Mass Spectrometric Approaches Using Electrospray Ionization Charge States and Hydrogen-Deuterium Exchange for Determining Protein Structures and Their Conformational Changes*

Electrospray ionization (ESI) mass spectrometry (MS) is a powerful analytical tool for elucidating structural details of proteins in solution especially when coupled with amide hydrogen/deuterium (H/D) exchange analysis. ESI charge-state distributions and the envelopes of charges they form from proteins can provide an abundance of information on solution conformations that is not readily available through other biophysical techniques such as near ultraviolet circular dichroism (CD) and tryptophan fluorescence. The most compelling reason for the use of ESI-MS over nuclear magnetic resonance (NMR) for measuring H/D after exchange is that larger proteins and lesser amounts of samples can be studied. In addition, MS can provide structural details on transient or folding intermediates that may not be accessible by CD, fluorescence, and NMR because these techniques measure the average properties of large populations of proteins in solution. Correlations between measured H/D and calculated parameters that are often available from crystallographic data can be used to extend the range of structural details obtained on proteins. Molecular dynamics and energy minimization by simulation techniques such as assisted model building with energy refinement (AMBER) force field can be very useful in providing structural models of proteins that rationalize the experimental H/D exchange results. Charge-state envelopes and H/D exchange information from ESI-MS data used complementarily with NMR and CD data provides the most powerful approach available to understanding the structures and dynamics of proteins in solution.

The globular or braided topology with attendant surface crevices and interior cavities creates a unique three-dimensional structure for proteins and, together with their dynamic properties mediated through local fluctuations or large-scale conformational changes, the singular functions of these lifesustaining biomolecules are executed. Conformations and conformational dynamics are key to the protein's functional integrity and these characteristic properties are highly dependent on environmental conditions. Thus, varying physiological conditions may play an important role in protein conformational changes. For example, pH adjustments in organisms can drive ligand-receptor dissociation and receptor recycling (1). Protein conformational changes are also manifest in the protein life cycles from expression to final post-activity phase, i.e. birth, function, and death (2), and improper protein conformations may be responsible for a number of diseases (3). Detailed information is required on protein higher-order structures and dynamics to fully understand how proteins perform their biological functions at the molecular level.
Protein conformational changes can be monitored by conventional biophysical and analytical methods, including circular dichroism (CD) 1 (4), tryptophan fluorescence (5), and infrared spectroscopy (6). Far-ultraviolet (UV) CD spectroscopy, for instance, provides information mainly on secondary structural elements of the polypeptide chains; near-UV-CD detects changes in the tertiary structure around aromatic amino acid side chains, and fluorescence techniques may be another more sensitive technique for this purpose. The low-resolving power inherent in these methods, however, limits the investigator's ability to focus on regions where conformational changes might occur. Detailed structural information at individual residues of a protein can, of course, be obtained from x-ray diffraction (7) and nuclear magnetic resonance (NMR) spectroscopy (8 -11). However, obtaining a suitable crystal for x-ray diffraction analysis or finding a compatible solvent for proteins, typically less than 30 kDa, to give high enough concentrations for NMR studies is often an obstacle with these methods. Both methods also require relatively large amounts of protein of high purity. Moreover, the conditions required are generally not compatible with the physiological environment of functional proteins, and the static picture of their structures usually is not sufficient for understanding the dynamics associated with performance of their functions.
The advantages of modern mass spectrometry (MS) for studying protein structure and dynamics are its high mass measuring accuracy and its sensitivity. Electrospray ionization (ESI) MS has become a powerful analytical tool not only to identify proteins and their post-translational modifications but also to analyze their higher-order structures, dynamics, and conformational changes, especially when combined with hydrogen/deuterium (H/D) exchange techniques. H/D exchange coupled with ESI-MS provides a relatively fast and sensitive approach for studying global and regional conformational changes as well as regional dynamics on small amounts of protein.
The presence of several conformations in equilibrium with different kinetic rates of interchange that depend on environmental conditions represents a typical scenario by which the biological functions may be governed. Thus, the stability, dynamics, and folding pathways of proteins are often studied by examining them under a variety of conditions to mimic protein-native or folded state and/or conformationally altered states. MS is uniquely suited for monitoring co-existing conformations of proteins, regardless of size and physical properties. These are issues that must be addressed when considering x-ray analysis or NMR spectroscopy for similar studies. However, there are potential pitfalls in using MS, and thus it should always be considered as a complementary tool to other well-established biophysical methods for elucidating protein structure and performing conformational analyses. In this article, an attempt is made to summarize some recent developments in methods for determining protein solution phase conformational properties by MS alone or coupled with H/D exchange techniques.

CONFORMATIONAL CHANGES BY CHARGE-STATE ANALYSIS
Multiply protonated protein ions are generated in solution and transferred into the gas phase of the mass spectrometer during the ESI process. The observed charge states or charge-state distributions (CSD) of proteins in mass spectra are influenced by various experimental conditions, including pH (12)(13)(14), temperature (15,16), solvent composition (13,17), and instrument settings (18). In addition, the intrinsic properties of a protein such as its gas-phase chemistry (19), sequence (20) and conformations (21,22) are important. The latter contribution to the CSD pattern is the basis for current applications and it may provide an alternative method to CD, for example, for determining overall protein conformational states. However, controversy in using ESI for these application still reigns, particularly with regard to whether there is a correspondence between protein solution and protein gasphase conformations subsequent to the ESI process and whether different conformations can actually be represented by charge-state envelopes.
Desorption and transport of multiply charged biomolecules from liquid droplets in the ion source to the gas phase have been studied extensively, but the influence on the threedimensional structures of proteins during the electrospray process is still unclear. In many cases, ESI mass spectra appear to reflect the expected solution phase conformations rather well (23)(24)(25)(26)(27)(28). For example, the N-terminal domain of cardiac muscle troponin C by Fourier transform ion cyclotron resonance MS (23) exhibited the same charge states and relative intensities when electrosprayed either from water in which the protein was in its native state or from 70% methanol/water/0.25% acetic acid solutions in which it had greater helicity. The identical charge states for the different conformations in these two solutions allowed a direct comparison of the conformations in the gas phase by H/D exchange. From these results, the authors were able to conclude that gasphase protein ions can retain at least some memory of their solution-phase conformations (24). On the other hand, other studies suggest there is little correspondence (25)(26)(27) between gas-and solution-phase conformations. Thus, gasphase H/D exchange profiles of lysozyme electrosprayed from water in which the protein should be in its native state, or from water/methanol 20:80 solution in which it should exist as a denatured helical state, showed little difference between the conformations from these two solutions. Apparently, unfolded lysozyme conformations in solution refold to compact ions in the gas phase (28) because of the presence of four intact disulfide bonds that allowed the protein to retain sufficient structure that it refolded quickly in the gas phase prior to deuterium exchange. The pH of the solution from which the protein is electrosprayed is probably important too, as it was reported that at neutral pH the charge-state distributions of ␣-dendrotoxin reflect the solution structure exactly, but under acidic conditions the solution conformations are only partly retained (29).
The relationship between charge states in the gas phase and those in solution has also received a considerable amount of attention (17, 23, 30 -33). The extent of protonation observed in an ESI spectrum is proposed to be dependent on the steric accessibility of basic amino acids (Lys, Arg, His, and the N terminus) to protonating solvents before entering the gas phase (34). Excellent agreement has been observed between the maximum charge states and the numbers of basic residues for many proteins (20,34). Recent studies suggested that the maximum number of charges and the CSDs depend on the relative gas-phase basicities, i.e. the solvents with higher gas-phase basicities have a propensity to remove protons from higher charge state ions and therefore influence the CSDs (35). For the most part, mass spectrometrists have adopted the view that charges on a given protein are not lost during desolvation in the gas phase and that the charge states observed in the gas phase are a reliable indicator of the charges on the protein in solution.
It is the consensus that ESI mass spectra of proteins in their unfolded states show higher multiply charged ion envelopes (lower m/z value) than the corresponding folded or compact forms (36), although there may be exceptions (28). In solution, proteins fold and unfold continuously, even under conditions that approach those of physiological environments, i.e. temperature, pH, and ionic strength, resulting in the continuous presence of transitional intermediates (37). However, a favored conformation governed by environmental conditions may represent the bulk of the protein population. The analysis of protein CSD patterns in ESI spectra is generally considered to produce only qualitative, low-resolution information. The numbers of conformational states may not be resolved by the numbers of charge-state envelopes in this straightforward method because the same degree of charging may occur with different conformational states (38). The mathematical deconvolution of the charge-state envelopes may help to enhance the structural details. Thus, the evolution of partially unfolded intermediates along with the folded and unfolded forms for holo-myoglobin and apo-myoglobin as a function of pH could be studied by ESI charge-state envelopes (39).
In practice, the monitoring of transitions between the native and denatured states of a protein in solution by charge-state envelope analysis can be very useful if the system is relatively uncomplicated, i.e. a two-state folding/unfolding model (40) with no intermediate states. Most single-domain proteins are believed to approach closely this ideal (41), and CSDs can be used to determine the ratio of folded and unfolded states as well as the kinetics of unfolding and protein stability. Thioredoxin (TRX), for example, has 13 basic sites distributed throughout its 108-aa residue sequence that are possible sites for protonation. The maximum charges observed in the ESI spectrum for unfolded TRX is 15 ( Fig. 1). Clearly, there are at least two other sites with sufficiently high proton affinity to be observed under ESI conditions. The analyses of TRXs (oxidized [Oxi-TRX], reduced [Red-TRX], C 32 -ethylglutathionylated [GS-Et-TRX], and C 32 -ethylcysteinylated [Cys-Et-TRX]) by ESI charge-state envelopes produced melting curves that provided substantial information on the structures of these proteins ( Fig. 2A) (42). The results indicate that Oxi-TRX is most stable (T m ϭ 67°F) while the reduced form and its derivatives show similar stabilities with melting temperature T m ϳ 54°C. There is reasonable agreement in the stability order of TRXs, but the melting temperatures are 3-10°C lower by near-UV-CD than when measured by ESI chargestate envelopes (Fig. 2B). The near-UV-CD signal represents the total of four aromatic amino acids in TRX, i.e. two tryptophans in the active site region and two in ␣ helices remote from the active site. The gradual decrease in molar ellipticity is likely due to minor changes in the asymmetric environment near the tryptophans and tyrosines as a result of increased There is significance to the fact that Red-TRX and Cys-Et-TRX show similar melting curves, whereas GS-Et-TRX has an apparent folded/unfolded composition more like that of Oxi-TRX in the temperature region T Ͻ T m , but nearly identical to that of Red-TRX in the region T Ͼ T m . Modeling of GS-Et-TRX by assisted model building with energy refinement (AMBER) force field suggested that two salt bridges are formed between the carboxylates of the ␥-glutamyl and glycylyl residues of the glutathionyl moiety and the basic residue side chains of Arg 73 and Lys 90 , thereby imposing structural compactness and restriction to protonation in the folded form (43). Such salt bridges are not present in Cys-Et-TRX, and the modification has no greater influence on the melting curve than that observed in the reduced from, i.e. Red-TRX.
Thus, it can be concluded that the charge-state envelopes reflect the overall conformational properties of these proteins whereas near-UV-CD is more descriptive of localized environ-mental conditions. The dramatic changes among the different temperature-dependent ellipticity profiles of TRXs appear to be most affected by changes in the active site region. The near superimposeability of the GS-Et-and Cys-Et-TRX profiles reflect common environmental conditions with changing temperature regardless of the nature of the alkylation, but the unusual profile for Red-TRX points to a more fundamental phenomenon. Generally, when melting profiles are not superimposable, transient intermediates are suspected. Indeed, H/D exchange in conformations above the T m in the unfolding profiles showed minor peaks between the protonated and deuterated peaks (42), indicating that none of the TRXs follow a true cooperative two-state unfolding model.
Organic solvent-or pH variation-induced protein unfolding can be observed by CSD analysis as demonstrated with ferricytochrome c (21). In 3% methanol near neutral pH, cytochrome c displayed a native folded conformation with an envelope centered on the ϩ8 charge state (Fig. 3A). At pH 4.2, the protein apparently retained a native-like conformation as indicated by the ESI spectrum but the envelope is now cen- tered on the ϩ9 charge state (Fig. 3B), indicating perhaps some loosening or a "secondary" solvent effect, but no real unfolding. A distinct second envelope is centered about the ϩ17 charge state for the unfolded state at lower pH ( Fig. 3, C and D). In 50% methanol and pH 3.8, the appearance of the folded charge-state envelope centered at ϩ10, and an unfolded state at ϩ14, indicated there was a different conformation present (Fig. 3E). Far-UV-CD showed retention of secondary structure in 50% methanol, whereas both tertiary and secondary structural breakdown occurred under acidic conditions in 3% methanol. This is a clear illustration that ESI charge-state envelopes can detect the presence of several protein conformers under a given set of experimental conditions. ESI-MS charge-state envelopes were also used to show the collapse of the tertiary structure of Zn 2ϩ complex colicin E9 endonuclease upon acid-induced release of the metal ion. Moreover, these studies revealed that the noncovalent protein-protein complexes, i.e. colicin E9 endonuclease and its cognate immunity protein, Im9, dissociate in the gas phase before the metal ion complex does, thereby confirming that electrostatics and not hydrophobic interactions are more important in the gas phase (44). Another example of the power of charge-state analysis to elucidate structural properties of a protein where other methods fail is illustrated by studies on human recombinant macrophage-colony stimulating factor ␤ (rhM-CSF␤) (45). This compact homodimer with nine disulfide bonds exhibits a charge-state envelope centered on ϩ21 with a maximum of ϩ29 charges (Fig. 4A). Selective reduction of two disulfide bonds (C 157, 159 -CЈ 157, 159 ) near the C-terminal tail showed little change in the secondary structure as suggested by far-UV-CD and fluorescence spectra, and even the three-dimensional integrity of the protein was unaltered as suggested by the unchanged biological activity. However, the ESI spectrum clearly revealed there were major changes as the charge states increased to a maximum of ϩ41 (Fig. 4B), suggesting that six additional basic sites became accessible in each subunit by removal of two disulfide bonds. However, only Lys 154 would appear to become available for protonation upon reduction of the disulfides, C 157, 159 -CЈ 157, 159 , as all other basic sites are in the N-terminal direction relative to the next stabilizing disulfide bond, C 102 -C 146 . Apparently, an overall loosening of the structure occurred as will be discussed later, thereby causing greater exposure of additional basic residues to the solvent.
Clearly ESI charge-state envelopes yield considerable information when proteins are denatured thermally or under acidic or varying solvent conditions and when chemical modifications are made as in the reduction of disulfide bonds. In fact, the method has advantages over CD because specific chromophores on the protein are not required and multiple interactions and conformational changes can be observed simultaneously. The tertiary structure and changes to the structure are the important features of proteins in the performance of their natural functions, and the simple but sensitive ESI charge-state envelopes can fortuitously reveal some of the details. Although ESI experiments can reveal structural changes in proteins affected by temperature, pH, and disulfide cleavages, the effects of denaturing agents, urea, and guanidine hydrochloride on protein unfolding or conformations cannot be determined directly by this method because of the intolerance of ESI for salts. When examining conformations or conformational dynamics under the influence of chemical denaturants, CD may be the better method, but If ESI-MS has the capability to probe protein conformational dynamics through analysis of CSDs that is at least as good as any other method, then the power of MS to probe conformational changes of proteins is advanced still further when H/D exchange and protein fragmentation techniques are applied. While charge-state analysis provides information on the overall compactness of the protein, regional or site-specific amide hydrogen exchange can be used to probe localized conformations and conformational dynamics. Generally, the level of exchanged hydrogens reflects the relative openness, solvent accessibility, or hydrogen bonding strength in protein structures (46 -49).
The H/D exchange method was introduced by Linderstrom-Lang and coworkers four decades ago (50). Since then, many fundamental studies have been undertaken directed at understanding protein dynamics using various analytical techniques to measure hydrogen isotope exchange. Among them, NMR has played a central role and continues to do so as newer high-field instruments and chromatographic techniques are applied. In the late seventies, hydrogen isotope exchange experiments involved tritium labeling and radiation counting to probe protein structure and dynamics (51). The techniques were developed further by Englander and coworkers, with low temperature (0°C) and adjusted pH (2.4 -2.8) to minimize the loss of isotopic label during the high-performance LC step (52). Deuterium was substituted for tritium when NMR, and later MS, emerged as tools in the analyses of H/D levels. MS provided a particularly important dimension as it allowed measuring accurate molecular masses of fragmented peptides and proteins, thereby providing information on the numbers of hydrogens that had been exchanged.
When protonated proteins are solubilized in a large excess of deuterated solvent, the hydrogens on proteins including those at amide sites, N and C termini, and other functionalities, e.g. in side chains containing oxygen, nitrogen, or sulfur, undergo exchange. The rate of exchange of amide hydrogens varies over a range 10 8 depending mostly on their solvent accessibility and hydrogen bonding status, while hydrogens on functional groups exchange too fast to be measured conveniently by most techniques. These back exchange quickly when exposed to excess protic solvents as, for example, during high-performance LC separation, but amide hydrogen isotopes remain (52). The information on protein backbones is then provided by the presence of the hydrogen isotope on the amide nitrogens.
The application of MS to measure H/D content in peptides and proteins was first introduced by Katta and Chait in 1991 (53). The mechanism of amide hydrogen exchange, which can be either acid-or base-catalyzed (54), is well understood from experiments performed with extended, unstructured peptides in solution. The exchange reaction is slowest at pH 2-3, which is about 10 3 -10 4 times slower as compared with the rate around pH 7 (55,56). The exchange reaction rate is also temperature-dependent, decreasing by about a factor of 10 as the temperature is reduced from 25 to 0°C. Generally, under pH 2-3 and 0°C, which is commonly referred to as "quench conditions," the half-life for amide hydrogen isotopic exchange in an unstructured polypeptide is 30 -90 min. This sensitivity to pH and temperature is the basis for conducting exchange under a given set of conditions and arresting it under another, thereby allowing the sample to be prepared for analysis by ESI-MS or other mass spectrometric approaches. It should be noted that the mechanism of amide hydrogen isotope exchange in the gas phase is thought to be very different from that in solution phase, making it difficult, if not unrealistic, to compare the processes in the two phases (57)(58)(59)(60).
Over the past three decades, H/D analysis by NMR, which also measures amide hydrogen isotope levels, has been the single most powerful approach for determining protein conformation and dynamics in solution (8,61). It has, in fact, become the standard for comparison when other techniques are applied. Recently, however, H/D analysis by MS has proved to be complementary to NMR and x-ray crystallographic methods for studying protein conformations and conformational changes. Dobson and coworkers were the first to show the unique advantages of ESI-MS analysis of hydrogen isotope ratios following pulse labeling when they performed a nonequilibrium time course study on the refolding pathway of hen egg-white lysozyme (62). Their studies showed fully protected, unprotected, or partially protected transient folding intermediates that, together with NMR rate data, were interpreted as involving cooperative folding of ␣ and ␤ domains in which the ␣ domain folds independently in an initial fast phase, and a sequential folding of the ␣ and then the ␤ domain follows in slower phases. NMR data alone could not have distinguished between this mechanism and one involving independent folding pathways for the two domains in the fast kinetic phase because the technique measures labeling isotopic levels averaged over all populations of protein in the sample. Further studies that coupled hydrogen exchange and MS proved that the ␣ domain is destabilized both at high temperature (50°C) (63) and when one of the disulfide bonds (C 6 -C 127 ) is reductively alkylated (64).
Smith and coworkers (65) established a general procedure (Fig. 5) for protein H/D exchange and digestion experiments in applications that utilize ESI-MS for analysis. This technique made it possible to probe protein conformational changes with a spatial resolution of better than 5-10 residues when not limited by proteolytic cleavages. Overall, deuterium levels in peptic peptides can be obtained from the increase in molecular mass after exchange-in. The centroids of the isotopic clusters are used to determine the backbone amide site H/D content, and the deuterium levels at individual amide sites can in principle be obtained in the same way when collisioninduced dissociation (CID) tandem MS (MS/MS) is used for sequencing.
CID MS/MS-fragmented peptides, dominated by cleavage at the peptide bonds resulting in b n and y n ions, yield the sequence information and potentially the H/D content at individual amide sites. A reluctance to relying too heavily on CID MS/MS to measure the isotopic content at individual amide linkages after hydrogen isotope exchange stems from studies that showed scrambling of the label during the collision process. Thus, significant hydrogen scrambling was observed during H/D analysis on cytochrome c after activation by sustained off-resonance irradiation in the ion cyclotron resonance cell of a mass spectrometer (66). However, other studies (49,66,67) suggested that scrambling was minimal for short helical peptides under typical CID conditions in triple quadrupole instruments. Similarly, there appeared to be no evidence for hydrogen scrambling during fragmentation of transmembrane peptides during CID in a hybrid (Q-TOF) mass spectrometer, but subsequent model studies of transmembrane peptides incorporated in a lipid bilayer indicated that the extent of scrambling was dependent on the nature of the charge carrier and amino acid sequence (67,68). Alkali-cationized peptide ions showed less scrambling than protonated ions (69). Complete scrambling of label was reported when peptides from a study to determine the binding interface between urokinase plasminogen activator and its cellular activator were collisionally activated in an attempt to map the specific binding residues (70). Thus, although there are examples where mapping of site-specific events by amide H/D analysis have apparently been quite successful, all experimental results involving CID MS/MS in such studies must be viewed with skepticism until this issue is resolved.
Ion formation studies suggest that the production of y n ions during CID depend on the transfer of a proton from the leaving group, while the b n ions form a protonated oxazolodone structures without proton transfer (Fig. 6), i.e. the amide site to which the proton was originally transferred becomes the leaving group and is of no further consequence in the analysis (71,72). Yet, b n and y n ion formation via the "mobile proton model" may still constitute proton migration from a protonated amino acid side chain to carboxyl oxygens and amide nitrogen (73). In fact, proton transfer from the side chains of basic amino acids to the less-basic amide nitrogen of the peptide bond (74) may be a necessary step for creating the labile bonds that result in the production of b n ions. The scrambling problem has been attributed partly to the conditions used for CID processes, i.e. the applied collision energy and the time scale for ion activation and fragmentation (75). Space charge limits may also cause problems by shifting masses during analysis (76). Rapid activation to high energy states from collision processes may reduce the extent of hydrogen scrambling (77). Deuterium levels at individual peptide amide linkages after exchange-in onto proteins as determined through CID MS/MS-generated b n ions have been found to correlate with data derived from NMR exchange rates (78,79). The key to successfully measuring H/D content through CID may depend on the mechanism and experimental conditions for formation of the b n ions, but y n ions have appeared to be particularly unreliable for H/D analysis in almost all studies (79). The information gained from site-specific amide H/D exchange for an entire protein, i.e. TRX, demonstrates the potential of this technique (78). Deuterium equilibrium exchange-in experiments with oxidized, reduced, and modified TRXs indicated resistance to exchange in the order observed by F/(F ϩ U) values in CSD (42). However, the influence on protein stability by the two alkyl groups, compared with Red-TRX conformation, are different, as suggested by site-specific amide hydrogen isotopic exchange (78). The ethylcysteinyl group in Cys-Et-TRX results in faster exchange at Thr 66 and Ala 67 than in both Red-TRX and GS-Et-TRX. The ethylcysteinyl group, although small, appears to perturb the conformation of Red-TRX globally as these two sites are distant both in sequence and spatially from the modified Cys 32 site. The ethylglutathionyl group in GS-Et-TRX, on the other hand, caused an increase in exchange at Lys 90 and Val 91 , but a decrease at Ile 75 , Gly 92 , and Ala 93 . The difference in behavior between the two alkylated structures is attributed to the salt bridges in GS-Et-TRX mentioned above, which results in a net increase in electrostatic stability calculated to be ϳ35 kcal relative to Red-TRX. Induced hydrogen bonding interactions also were predicted between carbonyl oxygens of the conformationally restricted ethylglutathionyl group amide hydrogens in the protein backbone, and H/D data provided strong evidence in support (43).
H/D measurements by LC-ESI-MS/MS analysis must take into account the inevitable back-exchange during LC separation if absolute values are needed. Corrections are made by assuming that deuterium loss throughout is constant with its measured level in comparison to the overall exchange in the control sample according to (65): where D is the deuterium content of a peptide with N peptide amide hydrogens and m t , m 0% , and m 100% are peptide molecular masses centroided for H/D exchange-in at t min, 0 min, and fully exchanged, respectively. Serious back-exchange may result in a failure to detect subtle isotopic differences and the ability to conclude anything about a protein's conformational changes (80). In addition, complete sequence information from CID MS/MS and at the ends of peptic fragments is often difficult to obtain reliably. Overlapping fragments can be used to increase spatial resolution, but selective back-exchange sometimes causes uncertainty. These problems may be solved if intact protein solutions are sprayed directly into the mass spectrometer. Akashi and co-workers used high-resolution and high-accuracy ESI-Fourier transform ion cyclotron resonance MS to analyze a small labeled protein (8.5 kDa) by capillary-skimmer CID. Less back-exchange (3-4%) was observed in comparison to the 10 -30% back-exchange that generally occurs by proteolysis and LC separation of the resulting peptides. The spatial resolution achieved in these studies was consistent with that obtained by NMR and x-ray studies (81).

CONFORMATIONAL CHANGES BY EXCHANGE RATE ANALYSIS
Although amide hydrogen exchange has long been studied and has been widely recognized as a powerful tool to probe protein conformational changes (82)(83)(84)(85)(86)(87) and the chemical process of exchange in unstructured peptides is quite clear, the mechanisms of exchange in folded proteins are still poorly understood. Proteins are continuously unfolding and refolding even under physiological conditions. Amide hydrogens may exchange directly from the folded state, after global unfolding, or from both according to a commonly accepted model (82,88): where k 1 and k Ϫ1 are the rate constants for unfolding and refolding, and k int is the exchange rate at this peptide in the random coil conformation. The observed exchange rate (k ex ) is the sum of contributions of exchange from the folded state (k f ) and the globally unfolding state (k u ): The exchange rate constant from the folded state is given by: where ␤ is the probability that amide hydrogens contact solvent and catalyst. The exchange rate constant after unfolding is given by: When k Ϫ1 Ͼ Ͼ k int , the folding-refolding process occurs many times before H/D exchange is complete. This limiting case is referred to as EX2 kinetics and k ex ϭ k int (k 1 /k Ϫ1 ). When k int Ͼ Ͼ k Ϫ1 , exchange occurs by the EX1 mechanism and k ex ϭ k 1 . When H/D exchange occurs by the EX2 mechanism, which generally describes the exchange in the native state where regional unfolding may occur or under relatively mild denaturing conditions, the mass spectrum shows a single mass peak that gradually shifts to higher mass with increasing exchange-in time (89). In contrast, two distinct mass peaks, one that is protonated and one that is largely deuterated, develop after short exchange-in time if exchange occurs via the EX1 mechanism. This mechanism is operative when global unfolding takes place, for example, at high temperatures, under acidic pH conditions, or in the presence of high denaturant concentrations (16,42,56). In contrast to EX2 kinetics, the rates of unfolding can be determined within the EX1 kinetic limit.
The pioneering studies of Dobson, Robinson, and coworkers demonstrated that EX1 and EX2 kinetics in H/D exchange studies could be monitored using MS (62). Simply measuring the width of the protein's mass peak provided information on whether exchange followed EX2 or kinetics intermediate between the two limits. Proteins may undergo isotopic exchange simultaneously by EX1 kinetics in certain regions and EX2 kinetics others. For example, oxidized and reduced TRX in 1% AcOD/D 2 O at room temperature underwent H/D exchange by EX2 kinetics at the N-terminal end (aa 28 -39) and by EX1 type kinetics at the C-terminal end (aa 59 -80 and aa 81-101) (Fig. 7) (90). In an elegant series of experiments, pulse-labeling hydrogen isotope exchange followed by ESI-MS analysis of a mixture in the 15 N-labeled wild-type human lysozyme and the amyloidogenic D67H mutant showed EX2 exchange for the former and EX1 kinetics in the latter (91). The results provided evidence for cooperative unfolding of the ␤ domain and adjacent C helix, allowing the authors to conclude that the extensive intermolecular interactions possible can lead to aggregation and fibrallar structures in patients with the mutant gene.
Two types of structural models have been developed to describe H/D exchange in native proteins: the "solvent penetration" model (84) and the "local unfolding" model (92). According to the solvent penetration model, protons exchange within the protein interior as catalysts enter the protein core through transiently formed channels and cavities. Therefore, the rate of exchange of a proton depends on its average accessibility to solvent, which is a function of its depth and the local mobility of the protein's structural elements, and on its reactivity with the catalyst, which is affected by local structural features, packing density, and hydrogen bonding. In the local unfolding model, exchange occurs on the surface of the polypeptide chain when a subregion transiently unfolds, exposing the surfaces to bulk solvent. In this model, the main barrier to exchange is considered to be the interior hydrogen bonding of amide protons, rather than their depth. Adjacent protons within subregions are predicted to exchange at roughly the same rate. It is generally accepted that the exchange rate of protein amide hydrogens is highly dependent on the combination of hydrogen bonding and the extent of shielding from solvent.
The deuterium exchange rate in various segments of proteins can be semiquantitatively estimated from the time course of deuterium incorporation. Under conditions of constant pH, temperature, and bulk deuterated buffer, deuterium incorporation is described as the sum of a series of first-order exponentials of rate (93): where D is the deuterium content of a peptide with N peptide amide hydrogens as described for Eq. 1 and k i is the rate constant for exchange for the peptide hydrogens. Nonlinear least-squares fitting is used to obtain exchange rate constants for different groups of amide hydrogens from the MS data (55). It is difficult to directly relate H/D exchange rates with protein structural features, though these may be revealed by judicious comparisons with calculated solvent accessibility in different segments of a protein or the conformational changes arising from environmental influences (94 -98) and on protein folding intermediates (62,99). Attempts have been made to find correlations between exchange rates and protein structural parameters, including most notably intramolecular hydrogen bonding and secondary structures (65), solvent accessibility surface areas (100), and B factors (98,101). Good correlations have been found for each in individual proteins. Such correlations were used to map three-dimensional topological details of the C terminus of rhM-CSF␤ (102), a protein that has not been crystallized.
This homodimeric protein contains nine disulfide bonds. Proteins with multiple disulfide bonds present some major challenges in studies involving H/D exchange and MS analysis. Nevertheless, isotopic exchange results were obtained for rhM-CSF␤ with a spatial resolution of 3-31 residues by simultaneously reducing the disulfide bonds and proteolytically digesting the native protein under quench conditions after exchange-in (102). X-ray crystallographic coordinates of atoms in aa 4 -149 are available for rhM-CSF␣, whose structure is identical with the core of rhM-CSF␤. The latter has an extended C terminus consisting of aa 150 -221. The crystallographic data allowed calculation of the average depth of residues and solvent-accessible surface area values. Deuterium exchange for all amide hydrogens in the long C-terminal tail, aa 150 -221, was fast (k ex Ͼ 4.16 min Ϫ1 ). It was clear that this segment was substantially solvent accessible and, therefore, believed to be largely unstructured in solution.
The dead time (10 s) for deuterium exchange corresponds to an exchange rate constant, k ex Ͼ 4.16 min Ϫ1 . The amide hydrogen exchange rates within peptides for rhM-CSF␤ were grouped into three categories, i.e. slow plus nonexchanging (k ex Ͻ 0.1min Ϫ1 ), intermediate (0.1 min Ϫ1 Յ k ex Յ 4.16 min Ϫ1 ), and fast exchange (k ex Ͼ 4.16 min Ϫ1 ). Residue depth was found to correlate with these hydrogen isotope exchange rates (105). Residues deeper than 6.0 Å have zero solvent accessibility, and the average calculated depth for the fast exchanging group in rhM-CSF␤ is 4.8 Å (104). Thus, individual residues in the peptides were categorized into three groups; deep (Ͼ6.0 Å), intermediate (4.8 Å Ͻ average Յ 6.0 Å), and shallow (Յ4.8 Å).
Generally, there was good correlation between the corresponding numbers of fast-exchanging amide hydrogens with shallow depths in 20 peptic fragments (Fig. 8A). However, in the plot of the intermediate and slow plus nonexchanging groups, the peptides in the regions aa 95-105, aa 114 -128, and aa 63-76 appear to be outliers (Fig. 8, B and C). The amide hydrogens in these peptides exchange too slowly relative to the depths calculated from the rhM-CSF␣ crystallographic data. The most obvious explanation for this discrepancy is that the long carboxyl-terminal tails, which are not present in rhM-CSF␣, provide protective interactions with these segments. It may reasonably be concluded that the two C-terminal tails, which extend from Val 152 in each subunit, meet at the center of the homodimer where two intermolecular disulfide bonds C 155 -C 155Ј and C 157 -C 157Ј link them together. Based on the crystal structure, the segment, aa 143-151, extends right below the region of the aa 95-105 segment, which includes the ␤2 strand, and provides some protection for it perhaps through induced hydrogen bonding. In addition, the C 155 -C 155Ј and C 157 -C 157Ј disulfide bonds in the native structure cause the C-terminal tails to extend across the surfaces of the aa 114 -128 segment in ␣D and the aa 63-76 segment of ␣C on either side of the two domains, shielding them against amide H/D exchange (Fig. 9) (102). Solvent-exposed surface area analysis showed the same results. The exposures of the amino acids, as calculated using the Connally method, were grouped into three categories: 1) highly exposed (Ն50 Å 2 ), 2) moderately exposed (30 -50 Å 2 ), and 3) buried (Ͻ30 Å 2 ). The numbers of amino acids in peptides for each of these categories correlated with the corresponding numbers of residues in the three H/D exchange categories, and peptide aa 114 -128 was again predicted to be significantly more accessible to solvent than the H/D exchange data suggested (102).
Modeling provided additional support for the proposed structure. A model was first constructed by extending the crystal structure of rhM-CSF␣ to 160, thereby providing the symmetrical C 157 -C 159 disulfide linkages in the dimer. The resulting structure was subjected to molecular dynamics and energy optimization by AMBER force field and the remaining sequence to Glu 177 was then modeled as a turn and a polyproline type II helix. In the resulting model (Fig. 9), the C-terminal helix buries the ␣D helix of the protein consistent with the experimental results. Moreover, the segments aa 95-105 lie beneath the C-terminal regions in the dimer that can otherwise become exposed when the C 157 -C 159 disulfide linkages are reduced.
An increase of 12 charges in the ESI spectrum when these two disulfide bonds are reduced can now be more fully ap- The primary anomaly was seen in peptide aa 114 -128, aa 63-76, and a second anomaly in aa 95-105. These segments are, therefore, protected from solvent to a greater extent than estimated on the basis of depth calculations from rhM-CSF␣ crystallographic data, which lead to their proposed interaction with the C-terminal tail. preciated (45). In the segment aa 95-105, there are three basic sites, His 98 , Lys 100 , and Arg 104 , which become exposed in the two subunits as a result of reduction of the disulfide bonds. Thus, three additional basic sites can be accounted for in each subunit for a total of six in the homodimer or eight if Lys 154 is included. Previously, it was somewhat difficult to rationalize the 12 additional charges (45) as only Lys 154 becomes exposed when the C 157 -C 159 disulfides are reduced. These results would not have been available from x-ray studies because the full-length protein has not been crystallized.

CONCLUSIONS AND PROSPECTS
MS is becoming an increasingly important tool to study protein structures and dynamics. It provides complementary in-formation for studying protein higher-order structure or conformational changes with other biophysical methods. It has a unique capacity to detect co-existing conformations and to monitor conformational changes in relatively fast processes (i.e. millisecond timescale) either by CSD in ESI-MS or in combination with H/D exchange techniques that can provide information on three-dimensional structures at 5-to 10-residue spatial resolution. Site-specific amide H/D exchange analyzed by CID MS/MS is potentially a very powerful approach for examining interactions at the individual residue level, if label scrambling can be eliminated and proteolytic digestion and incomplete collisional fragmentation can be improved upon. In addition, MS provides high sensitivity, relatively fast analysis, and a wide range of solvent, pH, and temperature conditions under which proteins and peptides can be examined. Protein size also is not limiting, and together with the inherent sensitivity of MS the risk of protein aggregation is reduced. Recent advances in mass resolving power and mass measurement accuracies on modern instruments will surely have an impact by removing any doubt about charge states and modifications that result in relatively small changes in mass, such as cyanylation, oxidation, methylation, etc. It is expected that in the future the structures of proteins that fail to crystallize may potentially be partially characterized by a combination of H/D exchange methods and judicious use of modeling, sequence-dependent secondary structure predictions, and comparisons with related or homologous proteins in databases. Future developments are likely also to involve electron capture dissociation on instruments that are accessible to the average user, thereby providing robustness in H/D analysis and structure determinations. Electron capture dissociation should reduce the potential for label scrambling and provide greater opportunities to perform structural analysis on whole proteins in the gas phase. Back-exchange attendant with current fragmentation methods may be minimized by conducting on-line H/D exchange, on-line fragmentation, and faster analyses, or by fragmenting intact proteins directly in the mass spectrometer to yield site-specific information. Automation in H/D analysis is a potentially powerful technique for rapid analysis of ligand, agonist, and antagonist binding to proteins in pharmaceutical chemistry and protein-protein and protein-DNA interactions for studies in signaling pathways.
Acknowledgments-The results from M. Kim and H. Zhang contributed in this review are gratefully acknowledged. * This work was supported by grants from the National Institute on Environmental Health Sciences (ES 00040 and ES 00210) to M. L. D. and from the National Institutes of Health (RIGM 62957A) and National Science Foundation (MCB-0090615) to P. S. H. We acknowledge the support of the nucleic acid and protein core and the mass spectrometry core in the Oregon State University Environmental Health Sciences Center. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
FIG. 9. Ribbon model of rhM-CSF␤. A, The structure of rhM-CSF␤ up to Glu 177 (of the full-length 218-residue protein) correlates better to H/D exchange data from MS than does the crystal structure of the truncated rhM-CSF␣ (truncated at His 151 ). The rhM-CSF␤ was constructed by first extending the crystal structure of rhM-CSF␣ to residue 160 (red). This allowed Cys 157 to be cross-linked to Cys 159 of the symmetry related dimer, and vice versa. The resulting structure was subjected to molecular dynamics and energy optimization using the AMBER force field. The remaining sequence to Glu 177 was modeled as a turn and a polyproline type II helix (orange). This C-terminal helix buries the ␣D helix of the core protein (gray), which helps to rationalize the anomalously slow exchange seen for these residues when compared with the crystal structure. In addition, the disulfide cross-link tethers residues 151-160 across the dimer interface. This is consistent with the observation that H/D exchange at residues 95-105 (violet) is less than predicted by depth analysis from rhM-CSF␣ crystallographic data, and reduction of the disulfide bonds C 157 -C 159 exposes more charge sites as observed by ESI-MS in rhM-CSF␤ than was originally expected (45). B is the top view of the model. ¶ To whom correspondence should be addressed. Tel.: 541-737-1773; Fax: 541-737-0497; E-mail: Max.Deinzer@orst.edu.