proteinsSTRUCTURE O FUNCTION O BIOINFORMATICS

Kinetic consequences of native stateoptimization of surface-exposed electrostaticinteractions in the Fyn SH3 domainArash Zarrine-Afsar,1,2 Zhuqing Zhang,1,2,3 Katrina L. Schweiker,4 George I. Makhatadze,4

Alan R. Davidson,1,2* and Hue Sun Chan1,2,3*1Department of Biochemistry, University of Toronto, Toronto, Ontario, M5S 1A8 Canada

2Department of Molecular Genetics, University of Toronto, Toronto, Ontario, M5S 1A8 Canada

3Department of Physics, University of Toronto, Toronto, Ontario, M5S 1A7 Canada

4 Center for Biotechnology and Interdisciplinary Studies and Department of Biology, Rensselaer Polytechnic Institute,

Troy, New York 12180

INTRODUCTION

Commencing with the seminal works of Linderstrøm-Lang

in the 1920s, it is now widely accepted that electrostatic inter-

actions play a significant role during protein folding. Optimiza-

tion of charge–charge interactions on protein surface has

emerged as an attractive strategy to enhance protein stability.1–3

Indeed, the increase in stability gained through in silico optimi-

zations of surface charge–charge interactions is seen to be com-

parable to optimizations of hydrophobic interactions in the

protein core.4 The effects of optimizing charge–charge interac-

tions on the equilibrium stability of proteins have been charac-

terized in considerable detail4,5; but it is generally unknown as

to how redesigning the surface charge distribution of a protein

will affect its folding and unfolding kinetics. Understanding the

interplay between specific and nonspecific electrostatic interac-

tions during folding—change in global charge density due to

compaction, for example—is of importance to elucidating the

biophysical principles that drive folding in general and, at the

same time, is invaluable to investigators who routinely use pro-

tein engineering method to selectively disrupt a specific interac-

Arash Zarrine-Afsar and Zhuqing Zhang contributed equally to this work.

Arash Zarrine-Afsar’s current address is Lash Miller Chemical Laboratories, Department of

Chemistry, University of Toronto, 80 St. George Street, Toronto, Ontario M5S 3H6, Canada

Zhuqing Zhang’s current address is College of Life Science, Graduate University of the

Chinese Academy of Sciences, 19A Yuquanlu, Shijingshan District, Beijing 100049, China

Katrina L. Schweiker’s current address is Air Force Research Laboratory, 3550 Aberdeen

Ave SE, Kirtland AFB, NM 87117 USA

Grant sponsor: Doctoral Canada Graduate Scholarship; Grant number: CGS-D3; Grant

sponsor: Canadian Institutes of Health Research; Grant number: MOP-13609, MOP-84281;

Grant sponsor: U.S. National Science Foundation (NSF); Grant number: MCB-0110396,

MCB-1051344; Grant sponsor: Natural Science and Engineering Research Council of Can-

ada; Grant sponsor: Canada Research Chairs Program.

*Correspondence to: Hue Sun Chan, Department of Biochemistry and Department of

Molecular Genetics, University of Toronto, Toronto, Ontario M5S 1A8 Canada. E-mail:

chan@arrhenius.med.toronto.edu or Alan R. Davidson, Department of Biochemistry and

Department of Molecular Genetics, University of Toronto, Toronto, Ontario M5S 1A8

Canada. E-mail: alan.davidson@utoronto.ca

Received 13 September 2011; Revised 24 October 2011; Accepted 29 October 2011

Published online 9 November 2011 in Wiley Online Library (wileyonlinelibrary.com).

DOI: 10.1002/prot.23243

ABSTRACT

Optimization of surface exposed charge–charge interac-

tions in the native state has emerged as an effective

means to enhance protein stability; but the effect of elec-

trostatic interactions on the kinetics of protein folding is

not well understood. To investigate the kinetic conse-

quences of surface charge optimization, we characterized

the folding kinetics of a Fyn SH3 domain variant contain-

ing five amino acid substitutions that was computation-

ally designed to optimize surface charge–charge interac-

tions. Our results demonstrate that this optimized Fyn

SH3 domain is stabilized primarily through an eight-fold

acceleration in the folding rate. Analyses of the constitu-

ent single amino acid substitutions indicate that the

effects of optimization of charge–charge interactions on

folding rate are additive. This is in contrast to the trend

seen in folded state stability, and suggests that electro-

static interactions are less specific in the transition state

compared to the folded state. Simulations of the transi-

tion state using a coarse-grained chain model show that

native electrostatic contacts are weakly formed, thereby

making the transition state conducive to nonspecific, or

even nonnative, electrostatic interactions. Because folding

from the unfolded state to the folding transition state for

small proteins is accompanied by an increase in charge

density, nonspecific electrostatic interactions, that is,

generic charge density effects can have a significant con-

tribution to the kinetics of protein folding. Thus, the

interpretation of the effects of amino acid substitutions

at surface charged positions may be complicated and con-

sideration of only native-state interactions may fail to

provide an adequate picture.

Proteins 2012; 80:858–870.VVC 2011 Wiley Periodicals, Inc.

Key words: protein folding; SH3 domains; protein

folding kinetics; coarse-grained modeling; electrostatic

interactions.

858 PROTEINS VVC 2011 WILEY PERIODICALS, INC.

tion in the folding transition state through mutagenesis.6,7

One of the major assumptions of the protein engineering

method is that single-point mutations must remove a spe-

cific set of native interactions in order for mutagenesis to

be useful as a probe of the transition state. Therefore, resi-

dues that influence the stability through nonspecific or

nonnative mechanisms are difficult to probe. This issue is

important in protein engineering studies of charge–charge

interactions, as the majority of studies investigating elec-

trostatic interactions have focused only on the role of these

interactions in the folded or unfolded states.8–11 Our

knowledge of the role of electrostatic interactions in fold-

ing transition states,12,13 however, is still quite limited.

Because the strength of electrostatic interactions is readily

modulated by solvent effects, charge–charge interactions

can be exploited to tune protein kinetic stability14 so as to

increase protein shelf-life for therapeutic or biotechnology

applications.13 A better understanding of the electrostatic

interactions in the folding/unfolding transition state is

expected to assist such practical applications as well.

In this context, we report below an investigation of the

impact of surface charge–charge interactions on the folding

kinetics of the SH3 domain of Fyn tyrosine kinase. Fyn SH3

is composed of two b-sheets orthogonally packed against

one another, a structure that can be stabilized significantly

through the rational design of surface charge–charge inter-

actions.4 This protein folds in an essentially two-state man-

ner,15 that is, the polypeptide chain folds into its folded

state by passing through a high-energy transition state bar-

rier ({) without any substantial transient population of in-

termediate states.16 The folding transition state of the SH3

domains has been studied extensively using experimental

F-value analysis17–20 as well as computational model-

ing.16,21–25 Because this class of proteins is amenable to

in vitro biophysical analyses, the extensive structural data

about SH3 folding transition states has greatly facilitated

our acquisition and interpretation of Fyn SH3 folding

kinetics data. To characterize the electrostatic interactions

operating in the folding transition state, we have also per-

formed folding simulations using a modified version of a

coarse-grained protein chain model that has recently been

shown to capture the sequence-dependent nonnative

hydrophobic interactions during Fyn SH3 folding.25

MATERIALS AND METHODS

Protein expression and purification

The Fyn SH3 domain variants were recombinantly

expressed as described previously.26

Differential scanning calorimetry (DSC)

The details of DSC experiments are as described previ-

ously.4 All DSC measurements were performed in

50 mM sodium phosphate, 100 mM NaCl, pH 7.0.

Folding kinetics studies

The folding kinetics of proteins was monitored by Trp

fluorescence using a Bio-Logic SFM-4 stopped flow de-

vice (BioLogic Instruments, Claix, France) as described

previously.26 For the Fyn SH3 domain variants, the fold-

ing kinetic data at 258C in 50 mM sodium phosphate,

100 mM NaCl, pH 7.0 were fit to the linear equation

ln kobs ¼ ln kf � ðmkf ½urea�Þ

where kobs is the observed rate constant at a given urea

concentration, kf is the folding rate in 0 M urea, and mkf

is the dependence of ln kf on the concentration, [urea],

of urea.

Values for changes in the free energy between the

unfolded state, transition state, and the folded state were

calculated as follows:

DDGz!u¼ �kBT lnðkMUT

f =kWTf Þ

DDGf!z ¼ �kBT lnðkWT

u =kMUTu Þ

DDGf!u ¼ DDGz!uþ DDG

f!z

where kB is Boltzmann constant, here T is absolute tem-

perature, and the superscripts ‘‘MUT’’ and ‘‘WT’’ stand

for mutant and wild type, respectively.

To determine the unfolding rates (ku) at room temper-

ature, we combined DDGf?u and DDG{?u values

obtained from DSC and stopped flow experiments,

respectively, to estimate DDGf?{ values at room tempera-

ture using the relations above. Here, since the kinetic mkf

values are not drastically different from the value

obtained for the WT protein, we do not expect signifi-

cant error to be incurred by the linear extrapolation of

folding rates to [Urea] 5 0 M. As an illustration of the

accuracy of this general procedure, consider the calcu-

lated unfolding rate for the WT protein using this com-

bined measurement (0.030 � 0.006 s21, Table I). In view

of the experimental uncertainties, this calculated rate is

close to the extrapolated value of 0.015 � 0.009 s21

obtained from previous stopped flow measurements on

WT protein, despite the short unfolding arm of its chev-

ron plot.27

Circular dichroism spectroscopy

Temperature-induced melting of the domain was

monitored by changes in the CD signal (ellipticity) at

220 nm on an Aviv circular dichroism spectrometer

(Aviv Associates, Lakewood, NJ). Melt profiles were fit to

obtain the Tm values, as described previously.28 Far UV

scans of variants using 50 lM protein sample (in 50 mM

sodium phosphate, 100 mM NaCl, and pH 7.0 as in the

Electrostatic Interactions in Folding

PROTEINS 859

DSC experiments) at T 5 258C were also recorded to

ensure that none of the substitutions grossly altered the

secondary structure in WT.

Coarse-grained chain model simulations

We used a coarse-grained Ca chain model to gain

insights into charge-changing mutations. In folding stud-

ies, simple computational models with explicit chain rep-

resentations29 are valuable for exploring effects that are

not readily accessible by experiment. Taking an approach

analogous to our treatment of nonnative hydrophobic

interactions,25 the potential energy function E in the

present model comprises of a native-centric, Go-like30

background interaction augmented by sequence-depend-

ent contributions, viz.,

E ¼ E0 þ Ee ð1Þ

where E0 denotes the native-centric potential used in

Ref. 31 with e 5 1 (which is equivalent to the potential

for the no-desolvation barrier model in Ref. 32),

Ee ¼X

i

X

j

Kij ½ðaij=rij Þ12 þ qiqjbij expð�jrijÞ=rij � ð2Þ

is introduced to capture rudimentary aspects of pairwise

electrostatic interactions between residues i and j with

electric charges qi and qj, respectively, rij is the Ca–Cadistance (in A) between the pair of residues, and the Kij’s

control the strength of the electrostatic terms relative to

that of E0. The first term in Eq. (2) is for excluded vol-

ume whereas the second term accounts for charge–charge

interactions. We set qi 5 1 for arginine and lysine, qi 521 for aspartic and glutamic acid, and qi 5 0 for all the

other types of amino acid residues. In particular, because

the typical pKa � 6.0–6.5 of the ionizable histidine side-

chain in proteins33 is close to our experimental pH 57.0, histidine is treated as neutral (qi 5 0) in Ee. Our

previous work utilized a pKa value of 6.3 for His, and a

correspondence between theory and experiment was

achieved.4 Aiming for a simple physical picture, we used

a Ca representation of the protein chain instead of using

a structurally more accurate but computationally more

intensive model with charged sidechains.34 The second

term in Eq. (2) takes the form of a screened Coulombic

interaction. The standard Debye–Huckel formula for the

inverse screening length j gives j 5 (8pIe2/er kBT)1/2,

where e2 is the square of the electronic charge, I is the

ionic strength, and er is the dielectric constant of the sol-

vent. Although the effective dielectric constant of water is

distance dependent and can be significantly smaller than

the bulk value for short separations <10 A (Ref. 35), in

the context of our highly coarse-grained approach we

simply use the bulk value of er 5 78.5. This er value

gives j � I1/2/3.04 A21 � 0.32 I1/2 A21 at T 5 258C 5298.15 K (see, e.g., Refs. 36 and 37). The ionic strength

of the 50 mM Na3PO4, 100 mM NaCl buffer in our

experiment is I 5 [3 3 0.05 M(11)2 1 0.05 M(23)2 10.1 M(11)2 1 0.1 M(21)2]/2 5 0.4 M, thus the inverse

screening length is 0.208 A21. Accordingly, we used j 50.2 A21 for Ee in Eq. (2).

The summation in Eq. (2) is over i < j – 2 and re-

stricted to i,j for which qi qj = 0. Unlike the expression

EHP we used previously that accounted only for nonna-

tive hydrophobic interactions,25 Ee encompasses both

native and nonnative charge–charge interactions regard-

less of whether a residue pair is already interacting

favorably in E0. For our model to be viable, Ee cannot

significantly distort the native structure favored by E0.

Thus, aij, bij, and Kij were chosen judiciously for pairs

of residues that are sequentially or spatially close to

each other in the native structure. The purpose is to

ensure that repulsion between like charges is not too

strong and the energy minima of the attractive potential

between opposite charges would coincide approximately

with the Ca–Ca distance rijn between these pairs of res-

idues in the PDB structure. We did so by setting aij 5rij

n (0.002 rijn 1 0.968) and bij 5 14.88 exp(0.293 rij

n)

Table IExperimental Data on Folding Thermodynamics and Kinetics of the Proteins Examined in This Study

Fyn SH3 domain variants kf (s21)a mkf ku (s

21)b Tm (8C)cDDG{?u

d

(kcal mol21)DDGf?{

d

(kcal mol21)DDGf?u

d

(kcal mol21)

WT 77 � 4 0.80 0.030 � 0.006 71.6 0.00 0.00 0.00E11K 84 � 4 0.81 0.031 � 0.006 70.6 20.06 � 0.04 0.04 � 0.17 20.02 � 0.18D16K 102 � 5 0.75 0.008 � 0.002 77.1 20.17 � 0.04 20.76 � 0.17 20.93 � 0.18H21K 97 � 5 0.82 0.012 � 0.002 76.6 20.14 � 0.04 20.53 � 0.18 20.67 � 0.18N30K 105 � 5 0.77 0.062 � 0.012 71.2 20.19 � 0.04 0.45 � 0.17 0.26 � 0.18E46K 232 � 12 0.82 0.013 � 0.003 77.7 20.66 � 0.04 -0.46 � 0.18 21.12 � 0.18E46K-E11K-D16K-H21K-N30K 648 � 32 0.88 0.014 � 0.003 83.3 21.27 � 0.04 20.43 � 0.17 21.70 � 0.18

aThe extrapolated folding rates of the Fyn SH3 domain variants were obtained by fitting the urea refolding data in Fig. 2 to lnkobs 5 lnkf – mkf [urea], as described in

Materials and Methods. The uncertainties reported for folding rates are �5% of the rate, based on repetitions.bThe unfolding rates were calculated by combining the urea folding data and overall stability data from DSC unfolding. The errors of the extrapolated unfolding rates

are �20% of the rates.cMelting temperature (Tm) values from DSC experiments. Some of the Tm values are those reported previously4 and are reproduced here to facilitate the present discus-

sion.dThe free energy changes were calculated from folding and unfolding rates of variants with respect to WT according to the relations given in Materials and Methods.

A. Zarrine-Afsar et al.

860 PROTEINS

for three classes of residue pairs: (i) all i,i13 pairs (j 5i 1 3), (ii) j > i 1 3 pairs for which the shortest spa-

tial distance, dmin, between two nonhydrogen atoms

(one from each residue) is within 4.5 A (these corre-

spond to the native contacts,25,31,32 see below), and

(iii) j > i 1 3 pairs with 4.5 A < dmin � 6 A. For

these pairs, Kij was chosen such that the energy of every

attractive (< 0) term at rijn is 20.8e and the energy of

every repulsive (> 0) term at the same position is 0.8e,where e is the well depth of the interaction energy

between a native pair in E0 (see Ref. 31). In addition to

these interactions, we also allow for nonnative charge–

charge interactions for j > i 1 3 residue pairs with

dmin > 6 A. For these pairs, we set aij 5 4.8 and bij 550. The resulting attractive well (energy minimum) for

favorable nonnative electrostatic interactions is at r0 �5 A, similar to that for the nonnative hydrophobic

interactions in our previous study.25 The Kij values

were also chosen so that attractive and repulsive nonna-

tive charge–charge interaction energies at r0 are equal to

20.8e and 0.8e, respectively.In addition to Ee, two alternate potentials, Ee

(His1) and

EeMR, were studied as controls. As stated above, the

charge of histidine is set to zero in Ee, as in the recent

model of Azia and Levy34 because the pKa of the histi-

dine sidechain is close to our experimental pH. To assess

the effects of a possible shift in the protonation state of

the histidine during the folding process of Fyn SH3, we

define Ee(His1) to be identical to Ee except the qi value

for histidine is set to 11 instead of 0. Because there is

only one histidine (at position 21) in the WT Fyn SH3

sequence, the WT and H21K single mutant become

equivalent in the Ee(His1) construct.

The other alternate potential EeMR is referred to as

medium-range (MR) because it stipulates a stronger

screening and thus a shorter spatial range than Ee; but

the spatial range of EeMR is still longer than that of the

EHP potential we used previously for nonnative hydro-

phobic interactions.38 EeMR also takes the general form

of Eq. (2) (with qi 5 0 for histidine) but with a larger

j 5 0.63 instead of the j 5 0.2 value for Ee. Conse-

quently, EeMR decays faster with increasing rij than Ee

(see below). As such, EeMR is useful for assessing the

impact of charge–charge interactions with shorter spatial

ranges, which can originate, for instance, from a dis-

tance-dependent dielectric effect.35 The aij values in

EeMR are identical to those in Ee, whereas the bij values

in EeMR were modified from those in Ee to maintain ev-

ery energy minimum at approximately the same rij sepa-

ration as that in Ee. This was accomplished by setting bij5 7.812 exp(0.693 rij

n) for the three aforementioned

classes of contacts that involve PDB distance rijn, and by

setting bij 5 180 for nonnative electrostatic interactions

with r0 � 5 A. As for Ee, the Kij values for EeMR were

chosen such that the energy of every term at rijn or r0 is

approximately �0.8e.

Conformational sampling for all models was per-

formed using Langevin dynamics as described

before.25,31,32 As in these references, a residue pair i,j is

defined to be in the native contact set if | i – j | > 3 and

dmin � 4.5 A. Only these pairs have favorable contact

interactions in E0 and are counted by the variable Q for

fractional number of native contacts.31 As before,25 a

native contact in this set is considered to be formed dur-

ing the Langevin dynamics simulation when rij < 1.2rijn.

Probabilities and related properties of these native con-

tacts are reported in the lower-right parts of the contact

maps below. As discussed above, we also monitored con-

tacts between those i,i13 pairs and j > i 1 3 pairs with

4.5 A < dmin � 6 A that are involved in charge–charge

interactions in either WT or in Fyn5. There are five such

i,i13 pairs and five such j > i 1 3 pairs with 4.5 A <dmin � 6 A. As for the native contacts, a contact between

any given one of these pairs is considered to have formed

when rij < 1.2rijn. Probabilities and related properties of

these contacts are also reported in the lower-right parts

of our contact maps. A nonnative electrostatic contact

between residues i,j is considered to exist if qi qj = 0, | i

– j | > 3, dmin > 6 A, and rij < 7 A. The probabilities

and related properties of these nonnative electrostatic

contacts are reported in the upper-left parts of our con-

tact maps. Contacts between all other residue pairs are

not reported in the present contact maps. Models for

WT and mutants are based on the same PDB structure

1SHF39 for Fyn SH3 [Fig. 1(A)]. Folding transition states

are modeled by chains with 0.4 < Q < 0.55, correspond-

ing to the conformations that populate the barrier

region30–32 of the free energy profiles. Our previous

study using a similar criterion for model transition state

ensembles indicated that they can provide useful predic-

tions for the folding kinetics of SH3 domain proteins.25

It is worth emphasizing that the parameter choices in

all three models of electrostatic interactions used in the

present study were based upon physical considerations

and intuitive, constructive adaptations to the limitations

of the present coarse-grained setup. No calibration

beyond what has been described above was conducted to

optimize agreement between our computational models

and experiment.

RESULTS

Using the procedure in Materials and Methods, we

examined the folding and unfolding rates of a designed

Fyn SH3 variant with optimized surface charge–charge

interactions4 and compared its behavior with its constit-

uent single-mutant variants by performing denaturant-

induced unfolding experiments probed by stopped-flow

Trp fluorescence. The designed protein involves five

mutations. Four of them are charge reversals and one

introduces a positive charge. We refer to this protein as

Electrostatic Interactions in Folding

PROTEINS 861

Fyn5, which exhibits an increase in equilibrium stability

of 1.8 kcal mol21 relative to the WT protein.4 Figure 1

shows the positions of the mutations and their effects on

stability. Because the stability of Fyn5 did not have the

same dependence on the ionic strength as WT, the inter-

pretation of rates obtained from GuHCl-mediated

unfolding/refolding experiments would be unduly com-

plicated. Urea, on the other hand, was required in satu-

rating amounts to bring about complete unfolding of

Fyn5 and some of the single-mutant variants. As a result,

only the folding rates could be directly measured using

stopped-flow methods. Unfolding rates were calculated

by combining the folding kinetics data with the equilib-

rium stability data (at the same temperature T 5 258C)obtained from temperature-induced unfolding monitored

by differential scanning calorimetry (DSC).

Figure 2 provides the denaturant dependence of the

observed folding rates for the mutants we studied. The

data show a linear dependence of lnkobs on the concen-

tration of urea (M). The folding and stability data

derived from these experiments are summarized in Table

I. Circular dichroism (CD) spectroscopy4 was used to

corroborate the DSC measurements. The transition mid-

point temperature (Tm) obtained from DSC and CD

experiments are in very good agreement [Fig. 1(B)], sug-

gesting that our charge optimization does not perturb

the essentially two-state folding mechanism of Fyn SH3.

Far-UV CD spectra indicate that the secondary structural

contents of the mutants are not significantly altered ei-

ther (data not shown). We therefore interpret all kinetic

data according to the principles of two-state fold-

ing,6,15,40–42 based on both kinetic (single exponential

relaxation to equilibrium without populating folding in-

termediate states) and the calorimetric (van’t Hoff en-

thalpy being equal to calorimetric enthalpy) criteria.43 In

addition, the absence of drastic changes in kinetic m-

value (mkf, slope of lnkobs vs. [Urea] in Fig. 2) suggests

that variations in the unfolded state structure44 are not

large among the Fyn SH3 mutants. The data presented in

Table I indicate that Fyn5 is stabilized relative to WT

through a dramatic eight-fold acceleration in its folding

rate accompanied by a two-fold deceleration in the

unfolding rate.

To identify pairs of charge residues with significant

interaction energies, we summarize in Table II the muta-

tional effects on charge–charge interactions quantified by

the change in Coulombic contribution DDGqq to the free

energy of folding from Tanford-Kirkwood modeling,45

Figure 1The structure of Fyn SH3 domain and the stabilities of its mutants. (A)

The sidechains mutated in this study are shown in red. Also highlighted

in the ribbon diagram are the structured regions of the folding

transition state consisting of b-strands b through d. This drawing was

generated by the program PyMol (DeLano Scientific, Palo Alto, CA)

using the coordinates of the Fyn SH3 domain from the PDB structure

1SHF. (B) The correspondence between the folding/unfolding midpoint

(melting) temperatures determined by DSC (Table I) and those by CD

spectroscopy for the WT and six mutant Fyn SH3 domains considered

in this study; r is the Pearson correlation coefficient.

Figure 2Folding kinetics of WT Fyn SH3 and its variants. As described in

Materials and Methods, solid lines are linear least-squares fits of the

data for obtaining the extrapolated folding rates under nondenaturing

conditions at zero urea concentration.

A. Zarrine-Afsar et al.

862 PROTEINS

where DDGqq 5 DGqqMUT 2 DGqq

WT is the difference in

the theoretical free energy of charge interaction between

a given mutant (MUT) and WT.4 As highlighted in Table

II, Tanford-Kirkwood modeling predicts significant

increases in favorability for a number of specific interac-

tions in MUT relative to that in WT. For example, K46 is

predicted to have a stronger specific interaction with E24

in MUT. We have also verified that the predicted DDGqq

values summed over all interacting residues (last column

on the right in Table II) are well correlated with the ex-

perimental stability DDGf?u values in Table I [Fig. 3(A)].

This agreement between theory and experiment argues

for a dominant role played by electrostatic interactions in

stabilizing the Fyn SH3 variants.4 However, the change

in folding and unfolding rates caused by the charge-

changing mutations are not well correlated [Fig. 3(B)].

In particular, the magnitudes of DDG{?u values for

N30K, H21K, and D16K are significantly smaller than

the magnitudes of the corresponding DDGf?{ values.

In the conventional perspective of F-value analysis,6,7

the latter finding suggests that the interactions involving

these residue positions are far weaker in the transition

state than in the native state of their respective single-

mutant variants. By comparison, the corresponding

interactions involving residue 46 appear to be signifi-

cantly stronger. These observations will be discussed

below.

Interestingly, the experimental DDG{?u value of Fyn5

(21.27 � 0.04 kcal mol21) is very similar to the sum of

the DDG{?u values of its constituent five single mutants

(21.23 � 0.09 kcal mol21), suggesting that the effects of

electrostatics on the stability of the folding transition

state is additive for Fyn SH3. This is in sharp contrast

with the trend seen in unfolding where the experimental

DDGf?{ value of Fyn5 (20.43 � 0.17 kcal mol21) is

quite different from the sum of the DDGf?{ values of the

constituent single mutants (21.26 � 0.18 kcal mol21).

To better understand the physical basis of the experi-

mental kinetic effects of mutations involving charged res-

idues, we developed a computational model for the fold-

ing transition states of Fyn SH3 to explore the interplay

between charge–charge interactions and explicit-chain

dynamics.29 As detailed above in Materials and Methods,

the construction of our model follows the framework we

introduced previously to treat nonnative hydrophobic

interactions,25 namely by augmenting a background

native-centric potential30 with sequence dependent non-

native terms.25,29,38 Conceptually, our approach is also

similar to the methods employed in Refs. 11, 34, and 46

to address electrostatic effects in protein folding and

interactions.

The free energy profiles simulated using Ee for the WT

and the six variants of Fyn SH3 we studied are shown in

Figure 4(A). As an example of the energy terms for elec-

trostatic effects used in our simulation, Figure 4(B)

depicts the potential for nonnative charge–charge interac-

tion in our Ee and EeMR models. As in Ref. 25, we limit

the application of our model to the rationalization of

Table IICharge–Charge Interaction Energies From the Tanford–Kirkwood Analysis

E11 R13 E15 D16 D17 H21 K22 E24 K25 E33 D35 E38 R40 E46 D59 C-tr Sum

E11K 0.0 0.0 0.0 0.0 20.2 0.0 0.0 20.1 0.0 0.0 0.0 0.0 0.0 20.1 0.0 0.0 20.4D16K 0.0 0.0 20.1 0.0 20.4 0.0 20.1 0.0 0.0 20.1 20.1 20.2 0.0 0.0 0.0 0.0 21.1H21K 0.0 20.1 0.0 0.0 20.1 0.0 0.0 20.2 0.1 0.0 20.1 0.0 0.0 0.0 0.0 0.0 20.7N30K 0.0 20.1 0.0 0.0 20.1 0.0 0.0 0.0 0.1 20.2 20.1 20.2 0.0 0.0 20.1 0.0 20.5E46K 20.1 0.0 20.1 20.1 20.2 0.0 0.0 20.4 0.1 0.0 0.0 20.1 0.1 0.0 0.0 0.0 20.9

For each of the five single-mutant Fyn SH3 variants, the DDGqq values for the interactions between the lysine (K) residue at the mutated sites and other charged resi-

dues in the protein were evaluated using the folded structure of Fyn SH3 as described previously4. Here we define DDGqq 5 DGqqMUT 2 DGqq

WT, where DGqqMUT and

DGqqWT are, respectively, the predicted contribution of charge–charge interaction to the free energy of folding in a given mutant (MUT) and in the WT. A negative

DDGqq value means that the interaction is more favorable in MUT than in WT. Significant decreases in interaction energy (DDGqq � 20.2 kcal mol21) are shown in

bold type. The residue pairs thus highlighted are predicted to have significantly stronger attractive interactions in MUT than in WT because of electrostatic effects. Data

for a given single-mutant variant are tabulated in the same row. The total change in charge–charge interactions predicted by the Tanford–Kirkwood analysis for the vari-

ant is equal to the sum of DDGqq values for all interacting residue pairs along the row and is listed as the last entry on the right.

Figure 3Equilibrium and kinetic effects of charge-changing mutations in Fyn

SH3. (A) Scatter plot showing the correspondence between the change

in native stability determined experimentally by stopped-flow

measurements (DDGf?u; Table I) and the predicted change in the free

energy of electrostatic contacts from Tanford–Kirkwood modeling

(DDGqq; Table II) for the five single-mutant variants; r is Pearson

correlation coefficient. (B) Scatter plot of the change in folding barrier

height (DDG{?u) versus the change in negative unfolding barrier height

(DDGf?{) caused by the same charge-changing substitutions in the five

single-mutant variants (data from Table I).

Electrostatic Interactions in Folding

PROTEINS 863

folding rates.29 This is because folding rates are deter-

mined largely by the properties of disordered or partially

disordered conformations in the unfolded and transition

states that are readily captured by coarse-grained models.

In contrast, as discussed in Ref. 25, our model is likely

insufficient for rationalizing mutational effects on

unfolding rate and native stability because it does not

fully account for the cooperativity29,47 that arises from

specific packing and other subtle effects in the folded

state. From each model protein’s free energy profile, the

barrier to folding, DG{/kBT, is taken to be the –lnP(Q)

value at the transition-state peak minus that at the

unfolded-state minimum. Previous analyses have demon-

strated that DG{/kBT values are well correlated with the

negative natural logarithm of folding rates (with slope �1) simulated near the transition midpoints of similar

models.31,32

Consistent with experiment, Figure 4(A) indicates that

all six variants have lower simulated DG{/kBT values

(ranging from 1.69 to 3.16) than that of WT (DG{/kBT

5 3.27). The theoretical DG{/kBT values for Ee correlate

well with the negative logarithm of experimental folding

rates (black squares in Figure 4(C), Pearson coefficient 5

0.98, slope of the fitted line 5 0.69). No significant out-

lier is observed for this set of data. However, because the

correlation is less than perfect, the experimentally

observed additivity of DDG{?u was not reproduced by

this set of simulated results (the sum of Ee-simulated

DDG{?u values for the five single mutants and the corre-

sponding simulated DDG{?u value for Fyn5 are –2.10

kBT and –1.58 kBT, respectively). Nonetheless, the agree-

ment between the general trends observed in theory and

experiment suggests that instructive insights can be

gleaned from further analyses of the simulation data,

especially with regard to how charge-changing mutations

may affect the conformational distribution in the folding

transition state.

The theoretical DG{/kBT values in the Ee(His1) and Ee

MR

models are shown in Figure 4(C) as controls. As in the

case of Ee, the alternate potentials Ee(His1) and Ee

MR did

not reproduce the experimentally observed additivity of

DDG{?u. Nevertheless, both sets of DG{/kBT values corre-

late reasonably well with the negative logarithm of experi-

mental folding rates (Pearson coefficient 5 0.93 and 0.92,

respectively, for Ee(His1) and Ee

MR), though the correlation

is not as good as that for Ee (Pearson coefficient 5 0.98).

Figure 4Modeling the kinetic effects of native and nonnative electrostatic interactions on the folding of Fyn SH3 and its variants. (A) Simulated free energy

profiles of WT and mutants of Fyn SH3 are shown respectively by curves with the different colors indicated. The profiles were obtained by

extensive sampling using the coarse-grained chain model with the Ee potential described in Materials and Methods. All profiles were simulated at

the same model temperature, which was chosen to be close to the transition midpoints of the single mutants. As in our previous studies,25,31 P(Q)

is normalized chain population as a function of fractional native contact Q; thus 2lnP(Q) is free energy in units of kBT. The light gray area

indicates the range of Q values of the transition-state conformations that were sampled to compute the contact probability maps in Figure 5. (B) A

term in the charge–charge interaction potential in our model. Shown here as an example is the term for nonnative electrostatic interaction (wherein

the closest nonhydrogen atoms of the two residues are >6 A apart in the PDB) in the Ee potential (solid curves) and the corresponding term in the

alternate medium-range EeMR potential (dashed curves) that we used as a control. The Ca–Ca separation rij is in units of A. The interaction

energies between opposite charges and between like charges are provided by the red and blue curves, respectively. The spatial range of the medium-range Ee

MR potential is shorter than that of the Ee potential (as rij increases in the present figure, the dashed curves approach zero at a faster rate

than the solid curves); but the spatial range of EeMR is longer than the EHP potential we used previously to model nonnative hydrophobic

interactions25 (cf., the curve with energy minimum at �5 A in Fig. S2 of Ref. 38) (C) Scatter plot of simulated barrier heights [DG{/kBT (sim)]

versus negative logarithms of the experimental folding rates [2ln kf (exp)] for WT, Fyn5 (as marked), and the five single mutants (not marked).

Results for the Ee potential are depicted by the black squares. The line is the least-squares fit to this set of data for Ee only. Results for the alternate

potentials Ee(His1) (magenta circles) and Ee

MR (open circles) are included for comparison; but no fitted line is shown for the Ee(His1) or Ee

MR data.

A. Zarrine-Afsar et al.

864 PROTEINS

Taken together, Figure 4(C) shows a robust trend of agree-

ment between theory and experiment despite the varia-

tions in the modeling setup we considered. It also suggests

that the Ee model is superior, which is not unexpected

because of the physical considerations that went into its

construction. By comparison, the simulation results from

the Ee(His1) and Ee

MR model exhibit increased scatter with

the experimental data. Notably, the medium-range EeMR

model failed to capture the much stronger kinetic effect

exhibited by the E46K single mutant compared to other

single mutants. Apparently, charge–charge interactions

that have longer spatial ranges similar to that of Ee are

necessary to rationalize the large increase in folding rate of

the E46K single mutant relative to WT.

Building on the agreement between experimental fold-

ing rates and simulation results in the Ee model [Fig.

4(C)], we proceed to analyze the simulated contact prob-

abilities of the transition state ({) of WT and the opti-

mized Fyn5 SH3 domain variant (Fig. 5) in the Ee model.

Similar to the results from our previous study of Fyn

SH3,25 the distributions of contacts in the transition

state ensembles of WT [Fig. 5(A)] and Fyn5 [Fig. 5(B)]

are essentially diffused versions of the native contacts

[Fig. 5(C)] showing relatively stronger favorable interac-

tions among b-strands b through d than those between a

and e [see Figs. 1(A) and 5(C) for the b-strand posi-

tions]. Nonnative charge–charge contacts are present (see

below) but their probabilities are low, and thus not con-

Figure 5Effects of charge-changing mutations on the folding transition state in the coarse-grained model using the Ee potential. Simulated contact

probability maps of the folding transition state (TS) ensembles of (A) WT Fyn SH3 and (B) the designed E46K-E11K-D16K-H21K-N30K (Fyn5)

variant were computed as described in Materials and Methods [see shaded area in Fig. 4(A)]. In each map, residue numbers i and j are represented

by the vertical and horizontal axes; native contacts as well as contacts between i,i13 pairs and between j > i 1 3 pairs with 4.5 A < dmin � 6 A

that are involved in charge–charge interactions in WT or Fyn5 (or in both) are shown in the lower-right (below the i 5 j diagonal), whereas

nonnative electrostatic contacts are shown in the upper-left (above the i 5 j diagonal). The probability of contact between any pair of residues (i,j)

is color coded according to the scale on the right of each map. (C) The contact map of the PDB native structure (N). Native contacts are shown in

black, the i,i13 pairs and j > i 1 3 pairs with 4.5 A < dmin � 6 A considered in (A) and (B) are shown in gray. The arrows mark the positions of

the five native b-strands a–e [see Fig. 1(A)]. (D) The values shown in the difference map Fyn5 – WT are the TS contact probabilities of Fyn5 in

(B) minus the corresponding TS contact probabilities of WT in (A). Negative values are possible in (D). The sites of charge-changing substitutions

in Fyn5 are marked by dotted lines. Note that the color code and the range of the scale in (D) are different from those in (A) and (B).

Electrostatic Interactions in Folding

PROTEINS 865

spicuous in the color scales for Figure 5(A,B). The differ-

ence contact map in Figure 5(D) offers insights into how

folding was speeded up by the charge-changing substitu-

tions (all mutated to K) at positions 11, 16, 21, 30, and

46 in Fyn5. Figure 5(D) shows that the substitutions

have broad effects over many contacts, not limited to

those involving the substituted residues (marked by dot-

ted lines). Some of the changes in the native contacts in

the lower-right of Figure 5(D) are stabilizing (increased

probabilities indicated by reddish hues) whereas others

are destabilizing (decreased probabilities indicated by

bluish hues). For example, the interactions between b-strands b and c are somewhat strengthened whereas those

between b-strands c and d are slightly weakened in the

Fyn5 transition state.

Figure 5(D) shows that the contacts in the Fyn5 transi-

tion state involving K11, K16, K21, and K30 are generally

enhanced relative to those involving the E11, D16, H21,

and N30 residues in the WT. For K46 in Fyn5, the transi-

tion-state contacts it forms with residues around position

40 are weaker, but those it forms with residues around

position 18 are stronger, than the corresponding contacts

involving E46 in the WT. Interestingly, a cluster of con-

tacts including K16-G48, K16-Y49, L18-K46, L18-T47,

L18-Y49 are strongly enhanced, apparently because of the

D16K and E46K substitutions. Although most of these

effects are seen among native contacts, the Fyn5 substitu-

tions also lead to changes in nonnative charge–charge

interactions [upper-left of Fig. 5(D)]: Substituting E with

K at position 11 leads to increased contact probability

between the opposite charges K11 and E15; substituting

N with K at position 30 brings K30 close to D35 as an

electrostatic attraction is created. The charge reversal sub-

stitution E46K brings K46 into closer contact with E15,

K16, and D17 because of the K46-E15 and K46-D17

attraction, even though the Ca positions of the residue

15, 16, and 17 are well separated from that of K46 in the

native structure, by 11.3, 11.2, and 10.6 A, respectively.

Notably, the H21K substitution brings K11 close to K21

despite their repulsive interactions because of a general

enhancement of favorable native interactions in the loop

connecting b-strands a and b. Among the six residue

pairs i,j satisfying j � i 1 3 that are predicted by the

equilibrium Tanford–Kirkwood analysis to have signifi-

cantly enhanced favorable charge–charge interactions in

Table II, five pairs are seen in Figure 5(D) to have

increased transition state contact probabilities in Fyn5

relative to WT (K11-D17, K16-E38, K21-E24, E24-K46,

K30-E33). However, even though K30-E38 is favored in

Table II, Figure 5(D) shows that the transition state con-

tact probability of K30-E38 is essentially the same for

Fyn5 and WT. Viewed as a whole, the simulation results

in Figure 5 show that the charge-changing substitutions

in Fyn5 affects many native and nonnative contacts in

the transition state. Their effect is not restricted to the

specific electrostatic interactions in the native structure,

nor is the impact of the charge-changing mutations on

the folding transition state a simple reflection of the

effects of the same mutations on the native state.

DISCUSSION

The goal of this study is to investigate the kinetic con-

sequences of native state optimization of surface-exposed

electrostatic interactions. Tanford–Kirkwood modeling

optimizes specific interactions in the folded state (Table

II). As detailed above, contact probability maps of the

folding transition state from our coarse grained chain

model suggest a variety of changes in native as well as

nonnative interactions arising as a consequence of native-

state optimization. As far as coarse-grained model devel-

opment is concerned, it is encouraging that a simple Cachain model without an explicit representation of side

chains is able to capture most of the effects of sequence-

dependent electrostatics on folding kinetics (Fig. 4).

Future improvements of our models will need to incor-

porate more structural and energetic details such as side-

chain packing and hydrogen bonding.

To gain further insight and to assess the robustness of

the present theoretical predictions, Figure 6 provides the

Fyn5 2 WT transition-state contact probabilities maps in

the alternate Ee(His1) and Ee

MR models. Comparing Fig-

ure 6 with Figure 5(D) for the Ee model shows quite

clearly that the mutational effects on the transition state

predicted by the three models are very similar. This is in

keeping with the robustness observed above in Figure

4(C). Nevertheless, several differences between Figures 5

(D) and 6 are noteworthy. Contrasting Figure 5(D) with

6(A) indicates that when the charge at position 21 is

considered to be unchanged at 11 [Fig. 6(A)] rather

than being changed from 0 to 11 [Fig. 5(D)], the Fyn5

mutations exhibits slightly less stabilization of the inter-

actions between b-strands a and b, but slightly more sta-

bilization of the interactions between b-strands b and c

and between b-strands c and d. This trend suggests that

a positive charge at position 21 tends to stabilize the

interactions around the RT-Src loop region that contains

a series of negatively charged residues but, for some rea-

son yet to be delineated, destabilize the interactions

around the N-Src loop and distal loop regions in the

transition state [see Fig. 1(A)]. Contrasting Figure 5(D)

with 6(B) shows a significantly stronger stabilization of

all interactions involving residues 16–20 in the Ee model

than in the shorter-range EeMR model, especially the

interactions of these residues with those sequentially close

to K46 (residues 46–51). The enhanced strength of these

interactions under a longer-range potential most likely

underlies the better performance of the Ee model in

rationalizing the folding rate of the E46K single mutant

[Fig. 4(C)]. K46 is also predicted by the present theoreti-

cal treatments to engage in electrostatic interaction with

A. Zarrine-Afsar et al.

866 PROTEINS

E24 in both the folded and the folding transition states

(Table II, Figs. 5 and 6). E24, like the 46 position itself,

is known experimentally to be highly structured in the

folding transition state of SH3 domains.17–19

Although parts of the folding transition state of Fyn

SH3 are highly structured, the transition state as a whole

is much less ordered than the folded state. The additivity

seen in the transition state stability of Fyn5 in conjunc-

tion with a lack of additivity in the folded state stability

suggest that the relatively simple dependence of folding

rate on electrostatic interactions is a result of the partial

conformational disorder in the transition state. Similarly,

the effects on folding rates caused by Fyn SH3 domain

amino acid substitutions have been shown to correlate

strongly to predicted changes in local structure propen-

sity while relatively poorer correlations were observed

with unfolding rates.26 In the same vein, a previous

study has shown that omitting data for residues with

many tertiary contacts from statistical analysis can signif-

icantly improve the correlation between unfolding rates

and local structure propensity.26

Prediction of the effects of charge-changing mutations

on native stability and unfolding rates are beyond the

scope of the present coarse-grained chain model (see

above); but it is instructive to note that the N30K single

mutant is the only variant listed in Table I that exhibits a

significant destabilization of the folded state (DDGf?u 50.26 kcal mol21) and a significant increase in unfolding

rate [DDGf?{ 5 0.45 kcal mol21; see also Fig. 3(B)]. The

N30 residue engages in a complex network of hydrogen

bonds with the S31 and S32 residues in the native state

of Fyn SH3.39 These hydrogen bonds can potentially sta-

bilize a turn in the N-Src loop region of the WT protein.

If this picture applies, the K substitution at position 30

may eliminate these hydrogen bonds and lead to an over-

all destabilization of the folded structure.

In addition to the specific interactions discussed above,

our coarse-grained model indicates that the charge-

changing substitutions in Fyn5 have broad effects on

many contacts in the transition state (Fig. 5), and that

these effects are sufficient to account for the general

trend exhibited by the folding rates of the Fyn SH3 var-

iants [Fig. 4(C)]. The plasticity arising from the partially

disordered nature of the dynamic transition state ensem-

ble may make it conducive to formation of less specific,

and potentially nonnative, contacts. Nonspecific charge–

charge interactions were the basis of an early mean-field

model of electrostatic effects on protein stability.48 The

importance of such interactions has been demonstrated

experimentally for the molten globule states of horse

cytochrome c49 as well as underscored by a theoretical

rationalization of unfolded state behaviors using a gaus-

sian-chain model that incorporated residual charge–

charge interactions.36 Recently, nonspecific electrostatics

were found to be critical also for understanding the

interactions11,50 and conformational dimension51–53 of

intrinsically disordered proteins. In protein folding, the

interplay between the driving forces for native topology

and sequence-dependent nonnative interactions can be

subtle.29,38,54 For Fyn5, our simulation results suggest

that charge-changing substitutions affect mostly the

native interactions in the transition state ensemble; but

weak nonnative charge–charge contacts are also present

in the ensemble. Future experiments could be aimed at

testing the predicted transition-state contact pattern of

Fyn5.

The observation that the designed variant folded eight-

fold faster than the WT protein contradicts previously

Figure 6Effects of variation in the model electrostatic potentials on the predicted transition state (TS) ensemble. WT 2 Fyn5 difference map of TS contact

probabilities were determined as in Figure 5(D) (for Ee) except results shown here are for the alternate potentials Ee(His1) (A) and Ee

MR (B).

Electrostatic Interactions in Folding

PROTEINS 867

held notions that improving the native state design of

two-state proteins inevitably results in dramatic altera-

tions to unfolding kinetics.15,55 The consequence of

improving native state design in the Fyn SH3 domain is

primarily a neutralization or reversal of the unfavorable

charge–charge repulsions that exist in the WT protein. In

addition to the stability conferred through formation of

favorable charge–charge interactions, the dramatic accel-

eration in the folding rate of the designed sequence may

in part be rationalized by taking into consideration the

changes in the global charge density during folding.

Because the unfolded state occupies a much larger vol-

ume compared to either the folded or the transition

states, folding from the unfolded state to the transition

state is accompanied by a nonspecific increase in the

charge density. For the Fyn SH3 domain, the folding

transition state structure is believed to be �70% as com-

pact as the folded state.56 The computational evaluation

via Tanford–Kirkwood modeling of the electrostatic inter-

actions suggests that in the folded state of the WT Fyn

SH3 domain, the charge–charge interactions are highly

repulsive and the domain possesses a net negative charge

under the experimental conditions.4 Therefore, transition

state formation of the WT Fyn SH3 domain entails an

increase in the negative charge density.

Consistent with this consideration, Debye–Huckle

screening of unfavorable electrostatic repulsions has been

shown to result in faster folding.27,57 Therefore, it is

likely that the eight-fold increase in the folding rate of

the optimized Fyn SH3 domain arises in large measure

from a favorable nonspecific charge-density effect associ-

ated with the removal or reduction of electrostatic

repulsion from its surface. By its physical nature, salt

screens electrostatic interactions irrespective of whether

they are specific or nonspecific. Therefore, we expect

salt to modulate the stability and folding kinetics of the

optimized Fyn5 variant even if favorable nonspecific

charge–charge interactions not present in the WT pro-

tein were introduced. However, weak nonspecific inter-

actions that can involve charges that are somewhat far

apart are likely to be screened at a lower salt concentra-

tion compared to specific interactions between charges

that are in close proximity. Inasmuch as these assump-

tions are valid, the extent to which salt modulates the

stability and folding kinetics of Fyn5 could be different

from that for the WT protein. This possibility remains

to be investigated. In a recent study of E. coli thiore-

doxin, extensive examination of the kinetic and thermo-

dynamic effects of NaCl has determined a F-value for

NaCl that describes the impact of salt on protein

unfolding kinetics. The FNaCl-value was found to be

close to unity at low salt, regardless of the site mutated,

but decreases at higher concentrations of salt.13 It will

be instructive to investigate whether this trend is related

to the screening of specific versus nonspecific electro-

static interactions.

CONCLUDING REMARKS

In summary, we have demonstrated that the kinetic

consequence of native state optimization of surface elec-

trostatic interactions could be complex, involving specific

native or nonnative interactions as well as nonspecific

mechanisms. As shown by the behaviors of the Fyn SH3

variants, electrostatic interactions can affect folding

kinetics through a charge density effect, which allows

mutations that reduce the net charge of a protein to sta-

bilize the partially disordered folding transition state.

Although the prevalence of this nonspecific electrostatic

effect in other proteins remains to be ascertained, the ex-

istence of such an effect suggests that the impact of any

charge-removing substitution in a protein engineering

experiment may not be limited to the removal of a spe-

cific set of native interactions. This consideration should

be taken into account in the interpretation of F-valuesof charged residues. A further complicating factor is that

conformational compaction during folding can cause a

significant change in charge density. In view of the pres-

ent findings, extra caution is warranted when F-valueanalysis is applied to study the kinetic roles of charged

residues.

ACKNOWLEDGMENTS

The authors thank Dr. Walid A. Houry of University

of Toronto for the use of his stopped-flow device. Z.Z.

and H.S.C. are grateful for the computing resources pro-

vided by SciNet of Compute Canada.

REFERENCES

1. Makhatadze GI, Loladze VV, Ermolenko DN, Chen X, Thomas ST.

Contribution of surface salt bridges to protein stability: guidelines

for protein engineering. J Mol Biol 2003;327:1135–1148.

2. Loladze VV, Ibarra-Molero B, Sanchez-Ruiz JM, Makhatadze GI.

Engineering a thermostable protein via optimization of charge-

charge interactions on the protein surface. Biochemistry 1999;38:

16419–16423.

3. Strickler SS, Gribenko AV, Keiffer TR, Tomlinson J, Reihle T,

Loladze VV, Makhatadze GI. Protein stability and surface electro-

statics: a charged relationship. Biochemistry 2006;45:2761–2766.

4. Schweiker KL, Zarrine-Afsar A, Davidson AR, Makhatadze GI.

Computational design of the Fyn SH3 domain with increased sta-

bility through optimization of surface charge-charge interactions.

Protein Sci 2007;16:2694–2702.

5. Halskau O, Jr., Perez-Jimenez R, Ibarra-Molero B, Underhaug J,

Munoz V, Martinez A, Sanchez-Ruiz JM. Large-scale modulation of

thermodynamic protein folding barriers linked to electrostatics.

Proc Natl Acad Sci USA 2008;105:8625–8630.

6. Matouschek A, Kellis JT, Jr., Serrano L, Fersht AR. Mapping the

transition state and pathway of protein folding by protein engineer-

ing. Nature 1989;340:122–126.

7. Matouschek A, Fersht AR. Protein engineering in analysis of protein

folding pathways and stability. Methods Enzymol 1991;202:82–112.

8. Cho JH, Raleigh DP. Mutational analysis demonstrates that specific

electrostatic interactions can play a key role in the denatured state

ensemble of proteins. J Mol Biol 2005;353:174–185.

A. Zarrine-Afsar et al.

868 PROTEINS

9. Cho JH, Raleigh DP. Electrostatic interactions in the denatured

state and in the transition state for protein folding: effects of dena-

tured state interactions on the analysis of transition state structure.

J Mol Biol 2006;359:1437–1446.

10. Trefethen JM, Pace CN, Scholtz JM, Brems DN. Charge-charge

interactions in the denatured state influence the folding kinetics of

ribonuclease Sa. Protein Sci 2005;14:1934–1938.

11. Weinkam P, Pletneva EV, Gray HB, Winkler JR, Wolynes PG. Elec-

trostatic effects on funneled landscapes and structural diversity in

denatured protein ensembles. Proc Natl Acad Sci USA

2009;106:1796–1801.

12. Oliveberg M, Fersht AR. Formation of electrostatic interactions on

the protein-folding pathway. Biochemistry 1996;35:2726–2737.

13. Pey AL, Rodriguez-Larrea D, Bomke S, Dammers S, Godoy-Ruiz R,

Garcia-Mira MM, Sanchez-Ruiz JM. Engineering proteins with tun-

able thermodynamic and kinetic stabilities. Proteins 2008;71:

165–174.

14. Sanchez-Ruiz JM. Protein kinetic stability. Biophys Chem

2010;148:1–15.

15. Jackson SE. How do small single-domain proteins fold? Fold Des

1998;3:R81–R91.

16. Ollerenshaw JE, Kaya H, Chan HS, Kay LE. Sparsely populated

folding intermediates of the Fyn SH3 domain: matching native-cen-

tric essential dynamics and experiment. Proc Natl Acad Sci USA

2004;101:14748–14753.

17. Martinez JC, Serrano L. The folding transition state between SH3

domains is conformationally restricted and evolutionarily con-

served. Nat Struct Biol 1999;6:1010–1016.

18. Grantcharova VP, Riddle DS, Baker D. Long-range order in the src

SH3 folding transition state. Proc Natl Acad Sci USA 2000;97:7084–

7089.

19. Riddle DS, Grantcharova VP, Santiago JV, Alm E, Ruczinski I, Baker

D. Experiment and theory highlight role of native state topology in

SH3 folding. Nat Struct Biol 1999;6:1016–1024.

20. Grantcharova VP, Baker D. Folding dynamics of the src SH3 do-

main. Biochemistry 1997;36:15685–15692.

21. Tsai J, Levitt M, Baker D. Hierarchy of structure loss in MD simula-

tions of src SH3 domain unfolding. J Mol Biol 1999;291:215–225.

22. Dokholyan NV, Li L, Ding F, Shakhnovich EI. Topological determi-

nants of protein folding. Proc Natl Acad Sci USA 2002;99:8637–

8641.

23. Klimov DK, Thirumalai D. Stiffness of the distal loop restricts the

structural heterogeneity of the transition state ensemble in SH3

domains. J Mol Biol 2002;317:721–737.

24. Lam AR, Borreguero JM, Ding F, Dokholyan NV, Buldyrev SV,

Stanley HE, Shakhnovich E. Parallel folding pathways in the SH3

domain protein. J Mol Biol 2007;373:1348–1360.

25. Zarrine-Afsar A, Wallin S, Neculai AM, Neudecker P, Howell PL,

Davidson AR, Chan HS. Theoretical and experimental demonstra-

tion of the importance of specific nonnative interactions in protein

folding. Proc Natl Acad Sci USA 2008;105:9999–10004.

26. Zarrine-Afsar A, Dahesh S, Davidson AR. Protein folding kinetics

provides a context-independent assessment of beta-strand propen-

sity in the Fyn SH3 domain. J Mol Biol 2007;373:764–774.

27. Zarrine-Afsar A, Mittermaier A, Kay LE, Davidson AR. Protein sta-

bilization by specific binding of guanidinium to a functional argini-

ne-binding surface on an SH3 domain. Protein Sci 2006;15:

162–170.

28. Maxwell KL, Davidson AR. Mutagenesis of a buried polar interac-

tion in an SH3 domain: sequence conservation provides the best

prediction of stability effects. Biochemistry 1998;37:16172–16182.

29. Chan HS, Zhang Z, Wallin S, Liu Z. Cooperativity, local-nonlocal

coupling, and nonnative interactions: principles of protein folding

from coarse-grained models. Annu Rev Phys Chem 2011;62:301–326.

30. Clementi C, Nymeyer H, Onuchic JN. Topological and energetic

factors: what determines the structural details of the transition state

ensemble and ‘‘en-route’’ intermediates for protein folding? An

investigation for small globular proteins. J Mol Biol 2000;298:

937–953.

31. Wallin S, Chan HS. Conformational entropic barriers in topology-

dependent protein folding: perspectives from a simple native-centric

polymer model. J Phys Condensed Matter 2006;18:S307–S328; Cor-

rigendum: 2009;21:329801.

32. Ferguson A, Liu Z, Chan HS. Desolvation barrier effects are a likely

contributor to the remarkable diversity in the folding rates of small

proteins. J Mol Biol 2009;389:619–636; Corrigendum: 2010;401:153.

33. Stryer L. Biochemistry. San Francisco: W. H. Freeman; 1981. pp 40,

80–82.

34. Azia A, Levy Y. Nonnative electrostatic interactions can modulate

protein folding: molecular dynamics with a grain of salt. J Mol Biol

2009;393:527–542.

35. Mazur J, Jernigan RL. Distance-dependent dielectric constants and

their application to double-helical DNA. Biopolymers 1991;31:

1615–1629.

36. Zhou HX. A Gaussian-chain model for treating residual charge–

charge interactions in the unfolded state of proteins. Proc Natl

Acad Sci USA 2002;99:3569–3574.

37. Vuzman D, Azia A, Levy Y. Searching DNA via a ‘‘Monkey Bar’’

mechanism: the significance of disordered tails. J Mol Biol

2010;396:674–684.

38. Zhang Z, Chan HS. Competition between native topology and non-

native interactions in simple and complex folding kinetics of natu-

ral and designed proteins. Proc Natl Acad Sci USA 2010;107:

2920–2925.

39. Noble ME, Musacchio A, Saraste M, Courtneidge SA, Wierenga RK.

Crystal structure of the SH3 domain in human Fyn: comparison of

the three-dimensional structures of SH3 domains in tyrosine

kinases and spectrin. EMBO J 1993;12:2617–2624.

40. Baker D. A surprising simplicity to protein folding. Nature

2000;405:39–42.

41. Barrick D. What have we learned from the studies of two-state fold-

ers, and what are the unanswered questions about two-state protein

folding? Phys Biol 2009;6:015001.

42. Sosnick TR, Barrick D. The folding of single domain proteins—

have we reached a consensus? Curr Opin Struct Biol 2011;21:12–24.

43. Chan HS, Shimizu S, Kaya H. Cooperativity principles in protein

folding. Methods Enzymol 2004;380:350–379.

44. Myers JK, Pace CN, Scholtz JM. Denaturant m values and heat

capacity changes: relation to changes in accessible surface areas of

protein unfolding. Protein Sci 1995;4:2138–2148.

45. Ibarra-Molero B, Loladze VV, Makhatadze GI, Sanchez-Ruiz JM.

Thermal versus guanidine-induced unfolding of ubiquitin. An anal-

ysis in terms of the contributions from charge–charge interactions

to protein stability. Biochemistry 1999;38:8138–8149.

46. Levy Y, Onuchic JN, Wolynes PG. Fly-casting in protein-DNA bind-

ing: frustration between protein folding and electrostatics facilitates

target recognition. J Am Chem Soc 2007;129:738–739.

47. Kaya H, Chan HS. Simple two-state protein folding kinetics

requires near-levinthal thermodynamic cooperativity. Proteins 2003;

52:510–523.

48. Stigter D, Alonso DO, Dill KA. Protein stability: electrostatics and com-

pact denatured states. Proc Natl Acad Sci USA 1991;88:4176–4180.

49. Hagihara Y, Tan Y, Goto Y. Comparison of the conformational sta-

bility of the molten globule and native states of horse cytochrome

c. Effects of acetylation, heat, urea and guanidine-hydrochloride. J

Mol Biol 1994;237:336–348.

50. Borg M, Mittag T, Pawson T, Tyers M, Forman-Kay JD, Chan HS.

Polyelectrostatic interactions of disordered ligands suggest a physical

basis for ultrasensitivity. Proc Natl Acad Sci USA 2007;104:9650–

9655.

51. Marsh JA, Forman-Kay JD. Sequence determinants of compaction

in intrinsically disordered proteins. Biophys J 2010;98:2383–2390.

52. Mao AH, Crick SL, Vitalis A, Chicoine CL, Pappu RV. Net

charge per residue modulates conformational ensembles of

Electrostatic Interactions in Folding

PROTEINS 869

intrinsically disordered proteins. Proc Natl Acad Sci USA 2010;

107:8183–8188.

53. Muller-Spath S, Soranno A, Hirschfeld V, Hofmann H, Ruegger S,

Reymond L, Nettels D, Schuler B. From the cover: charge interac-

tions can dominate the dimensions of intrinsically disordered pro-

teins. Proc Natl Acad Sci USA 2010;107:14609–14614.

54. Enciso M, Rey A. Improvement of structure-based potentials for

protein folding by native and nonnative hydrogen bonds. Biophys J

2011;101:1474–1482.

55. Zarrine-Afsar A, Davidson AR. The analysis of protein folding ki-

netic data produced in protein engineering experiments. Methods

2004;34:41–50.

56. Northey JG, Di Nardo AA, Davidson AR. Hydrophobic core pack-

ing in the SH3 domain folding transition state. Nat Struct Biol

2002;9:126–130.

57. de Los Rios MA, Plaxco KW. Apparent Debye-Huckel electrostatic

effects in the folding of a simple, single domain protein. Biochemis-

try 2005;44:1243–1250.

A. Zarrine-Afsar et al.

870 PROTEINS