Dominant Forces in Protein Folding_Biochem_1990

8/14/2019 Dominant Forces in Protein Folding_Biochem_1990

1/23

0 Copyright 1990 by the American Chemical Society Volume 29, Number 31

Perspectives in Biochemistry

August 7, 1990

Dominant Forces in Protein FoldingKen A. Dill

Department of Pharmaceutical Chem istry, University of Calif ornia , San Francisco, Californ ia 941 43 - 1204Received April 3, 1990; Revised Manuscript Received M ay 2, 1990

T e urpose of this review is to assess the nat ure a nd ma g-nitudes of the dom inan t forces in protein folding. Sinc eproteins are only marginally stable at room temperature, notype of molecular interaction is unimportant, and even smallinteractions can contribute significantly (positively or nega-tively) to stability (Alber, 1989a,b; Matthews, 1987a,b).However, the present review aim s to identify only the largestforces that lead to the structural features of globular proteins:their extraordinary compactness, their core of nonpolar resi-dues, and their considerable amou nts of internal architecture.This review explores contributions to the free energy offolding arising from electrostatics (classical charge repulsionsand ion pairing), hydrogen-bonding and van der Waals in-teractions, intrinsic propensities, and hydrophobic interactions.An earlier review by Kauzmann (1959) introduced the im-port anc e of hydrophobic interactions. His insights wereparticularly remarkable considering that he did not have thebenefit of known protein structures, model studies, high-res-olution calorimetry, mutational methods, or force-field orstatis tical mechanical results. Th e present review aims toprovide a reassessment of the factors important for folding inlight of curr ent knowledge. Also considered here are theopposing forces, conformationa l entropy and electrostatics.The process of protein folding has been known for about60 years. In 1902, Emil Fischer and Fr anz Hofm eister in-dependently concluded that proteins were chains of covalentlylinked amino acids (Haschemey er & Haschemeyer, 1973) butdeeper understanding of protein stru cture and conformationalchange was hindered because of the difficulty in findingconditions for solubilization. Chick and Ma rtin (191 1) werethe first to discover the process of denaturation and to dis-tinguish it from the process of agg rega tion. By 1925, thedenaturation process was considered to be either hydrolysisof the peptide bond (Wu & Wu, 1925; Anson & Mirsky,1925) or dehydration of the protein (Robertson , 1918). Theview that protein denaturation was an unfolding process was

I The f ree energy AGunfold= Gdcnaturcd Gnativcs typically 5-20kcal/mol of protein, less than (l/lO)kT per residue, where k = Boltz-manns constant and T is tempera ture (Pace, 1975; Privalov, 1979).0006-2960/90/042 9-7 133$02.50/0

first put forward by Wu (1929,1931). H e proposed that nativeproteins involve regular repeated patterns of folding of thechain into a three-dimensional network somewhat resemblinga crystal, held togethe r by noncovalent linkages. Denatu rationis the breaking up of these labile linkages. Inste ad of beingcompa ct, the protein now becomes a diffuse structure. Thesurface is altered and the interior of th e molecule is exposed(W u, 1929). Dena turation is disorganization of the naturalprotein molecule, the change from the regular ar rang eme ntof a rigid structure to the irregular, diffuse arrangement ofthe flexible open chain (Wu, 1931).Before discussing forces, we ask: Is the native structurethermodynamically stable (the thermodynamic hypothesis;Anfinsen, 1973) or metasta ble, determined, for example, asthe protein leaves the ribosome? To prove thermodyn amicstability, it is sufficient to demonstra te tha t the na tive structureis only a function of state a nd does not depend on the processor initial conditions leading to that state . By definition, sucha state would be at the global minimum of free energy relativeto all other states accessible on that time scale. Experimentsof Anson and Mirsky (1931) and Anson (1945) showed thathemoglobin folding is reversible as evidenced by similaritiesin the following properties of native and renatured protein:solubility, crystallizability, characteristic spectrum, bindingto O 2an d CO , and inaccessibility to trypsin digestion. Th efolding of serum album in and other proteins was shown to besimilarly reversible by these coa rse measures of native struc ture(N eur ath et al., 1944; Anson, 1945; Lumry & Eyring, 1954).It was then demonstrated that denaturation is also thermo-dynamically reversible for some proteins (Eisenberg &Schw ert, 1951; Brandts & Lumry , 1963) an d involved largeconformational changes (Harrington & Schellman, 1956;Schellman & Schellm an, 1958). Recent high-resolution ca-lorimetry experiments show thermodynamic reversibility formany small single-domain globular proteins (Privalov, 1979,1989; Santoro & Bolen, 1988; Bolen & Santoro, 1988; Pace,1975) and also for some multidomain and coiled-coil proteins(Privalov, 1982). Reversibility was tested much more spe-cifically by the experiments of Anfinsen et al. (1973) in whichthe disulfide bonds of bovine pancreatic ribonuclease were

0 1990 American Chemical Society


2/23

7134 Biochemistry, Vol. 29 , No. 31, 1990 Perspectives in Biochemistryscrambled to random distributions of the 105 possiblebinding patterns and reacquisition of native structure andactivity was observed upon renaturation (Haber & Anfinsen,1962; Anfinsen, 1973). Th e ad vantage of m onitoring disulfidebonds is that they are uniquely trappable and identifiable.Similarly, two circularly permuted proteins refold to theiroriginal native states (Luger et al., 1989; Goldenberg &Creighton, 198 3). There fore despite often extrem e difficultiesin the achieveme nt of reversibility, the therm odynam ic hy-pothesis has now been widely established. These expe rimentsdo not necessarily imply rev ersibility is completely general forother cond itions, for other proteins, or even for other parts ofa given protein than those monitored by the given experiment.It is clear that the folding of some proteins can be catalyzedby other assisting proteins, such as polypeptide binding orchaperone proteins (Rothm an, 1989; Ostermann et al., 1989;Ellis, 1988; Anfinsen, 1973). Nevertheless, the existence ofchaperones bears only on the rate that a protein is folded(provided the chapero ne is a true catalyst) and has no bearingon the thermod ynam ic hypothesis, on the natur e of the nativestate (if the native structure is otherwise reversible on theexperimental time scale), or on the driving forces that causeit. Th e present discussion address es only those proteins andconditions for which reversibility holds.

In this discussion of the nature of forces, it is useful todistinguish long-ranged and short-ranged forces, on the onehand, from local and nonlocal forces, on the othe r. Th e dis-tance dependence defines the range: energies that depend ondistance r as r-P are long-ranged if p I (ion-ion and ion-dipole interactions, for example) or short-ranged if p > 3(Lennard-Jones attractions and repulsions, for example). Thisinverse third-power dependence is the natural division becausefor simple pure media the integral that gives the total energyof a system diverges, according to this definition, for long-ranged forces and converges for short-ranged forces (Hill,1960). For polymer chains such as proteins, segment positionin the chain is also imp ortant, in add ition to the rang e of force.Local interactions are those among chain segments that a reconnected neighbors (i, i + l ) , or near neighbors, in the se-quence (see Figure 1). Nonlocal refers to interactionsamong residues that a re significantly apart in the sequence.Local interactions can arise from eithe r long- or short-rangedforces, as can nonlocal interactions.( I ) LONG-RANGEDNTERACTIONS:LECTROSTATICS

Because acids and bases were among the earliest knowndenatu rants of proteins, the folding forces were first assumedto be electrostatic in nature. The signature of electrostaticallydriven processes is a depende nce on pH and /or ionic strength.Whereas the pH determines the total charge on the protein,the salt determines the extent of interaction among thosecharges since salts shield charges. The first quantitative modelof electrostatic interactions in native proteins was proposedby Linderstrom-Lang (1924) (when he was 27 years old!).This work appea red less than 1 year after the Debye-Huckeltheory on which it was based. By treatin g a native proteinas a multivalent impenetrable spherical particle with its netcharge uniformly distributed at the surface, Linderstrom-Langpredicted the number of protons bound and the net charge asfunctions of the hydrogen ion concentration, Le., the pH ti-tration curve. The view tha t protein electrostatics can berepresented in terms of charges on a sphere of low dielectricconstant in a higher dielectric medium has remained useful.Recent improvem ents have included (i) the consideration ofdiscrete charges located a t specific positions on the spheric alnative protein (Tanford & Kirkwood, 1957a,b; Matthew &

TOPOLOGICAL

CONNECTED -eighborsinteraction

FIGURE 1: Spatially neighboring residues (id) re defined as connectedneighbors if they share a backbone bond, j = i + 1; therwise, theyare topologicalneighbors. Interactions are local or nonlocal dependingon the separation along the chain of the interacting residues.Gurd, 1986; Matthew & Richards, 1982), (ii) modeling nativestructural deviations from spherical shape (Gilson & Honig,1988a ,b), and (iii) the development of ele ctrostatic theory forthe unfolded state and therefore for free energy contributionsto stability (S tigter & Dill, 1990; Stigter et al., 1990).The re ar e two different ways in which electrostatic inter-actions can a ffect protein stability. (1) Classicalelectrostaticeffects are the nonspecific repulsions that arise when a proteinis highly charged, for example, at extremes of pH. The tra-ditional view (Tan ford, 1961; Kauzm ann 1954; Linderstrom-Lang, 1924) of these effects has been that the electrostaticfree energy depends on the square of the net charge. Henc e,no electrostatic contribution to protein stability is expectednear the isoelectric point. As the net char ge on the nativeprotein is increased by increasing acidity or basicity of thesolution, the increasing charge repulsion will destabilize thefolded protein because the charge density on the foldedmolecule is greater th an on the unfolded molecule. Thus, theprocess of unfolding leads to a state of lower elec trostatic freeenergy. Hen ce, acids and bases destab ilize native proteins (seeFigure 2).(2 ) Specific charge interactions can also affect stability. Forexample, ion pairing (salt bridging) occurs when oppositelycharge d amin o acid side chains a re in close spatial proximity.Wher eas the classical mechanism predicts that increasing thecha rge could only destabilize the folded state , ion pair ing couldstabilize it. It has traditionally been held that classical andion-pairing effects could be distinguished by experiments onthe ef fects of salt conc entration (below about 0.1-1.0 M) orthe dielectric constant of the solution. Only at low concen-trations does salt predominantly af fect electro static shielding;at higher concentrations the electrostatic shielding is saturated ,so that then the dom inant effects of salt, like any other ad-ditives, ar e on the solvent prope rties of the solution (M orrison ,1952). In the traditional view, it is assumed that salts andthe dielectric constant of the medium do not affect the netcharg e on the molecule and tha t they affect the native state


3/23

Perspectives in Biochemistry Biochemistry, Vol. 29 , N o. 31, 1990 7135

2 4 6 8 1 0 1 2PH

FIGURE 2: Denatura tion temperature vs pH for ( 0 )metmyoglobin,(A) ribonuclease A, (0 ) ytochrome C , ( 0 ) -chymotrypsin and (0 )lysozyme. Increased charge on the protein at extremes of pH (lowor high) favors unfolding. Reproduced from Privalov, P. L., &Kechinashvili, N. N. (1974) J . Mol. Biol. 6,6 65 . Copyright (1974)Academic Press Inc. (London) Ltd.more than the denatured state. It has often been assumedtherefore that either adding sa lt or increasing the dielectricconstant of the solution should stabilize proteins if classicaleffects are dominant or destabilize them if ion pairing is im-portan t. It is now clear, however, that effec ts of neither saltnor dielectric constant can be interpreted so simply, for thefollowing reasons. First, salts strongly affect the unfolded sta te(see below). Second , ion-pair bonds are genera lly much shorterthan the Debye lengths in salt solutions, so salt should havelittle effect on ion-pairing stability. [O ne exception is aGlu-2--*Arg- 10' salt bridge in the C -pepti de helix (Shoemak eret al., 199 0), which is screened by 1 M sa lt, but this m ay bea solvent-separated ion pair (R . L. Baldwin, personal com-munication).] Third, althoug h a decreased dielectric constan twill lead to increased charg e interac tions, it will also decreasethe total ionization since charging is energetically more costlyin a low-dielectric medium. Moreover, the dielectric constan tis also correlated with other solvent properties such as hy-drophobicity and is not a simple diagnostic for charge effectsalone. Therefore, discriminating between classical and ion-pairing electrostatics contributions to stability has been dif-ficul t.During the 193Os, ion pairing was considered to be thedominant contributor to protein stability (Cohn et al., 1933;Mirsky & Pauling, 1936; Eyring & Stearn, 1939). Mirskyand Pauling suggested that folding was driven by the ionpairing of carboxyl and amino groups on the side chains ofthe charged amino acids.If ion pairing is important for protein stability, then suchstability must arise from charged pairs at protein surfacesrather than from charged pairs buried in the protein core. Thefirst evidence that few ion pairs ar e buried was due to Jaco bsenand Linde rstrom-L ang (1 949) on model compounds. An im-portant signature of electrostatic effects in solution is a changein volume: the local volume of water decrease s around amolecule of increasing charge. Th e electrostatic field of thecharged molecule o rients and orders neighboring water dipoles(electrostriction), decreasing the entropy and volume of thelocal water molecu les. At low pH wh ere only the carbox ylgroups are titratable, Jacobsen and Linderstrom-Lang notedthat the volume increase upon protonation of C OO- to CO OH ,of about 10 mL/mol in proteins, is the same as in modelcarboxyl compounds in water, suggesting that ion pairs inproteins mus t be exposed. Mo re recent studies of knownprotein structure s show th at indeed few ion pairs ar e buried

Unfolded FoldedFIGURE 3: Early model in which protein folding was proposed to bedriven by ion-paired hydrogen bonding among side chains (M irsky& Pauling, 1936; Eyring & Stearn, 1 939), shown by Jacobsen andLinderstrom-Lang (1 949) to be inco nsistent with partial molar vol-umes.(on average, only about one ion pair per 150-residue proteinis buried) (Barlow & Thorn ton, 1983). This follows from thevery high Born energy required to transfe r a charged ion fromaqueo us solution to the low-d ielectric interior of the prote in,ranging from 19 kcal/mol for full burial to 4 kcal/mol for ahalf-exposed ion a t the surfac e, 7 kcal/mol for complete burialof an ion pair (H onig et al., 1986; Ho nig & Hubbell, 1984).Thus, unless other specific interactions are involved, onlysurface ion pairs could generally stabilize native states.

It is clear that ion pairing can contribu te to protein stability.Studies of X-ray crystal structures of known proteins (Wada& Naka mura, 1981; Barlow & Thorn ton, 1983) show that ionpairing is common on the surfaces of proteins. Also, varia tionsin sequence that affect ion pairing can change stability byabou t 1-3 kcal/mol of ion pairs (Fe rsht , 1972; Peru tz & Raidt,1975 ); Asp-70-H is-3 1 in T 4 lysozyme has recently been foundto stabilize by 3-5 kcal/m ol (Anderso n et al., 1990). Simila rly,ion binding sites designed into proteins can affect stability(Pace & Grimsley, 1988). However, it is clear that ion pairingis not the domin ant force of protein folding. Th e first evidenceemerged from the pivotal paper of Jacobsen and Linder-strom-Lang (1949). They interpreted the models of Eyringand Stearn (1939) and Mirsky and Pauling (1936) as shownin Figure 3: the hydrogen protonates the carbo xyl group inthe unfolded state, so both carboxyl and amino groups areuncharged, whereas in the folded state the hydrogen protonatesthe am ino group, so that the carboxyl and amin o groups forman ion pair. Jacobsen and Linderstrom-Lang presumed thecharg es remained solvated upon folding. Model compoundsshow that the protonation of N H 2 to NH 3+ eads to an elec-trostriction of about -4 mL /mol and deprotonation of C OOHto COO- leads to -10 mL/mo I, as noted above. Folding shouldthen resu lt in a volume change of -14 mL /mo l per ion pair.In con trast to this model, experiments show that folding leadsto an increase in volume (Jacobsen & Linderstrom-Lang, 1949;Zipp & Kauz mann, 1973; Brandts et al., 1970; Edelhoch &Osborne, 1976).

Also inconsistent with ion pairing as the d omina nt force offolding is the observ ation that the stab ilities of proteins showlittle dependence on pH or salt (a t low salt conce ntrations)near the isoelectric point (Tanfo rd, 1968; von Hippel &Schleich, 1968; Herman s & Scheraga , 1961; Acampora &Hermans, 1967). [For some proteins, the pH of maximumstability does not coincide with the isoelec tric pH, but this canbe accounted for within the classical model by the burial inthe hydroc arbon core of some of the titratable groups (oftenhistidines) (Stigter & Dill, 1990).] As further evidence tha tcharge generally contributes only weakly to protein stability,Hollecker and Cre ighton (1982) found little effect of changingthe charg es on several different amino groups in three differentproteins.


4/23

7 136 Biochemistry, Vol. 29, No. 31, 1990 Perspectives in Biochemistrymay play an imp ortant role in protein folding, but the mag-nitudes, am ong all the types of force contrib uting to proteinfolding, are currently perhaps the most difficult to assess.

Mirsky and Pauling (1 936) were the first to suggest thathydrogen bonding was the dominant force of protein folding.Although their focus appears to have been the electrostatichydrogen bonds arising from ion-paired side chains (see pre-ceding section), they also suggested that hydrogen bondingcould occur between the carbonyl C 4 nd amide NH groupsof the peptide backbo ne. It is the peptide hydrogen bonds weconsider here. Their proposal led to the X-ray crystallographystudies of amino acid crystals by Pauling et al. begun in 1937,culminating in the discovery of th e a-helix an d parallel andantiparallel sheets i n 1951 (Pauling et al., 1951; Pauling &Corey, 195 1a-d ). These were first called "secondarystructures" by Linderstrom -Lang (1 952).During the 1950s, Doty and his colleagues found a modelsystem, poly(?-benzyl-L-glutamate), for studying th e drivingforces in the formation of polypeptide helices in nonaqueoussolution (D oty & Yang, 1956; Doty et al., 1954, 1956, 1958).Soon thereafter a theoretical framework emerged for under-standing the balance of forces driving the helix-coil transition.The first theoretical model was due to Schellman (1958a).

Many other elegant treatments followed, principally based onthe one-dimensional Ising model (Peller, 1959; Gibbs & Di-Marzio, 1959; Zimm & Bragg, 1959; Zimm & Rice, 1960;Flory, 1969; Poland & Scheraga, 1970). In these models, theintrachain hydrogen bond is considered energetically favorablerelative to the hydrogen bond with the solvent. However, toform the first such bond requires overcoming configurationalentropy to arrange the immediately adjacent bonds into ahelical configuration. At low tempe ratur es with simple sol-vents, the enthalpic contribution dominates, and the moleculeforms a helix; at high temperatures, the entropy dominatesand the m olecule is configured as a ra ndom coil (Shoemakeret al., 1987; Marqu see et al., 1989; Lupu-L otan et al., 1965;Platzer et al., 1972). A sharp transition between these statesresults from this subtle balance between the large forces. Th eentropy is local insofar as it involves the co nfigurations of onlyimmediately neighboring bonds along the chain and thus isassumed to be independent of aspects of the chain configu-rations more distan t. Theore tical helix-coil transition modelssuccessfully predict (i) this temperature dependence and (ii)that helices become more stable and that transitions sharpenwith increasing chain length (Poland & Scheraga, 1970). Themodels also predict the influence of pH on the helix-coiltransition: greater charge on the molecule destabilizes the helixsince the coil has lower charge density and thus lower elec-trostatic free energy (Peller, 1959; Zimm & Rice, 1960). Thehelix-coil transition has inverted tem pera ture dependence insome mixed solvents (Zi mm et al., 1959; Lupu-L otan et a l.,1965). Solvents tha t bind to the peptide bond will favor thecoil; one example is formic acid, which protonates the bond(Lotan et al., 1967). Consistent with the view that hydrogenbonding is the principal driving force of th e helix-coil tran -sition, solvents tha t form stron g hydrogen bonds compe te moreeffectively with the peptide and destabilize th e helix relativeto the coil. For example, chloroform, dimethylformamide,2-chloroethanol, trifluoroethanol, and other alcohols favor thehelix, relative to formic acid, dichloroacetic acid, or tri-fluoroacetic acid (Doty & Yang, 1956; Doty et al., 1954, 1956,1958; Lupu-L otan et al., 1965; Conio et al., 1970; Nemethyet al., 1981; Nelson & Kallenbach, 1986). Similar theory hasbeen developed for P-sheet formation (Mattice & Scheraga,1984).

Third, perhaps the most persuasive evidence that ion pairingis not the dominant force of folding comes from the structuralstudies of Barlow and Thornton (1983). They have observedthat ion pairs are not highly conserved i n evolution. Moreimportantly, the nu mber of ion pairs in proteins is small. Theyobserve about five ion pairs per 150 residues of protein (abo utone of which is buried, noted above). It is unlikely th at anyinteraction involving only 10 residues, less than 10% of themolecule, could be the domin ant folding force. Using theestimate of 1-3 kcal/mol (Fer sht, 1972; Perutz & Raidt, 1975)for the stabilization per ion pair leads to a value of 5-15kcal/mol stabilization. Even though ion pairing would thuscontribute a free energy equal to that of the net stability ofthe protein, this is still 5-10-fold smaller than the hydropho bicinteraction discussed below.A similar estimate, of about 10 kcal/mol stabilization dueto ion pairing, has been made by Friend and Gurd (1979;Matthew & Gurd, 1986). They observed decreased stabilityof sperm whale ferrimyoglobin with increased salt and i n -terpreted this as evidence for ion pairing. They assumed saltpredominantly affects the native state, on the basis of thedifference i n titration behavior of native and de natured states.However, these results do not necessarily imply the electrostaticstabilization comes from ion pairing. Sa lt can affect therelative free energies differently than it affects the titrationbehavior. A recent polyelectrolyte model of proteins (Stigter& Dill, 1990; Stigte r et a] ., 1990) shows instead that increasingsalt, by classical effects alone, will reduce the electrostatic freeenergy of the unfolded s tate of myoglobin more than th e foldedstate. Increased salt shields the charge repulsions i n the un-folded molecule more effectively than i n the folded moleculeat low pH, probably because of better penetration of the saltsolution into the unfolded m olecule. Th e model is consistentwith an additional experimental observation th at is otherwisedifficult to explain on the basis of ion pairing. For 0-lactam ase,similar to the myoglobin experiments of Friend and G urd , Gotoand Fink ( 1 989) observe that salt destabilizes the native statewhen the m olecule is highly charged a t low p H, but they alsofind that salt stabilizes the native structure when the m oleculeis charged a t high pH. It is interesting tha t a significantfraction of the electrostatic free energy is predicted to arisefrom the entropy of proton release (Stigter & Dill, 1989,1990), rather than simply from the charge energetics.INTERACTIONS

van der Waals attractions arise from interactions amongfixed or induced dipoles. A hydrogen bond occurs when ahydrogen atom is shared between generally two electronegativeatoms. Hydrogen-bond str eng th, which depends on the elec-tronegativity and o rientation of the bonding atoms, is in th erange of 2-10 kcal/mol (Pauling, 1960). For example, thewater-water hydrogen bond i n the vapor phase is -6.4kcal/mol (Weiner et al., 1984). A hydrogen bond is primarilya linear arrangement of donor, hydrogen, and acceptor andis comprised of electrostatic, dispersion, cha rge- tran sfer, andsteric repulsion interactions (Vinogradov & Linnell, 1971 ) .The dominant component of a hydrogen bond is electrostatic(Pauling, 1960: Cybulski & Sheiner, 1989; Vinogradov &Linnell, 1971). I n this section we ask: do hydrogen bondsand van der Waals interactions contribute differently to foldedand unfolded states of proteins, and therefore to stability?While these two types of force ar e microscopically qu ite dif-ferent, there are few simple macroscopic diagnostics that candistinguish between them; hence, in this section we considerthem toge ther. The evidence cited below suggests that they

( 2 ) H Y D R O G E NONDINGN D VA N DE R WAALS


5/23

Perspectives in BiochemistryFor three reasons, it is natural to assume that hydrogenbonding and van der Waals interactions will be important forthe conformational changes of proteins. First, the amino acidsthat com prise proteins are dipolar and ar e capable of hydrogenbonding. Second , helices are common feature s of globularproteins, and the studies cited above show that the helix-coiltran sition is largely driven by hydrogen bonding. Sim ilarly ,intramolecu lar sheets are also formed by hydrogen bonding(Anufrieva et al., 1968). Third, the conformational forces fornonelectrolyte polymers in nonelectrolyte solvents are shortranged, arising from differences in monomer-monomer at-tractions of the chain relative to monomer-solvent attractions(Flory, 1953; deGennes, 1 979). If monomer and solvent in-teractions are short ranged, then classical polymer theorieswould predict that chains should usually be relatively self-attractive, with radius changes characterized by a tempera-ture-independent enthalpy (Flory, 1953). Such a tempera-ture-independent e nthalpy has been inferred to contribute toprotein folding on the basis of model assumptions about thecontribution of the hydrophobic interaction (Baldwin, 1986;Privalov, 1979; Dill et al., 1989). This residual folding forcebecomes more favorable as the number of polar groups in-creases (Privalov & Kechinashvili, 1974). For hen lysozyme,

the m agnitude of this enthalpy is 0.43 kcal/mol of residues(Baldwin, 1986).Although these short-ra nged forces ar e therefore undoub-tedly important, Kauzmann (1954, 1959) concluded that theyare probably not the dominant forces that fold proteins inwater. A f undam ental criterion for a dominant driving forceis that it must explain why the folded state is advantageou srelative to the unfolded s tate. H e argued tha t hydrogenbonding would not satisfy this criterion, because there was nobasis for believing that the intrachain hydrogen bonds in thefolded state would have lower free energy than those of theunfolded chain to wate r. In suppo rt of this view, the distri-bution of hydrogen-bond angles in proteins is observed to beabout the same as i n small-molecule compounds (Bake r &

Hub bard , 1984). It follows however that folded proteins mustcontain m any hydrogen bonds; for otherwise, the protein woulddenature.Kauzmanns hypothesis led to model studies on analoguesto determine the free energy of peptide hydrogen bonds inwa ter. Th e several models of the peptide hydrogen bondinclude urea (Schellman, 1 9 5 9 , valerolactam (Susi, 1969),N-methylacetamide (NMA) (Klotz & Franzen, 1962;Kresheck & Klotz, 1969), and cyclic dipeptides, the diketo-piperazines (Gill & Noll, 1972). For reasons described below,however, these model studies have not yet yielded definitiveestimates for th e contribu tion of hydrogen bonds to proteinstability. Th e dimerization binding constants and their tem-perature dependences have been measured for these molecules

i n water, in order to obtain free energies and enthalpies ofdimerization. Because the concentration dependences in theseexperiments are linear at low concentrations, the bound speciesis presumed to be predominantly in the form of dimers, ratherthan higher multimers. At 25 OC, imerization in water isdisfavored; the equilibrium ratio of dimers to monomers is only4.1 X for urea and 5.0 X for diketopiperazine. Thus,the free energy of dimerization is positive (AGdimnizatim +1.89kcal/mol for urea) . However, the enthalpy of dimer formationis negative (-2.1 kcal/mol for diketopiperaz ine, -2.1 kcal/m olfor urea, -2.8 kcal/mol for valerolactam), except for N -methylacetamide for which it is approximately zero (Klotz &Franzen, 1962). On the assumption that the hydrogen bondis the only attraction driving dimerization, it has been generally

Biochemistry, Vol. 29, No. 31, 1990 7137concluded that the hydrogen bond in water is enthalpicallyfavored relative to the monomer-water bond. For N-methylacetamide, Jorgensen (1989) has shown by Monte Carlosimulation that this assumption is probably not valid: in water,the amides stack rather tha n form hydrogen bonds. In con-trast, in chloroform, the am ides form good hydrogen bonds(Jorgensen, 1989). Susi and Ard (1969) have suggested thatc-caprolactam dimerization may also be driven by somemechanism other than hydrogen bonding. Thu s in additionto hydrogen bonding, van der Waals and other interactionsmay also contribute significantly, but their importance for th eother model compounds is not yet clear.An add itional problem prevents unequivocal determinationof the free energy of hydrogen-bond formation from thesemodel studies. In all the model compounds, there a re two waysa dimer can form: singly bonded, wherein one partn er in thedimer will have considerable rotational freedom relative to theother, or doubly bonded, with one partner considerably re-stricted in its rotation relative to the other. The experimentsfind only the ratio of com plexed molecules (of singly-bondedplus doubly-bonded dimers) to unbound m onomers. To obtainthe free energy of hydrogen-bond formation from these datarequires additional knowledge of the relative numbers ofsingly-bonded and doubly-bonded dimers, not currentlyavailable for these model compounds. Th e dilemma is illus-trated by the following comparison. Suppose, on the one hand,that the only species in solution was known to be the singly-bonded dimer; then t he mea sured positive free energy implieshydrogen bonding is disfavored in water. Suppose alternativelythat the only species in solution was known to be the dou-bly-bonded dimer. The n the binding free energy will includecontributions from the two hydrogen bonds and an unfavorableentropy of rotational restriction. If this rotational restrictionis sufficiently unfavorab le, contributing a larg e enoug h positivefree energy to the overall dimerization free energy, then th eintrinsic free energy of hydrogen-bond formation will be in-ferred to be negative, implying that hydrogen bonding is f a -vored in water. Schellm an (1955) estimated this entropicrestriction to be 3-6 eu and concluded that th e free energyof formation of a hydrogen bond is negative but probablysmall. Thus th e inference as to whether hydrogen-bond for-mation is favored or disfavored in water depends on (i) w hichcompound is chosen as a model, (ii) the importance of in-teractions other than hydrogen bonding for the associationprocesses of those model compounds in water, and (iii) esti-mation of the m agnitude of a rotational restriction entropy ,presently unknown but probably of about the sa me magn itudeas the free energy of hydrogen bonding itself.Moreover, this class of experiments has largely been re-stricted to aqueous solvents. But the peptide hydrogen bondin a globular protein is in a more hydrocarbon-like medium.We ar e therefore interested in the following equilibrium:

A, + 6,k AB),

& + Bw7AWwwhere A and B are the hydrogen-bond donor and acceptor,w is water, and n is the nonpolar solvent. W e aim to determineAG,, the hydrogen-bond contribution to protein stability. Weobtain this by using other steps of the thermodynamic cycle.(1) A wide range of hydrogen-bonding species, includingNMA, formamide, alcohols, carboxylic acids, and phenols,tend to associate in nonpolar solvents ( K , > 1); for example,for the dimerization of N M A in CC14,AG , = -2.4 kcal/mol


6/23

7138(Vinogradov & Linnell, 1971; Klotz & Farnham, 1968;Kresheck & Klotz, 1969; Roseman, 1988; Sneddon et al.,1989). However, this free energy is solvent dependent. Hy -drogen bonding strengthens in nonpolar solvents either if (i)the dielectric constant of the solvent is reduced, with othersolvent properties held fixed (F ranz en & Stephens, 1963), asexpected for Coulomb interactions, or if (ii) the electron- orproton-donating or -accepting cap acity of th e solvent is varied,with the dielectric constant held fixed (Allen et al., 1966;Krikorian, 1982). (2) Transfe rring a hydrogen bond into anonpolar medium is generally disfavored: AGz = +6.12kcal/mol has been estimated for NMA from water to CCl,(Roseman, 1988). (3) As noted above, AB dimerization isdisfavored in water; for NM A, AG, = +3.1 kcal/mol (Klotz& Farnha m, 1968; Roseman, 1988). (4) It follows that AG4= +0.62 kcal/mol for NMA in CCl, (Roseman, 1988); AG4is also near zero for carboxylic acids in benzene (Klotz &Farnham, 1968) and is predicted to be +2.2 kcal/mol fortransferring the formamide dimer from water to CCl,(Sneddon et al., 1989). ( 5 ) It follows from these estim atestha t hydrogen bonding opposes folding. For NM A in CCI4,AG5 = +3.72 kcal/mol; for formamide, AG5 = +1.9 kcal/mol.However, if the transfer of A and B from water into a nonpolarmedium is driven by some other force, such as hydrophobicity(see below), then hydrogen-bond formation, process 1, will bestrongly favored within the folded struc ture. Thu s whereashydrogen bond ing may not assist the collapse process, it wouldfavor internal organization within the compact protein. Twoadditional uncertainties make it difficult to estimate the me-dium effect on the hydrogen-bond strength: (i) the nonpolarcore of a protein is not a homogeneous dielectric (Honig etal., 1986; Wars hel, 1 984), and (ii) hydrogen-bond strengthis extremely sensitive to geometric details of bond angles(Scheiner & Hillenbrand, 1985; Scheiner et al., 1986). Sincethere are so many hydrogen bonds in native proteins, then evensmall errors in estimating their strength will lead to large errorsin determining their effects on protein stability. Only 11%of all C=O groups and 12% of all N H groups have no hy-drogen bonds (Baker & Hubbard, 1984). Of all the hydrogenbonds to C=O groups, 43% are to water, 11% are to sidechains, and 46% are to main-chain N H groups. Of all thehydrogen bonds to N H groups, 21% are to water, 11% are toside chains, and 68% are to main-chain C= O groups. Toreliably estimate the stability of a protein would thereforerequire model studies more accurate than about (1 / 5 ) k T perintra cha in peptide bond. For the reasons noted above, thisaccuracy is not yet available from the current models.

Solvent denaturation studies indicate that hydrogen bondingis not the dom inant folding force (Singer, 1962; Edelhoch &Osborne, 1976). I f it were, then solvents that form stronghydrogen bonds to the peptide backbone should compete ef-fectively and unfold the protein. Tho se solvents tha t do notaffect hydrogen bonding should not affect stability. In thislight, several observations ar e of importance. (1) It would bedifficult to rationalize the observation that very small con-centrations of surfac tants (1% dodecyl sulfate, for example)unfold proteins (Tanford, 1968) since they do not destabilizehelices (Lupu-Lotan et al., 1965). Moreover, the effectivenessof tetraalkylammonium salts to denature proteins depends onthe number of methylene groups, indicative that it is the hy -drophobic interaction, rather than hydrogen bonding, whichdetermine s stability. (2) Since the C=O group is a stronghydrogen acceptor and the N H group is a weak donor, Singer(1962) has pointed out t hat the solvents useful for competingwith the peptide hydrogen bond would be those which are

Biochemistry, Vol. 29 , No. 31, 1990 Perspectives in Biochemistrystronger donors than NH; the peptide bond will generallycompete effectively against solvents that a re weaker hydrogenacceptors tha n C=O. Dioxane is only a hydrogen-bond ac-ceptor and therefore should not den atur e proteins if hydrogenbonding were the dominant folding force. However, dioxanedenatures proteins (Sing er, 1962). (3) Alcohols are morehydrophobic than water, but they enhance helix formation(Conio et al., 1970). If hydrogen bonding were the dominantfolding force, alcohols should stabiliz e proteins. Hen ce, theobservation that at low concentrations they destabilize proteinsis inconsistent with hydrogen bonding as the principal drivingforce (Herm ans, 1966; von H ippel & Schleich, 1969a,b; Parcdiet al., 1973). The caveat is that alcohols have complex effectson protein stab ility, depending on concentration and tem-perature (Br andts, 1969). Finally, a particularly importantcomparison involves the effect on protein stability of alcohols(ethanol and propanol) vs the corresponding glycols (ethyleneglycol and propylen e glycol). Th e glycols ar e less hydrophobicand have more hydrogen-bond sites than the correspondingalcohols. The observation th at th e glycols are worse denatu-rants is strong evidence that hydrogen bonding is less importantthan t he hydrophobic interaction (T anford, 1968; Herskovitzet al., 1970; von Hippel & Schleich, 1969a).Mutation studies show tha t hydrogen-bonding groups affectstability, but by an amount which can differ considerablydepending on the site and nature of the mutation (B artlett &Marlowe, 1987). Fersht et al. (1985) have estimated fromactivation free energy measurements of tyrosyl-tRNAsynthetase/substrate interactions that breakage of a hydrogenbond increases the free energy by 0.5-1.5 kcal/mol for u n -paired uncharged donor and acceptor or about 3.5-4.5kcal/mol if the donor or acceptor is charged. Site-directedmutagenesis experiments in which nonhelical proline 86 inphage T4 lysozyme is replaced by other amino acids whichextend the helix and add new backbone hydrogen bonds showmarginal reduction, rather than increase, in stability (Alberet al., 1988). On the other hand, side-chain hydrogen-bondinggroups are found to stabilize T4 lysozyme (Alber et al., 1987a;Gr utte r et al., 1987). However because site-directed muta -genesis experiments measure only the total change in stabilityupon mutation, AAG, and not the individual molecular com-ponents of th at chang e, then these mutations m ay involve morethan just hydrogen bonding. For example, for one of them(Th r replacing V al- 157), free energy perturbation calculationsshow tha t the added stability arises from better van der W aalsinteractions rathe r than from th e difference in hydrogen-bondstrength (Dang et al., 1989).(3 ) LOCAL NTERACTIO NS:NTRINSICROPENSITIES

Th e term intrinsic propensity! does not desc ribe any sing letype of force. Rathe r it is intended to convey the idea tha tthere are certain conformational preferences of di- or tri-peptides, depending on the sequence, which arise from the sumof short- and long-ranged forces that are local among con-nected neighboring residues. (Local may extend to residuesthree to four monomers distant and may therefore also includehydrogen bonds involved in turns or helices.) Intrinsic prop-ensities have been studied by the measurement of helix/coilequilibria of peptides in water (Sueki et al., 1984; Marquseeet al., 1989; Marqusee & Baldwin, 1987) and turn/coilequilibria (Dyson et al., 1985, 198 8a; Wr ight et al., 1988).The stabilities of long polypeptide helices in aqueous solutioncan be attr ibu ted to intrinsic propensities. Helix stabilityincreases with chain length (Goodman et al., 1969; Zimm &Bragg, 1959; Poland & Scheraga, 1970). Therefore, althoughthe free energy contribution from each individual residue may


7/23

Perspectives in Biochemistrybe small, summed over many residues, the helix can b e stronglyfavored relative to the coil in a long chain.

The traditional view has held that the individual residuehelix/coil equilibrium constants are so nearly equal to one,however, and the initiation constants so small that short helices(less than abou t 15-20 residues) ar e not stable in aqueoussolution. Th ere have been two bases for this view. First, shorthelices extracted from stable globular proteins have been foundto be unstable in isolation in aqueous solution (Epand &Scheraga, 1968; Taniuchi & Anfinsen, 1969; Dyson et al.,1988b). Second, using guest am ino acids randomly dopedinto host copolymers of hydroxypropyl- and hydroxy-butyl-L-glutamine, Scheraga and his colleagues (Sueki et al.,1984) showed that the intrinsic propensities of amino acidsto form helices in water are small (w ith helix/coil equilibriumconstants ranging from 0.59 to 1.39 at 20 C). (Helicalpropensity can be increased considerably by reducing thetemperature to near 0 C .) Howeve r, these equilibriumconstants will be universal, in principle, only if the host helixitself is completely inert in its effect on the helix/coil equi-librium of the guest residue. Th e following recent evidencewith oth er hosts, however, suggests that the helix is not com-pletely inert, Le., tha t there are context effects. (1) Thehelix/coil constants differ in other sequences and can be aslarge as nearly 2 for alanine in alanine-based helices (Ma r-qusee et al., 1989; Padmanabhan et al., 1990). Although thehelix/coil constant for uncharged guest residue i appearstherefore to depend on the local sequence through residues i-1an d i+ 1, it does not appe ar to furth er depend on i-2 and i+ 2or otherwise on the position in the cha in (M erutk a & Stell-wagen, 1990 ). (2) Additional stability results if helix for-mation leads to burial of nonpolar surface (Tanford, 1968;Chou et al., 1972; Richards & Richmond, 1978). (3) Helicescan be stabilized considerably by reducing the helix dipolemoment through reduction of the charges at the helical ends(Shoemaker et al., 1985, 1987; Marqusee & Baldwin, 1987).(4) Salt bridges and aromatic interactions can also affectstability (Marquse e & Baldwin, 1987; Shoem aker et al., 1 990).In addition, intrinsic propensities can vary with the solvent(Rich & Jasensky, 1980). Recent evidence (Merutka et al.,1990) suggests that context effects may be at least as importantas intrinsic propensities.

On the basis of these observations, considerable progresscontinues to be ma de in improving stabilities, so that higherhelix/coil eq uilibrium constants are achiev ed, in shorter chains,and at increasing tempera tures up to near room temperatu re(Marqusee & Baldwin, 1987; Bradley et al., 1990). Never -theless, intrinsic propensities, in the absence of other forces,appear to be insufficient to account for the full helical stabilityin globular proteins. Helices in globular proteins are s hort.Th e average length is about 12 residues, and the most probablehelix length (peak of the distribution) is less than 6 residues(Kabsch & San der, 1983; Levitt & Greer, 1977; Srinivasan,1976). Yet protein helices remain 100%helical up to tem-peratures near the denatura tion point. Othe r forces musttherefore also be im porta nt for stabilizing helices in globularproteins. On e possibility is tha t there may be add itiona lcontext effects due to the environment provided by theprotein interior. For example, charges are distributed inproteins so as to stabilize the helix dipole (Blagdon & Good-man , 1975; Richardson & Richardso n, 1988). Helices withmodified charges at the ends can affect protein stability(Mitchinson & Baldwin, 1986). In principle, helices couldpack in pairs, antialigned, to reduce th e net dipole mom ent;this probably c ontributes little to stability, however, since the

Biochemistry, Vol. 29, No. 31, 1990 7139ends of helices are genera lly found in a high-die lectric mediumat protein surfaces (Rogers, 1989; Gilson & Honig, 1989;Presnell & Cohen, 1989). In contrast to these effects, theprotein environment may not always stabilize helices: Alberet al. (1988) found that added hydrogen-bonded helix-ex-tending residues in T4 lysozyme had little effect or destabilizedthe protein.Othe r evidence suggests that protein architecture does notarise principally from intrinsic propensities. First, distributionsof secondary structures predicted by intrinsic propensities areinconsistent with those in known protein crystal structures.Any model of protein stability based only on local interactionswould predict, as is observed in helix/coil eq uilibria, that longerhelices should be more stable, and therefore more probable,than shorte r helices. In contrast, studies of the crystalstructures of globular proteins (Kabsch & Sander, 1983; Levitt& Greer, 1977; Srinivasan, 1976) show just t he opposite: helixprobability decreases monotonically with length (see Figure11). Similarly, longer sheets ar e observed to be less probablethan shorter sheets, for both parallel and antiparallel sheets.These discrepancies ar e not repaired by local factor s alone.For example, it is known that helices can be term inated bystop residues (K im & Baldwin, 1984). Sto p signals will notaccoun t for the da ta base trends, however, which are grandaverages over residues, positions, and proteins, since it wouldthen follow that most amino acids must be helix destabilizing,in contradiction to the basic premise. Moreover, even giventhat local interactions impart some stability to helices andturns, it is difficult to rationalize how they would give rise tosheets, which are intrinsically nonlocal. Therefore, alternativeexplanations for these distrib ution s involve nonlocal factors.For example, at high densities short peptides can form stablehelices in crystals (Karle et al., 1990), suggesting the im-portance of packing effects. It is shown in section 8 that th edistributions of internal architecture can be accounted for bysteric forces of nonlocal origin.Second, attempts to predict protein structures by use of onlyintrinsic propensities have had limited success (Schulz &Schirmer, 1979; Kabsch & Sander, 1984; Argos, 1987;Thornton, 1988; Rooman & Wodak, 1988; Qian & Sejnowski,1988). Success rates ar e about 64% when averaged over manyproteins (Thornton, 1988; Rooman & Woda k, 1988; Qian &Sejnowski, 1988; Holley & Kar plus , 1989). Because this isconsiderably better than chance, it implies that intrinsic pro-pensities are significant determinants of protein structure.However, according to Qian and Sejnowski (1988), who haveused neural net methods, no method based solely on localinformation is likely to produce significantly better results fornon-homologous proteins. With in a given class of proteins,success rates may be higher (Kneller et al., 1990). Limitationsof intrinsic propensities are also found in studies of confor-mations of identical pentapeptides in different proteins (Kabsch& Sander, 1984; Argos, 1987). Kabsch and Sander found thatin 6 of 25 cases one pentamer could be found in a helix whereasthe identica l sequence in a differen t protein would be in a sheet,implying that local information alone is not sufficient to fullyspecify the conformation in the protein. W ha t does the 64%success rat e tell us about the ma gnitud e of the nonlocal factorsmissing from these prediction metho ds? If we assume tha tthe conformation of any residue can be predicted with a priorisuccess rate p o = 1/3 (Chou & Fasman, 1978), as a roughestimate for classification as a helix, sheet, or other confor-mation, then a success rate p carries an amount of information,(1) :


8/23

7 140 Biochemistry, Vole29, N o. 31, 1990Thu s a success rate of p = 0 .6 0 4 7 0 implies that local factorsalone account for only about 15-3076 of the total informationrequired to ma ke a perfect prediction. This agrees with es-timates (Chan & Dill, 1990b) that the local contributions tothe stability of a six-residue helix (based on a 1.05 equilibriumconstant) a re about 12% of the magnitude of the nonlocal stericfree energy that drives helix formation, described in section8.(4 ) HYDROPHOBICFFECT

Following the elimination of electrostatics as a principalforce of folding by Jacobsen an d Linde rstrom-La ng (1949),it was suggested that protein folding was driven by the aversionfor water of the nonpolar residues (Linderstrom-Lang, 1952;Lumry & Eyring, 1954; Kauzmann, 1954). The same aversionwas known to drive micelle formation; it was then assumedto be due to van der Waals interactions (Debye, 1949). Intwo remarkably insightful papers, Kau zmann (1 954, 1959 )made th e first strong case for the im portance of the hydro-phobic interaction in protein folding. H e reasoned tha t theformation of one hydrophobic bond (which he called anantihydrogen bond) upon folding involves the gain of a fullhydrogen bond among water molecules, which should be moreimportant by an order of magnitude than simply a change ofstrength of a hydrogen bond upon folding if hydrogen bondingwere the dominant folding force. In support of this view,Ka uzm ann offered the following evidence. First, nonpolarsolvents denature proteins (Singer, 1962; von Hippel &Schleich, 1969a). According to a hydrophobic mechanism,the nonpolar solvent reduces the free energy of the unfoldedstate by solvating the exposed nonpolar amino acids. Second,experiments of Christensen (1952) had shown an unusualtemp era ture dependence in which stability not only decreasesat high temperatures but also decreases at low temperatures.Kauzmann observed that this cold destabilization resemblesnonpolar solvation: nonpolar solutes become more soluble inwater at low temperatures (Privalov & Gill, 1988).A considerable body of more recent evidence continues tosupport the view th at hydrophobicity is the domin ant forceof folding. First, spectroscopic and high-re solution differentialscanning calorimetry experiments show the resemblance of thetem pera ture dependence of the free energy of folding and th etemperature dependence of the free energy of transfer ofnonpolar model compounds from water into nonpolar media(Pace, 1975; Privalov, 1979; Privalov & Gill, 1988). Bothinvolve large decreases in heat capacity. Second, a largenumber of crystal structures of proteins have become availablesince Kauzmanns predictions. They show that a predominantfeature of globular protein structures is that the nonpolarresidues are sequestered in to a core where they largely avoidcontact with water (Perutz et al., 1965; Chothia, 1974, 1976;Wertz & Scheraga, 1978; Meirovitch & Scheraga, 1980; Guy,1985). Third, protein stability is affected by different saltspecies (particu larly at high salt concentrations) in the samerank or der as the lyotropic (Hofm eister) series (von Hippel& Schleich, 1969a,b; Arakawa & Timashe ff, 198 4); this isgenerally taken as empirical evidence for hydrophobic inter-actions (von Hippel & Schleich, 1969; Collins & Washabaugh,1985; Morrison, 1952; Morrison & Billet, 1952). From themost stabilizing (for folded ribonuclease) to the most desta-bilizing, the rank order of anions is found to be SO-CH,COO-, CI-, Br-, C lod -, CN S- and the rank order ofcations is NH4+, K+ , Na+, Lie, CaZ+. The solubilities ofbenzene and acetyltetraglycyl ethyl ester in aqueous salt so-lutions increase in the same rank order. Fourth, accessiblesurface studies (Richards, 1977) and site-directed mutagenesis

Perspectives in Biochemistry

c ys (S-S)O L ? / l

0 Tyr

OTrp-4 - 1 0 1 2 3 4-aGt, (kcal mol.)FIGURE 4: Change in free energy of unfolding, AAG, of mutant T4lysozymes a t position 3 (wild type is Ile) by substitution of otherresidues, compared to the corresponding free energy of transfer fromwater to ethanol, AG,. Reprinted with permission from Matsumura,M., Becktel, W. J. , & Matthews, B. W. (1988) Nature 334, 406.Copyright (1988) Macmillan Magazines Limited.experiments involving the replacem ent of a given residue byother amino acids show that the stability of the protein isproportional to the oil-water partitioning propensity of theamino acid (see Figure 4) (Yuta ni et al., 1984, 1987; Ma t-sum ura et al., 1988a,b; Kellis et al., 1989). Fifth , the hy -drophobicities of residues in the cores of proteins appear tobe more strongly conserved and correlated w ith structure thanother types of interactions (Lim & Sau er, 1989; Bowie et al.,1990; Kelly & Holladay, 1987; Sweet & Eisenberg, 1983;Bashford et al., 1987). Sixth, computer simulations of in-correctly folded proteins show that the principal diagnosticof incorrect folding of proteins, apar t from inappropriate bu rialof charge, is the interior/exterior distribution of hydrophobicresidues (Novotny et al., 1984, 1988; Baumann et al., 1989).Wh at are hydrophobic interactions? There has been somedisagreement about the meaning of hydrophobic (effect, force,interactions, etc.) (H ildebrand, 1968, 1979; Nem ethy e t al.,1968; Tanford, 1979; Ha et al., 1989). At least three differentmeanings of these terms have been used. (1) Hyd rophobichas been used to refer to the transfer of a nonpolar solute toany aqueo us solution. (2) Alternatively, it has been used morespecifically to refer to transfers of nonpolar solutes into anaqueous solution only when a p articular c haracteristic tem-pera ture dependence, described below, is observed. These twomeanings describe only experimental observations and makeno reference to any particular molecular interpretation. (3)Hydrophobicity has also been used to refer to particularmolecular models, generally involving the ordering of watermolecules around nonpolar solutes. In this review, hydro-phobicity will be defined by (2) for reasons discussed below.Wh at is unusual about the tempe rature dependence of thehydropho bic interaction? First it is useful to describe normalsolutions. There are two driving forces relevant to the mixingof simp le solutions of A with B, of spherical particles governedby dispersion forces. Th e tendency to mix is driven by anincrease in the translational entropy since there are moredistinguishable spatial arrangements of the A and B moleculesin the mixed system than of the individual pure systems. Onthe other hand, mixing in simple systems is opposed by theenthalpies of interaction; ordinarily, dispersion forces leadingto A B attractions a re smaller than those leading to the cor-responding AA and BB attractions. The latter is captured inthe general rule that like dissolves like (Hildebrand & Scott,1950). For simple solutions, the transfer of B from pure Binto A is therefore g enerally opposed by entha lpic interactions.When these interactions are stron g, i.e., when A an d B are


9/23

Perspectives in Biochemistryrelatively insoluble in each other, then the free energy oftran sfe r is dominated by this opposing enthalpy, which is muchlarger tha n the mixing entropy, and transfer is disfavored. Thisis the ordinary form of incompatibility of two components Aand B .Oil and water a re also incompatible a t 25 OC. Oil/w aterincompatibilities tend to be stronger tha n normal incom-patibilities. However, th e free energy alone (Le., the solubilityor the partition coefficient) is not the principal distinction ofnormal from hydrophobic processes; in both cases the transfercan be disfavored. Th e distinction between normal and hy-drophobic processes is in the tempera ture dependence. W ha twas first recognized as unusual ab out nonpolar transfers towater at 25 C (Edsall, 1935; Butler, 1937; Frank & Evans,1945) was that they are not principally opposed by the en-thalpy; they are principally opposed by an excess (Le.,unitary) (Gurney, 1953; Tanford, 1970) entropy (Gill &Wads o, 1976; Tan ford, 1980; Privalov & Gill, 1988). Theexcess entropy is that which remains afte r the mixing entropy(the RT In x term in the chemical potential) is subtracted fromthe total measured entrop y. Th e enthalpy of mixing oil andwater is generally small, sometimes even negative (favorable),at 25 OC (Privalov & Gill, 1988). Because these conclusions

derive from experiments with solutes at high dilution, theyimply that the excess entropy must arise from the watersolvation around the solute rather than from some possiblesolute-solute interaction. A molecular interpretation of thesedata, which is supported by computer simulations (Geiger etal., 1979; Pangali e t al., 1982; Ravishanker e t al., 19 82), isthat a t 25 OC nonpolar solutes are surrounded by orderedwaters. Wa ters surrounding the nonpolar solute prefer tohydrogen bond with other waters rather tha n to waste hy-drogen bonds by pointing them toward the nonpolar species(see top of Figure 6) (Stillinger, 1980; Geiger et al., 1979).However, the av ersion of nonpolar solutes for water becomesmore ordinary, an d less entropy driven, at higher temperatures.This is because there is a second fundamental difference be-

tween simple incompatibility on the o ne hand a nd oil/waterincompatibility on the other . For simple solutions, the hea tcapacity chan ge upon transfer is small. For nonpolar solutes,the hea t capacity change upon transfer from th e pure liquidto water is large and positive (Frank & Evans, 1945; Christian& Tuck er, 1982; Privalov & Gill, 1988; Edsall & McKenzie,1983). This means th at for simple solutions the transfer ofA into B is character ized by an enthalpy and en tropy whichare temperature independent and a free energy which isconstant or linear with tempera ture. However, the situationis quite differen t for the transfer of nonpolar solutes into water.A large heat capacity implies the enthalpy and entropy arestrong functions of temperatu re, and the free energy vs tem-perature is a curved function, increasing at low temperaturesand decreasing at higher temperatures. Hence, there will bea temp eratu re a t which the solubility of nonpolar species inwate r is a minimu m ( Crov etto et al., 1982; Becktel &Schellman, 1987; Edsall & McKen zie, 1983; Privalov & Gill,1988) (see Figure 5 ) . A striking consequence follows, onewhich i s at variance with the view th at w ater ordering is theprincipal featu re of the aversion of nonpolar residues for water.The free energy of transferring nonpolar solutes into wateris extrapolated to be most positive in the tempe rature ra nge130-160 OC (Privalov & Gill, 1988). Therefore, th e aversionof nonpolar species for water, whatever its molecular nature,is greatest at these high temperatures. Because this maximumaversion, by definition, arises where the free energy of transferis a maximum, and thus where the entropy (the temperature

Biochemistry, Vol. 29, N o. 31, 1990 7141

-05 1 , ,Benzene 1 1 , , , 1-40250 300 350 400 450A. Aqueous Solution

T, K TB. Simple Solution

FIGURE : Comparison of the enthalpy (AH),entropy (AS ) , and freeenergy (AG) of solute transfer from the pure liquid into a hypotheticalregular solution and into an aqueous solution. The data in (A ) arefor benzene, from Privalov and Gill (1988); the entropy at hightemperature is extrapolated on the basis of assumed constant heatcapacity. The figure on the right represents an idealization accordingto regular solution theory. Reprinted with permission from Privalov,P. L., & Gill, S. J. (1988) Ado. Protein Chem. 39, 191. Copyright(1988) Academic Press.derivative of the free energy) equals zero, then the m aximumaversion of nonpolar solutes for water must be driven by en-thalpy (Privalov & Gill, 1988; Baldwin, 1986). In other words,at the temperature for which hydrophobicity is strongest, theentropy of transfer is zero! A t those tempera tures, the aversionof nonpolar solutes for wa ter is enthalpic, as in simple classicalsolutions. It is therefore inapprop riate to refer to nonpolarsolvation processes in water, with large heat c apacity changes,as entropy driven or enthalpy driven, since either de-scription is only accura te within a given temper ature ra nge.A t 25 C , the hydrophobic effect is entropic; a t 140 OC, it isextrapolated to be enthalpic (Privalov & Gill, 1988).For processes of constant ACp, the enthalpy and entropy oftransfer areA H ( T ) = A H ( T I )+ s T > C p d T =

AH(7-J + A C p ( T - T I) (2 )TA S ( T ) = A S ( T 2 )+ s A C p / T d T =T2 W T 2 ) + ACp In ( T / T 2 ) (3 )

and therefore the free energy of transfer isAG(T) = AH - TAS =A H ( T l )- T A S ( T 2 )+ AC,[ (T - T I)- T In ( T / T 2 ) ] 4)These quantities are defined in terms of two arbitrary referencetemperatures: T I , or which the enthalpy is known, and T 2 ,for which the entropy is known. Th ere ar e three alterna tiveways to express this free energy in terms of two particularlyconvenient reference temperatures, Th, he temperature a twhich t he enthalpy is zero, and T,, the temperature for whichthe entropy is zero:

TI = Ti, T2 = T ,AG(T) = A C p [ ( T- Th) - T In ( T / T , ) ] ( 5 )

TI = T2 = T ,AG(T) = A H ( T , ) + A C p [ ( T- T , ) - T In ( T / T , ) ] ( 6 )

T I = T2 = Th


10/23

7142 Biochemistry, Vol. 29, No. 31 , 1990 Perspectives in Biochemistry

noSO

Orientational Sta ies of Water

noso

Additional Orientational States (at High Temperature)FIGURE : Iceberg model for the large heat capacity of transferof nonpolar solutes into water (Frank & Evans, 1945; Gill et al., 1985) .At low temperatures (near room temperature for benzene, for ex-ample), the water molecules surrounding the nonpolar solute adoptonly a few orientations (low entropy), to avoid w asting hydrogenbonds; thus, all water configurations are fully hydrogen bonded (lowenergy). At higher temperatures, more conformations are accessible(higher entropy), but some of them have weakeror unformed hydrogenbonds and/or van der Waals interactions (higher energy). Thiscontributes to the heat capacity because the system energy increaseswith temperature.These three expressions have identical content. Which formis used depends on which set of parame ters is most convenient:Ordinary thermodynamic convention is to choose a singlereference temperature, rather than two, but the mixed ex-pression is included here because it has been used for proteinfolding (see below). Equation s 5-7 predict the type of tem-perature dependence shown in Figure 5A when ACp is largeor in Figure 5B when ACp = 0.What is the molecular basis for the large heat capacity oftransfe r of nonpolar solutes into water? Two observations havecontributed significantly to a molecular picture. First, theentropy and heat ca pacity of transfe r are linearly proportionalto the surface ar ea of the nonpolar solute (Miller & Hilde-brand, 1968; Gill et al., 1985; Jorgensen et al., 1985; Jolicoeuret al., 1986). Second, the large heat capacity of transfer forthe simplest solutes (Gill et al., 1985) decreases slightly withincreasing temperature. This leads to the view (Frank &Evans, 1945; Nemethy & Sche raga , 1962; Gill et al., 1985;Muller, 1990 ) that the organ ization of water molecules in thefirst shell surrounding the solute is like an iceberg, aclathrate, or a flickering cluster; see Figure 6. At roomtemperature, the water molecules surrounding the nonpolarsolute principally populate a low-energy, low-entropy state:the waters are ordered so as to form good water-water hy-drogen bonds. With increasing temperature, the waters sur-ronding the solute increasingly populate a higher en ergy, higherentropy state: they are less ordered and have weakened at-tractions. Henc e, increased tempera ture cau ses melting ofthe surrounding water structure, insofar as the entropy andenergy are increased. [This melting process is probably be tterrepresented as bent hydrogen bonds than broken bonds, sincethe bending energy is much sm aller than the breaking energy(Lennard-Jones & Pople, 1951).] The reason this results ina large heat ca pacity is that the two different energetic statesof water provide an energy storage mechanism. The higherenergy state becomes more populated with temperature. Thereason this heat capacity is so large per solute molecule isbecause each solute molecule is surrounded by a large numb er(more tha n 10) of first-shell water molecules, each of whichcan participate in this energy storage mechanism. No t yetknown is the detailed breakdown of the nature of these en-ergies, although they undoubtedly include some combination

( A c p , Ts,Th), ( A c p Ts,AH(Ts) ) , or ( A c p Thy B ( T h ) ) *

of solute-water and water-water hydrogen bonding and vander Waals and dipolar electrostatic interactions.What is the molecular interpretation of T, an d Th?Equation s 2-7 each have temper ature-dep endent and tem-perature-independent terms. The slope of the tempera turedependence in eqs 5-7 is given by AC,. T, an d Th can beinterpreted as representing a reference enthalpy or entropy atsome given temperature . T, and Th ar e diagnostic for liq-uid-state nonpolar transfer processes (Baldwin, 1986).Sturtevant (1977) observed that several different biomolecularprocesses at 25 C have nearly identical values of the ratioAS/AC,. Baldwin (1986) showed that the constancy of thisratio, taken together w ith T2= T , an d T = 298 K, substitutedinto eq 3 implies that these various processes can all becharacterized by a single temperature, T, = 114 O C . Usingthe data of Gill and Wadso (1976) for AS an d ACp for th etransfer of liquid nonpolar compound s to water a t differenttemperatures, T, substituted into eq 3, Baldwin again founda single characteristic temperature, T, = 112.8 O C , implyingthat Sturtevants biomolecular processes resemble nonpolarsolvation. Baldwin (1986 ) preferred the use of eq 5. For eqs5-7, Ts nd Th are convenient reference quantities because m,

AS, an d AC, depend linearly on solute surface ar ea, so ratiossuch as AS/C,, and hence Ts,should be independent of solutesize.On the other hand, Murphy, Gill, and Privalov (Murphyet al., 1990; Privalov & Gill, 1988) have preferred the con-vention given by eq 6. They assumed that at tempe rature T,nonpolar solvation is identica l with classical solvation an d tha tA H( T , ) is due only to van der Wa als forces. The y refer tothe factors other than AH( ,) in eq 6 as the hydration effect.They note that A H ( T, ) is a positive enthalpy disfavoringtransfer and t hat the hydration term is always negative (o rzero at T = T, ) . It follows that the hydration effect favorsnonpolar transfers into water. However, it is not clear thatthis separation into these molecular factors is warra nted. AtT = T,, nonpolar solvation is not identical with simple sol-

vation. Even though the entropy of transfer may extrap olateto zero at T = T,, the heat capacity remains large (Muller,1990). The heat capacity is probably a more fundamentalcharac teristic of nonpolar solvation than the en tropy, becausethe entropy depends on the choice of concentration units, asdo the fre e energy and partition coefficient, whereas that heatcapacity and enthalpy do not. In addition, these thermody-namic models predict that for T > T , the entropy of transferbecomes positive, leading to the question able prediction tha tthe entropy would then favor solvation at high temperatures(see Figure 5A for benzene above 400 K). Also, since waterhydrogen b onding persists to beyond 500 K (Crovetto et al.,1982; Franks, 1983), AH( T,) probably includes water-waterhydrogen bonding in addition to van der Waals interactions.Thus, it is not clear that this thermodynamic separation ofterms corresponds to a simple molecular picture. Equation7 provides yet a different view. Wh erea s AG for transferringbenzene to water is most positive at 7 T , (near 100 C),A G / R T is most positive at T = Th (near room temperature).A G / R T correspo nds directly to a so lubility, partition coeffi-cient, or Boltzmann population. Benzene is least soluble inwater around room temperature. In eq 7, -AS(Th)/R, rep-resenting water ordering, is the most positive contribution toA G I R T . The remaining terms, whatever their molecularinterpre tation, favor solvation. T hus at present, Thand T, serveto identify nonpolar transfer processes, but their breakdowninto molecular components awaits further theory and exper-iments.


11/23

Perspectives in BiochemistryThe th erma l unfolding of proteins shows impo rtant simi-larities and important differences in comparison with thesolvation processes of small nonpolar solutes. Th e similarityis that protein unfolding involves a large increase in heatcapacity,2 characteristic of nonpolar exposure (P ace, 1975;Privalov, 1979; Baldwin, 1986; Becktel & Schellman, 1987;Schellm an, 1987; Privalov & Gill, 1988; Ooi & Oobatake,1988). Therefore, the enthalpy and entropy of folding stronglydecrease with temperature, and the free energy is curved:maximum protein stabilities are in the ran ge of 0-30 C (seeFigure 8). The h eat capa city of protein unfolding is itselfapproximately independent of temperature, although it appearsto decrease somewhat at higher temperatures (Becktel &Schellm an, 1987; Gill & Privalov, 1988).The thermal unfolding of proteins differs from nonpolarsolvation in the particular values of entha lpy and entropy atany given tempera ture. For example, as noted above, at 25

O C , the transfer of small nonpolar compounds into water hasapproximately zero enthalpy change and a large negativeexcess entropy chan ge. In contrast, a t 25 C ,protein unfoldinghas a positive entha lpy change (except for myoglobin for whichthe enthalpy is about zero) and a small or positive excessentropy (Baldwin, 1986; Privalov & Gill, 1988). There is anadditional positive entropy a nd enth alpy of unfo lding, in excessof that predicted from nonpolar solvation experiments.What is the origin of the residual enthalpy and entropy offolding? As noted in section 2, the residual enth alpy of un-folding becomes more positive with increased content of polarresidues (Privalov, 1979; Privalov & Gill, 1988; Dill et al.,1989). If it is assumed to be approximately temp eratu reindependent (Baldwin, 1986), then this result is expected fromclassical polymer solution behavior wherein chain m onomersof type A have normal dispersion-force-driven ncompatibilitieswith solvent B (Flory, 19 53). This enthalpy may arise fromvan der Waals or hydrogen-bonding interaction s among th ebackbone or polar residues. However, thermodynam ic modelsof protein unfolding have generally been based on the as-sumption th at th e large heat ca pacity change is fully attrib-utab le to nonpolar solvation. This is open to question. Hy-drogen bonding to water weakens with temperature, accordingto valerolactam dimerization and thymine dissolution exper-iments (Alvarez & Biltonen, 1973); this would also contributeto a change in heat capacity. Also, increased temp eratu recauses the unfolded states of proteins to expand, reducingnonpolar solvation an d contributing ad ditional temperaturedependence to the enthalpy and entropy of folding (Dill et al.,1989).The large residual entropy represents an increased disor-dering upon unfolding, relative to that expected for nonpolarsolvation. In sections 5 and 6 , it is suggested that at least alarge compon ent of this is due to a difference in the freedomof configu ratio ns of the chain backb one, more severely re-stricted by steric constraints in the folded than in the unfoldedstates. However, ther e may also be an extra residual entropydue to additional configurational restrictions of the side chainsin the folded state. The side chains may be partially frozenif the folded protein resembles a solid-like state (Shakhnovich& Finkelstein, 198 9a,b). The residual entropy of proteindenaturation is ASr(112 C )= 18 J/(K.mol) (Baldwin, 1986;Privalov, 1979; Murph y et ai., 1990), similar to that of thedissociation of solid diketopiperazines, ASr( 12 C) 16J/(K.mol), and considerably different than that of liquid

Biochemistry, Vol. 29, No. 31, 1990 7143hydrocarbon dissolution for which AS,(112 C ) = -0.5 J/(Kamol) (Murphy et al., 1990). The hea t capacity changeincrement upon dissolution of the solid diketopiperazines isthe sam e as for the liquid-state transfe r process (M urph y &Gill, 1989a,b). From those thermody namic experiments,however, it is not possible to determine how much of theentropy difference originates with the chain expansion andsolvation and how much originates from side-chain unfreezing.Bendzko et al. (1 986) have suggested tha t if side-chain freezingis important, then protein den aturation should lead to an in-crease in partial molar volume of the protein; instead, theyand others (see section 5 ) find a decreased volume upon un-folding. Ano ther test of side-chain restrictions in the foldedcore is to compare crystal stru ctur e distributions of side-chainrotamers (Ponder & Richards, 1987) with computer simula-tions of the mean position and fluctuations of side chains thatare attache d to spatially unconstrained backbones. Suc hcomparisons (Ja nin et al., 1978; Piela et al., 198 7) show thatequilibrium side-chain positions are predicted relatively wellby the unconstrained simulations, but they a re constrainedsomewhat differently in different secondary structures (Pielaet al., 1987; McGregor et al., 1987).

Another class of experiment bears on the issue of th e sol-id-like vs liquid-like nature of the protein core. Wh erea s thecore is solid-like in its density and compressibility, it maybehave differently insofar as the transfer process is concerned.Th e experiments involve multiple amin o acid substitutions a ta given site and the measurement of the change in proteinstability, A A G , ue to each of the replacements (Yut ani et al.,1984, 1987; Mats umu ra e t al., 1988a,b; Kellis et al., 1989;Sandberg & Terwilliger, 1989). Th e slope of A A G vs the freeenergy of transfer AG, for the corresponding amino acid fromwater to oil (see Figure 4), provides information about sim-ilarities and differences of the protein folding process relativeto the simpler process of amino acid transfer from water intoliquid oil. This slope depends on a combination of factors: (i)the deformability of the native-state cavity, the energeticand entro pic constraints affecting th e freedom of th e cavitywall residues to move to accomm odate the muta ted am ino acid;(ii) the interactions, w, f the re sidue with the cav ity, includingboth its entropic restrictions (side-chain freezing) and theresidue/cavity energetics; (iii) the degree to which the dena-tured-state environment of the specified residue resembles puresolvent; an d (iv) the exposure of the residue in the native sta te.A slope of 1 is consis tent with a process in which the aminoacid is exposed in the unfolded state and is transferred intoa native cavity that resembles the reference liquid oil. Simplesolute transfer to a liquid involves (a) opening the cavity(unfavorable by approximately free energy w/2) nd then (b)transferring th e solute (favorable by a free energy, w). Onthe other hand, if a solute is transferred instead into an alreadyopened cavity (preformed), then only (b) is involved (Kelliset al., 1989). Thus, the free energy of transfer into a simpleliquid is only half that of transfer into a preformed cavity.Therefore, a slope of 2 is expected from these experiments ifthe cavity wall residues ar e so constrained that they do notmove to accommoda te the muta nt residue. Protein cavitiesmay differ. Som e may have much deformability, as in aliquid, and lead to slopes near 1. Othe rs may have littledeformability, if n eighboring residues ar e constrained, and leadto slopes near 2. Cavity deformabilities may also depend onresidue size; cavity wall residues may move to accommodatelarge occupa nts but not small ones, for example. Overall,slopes ranging from 1 to 2 ar e expected from factor (i). Slopessmaller tha n 1 imply that the residue may be exposed in the

On e interesting c ounterexam ple, however, is a highly stable phos-phoglycerate kinase from a thermostable bacterium (Nojima et al., 1977)in which both the enthalpy and heat capacity of folding are small.


12/23

7 144native state or may be significantly buried in the denaturedstate. Slopes considerably greater tha n 2 must arise at leastpartly from (ii). A negative slope could arise in principle ifthe residue is more exposed at the surface of the nativestructure than it is in the denatured state. A slope of ap-proximately 1 is observed for residue 3 in T4 lysozyme(Ma tsum ura , 1988a). A slope of approximately 2 is observedfor residues 88 and 96 i n barnase and for a pocket involvingresidues 35 and 47 of gene V protein of phage f l (Ke llis etal., 1989; Sandberg & Terwilliger, 1989). For two sitesnormally containing charged residues, Asp-80 in kanamycinnucleotidyl transferase (M atsum ura et al., 1988b) and Glu-49in Tr p synthase (Yu tani et al., 1984, 19 87), the slopes varywith pH, reaching a maximum of about 3.8. These latterexperiments are more complex insofar as the proteins havesignificant populations of intermediates (and do not havetwo-state transitions) and they have an electrostatic componentfor the transfer. Although it would appear that curren t resultson noncharged residues can be explained largely by differentcavity deformabilities (i), nothing rules out the possibility thatthe other factors (ii and i i i ) are importa nt. This type ofexperiment has not yet established whether side-chain freezingis an important component of protein stability.Which liquid oil best characterizes the native core? Whilethere is some evidence that cyclohexane is good (Radzicka &Wolfen den, 1988), the best correlations of transfer stud ies withamino acid distributions in proteins appear to involve nonpolarhydrogen-bonding solvents including octanol, ethanol, anddioxane (Faucher e & Pliska, 1983; Nozaki & Tanford, 1971;Kyte & Doolittle, 1982; Rose et a]. , 198 5a). It may be,therefore, that some fraction of the temperature-independententhalpy, attributed to van der Waals and hydrogen-bondinginteractions, is also present i n these transfer experiments andthat it contributes differently for different oils.Finally, I return to the meaning of hydrophobicity. I believethe most useful, common (Tanford, 1980; Ha et al., 1989),and unambiguous meaning of this concept is simply in ref-erence to nonpolar transfers from nonaqueous media intoaqueous media: (i) tha t are strongly disfavored and (ii)whenever there is a large associated increase i n heat capacity[definition (2) at the beginning of this section]. It was thisremarkable feature of nonpolar solvation that was first iden-tified as unusual (Edsa ll, 1935; Butler, 1937; Frank & Evans,1945) and which merits special terminology. Definition ( l ) ,on the other hand, needs no special term because it otherwisedescribes ordinary solution processes. There are two problemswith (3) , hydrophobicity as water ordering or other molecularmodels: (i) the molecular mechanism is still not fully un -derstood, and ( ii) water ordering is an appropriate descriptionof the entropic repulsion of nonpolar solutes near room tem -perature, but not over a broader tem pera ture range. Thiswould lead to unnecessa ry hairsplitting : benzene insolubilityin water would be referred to as hydrophobic at 25 C but nota t 100 C, where it is even more strongly expelled. Thismeaning is not subject to the Hildebrand objection (1968,1979 ); he noted tha t hydrophobicity is not an enthal pic dis-affinity of nonpolar solutes for water but instead is due to awater-water affinity. Also, because definition (2) describeshydrophobicity in terms of the full transfer process, representedby the total free energies in eq s 5-7 rather than by a particularterm in those expressions, it is not subject to difficulties ofmolecular interpretations (Privalov et al., 1990; Dill, 1990).( 5 ) WHAT s MISSING?

Th e dominan t force of folding is only half the story. Nea rlyequal i n magnitude is a large opposing force. The structure s

Biochemistry, Vol. 29 , N o. 31, 1990 Perspectives in Biochemistryand stabilities of globular proteins result from the balancebetween driving and opposing forces. Only recently have themain opposing contributions become better understood. Th athydrophobicity is not the whole story becomes immediatelyappa rent from certain puzzles. First, Tanford ( 1 962) andBrandts (1964a,b) showed tha t hydrophobicity alone wouldpredict protein stability an ord er of magnitude greater thanmeasured values. They estimated that the free energy ofunfolding would be abou t 100-200 kcal/m ol a t 25 OC. Thisestimate is based on free energies of transfer of hydrophobicamino acids from water into ethanol or dioxane, representativeof the folded core of the protein, multiplied by the number ofnonpolar residues. Howeve r, free energies of unfolding areobserved to be only abou t 5-20 kcal /mol (Pace, 1975; Frivalov ,1979; Privalov & Gill, 1988). This implies that there mustbe a large force, of magnitude nearly equal to the hydrophobicdriving force, which opposes folding. Second, while the tem-perature dependence of folding was an important clue forhydrophobicity, Tan ford (1 962 ) suggested t hat it also poseda paradox. If nonpolar components associate more stronglyas the temperature is increased, then proteins should fold moretightly with increasing tempe ratur e. Just the opposite is ob-served above room temp erature: increasing tem perature un-folds proteins. Proteins are typically thermally unfolded inthe range of 50-100 C; the free energies of transfer of hy-drocarbons into water have a maximum extrapolated to be inthe range of 130-160 C. Third, if hydrogen bonding is nota dominant force of folding, then what is the origin of theconsiderable amounts of internal architecture, of secondaryand tertiary structures, in proteins? Through what forces dothe amino acid sequences so uniquely determine the nativestruct ure? The hydrophobic effect would seem to be toononspecific and an unnatural candidate as the origin of helicesand sheets.

Fourth, the pressure dependence of protein stability doesnot resemble that of model hydrophobic compounds (Brandtset al., 1970; Zipp & Kauzmann, 1973; Kauzma nn, 1987). Forexample, the partial m olar volume of m ethane decreases from60 to 37.3 mL/mol upon transfer from hexane to water(Masterso n, 1954). Th e decreased volume arises becausewater molecules pack more efficiently surrounding a nonpolarsolute molecule than in its absence. Sinc e the unfolding ofa protein leads to increased nonpolar exposure to water, thenthese model studies would suggest that the partial molarvolume of a protein should decrease considerably upon un-folding du e to similar contraction of solvating water molecules(by about 20 mL/mol multiplied by a number in the rangeof 10-40, representing the total hydrophobic exposure uponunfolding). While the volume chan ge of protein unfolding isindeed generally observed to be negative, it is only in the rangeof -30 to -300 mL /mo l (abo ut 0.5% of the total volume)(Bra ndts et al., 1970; Zipp & Kauzm ann, 1973; Edelhoch &Osborne, 1976; Richards, 1977), which is somewhat smallerthan the methane model would suggest. Brandts et al. (1970)suggested that other simple factors would not acc ount for thisdiscrepancy; for example, if unfolding leads to exposure ofcharged groups, then the volume change upon unfolding wouldbe predicted to be even more negative, increasing the dis-crepancy. There is an additional problem. For model com-pounds, increasing the pressure leads to more normal watersolvation, so the volume change of transfer to water d iminishesand ultimately becomes positive a t about 1500-2000 atm . Forproteins, on the other h and , the negative volume of unfoldingdoes not change much with pressure. Two simple explanations,however, can account for these discrepancies between the


13/23

Perspectives in Biochemistrymodel transfer experiments and protein unfolding. First,me than e is a poor model ami no acid. Better models includealcohols, ketones, and amides, which c a n form hydrogen bonds;these model compounds have much smaller negative volumesof transfer to water, more closely predictive of the proteinexperiments (Friedman & Scheraga, 1965; Hvidt, 1975).Second, the pressure dependence is at least qualitatively ac-counted for by recognizing (i) t ha t model nonpolar solutes inwater a re less compressible than in the pure liquid (Brandtset al., 1970) and (ii) that the folded state of a protein is muchless compressible than the reference liquid hydrocarbon towhich it is gene rally compared. A folded protein is typically10-fold less compressible than organic liquids (or about halfas compressible as ice!) (Ku ndro t & Richards, 1987; Gavishet al., 1983; Eden et al., 1982; Klapper, 1971; Fahey et al.,1969). Due to (i), the volume change in small-moleculetransfers should diminish with increasing pressure; due to (ii),the volume change in protein folding should diminish muchless with increasing pressure than in the small-molecule ref-erence experiment. Therefore, it is important to recognize thata protein is not jus t a sum of transfers of sma ll-molecule modelside chains. Proteins are polymers. It is described below howthe chainlike n ature of proteins and the resultant conforma-tional freedom lead to a strong force that opposes folding.( 6 ) PRINCIPLE OPPOSING FORCEIS ENTROPIC

Since the 1930s it has been known that the main forceopposing protein folding is entropic. Nor throp (1932) was thefirst to observe a sharp thermal denaturation transition; theequilibrium constant depends strongly on temperature. Onlymore recently has the molecular basis for the opposing entropybecome clear. Just as there ar e translational, rotational, andvibrational entropies of small molecules, depending on therelevant degrees of freedom, likewise there are differentpossible molecular origins of the entropy gain upon proteinunfolding. For example, Mirsky and Pauling (1936) suggestedthat folding would be opposed by an entropy arising from theproper mating of specific ion pairs. As anoth er example, thehelix-coil transition theories (Schellm an, 1958a ; Zimm &Bragg, 1959; Poland & Scheraga, 1970) showed that localdegrees of freedom could be an important source of entropyopposing helix formation.However, it has long been known that polymers are alsosubject to anot her type of configurational entropy, one whichis nonlocal (Flory , 1953; deGennes, 1979). It arises fromexcluded volume, the impossibility that two chain segmentscan simultaneously occupy the same volume of space. A chaincan occupy a large volume of space in any of a large numberof different configurations. However, there are relatively fewways the chain can configure if it is constrained to occupy asmall volume of space, simply du e to severe steric constraints.Excluded volume (steric constraints) will play a role in anyprocess tha t involves a chang e in the spatial d ensity of polymersegments. These include solution thermodynamic propertiesof chain molecules, particularly their dependence on concen-tration, including solubilization and phase

Dominant Forces in Protein Folding_Biochem_1990

Documents