|Home | About | Journals | Submit | Contact Us | Français|
19F NMR has proven to be a powerful technique in the study of protein structure and dynamics because the 19F nucleus is easily incorporated at specific labeling sites, where it provides a relatively nonperturbing yet sensitive probe with no background signals. Recent applications of 19F NMR in mapping out structural and functional features of proteins, including the galactose-binding protein, the transmembrane aspartate receptor, the CheY protein, dihydrofolate reductase, elongation factor-Tu, and D-lactose dehydrogenase, illustrate the utility of 19F NMR in the analysis of protein conformational states even in molecules too large or unstable for full NMR structure determination. These studies rely on the fact that the chemical shift of 19F is extremely sensitive to changes in the local conformational environment, including van der Waals packing interactions and local electrostatic fields. Additional information is provided by solvent-induced isotope shifts or line broadening of the 19F resonance by aqueous and membrane-bound paramagnetic probes, which may reveal the proximity of a 19F label to bulk solvent or a biological membrane. Finally, the effect of exchanging conformations on the 19F resonance can directly determine the kinetic parameters of the conformational transition.
Since the initial application of solution 19F NMR to protein systems in the late 1960s (80), the technique has evolved to become a versatile tool in the study of protein structure and dynamics and, thus, has provided an approach that complements established structure determination techniques. When high-resolution structural methods cannot be used, 19F NMR generates valuable low-resolution structural information. Perhaps the most powerful application of 19F NMR, however, is to probe proteins of known structure in their native solution environment, where the extreme sensitivity of the 19F resonance to its molecular surroundings can reveal important structural and kinetic features of protein conformational changes.
The rapidly expanding number of high-resolution protein crystal structures has emphasized the need for alternative approaches that can be used to monitor specific locations within a known structural frame-work in solution. For smaller proteins (Mr < 30 kDa), whose solution structures can be probed effectively by multidimensional NMR, such techniques clearly provide the maximum obtainable information and represent the method of choice to complement crystallographic data. Many proteins of known crystallographic structure, however, are too large for multidimensional NMR but still fall within the range of molecular weights accessible to solution 19F NMR (Mr < 100 kDa). For such a protein, one or more fluorine labels can be introduced, either biosynthetically or chemically, into specific sites, thereby enabling the use of 19F NMR to map out the regions of the molecule involved in a conformational change, characterize the kinetics of the event, and, in some cases, measure the intramolecular distance changes that result.
In this review, we use recent 19F NMR studies to illustrate the analysis of both proteins of known crystal or NMR structure and proteins for which no high-resolution structural information exists. We focus primarily on studies published during 1993–1995, in which the fluorine probes are introduced covalently into the protein itself. For a comprehensive summary of earlier applications, including theoretical aspects and studies that introduce fluorine into a small molecule or peptide ligand, see the excellent reviews by Gerig (30, 31), de Dios et al (21), Ho et al (37), and Sykes & Hull (85).
A wide variety of fluorine labels, including fluorinated amino acids, fluorinated reagents that react covalently with specific sidechains in proteins, and fluorinated ligands, are available. The most commonly used fluorinated amino acids are analogues of the aromatic amino acids, all of which are available commercially as mixtures of the D and L enantiomers (30). These amino acids include the ortho, meta, and para derivatives of phenylalanine (2-, 3-, and 4-F-Phe), the meta derivative of tyrosine (3-F-Tyr), and tryptophan fluorinated at specific indole ring positions (4-, 5-, and 6-F-Trp).
Several methods exist for incorporating fluorinated amino acids into a selected protein. Although chemical synthesis is impractical for any protein of appreciable size, several proteins have been labeled using a “semi-synthetic” approach (11, 46, 88). In this method, a peptide that contains a fluorinated amino acid is synthesized and combined with the remainder of the protein that has been produced biosynthetically, thereby producing an active, labeled protein. Recent advances in protein splicing techniques may soon revive this approach, which has the advantage of labeling a unique region of the protein (98).
More commonly, fluorinated amino acids are incorporated biosynthetically by microbial protein expression in the presence of the desired amino acid analogue. A fluorinated aromatic amino acid can be incorporated into a bacterial expression system easily, for example, by including it in the growth medium and eliminating the ability of the cell to synthesize that amino acid endogenously. The latter is accomplished either by using a bacterial strain auxotrophic for the amino acid of interest (85) or by adding glyphosate, which inhibits the synthesis pathways of all three aromatic amino acids (44). Depending on the application, the level of incorporation can be adjusted from less than 5% (22) to greater than 90% (50). Similarly, biosynthetic incorporation can be carried out by using a yeast expression system (96) or by including the amino acid analogue in the diet of such vertebrates as rabbit (95) avians (51), or primates (28).
When biosynthetic incorporation of fluorinated amino acids is impractical, or when alternative probe locations are desired, a fluorinated modification reagent can be used to label specific protein sidechains covalently. The most popular of these reagents are the cysteine-reactive probes, which target the unique chemistry of the cysteine thiol. Conveniently, the low frequency of cysteine usage in natural protein sequences generally guarantees a manageably small number of labeling sites or enables the engineering of labeling sites into important, selected locations in the protein structure. Other amino acids that may be labeled with fluorinated reagents include lysine (as well as the N-terminus), serine, tyrosine, and histidine. These reagents have been reviewed recently (30).
Finally, a wide variety fluorinated analogues can be used to substitute for natural ligands, substrates, and intermediates. For example, several fluorinated compounds have been found to be effective enzyme inhibitors (reviewed in 1). Other sources provide further information regarding these small molecules, as well as nucleic acid applications (16–18, 23, 36, 43, 58, 73–75, 81, 86).
Fluorine incorporation, particularly at a ring position of an aromatic sidechain, generally has little effect on the structure and function of a protein. Escherichia coli, for example, has been labeled with 5-F-Trp to a level of 80% at every tryptophan position within the cell without affecting growth seriously (44, 54), whereas higher order animals can tolerate at least 25% incorporation of 4-F-Phe (95). Similarly, at least 15 proteins have been shown to be structurally or functionally unperturbed by incorporation of fluorine (2, 11, 19, 22, 29, 33, 42, 49, 53, 54, 64, 71, 72, 77, 87, 91, 97). In contrast, detectable perturbations have been observed for a pentafluoro analogue of phenylalanine (8), as well as for monofluoro derivatives of tyrosine (47) and histidine (89), in which the ring fluorine alters the sidechain pKA substantially. Such perturbations, however, appear to be the exception. More caution must be used when fluorine is introduced by chemical modification, because the relatively large size of the modification reagent itself can generate a significant structural perturbation.
Several factors likely contribute to the surprisingly nonperturbing nature of fluorine when it is substituted for hydrogen within amino acid sidechains. The fluorine atom is similar in size to hydrogen [covalent radii of 1.35 Å and 1.2 Å, respectively (68)], such that fluorine substitution has little steric effect. Incorporation of a single fluorine atom into phenylalanine, for example, increases the volume of the side-chain only 0.7%. Moreover, the aliphatic C-F bond is only moderately polar, such that the fluorine atom is a weak hydrogen bond acceptor, at best (66). Aromatic C-F bonds, in which the fluorine electron density is significantly reduced by backbonding to the ring (68), are even less polar and, therefore, are correspondingly less able to accept a hydrogen bond. In fact, no case of hydrogen bond formation that involves aromatic fluorine has been documented. Finally, extensive mutagenic studies have shown that protein structures are relatively plastic and can accommodate a variety of amino acid substitutions, except for those targeted to the active site or locations critical for folding. Substitution of fluorine into a sidechain for hydrogen represents perhaps the most subtle type of “single-atom mutagenesis” possible.
To draw conclusions regarding the native protein from 19F NMR studies, however, it remains essential to establish that the fluorine label is nonperturbing. Two classes of approach have been used to test for perturbations in different types of applications. For those cases in which the level of fluorine incorporation is high enough, the effect of fluorine on protein structure and activity has been assayed directly (2, 11, 29, 33, 42, 49, 53, 54, 64, 71, 72, 77, 87, 91, 97). When the level of incorporation is low, however, an alternative mutagenic approach has been used (19, 22). The latter approach substitutes a natural amino acid, rather than a fluorinated analogue, at a position to be probed by fluorine incorporation. For example, tyrosine has been substituted at phenylalanine positions targeted for labeling with 4-F-Phe, thereby enabling the effect of substituents at the para position of specific phenylalanine residues to be measured directly (19, 22). In this example, the substitution of tyrosine is expected to overestimate any perturbation caused by the incorporation of 4-F-Phe, because the hydroxyl group is considerably larger, more polar, and better able to form hydrogen bonds than is fluorine.
Several methods have been used to assign 19F NMR resonances. The most rigorous assignments have used site-directed mutagenesis either to replace or to “nudge” each specific fluorine label. In the first method, the residue targeted for assignment is substituted with the most similar sidechain available, thereby causing the corresponding resonance in the 19F NMR spectrum to disappear while the remaining resonances are unperturbed (19, 22, 39, 50, 54, 76, 77). The direct replacement method fails occasionally, because each attempted substitution has perturbed the nontarget resonances. In proteins of known structure, the nudge method can then be used (19, 22). This approach identifies a sidechain in van der Waals contact with the target residue for conservative replacement, thereby generating a local nudge. In favorable cases, this nudge is not transmitted to the remaining resonances. Figure 1 illustrates both types of assignments in the 4-F-Phe-labeled CheY protein (22). A nonmutagenic, but similar, method of assignment has been illustrated in the egg white lysozyme system, for which 19F NMR spectra were obtained with the use of nonidentical lysozymes from several avian species (51).
Additional information, including solvent exposure (see below), perturbation of a resonance by a small molecule ligand or a substrate that binds to a specific site, or chemical modification of an amino acid near a fluorinated residue, can be used to confirm or tentatively assign resonances. These approaches can be complicated, however, by the ability of 19F NMR to detect the local structural and electrostatic effects triggered by a long-range conformational change that originates in a different region of the protein.
Solvent exposure can be detected either by a solvent-induced isotopic shift (SIIS) or by paramagnetic broadening that stems from a phase-specific probe. The SIIS effect generates up to a 0.25-ppm chemical shift of a 19F resonance exposed to aqueous solution when the solvent is changed from H2O to D2O (30,42). In contrast, a 19F resonance that emanates from the protein interior does not experience solvation and the accompanying chemical shift change, unless the residue interacts with buried water molecules. The alternative method uses an aqueous or hydrophobic paramagnetic probe to broaden the resonances of water-exposed and lipid-proximal fluorine labels, respectively. The aqueous reagent Gd3+ EDTA4−, in which the metal ion possesses seven unpaired electrons, and the membrane probe 8-doxylpalmitic acid have proved useful in many studies (19, 20, 70, 77, 84, 90). Finally, a 19F NMR method has been developed recently to detect bound water molecules within a macromolecule (16).
The large covalent chemical shift range of 19F, which approaches 1000 ppm, stems from the paramagnetic shielding generated by the fluorine lone-pair electrons. This paramagnetic term of the shielding equation also causes the chemical shift to be highly sensitive to the noncovalent tertiary interactions in a folded protein structure. In an unfolded protein that is labeled homogeneously with a given fluorine probe, the resonances from multiple probe positions collapse to give a single peak, because of the equivalent solvent environment experienced by each probe position. In a folded protein, however, the local packing and electrostatic fields perturb the lone-pair electrons of each probe differently, thereby providing a noncovalent chemical shift range as large as 17 ppm (69).
Several theoretical descriptions have been offered to explain observed range of environmental 19F chemical shifts in proteins. Millet & Raftery (61) divided fluorine shielding into contributions from (a) local magnetic fields that arise from electronically anisotropic groups, (b) hydrogen bond formation, (c) electrostatic fields from dipoles or formal charges, and (d) van der Waals interactions. Local magnetic fields from aromatic rings, carbonyls, and other electronically anisotropic groups generate chemical shift changes less than 2 ppm in magnitude and, therefore, cannot explain the observed protein chemical shift range (32). The contribution of hydrogen bonding is uncertain, because fluorine hydrogen bonds in proteins are weak or nonexistent. The two forces that dominate environmental 19F chemical shifts in proteins, therefore, are likely to be van der Waals interactions and electrostatic fields.
Considerable evidence suggests that van der Waals interactions play an important role. Hull & Sykes (42) observed a rough correlation between chemical shift and T1 relaxation for fluorines in alkaline phosphatase labeled with 3-F-Tyr. Because T1 relaxation and the van der Waals force have the same r−6 dependence on the distance to nearest neighbor protons, T1 relaxation was regarded as an indicator of the extent of van der Waals contacts or the extent of burial within the protein interior. The latter relationship was verified by solvent exposure measurements with the use of SIIS. These authors therefore concluded that van der Waals contacts were the primary factor influencing the chemical shift. Gregory & Gerig (32) reached a similar conclusion in a study comparing the results of molecular dynamics calculations with experimental data obtained for fluorinated ribonuclease-S (11). van der Waals interactions were proposed to be the dominant factor controlling 19F chemical shift, together with smaller contributions from local magnetic and electric fields. This study demonstrated the importance of molecular dynamics in chemical shift calculations, because calculations based on static structures yielded incorrect chemical shifts. More recently, van der Waals interactions within a molecular dynamics simulation have been used to explain the 19F chemical shifts of the E. coli galactose-binding protein in terms of its known high-resolution structure (12).
If van der Waals interactions were the sole effector of fluorine chemical shift changes, however, one would expect all 19F NMR resonances from buried probe positions to be shifted downfield from the denatured protein resonance. The fact that many 19F NMR resonances are observed upfield of the denatured protein resonance (30) indicates that another factor must also contribute to the observed chemical shifts. On the basis of electric field effects alone, Augspurger et al (3) could predict accurately chemical shift ranges in several nuclei, including 19F. Subsequently, a self-consistent field approach, denoted “charge field perturbation-gauge including atomic orbital,” was used to predict the chemical shifts of the E. coli galactose-binding protein solely on the basis of electric field effects (21, 69). A strong correlation was seen between the predicted and experimental chemical shifts, which led to the conclusion that weak electrical interactions dominate the fluorine chemical shielding term. Surface charge does not significantly contribute to the shielding (51, 69).
Overall, it appears likely that van der Waals and electrostatic interactions together control the environmental 19F chemical shifts observed in proteins. Significantly, such chemical shifts are regulated by local, rather than long-range, factors. Thus, although it remains difficult to convert either the magnitude or the sign of a chemical shift into a structural parameter, the observation of a chemical shift change indicates that the protein structure or electrostatics has been altered in the vicinity of that probe. In fact, the extreme sensitivity of the 19F chemical shift to its environment provides what is perhaps the most sensitive physical method available for the detection of local conformational changes.
X-ray crystallography and 19F NMR have proven to be splendidly complementary techniques. Starting with a high-resolution structure solved under relatively static conditions, 19F NMR allows us to observe the dynamics of the protein, as well as to map out areas where changes occur as the protein enters a conformation inaccessible to crystallography. Studies to date have primarily used proteins labeled with fluorinated aromatic amino acids.
Because of the importance of aromatic sidechains in many aspects of protein function, aromatic fluorine labels are often located in critical regions of the target protein, including most ligand and effector protein binding sites. For the most part, two 19F NMR approaches have been used to probe conformational changes. The first approach maps out the regions involved in the conformational change by observing which 19F NMR resonances undergo chemical shift changes. The second approach uses paramagnetic broadening to measure distances and changes in solvent exposure.
In addition to providing spatial maps of conformational changes in proteins, 19F NMR has been used to characterize the kinetic parameters of these conformational changes. The standard relationships of NMR exchange averaging places limits on the frequency of a conformational change, or, in favorable cases, can determine this rate directly (94):
When two conformations, A and B, which yield NMR frequencies vA and vB, interconvert over time at a frequency of vic two distinct resonances are observed if the interconversion is slow relative to the difference in their NMR frequencies (Equation 1). If, however, the interconversion rate is rapid relative to the NMR frequency difference, then a single exchange-averaged resonance is observed (Equation 3). In the intermediate exchange limit, where the interconversion rate approximately equals the NMR frequency difference, the two resonances become very broad and disappear into the baseline. Observation of the latter intermediate exchange limit enables direct determination of the interconversion rate, whereas the slow and rapid exchange cases provide upper and lower limits, respectively, on the interconversion rate. Spectral simulation can provide even more precise information regarding both interconversion kinetics and the relative populations of different conformational states.
The chemotaxis pathway of E. coli and Salmonella typhimurium provides a well-suited system for the study of conformational changes involved in signal transduction (reviewed in 6, 35, 67, 83). The protein components of the pathway are defined fully, and several have been characterized structurally by crystallography or NMR. Such structural studies have helped define key mechanistic questions, the answers to which will provide a molecular understanding of signaling protein activation. To address these questions, 19F NMR has been utilized to augment high-resolution structures. Studies of the galactose-binding protein (54–56) represent the earliest application of 19F NMR to probe intrinsic sidechain labels in a protein of known crystal structure. The same approach has been extended to studies of the aspartate receptor (19, 26) and CheY (7, 22). Together, these three proteins illustrate how solution 19F NMR can complement the information provided by high-resolution structure determination. In some cases, the power and limitations of the 19F NMR method have been defined by direct comparison with results from independent approaches.
Chemosensing in bacteria often begins with a soluble periplasmic receptor, also termed a binding protein, which engulfs a specific small molecule ligand, then docks to and activates one of several transmembrane receptors that span the cytoplasmic membrane. In general, the soluble receptor also transports its ligand by associating with a membrane-bound transport system distinct from the sensory pathway. As a representative member of the soluble receptor family, the galactose-binding protein plays both a sensory and transport role for its two ligands, D-galactose and D-glucose. The crystal structure of the monomeric 32-kDa protein has been solved to 1.9 Å resolution for the conformation that contains bound D-glucose, as illustrated in Figure 2 (93), but the apo-structure is undefined. Thus, little is known about the conformational change that the protein undergoes when it binds ligand. Overall, the structure consists of two homologous domains, each consisting of a β-sheet sandwiched between layers of α-helices. The two domains are connected by three polypeptide strands, and the sugar-binding site lies within the cleft between the domains. The protein also possesses an EF-hand-like calcium-binding site, which stabilizes the protein structure. Luck & Falke (54–56) have fluorine-labeled the protein with 5-F-Trp and 3-F-Phe to investigate the conformational changes induced by sugar and calcium binding.
As seen in Figure 2, the targeted 12 probe locations (five tryptophans and seven phenylalanines) are well distributed throughout the molecule, including the sugar- and calcium-binding site regions. Biosynthetic incorporation of fluorinated amino acids into the appropriate E. coli auxotrophs yields 65–80% incorporation of 5-F-Trp or 20% incorporation of 3-F-Phe (54). The resonance that corresponds to each 5-F-Trp probe has been assigned by the direct-replacement mutagenesis method (see above).
The effect of fluorine labeling on the structure has been assayed directly, thereby revealing relatively minor effects caused by fluorine substitution (54). The estimated upper limits of these effects, calculated for the fully labeled protein, are as follows: (a) a 2.2- to 3-fold decrease in the D-galactose affinity caused by 5-F-Trp substitution, or a 1.5- to 4-fold affinity decrease caused by 3-F-Phe substitution, respectively; (b) a 1.5- to 2.5-fold increase in the calcium dissociation rate caused by 5-F-Trp substitution, whereas 3-F-Phe substitution has no detectable effect; and (c) a measurable decrease in protein stability, as revealed in urea denaturation curves yielding ΔΔGU(H2O) = −1.2 kcal mol −1 for 5-F-Trp, or −0.8 kcal mol −1 for 3-F-Phe. For this protein, then, the effects of fluorine incorporation on protein structure and function are quite small, especially considering that bound ligand is stabilized by direct van der Waals interactions with tryptophan and phenylalanine sidechains (one of each).
At the time the 19F NMR studies were carried out, it was proposed that a key feature of binding protein activation was closure of the ligand-binding cleft through bending of the interdomain linkage (57a). Such a hinged-cleft mechanism, however, had not been demonstrated experimentally for any member of the binding protein family. To test this mechanism, and to measure the minimum angle of cleft opening on removal of sugar, 19F NMR paramagnetic broadening measurements were carried out with the use of the aqueous paramagnet Gd3+ EDTA4−(56). With this approach, the inverse sixth power dependence of paramagnetic broadening is used to estimate the separation between the paramagnet and the 19F nucleus of a 5-F-Trp probe located at position 183 within the cleft. Addition of the paramagnet to the sugar-empty protein causes selective broadening of the probe resonance, as shown in Figure 3, whereas the paramagnet has no effect on the probe resonance when the cleft is closed with D-glucose trapped inside. The broadening of the probe resonance can be used to place an upper limit on the probe–paramagnet separation for the empty cleft. These results demonstrate that the cleft can open by at least 18 degrees in the absence of sugar, thereby enabling the paramagnetic probe to enter, as illustrated schematically in Figure 4 (56). Subsequent crystallographic studies have confirmed large, hinged-cleft motions in related binding proteins (27, 79).
More subtle conformational changes are also revealed by 19F NMR chemical shift changes within the individual protein domains (Figure 5). Inside the sugar cleft itself, the binding of D-galactose causes a dramatic chemical shift change (+ 3.8 ppm) in the resonance that corresponds to Trp 183, which contacts the bound sugar. The binding of D-glucose, which differs from D-galactose by the stereochemistry about a single carbon atom, induces a slightly smaller chemical shift change (+ 2.8 ppm) in the same resonance. In contrast, at more distal 5-F-Trp or 3-F-Phe probe positions, the two sugars induce identical chemical shift changes, with the exception of resonances in the vicinity of the calcium-binding site. These resonances are unaffected by sugar binding (Figure 2). Conversely, calcium binding alters the chemical shifts of only those residues located in the vicinity of its binding site and does not affect the resonances coupled to the sugar site (55). Calcium binding, therefore, induces only a local structural change, whereas sugar binding generates a long-range conformational change; the two sites are allosterically independent of one another. The long-range nature of the sugar-induced conformational change is consistent with its function, which is to regulate protein-protein interactions that involve large regions of the binding protein surface.
Further analysis of 19F NMR spectra places a lower limit on the kinetics of sugar release. At substoichiometric concentrations of D-glucose and D-galactose, the spectrum of the 5-F-Trp-labeled protein shows distinct resonances that correspond to each apo- and sugar-bound state (54), thereby placing sugar release in the slow exchange limit. Equation 1 indicates that the lifetime of the D-galactose-bound state is τic 0.6 ms, whereas the lifetime of the D-glucose-bound state is τic 0.8 ms, which is consistent with direct kinetic measurements that yield bound sugar lifetimes of 220 ms for D-galactose and 710 ms for D-glucose (60).
Altogether, the 19F NMR data indicate that the sugar-binding cleft does, indeed, close on sugar binding, thereby trapping the sugar inside until the complex can diffuse to the appropriate membrane receptor. Moreover, subtle, long-range conformational changes are induced in both domains by sugar, but not calcium, binding. These intradomain conformational changes, which are not easily detected by crystallography, likely stem from the conserved architecture of the binding protein family, in which most secondary structure elements originate or terminate in the substrate cleft. Such an architecture may transmit conformational information from the buried ligand-binding site to the protein surface, where it can be used to regulate the critical membrane docking sites (45).
The transmembrane aspartate receptor is activated either directly, by the binding of the chemoattractant aspartate to its periplasmic ligand-binding domain, or indirectly, by association with the maltose binding protein. The resulting transmembrane signal regulates an associated cytoplasmic histidine kinase, which ultimately controls the swimming behavior of the cell. The receptor is a homodimer of identical 60-kDa subunits in both the presence and the absence of ligand (25, 62). The structure of the isolated periplasmic domain has been determined to 2.0 Å resolution by X-ray crystallography, as displayed in Figure 6 (59). This domain consists of a homodimer of four-helix bundles, with two symmetric, nonoverlapping attractant binding sites at the dimer interface. In the intact receptor, the structure of the transmembrane domain has been defined by disulfide mapping, which indicates that the N- and C-terminal α-helices of the ligand-binding domain continue uninterrupted through the membrane.
To map out the spatial and kinetic features of the aspartate-induced conformational change, both the intact receptor and the isolated ligand-binding domain have been probed by solution 19F NMR. The intact receptor has been labeled with 5-F-Trp in its native membrane (26). Because of the very slow tumbling of the protein-membrane complex, all but one of the receptor 5-F-Trp resonances are too broad to detect. The lone resonance observed lies near the mobile C-terminus of the receptor and is exposed to the cytoplasmic compartment, where it possesses considerable local motion. Aspartate binding to the periplasmic ligand-binding site induces a long-range, transmembrane conformational change that is transmitted to the C-terminal 5-F-Trp probe located more than 90 Å away, thereby perturbing the chemical shift of this resonance.
A higher resolution picture of the conformational change has been obtained with the use of the 36-kDa isolated ligand-binding domain, which has been cloned and shown to retain native aspartate binding (19, 63). The homodimeric domain has been biosynthetically labeled with 4-F-Phe, with the use of glyphosate to inhibit endogenous synthesis of phenylalanine (19). Each identical subunit possesses six phenylalanine labeling positions, which are distributed throughout the known structure (Figure 6). Significantly, two probe positions lie in or near the ligand-binding pocket, whereas the other four probes lie at positions distal to the bound aspartate, including distal sites on the N- and C-terminal α-helices. Assignment of the resulting resonances has been carried out with the use of a combination of the direct replacement and the nudge mutagenic techniques.
The chosen low level of fluorine incorporation (7%) ensures that the labeled domain typically possesses no more than one fluorine atom, thereby preventing pairwise or higher order interactions. To estimate the effects of such fluorine incorporation on domain structure and function, each of the six phenylalanine residues has been individually mutated to tyrosine (19). The resulting para-hydroxyl moiety is expected to be even more perturbing than the corresponding parafluorine (see above), yet the tyrosine substitutions yield little effect on the aspartate binding site, which retains glutamate affinities ranging from 4-fold to 0.6-fold of native. These effects are quite small on a free energy scale (≤0.8 kcal mol−1), which suggests that hydroxyl substitutions at the para position of each phenylalanine, as well as the corresponding fluorine substitutions, cause little or no perturbation of ligand binding and the ensuing conformational change.
The 19F NMR spectrum of the ligand-binding domain has been obtained for the apo-, aspartate-bound, and glutamate-bound states (19). Only six resonances are observed for the 12 4-F-Phe positions in the dimer, which indicates that the monomers are equivalent, on average, owing to rapid exchange of ligand between the two symmetric binding sites. As observed for the galactose-binding protein, different ligands elicit contrasting chemical shift changes for a resonance within their shared binding pocket (4-F-Phe150) but yield the same ligand-induced pattern of chemical shift changes at more distant probe positions. These ligands therefore generate similar or identical long-range conformational changes outside the binding site itself.
Figure 6 summarizes the probe positions for which significant ligand-induced chemical shift changes have been detected (19). Importantly, chemical shift changes are observed for probe positions lying on the C-terminal helix but not for those on the N-terminal helix, which implicates the former helix as the transmitter of the transmembrane signal. Subsequent analyses of data from X-ray crystallography and disulfide engineering have confirmed this identification of the C-terminal helix as the key transmembrane signaling element, whereas the N-terminal helix is proposed to play a structural role in stabilizing the dimer (13–15, 48).
Additional stoichiometric and kinetic information has been extracted from a titration of the 19F NMR spectrum with ligand (19), as illustrated in Figure 7. Note that the maximum effect of ligand on the resonances is attained at a ligand:dimer mole ratio of 1:1, although there are two symmetric binding sites in the apo-dimer. This result confirms the earlier conclusion of the crystallographic and direct binding measurements (5, 59), which indicated strong negative cooperativity between the two aspartate binding sites. The titration also reveals that one of the resonances, which arises from 4-F-Phe150 in the ligand-binding site, disappears because of exchange broadening at a nonsaturating aspartate concentration, thereby enabling the kinetics of aspartate binding and dissociation to be measured. The resulting calculation indicates that aspartate binding and release is rapid, approaching the diffusion controlled limit observed for the fastest enzymes (19). Ligand binding and release, therefore, are not rate limiting in the chemosensory pathway, because subsequent phosphorylation and/or protein diffusion events in the cytoplasm are significantly slower (78).
CheY is the response regulator protein of the chemotaxis pathway. It receives a phosphate from the histidine kinase associated with the transmembrane receptors and subsequently docks to and regulates the behavior of the flagellar motor. The structure of the unphosphorylated 13-kDa protein has been determined crystallographically by several groups (4, 82, 92), and independent solution NMR structures have also been completed (9, 52, 65). CheY folds as a compact (α/β)5 motif illustrated in Figure 8, with a central antiparallel β-sheet sandwiched between layers of α-helices. The activation site, where phosphorylation occurs and activating mutantations are found, lies at one end of the β-sheet and includes four residues that exhibit 70–100% conservation across the response regulator family: Asp12, Asp13, Asp57, and Lys109. The Asp57 sidechain is the site of phosphorylation and provides one of the coordinating oxygens for a bound magnesium ion, which serves as a cofactor in phosphorylation and dephosphorylation of Asp57. A salt bridge has been proposed to exist between the highly conserved Asp57 and Lys109 residues (92). This salt bridge could provide a structural on-off switch in response regulator activation; alternatively, the Lys109 sidechain could play a role in the active site chemistry.
The short lifetime of the phospho-Asp57 linkage (τ1/2 ~ 10 s) has prevented the crystallization of the protein in the phosphorylated state. Instead, 19F NMR has been used to probe CheY labeled with 4-F-Phe, thereby enabling the conformational changes triggered by both phosphorylation and activating mutations to be mapped out (22). Using an E. coli strain auxotrophic for phenylalanine, 4-F-Phe has been incorporated to a level of 5%. The resulting six fluorine probe positions (Figure 8) are found at interesting locations, including the active site (Phe14), the loop containing the putative “switch” residue Lys109 (Phe111), and a phenylalanine cluster on the opposite side of the molecule from the active site, which can serve as an antenna for long-range conformational changes. The six resonances have been assigned with the use of direct replacement and the nudge method (Figure 1), and the maximal effect of fluorine incorporation at each phenylalanine position has been estimated by the corresponding tyrosine substitutions (see above). Each of these tyrosine substitutions has a negligible effect on the chemotaxis pathway in vivo, which indicates that perturbations of CheY caused by para substituents on its phenylalanine rings are minimal or nonexistent (22).
To probe the conformational change induced by phosphorylation, the short lifetime of the phosphorylated state has been overcome by chemically phosphorylating Asp57 with the small molecule phosphodonor, acetyl phosphate (57). Excess acetyl phosphate can maintain a high steady-state level of phosphorylation for 10 min—sufficient time to obtain a one-dimensional 19F NMR spectrum. Phosphorylation induces a global conformational change in the CheY molecule, which extends from the active site to the distant phenylalanine cluster (Figure 8). Subsequent multidimensional solution NMR studies have confirmed this long-range conformational change and have provided a higher resolution map of its extent (52). The long-range nature of the phosphorylation-induced structural change may play an important role in the large family of two-domain response regulator proteins (67), in which phosphorylation of the regulatory domain could transmit an allosteric signal to the effector domain.
In contrast to the long-range structural change triggered by phosphorylation, activation of CheY by specific “lock-on” mutations generates a much more localized conformational perturbation (7). Six single or double mutations that activate CheY constitutively all generate an upfield chemical shift change for the resonance corresponding to Phe111, which lies in the same loop as Lys109 (Figure 8). The structural change detected for this loop could represent a key component of the on-off “switch.” On the basis of circumstantial evidence, it has been proposed (7) that the Asp57-Lys109 salt bridge is formed in the off state and broken in the on state; this conclusion remains controversial, however, and must be tested by other approaches. The localized nature of the conformational change triggered by one of the lock-on mutations has recently been confirmed by multidimensional NMR (FW Dahlquist, personal communication).
Nuclear magnetic resonance methods, including 19F NMR, have begun to reveal key aspects of the protein–protein interaction between CheY and its kinase protein, CheA. Interestingly, the high-affinity kinase binding site on CheY appears to be located at a surface distal from the phosphorylated residue Asp57. Thus, when the CheA fragment responsible for CheY docking is added to the labeled protein, large chemical shift changes at positions distant from the active site have been observed in both 19F (20; TB Morrison, M Welch, Y Blat, SL Butler, JJ Falke, M Eisenbach, and JS Parkinson, submitted for publication) and multidimensional (84a) NMR studies. The phosphorylation domain of the kinase, therefore, must be large enough to bridge this distance.
The dihydrofolate reductase (DHFR) enzyme of E. coli is an 18-kDa monomeric protein that catalyzes the NADPH-dependent reduction of 7,8-dihydrofolate to 5,6,7,8-tetrahydrofolate, an important cofactor for several biosynthetic pathways. Because of its small size, well-characterized enzymatic mechanism, well-refined structure, and the reversibility of its folding reaction in the presence of chemical denaturants, Hoeltzi & Frieden (39) have deemed this a good model for studies of protein folding. The use of 19F NMR provides an advantage over fluorescence, absorbance, or circular diochroism in that it can provide information regarding specific sites within the protein. The five tryptophan residues, which have been labeled with 6-F-Trp, are distributed throughout the molecule illustrated in Figure 9 and give rise to well-resolved resonances that have been assigned by direct replacement mutagenesis (39).
Dihydrofolate reductase possesses two ligand-binding sites, one for NADPH and the other for 7,8-dihydrofolate. The latter site also binds methotrexate, a clinically important anticancer drug. A comparison of the spectra observed for the apo, NADPH-bound, methotrexate-bound, and doubly liganded (NADPH and methotrexate) forms of the protein reveal that NADPH significantly shifts two of the five probe resonances in or near its binding site (Trp74 and Trp22), whereas methotrexate triggers a longer range conformational change revealed by large chemical shift changes of three probe resonances (Trp22, Trp133, and Trp47; see Figure 9). Moreover, methotrexate sharpens the resonances of two probe resonances (Trp22, Trp30) considerably, whereas NADPH broadens two resonances (Trp22, Trp74) relative to the apo state. The doubly liganded protein exhibits the maximum chemical shift dispersion and sharpest resonances, which suggests that this conformation is the most stable or least flexible and possesses less structural heterogeneity than the lower ligation states. Of the two ligands, the methotrexate appears to generate the greater increase in structural homogeneity (39). Crystallographic evidence also supports the ordering effect of this ligand, which is observed to order the loop containing one of the fluorine probe positions, Trp22 (10).
To study the unfolding process of this protein further, Hoeltzi et al (40, 41) have constructed a stopped-flow 19F NMR device to monitor unfolding in real time with a resolution of 1.5 s. They observe that immediately after the addition of urea, the resonances from the native protein disappear, but the denatured resonances exhibit only 20% of their final intensity. Together with fluorescence and circular dichroism data, these stopped-flow 19F NMR results indicate that the protein unfolds by two pathways. Approximately 20% of the population denatures rapidly, whereas 80% of the protein forms a molten globule intermediate that retains native-like secondary structure but loses tertiary contacts, thereby causing the tryptophan sidechains to exhibit considerable structural heterogeneity (40). Interestingly, when the protein is titrated with increasing concentrations of urea in a separate equilibrium experiment, the resonance corresponding to Trp22 narrows and moves toward its denatured chemical shift at urea concentrations well below the denaturation midpoint (39). Thus, as suggested by the 19F NMR ligand-binding and crystallographic results, the region of the methotrexate binding site, including the Trp22 residue, appears to be less folded than the rest of the molecule in the absence of its ligand.
In the GTP-bound form, elongation factor Tu (EF-Tu) delivers aminoacyl-transfer RNA to the ribosome. After hydrolysis of the GTP to GDP, EF-Tu dissociates, and the GTP-bound form is regenerated by a different elongation factor. The structure of each of the three domains of trypsin-treated EF-Tu have been determined crystallographically. Subsequently, 19F NMR has been used to monitor the effects of various ligands on the conformation of full-length EF-Tu and to examine the individual domains for perturbations that arise from their isolation through trypsin cleavage (24). Elongation factor Tu has been labeled with 3-F-Tyr at its ten tyrosine positions, and tentative assignments are postulated for two of the resulting resonances that exhibit the largest upfield shifts for the GDP-bound state relative to other conformations. One of the assigned resonances stems from a buried tyrosine in domain III of the known structure, which suggests that this domain interacts with domain I containing the GDP/GTP binding site. No major chemical shift changes are seen between the native and trypsin-treated forms of the protein, which indicates that, in all likelihood, the crystal structures represent native domain conformations.
We have seen how 19F NMR can complement a high-resolution protein structure in probing conformational changes. Even for proteins that have not been characterized structurally, however, 19F NMR can be used to define the solvent exposure or separations of specific probe positions. The best example of such an application is E. coli D-lactate dehydrogenase (D-LDH), a peripheral membrane protein for which Ho and coworkers have developed a low-resolution structural model on the basis of extensive 19F NMR data (38, 70, 77, 84, 90).
Rule et al (77) have incorporated 4-, 5-, and 6-fluorotryptophan into the protein to levels reaching 90% and assigned the 19F NMR resonances by direct replacement mutagenesis. None of the substitutions alter the kinetic parameters of the protein substantially, and stability studies at 55°C have demonstrated that both the 4- and 6-fluoro-Trp labels retain the native activity decay constant of 10 s. Moreover, 5-F-Trp labeling actually increases this decay constant to 40 s (77). The secondary structure of the 4- and 5-F-Trp-labeled D-LDH appears, by circular dichroism measurements, to be unaltered from native, although the 6-F-Trp-labeled D-LDH shows a minor perturbation (77). Because the 5-F-Trp resonances exhibit wider dispersion than the 4- or 6-F-Trp resonances (7 ppm vs 6 or 3 ppm, respectively), studies have focused on this analogue. To increase the molecular tumbling rate and thereby minimize the linewidths of the probe resonances, the protein was solubilized from its native membrane to yield a protein–detergent micellar complex, in which the resonances were resolved easily. Although this solubilization narrowed the linewidths significantly, the chemical shifts of the resonances were unperturbed, which indicated that detergent micelles did not induce a conformational change (77). This finding was verified later by repeating the studies in a small unilamellar vesicle system, which led to the same conclusions drawn in the detergent-solubilized system (90).
Of the five labeled tryptophan positions, two (Trp384 and 567) are revealed by their solvent-induced isotopic shifts to be near the aqueous surface of the molecule. One position (Trp369) displays an altered chemical shift and line broadening on substrate binding and the accompanying reduction of the bound flavin cofactor (FAD), which suggests that this position may lie in or near the lactate-binding site (77). The observed line broadening is attributed to exchange of the probe between multiple environments, presumably owing to fluctuations of the local structure (38). In further studies, the proximity of the fluorine positions to the detergent micelle or membrane has been examined with the use of the nitroxide-containing fatty acid 8-doxylpalmitic acid, a paramagnetjc probe that will broaden any resonance within approximatelyo15 Å. Because the nitroxide lies in the middle of the fatty acid, 7 Å from either end, any resonance broadened would lie in or near the hydrocarbon phase. None of the intrinsic tryptophan positions are broadened by this probe, which suggests that they are not located near the membrane-binding surface (77).
To elaborate on these results, the repertoire of labeling positions has been expanded by substituting tryptophan for other aromatic sidechains within D-LDH through site-directed mutagenesis (70, 84). Only those mutants that retain significant enzymatic activities were used in the subsequent structural analysis. Figure 10 summarizes the resulting engineered 5-F-Trp positions, as well as the intrinsic 5-F-Trp probes, located within in a region of D-LDH proposed to contain the membrane-binding site. Also summarized are three types of structural information obtained by 19F NMR: (a) chemical shift changes induced by substrate binding and accompanying FAD reduction, (b) line broadenings caused by the paramagnetic fatty acid probe placed in the associated micelle or bilayer, and (c) solvent-induced isotopic shifts. The identified membrane-binding region extends from Tyr226 to Trp384. Within this region, the 19F NMR data have allowed the development of a low-resolution structural model (84), as illustrated in Figure 10.
In regions of D-LDH lying outside the proposed membrane-binding domain, limited structural information has been obtained from the chemical shift and linewidth changes induced by sidechain substitutions. In general, the effects of such point mutations stem from local structural perturbations of the region surrounding the substitution, thereby identifying fluorine probes in the vicinity of the altered sidechain (70). Such identifications must be regarded as tentative, however, because of the ability of 19F NMR to detect long-range conformational changes triggered at a distant site.
In summary, the systematic use of solution 19F NMR has shed light on the kinetics and spatial ranges of conformational changes in several proteins of known structure and has been used to develop a low-resolution conformational map for one protein of unknown structure. In cases where the conclusions of 19F NMR studies can be checked against newer information from crystallographic or multidimensional NMR structures, it generally has been observed that 19F NMR provides an accurate and powerful method for probing protein conformations. As with any method, the main caution is to avoid the overinterpretation of a limited data set.
Challenges for the future include the use of 19F NMR to probe specifically labeled proteins in vivo. One example of such an application has already been described in living yeast cells (96). Further, it should be possible to use paramagnetic probes that are attached covalently to a protein functionality, or paramagnetic metal ions bound in an intrinsic site, to map out multiple fluorine-paramagnet distances, thereby facilitating the development of a low-resolution structure for macromolecules inaccessible to other structural techniques. Such an approach has been used successfully in 1H NMR studies (34) but still needs to be tested for 19F NMR in a protein of known structure. Finally, the assignment of 19F NMR resonances in proteins of known structure could be facilitated by 19F-1H nuclear Overhauser effect experiments. Such approaches could, in favorable cases, allow full assignment of the 19F NMR spectrum without the aid of protein engineering. Overall, as additional high-resolution protein structures become available, and as the techniques of molecular cloning make new proteins of unknown structure accessible, 19F NMR will continue to be a useful technique in the toolbox of the protein chemist.
The authors thank a large number of colleagues in the field for sending reprints and preprints. We apologize to those whose work was, unfortunately, outside the scope of the present review. Special thanks are due to Steven Drake, Dr. Linda Luck, and Dr. Olve Peersen for helpful discussions and important experimental contributions, and to the National Institutes of Health for funding (GM40731 and GM48203 to JJF).