HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

J. Biol. Chem., Vol. 277, Issue 24, 21983-21997, June 14, 2002

This Article Abstract Full Text (PDF) All Versions of this Article: 277/24/21983    most recent M201994200v1 Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Franzoni, L. Articles by Rüterjans, H. Search for Related Content PubMed PubMed Citation Articles by Franzoni, L. Articles by Rüterjans, H. Social Bookmarking                      What's this?

Structure and Backbone Dynamics of Apo- and Holo-cellular Retinol-binding Protein in Solution^*

Lorella Franzoni, Christian Lücke^§¶, Carlos Pérez^§, Davide Cavazzini^**, Martin Rademacher^§, Christian Ludwig^§, Alberto Spisni, Gian Luigi Rossi^**§§, and Heinz Rüterjans^§

From the  Department of Experimental Medicine, Section of Chemistry and Structural Biochemistry, University of Parma, 43100 Parma, Italy, ^§ Institute of Biophysical Chemistry, University of Frankfurt, 60439 Frankfurt, Germany, and the ^** Department of Biochemistry and Molecular Biology, University of Parma, 43100 Parma, Italy

Received for publication, February 28, 2002, and in revised form, March 29, 2002

	ABSTRACT

TOP ABSTRACT INTRODUCTION EXPERIMENTAL PROCEDURES RESULTS AND DISCUSSION REFERENCES

Retinoid-binding proteins play an important role in regulating transport, storage, and metabolism of vitamin A and its derivatives. The solution structure and backbone dynamics of rat cellular retinol-binding protein type I (CRBP) in the apo- and holo-form have been determined and compared using multidimensional high resolution NMR spectroscopy. The global fold of the protein is consistent with the common motif described for members of the intracellular lipid-binding protein family. The most relevant difference between the NMR structure ensembles of apo- and holoCRBP is the higher backbone disorder, in the ligand-free form, of some segments that frame the putative entrance to the ligand-binding site. These comprise -helix II, the subsequent linker to -strand B, the hairpin turn between -strands C and D, and the E-F turn. The internal backbone dynamics, obtained from ¹⁵N relaxation data (T₁, T₂, and heteronuclear nuclear Overhauser effect) at two different fields, indicate several regions with significantly higher backbone mobility in the apoprotein, including the C-D and E-F turns. Although apoCRBP contains a binding cavity more shielded than that of any other retinoid carrier, conformational flexibility in the portal region may assist retinol uptake. The stiffening of the backbone in the holoprotein guarantees the stability of the complex during retinol transport and suggests that targeted retinol release requires a transiently open state that is likely to be promoted by the acceptor or the local environment.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION EXPERIMENTAL PROCEDURES RESULTS AND DISCUSSION REFERENCES

Vitamin A derivatives play important roles in a variety of biological processes including vision, cell growth, cell differentiation, and morphogenesis (1). Plasma transport of retinol to target cells and intracellular transport for either storage or metabolic conversion are performed by binding proteins that belong to the calycin superfamily. The cytosolic carriers are members of the intracellular lipid-binding protein (i-LBP)¹ family, characterized by molecular masses of around 15 kDa. Their structure consists of a 10-stranded -barrel, formed by two orthogonal -sheets, and two short -helices (2). The two best known intracellular carriers of retinol are cellular retinol-binding protein type I (CRBP), widely distributed in various tissues (3, 4), and cellular retinol-binding protein type II (CRBP-II), present in the enterocytes of the small intestine and in neonatal hepatocytes (5, 6). The structures of rat apo- and holoCRBP-II have been solved both in the crystal (7) and in solution (8, 9), whereas the only structure of CRBP available to date was that of the holoprotein in the crystal (10). More recently, two other retinol carriers have been identified as follows: murine CRBP-III, expressed primarily in heart, muscle, and adipose tissue (11); and human CRBP-III, most abundant in liver and kidney, whose structure in the retinol-free form has been solved by x-ray crystallography (12).

In the cell, the poorly water-soluble retinol is stored within membranes as a retinyl-ester derivative of long-chain fatty acids, whose CRBP-dependent synthesis is catalyzed by the microsomal enzyme lecithin:retinol acyltransferase (13). The role of the carrier in efficient retinyl-ester synthesis has been confirmed by an in vivo study (14). For the metabolic utilization of vitamin A, apoCRBP promotes retinyl-ester hydrolysis and binds the released retinol (15). CRBP-delivered retinol is converted to retinaldehyde by an enzyme located on the cytosolic side of endoplasmic reticulum, which was first isolated from the microsomal fraction of rat liver (16, 17). Retinaldehyde can be further oxidized to retinoic acid, the biologically most active derivative of vitamin A (18, 19). The latter metabolite is specifically recognized by two other members of the i-LBP family, the cellular retinoic acid-binding proteins CRABP-I and CRABP-II (6, 20).

The three-dimensional structure of retinoid carriers accounts for their ligand specificity and affinity. CRBP forms the most stable complex with all-trans-retinol and a weaker one with retinaldehyde, whereas it does not bind retinoic acid (21). The crystal structure of holoCRBP shows that the internal binding cavity is lined by both polar and hydrophobic side chains. The bound all-trans-retinol has a planar conformation with its hydroxyl group innermost and hydrogen-bonded to the side chain of glutamine 108 (10).

Ligand uptake and targeted release, besides transport, are the fundamental functional features of hydrophobic ligand carriers. Despite the wealth of structural information, neither process is well understood as yet. Crystallographic studies on various members of the i-LBP family have evidenced the general lack of an obvious route for the ligand to enter and exit from its binding site. Nevertheless, a possible "portal" has been identified in the region located between -helix II and the two turns C-D and E-F. This hypothesis was initially based on the occurrence of a small opening on the surface of intestinal fatty acid-binding protein (I-FABP) (22). Further evidence was provided by the crystal structures of the oleate:adipocyte lipid-binding protein and the oleate:liver fatty acid-binding protein complexes, in which one end of the bound ligand is protruding into the surrounding solvent (23, 24).

The crystal structures of most i-LBPs reveal only minimal differences between the apo- and the holo-form. This feature could in principle be attributed to lattice interactions that select the same protein conformation both in the presence and in the absence of the ligand, masking differences that possibly exist in solution. However, the NMR data confirm the structural similarities and suggest that a crucial role in ligand binding is likely to be played by the higher conformational flexibility of the portal region in the apo-form. According to the "dynamic portal hypothesis," suggested by NMR studies on I-FABP (25, 26), some residues can undergo large movements that enable the opening or closing of the portal.

The structures of the retinoid-binding proteins represent a somewhat different situation, because the apo-forms exhibit varying degrees of accessibility to the ligand-binding site. The solution structure of human apoCRABP-II shows that the binding cavity is readily accessible to retinoic acid, as a result of a concerted conformational change of -helix II and the two turns C-D and E-F with respect to their position in the crystal structure of the holo-form (27, 28). Such a concerted displacement of three structural elements is unique to CRABP-II and appears to be partially suppressed by the R111M mutation, more so in solution (29) than in the crystal (30). Murine apoCRABP-I in the crystalline state is a head-to-head dimer, forming a double -barrel joined at the portal regions and thus creating a possible route for retinoic acid to enter the binding cavity (31). Although it is not known whether such a structure occurs in vivo, NMR studies on CRABP-I confirm that this protein self-associates (32). The solution structure of apoCRBP-II, when compared with the holo-form, is characterized by the presence of a small aperture and higher local flexibility, two features that could facilitate the entry of retinol (8); once bound, the ligand would induce the changes that close the entrance and stabilize the complex (9). However, CRBP-II in the crystalline state displays nearly the same structure in the apo- as in the holo-form (7).

The crystal structure of holoCRBP (10) shows that the retinol-binding site is fully shielded from the outside medium. This observation raised the question whether the structure of the ligand-free protein would indicate significant conformational differences. The present study on CRBP by multidimensional high-resolution NMR spectroscopy provides the first opportunity to compare structure and internal backbone dynamics of the protein in the apo- and the holo-form, and to verify whether it exhibits the characteristic differences that have been reported for its close homologue CRBP-II (8, 9).

	EXPERIMENTAL PROCEDURES

TOP ABSTRACT INTRODUCTION EXPERIMENTAL PROCEDURES RESULTS AND DISCUSSION REFERENCES

Protein Expression and Purification-- The rat CRBP cDNA has been cloned into a pET-11b vector and transformed into the Escherichia coli strain BL21(DE3). For expression of the unlabeled recombinant protein, bacteria were grown at 37 °C in Luria broth medium containing carbenicillin (0.1 mg/ml) until the absorbance at 600 nm reached ~0.6 OD. The expression of CRBP was induced by adding isopropyl-1-thio--D-galactopyranoside up to 1 mM final concentration. After additional 5 h of growth, the bacterial cells were harvested by centrifugation at 9,000 × g for 10 min. The same procedure was used for the expression of uniformly ¹⁵N-enriched CRBP, except that the bacteria were grown in M9 minimal medium containing ¹⁵NH₄Cl (Cambridge Isotope Laboratories, Andover, MA) as the sole nitrogen source.

The recombinant protein was purified to homogeneity as described elsewhere (33). Briefly, the supernatant obtained from the bacterial lysate after setting the ammonium sulfate concentration to 50% saturation was first subjected to an Ultrogel AcA54 gel filtration chromatography and subsequently purified to homogeneity on a QMA anion-exchange fast protein liquid chromatography column, eluted with a linear NaCl gradient (0-400 mM) in 25 mM tris-(hydroxymethyl)aminomethane hydrochloride buffer, pH 7.3. The purity of the protein was estimated by SDS-PAGE with 15% polyacrylamide gels.

Both the unlabeled and the ¹⁵N-enriched apoCRBP samples were at least 95% free of fatty acids and other potential ligands endogenous to E. coli, as shown by gas chromatography and mass spectrometric analysis (34). The protein concentration was estimated using an extinction coefficient of 28,800 M¹ cm¹ at 280 nm (35).

To obtain the holoCRBP samples, the apoprotein was saturated (5 min incubation in the dark) with 1.5 M excess of freshly prepared all-trans-retinol, delivered in dimethyl sulfoxide (maximum 1% v/v). The unbound retinol was removed by gel filtration chromatography (Econo-Pac 10DG column, packed with Bio-Gel P-6DG gel, Bio-Rad). The reconstituted holoCRBP complex exhibited an A₃₅₀/A₂₈₀ ratio of about 1.65, close to the best value reported for the natural protein from rat tissues (4).

NMR Data Collection and Processing-- For NMR measurements, 1.6 mM protein samples were prepared in potassium phosphate buffer (20 mM, pH 6.0; H₂O/D₂O = 90:10, v/v) containing 0.05% NaN₃. All NMR experiments were carried out at 25 °C on Bruker DMX spectrometers operating at ¹H resonance frequencies of 499.87 and 600.13 MHz, both equipped with triple resonance (¹H/¹³C/¹⁵N) XYZ gradient probes. Homonuclear two-dimensional spectra (TOCSY and NOESY) as well as ¹⁵N-edited multidimensional spectra were acquired; the latter included HSQC, HTQC, TOCSY-HSQC, and NOESY-HSQC. The TOCSY experiments were performed with spin-lock times of either 80 or 5.5 ms (to obtain COSY-type information). For the NOESY experiments, mixing times (_m) of 150 and 200 ms were used. The homonuclear spectra were recorded in a phase-sensitive mode with time-proportional phase incrementation of the initial pulse. Quadrature detection was used in both dimensions with the carrier placed in the center of the spectrum on the water resonance. All three-dimensional experiments made use of pulsed field gradients for coherence selection and artifact suppression, as well as gradient sensitivity enhancement schemes wherever appropriate (36, 37). Quadrature detection in the indirectly detected dimensions was achieved by either the States-TPPI (38) or the echo/antiecho (37) method. Chemical shifts were referenced to external sodium 2,2-dimethyl-2-silapentane-5-sulfonate (Cambridge Isotope Laboratories, Andover, MA) in order to ensure consistency among all spectra (39).

The ¹⁵N relaxation experiments were recorded at two different fields (500 and 600 MHz). Standard sets of sensitivity-enhanced HSQC-type pulse sequences were used to carry out the NMR experiments for the determination of longitudinal (R₁) and transverse (R₂) relaxation rates as well as the heteronuclear NOE (40-42). Series of HSQC spectra were recorded in an interleaved mode, each using 9-12 relaxation periods between 8 and 1600 ms. The spectral widths were set to 15/15 ppm in the ¹H dimension and 31/32 ppm in the ¹⁵N dimension for apo- and holoCRBP, respectively. Echo/antiecho-type gradient selection was used for phase-sensitive detection in the ₁ dimension. 512 t₁ increments were recorded in the indirect dimension, and 1024 data points were recorded in the t₂ dimension.

The spectral data were processed on a Silicon Graphics O2 work station using the Bruker XWIN-NMR 1.3 software package. A 90° phase-shifted squared sine-bell function was used for apodization in all dimensions. In the three-dimensional spectra, forward linear prediction to extend the time domain data as well as zero-filling were applied in the indirectly detected dimensions. Polynomial base-line correction was applied to the processed spectra wherever necessary. The final matrices consisted of 2048 × 2048, 2048 × 1024, or 1024 × 128 × 256 real data points for homonuclear two-dimensional, heteronuclear two-dimensional and three-dimensional spectra, respectively. Peak picking and data analysis of the transformed spectra were performed using the AURELIA 2.5.9 (Bruker) and FELIX 97 (Molecular Simulation, Inc., San Diego, CA) software packages.

For relaxation data analysis, following the measurement of individual signal intensities the relaxation rates were estimated through non-linear least squares optimization using the RMX program (Dr. J. Schmidt, National Institute for Medical Research, London, UK). The resulting relaxation rates R₁ and R₂ together with the heteronuclear NOE were analyzed by applying a Lipari-Szábo-type model-free formalism using the Modelfree 4.01 software package (43, 44). The relaxation data were fitted for each residue to several models of internal dynamics assuming isotropic molecular motion. A grid search was performed using the entire set of experimental data in order to determine the molecular correlation time _m.

Restraints Generation and Structure Calculation-- The NOE-derived distance constraints were determined from homonuclear two-dimensional NOESY and three-dimensional ¹⁵N-edited NOESY-HSQC spectra. Automated assignments of the NOEs, based only on chemical shifts, were obtained with the program nmr2st (45). An internal calibration, based on the intensities of characteristic intra- and inter-strand NOEs for residues within the -sheet structure as well as sequential and medium range NOEs for residues belonging to the -helices, was used to set the upper distance limits. The integrated peak volumes were converted into approximate inter-proton distances by normalizing against the calibrated volumes. The upper distance bounds were subsequently set to different categories of 2.7, 3.5, 4.2, 5.0, and 6.0 Å; lower bounds were taken in all cases as the sum of the van der Waals radii (1.9 Å) of the interacting protons.

The structures were calculated on a Silicon Graphics O2 work station with the program DYANA (46), which uses a simulated annealing algorithm combined with torsion angle dynamics. Starting ab initio, 100 conformers were calculated in 8000 annealing steps each. Initial structures were calculated by using only uniquely assigned distance constraints; subsequently, an iterative strategy was used for the structure refinement. In each round of structure refinement, the newly computed NMR structures were employed to assign ambiguous NOE cross-peaks, to correct erroneous assignments, and to loosen the NOE distance bounds if spectral overlap was deduced.

A retinol template, derived from the x-ray coordinates of holoCRBP (10), was introduced to the DYANA residue library. By using the standard procedure for DYANA calculations with several molecules, the ligand was attached to the protein via one PLQ and 39 LLQ pseudo-atom linkers of the DYANA library, spanning a maximal distance of ~85 Å from the C terminus of the polypeptide chain to the C-1 atom of the retinol. The retinol was kept in an all-trans conformation, whereas the single bonds of the methyl groups (C-16, C-17, C-18, C-19, and C-20) and the C-15-OH bond were allowed to rotate.

Stereospecific assignments were obtained for almost all prochiral methylene and isopropyl groups using the program GLOMSA (47). Pseudo-atom correction for magnetically equivalent protons was applied as proposed by Wüthrich et al. (48). No hydrogen bond constraints were used in the structure calculation. In the analysis of the final structures, a purely geometrical criterion for the existence of a hydrogen bond was applied (d_HO <2.3 Å, _NHO >135°).

In the last run of calculation, the structures were computed using a total of 2409 and 2826 meaningful distance restraints for apo- and holoCRBP, respectively (Table I). Subsequent energy minimization in the presence of the NOE-derived distance restraints, carried out with the DISCOVER module of the INSIGHT 97 software package (Molecular Simulation, Inc.), was performed on the 20 best DYANA conformers. The consistent valence force field (49) was used with a dielectric constant equal to r (distance in Å). A force constant of 20 kcal Ų mol¹ was applied in the NOE restraint term. The final selected structures were analyzed with PROCHECK-NMR (50).

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION EXPERIMENTAL PROCEDURES RESULTS AND DISCUSSION REFERENCES

Sequential Resonance Assignments-- The sequence-specific ¹H and ¹⁵N resonance assignments of both apo- and holoCRBP have been achieved by a combination of homonuclear and heteronuclear multidimensional NMR experiments. The sequential backbone connectivities were obtained by a combined analysis of the three-dimensional ¹⁵N-edited TOCSY-HSQC and NOESY-HSQC spectra.

Residues Glu⁷², Asp⁷³, Leu⁷⁴, Thr⁷⁵, Gly⁷⁶, Asp⁷⁹, and Arg⁸⁰ could not be identified in the ¹H/¹⁵N-HSQC spectrum of apoCRBP (Fig. 1). Because the amide protons of these residues also showed no detectable signals in any of the acquired TOCSY and NOESY spectra, the assignments of the corresponding backbone atom resonances are missing. We believe this is an indication of a high degree of conformational dispersion and/or mobility of the corresponding residues, rather than fast exchange with water. In fact, in the case of the holo-form we could obtain the complete resonance assignments for the entire protein sequence. A similar situation has been reported for other i-LBPs; very weak or missing resonance signals were described in the apo-forms, most of which became better observable upon ligand binding (25, 27).

View larger version (34K): [in this window] [in a new window]   Fig. 1.   Gradient- and sensitivity-enhanced two-dimensional 1H/15N-HSQC spectrum of uniformly 15N-enriched rat apoCRBP (20 mM phosphate buffer, pH 6.0, 25 °C) recorded at a 1H frequency of 600 MHz. Backbone and side-chain (sc) amide resonance assignments are indicated for each residue. The arginine side-chain resonances are folded. Due to spin-system heterogeneities, some of the residues show multiple peaks that are not separately labeled. Low intensity peaks that lie below the plot level are marked by squares, whereas peaks that reside outside the 1H display limits are indicated by solid lines that intersect the nitrogen axis. The side-chain amide protons of Gln108 are shifted unusually far to 3.85 and 3.66 ppm.

Other signals found to be rather weak in the ¹H/¹⁵N-HSQC spectrum of the apo-form (see Fig. 1) belong to the side chains of Asn¹³, Gln⁹⁷, Arg¹⁰⁴ (N2 group), and Trp¹⁰⁶. The side-chain amide resonances of all asparagine and glutamine residues were assigned via a combination of HTQC and ¹⁵N-edited NOESY-HSQC spectra. Arginine and lysine side-chain amides were identified in a HSQC spectrum of large spectral width in the ¹⁵N dimension (data not shown) and assigned to the corresponding residues using the TOCSY and/or NOESY information.

In the holo-form, the proton signals of all-trans-retinol were assigned without isotope labeling, based on analysis of the homonuclear two-dimensional spectra and on comparison with previously published (9, 51) chemical shift data.

Solution Structure Determination and Analysis, Description of the Global Fold-- The pattern of NOE connectivities between backbone protons was used to determine the secondary structure elements of the protein, as indicated in Figs. 2 and 3 for apoCRBP. For most of the amino acid sequence, concomitant with the absence of strong sequential d_NN NOEs and medium range correlations, strong sequential d_N NOE connectivities were observed (Fig. 2), as expected for a protein rich in -sheet structure. The inter-strand NOEs defining the repeated +1 topology of the -sheet structure (Fig. 3) evidence the presence of 10 antiparallel -strands (Fig. 3, A-J). The -strands A-D form a continuous -sheet, just as the -strands E-J. No backbone-to-backbone NOE contacts have been found between -strands D and E, but they are present between the N- and C-terminal -strands A and J, thus interconnecting the two -sheets (Fig. 3). Most -strands are sequentially connected by hairpin turns, as indicated by the NOE patterns. Furthermore, based on a series of strong sequential d_NN, weaker d_N, as well as a dense network of medium range d_NN(i,i + 2), d_N(i,i + 2), d_N(i,i + 3), d(i,i + 3) and d_N(i,i + 4) connectivities (Fig. 2), two well defined -helices have been identified between the first and the second -strand. Finally, the N terminus is extended, different from several other members of the i-LBP family where it forms a helical turn.

View larger version (18K): [in this window] [in a new window]   Fig. 2.   Sequence of rat CRBP with the sequential and medium-range NOE pattern of the apo-form. For sequential connectivities, the thickness of the bars indicates the NOE intensities; medium-range NOEs are identified by lines connecting the two coupled residues. Residues with backbone amide proton resonances not assigned are set in boxes.

View larger version (29K): [in this window] [in a new window]   Fig. 3.   Schematic representation of the CRBP -sheet structure showing the 10 antiparallel -strands (A-J) together with the inter-strand NOE connectivities identified in the apo-form. Unambiguous inter-strand NOEs are depicted by arrows with solid lines, whereas dashed lines indicate expected NOEs that are ambiguous due to chemical shift degeneracies, spectral overlap, or proximity to the water. Amide groups with missing resonance assignments are set in boxes.

On the basis of 2409 and 2826 NOE-derived non-redundant distance restraints obtained for apo- and holoCRBP, respectively (Table I), 100 structures were generated in both cases as described under "Experimental Procedures." It should be noted that the structure of CRBP complexed with all-trans-retinol was not obtained by docking the ligand molecule into the protein structure. Rather, the structure calculations were performed entirely by using the torsion angle dynamics program DYANA (46), with the retinol molecule built into the program library as an additional residue. Thus, intra-ligand, intra-protein, and ligand-to-protein distance constraints were applied simultaneously by the algorithm to determine the global structure of the protein complex ab initio. For both apo- and holo-form, the 20 best structures with the lowest target function values (all below 0.3 Å²) were subsequently refined by restrained energy minimization.

The global fold of CRBP is similar to the one typically observed for most of the i-LBPs. It consists of a flattened -barrel formed by two nearly orthogonal -sheets; one end of the -barrel is blocked by the N terminus of the protein, and the other end is covered by a helix-turn-helix domain (Fig. 4). The proposed entry site of retinol is located between -helix II, the subsequent linker to -strand B, the C-D turn, and the E-F turn (2). Both -helices are amphiphilic, as the hydrophobic residues interact with the ligand bound inside the protein and the charged residues point outwards into the solvent. A salt bridge (Glu¹⁷-Lys³¹) might help to stabilize the turn between the two -helices (10). As in the case of many other cellular proteins, disulfide bridges are absent.

View larger version (62K): [in this window] [in a new window]   Fig. 4.   Ribbon diagram representing the solution structure of apoCRBP. The antiparallel -strands, labeled A-J, form a -barrel structure that defines the cavity in the protein interior. The proposed entry site of retinol resides between -helix II, the subsequent linker to -strand B, the C-D turn, and the E-F turn (2). The latter turn, which includes 7 residues with some missing resonance assignments that result in a high backbone dispersion, is displayed in an average conformation out of the 20 conformers. (The figure was prepared using MOLSCRIPT (73) and Raster3D (74).)

There is a G₁ bulge at the highly conserved Gly⁶⁷; this bulge separates -strands D and E, thus forming a gap in the -sheet structure. However, this does not provide a ligand entrance to the interior of the protein, because the side chains of several adjacent residues (i.e. Tyr⁶⁰, Met⁶², Phe⁶⁴, Phe⁷⁰, Glu⁷², and Thr⁷⁵) fill the gap, thereby blocking the access to the cavity. The function of this gap is still obscure, but it is a characteristic feature of the i-LBP family that has been postulated as a second portal for the aqueous solvent (52, 53).

Fig. 5 shows the two final ensembles, 20 energy-minimized conformers each, that have been selected to represent the solution structures of CRBP in the ligand-free form and complexed with all-trans-retinol. The structural statistics information is summarized in Table I. It shows no significant violations of single distance restraints, thus indicating a good agreement between the experimentally obtained NOE data and the calculated conformers.

View larger version (52K): [in this window] [in a new window]   Fig. 5.   Stereo representation of the solution structure ensembles (as C traces) of apoCRBP (top panel) and CRBP complexed with all-trans-retinol (bottom panel). The structural statistics of these structures are given in Table I. The apoCRBP ensemble is characterized by increased backbone disorder, in particular in the domains framing the portal region, such as the E-F turn. The structure of holoCRBP shows the retinol (in gray) bound inside the protein cavity.

The structural quality of the energy-minimized models was examined with the PROCHECK-NMR program (50). Analysis of the backbone dihedral angles  and  showed that 98.9% of all non-glycine/non-proline residues in apoCRBP fall within the allowed regions of conformational space (Table I). Considering that this analysis includes the poorly defined E-F turn, the result is quite satisfactory. In the holoprotein, on the other hand, the allowed regions are populated by 99.9% of the residues. Moreover, the backbone dihedral angles of those residues located within the main secondary structure elements show well defined values with low standard deviations for both protein forms.

Comparison of the Apo- and HoloCRBP Solution Structures-- The structure of CRBP complexed with all-trans-retinol in solution is highly similar to the ligand-free protein. The 10 antiparallel -strands and the 2 -helices are almost identical in terms of both length and sequence location for both protein forms. The only apparent difference occurs in -strands E and F. In the apo-form, which is incompletely assigned for residues 72-80 as mentioned previously, these two -strands range from Lys⁶⁸ to Glu⁷¹ and Cys⁸² to Asp⁸⁹, whereas in the holoprotein they are slightly longer, spanning the segments Lys⁶⁸ to Asp⁷³ and Arg⁸⁰ to Asp⁸⁹, respectively.

The tertiary structures of apo- and holoCRBP are almost superimposable; the average r.m.s.d. between the two ensembles is 1.04 ± 0.16 Å for the backbone atoms and 1.69 ± 0.14 Å for all heavy atoms (Table I). Excluding the E-F turn, however, the average r.m.s.d. values (residues 1-71 and 81-134) drop down to 0.73 ± 0.09 and 1.34 ± 0.09 Å for the backbone and heavy atoms, respectively, thus evidencing the close structural similarity between the two protein forms. The regions showing more extensive backbone variations between holoCRBP and the ligand-free protein, with local r.m.s.d. values higher than average (i.e. >0.73 Å), involve the N terminus (residues 0-2), the loop between the two -helices (residues 23-26), the entire -helix II, the subsequent linker to B (residues 36 and 37), and almost all the turns connecting the -strands, namely B-C, C-D, E-F, F-G, and G-H. Finally, the size of the gap between D and E is the same in apo- and holoCRBP ensembles, when measuring the C-C distance between residues Phe⁶⁴ and Phe⁷⁰. These two residues are located in the lower part of the gap with an average distance of ~7.4 Å. However, measuring the C-C distance between residues Arg⁵⁸ and Gly⁷⁶ located in the upper part of the gap, it appears more narrow in holoCRBP by more than 1 Å, with average distances of 8.5 Å (apo) versus 7.2 Å (holo). But the latter comparison is certainly biased by the high conformational dispersion of the E-F turn in apoCRBP as a result of its missing resonance assignments.

Molecular dynamics calculations and essential dynamics analyses, using the crystal structure of the holoprotein as starting structure, led to the prediction that removal of retinol from the protein would cause -helix II of the helix-turn-helix domain to bend away from the rest of the protein and the C-D and E-F turns to undergo large conformational changes that widen the cleft between -strands D and E, with Gly⁶⁷ serving as a sort of hinge (54). Neither of these predictions is confirmed by the current study. Whereas a role for the highly conserved Gly⁶⁷ is supported by an experiment showing that its replacement with a bulkier, less flexible residue abolishes retinol binding (55), it has recently been argued, in the case of I-FABP, that the two neighboring -strands D and E are likely to bend outwards during a molecular dynamics simulation just because they lack the typical -sheet stabilization due to hydrogen bonds (53).

The most striking difference between the two families of conformers (apo versus holo, Fig. 5) is the presence of discrete regions, especially around the entry portal, that display a higher backbone disorder in the ligand-free form, as indicated by increased local r.m.s.d. values. In general, this higher backbone dispersion in apoCRBP corresponds to residues that show fewer distance restraints, because of either conformational dispersion/mobility or lack of assignments. In the case of the E-F turn, the poor definition of the backbone certainly originates from a combination of missing resonance assignments and subsequent lack of NOEs, which can in turn be due to local backbone disorder and/or mobility. Similar observations of localized disorder have been reported for several other members of the i-LBP family (8, 25, 27, 56, 57) suggesting a higher degree of molecular flexibility at the portal.

The backbone conformation of holoCRBP with an r.m.s.d. of 0.49 ± 0.07 Å is significantly better defined than the apo-form, which has an r.m.s.d. of 0.95 ± 0.16 Å. This difference in structure convergence is much less pronounced when the E-F turn region is excluded, with backbone r.m.s.d. values of 0.47 ± 0.06 and 0.70 ± 0.10 Å for holo- and apoCRBP, respectively. Of course, the presence of over 400 additional meaningful distance restraints (see Table I), including 117 intermolecular protein-to-retinol distance restraints inside the protein, causes the holoCRBP structure to be much better defined than apoCRBP.

In the holoprotein, the -ionone ring of retinol is located in a hydrophobic niche defined by the side chains of Leu²⁹, Ala³³, Leu³⁶, Phe⁵⁷, Arg⁵⁸, and Ile⁷⁷, which belong to -helix II and to the hairpin turns B-C and E-F. Even though these side chains are less defined in apoCRBP (as seen in Fig. 6, top panel), they all adopt comparable conformations in both protein forms, with the exception of residues Arg⁵⁸ and Ile⁷⁷. In several conformers of the apo-form the mutual disposition of the two latter side chains leaves a small aperture, but a space large enough for retinol to enter the binding site could not be detected in any of the 20 final conformers.

View larger version (31K): [in this window] [in a new window]   Fig. 6.   Comparison of the side-chain orientations of residues either framing the proposed ligand entry portal (top panels) or present inside the ligand-binding cavity (bottom panels) in the solution structures of apoCRBP (left) and holoCRBP (right). The side chains lining the ligand-binding site (bottom panels) are well defined in both protein forms, whereas an increased conformational dispersion is evident in the portal region of the apo-form (top left panel), in particular for residues Arg58 and Ile77. In the top right panel, retinol is displayed in only one conformer.

The side-chain orientations of the residues lining the ligand-binding cavity are well defined in both families of structures (Fig. 6, bottom panel), resulting in relatively small circular variances of the dihedral angles ₁ and ₂. In most cases, the circular variances of the ₁ angles are lower than 0.01°. Residues Lys⁴⁰, Thr⁵³, Arg¹⁰⁴, Trp¹⁰⁶, Gln¹⁰⁸, Gln¹²⁸, and Phe¹³⁰, among others, retain basically the same orientation (see Fig. 6; Gln¹²⁸ and Phe¹³⁰ are omitted for clarity). As suggested by the crystal structure analysis of holoCRBP (10), in most of the holo-form NMR conformers (15 of 20) the hydroxyl group of retinol acts as a hydrogen donor to the carbonyl group of the Gln¹⁰⁸ side chain. At the same time, the Gln¹⁰⁸ NH₂ group is located close to the center of the Phe⁴ ring, probably forming an aromatic hydrogen bond (except in two of the conformers). In the apoCRBP conformers, on the other hand, both the side-chain amide group of Gln¹⁰⁸ and the Phe⁴ ring seem to have a higher degree of freedom and do not converge to a single orientation in the structure calculations. However, the highly unusual and nearly identical chemical shift values of the Gln¹⁰⁸ H21/H22 amide protons (3.85/3.66 ppm and 3.85/3.69 ppm for apo- and holo-form, respectively) can only be explained by a very close proximity of the Phe⁴ ring to the Gln¹⁰⁸ side-chain amide group in both protein forms.

Spin-system Heterogeneities-- A total of 33 residues showed multiple resonances for the backbone and/or side-chain atoms in apoCRBP. Although visible in the ¹H/¹⁵N-HSQC spectrum, these peaks have not been labeled separately in Fig. 1. They were also detected in the homonuclear two-dimensional and heteronuclear three-dimensional spectra. Table II lists all residues that show more than one set of resonances. The presence of such spin-system heterogeneities has been reported for other members of the i-LBP family as well (8, 57-59).

According to mass spectrometric analysis, the additional N-terminal methionine, Met⁰, is retained in 80% of the recombinant protein. Thus, chemical micro-heterogeneity in the NMR sample (60) may be the origin of the multiple sets of resonances observed for three residues at the N terminus (Val²-Asp³-Phe⁴). This could furthermore explain the triple spin-systems observed for other residues (Gly⁴⁶, Asp⁴⁷, Lys⁹², and Leu⁹³) located in the B-C and F-G turns, which are both spatially close to the N-terminal region (Fig. 4). Most of the other residues listed in Table II reside in regions with large local backbone r.m.s.d. (Fig. 5, top panel), which are expected to undergo conformational changes in solution. More precisely, these regions include the start and the end of -strands, the turns connecting -strands, as well as -helix II. Only six of the residues are located elsewhere, with their side chains exposed to the bulk solvent.

In the holoprotein, multiple sets of resonances were observed for only 8 residues, all of which are situated at the N terminus or in the B-C and F-G turns (Table II). The data would therefore support the hypothesis that, although the multiple spin-systems observed around the N-terminal region originate from chemical micro-heterogeneity in the samples (60), the additional heterogeneities found in apoCRBP may arise from a higher local conformational flexibility.

Because the differences in chemical shift values between spin-systems of the same residue are rather small, the conformational deviations are assumed to be subtle. However, it should be pointed out that the structural calculations were based only on distance restraints derived from the primary spin-systems that represent the major and most populated conformational state.

Backbone Dynamics of CRBP-- To determine whether the lower local disorder observed in the NMR structures of holoCRBP is due to a change in molecular flexibility, the internal backbone dynamics were studied for both protein forms by measuring ¹⁵N relaxation rates (R₁ and R₂) and steady-state heteronuclear ¹⁵N{¹H} NOE at 500 and 600 MHz. In the apo-form, the protein samples displayed a considerably reduced stability, indicated by the appearance (after several days of measurement at room temperature) of additional amide signals belonging to a denatured form of the protein. These distinct protein stabilities are also reflected by the higher experimental error values obtained for apoCRBP with respect to holoCRBP. Moreover, the difference in their molecular correlation times (_m = 8.0 ns (apo) and 8.5 ns (holo)) may be attributed to the respective sample conditions as well.

Based on the complete set of experimental relaxation data, the microdynamic parameters (S², _e, and R_ex) have been subsequently obtained for both apo- and holoCRBP. Viewed over the entire protein sequence, the average order parameter (S²) values are 0.840 ± 0.062 and 0.869 ± 0.041 for the apo- and holo-form, respectively. However, a comparison of the S² values for each residue, as shown in Fig. 7, indicates a higher backbone mobility on the nanosecond-to-picosecond time scale in several segments of the ligand-free protein with respect to the retinol-bound form.

View larger version (17K): [in this window] [in a new window]   Fig. 7.   Line plot showing the order parameter (S2) values obtained for apo- (gray squares) and holoCRBP (black triangles). ApoCRBP displays on average lower order parameter values than holoCRBP, indicating an overall higher internal backbone mobility. Especially pronounced are the decreased S2 values of residues Asn59, Ile77, and Asp78 in the portal region of the apo-form. Error bars indicate S.D.

The above-mentioned missing resonance assignments in the E-F turn already indicated an abnormal dynamic behavior in that part of the portal region for apoCRBP, especially because all resonances could be assigned upon ligand binding. The relaxation data analysis of the E-F turn and of the adjacent C-D turn show significantly decreased S² values for Asn⁵⁹ (0.684), Ile⁷⁷ (0.683), and Asp⁷⁸ (0.733) in the apo-form, relative to values of S² >0.85 obtained for the same residues in the holo-form. Hence, all these data suggest that upon retinol binding there is a stiffening of the CRBP structure around the portal region, thus stabilizing the complex.

Other regions that display a higher backbone mobility in apoCRBP compared with the holo-form include the segments Glu¹⁷-Asp²⁴ (-helix I), Lys⁴⁰-Glu⁴¹ (-strand B), Asp⁴⁵-His⁴⁸ (turn B-C), Trp⁸⁸-Asp⁹¹ (turn F-G), Trp¹⁰⁶ (-strand H), and Leu¹¹⁷-Ala¹²¹ (-strand I).

Differently from other members of the family (8, 26, 61), apoCRBP contains an -helix I that appears to be more mobile, as a whole, than -helix II (Fig. 7). This behavior is quite unusual; in fact, whereas -helix II shows a certain degree of variability in the amino acid sequences of the various i-LBPs, -helix I appears to be evolutionarily well conserved and also better defined within the NMR structure ensembles. Following retinol binding, there is a stiffening of the backbone of -helix I, with no change in the degree of mobility of -helix II. This feature, which is unique among retinoid carriers, suggests a possible role for -helix I in retinol exchange. However, any model explicitly involving this segment must await further experimental evidence.

As for residues with side chains located inside the ligand binding cavity, a change in backbone mobility may, for example, be deduced for Trp¹⁰⁶, which shows a significantly increased S² value in the holo-form (0.899) with respect to the apo-form (0.773).

Overall, the relaxation data provided additional evidence that the retinol binding process produces a certain increase in local backbone stability that could also justify the longer lifetime of the holoCRBP samples. A similar behavior has been reported previously (56) for another i-LBP, the porcine ileal lipid-binding protein.

Comparison between the Solution and Crystal Structures of HoloCRBP-- The solution structure of holoCRBP is essentially identical to the 2.1-Å resolution x-ray structure (Protein Data Bank code 1CRB) (10). Superposition of all 20 NMR conformers to the crystal structure gave average r.m.s.d. values of 0.61 ± 0.05 and 1.12 ± 0.04 Å for the backbone and heavy atoms, respectively (Table III). Protein regions characterized by above average local backbone r.m.s.d. values consist of mostly turns as well as residues 59-61 at the start of -strand D. Different side-chain orientations (heavy atom r.m.s.d.  1.50 Å) are found only for charged residues that generally appear disordered at the protein surface in the solution structure: Arg²¹, Asp²⁴, Lys³⁷, Arg⁵⁸, Lys⁶⁸, Glu⁷¹, Lys⁸¹, Asp⁹¹, Glu¹⁰⁰, Glu¹⁰², Arg¹²⁰, and Lys¹²⁷. We have recently found a similar behavior in another -barrel protein belonging to the lipocalin protein family (62).

Retinol-to-protein NOEs were observed for residues Phe¹⁶, Tyr¹⁹, Leu²⁰, Leu²³, Val²⁵, Leu²⁹, Ile³², Ala³³, Leu³⁶, Pro³⁸, Lys⁴⁰, Ile⁵¹, Thr⁵³, Ser⁵⁵, Phe⁵⁷, Arg⁵⁸, Asn⁵⁹, Tyr⁶⁰, Met⁶², Ile⁷⁷, Trp¹⁰⁶, Leu¹¹⁷, and Met¹¹⁹. With the exception of Leu²³ and Ile³², both of which show only a single NOE connectivity to retinol, all of the above coincide with those residues that change their accessible surface area by more than 1 Å² upon removal of the ligand from the crystal structure model (10). The side chains of these residues present the same orientation in solution and in the crystal, with heavy atom r.m.s.d. ranging from values as low as 0.32 (Ile⁵¹ and Leu¹¹⁷) to 0.99 Å (Phe¹⁶ and Leu²⁹). Only four residues at the portal region show r.m.s.d. values higher than 1.0 Å, namely Phe⁵⁷, Arg⁵⁸ (both in the C-D turn), Asn⁵⁹, and Tyr⁶⁰ (both at the start of -strand D). In the case of the latter two residues, the side-chain displacement is a consequence of the previously mentioned altered backbone, whereas for Phe⁵⁷ and Arg⁵⁸ only the side chains are reoriented, especially in the case of Arg⁵⁸ with a heavy atom r.m.s.d. of 2.31 Å (Fig. 8, top panel).

View larger version (27K): [in this window] [in a new window]   Fig. 8.   Superposition of the crystal and solution structures of holoCRBP, showing residues that either frame the proposed ligand entry portal (top panel) or line the ligand-binding site (bottom panel). The 20 conformers of the solution structure display good alignment with the crystal structure (colored red, Protein Data Bank code 1CRB) (10). In the top panel retinol is displayed in only one conformer of the solution structure.

The position and conformation of the bound retinol in solution was determined based on a total of 117 intermolecular ligand-protein restraints (Table I) and by inspection of the intra-ligand NOE pattern. The ligand assumes a conformation, where its -ionone ring is approximately planar relative to the fully extended polyisoprene chain. Its orientation inside the binding cavity is very close to that found in the crystal structure (Fig. 8, bottom panel).

Of particular interest is also the position of the side-chain amino group of Lys⁴⁰, a residue conserved in all CRBP types. The observation of the NH resonances in the NMR spectra (33.12 and 6.98 ppm for N and H, respectively) is highly unusual, and more so because the ¹⁵N chemical shift value suggests that the amino group is positively charged, despite the lack of a compensating salt link. Presumably, the exchange of these amino protons is significantly reduced either because of its solvent-inaccessible location or because of hydrogen bond interactions with the Thr⁵³ O and/or the conjugated -electron field of the retinol polyisoprene tail. The latter interaction was also suggested by the crystal structure (10), and it could be considered to mimic an aromatic hydrogen bond where the polyisoprene moiety acts as -acceptor (63), thus providing additional stability to ligand binding. In apoCRBP, where the interaction between Lys⁴⁰ and retinol is missing, the NH resonances of Lys⁴⁰ are still observable at 31.04 and 7.39 ppm for N and H, respectively, indicating a charged amino group with slow-exchanging protons, even in the absence of the ligand. Because the local protein structure is nearly identical in both CRBP forms, the slow exchange of these amino protons in the absence of ligand may be attributed either to a hydrogen bond interaction with Thr⁵³ O or to a rather stable local micro-environment consisting of a network of well ordered water molecules inside the cavity. The latter setting has been reported previously (7, 28, 31, 53, 64, 65) for the apo-forms of other i-LBP family members.

Comparison of apoCRBP with Its Close Homologue Rat ApoCRBP-II-- CRBP and CRBP-II, the first intracellular retinol-binding proteins to be discovered (3-5), show 56% sequence identity. Both proteins interact with all-trans-retinol and all-trans-retinal, with the binding affinity of CRBP toward retinol reported to be about 100-fold higher than that of CRBP-II (6, 66). Because the solution structure of CRBP-II in the ligand-free state shows rather peculiar differences in the portal region with respect to the crystal structure (7, 8), we have compared the structure of apoCRBP with both the solution and crystal structures of rat apoCRBP-II.

Remarkably, the structural features of apoCRBP in solution are more similar to the ones observed in the crystal structure of apoCRBP-II (Protein Data Bank code 1OPA) (7) than in the corresponding solution structure (Protein Data Bank code 1B4M) (8). This is shown in Fig. 9 and confirmed by the backbone atom r.m.s.d. values reported in Table III. Comparing the two proteins in solution, the differences in the backbone conformation reside mainly in the regions framing the portal. In particular, the solution structure of apoCRBP-II lacks -helix II (segment 27-34)² (8), which is present in both its crystal structure (7) and the solution structure of apoCRBP (residues 27-35). Other domains of significant displacement in the apoCRBP-II solution structure are the N terminus, residues 35-39, the C-D turn, and the E-F turn.

View larger version (23K): [in this window] [in a new window]   Fig. 9.   Superposition of the solution structure of apoCRBP (one model colored blue) with the following structures: crystal structure of rat apoCRBP-II (molecules A and B colored dark and light green, respectively, Protein Data Bank code 1OPA) (7); solution structure of rat apoCRBP-II (best model colored yellow, Protein Data Bank code 1B4M) (8); crystal structure of human apoCRBP-III (molecules A and B colored dark and light magenta, respectively, Protein Data Bank code 1GGL) (12). The side-chain orientations of those residues, which either line the putative entrance to the binding site (top) or are in close proximity to the bound retinol (bottom), are shown for all structures. The retinol ligand (colored red) is represented as ball-and-stick model the way it is bound to holoCRBP.

The present work thus provides a negative answer to the question whether CRBP exhibits similar structural differences between the apo- and holo-form as reported for CRBP-II (9). This implies that the unwinding of -helix II is not a general feature in the apo-forms of cellular retinol carriers.

Aside from the unwinding of -helix II relative to the crystal structure, the Phe⁵⁷ side chain in the solution structure of apoCRBP-II is oriented in such a way that it interacts with Leu³⁶ and does not block the entrance to the ligand-binding site (Fig. 9, top panel) (7, 8). In the case of apoCRBP, on the other hand, the phenyl ring of Phe⁵⁷ retains an orientation very similar to the one observed in the holo-form, thus blocking the access to the binding cavity (Fig. 6, top panels). It should be noted that the position of the Phe⁵⁷ side chain in the solution structures of CRBP was affirmed by uniquely assigned NOE connectivities between the  and  protons of the phenyl ring and the side-chain protons of residues Leu²⁹ (H1/H2) and Ile³² (H1 and H2).

The orientation of the phenyl ring of Phe⁵⁷, a highly conserved residue in the i-LBP family, is a question of interest because it has been implicated in the control of the access to the binding cavity, either by directly blocking the portal (2, 22) or through stabilizing "capping interactions" with -helix II (25). Structural comparison of the various retinol carriers (see Refs. 7-10, 12, and current work) shows that it is not possible to identify a single orientation of Phe⁵⁷ that is unique to either the apo- or the holo-form. We think that Phe⁵⁷ has a flexible side chain whose movement toward Leu³⁶ may be a prerequisite to provide the ligand access to the binding cavity. It must be noted that a concerted movement of both Phe⁵⁷, to reach the position occupied in the solution structure of apoCRBP-II, and Arg⁵⁸, to assume the orientation detected in a few models of apoCRBP, would enlarge the aperture in the portal region. Such side-chain movements, together with moderate backbone structure alterations in the C-D and E-F turns, would probably be sufficient to allow retinol to enter the binding cavity.

The orientations of side chains that are spatially close to the bound retinol in holoCRBP and holoCRBP-II (9) have been compared in the corresponding apo-structures. Again, differences were found between the apoCRBP and apoCRBP-II structures in solution for residues Lys⁴⁰, Thr⁵³, Trp¹⁰⁶, and Gln¹⁰⁸, whereas there is a good agreement in the orientation of these residues between apoCRBP in solution and apoCRBP-II in the crystal (Fig. 9, bottom panel). The hydrogen bond interaction of the Lys⁴⁰ NH group with Thr⁵³ O is missing in the solution structure of apoCRBP-II, where these groups are placed much farther apart. Moreover, in the apoCRBP-II solution structure, different positions were also observed for the amide group of Gln¹⁰⁸, which forms a hydrogen bond with the hydroxyl group of retinol in both holoCRBP and holoCRBP-II, and for the neighboring Trp¹⁰⁶ indole ring, which is significantly rotated.

Comparison of ApoCRBP with the Crystal Structure of Human ApoCRBP-III-- The backbone atom superposition of our 20 conformers with human CRBP-III (56% sequence identity; Protein Data Bank code 1GGL) (12) shows that the structures of the two apoproteins are very similar (Table III), especially in the domains characterized by regular secondary structure. The main differences in backbone conformation are observed at the N terminus and in the loops connecting the -strands. Probably because of the presence of two consecutive proline residues, Pro¹ and Pro², the four N-terminal residues in CRBP-III form a sort of turn (only in chain A) that causes the segment to move farther away from the end of -strand B and from the F-G turn. In apoCRBP, on the other hand, the ring of Pro¹ stacks on the ring of Trp⁸⁸ (located near the F-G turn). Helix II is fully retained in retinol-free CRBP-III, but it is displaced with respect to its position in apoCRBP. The deviations between the corresponding C positions of residues 27-35 range from 0.72 to 1.89 and 1.37 to 1.98 Å for chains A and B, respectively.

A comparison of the residues in apoCRBP, which either are in close proximity to the bound retinol (Lys⁴⁰, Thr⁵³, Arg¹⁰⁴, Trp¹⁰⁶, and Gln¹⁰⁸) or line the entrance to the binding cavity (Leu²⁹, Ala³³, Leu³⁶, Phe⁵⁷, Arg⁵⁸, and Ile⁷⁷), with the corresponding residues in apoCRBP-III (Lys⁴⁰, Thr⁵³, Arg¹⁰⁴, Trp¹⁰⁶, His¹⁰⁸, Val²⁹, Ala³³, Leu³⁶, Phe⁵⁷, Arg⁵⁸, and Val⁷⁷, respectively) shows that their positions are nearly the same (Fig. 9). As in the other CRBP structures, the phenyl ring of Phe⁵⁷ blocks the entrance to the ligand-binding cavity of CRBP-III. However, a difference is found for the side chain of Arg⁵⁸, whose orientation in apoCRBP-III is very similar to that observed in the crystal structure of holoCRBP but slightly shifted in the apoCRBP solution structure (Fig. 9, top panel).

In both murine and human CRBP-III, Gln¹⁰⁸ is replaced by a histidine (11, 12). The nature of the residue at position 108 is critical in determining the ligand-binding properties of i-LBPs. In the case of both CRBP and CRBP-II, the Q108R mutation results in a decreased affinity for all-trans-retinol and, at the same time, a measurable affinity for retinoic acid (67, 68). Conversely, the replacement of the corresponding Arg¹⁰⁶ in I-FABP with glutamine increases the binding affinity of the FABP mutant for all-trans-retinol (69). The Gln¹⁰⁸/His¹⁰⁸ substitution in CRBP-III may have yet another functional significance. The side chains of Gln¹⁰⁸ in apoCRBP and His¹⁰⁸ in human apoCRBP-III are arranged similarly in the ligand-binding site (Fig. 9, bottom panel). Thus, the histidine N 2 can form a hydrogen bond to the retinol molecule in a manner similar to, but possibly less effective than, the interaction between retinol and glutamine 108 in CRBP (11). Any change of the ionization state of His¹⁰⁸ might affect retinol binding to CRBP-III (12). However, the fact that CRBP-III carries a leucine, instead of a phenylalanine, at position 4 excludes the possibility of an aromatic hydrogen bond formation with the side chain of residue 108, a bond that might additionally stabilize retinol binding in holoCRBP.

Retinol Uptake and Release by CRBP-- Although the apoCRBP binding cavity is more shielded than that of any other retinoid carrier in solution, binding of retinol occurs very efficiently both in vivo and in vitro. How does the cavity become available to retinol? In vivo, ligand uptake by the apoprotein occurs in proximity to a membrane, following either retinol internalization or hydrolysis of the stored retinyl-esters. It is known that both local pH and dielectric constant in proximity to a biomembrane are lower than in the bulk solution. Such a micro-environment might assist a transient opening of the more flexible structure elements in the portal region. In vitro, holoCRBP is readily obtained either by mixing the apoprotein with retinol dissolved in a minimal amount of an organic solvent or by mixing apoCRBP with holoCRBP-II (21). In the first instance, the organic solvent might play a direct role in the opening of the portal and the transfer of retinol into the protein, by generating a suitable interface that alters the local micro-environment. In the second instance, a transient protein-protein interaction, possibly similar to the dimeric structure observed in the crystal of apoCRABP-I (31), might create a route for retinol exchange; alternatively, retinol bound to CRBP-II might dissociate and diffuse through the aqueous medium, as reported for its transfer from this carrier to small unilamellar vesicles (70), and interact with CRBP, inducing itself the local changes that are necessary to permit its access to the binding cavity.

Even more than retinol uptake, retinol dissociation from CRBP seems to require the