Structures of active site histidine mutants of IIIGlc, a major signal-transducing protein in Escherichia coli. Effects on the mechanisms of regulation and phosphoryl transfer.

IIIGlc (also called IIAGlc), a major signal-transducing protein in Escherichia coli, is also a phosphorylcarrier in glucose uptake. The crystal and NMR structures of IIIGlc show that His90, the phosphoryl acceptor, adjoins His75 in the active site. Glutamine was substituted for His-, giving H75QIIIGlc and H90QIIIGlc, respectively (Presper, K. A., Wong, C.-Y., Liu, L., Meadow, N. D., and Roseman, S. (1989) Proc. Natl. Acad. Sci. U. S. A. 86, 4052-4055), but the mutants showed unexpected properties. H90QIIIGlc loses regulatory functions of IIIGlc, and the phosphoryltransfer rates between HPr/H75QIIIGlc are 200-fold less than HPr/IIIGlc (Meadow, N. D., and Roseman, S. (1996) J. Biol. Chem. 271, 33440-33445). X-ray crystallography, differential scanning calorimetry, and NMR have now been used to determine the structures of the mutants (phospho-H75QIIIGlc was studied by NMR). The three methods gave completely consistent results. Except for the His to Gln substitutions, the only significant structural changes were in a few hydrogen bonds. H90QIIIGlc contains two structured water molecules (to Gln90), which could explain its inability to regulate glycerol kinase. Phospho-IIIGlc contains a chymotrypsin-like, hydrogen bond network (Thr73-His75-O−-phosphoryl), whereas phospho-H75QIIIGlc contains only one bond (Gln75-O−-phosphoryl). Hydrogen bonds play an essential role in a proposed mechanism for the phosphoryltransfer reaction.

III Glc loses regulatory functions of III Glc , and the phosphoryltransfer rates between HPr/ H75Q III Glc are 200-fold less than HPr/III Glc (Meadow, N. D., and Roseman, S. (1996) J. Biol. Chem. 271, 33440 -33445). X-ray crystallography, differential scanning calorimetry, and NMR have now been used to determine the structures of the mutants (phospho-H75Q III Glc was studied by NMR). The three methods gave completely consistent results. Except for the His to Gln substitutions, the only significant structural changes were in a few hydrogen bonds. H90Q III Glc contains two structured water molecules (to Gln 90 ), which could explain its inability to regulate glycerol kinase.
Phospho-III Glc contains a chymotrypsin-like, hydrogen bond network (Thr 73 -His 75 -O ؊ -phosphoryl), whereas phospho-H75Q III Glc contains only one bond (Gln 75 -O ؊phosphoryl). Hydrogen bonds play an essential role in a proposed mechanism for the phosphoryltransfer reaction.
The protein III Glc (or IIA Glc ) 1 plays a key role in many bac-terial cell functions (see Refs. 1-6 for reviews), and these are effected by its interactions with other proteins, both covalent (the transfer of the phosphoryl group) and non-covalent (6). Two of many unresolved questions concerning this important signal-transducing protein are: (a) the mechanism of phosphoryl transfer to other proteins of the phosphoenolpyruvate:glycose phosphotransferase system (PTS), 2 such as HPr; and (b) how III Glc "recognizes" or binds to its target proteins, despite the fact that there is no apparent common sequence in these targets.
III Glc from Escherichia coli is 18.1 kDa, and contains two His residues, His 75 and His 90 , but not Trp, Cys, or Tyr. The structure of III Glc has been extensively studied and has been established by NMR and x-ray crystallography (7)(8)(9)(10)(11), as has the structure of the cytoplasmic (or IIA) domain of the Bacillus subtilis II Glc membrane protein (12)(13)(14)(15)(16)(17)(18). E. coli III Glc and the B. subtilis IIA domain exhibit many similarities.
In E. coli III Glc , the N-terminal domain (residues 1-18) is unstructured both in solution and in the crystal. The remainder of the molecule, residues 19 -168, is a compact structure, shaped approximately like a parellopiped, about 30 Å on a side, and much thinner between the two faces. The major structural unit in each face of the protein is a ␤-sheet with six antiparallel strands.
Both His 90 and His 75 lie in a depression located off center of one of the faces, and His 90 is the residue that accepts the phosphoryl group from phospho-HPr. A space-filling model of the protein is shown in Fig. 1. The depression is surrounded by a very hydrophobic group of amino acids, and it is these residues that comprise the principal binding sites of III Glc to one of its target proteins, glycerol kinase (19 -21). 3 It has also been suggested that the hydrophobic patch, which is about 18 Å in diameter, is intimately involved in the binding of B. subtilis phospho-HPr to the IIA Glc domain (22). Within the active site depression of E. coli III Glc , the N-3 (or N ⑀2 ) atoms of each imidazole ring in His 75 and His 90 lie so close that they are cross-linked by chloroplatinate (7).
In an attempt to understand the pleiotropic effects of III Glc , two mutant proteins were constructed by substituting glutamine either for His 90 or for His 75 , yielding proteins designated H90Q III Glc and H75Q III Glc , respectively (23). The H90Q III Glc mutant could not be phosphorylated, as expected, but showed only some regulatory properties of the wild type protein, which we did not expect. Thus, His 90 is critical in both glucose translocation and regulation. H75Q III Glc also behaved unpredictably. While it accepts the phosphoryl group from phospho-HPr, phospho-H75Q III Glc is inactive in sugar phosphorylation (23), presumably because it does not transfer the phosphoryl group to the membrane protein II Glc . We suggested (23) that His 75 may be directly involved in the phosphoryl transfer sequence: PEP 3 Enzyme I 3 HPr 3 III Glc (His 90 ) 3 III Glc (His 75 ) 3 II Glc (Cys 421 ) 3 glucose.
Alternatively, the rate of phosphoryl transfer to the membrane protein II Glc (and then to sugar) is too slow to be detected. That is, phosphoryl transfer proceeds from III Glc (His 90 ) directly to II Glc (Cys 421 ), and His 75 plays an essential role in determining the rate of the reaction. In the accompanying paper (24), we report that replacing His 75 with Gln has a 200-fold effect on the rate of phosphoryl transfer between HPr and III Glc .
The results obtained with the mutants are not easily explained by the structural information on the wild type protein.
We have therefore conducted detailed analyses of the consequences of the two mutations in the active site of III Glc , and these provide information on the mechanism of action of this important signal-transducing protein.

EXPERIMENTAL PROCEDURES
Preparation of Mutant Proteins-The III Glc mutants H75Q III Glc and H90Q III Glc were overexpressed as described in the accompanying report (24). For NMR studies the cells were grown as described (9,25) in the minimal medium of Neidhardt et al. (26) using 15 NH 4 Cl as the source of nitrogen. In addition, the medium was supplemented with 0.2% glucose, 2 mg/ml thiamine, and 50 mg/ml ampicillin. Approximately 50 mg of H75Q III Glc and 20 mg of H90Q III Glc were obtained from 1.5 liters of medium. The purity of each sample was found to be greater than 97%, based on SDS-PAGE followed by quantitative densitometric scanning of the Coomassie-stained gel. Enzyme I and HPr were obtained as described (27,28).
Crystallization of Proteins and X-ray Diffraction Analysis-Crystals of H75Q III Glc and H90Q III Glc were prepared as reported previously for wild type III Glc (7). Diffraction data were collected using a Xuong-Hamlin multiwire detector and reduced using the supplied software (29). (2F o Ϫ F c ) and (F o Ϫ F c ) electron density maps were inspected, and model building was performed using the FRODO program (30). In some instances, electron density was not apparent for some or all side chain atoms, so these residues were truncated. The initial electron density maps were calculated using the refined model of wild type III Glc for phase calculation, after removal of the mutated side chain. Crystallographic refinement was performed using the TNT program suite with the conjugate direction option for function minimization (31,32).
Calorimetry-Differential scanning calorimetry experiments were performed in a DSC92 which is a DASM1 calorimeter modified at the Biocalorimetry Center (The Johns Hopkins University, Baltimore, MD). The calorimeter is interfaced to a PC computer via a Data Translation DT-2801 A to D board. The cell volume was 1.10 ml. Samples were scanned from 25°C to 100°C at a scan rate of 1°C/min. Recovery of structure upon cooling was checked by rescanning the samples.
from an unstructured chain to an ordered structure in the complex. This information is transmitted through the polypeptide backbone to the catalytic cleft in the enzyme, about 30 Å away, resulting in inhibition of catalysis. (d) The addition of Zn 2ϩ results in even greater inhibition of GK catalytic activity. The Zn 2ϩ complexes to the two His residues in III Glc , a Glu residue in GK, and a molecule of water (20,21). Both His residues are required for formation of the Zn 2ϩ complex. (e) Phospho-III Glc cannot bind to GK for steric reasons, and perhaps because of the negative charge. Thus, it is the ratio of phospho-III Glc to III Glc that determines the degree of inhibition of the enzyme, and we believe that this ratio is crucial to all physiological phenomena involving III Glc (3, 6).

FIG. 1. CPK representations of III Glc and phospho-III Glc .
One face of the protein is shown with emphasis on the hydrophobic patch and the active site. The orientation is similar to those shown in Fig. 9. A, a view of III Glc derived from the x-ray coordinates of the crystal structure. The hydrophobic patch is black, the imidazole ring of His 75 lies above that of His 90 (carbon atoms red, N atoms green), the hydrogen bonding atoms are white: Thr 73 O, top, next to N␦1 of His 75 ; Gly 92 carbonyl O, next to N␦1 of His 90 , at the lower end of the active site; Asp 94 N is to the right of Gly 92 . B, simulation of phospho-III Glc . The coordinates of the phosphoryl group (P atom, yellow; O atoms, blue) were used to construct the model, where it is linked to N⑀2 of His 90 . The simulation is only an approximation (a crystal structure of phospho-III Glc is not available), and does not permit, for example, a hydrogen bond to form between Asp 94 and a phosphoryl O atom, as indicated by the NMR spectrum (see Fig. 9). The NMR data also suggest that Val 96 may hydrogen-bond to an O atom of the phosphoryl group.
Data were analyzed using the CPPLUS6 program for the Macintosh developed at the Biocalorimetry Center. After normalizing for concentration, the melting temperature, T m , the enthalpy change, ⌬H o , and the heat capacity change, ⌬C p , for unfolding were determined by fitting the endotherms to a two-state transition.
NMR Spectroscopy-The preparation of protein samples for NMR experiments has been reported (8,9). Briefly, for experiments con- To study the pH dependence of the NMR signals, the pH of the sample was adjusted by adding small aliquots of aqueous potassium hydroxide or HCl. Meter readings were not corrected for deuterium isotope effects. 1 H-15 N HMQC NMR experiments (34) were used to identify the imidazole 2 J NH couplings of the single histidine residue of each mutant protein (11). The spectra were acquired on a Bruker AMX-500 spectrometer with the sample dissolved in D 2 O at 36.5°C and at pH values ranging from 6.12 to 8.85 H75Q III Glc , 6.12 to 8.70 phospho-H75Q III Glc , and 6.1 to 9.1 H90Q III Glc . The delay during which 1 H and 15 N signals become antiphase was set to 22 ms to refocus magnetization arising from 1 J NH couplings. For each spectrum 32 scans (512 complex points) were signal-averaged for each of 64 complex t 1 points using the States-TPPI method (35) of quadrature detection in t 1 . The 1 H transmitter was set to 4.67 ppm (H 2 O), and the 15 N carrier was set to 180 ppm. Acquisition times were 85.1 ms and 7.9 ms for the t 2 and t 1 time periods, respectively. 1 H-15 N HMQC NMR experiments utilizing a 1:1 echo H 2 O suppression sequence (34) were used to identify imidazole single-bond 1 J NH couplings. The spectra were acquired on a Bruker AM-500 spectrometer at 36.5°C and pH values of 8.8 ( H75Q III Glc ), 8.70 (P-H75Q III Glc ). The excitation maximum was set to 10.0 ppm, with the 1 H transmitter set to 4.67 ppm and the 15 N transmitter set to 180 ppm. The delay during which 1 H and 15 N signals become antiphase was set to 4.5 ms. For each spectrum 16 scans (1024 real points) were signal-averaged for each of 256 t 1 points using the TPPI method (35) to achieve quadrature detection.
Two-dimensional spectra were processed as described (11) using commercial (NMRi, Syracuse, NY) and in-house software (36). Chemical shifts are referenced to H 2 O ( 1 H, 4.67 ppm from TSP at 36.5°C) and external liquid ammonia ( 15 N) (37). Uncertainties in chemical shifts are Ϯ 0.02 ppm and Ϯ 0.2 ppm for 1 H and 15 N signals, respectively.

RESULTS 4
Crystal Structures of Mutant Proteins-H75Q III Glc and H90Q III Glc crystallized essentially isomorphously to wild type in space group R3. Diffraction data were collected from a single crystal of each. Data merging and atomic model statistics are summarized in Table I. The H75Q III Glc crystal diffracted to slightly better than 2.2 Å resolution, and 73% of the theoretically possible reflections were observed. The H90Q III Glc crystal diffracted to somewhat higher resolution, and the data set is 81% complete to 2.1 Å resolution. The initial electron density maps clearly revealed the position and orientation of the mutated side chains, so model building and refinement were straightforward. The final atomic model statistics are excellent, with good stereochemistry and crystallographic R-factors of about 0.18 in each case. Portions of the final (2F o Ϫ F c ) electron density maps in the vicinity of the mutated side chain are presented in Fig. 2.
The models in the vicinity of the mutation are shown in more detail in Fig. 3, revealing proposed hydrogen bond interactions with neighboring groups. The electron density maps do not reveal an unambiguous orientation for the terminal amidocarboxy group of glutamine. However, for the mutant H90Q III Glc this could be deduced from the hydrogen bonding pattern. The N⑀2 donates a hydrogen bond to the carbonyl of Gly 92 , much as His 90 N␦1 does in wild type (7,11). The O⑀1 may accept a hydrogen bond from His 75 N⑀2, but the distance between these atoms is 4.4 Å, which is generally considered to be too long for a hydrogen bond (38). Therefore the orientation is very likely as shown in the figures and is consistent with the chemical shift observations. This side chain makes two hydrogen bonds with well ordered solvent molecules so that its hydrogen bonding potential is nearly saturated. On the other hand, glutamine 75 does not appear to make good hydrogen bonds to neighboring groups. The most likely orientation, which is also consistent with the NMR data, is shown with the Gln 75 N⑀2 2.7 Å from His 90 N⑀2. This implies a hydrogen bond between the two groups, but the geometry is extremely poor. The Gln 75 O⑀1 is also 2.9 Å from the hydroxyl of Thr 95 , and might accept a hydrogen bond as suggested in Fig. 3A. However, the geometry is again extremely poor, with the putative bond approximately perpendicular to the plane of the amidocarboxy group. A further caveat is that the electron density for the Thr 95 side chain is poorly defined, and the orientation shown must be considered tentative.
Finally, a superposition of the mutant and wild type structures are shown in Fig. 4 (A and B). H75Q III Glc superimposes onto wild type with an r.m.s. deviation of 0.25 Å for the main chain atoms, which is within the estimated error of the two structure determinations. Likewise, H90Q III Glc superimposes on wild type with an overall r.m.s. deviation of 0.19 Å. However, there are small differences in the structure indicative of a concerted shift that is significant. In Fig. 4A, the main chain and side chains of residues 90 -92 move slightly relative to wild type to accommodate the somewhat less bulky Gln 75 . Fig. 4A further suggests that the rotamer configuration of the side chains of Thr 73 and Thr 95 have changed, but the electron density of each is poorly defined. In Fig. 4b, a portion of the overlay of H90Q III Glc and wild type is shown, which reveals no systematic differences except for the configuration of the mutated side chain.
Calorimetry-DSC melting curves for the wild type and mu- tant proteins are shown in Fig. 5. The fitted melting parameters are listed in Table II. The fitted values of T m , ⌬H°, and ⌬C p can be used to calculate the values of ⌬H°, ⌬S°, and ⌬G°at any temperature using the following equations.
T is the temperature of interest, and T m is the melting temperature, both in units of Kelvin. The thermodynamics of unfolding at 69.1°C (298 K), the T m of wild type III Glc , are given in Table III. The energetic changes summarized in Table III can be rationalized in view of the structures of the wild type and mutant proteins. In the H90Q III Glc mutation, the glutamine side chain appears to make good hydrogen bonds to neighboring groups and is not fully desolvated. Hence, this mutation is enthalpically stabilizing (ϳ6 kcal mol Ϫ1 ). However, the configurational entropy lost upon fixing the glutamine side chain is greater than that for histidine (39) so that the mutation is entropically destabilizing. The enthalpy-entropy compensation results in a marginal decrease in the stability of the protein, even though the hydrogen bonds are enthalpically stabilizing. Similar effects are seen in T4 lysozyme mutants (40), as has recently been discussed (41).
In the H75Q III Glc mutation, the glutamine side chain does not form good hydrogen bonds. This is reflected in the unfolding enthalpy, which is decreased by about 6 kcal mol Ϫ1 . However, the destabilization in ⌬H°is offset by a stabilizing change in ⌬S°. The stabilizing entropy probably reflects the mobility of the glutamine side chain, which is not restricted by hydrogen bonds.
The atomic models of the two mutant III Glc proteins provide support for the conclusions of the thermodynamic analyses. Inspection of the vibrational parameters, or "B factors" of the two mutant side chains reveals very high values (in the range of 80 -100 Å 2 , corresponding to an r.m.s. vibrational amplitude of approximately 1 Å and consequent poor localization) for the terminal amidocarboxyl group of Gln 75 , while the terminal group of Gln 90 has values of 20 -30 Å 2 , due to its rigid fixation in the active site by the hydrogen bond to the carbonyl oxygen of Gly 92 .
NMR Spectroscopy: Tautomeric States of the Active Site Histidines-15 N nuclear magnetic resonance spectroscopy can provide detailed information about the protonation states of histidine rings in proteins (42)(43)(44)(45)(46)(47)(48)(49). The great utility of the method is based on the sensitivity of the 15 N shift to the protonation state of the nitrogen (43-45, 48, 50). To simplify the discussion, protonated nitrogens that are incorporated into neutral and positively charged rings have been denoted type-␣ and type-␣ϩ, respectively, and unprotonated nitrogens have been denoted type-␤ (47,50). More recently, van Dijk et al. (51) have extended the nitrogen chemical shift measurements to include phosphorylated imidazole rings. The nomenclature and characteristic 15 N chemical shifts for the various nitrogen species are summarized in Table IV. These results have been used to characterize the tautomeric states of the histidines in ␣-lytic protease (46), triosephosphate isomerase (52), HPr (51), and E. coli ␤-hydroxydecanoyl thiolester dehydrase (53). In each of these studies, however, specific incorporation of 15  labeled histidine was required to assign the nitrogen signals.
Recently van Dijk et al. (54) and we (11) have utilized relatively strong 2 J NH couplings (43) to assign the histidine nitrogen signals in uniformly 15 N-labeled proteins via two-dimensional 1 H-15 N HMQC spectra. In particular, the 15 N ⑀2 nitrogen can be assigned on the basis of strong correlations with both H ␦2 and H ⑀1 via 2 J N ⑀2 H ⑀1 and 2 J N ⑀2 H ␦2 couplings, and the N ␦1 signal can be assigned on the basis of a strong correlation with H ⑀1 , and in some cases a weak correlation with H ␦2 via 2 J N ␦1 H ⑀1 and 3 J N ␦1 H ␦2 couplings, respectively. This experiment, along with the previously determined canonical 15 N chemical shift data (Table IV) were used previously to characterize the active site histidines in IIA Mtl (54) and both III Glc and P-III Glc (11).
Two-dimensional 1 H-15 N HMQC spectra for H75Q III Glc and phospho-H75Q III Glc are shown in Fig. 6A. For H75Q III Glc the downfield-shifted nitrogen signal (242.6 ppm) is correlated to two protons (H ␦2 and H ⑀1 ) and can therefore be assigned to N ⑀2 . Conversely, the upfield-shifted signal (168.6 ppm) is correlated to only H ⑀1 and can therefore be assigned to N ␦1 . Comparison of the nitrogen chemical shifts with the canonical values (Table  IV) shows that H90 N ␦1 and N ⑀2 are type-␣ and type-␤, respectively, and thus, H90 in H75Q III Glc exists predominantly as the N ␦1 -H tautomer. These results are shown schematically in Fig.  6 (inset).
For phospho-H75Q III Glc the nitrogen assignments are complicated by the fact that both N ␦1 and N ⑀2 correlate strongly with both protons (Fig. 6B). The additional N ␦1 -H ␦2 correlation, which results from a 3 J N ␦1 H ␦2 coupling, was also observed for phospho-H90 in spectra of wild type III Glc (11) and may be characteristic of phosphohistidine residues. To resolve the ambiguity in nitrogen assignments, the rates of hydrolysis of the phosphoproteins were determined as a function of pH. Phospho-histidinyl proteins are sensitive to acid hydrolysis, and phospho-N ␦1 , which is found in phospho-HPr, is much more labile than phospho-N ⑀2 (55). The rate constants for the hydrolysis of phospho-III Glc and phospho-H75Q III Glc are shown in Fig.  7. The shapes of the curves obtained with the two phosphoproteins are essentially the same. Furthermore, in the pH range above 5, phospho-H75Q III Glc is hydrolyzed about 4-fold more slowly than phospho-III Glc , rather than much more rapidly as would be the case were the phosphoryl group linked to the His N ␦1 atom. Thus, His 90 in the H75Q III Glc mutant is phosphorylated at the same position as the wild type protein (N ⑀2 ).
It follows that the signal at 211.0 ppm, which is close to the chemical shift expected for a phosphorylated nitrogen in a charged ring (Table IV), is due to N ⑀2 and that the signal at 172.0 ppm is due to N ␦1 . Moreover, comparison of the chemical shift of N ␦1 with the canonical values indicates that this nitrogen is protonated and that the ring is charged at pH 8.4. These data are shown schematically in Fig. 6B (inset).
An 1 H-15 N HMQC spectrum of the single histidine (H75) of the H90Q III Glc mutant (pH 7.66) is shown in Fig. 8. In this spectrum it can be seen that N ⑀2 (245.0 ppm), which is coupled to two protons, is type-␣ while N ␦1 (164.9 ppm), which is coupled to a single proton, is type-␤. Hence, H75 exists predominantly in the N ⑀2 -H state in the H90Q III Glc mutant as was the case in the wild type protein (11).
Effect of pH-As stated above, the protonation state of a nitrogen strongly influences its chemical shift (Table IV), and the imidazole 15 N signals therefore serve as particularly sensitive parameters for monitoring the titration of a histidine residue. To measure the pK a values of the histidines in the mutant proteins, imidazole 15 N and C-H chemical shifts were obtained from 1 H-15 N HMQC spectra acquired at pH values ranging from 6.10 to 9.0. Data were not collected outside this range due to aggregation of the mutant protein at lower pH values and inefficient regeneration of the phosphorylated protein at higher pH values. The data for H75Q III Glc , P-H75Q III Glc , and H90Q III Glc revealed only minimal changes in both the 1 H and 15 N signals over the stated pH range (see Figs. S4 and S5 in supplementary material). 4 The largest change was observed for His 90 N ⑀2 in unphosphorylated H75Q III Glc , which shifted upfield 5.0 ppm. By comparison, a type-␤ nitrogen is expected to shift upfield approximately 72 ppm (to type-␣ϩ) upon protonation. The invariance of 1H and 15N shifts with pH indicate that the protonation state of the single histidine in each mutant of III Glc does not change over the pH range 6.1 to 9.0. Furthermore, these data show that the pK a values for His 90 in H75Q III Glc and His 75 in H90Q III Glc are both less than 5.5 and that the pK a for phospho-His 90 in phospho-H75Q III Glc is greater than 10.
Hydrogen Bonding Pattern Associated with Active Site Histidines-The x-ray data presented above for H75Q III Glc suggested the presence of a hydrogen bond between His 90 N ␦1 and the carbonyl group of Gly 92 . In addition, the NMR results show that this nitrogen is protonated and can therefore act as a hydrogen bond donor. To further elucidate this issue, attempts were made to identify this signal in 1 H-15 N HMQC spectra. In general nitrogen-attached imidazole protons cannot be observed because of rapid exchange with solvent (56). In favorable cases, however, such signals can be observed in spectra that utilize a 1:1 echo H 2 O suppression sequence (34) if the exchange rate is reduced, for example through hydrogen bonding. Analysis of 1 H-15 N HMQC spectra revealed peaks at nitrogen chemical shifts of 169.2 ppm ( H75Q III Glc ) and 173.1 ppm (phospho-H75Q III Glc ) (see Fig. S1 in supplementary material). 4 After correction for 2 H isotope effects (57), these signals could be assigned to His 90 N ␦1 -H in the two forms of the protein. The fact that these signals can be observed provides further evidence for hydrogen bonding between His 90 and Gly 92 in both the phosphorylated and unphosphorylated mutant proteins.   15 N Chemical Shifts are referenced to liquid NH 3 and from NMR solution studies of model compounds as described in the text.
FIG. 5. Differential scanning calorimetry of mutant and wild type proteins. The thermal stabilities of III Glc and its mutants were determined by differential scanning calorimetry. Shown are the excess heat capacities versus temperature. The buffer and native state base lines have been subtracted from the data, and the data have been normalized to the protein concentration. The decreased stability of the mutants is indicated by their lower T m values (approximated by the maximum in the excess heat capacity). The decreased enthalpy change for H90Q III Glc is apparent from the decreased area under the curve. The x-ray data for H75Q III Glc indicate that the side chain of Gln 75 points toward the active site of the protein in a similar manner to the side chain of His 75 in wild type III Glc . Because of this, it is of interest to assign the amide signals of Gln 75 as a probe for interactions of this group with the phosphoryl oxygen atoms. The six pairs of 1 H-15 N signals (4 Asn, 2 Gln) for both H75Q III Glc and phospho-H75Q III Glc were identified by comparison of heteronuclear single quantum coherence spectroscopy (HSQC) spectra of the mutant and wild type proteins (9, 10). For H75Q III Glc the NH 2 signals of Asn 32 , Asn 57 , Gln 111 , and Asn 142 were identical in spectra of H75Q III Glc and wild type III Glc and in spectra of the respective phosphorylated proteins, allowing for their unambiguous assignment (see Fig. S2 and Tables S2 and S3 in supplementary material). 4 The signals for the neighboring residues Asn 74 and Gln 75 were also assigned by comparison with wild type 15 NH 2 chemical shifts. Specifically, for H75Q III Glc the pair of signals assigned to Asn 74 differed significantly less from the wild type Asn 74 chemical shifts (1.0 ppm 15 N, 0.04 and 0.07 ppm 1 H) than the pair of signals assigned to Gln 75 (Ϫ4.5 ppm 15 N, Ϫ0.78 ppm and Ϫ0.60 ppm 1 H). Similarly, for phospho-H75Q III Glc the pair of signals assigned to Asn 74 differed significantly less from those of Asn 74 in phospho-III Glc (difference Ϫ0.5 ppm 15 N; 0.04 ppm and 0.05 ppm 1 H) than from the pair of signals assigned to Gln 75 (difference 0.0 ppm 15 N; 0.62 and 1.51 ppm 1 H) (see Fig. S3 and Tables S2 and S3 in supplementary material). 4 Strikingly, the amide nitrogen of Gln 75 shifts downfield 4.6 ppm and one of the two amide protons shifts downfield 2.08 ppm upon phosphorylation of H75Q III Glc . By comparison, no significant changes were observed in either the amide nitrogen or proton signals upon phosphorylation of wild type III Glc (10), and the shifts in the amide resonances of the other asparagine and glutamine residues of H75Q III Glc were much smaller (Tables S2 and S3). 4 These data, along with the close proximity of the side chain of Gln 75 to that of His 90 in the x-ray structure of H75Q III Glc suggest that the side chain Gln 75 amide group interacts with the phosphoryl oxygens (see below).

DISCUSSION
Substituting Gln for His 75 or His 90 in E. coli III Glc has virtually no effect on the structure of the protein outside of the active site. Within experimental error, the crystal structures show that all of the atoms, except Gln and His, are superimposable in the wild type and mutant proteins. The only significant differences in the active sites among the three proteins and the two phosphorylated derivatives are illustrated in Fig.  9. The small changes in the thermodynamics of unfolding of wild type, H90Q III Glc , and H75Q III Glc proteins are explained by the differences in hydrogen bonding and are fully consistent with the crystal and NMR structures.
While H90Q III Glc cannot accept the phosphoryl group from phospho-HPr, we predicted that it would behave similarly to wild type III Glc in regulation. However, in vivo data (23) showed that H90Q III Glc did not significantly inhibit utilization of non-PTS sugars, whereas H75Q III Glc was as effective as wild type III Glc . In vivo data are difficult to interpret, especially when they involve expression of operons, but definitive information comes from in vitro inhibition studies with one of the target proteins, glycerol kinase (GK). 3 III Glc binds to GK primarily via the "hydrophobic patch," and this does not involve the His residues in the absence of Zn 2ϩ . In recent experiments, Dr. Donald W. Pettigrew (Texas A&M University) did not detect inhibition of GK by H90Q III Glc , i.e. it was less than 3% as effective an inhibitor as the wild type protein or H75Q III Glc . From this, it appears that either H90Q III Glc binds poorly to GK or it binds as well as wild type III Glc , but is unable to constrain the GK polypeptide to the manner required for inhibition of catalytic activity. We predict that the binding constant of the mutant to GK will be significantly less than that of wild type III Glc . The crystal structures appear to show that H90Q III Glc contains two bound, structured water molecules (to the Gln amide), whereas neither wild type III Glc nor the III Glc /GK complex contains structured water. Thus, we suggest that the energy required for desolvation of the Gln in H90Q III Glc leads to a decreased binding constant of the mutant to GK, making it a poor inhibitor.
The remainder of this discussion concerns the mechanism of phosphoryl transfer between III Glc and HPr.
It is beyond the scope of this paper to consider all the factors involved in catalyzing this reaction, but it appears that the active sites of HPr and III Glc must complement each other almost perfectly to accommodate a very rapid phosphoryl transfer. In wild type III Glc , the required tautomers of His 90 and His 75 are fixed by their respective hydrogen bonds to Gly 92 and Thr 73 . Also, the backbone amide of Asp 94 (and possibly Val 96 ) is positioned to form an H bond to the phosphoryl group.
Assuming an associative phosphoryl transfer mechanism, the P atom passes from a tetrahedron in phospho-HPr, through a trigonal bipyramid in the transition state, to an inverted tetrahedron in phospho-III Glc (Fig. 10). To bring the hydrated phosphoryl group of phospho-HPr 5 into the transition state, we assume that hydrogen bonds to water are replaced by H bonds to N ⑀2 of His 75 and the amide of Asp 94 (and possibly Val 96 ). Since these hydrogen bonds are presumably optimized to stabilize the transition state, they are likely to be stronger than those from water to the phosphoryl group, and this should enhance the electrophilicity of the P atom. In turn, the H bond from the carbonyl of Gly 92 to His 90 N ␦1 should make the reactive N ⑀2 more nucleophilic.
In short, it appears that the hydrogen bonds provide many of the factors required for optimum transition state formation (59): charge shielding and stabilization, proximity and orientation, immobilization of the reactive groups, increased nucleophilicity of the reactive N ⑀2 , and electrophilicity of the P atom.
But why are the rate constants 200-fold lower when His 75 is replaced by Gln? We surmise that the relative rates are determined by differences in the number and/or strengths of hydrogen bonds in the mutant and wild type proteins. A 200-fold rate difference translates to a difference of about 3 kCal/mol in the free energies of activation ⌬G ‡ for the two reactions, A and B, well within the range (1-5 kCal/mol) for formation of "normal" hydrogen bonds (60).
Hydrogen bonds can influence reaction rates by delocalizing unfavorable charge (61). As the phosphoryl group approaches bipyramidal geometry in the transition state, the total charge on the O atoms changes from Ϫ2 to a value between Ϫ2 and Ϫ3. The hydrogen bond to His 75 can serve to delocalize this increased charge by resonance stabilization to a greater extent 5 Unlike His 75 and His 90 in III Glc , which show abnormally low pKЈ a values, the single His residue of HPr, His 15 , is exposed to the solvent and exhibits a pKЈ a of 5.6 (63). In phospho-HPr, the phosphoryl group is linked to N ␦1 of the His, and the phosphoryl O atoms are undoubtedly hydrogen-bonded to water.

FIG. 7.
Effect of pH on the rate of hydrolysis of phospho-III Glc and phospho-H75Q III Glc . Samples of the proteins were labeled with [ 32 P]PEP as described in the accompanying paper (24). Each phosphoprotein was dissolved in 17 mM KHCO 3 , 3 mM K 2 CO 3 , warmed to 37°C, and diluted with 4 volumes of one of the following 0.2 M buffers: pH 1 or 2, HCl/KCl; pH 3 to 6, citrate/PO 4 ; pH 7 or 8, PO 4 ; pH 9 or 10, KHCO 3 /K 2 CO 3 . Triplicate samples were taken from each reaction mixture at time intervals that ranged from 22 min to 6 h. The aliquots were immediately transferred to one third their volumes of the mixture (5 M urea, 3 M KOH) used for the rapid quench experiments described in the accompanying paper (24), and stored at Ϫ70°C. To measure the baseline level of hydrolysis, protein samples were mixed with the quench solution and frozen immediately. The pH of each incubation mixture was measured at the beginning and end of each time course, and the quantity of phospho-protein and inorganic phosphate in each sample was analyzed by gel filtration chromatography (24). Rate constants were estimated from semi-logarithmic plots of remaining phospho-protein versus time. ϩ, phospho-III Glc ; f, phospho-H75Q III Glc . than is possible with Gln 75 , which has less side-chain conjugation.
Hydrogen bonds can also show cooperativity (62), and it is important to note that there is a hydrogen bond network in phospho-III Glc (Fig. 9), Thr 73 -His 75 -O Ϫ -phosphate, analogous to the Ser 195 -His 75 -Asp 102 network in chymotrypsin and related serine proteases. Thr 73 does not hydrogen-bond to Gln 75 in the mutant protein, which is therefore deficient in one such bond compared to the wild type. Conceivably, this network makes a significant contribution to the phosphotransfer reaction mechanism in the wild type protein.
In sum, the properties of the mutants relative to wild type III Glc are primarily ascribed to differences in the strength and/or number of a few hydrogen bonds within the active site. These theories can be tested. We are currently attempting to replace Thr 73 by Ala, Val, and Ser, and His 75 and His 90 by Leu. A quantitative study of these mutant proteins may provide direct evidence, for or against the hypotheses offered above.