NMR Dynamic Studies Suggest that Allosteric Activation Regulates Ligand Binding in Chicken Liver Bile Acid-binding Protein*

Apo chicken liver bile acid-binding protein has been structurally characterized by NMR. The dynamic behavior of the protein in its apo- and holo-forms, complexed with chenodeoxycholate, has been determined via 15N relaxation and steady state heteronuclear 15N(1H) nuclear Overhauser effect measurements. The dynamic parameters were obtained at two pH values (5.6 and 7.0) for the apoprotein and at pH 7.0 for the holoprotein, using the model free approach. Relaxation studies, performed at three different magnetic fields, revealed a substantial conformational flexibility on the microsecond to millisecond time scales, mainly localized in the C-terminal face of the β-barrel. The observed dynamics are primarily caused by the protonation/deprotonation of a buried histidine residue, His98, located on this flexible face. A network of polar buried side chains, defining a spine going from the E to J strand, is likely to provide the long range connectivity needed to communicate motion from His98 to the EF loop region. NMR data are accompanied by molecular dynamics simulations, suggesting that His98 protonation equilibrium is the triggering event for the modulation of a functionally important motion, i.e. the opening/closing at the protein open end, whereas ligand binding stabilizes one of the preexisting conformations (the open form). The results presented here, complemented with an analysis of proteins belonging to the intracellular lipid-binding protein family, are consistent with a model of allosteric activation governing the binding mechanism. The functional role of this mechanism is thoroughly discussed within the framework of the mechanism for the enterohepatic circulation of bile acids.

Recent studies have shown that bile acids not only serve as the physiological detergents that facilitate absorption, transport, and distribution of lipid-soluble vitamins and dietary fats but also are the signaling molecules that activate nuclear receptors and regulate bile acid and cholesterol metabolism. In addition, bile acids induce the cytochrome P450 3A family of cytochrome P450 enzymes that detoxify bile acids, drugs, and xenobiotics in the liver and intestine, induce hepatocyte apoptosis, and activate the gene encoding a candidate bile acid transporter protein (1). Given the important role of bile acids, the study of their transport at a molecular level is of special medical and pharmacological interest. In this line it is essential to gain insight into the three-dimensional structures and dynamic behavior of proteins, in their free and complexed forms, involved in bile acid recycling.
Interestingly, bile acids have been suggested to be the putative ligands of a group of intracellular lipid-binding proteins (iLBPs) 2 or fatty acidbinding proteins (FABP), expressed in the liver of nonmammalian species, and referred to previously as liver basic FABP. FABPs have been classified and described on the basis of the organ that they were initially isolated from, but several instances are known in which more than one FABP type has been shown to be produced by a single tissue. We have reported previously on the higher similarity of liver basic FABPs from nonmammalian species with ileal lipid-binding protein (ILBP) rather than with mammalian liver FABP (2). In agreement with this observation, bile acid binding and transport is emerging as the specific function of the liver nonmammalian subfamily, hence called liver bile acid-binding protein (BABP) (2,3). At variance, the paralogous proteins expressed in the same tissue but in mammals play a role in fatty acid binding and transport (4). A multiple alignment of all the known sequences of nonmammalian liver BABPs with ILBPs is reported in Fig. 1.
It has been proposed that internal protein dynamics in iLBPs could be intimately connected with ligand recognition and interaction (2,(5)(6)(7)(8). We report here a structural and dynamic study on chicken liver BABP (cl-BABP), in its apo-and holo-form, combining heteronuclear NMR experiments and 15 N NMR relaxation measurements with MD simulations. We investigate the role of the protonation state of a buried histidine on protein dynamics. We discuss here the observed change in dynamics upon ligand binding in terms of an allosteric activation mechanism, i.e. a shift between inactive and active conformations (9). The proposed mechanism for ligand binding in cl-BABP is further analyzed FIGURE 1. ClustalW multiple alignment of proteins belonging to iLBP family. The alignment includes the 13 "liver basic" fatty acid-binding proteins from nonmammalian species and the five known ILBPs. Secondary structure elements are highlighted at the top of the sequences.
in light of data reported for other members of the iLBP family and discussed as functional to bile acid enterohepatic circulation.

MATERIALS AND METHODS
Protein Expression and Purification-Recombinant cl-BABP was expressed as soluble protein in Escherichia coli BL21 (DE3) bearing the recombinant plasmid pET24d. Transformed cells were grown on plates containing 50 g/ml kanamycin. One liter of LB was inoculated with an overnight culture and incubated at 310 K until cells reached an A 600 of 0.8. Protein expression was induced by addition of 0.7 mM isopropylthiogalactopyranoside and incubation continued overnight at 293 K. The cells were harvested and resuspended in lysis buffer (50 mM Tris, 10% sucrose, 1 mM EDTA, 10 mM ␤-mercaptoethanol, pH 8.0). After lysis, the supernatant, containing cl-BABP, was loaded on a DEAE-cellulose (Whatman) anion exchange column equilibrated with 50 mM Tris acetate, pH 7.8. The same buffer was used for protein elution. Fractions containing cl-BABP were concentrated and resolved on a Sephacryl S-100 HR (Amersham Biosciences) column equilibrated with 50 mM Tris-HCl, 0.2 M NaCl, pH 7.2. cl-BABP was delipidated as described (10). The protein purity was checked by the presence of a single band on SDS-PAGE and by mass spectrometry. The protein yields were 90 mg/liter of bacterial culture. 15 N isotope labeling was achieved using M9 minimal media containing 1 g/liter 15 NH 4 Cl, following protocols reported in the literature (11). The extent of 15 N labeling was verified by MALDI mass analysis, and the isotope incorporation was found to be more than 92%. [ 15 N]cl-BABP was obtained in a yield of 50 mg/liter of minimal media. 13 C, 15 N double labeling was obtained with the same procedure using M9 minimal media containing 1 g/liter 15 NH 4 Cl and 4 g/liter 13 C-enriched sucrose. The extent of labeling, verified by MALDI mass analysis, was Ͼ90%, and yields of 25 mg/liter of minimal media were obtained. Commercial chenodeoxycholic acid (Sigma) was employed for the preparation of holo-cl-BABP with a ligand to protein ratio 5:1, as described previously (8).
NMR Experiments-NMR data were recorded on Bruker Avance 500, 600, and 700 MHz spectrometers equipped with pulse field gradient triple-resonance probes. 0.5 mM protein samples in phosphate buffer at pH 7.0 and 5.6 and 298 K were employed for structure determination and relaxation measurements.
The following triple resonance experiments, using standard parameter sets (16), were recorded on the doubly labeled [ 15 N, 13 (17). Two NOESYtype three-dimensional experiments (mixing 100 ms) were acquired, one optimized for aliphatic and one for aromatic residues.
A series of two-dimensional 1 H-15 N HSQC experiments was performed for the apoprotein at different pH values (in the range 4. 2-7.4) to allow for measurement of the midpoint of the chemical shift pH-driven titration. Spectra were assigned on the basis of the assignments obtained at pH 7.0 and 5.6.
The 15 N chemical shift titration data were fitted to Equation 1 to evaluate pK a values (18), where ␦ p and ␦ d are the chemical shifts of the protonated and the deprotonated state, respectively. Calculation of 1 H and 15 N secondary shifts was performed according to d ϭ ((⌬d HN 2 ϩ ⌬d N 2 /25)/2) 1/2 (19). 15 N relaxation experiments (20), run as water flip-back version, were acquired at 600 and 700 MHz both at pH 7.0 and 5.6. Eleven delays (2.5, 20, 60, 100, 150, 200, 300, 400, 600, 800, and 1000 ms) were used for T 1 measurements, and nine delays (16. (20). To analyze the exchange contribution to relaxation at pH 7, T 1 , T 2 , and 1 H-15 N NOEs were also measured at 500 MHz, in the same conditions as described for higher field measurements.
Relaxation measurements were identically performed at pH 7.0 for holo-cl-BABP complexed with chenodeoxycholate. Data were processed with XWINNMR and NMRPipe (21) and analyzed with NMR-View 5.0.3 software package (22).
Structure Calculation of Apo-cl-BABP-Volume integration was performed on the three-dimensional 15 N-13 C NOESY and 1 H-15 N HSQC-NOESY spectra using NMRView (22). Peak volume calibration was performed using the median method and a routine of NMRView program, and the obtained list of distances was used as input for DYANA (23) calculations. angle restraints were derived from J HN,Ha coupling constants estimated from three-dimensional HNHA experiments (15). angle restraints of 139 Ϯ 30°for J HN,Ha coupling constants greater than 8.0 Hz and 60 Ϯ 30°for J HN,Ha coupling constants smaller than 5.0 Hz were used as restraints.
Amide proton exchange rates were estimated from a series of 1 H-15 N HSQC spectra performed at different times after dissolving the protein in D 2 O (data not shown). The partners for all hydrogen bonds were assigned on the basis of preliminary structures obtained by imposing only NOE restraints. Each hydrogen bond was introduced as a restraint on O-N distance of 3.00 Å and HN-O distance of 2.00 Å. The decision was taken to introduce in the calculation only totally unambiguous restraints, i.e. those correlations that were not affected by overlap in any spectra.
The restraints were re-examined to check for consistent violations. One hundred calculations were run employing DYANA (23), and the 20 conformers with the lowest residual target function were analyzed. The 20 final DYANA structures were further refined using the AMBER force field, as implemented in the program DISCOVER (Molecular Simulations, San Diego). A dielectric constant of 4 ϫ r was used, and a scaling factor of 10 was used for out-of-plane interactions. Each structure was minimized performing 100 steps of steepest descent and 300 steps of conjugate gradient. The 10 structures with the lowest potential energy were selected for further analysis. The structures were deposited in the PDB with code 1zry.
Relaxation Data Analysis-Relaxation times were calculated via least squares fitting of peak intensities, using the rate analysis routine of NMRView program (22). The heteronuclear NOE effects were calculated from the ratio of cross-peak intensities in spectra collected with and without amide proton saturation. The principal components of cl-BABP inertia tensor were calculated using Pdbinertia (A. G. Palmer III, Columbia University). The principal moments of inertia of apo-cl-BABP at pH 7 were calculated on the basis of our NMR structure (PDB code 1zry), whereas at pH 5.6 the representative coordinates from MD simulations were used (see below). For holoprotein, the x-ray structure (PDB code 1tw4) was employed. Isotropic and anisotropic models were tested for apo-and holo-cl-BABP.
An initial estimate of the overall correlation time and of principal components and orientation of the diffusion tensor can be reliably determined from the angular dependence of the relaxation rates of a subset of NH vectors assumed to have a negligible component of internal motion and/or exchange contribution to 15 N relaxation. The selection of the subset of residues was made following the procedures described in the literature (24): residues with NOE Ͻ0.65 were removed from the data set, and residues with low T 2 values (T 2 Յ ͗T 2 ͘ Ϫ T2 ) were removed from data set unless their corresponding T 1 values were high (T 1 Ն ͗T 1 ͘ ϩ T1 ), indicating that they could be affected by anisotropic tumbling.
For the axially symmetric model D ʈ and D Ќ , and initial estimates were evaluated using the Quadric Diffusion program (A. G. Palmer III, Columbia University) that uses the quadratic representation approach (25). Relaxation of amide 15 N nuclear spins were analyzed using the standard equations assuming, for a diamagnetic protein, dipolar coupling with directly attached protons and a contribution from the 15 N chemical shift anisotropy (26) evaluated as ⌬ ϭ Ϫ170 ppm. The experimental data were fitted to the Lipari-Szabo model (27) using the program MODELFREE (version 4.0). The extended Lipari-Szabo formalism proposes five spectral density functions that depend upon S 2 (the generalized motional order parameter), m (the overall correlation time of rotational diffusion), e (the effective correlation time), and R ex (the rate of conformational exchange). The five models of motion were iteratively tested in order of increasing complexity, and the model that best fitted the data were selected as described elsewhere (28). After model selection, the overall rotational diffusion model parameters and the internal motional parameters for each spin were optimized simultaneously. At pH 7, the exchange contributions were extracted from the relaxation data at three frequencies using the approach described in Ref. 29. The parameter R 2 Ϫ (R 1 /2) can be expressed as shown in Equation 2, in the assumption of an exponentially decaying autocorrelation function, where Using a plot of R 2 Ϫ (R 1 /2) versus B 0 2 , the spectral density function J(0) can be calculated from the intercept, I 0 ϭ (d 1 2 / 3)J(0), and the exchange constant A can be deduced from the slope, m ϭ (J(0)c 1 2 /3) ϩ A, of the line. In principle, any spin for which the data has a slope m Ͼ c 1 2 I 0 /d 1 2 will have an exchange contribution. However, taking in consideration experimental and fitting errors, a threshold of 1.3 ϫ ͗m͘ was used to determine residues subject to exchange (29), where ͗m͘ is the average slope.
Theoretical pK a Calculations-All pK a calculations have been performed as described previously (30,31). The linear Poisson-Boltzmann equation was solved for different charge states, and the electrostatic free energy was used to estimate pK a shifts. The mid-point of the titration for each site is taken as its pK a value. All Poisson-Boltzmann calculations have been performed using the program UHBD (32).
Molecular Dynamics Simulations-Molecular dynamics simulations were performed using the program GROMACS (version 3.2.1) employ-ing the GROMACS force field (ffgmx2) (33). The protocol used was essentially as described previously for ␤-lactoglobulin (34). The structure of the bile acid-binding protein was taken from PDB code 1zry, model 1. Protons were added using the program pdb2gmx, in the GRO-MACS suite of programs, for optimization of the hydrogen bond network. The protein was first minimized by 200 steepest descent minimization steps, followed by 200 conjugate gradients steps. Because of lack of solvent in this step, the dielectric constant used was 10. The Poisson Boltzmann equation was used to compute the electrostatic potential around the molecule. The lowest potential region at 0.7 nm from any protein atom was chosen for placing a counterion. The procedure was repeated on the protein and ion(s) until the net charge of the system was 0. The minimized protein and ions were then solvated in a box of SPC water with boundaries at least 1.6 nm away from any protein or ion atom. After addition of solvent molecules and ions to the system, long range electrostatic interactions were treated by particle mesh Ewald method with the following parameters: distance for non-bond interaction cutoff 12 Å and spacing for the fast Fourier transform grid 1.2 Å.
The solutes were fixed, and water was energy-minimized by 100 steepest descent minimization steps. A short molecular dynamics run (50 ps) keeping the solutes fixed was performed to let the water soak the system. During this run the time step was set to 1 fs. Finally, the unrestrained system was energy minimized by 200 steepest descent steps and equilibrated in the NTP ensemble for 100 ps.
In all molecular dynamics simulations the system was in equilibrium with a temperature bath at 300 K, with relaxation time constant of 0.1 ps. The system compressibility was that of water, 4.5 ϫ 10 Ϫ5 bar Ϫ1 . The relaxation time for pressure equilibration was 0.5 ps. The initial velocities were set to 0. Two 3.6-ns MD simulations were performed for the low pH form (with the two histidines protonated) and the neutral pH form (with both histidines deprotonated) of cl-BABP. In both cases 100 ps of equilibration time were employed.
The r.m.s.d. from starting structure could be fitted by an exponential with a time constant of 150 ps for both simulated forms, although for the protonated form a much slower, very small but detectable, increase in r.m.s.d. was observed throughout the run. The backbone r.m.s.d. from native, including protein ends and loops, is fluctuating around 2.2 Å after few hundred ps. To make sure that the system was equilibrated (at least in this time range), we repeated all analyses of local fluctuations for the same trajectories truncated at 1.8 ns. No significant difference was found.
Snapshots were taken at 100-ps intervals along the simulations, and these 37 snapshots were used for structural analysis. The snapshot exhibiting the smaller average r.m.s.d. with respect to all other snapshots has been taken as the most representative structure in the ensemble.
All structural analysis have been performed using the program Mol-Mol (36) and the analysis programs of GROMACS. Pairwise superposition has been performed using the program ProFit (A. C. R. Martin; www.bioinf.org.uk/software/profit/).

RESULTS
Apo-cl-BABP NMR Assignment and Structure Calculation-Recombinant cl-BABP has been characterized by 1 H, 13 C, and 15 N NMR. The choice of working at pH 7.0 was dictated by the need to perform structural and dynamic comparisons with the protein in its holo-form at neutral pH. Backbone assignment, performed by a combination of classical three-dimensional NMR experiments, was not straightforward especially for the C-terminal region of the protein corresponding to strands F-I. In this region, breaks in the process of assignment were caused by missing correlations due either to fast exchange of amide protons with solvent and/or to conformational exchange (see below). It was therefore necessary to combine the standard three-dimensional backbone assignment strategy with the sequential assignment strategy. Three-dimensional 1 H-15 N TOCSY/NOESY, performed at pH 5.6, guided the assignment of those amide resonances in fast exchange with solvent at pH 7.0. In this way the assignment was possible for all but six residues, namely Met 73 , Val 90 , Ser 93, Lys 95 , Glu 99 , and Gln 100 , located in a region of the protein mostly affected by conformational exchange, as revealed by 15 N relaxation analysis (see below). The 1 H, 13 C, and 15 N assignments of apo-cl-BABP have been deposited in the BioMagRes-Bank (entry code 6642).
Three-dimensional 1 H-15 N TOCSY/NOESY spectra obtained at pH 5.6 revealed the presence of double peaks for several residues, and unambiguous assignment was possible for Ser 3 (A strand), Gly 44 (BC loop), Phe 47 (C strand), Asp 74 (EF loop), Ala 85 (FG loop), Leu 89 (G strand), and Gly 104 (HI loop). The small difference in chemical shift of major and minor peaks of ϳ20 -120 Hz indicated a time scale of exchange of the order of 0.001-0.01 s. These double peaks provide an indication of slow exchange processes affecting the protein backbone.
Only totally unambiguous restraints, i.e. those correlations that were not affected by overlap in any spectra, were used for structural calculation. In this way a set of 1000 nonredundant NOEs was supplemented as follows: (i) by 26 distance restraints for 13 backbone hydrogen bonds defined on the basis of deuterium hydrogen exchange studies (data not shown), and (ii) by 48 angle constraints derived from J HN-Ha coupling constants. It should be stressed that this protein is highly flexible, as revealed both by H/D exchange and relaxation measurements, and several residues did not exhibit long range NOE correlations (see below). The superposition of the 10 best NMR structures, as obtained after DYANA molecular dynamics simulations followed by energy minimization, reported in Fig. 2, affords an r.m.s.d. backbone (3-125) value of 2.02 Ϯ 0.26 Å. The structural quality of the minimized structures was examined with the PROCHECK-NMR (35). Analysis of the backbone dihedral angles showed that 95% of all non-glycine and non-proline residues in apo-cl-BABP fall within the additional allowed regions of conformational space. Considering that this analysis includes some poorly defined regions located in the C-terminal end, this result can be considered reasonable. The NMR structures have been deposited in the Protein Data Bank as code 1zry. A survey of the quality of structure determination is reported in Table 1.
The distribution of distance restraints per residue accounts for the observed distribution of average global displacement (supplemental Fig.  1). The high backbone dispersion of certain segments of cl-BABP essentially corresponds to residues that showed fewer distance restraints because of either conformational dispersion/mobility or lack of assignment. Even if a few more amides could be detected at pH 5.6, the number of collected restraints did not exceed the 5% of the total restraints obtained at pH 7.0, thus reinforcing the picture of a highly flexible molecule.
Holo-cl-BABP NMR Assignment-1 H and 15 N assignments of cl-BABP complexed with chenodeoxycholate were obtained following the same strategy described for the apoprotein. The assignments of holoprotein are reported in supplemental Table 1. The comparison of apo and holo 1 H and 15 N chemical shifts indicated that regions mostly affected by binding are located in the C-terminal FGHIJ strands (Fig. 3). 15 N Relaxation Data and Model Free Analysis for Apo-cl-BABP-The R 2 /R 1 ratios and heteronuclear NOEs at 700 and 600 MHz for apoprotein at pH 7 are reported in Fig. 4, a and b. The same data obtained at pH 5.6 are reported in supplemental Fig. 2. At both pH values significantly high R 2 /R 1 ratios, indicative of conformational exchange processes, were found for residues located in the C-terminal end of the protein.
Heteronuclear NOE values lower than 0.65, indicative of protein regions with fast internal mobility, were detected mostly for helix II, the loop connecting helix II to strand B, CD, and FG loops.
The principal moments of inertia of apo-cl-BABP were in the ratio 1.0:0.94:0.59 (pH 7) and 1.0:0.85:0.66 (pH 5.6), suggesting that the shape of the molecule does not deviate appreciably from the sphere. However, D ʈ /D Ќ values obtained from the Quadric Diffusion program suggested a slightly different degree of anisotropy for apoproteins (1.2 at both pH values) and holoprotein (1.4) (see below). The relaxation data were therefore analyzed both with the isotropic and axially symmetric model. The results were substantially unchanged for the two models; in the text, the data obtained with the axially symmetric diffusion model are presented to take into account even minor effects because of anisotropy. At neutral pH, correlation times ( m ) of 6.9 Ϯ 0.4, 7.2 Ϯ 0.3, and 7.0 Ϯ 0.5 ns were estimated (20) at 700, 600, and 500 MHz, respectively. The data sets at the three magnetic fields were simultaneously used to perform Lipari-Szabo model free analysis for 86 residues. The final optimized values were m ϭ 7.1 ns and D ʈ /D Ќ ϭ 1.2, and the values for internal motion parameters of the single spins are reported in supplemental Table 2 Table 3. A calculated S 2 average value of 0.91 Ϯ 0.04 was observed. Residues Asp 33 , Thr 57 , and Asp 74 could not be fitted to any model. S 2 , e , and R ex contributions, obtained from model free analysis of the available data at two fields (600 and 700 MHz) and at two pH values, are reported in Fig. 5, and residues affected by motions are mapped in color onto the protein structure (Fig. 6).
Validation of R ex Contributions-To evaluate possible artifacts on R ex estimate, the R ex figures obtained from the model free approach, using model 3 of the spectral density function, were compared with data obtained from two strategies. In the first approach, R 1 and NOE data of residues showing large R 2 values, were fitted to Lipari-Szabo model 1. R ex contributions were derived as R ex ϭ R 2 (experimental) Ϫ R 2 (fitted), and the obtained data are reported in supplemental Table 4. In the second approach, additional relaxation experiments were acquired at 500 MHz, and R 2 Ϫ (R 1 /2) was plotted as a function of the static mag-netic field (29) to determine dR ex /dB 0 . The advantage of this approach is that no model-based assumption is made for the spectral density function. The described analysis was possible for a total of 64 residues, and exchange contributions were detected for 12 residues located in the C-terminal half of the protein, namely in DE and EF loops and FGHIJ strands ( Fig. 3 and supplemental Table 5). A summary of conformational exchange contributions obtained for cl-BABP at pH 7 with all the discussed approaches is presented in Fig. 7.  Table 6. An S 2 average value of 0.90 Ϯ 0.06 was obtained. Residues Asp 33 and Lys 95 did not fit to any model. Residues affected by e and R ex contributions are mapped in color onto the protein structure (Fig. 6).

N Relaxation Data and Model Free Analysis of Holo-cl-BABP-The
Histidine Protonation Equilibrium-A series of 15 N-1 H HSQC spectra recorded in the pH range 4.2-7.4 allowed the determination of the midpoint of the chemical shift pH-driven titration for some residues highly influenced by pH (supplemental Table 7).
The average titration midpoint of Leu 89 , Phe 96 , Ser 97 , and Ile 111 (close to His 98 ) is 5.1 Ϯ 0.1 and that of Gly 65 , Ile 84 , and Ala 85 (close to His 83 ) is 6.2 Ϯ 0.1. It was not possible to obtain data relative to His 83 and His 98 , because of broadening and overlap of their resonances upon lowering pH below 5.5.
Titration curves relative to the mentioned residues are reported in supplemental Fig. 5. Theoretical pK a calculations suggested that only His 98 exhibited a shifted mean pK a of 4.7, whereas a mean pK a value of 5.7 was calculated for His 83 .
MD Simulations-Molecular dynamics simulations were performed for the low pH and neutral pH forms of cl-BABP to investigate the possible role of the equilibrium between the protonated and deprotonated form of the two histidines (His 83 and His 98 ) in affecting the observed dynamics, as reported for other proteins (38) (see below). The limit of 3.6 ns was chosen to sample protein movements taking place in times of the order of 1 ns.
The most representative structures derived from MD simulations performed at acidic and neutral pH values have been superimposed globally. The largest differences involve residues 72-76 and 114 -117 (Fig. 8). These two stretches of the protein partially hinder access of ligands to the cavity of the protein (Fig. 9a). The creation of a net charge inside a protein, as is the case for protonation of the buried His 98 , is not favorable, and it is usually accompanied by solvent exposition of the charged group. Here the charged His 98 remains buried; however, it is involved in a salt bridge with Glu 109 , which in turn loosens to some extent its salt bridge with Arg 120 . This is consistent with the evidence that buried salt bridges mostly occur within salt bridges networks that favor charge dispersal (36).
One striking difference between protonated and deprotonated structures is a hydrogen bond between hydroxyl of Thr 72 and carboxyl of Asp 74 , which is conserved in all snapshots in the deprotonated simulation, but it is never found in the protonated simulation. Residue Asp 74 is instead loosely interacting through a salt bridge interaction with Lys 95 in most of the protonated simulations (Fig. 9b). In addition to the major conformational change observed for Asp 74 , a further change at residues Glu 94 and Lys 95 is observed after 1.3 ns of simulation. This transition does not alter the overall direction of the main chain but enables different interactions for the side chains of Lys 95 . Moreover, upon protonation, a rearrangement of His 98 H-bonds with Glu 109 and Arg 120 takes place, concomitant with the movement of Asp 74 and Lys 95 enabling the formation of a loose salt bridge.
Root mean square fluctuation analysis of backbone atoms, after superposition on the starting (reference) structure, has been performed using the program g_rmsf in GROMACS (www.gromacs.org). For both simulations, the first two N-terminal residues and loops FG, HI, and IJ in the C-terminal part of the molecule are not conformationally well defined. The largest differences in conformational flexibility between the two simulations are observed for the segment 72-80 (entailing loop EF), which shows very large fluctuations only in the protonated simulation. These results do not depend on the choice of reference structure. Indeed, almost identical results are obtained by the analysis of the average contribution to global r.m.s.d. in pairwise superposition of all snapshots on each other, performed using the program MolMol (37).
Average distances and computed J-couplings were compared with the available experimental data (which were not used in MD simulations). For both simulations less than 10% of the J coupling constants were found to differ more than 2 Hz from the corresponding experimental restraints. Similarly, only 5% of interatomic distances showed violations of the upper boundaries derived from NOE larger than 2 Å. The average upper bound violation is rather limited (0.38 and 0.28 Å for the deprotonated and protonated simulation, respectively), and it is mostly contributed to by very large violations involving atoms in the most mobile regions and/or involving longer distance boundaries. It is worth noting that these violations are greatly reduced when using third power averaging.

DISCUSSION
The three-dimensional structure, obtained for the apoprotein on the basis of NMR data (Fig. 2), is typical of all the proteins of the iLBP family and consists of 10 antiparallel ␤-strands (A-J) organized in two nearly orthogonal ␤-sheets that form a ␤-clam-type structure with a gap between D and E strands. Helices I and II, inserted between A and B strands, close the protein cavity where bile acids are bound.
Protein dynamics were investigated at the following two pH values characterizing the two functional states of cl-BABP: pH 7, where the binding can take place (active conformation), and pH 5.6, where a substantial decrease of the bound ligand is observed (inactive conformation). The dynamics analysis afforded average order parameters (S 2 ), viewed over the entire protein sequence, substantially unchanged going from pH 5.6 (0.91 Ϯ 0.04) to 7.0 (0.90 Ϯ 0.04). At both pH values, the same protein segments, namely helix II, all loops, and E strand, experience fast internal perturbations (ps-ns time scale), whereas R ex contributions are observed only for residues located in the C-terminal half of the protein (Fig. 6). It is important to stress here that the conformational exchange contributions were obtained for the same protein regions applying both Lipari-Szabo and model-independent approaches (Fig.  7). Upon changing pH, a few differences in the dynamic behavior of  Residues whose amide signals were broadened beyond detection are shown in orange, and a dark gray stretch of ribbon is related to those residues that could not be included in the analysis because of resonance overlap. cl-BABP were observed, mostly located in the C-terminal half of the protein, where two histidines are present, i.e. the buried His 98 (H strand) and the more exposed His 83 (F strand). The observed protein dynamics might therefore be coupled to the exchange between their protonated and deprotonated states.
Dramatic pH-dependent variations of the R 2 rates are observed for three residues close to His 98 , namely Phe 96 , Ser 97 , and Ile 111 (supplemental Fig. 6), indicating large changes in their s-ms dynamics in the investigated pH range. These residues have large R 2 rates at pH ϳ5 where the interconversion between the protonated and deprotonated forms of His 98 takes place, as inferred from NMR titration experiments (supplemental Fig. 5) and theoretical pK a calculations. These results strongly suggest that the observed conformational exchange in cl-BABP is closely related to the protonation state of buried His 98 . Further support for this hypothesis comes from the observation that missing 1 H-15 N HSQC cross-peaks, namely Met 73 , Val 90 , Ser 93 , Lys 95 , Ser 97 , Glu 99 , and Gln 100 , are all from the same flexible C-terminal half of the protein, and most of them are close to His 98 . In addition Thr 91 and Thr 110 , close to His 98 , exhibit an R ex contribution at low pH, which was not detected at neutral pH. It is worth mentioning that His 98 , differently from His 83 , is highly conserved in the iLBP family (Fig. 1).
The role of histidine protonation on protein conformational change was further investigated by 3.6-ns MD simulations. In the presence of conformational exchange and pronounced flexibility, it is indeed true that conformational sampling provided by molecular dynamics may be inadequate for explaining NMR experimental data obtained by sampling processes on much longer time scales. Even the processes taking place on the picosecond to nanosecond time scales may not be sampled by MD simulation simply because the conformation that enables those motions is not sampled. Nevertheless, it is worthwhile examining molecular dynamics trajectories to understand possible conformational trends. In this line it is worth mentioning that two molecular dynamics simulations of bovine ␤-lactoglobulin (which belongs to the same superfamily of cl-BABP) were able to sample a pH-driven transition in even shorter simulation times (34,38).
Both simulations indicate a larger flexibility in the C-terminal half of the protein compared with the N-terminal half, in agreement with experimental data. The extent of conformational variability in the 10 NMR-derived structures is much higher than that emerging from MD simulations, but this is linked to different time scale sampling and might also be related to lack of experimental data.
The most important suggestion coming from MD simulation is that protonation of His 98 has a rather dramatic effect on interactions involving residues close to residue Asp 74 , which are able to trigger (at least in the simulation) the large conformational change involving the open end of cl-BABP (Fig. 9). This large rearrangement is consistent with the pK a shift computed for His 98 , which points out the energetic cost for the neutral pH protein environment to accommodate the titration event. Within the simulation a clear closure movement of the EF loop at the open end of the protein is observed upon protonation. This conformational rearrangement finds experimental ground in the appearance of double peaks for Asp 74 at low pH, as shown in Fig. 10. The structural basis for the EF loop opening/closure mechanism can be identified in the presence of a network of H-bonds and salt bridges involving buried residues defining a sort of continuous polar "spine" going from E to J strand (Thr 72 , Cys 80 , Ser 93 , His 98 , Glu 109 , and Arg 120 ) (Fig. 9c). Upon lowering the pH, the first residue changing its protonation state is His 98 , and the presence of this new charge could induce side chain reorientations of the cited residues, transmitting motion to the EF loop region, across the whole C-terminal ␤-sheet. This is further confirmed by 15 N chemical shift changes Ͼ0.2 ppm, observed upon lowering the pH, for residues Thr 72 , Cys 80 , and His 98 and neighbors of Ser 93 and Glu 109 pointing to a conformational change even at the level of backbone.
To correlate the dynamic data obtained for apo-cl-BABP to a biological function, the dynamic behavior of the protein complexed with a physiological ligand was investigated. Cholate, deoxycholate, and their glycoconjugated derivatives are the most abundant bile salts, as they constitute the 80% of the natural pool (7). Interaction studies of cl-BABP with chenodeoxycholate (1:2 stoichiometry ratio) indicated that protein regions mostly affected by binding are located in the C-terminal FGHIJ FIGURE 7. R ex contributions of apo-cl-BABP at pH 7.0, 298 K derived with different approaches. Results of Lipari-Szabo approach using model 3 for data fitting (circles); R ex derived as R 2 (experimental) Ϫ R 2 (fitted), where data fitting was done with model 1 of Lipari-Szabo approach (gray squares, 600 MHz; white squares, 700 MHz); R ex values deduced from data at three frequencies following the approach described in Ref. 29 (triangle). Only the upper part of the error bars is shown for clarity purposes. strands, as deduced by significant secondary 1 H and 15 N shift changes (Fig. 3). Interestingly, resonances of residues Met 73 , Val 90 , Ser 93 , Lys 95 , Glu 99 , and Gln 100 , not present in 1 H-15 N HSQC of apoprotein, appeared in the spectra of the holoprotein, thus suggesting a change in the dynamic behavior in this region of the protein.
The comparison of the dynamic behavior of apo-and holo-cl-BABP indicated that although fast motions were similarly observed in the helical regions of the protein, conformational exchange contributions, observed for apo-cl-BABP at the level of EFGHI strands, substantially disappeared upon binding (supplemental Tables 2, 3, and 6). In holo-cl-BABP, all the residues showing vanished R ex contribution map to the regions that exhibited the highest secondary 1 H and 15 N shifts (Ͼ0.5 ppm) upon chenodeoxycholate binding (Fig. 3). These results indicate the following: (i) the ligand is capable of stabilizing one conformation, and (ii) in the apoprotein exchange takes place between the active and inactive conformations, having high and low affinity for the ligand, respectively. Such a mechanism is consistent with an allosteric activation, where the histidine protonation state modulates a functionally important motion, i.e. the opening/closing of loops at the entrance of the cavity, and ligand binding shifts a pre-existing equilibrium. It has already been suggested, in a study on nitrogen regulatory protein (9), that the stabilization of pre-existing conformations may be a fundamental paradigm for ligand binding. Our model nicely parallels the results on interactions of human ILBP where the binding of glycocholate has been reported to be characterized by two intrinsically weak binding sites and strong positive cooperativity, i.e. by an allosteric mechanism where the binding of the first ligand is energetically communicated to the second site through a conformational change in the protein (7,8,39).
To investigate whether the allosteric mechanism proposed for cl-BABP could be extended to the other liver and ileal BABPs, an analysis of the conservation of residues involved in this allosteric mechanism (Thr 72 , Asp 74 , Lys 95 , His 98 , Glu 109 , Arg 120 , and Ser 122 ) was performed. From the sequence alignment ( Fig. 1), it is clear that these residues are always conserved in liver proteins, pointing to a common binding mechanism. When the comparison is extended to ILBPs, it appears that residues 74 and 95 are mutated to glycine and asparagine, respectively, whereas His 98 is conserved only in pig and rabbit species, even if a histidine is present at position 97 in the human, mouse, and rat species. Structures of both apo-and holoproteins have been resolved for an isoform of cl-BABP (T91C) 3 (PDB codes 1tvq and 1tw4) (3) and for human (PDB codes 1o1u and 1o1v) (40) and pig (PDB codes 1eal and 1eio) (41,42) ILBPs. Average global r.m.s.d. differences obtained for these apo-and holo-structures have been compared with the average global r.m.s.d. differences between our protonated and deprotonated simulations (Fig. 11). It is clear from this comparison that the EF loop (residues 72-76) is similarly affected by ligand binding or histidine deprotonation in all the structures examined, suggesting that cl-BABP and ILBPs share the same conformational switch upon binding. Interestingly, in human ILBP another loop is strongly affected by binding, namely the CD loop (52)(53)(54)(55)(56)(57) where two histidines (His 52 and His 57 ) are located. In this line an analysis of the dynamic properties of human ILBP, together with NMR titration experiments, could clarify whether a similar pH-dependent triggering mechanism is at work for liver and ILBP proteins.
The mechanism described in this paper can be regarded as an extension of the "dynamic portal hypothesis" model (6,43,44), which implies that residues in the portal region exhibit large movements enabling the opening or closing of the portal. In the present view, the event triggering this functional rearrangement is correlated with histidine protonation equilibria, and a network of polar buried side chains is likely to provide the long range connectivity needed to allosterically communicate motions from His 98 to the EF loop region. Further NMR titration and relaxation experiments, together with dynamics simulations, are in progress in our laboratory on selected cl-BABP mutants and other ileal lipid-binding proteins to study in detail the conformational switch mechanism controlling protein activity and function.
It should be noted that a pH-driven conformational change, functional to ligand binding, has been reported for other proteins belonging to the same calycin superfamily, namely ␤-lactoglobulins, where the closure of the binding cavity lid is triggered by the protonation of a highly conserved glutamic acid residue (45).
In mammals bile acid circulation from the gut lumen to the ileum is mediated by ASBT and OATP3 proteins, present on the brush borders, and by the ILBPs that bring them, through the cytoplasm, to the basolateral ileal membranes. Here a truncated ASBT secretes bile salts into portal capillaries, where they bind to albumin and flow to the liver where they are recognized by sodium-taurocholate cotransport protein and OATP. In the liver cytosol, bile salts are bound by carrier protein(s) and shuttled to the canalicular membrane. Bile salts conjugated with taurine or glycine are directed for immediate secretion into bile by an ATP-dependent transporter, BSEP, located in the canalicular membrane. The transport across this membrane is the rate-limiting step in the transfer of bile salts from blood to bile. Bile salts finally pass down the biliary ducts into the gallbladder for storage and ultimate expulsion into the 3 H. L. Monaco, personal communication.  duodenum. Over 90% of bile salts are efficiently reabsorbed from the small intestine (46). In Fig. 12, a schematic representation of the described enterohepatic circulation is given.
There is ample evidence that the liver of lower vertebrates such as chicken, frog, turtle, little skate, and rainbow trout, has evolved specific transport proteins for mediating bile salt uptake and excretion, although the molecular basis of this transport remains to be clarified. In chicken, as in mammals, chenodeoxycholic acid is the predominant primary bile acid deriving from cholesterol catabolism, followed by cholic acid, both conjugated with taurine. Interestingly, orthologues of Slc10a1 genes of Ntcp have not been found in lower vertebrates; hence, only Oatps must mediate the bile salt uptake in these species (47,48). Moreover, the functional similarity in ATP-dependent taurocholate transport between the liver membrane vesicles of rat and those of the lower vertebrates, including chicken, indicates that an evolutionary conserved protein may be mediating the mechanism for bile acid excretion into bile (49,50). Very little is known about how bile salts are shuttled from one subcellular compartment to another. Conjugated bile acids, which carry a negative charge at physiological pH, require carrier-mediated transport to move within the enterohepatic tissues. Photoaffinity labeling experiments of ileal brush border membrane vesicles (51-53) identified a 14-kDa peripheral membrane protein, the ILBP, as a component of the ileal Na ϩ /bile acid transporter system. In summary, in the bile acid enterohepatic circulation three key steps are mediated by ASBT, ILBP, and truncated ASBT in the ileum, and three are mediated by sodium-taurocholate cotransport protein, cytosolic carrier(s), and BSEP in the liver (54). Hence, similar molecules might constitute the active players in both ileocytes and hepatocytes, i.e. (i) a receptor system, which binds bile salts on one surface and translocates them into the cell; (ii) a cellular bile salt-binding protein, which moves them across the cell; and (iii) an exit system, which moves bile salts out of the other side of the cell. Our hypothesis is that cl-BABP is the cytosolic protein carrying bile salts in liver in the same way as ILBP does in ileocytes. Structure and amino acid sequences of ILBPs are very similar to those of the liver BABPs (Fig. 1). In particular, the identity of the bile acid-binding residues shown by the two kinds of proteins (3) suggests that the two families could share a common function in ileocytes and hepatocytes, respectively, related to bile acid intracellular trafficking and targeting toward membranes (2, 3).
If this hypothesis is correct, the release of bile salts by cl-BABP at acidic pH, triggered by His 98 protonation, could be thought to occur at the canalicular membrane where a pH gradient can be generated by the H ϩ -pumping ATPase operating at the level of the bile acid export pump BSEP (49,55). In several cases, protein pH-dependent conformational changes involve histidine residues that, having a pK a of 6.3 in the free form in solution, are likely to participate in structural changes around the physiological pH. A cascade of electrostatic interactions can be induced by their pH-dependent protonation/deprotonation equilibrium mechanisms. As an example, the nuclear receptor farnesoid X receptor that transcriptionally regulates production, movement, and absorption of bile acids (Fig. 12), upon binding a bile acid molecule, is activated by His 464 that acts as a molecular switch through a -cation interaction with the orthogonally oriented Trp 466 . In the case of cl-BABP, the protonation/deprotonation mechanism seems to be strongly related to the modulation of the opening and closing at the protein open end and hence to the bile acid release/uptake process.