretching vibration of both HP35 variants. To circumvent thisproblem, we evaluated the effect of TMAO in the presence ofurea, which is known to help solubilize the protein. Specifically, weused a solution condition that was used in the study of GFCNG[i.e., TMAO/urea (3 M/3 M)]. As indicated, both the linear and2D IR spectra of HP35-P76 (Figs. S1 and S2 and Table S1) showthat the cosolvent-induced changes in the spectral and dynamicproperties of the C≡N stretching vibration of HP35-P76 are verysimilar to those observed for the GFCNG peptide under the samesolution conditions. For example, the 2D lineshape, at T = 500 fs,obtained in pure water is distinctly more circular than that mea-sured in the presence of cosolvent. A more quantitative analysisusing the method discussed above reveals that the cosolvent-

induced spectral property changes in this case are similar to thoseobtained with GFCNG (Table 3), indicating that the findings forGFCNG, in terms of the effect of TMAO, are not unique to shortpeptides, but rather are generally applicable to proteins. Similarto that observed for HP35-P76, addition of the same cosolventmixture results in a shift of the C≡N stretching frequency of HP35-P58 to lower wavenumbers (Fig. S3 and Table S2). However, theCLS of HP35-P58 obtained from its 2D IR spectra (Fig. S4) inwater shows a slower decay and also a significant offset (Fig. 4 andTable 3). This is consistent with the study of Fayer and coworkers(37, 53, 61), which showed that the frequency fluctuation dynamicsof the C≡N probe inside the hydrophobic core of HP35 are sloweddown. However, addition of TMAO/urea (3 M/3 M) leads toa further increase in the amplitude of this offset (Fig. 4), indicatingthat the nitrile probe samples a more slowly varying or rigid en-vironment (53). Because, in this case, the PheCN residue is unlikelyto directly interact with cosolvent molecules, this increase inprotein rigidity thus most likely arises from the TMAO-inducedcrowding effect, as discussed above.

DiscussionThe current view on how TMAO and urea affect protein stabilityhas been recently reviewed by Canchi and García (62). Owing tovarious quantitative thermodynamic measurements, such as va-por pressure osmometry (63–65), volumetric measurements (66),and equilibrium dialysis (67), we now know a great deal aboutwhy urea and TMAO act differently. Urea’s protein denaturationeffect is generally believed to arise from the molecule’s tendencyto accumulate around both the protein backbone and sidechains,owing, for example, to a stronger dispersion interaction betweenurea and the protein backbone (68), which preferentially stabilizes

Table 3. Parameters obtained from fitting the C≡N stretching bands of HP35-P76 andHP35-P58 to Eqs. 2–4

HP35-P76 (HP35-P58) Γ, cm−1 Δ, cm−1 Δs, cm−1

H2O 5.6 ± 0.4 (4.8 ± 0.6) 1.9 ± 0.3 (4.0 ± 0.4) 0.4 ± 0.3 (2.0 ± 0.4)TMAO/urea

(3 M/3 M)5.1 ± 0.4 (4.2 ± 0.4) 2.2 ± 0.3 (2.3 ± 0.4) 2.3 ± 0.4 (4.1 ± 0.6)

Fig. 4. CLS versus T plots of HP35-P76 and HP35-P58 obtained under dif-ferent solvent conditions, as indicated. The solid lines are fits of these data toEq. 1 with those fitting parameters listed in Table 3. The relatively largeuncertainties in the CLS data and the apparent asymmetric 2D lineshapein the spectra (Figs. S2 and S4) are due to distortions from background noise,because the absorbance of the nitrile group in these experiments was rela-tively low (∼2 mOD).

Ma et al. PNAS | June 10, 2014 | vol. 111 | no. 23 | 8479

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Mar

ch 2

6, 2

020

the unfolded state. However, TMAO molecules are found to beexcluded from various protein backbone and sidechain units, thusdestabilizing the unfolded state (59). Moreover, many studies haveshown that both TMAO and urea can alter the structure anddynamics of the H-bonding network of water, thus resulting in achange in protein stability (69). Although previous studies havesignificantly enhanced our understanding of the mechanisms bywhich TMAO/urea increase/decrease protein stability, especiallyfrom a thermodynamic point of view, many important microscopicdetails have yet to be revealed or verified by experiments. Herein,we use linear and nonlinear IR measurements to directly probe,from a sidechain’s perspective, how TMAO and urea affect theH-bonding dynamics of water at the protein–solvent interface.Two key findings that emerged from the current study are

worthy of further discussion. First, the spectral diffusion data ofthe C≡N stretching vibration clearly show that addition of eitherurea or TMAO makes the local H-bonding environment fluc-tuate faster (i.e., the smaller time constant of the Kubo term inTable 2). This picture is consistent with our early observationthat TMAO and urea can weaken the H-bond strength betweenthe nitrile probe and water molecules (42). It is obvious that aweakened H-bonding interaction will result in a shallower po-tential energy surface along the H-bond coordinate, thus makingthe diffusion dynamics along this coordinate faster (at the sametemperature). In addition, weakening the H-bonding interactionswill lead to a smaller vibrational energy splitting, which, in turn,would lead to a smaller contribution to the total FFCF amplitudeby the corresponding H-bonding dynamics, as observed (A in Table 2and Δ2 in Table 1). Furthermore, we can rule out the possibility thatthe observed changes in the H-bonding dynamics arise from anincrease in viscosity (e.g., 6 M urea increases the solution viscosityby only 40%), because Kubarych and coworkers have recentlyshown that increasing bulk viscosity slows down spectral diffusion(70). Because water always competes, whenever possible, witha protein’s intrinsic H-bonding donors/acceptors to form additionalH bonds with, for example, the backbone amides, the decreasedstrength of H bonds involving water would preferentially destabilizethe unfolded state of proteins, where more amide units are exposedto the solvent. Thus, this is an important finding because it provides,to the best of our knowledge, the first experimental evidence in-dicating that TMAO can enhance protein stability by attenuatingthe strength of H bonds formed between protein polar groups andwater. Interestingly, our IR measurements suggest, although to alesser degree in comparison with TMAO, that urea can also weakenthe H-bonding ability of water with proteins (42). At first glance,this finding seems to be inconsistent with the denaturing role ofurea. However, upon consideration of the fact that urea, unlikeTMAO, shows preferential accumulation in the vicinity of pro-tein backbones and sidechains (14), this result provides strongevidence in support of the notion that urea’s ability to denature

proteins does not arise from its effect on the structure of water, butis instead achieved through a more direct mechanism. This con-clusion is in agreement with that reached by Pielak and coworkers(71), who showed that there is no correlation between a solute’simpact on water structure and its effect on protein stability.The second major finding is that the static component ðΔ2

s Þ in theFFCF, which is absent (or insignificant) in pure water, increasessignificantly upon addition of TMAO. This static component, to-gether with the aforementioned Kubo term, determines the in-homogeneous broadening of the C≡N stretching vibration. Thus,the fact that TMAO increases the amplitude of Δ2

s provides strongevidence indicating that it either slows down the peptide’s con-formational motions, creates new and slowly moving conforma-tional states that are not populated in pure water, or both. Thispicture is entirely consistent with the NMR study of Loria andcoworker (72), which showed that TMAO can significantly rigidifythe protein backbone, and with the simulations of Thirumalai andcoworkers (44), which demonstrated that TMAO can act as anano-crowder, thus increasing protein stability via the excludedvolume effect. Despite this consistency, we note that a study by Quand Bolen (73) indicated that the efficacy of the macromolecularcrowding in forcing proteins to fold may be modest and may alsodepend on the crowding agents. Thus, it would be very useful ifnew experiments can be designed to directly quantify the crowdingefficacy of TMAO.Taken together, the effect of TMAO on the Kubo and static

terms of the FFCF of the nitrile probe provides insights into themechanism of action of TMAO in increasing the stability ofproteins. Primarily, our results indicate that the protein stabi-lizing ability of TMAO arises from both enthalpic and entropiccontributions. The enthalpic factor results from a decreasedH-bonding ability of water, whereas the entropic factor stemsfrom the crowding effect of TMAO; both act to destabilize theunfolded state ensemble. Because the tripeptide used in thecurrent study is relatively small, it is possible to use moleculardynamics simulations (68, 74, 75), in conjunction with vibrationalfrequency calculations (76), to directly calculate the FFCFs ofthe nitrile probe under different cosolvent conditions and to providefurther molecular insights. We hope that the current study will in-spire new computational efforts in this regard.

Materials and MethodsThe details of sample preparation are given in SI Text. The linear FTIR spectraare collected at a resolution of 1 cm−1 on a Magna-IR spectrometer. Thedetails of the 2D IR measurements are also given in SI Text.

ACKNOWLEDGMENTS. We thank Dr. Arnaldo Serrano for synthesizing theHP35 variants. This work was supported by National Institutes of Health(NIH) Grant GM012592. The 2D IR instrumentation was developed with NIHResearch Resource Grant P41 GM104605. I.M.P. was supported by NationalScience Foundation Graduate Research Fellowship DGE-0822.

1. Yancey PH (2005) Organic osmolytes as compatible, metabolic and counteractingcytoprotectants in high osmolarity and other stresses. J Exp Biol 208(Pt 15):2819–2830.

2. Wang A, Bolen DW (1997) A naturally occurring protective system in urea-rich cells:Mechanism of osmolyte protection of proteins against urea denaturation. Bio-chemistry 36(30):9101–9108.

3. Zou Q, Bennion BJ, Daggett V, Murphy KP (2002) The molecular mechanism ofstabilization of proteins by TMAO and its ability to counteract the effects of urea.J Am Chem Soc 124(7):1192–1202.

4. Sharp KA, Madan B, Manas E, Vanderkooi JM (2001) Water structure changes inducedby hydrophobic and polar solutes revealed by simulations and infrared spectroscopy.J Chem Phys 114(4):1791–1796.

5. Di Michele A, et al. (2006) Modulation of hydrophobic effect by cosolutes. J PhysChem B 110(42):21077–21085.

6. Rezus YLA, Bakker HJ (2007) Observation of immobilized water molecules aroundhydrophobic groups. Phys Rev Lett 99(14):148301.

7. Rezus YLA, Bakker HJ (2009) Destabilization of the hydrogen-bond structure of waterby the osmolyte trimethylamine N-oxide. J Phys Chem B 113(13):4038–4044.

8. Koga Y, Westh P, Nishikawa K, Subramanian S (2011) Is a methyl group alwayshydrophobic? Hydrophilicity of trimethylamine-N-oxide, tetramethyl urea andtetramethylammonium ion. J Phys Chem B 115(12):2995–3002.

9. Munroe KL, Magers DH, Hammer NI (2011) Raman spectroscopic signatures of

noncovalent interactions between trimethylamine N-oxide (TMAO) and water. J Phys

Chem B 115(23):7699–7707.10. Sagle LB, et al. (2011) Methyl groups of trimethylamine N-oxide orient away from

hydrophobic interfaces. J Am Chem Soc 133(46):18707–18712.11. Bakulin AA, Pshenichnikov MS, Bakker HJ, Petersen C (2011) Hydrophobic molecules

slow down the hydrogen-bond dynamics of water. J Phys Chem A 115(10):1821–1829.12. Larini L, Shea J-E (2013) Double resolution model for studying TMAO/water effective

interactions. J Phys Chem B 117(42):13268–13277.13. Athawale MV, Dordick JS, Garde S (2005) Osmolyte trimethylamine-N-oxide does not

affect the strength of hydrophobic interactions: Origin of osmolyte compatibility.

Biophys J 89(2):858–866.14. Auton M, Bolen DW (2005) Predicting the energetics of osmolyte-induced protein

folding/unfolding. Proc Natl Acad Sci USA 102(42):15065–15068.15. Paul S, Patey GN (2007) Structure and interaction in aqueous urea-trimethylamine-N-

oxide solutions. J Am Chem Soc 129(14):4476–4482.

16. O’Brien EP, Ziv G, Haran G, Brooks BR, Thirumalai D (2008) Effects of denaturants and

osmolytes on proteins are accurately predicted by the molecular transfer model. Proc

Natl Acad Sci USA 105(36):13403–13408.

8480 | www.pnas.org/cgi/doi/10.1073/pnas.1403224111 Ma et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

6, 2

020

17. Meersman F, Bowron D, Soper AK, Koch MHJ (2009) Counteraction of urea by tri-methylamine N-oxide is due to direct interaction. Biophys J 97(9):2559–2566.

18. Hu CY, Lynch GC, Kokubo H, Pettitt BM (2010) Trimethylamine N-oxide influence onthe backbone of proteins: An oligoglycine model. Proteins 78(3):695–704.

19. Wei H, Fan Y, Gao YQ (2010) Effects of urea, tetramethyl urea, and trimethylamineN-oxide on aqueous solution structure and solvation of protein backbones:A molecular dynamics simulation study. J Phys Chem B 114(1):557–568.

20. Rösgen J, Jackson-Atogi R (2012) Volume exclusion and H-bonding dominate thethermodynamics and solvation of trimethylamine-N-oxide in aqueous urea. J AmChem Soc 134(7):3590–3597.

21. Fiore A, Venkateshwaran V, Garde S (2013) Trimethylamine N-oxide (TMAO) and tert-butyl alcohol (TBA) at hydrophobic interfaces: Insights from molecular dynamicssimulations. Langmuir 29(25):8017–8024.

22. Macdonald RD, Khajehpour M (2013) Effects of the osmolyte TMAO (Trimethylamine-N-oxide) on aqueous hydrophobic contact-pair interactions. Biophys Chem 184:101–107.

23. Bru�zdziak P, Panuszko A, Stangret J (2013) Influence of osmolytes on protein andwater structure: A step to understanding the mechanism of protein stabilization.J Phys Chem B 117(39):11502–11508.

24. Yancey PH, Clark ME, Hand SC, Bowlus RD, Somero GN (1982) Living with water stress:Evolution of osmolyte systems. Science 217(4566):1214–1222.

25. Uversky VN, Li J, Fink AL (2001) Trimethylamine-N-oxide-induced folding of alpha-synuclein. FEBS Lett 509(1):31–35.

26. Mello CC, Barrick D (2003) Measuring the stability of partly folded proteins usingTMAO. Protein Sci 12(7):1522–1529.

27. Qu YX, Bolen DW (2003) Hydrogen exchange kinetics of RNase A and the urea:TMAOparadigm. Biochemistry 42(19):5837–5849.

28. Panuszko A, Bru�zdziak P, Zielkiewicz J, Wyrzykowski D, Stangret J (2009) Effects ofurea and trimethylamine-N-oxide on the properties of water and the secondarystructure of hen egg white lysozyme. J Phys Chem B 113(44):14797–14809.

29. Auton M, Bolen DW, Rosgen J (2008) Structural thermodynamics of protein prefer-ential solvation: Osmolyte solvation of proteins, aminoacids, and peptides. Proteins73(4):802–813.

30. Hamm P, Zanni MT (2011) Concepts and Methods of 2D Infrared Spectroscopy(Cambridge Univ Pres, New York), p ix.

31. 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.

32. Ma J, et al. (2013) Intrinsic structural heterogeneity and long-term maturation ofamyloid β peptide fibrils. ACS Chem Neurosci 4(8):1236–1243.

33. Kuroda DG, et al. (2013) Snapshot of the equilibrium dynamics of a drug bound toHIV-1 reverse transcriptase. Nat Chem 5(3):174–181.

34. Strasfeld DB, Ling YL, Shim S-H, Zanni MT (2008) Tracking fiber formation in humanislet amyloid polypeptide with automated 2D-IR spectroscopy. J Am Chem Soc130(21):6698–6699.

35. Strasfeld DB, Ling YL, Gupta R, Raleigh DP, Zanni MT (2009) Strategies for extractingstructural information from 2D IR spectroscopy of amyloid: Application to islet am-yloid polypeptide. J Phys Chem B 113(47):15679–15691.

36. Ghosh A, Qiu J, DeGrado WF, Hochstrasser RM (2011) Tidal surge in the M2 protonchannel, sensed by 2D IR spectroscopy. Proc Natl Acad Sci USA 108(15):6115–6120.

37. Chung JK, Thielges MC, Fayer MD (2011) Dynamics of the folded and unfolded villinheadpiece (HP35) measured with ultrafast 2D IR vibrational echo spectroscopy. ProcNatl Acad Sci USA 108(9):3578–3583.

38. Bagchi S, Boxer SG, Fayer MD (2012) Ribonuclease S dynamics measured using a nitrilelabel with 2D IR vibrational echo spectroscopy. J Phys Chem B 116(13):4034–4042.

39. 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.

40. Getahun Z, et al. (2003) Using nitrile-derivatized amino acids as infrared probes oflocal environment. J Am Chem Soc 125(2):405–411.

41. Waegele MM, Culik RM, Gai F (2011) Site-specific spectroscopic reporters of the localelectric field, hydration, structure, and dynamics of biomolecules. J Phys Chem Lett2(20):2598–2609.

42. Pazos IM, Gai F (2012) Solute’s perspective on how trimethylamine oxide, urea, andguanidine hydrochloride affect water’s hydrogen bonding ability. J Phys Chem B116(41):12473–12478.

43. Bennion BJ, Daggett V (2004) Counteraction of urea-induced protein denaturation bytrimethylamine N-oxide: A chemical chaperone at atomic resolution. Proc Natl AcadSci USA 101(17):6433–6438.

44. Cho SS, Reddy G, Straub JE, Thirumalai D (2011) Entropic stabilization of proteins byTMAO. J Phys Chem B 115(45):13401–13407.

45. Ghosh A, Remorino A, Tucker MJ, Hochstrasser RM (2009) 2D IR photon echo spectroscopyreveals hydrogen bond dynamics of aromatic nitriles. Chem Phys Lett 469(4-6):325–330.

46. Urbanek DC, Vorobyev DY, Serrano AL, Gai F, Hochstrasser RM (2010) The twodimensional vibrational echo of a nitrile probe of the villin HP35 protein. J Phys ChemLett 1(23):3311–3315.

47. Park K-H, et al. (2013) Infrared probes based on nitrile-derivatized prolines: Thermalinsulation effect and enhanced dynamic range. J. Phys. Chem. Lett 4(13):2105–2110.

48. Kwak K, Park S, Finkelstein IJ, Fayer MD (2007) Frequency-frequency correlationfunctions and apodization in two-dimensional infrared vibrational echo spectroscopy:A new approach. J Chem Phys 127(12):124503.

49. Mukamel S (1995) Principles of Nonlinear Optical Spectroscopy (Oxford Univ Press,New York), p xviii.

50. Cho M (2009) Two-Dimensional Optical Spectroscopy (CRC, Boca Raton, FL), p 378.51. Volkov VV, Palmer DJ, Righini R (2007) Distinct water species confined at the interface

of a phospholipid membrane. Phys Rev Lett 99(7):078302.52. Ghosh A, Hochstrasser RM (2011) A peptide’s perspective of water dynamics. Chem

Phys 390(1):1–13.53. Chung JK, Thielges MC, Fayer MD (2012) Conformational dynamics and stability of

HP35 studied with 2D IR vibrational echoes. J Am Chem Soc 134(29):12118–12124.54. Hochstrasser RM (2001) Two-dimensional IR-spectroscopy: Polarization anisotropy

effects. Chem Phys 266(2-3):273–284.55. Huang CY, Getahun Z, Wang T, DeGrado WF, Gai F (2001) Time-resolved infrared

study of the helix-coil transition using (13)C-labeled helical peptides. J Am Chem Soc123(48):12111–12112.

56. Wang T, Du DG, Gai F (2003) Helix-coil kinetics of two 14-residue peptides. Chem PhysLett 370(5-6):842–848.

57. Lin YS, Shorb JM, Mukherjee P, Zanni MT, Skinner JL (2008) Empirical amide I vibra-tional frequency map: Application to 2D-IR line shapes for isotope-edited membranepeptide bundles. J Phys Chem B 113(3):592–602.

58. Asbury JB, et al. (2004) Water dynamics: Vibrational echo correlation spectroscopyand comparison to molecular dynamics simulations. J Phys Chem A 108(7):1107–1119.

59. Canchi DR, Jayasimha P, Rau DC, Makhatadze GI, Garcia AE (2012) Molecular mech-anism for the preferential exclusion of TMAO from protein surfaces. J Phys Chem B116(40):12095–12104.

60. Hu KN, Havlin RH, YauWM, Tycko R (2009) Quantitative determination of site-specificconformational distributions in an unfolded protein by solid-state nuclear magneticresonance. J Mol Biol 392(4):1055–1073.

61. Chung JK, Thielges MC, Lynch SR, Fayer MD (2012) Fast dynamics of HP35 for foldedand urea-unfolded conditions. J Phys Chem B 116(36):11024–11031.

62. Canchi DR, García AE (2013) Cosolvent effects on protein stability. Annu Rev PhysChem 64(1):273–293.

63. Guinn EJ, Pegram LM, Capp MW, Pollock MN, Record MT, Jr. (2011) Quantifying whyurea is a protein denaturant, whereas glycine betaine is a protein stabilizer. Proc NatlAcad Sci USA 108(41):16932–16937.

64. Cannon JG, Anderson CF, Record MT, Jr. (2007) Urea-amide preferential interactionsin water: Quantitative comparison of model compound data with biopolymer resultsusing water accessible surface areas. J Phys Chem B 111(32):9675–9685.

65. Zhang W, Capp MW, Bond JP, Anderson CF, Record MT, Jr. (1996) Thermodynamiccharacterization of interactions of native bovine serum albumin with highly excluded(glycine betaine) and moderately accumulated (urea) solutes by a novel application ofvapor pressure osmometry. Biochemistry 35(32):10506–10516.

66. Lee S, Shek YL, Chalikian TV (2010) Urea interactions with protein groups: A volu-metric study. Biopolymers 93(10):866–879.

67. Arakawa T, Timasheff SN (1982) Preferential interactions of proteins with salts inconcentrated solutions. Biochemistry 21(25):6545–6552.

68. Hua L, Zhou R, Thirumalai D, Berne BJ (2008) Urea denaturation by stronger disper-sion interactions with proteins than water implies a 2-stage unfolding. Proc Natl AcadSci USA 105(44):16928–16933.

69. Kuffel A, Zielkiewicz J (2010) The hydrogen bond network structure within the hy-dration shell around simple osmolytes: Urea, tetramethylurea, and trimethylamine-N-oxide, investigated using both a fixed charge and a polarizable water model. J ChemPhys 133(3):035102.

70. King JT, Baiz CR, Kubarych KJ (2010) Solvent-dependent spectral diffusion ina hydrogen bonded “vibrational aggregate” J Phys Chem A 114(39):10590–10604.

71. Batchelor JD, Olteanu A, Tripathy A, Pielak GJ (2004) Impact of protein denaturantsand stabilizers on water structure. J Am Chem Soc 126(7):1958–1961.

72. Doan-Nguyen V, Loria JP (2007) The effects of cosolutes on protein dynamics: Thereversal of denaturant-induced protein fluctuations by trimethylamine N-oxide.Protein Sci 16(1):20–29.

73. Qu Y, Bolen DW (2002) Efficacy of macromolecular crowding in forcing proteins tofold. Biophys Chem 101–102:155–165.

74. Zangi R, Zhou R, Berne BJ (2009) Urea’s action on hydrophobic interactions. J AmChem Soc 131(4):1535–1541.

75. Kohlhoff KJ, et al. (2014) Cloud-based simulations on Google Exacycle reveal ligandmodulation of GPCR activation pathways. Nat Chem 6(1):15–21.

76. Choi J-H, Cho M (2011) Vibrational solvatochromism and electrochromism of infraredprobe molecules containing C≡O, C≡N, C=O, or C-F vibrational chromophore. J Chem Phys134(15):154513.

Ma et al. PNAS | June 10, 2014 | vol. 111 | no. 23 | 8481

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Mar

ch 2

6, 2

020