The NMR structure of a stable and compact all-β-sheet variant of intestinal fatty acid-binding protein

Abstract

Intestinal fatty acid-binding protein (I-FABP) has a clam-shaped structure that may serve as a scaffold for the design of artificial enzymes and drug carriers. In an attempt to optimize the scaffold for increased access to the interior-binding cavity, several helix-less variants of I-FABP have been engineered. The solution-state NMR structure of the first generation helix-less variant, known as Δ17-SG, revealed a larger-than-expected and structurally ill-defined loop flanking the deletion site. We hypothesized that the presence of this loop, on balance, was energetically unfavorable for the stability of the protein. The structure exhibited no favorable pairwise or nonpolar interactions in the loop that could offset the loss of configurational entropy associated with the folding of this region of the protein. As an attempt to generate a more stable protein, we engineered a second-generation helix-less variant of I-FABP (Δ27-GG) by deleting 27 contiguous residues of the wild-type protein and replacing them with a G-G linker. The deletion site of this variant (D9 through N35) includes the 10 residues spanning the unstructured loop of Δ17-SG. Chemical denaturation experiments using steady-state fluorescence spectroscopy showed that the second-generation helix-less variant is energetically more stable than Δ17-SG. The three-dimensional structure of apo-Δ27-GG was solved using triple-resonance NMR spectroscopy along with the structure calculation and refinement protocols contained in the program package ARIA/CNS. In spite of the deletion of 27 residues, the structure assumes a compact all-β-sheet fold with no unstructured loops and open access to the interior cavity.

Despite the considerable diversity in their amino acid sequence, all members of the intracellular lipid-binding protein family adopt a β-clam fold comprised of two five-stranded anti-parallel β-sheets and a helix-turn-helix motif (Sacchettini and Gordon 1993; Banaszak et al. 1994). The high degree of topological conservation of the helical domain within the family of lipid-binding proteins may imply important structural and/or functional roles. Prior to this work, a helix-less variant of I-FABP was engineered in order to investigate the roles of the helical domain. This protein, called Δ17-SG, was engineered by deleting 17 residues that span the helical domain and replacing them by a Ser–Gly linker (Cistola et al. 1996; Kim et al. 1996; Steele et al. 1998).

Comparative structural and ligand-binding studies of Δ17-SG and the wild-type protein showed that, notwithstanding the changes in protein stability, the helical domain is not a required element of the overall topology of I-FABP (Kim et al. 1996; Steele et al. 1998). However, ligand-binding kinetics results revealed that the helical domain functions by regulating the ligand entry into and release from the binding cavity (Cistola et al. 1996). Investigation of the determinants of fatty-acid transfer from fatty-acid-binding proteins to membranes also showed that the surface charges of the α-helical domain influences the mode of fatty acid transfer (Kim and Storch 1992; Hsu and Storch 1996; Corsico et al. 1998). These results have implications on the involvement of the α-helical domain of I-FABP in mediating fatty acid transfer inside the cell.

One unexpected feature of the tertiary structure of Δ17-SG was the presence of a long loop flanking the deletion site (Fig. 1A ▶). The wild-type I-FABP structure (Fig. 1B ▶) revealed that strand A kinks and makes hydrogen bonds with both β-strands B and J. The first four residues of strand A make hydrogen bonds with strand B while the residues in the distal half of strand A are engaged in hydrogen bonds with strand J. However, in Δ17-SG, the pairwise interactions between β-strands A and J are missing, and the residues that were in the distal half of strand A form the proximal part of the loop between strands A and B. This 10-residue loop, designated by the arrow in Figure 1A ▶, was longer than anticipated and replaces the helix-turn-helix domain of the wild-type protein. The amide proton resonances corresponding to these 10 residues were selectively missing in gradient- and sensitivity-enhanced two-dimensional ¹⁵N-HSQC spectra of Δ17-SG. No structural or dynamical constraints could be detected for the loop, suggesting a lack of structure in this region of the molecule. We hypothesized that the presence of this lengthy loop in Δ17-SG contributes unfavorably to its global stability. To first-order approximation, the loss of configurational entropy associated with the folding of this region of the protein was not compensated by favorable pairwise or nonpolar interactions in this loop. Moreover, in previous studies concerning loop-entropy, the entropic cost associated with the closure of loops was proven to be higher for longer loops than shorter ones (Nagi and Regan 1997). Therefore, as suggested earlier by Steele et al. (1998), further deletion of the residues spanning the floppy loop in Δ17-SG might result in a more compact and stable helix-less I-FABP. We engineered a second-generation helix-less variant of I-FABP, called Δ27-GG, by deleting a total of 27 residues from the wild-type protein, including the 10 residues spanning the unstructured loop between strands A and B, and inserting a Gly–Gly linker.

Figure 1.

NMR Structures of (A) first generation helix-less variant of I-FABP (Δ17-SG), and (B) wild-type I-FABP complexed wilth palmitate. In Δ17-SG, the position of the unstructured loop indicated by the arrow.

The Δ27-GG variant was engineered with the intent of designing a stable and compact all-β-sheet protein that could be used as a catalytic scaffold. Cysteine variants of the wild-type I-FABP have been used as artificial enzymes by covalently conjugating a catalytic active group to their binding cavity. Distefano and coworkers (Kuang and Distefano 1998; Haring and Distefano 2001; Tann et al. 2001) showed that the pyridoxamine derivatives of a V60C variant of the wild-type protein could catalyze the enantioselective reductive amination of α-keto acids to α-amino acids, increasing the rate of reaction by more than three orders of magnitude. The removal of the helical domain may give the substrates undeterred access to the binding cavity, thereby increasing the potency of the helix-less variants of I-FABP as catalytic scaffolds. However, the marginal stability of the first generation helix-less variant Δ17-SG made the investigation of its use as a catalytic scaffold difficult. Therefore, a variant with open access to the binding cavity, but increased stability, could serve as a better scaffold for the design of artificial enzymes or drug carriers.

In this report, we describe the stability and three-dimensional structure of Δ27-GG in comparison to the wild-type and first generation helix-less proteins.

Results

Protein stability

The urea denaturation curves for the apo forms of Δ27-GG, wild-type, and Δ17-SG are displayed in Figure 2 ▶. All the unfolding studies were performed under equilibrium conditions, and the unfolding process is reversible for Δ27-GG, as with the other two forms of the protein (Kim at al. 1996). The data obtained upon unfolding and refolding of Δ27-GG are displayed as closed and as open triangles, respectively. The data were analyzed by linear extrapolation methods (LEM) as implemented in Santoro and Bolen (1988). The least squares-fitted parameters indicative of the relative stability of the proteins as determined from both urea and guanidine hydrochloride (GdnHCl) denaturation experiments are summarized in Table 1.

Figure 2.

Equilibrium unfolding of Δ27-GG (filled triangles), Δ17-SG (filled squares), and wild-type I-FABP (filled circles) using urea (A) and GdnHCl (B) as denaturants. (Open triangles) The refolding process of Δ27-GG as a function of denaturant concentration.

Table 1.

Stability of I-FABP variants as determined by chemical denaturation using urea (A) and GdnHCl (B) as denaturants

Protein	ΔGou(H2O) (kcal mole−1)	mG (kcal mole−1 M−1)	Midpointa (M)
A. wild-type
I-FABP	5.38 ± 0.12	−3.99 ± 0.19	5.38 ± 0.16
Δ17-SG	4.17 ± 0.07	−4.29 ± 0.17	4.11 ± 0.22
Δ27-GG	4.81 ± 0.28b	−4.06 ± 0.24	4.67 ± 0.11b
B. wild-type
I-FABP	5.61 ± 0.21	−3.77 ± 0.19	1.36 ± 0.07
Δ17-SG	4.07 ± 0.20	−4.18 ± 0.25	0.86 ± 0.09
Δ27-GG	4.72 ± 0.25b	−4.55 ± 0.27	1.13 ± 0.06b

^a The midpoint represents the denaturant concentration at the midpoint of the transition.

^b Students T-test (P < 0.1) showed that the differences between Δ27-GG and the other variants are statistically significant.

The trends of midpoints of denaturation and free energies of unfolding are the same for the experiments using urea and GdnHCl as chemical denaturants. The midpoints of denaturation of Δ27-GG, wild-type, and Δ17-SG using GdnHCl as denaturant were 1.13 ± 0.06, 1.36 ± 0.07, and 0.86 ± 0.09 M, respectively. The previously reported values for wild-type and Δ17-SG under the same experimental conditions are 1.35 ± 0.11 and 0.89 ± 0.05 M, respectively (Kim et al. 1996). The Student’s t-test revealed that the differences in free energies among the three proteins are statistically significant.

Ligand binding

Fatty-acid binding to Δ27-GG, wild type I-FABP, and Δ17-SG was monitored by changes in steady-state intrinsic tryptophan fluorescence as a function of ligand concentration. All of the experiments were performed at 20°C in 20 mM sodium pyrophosphate, 10 mM NaCl, and 135 mM KCl (pH 9.0). Under these experimental conditions, the ligandbinding properties of the proteins are similar to those at physiological pH. However, the monomer solubility or critical micellar concentration (CMC) of oleate increases from ~0.01 mM at pH 7 to 1 mM at pH 9.0 (Cistola et al. 1986; Cistola and Small 1990) lending a wider oleate concentration range for performing binding assays without ligand self-association. Dynamic light-scattering experiments performed on the proteins complexed with oleate also showed no apparent protein self-association.

Figure 3 ▶ shows the stoichiometric fatty-acid-binding isotherms of all three proteins. All of the data sets were corrected for dilution. The binding isotherms for the wild-type and Δ17-SG proteins are biphasic, in which an initial fluorescence enhancement component is followed by a quenching component. Each component (enhancement or quenching) represents change in the microenvironment of the tryptophan side chain inside of the binding cavity, resulting from the ligand-binding process. Therefore, each individual phase in the binding isotherm of the proteins could represent a ligand-binding step. These biphasic isotherms were fit to a two-step binding model equation in order to determine the lower limit of the number of binding sites in the proteins. The results imply that the wild-type I-FABP and Δ17-SG have two binding sites. For the wild-type protein, the second binding step is too weak to be detected for fatty acid-to-protein concentration ratios less than 30 : 1. However, the second binding step is more prominent with Δ17-SG.

Figure 3.

Stoichiometric oleate-binding isotherms for the wild-type (A,B), Δ17-SG (C), and Δ27-GG (D). B is an inset of A for oleate concentration range of 0–30 μM. In this concentration range, the titration curve appears to be monophasic.

In contrast to wild-type and Δ17-SG, the oleate binding isotherm of Δ27-GG exhibits more than two phases, an initial fluorescence enhancement, a second quenching component followed by a third enhancement, and finally a fourth one that quenches (Fig. 3D ▶). The overall dynamic range for the fluorescence enhancement and quenching phases was 60%.

The stepwise binding affinities could not be accurately determined, as the experimental conditions were optimized for measuring binding stoichiometry, not affinity. Because of the multiple sites and the weak nature of some of the sites, it was not possible to quantitate the binding affinities in a definitive manner.

Resonance assignments and secondary structure

The gradient- and sensitivity-enhanced ¹⁵N-HSQC spectrum of apo-Δ27-GG protein in phosphate buffer at 25°C and pH 7.2 showed that the amide resonances were well dispersed. Overall, the backbone amide resonances of 100 of the 106 residues were assigned. The six residues that could not be assigned include A1, F2, G9, G10, and N54, A73 (wild-type numbering). Residues G9 and G10 represent the linker introduced between strands A and B, whereas the residues N54, A73 are located at the C-D, and E-F turns, which are part of the dynamic portal region in the wild-type protein (Hodsdon and Cistola 1997a). The amide resonances of these residues were missing even in gradient- and sensitivity-enhanced HSQC experiments, possibly due to rapid hydrogen exchange of the amide hydrogens at 25°C and pH 7.2. Complete side-chain aliphatic ¹H and ¹³C assignments for 103 of 106 residues has been established using the series of proton and carbon TOCSY-type experiments.

Secondary structure

A comparison of the chemical shift-derived secondary structures of Δ27-GG, Δ17-SG, and wild-type I-FABP is displayed in Figure 4 ▶. The chemical shift index (CSI) analysis was performed as described previously by Wishart and coworkers (Wishart et al. 1992; Wishart and Sykes 1994a,b) using ¹H^α, ¹³C^α, ¹³C^β, and ¹³CO chemical shift values. Overall, the positions of the secondary structure elements for Δ27-GG are the same as that of the β-sheet domain of the wild-type protein. The positions of the β-strands match those of the wild-type apo-protein as well as that of the first-generation helix-less I-FABP (Δ17-SG). The major difference between the secondary structures of Δ27-GG and the wild-type protein is the absence of the helical domain. This difference is also apparent from the far-UV circular dichroism (CD) spectra collected for the three proteins (data not shown). The far-UV CD measurements reveal that both Δ17-SG and Δ27-GG appear to have as much β-sheet content as the wild type I-FABP. The spectra also displayed a decrease in molar ellipticity at 196 nm for both Δ27-GG and Δ17-SG, which implies the loss of α-helical content of the two proteins compared with that of the wild-type protein.

Figure 4.

Alignment of chemical shift-derived secondary structures of (A) wild-type I-FABP, (B) Δ17-SG, and (C) Δ27-GG. The values along the Y-axis represent the chemical shift indices of each residue along the sequence. Positive indices indicate β-strand secondary structure, whereas negative values represent α-helix. The absence of a bar indicates a CSI value of 0 for an unstructured part of the protein or the lack of chemical-shift assignment.

The main difference between the secondary structures of Δ27-GG and Δ17-SG is the length of the loop between the β-strands A and B. This loop consists of only three residues in Δ27-GG as opposed to 10 residues in Δ17-SG.

Tertiary structure

The final structure calculation included 1151 manually assigned unambiguous restraints and 1022 automatically assigned restraints using ARIA (Ambiguous Restraints for Iterative Assignment, version 1.1.2; Nilges et al. 1997^; Brünger et al. 1998^; Linge 2000^; Linge et al. 2001). Of the 1022 that were assigned using ARIA protocol, 804 were unambiguous and 218 were ambiguous distance restraints. The profile of the distribution of the long-range and total restraints is very similar to what was observed for the wild-type protein, with generally higher number of restraints for the residues in the β-strands and relatively lower number of restraints-per-residue for β-turns and loops. A plot of the distribution of restraints is deposited as supplementary material.

The stereo diagram of 10 final superposed structures that were calculated using ARIA/CNS is shown in Figure 5A ▶. The RMS deviation of the overlay of the β-sheet residues of the 10 lowest energy structures is 0.56 Å. Structural quality statistics as evaluated using PROCHECK and PROCHECK-NMR (Laskowski et al. 1993, 1996) are given in Table 2. All 20 structures converged and exhibited good geometry, with minimal distance violation >0.5 Å or dihedral violations >5°. PROCHECK-NMR analysis of the 10 ensemble structures revealed that 94% of the backbone angles lie in the regions of the Ramachandran plot classified as allowed, and 37% of the residues were in the favored regions. The three residues that are classified in the disallowed region are in the reverse-turn segments of the protein.

Figure 5.

(A) Stereo diagram of the backbone Cα traces of 10 superimposed structures of apo-Δ27-GG. (B) Ribbon diagram of the average coordinates of the 10 NMR ensemble structures of Δ27-GG. The deletion site is indicated by the arrow. The secondary structures displayed were obtained from the CSI analysis.

Table 2.

Structural and stereochemical statistics for apo-Δ27-GG as assessed by MOLMOL and PROCHECK-NMR

Ensemble RMSD values	All residues	Ordered residues
Average pairwise Cα RMSD (Å)	1.57 ± 0.35	1.16 ± 0.25
Average main-chain RMSD from mean coordinates (Å)	1.23 ± 0.33	1.21 ± 0.29
Statistic		Ensemble average
Ramachandran plot statistics
Residues in all allowed regions		88 (94%)
Residues in most favored regions [A, B, L]		37 (40.0%)
Residues in additionally allowed regions [a, b, l, p]		39 (41.2%)
Residues in generously allowed regions [~a, ~b, ~l, ~p]		9 (9.6%)
Residues in disallowed regions		3 (3.2%)
Main-chain statistics
Standard deviation of ω(deg)		2.8
no. of bad contacts/100 residues		9.9
Standard deviation of “ζ-angle” (deg)		1.2
Standard deviation of H-bond energy (kcal/mole)		0.8
deviations from ideal bond lengths (Å)		0.010 ± 0.006
deviations from ideal bond angles (deg)		1.15 ± 0.76
Side-chain statistics
χ1 gauche minus		25.8
χ1 trans		25.1
χ1 gauche plus		24.2
χ1 “pooled”		22.7
χ2 trans		26.8

The numbers in parentheses for the Ramachandran plot statistics represent the percentile values of the number of residues.

Structural description

The structure comprises 10 antiparallel β-strands that make up two β-sheets, which are nearly orthogonal to each other (Fig. 5B ▶). This structure is generally consistent with the β-sheet structure of the wild-type I-FABP (Hodsdon et al. 1996) and that of the first generation helix-less protein (Δ17-SG; Steele et al. 1998). As could be seen from the solvent accessible surface maps of Δ27-GG and the wild-type protein (Fig. 6 ▶), one end of the ligand-binding cavity is open and exposed to solvent in Δ27-GG (Fig. 6A ▶). In the wild-type protein, this end of the cavity is covered by the helical domain (Fig. 6B ▶). In Δ27-GG, all H-bonding networks and van der Waals contacts involving the helix-turn-helix and C-D and E-F turns that help tether the portal region are disrupted by the deletion of the entire helical domain.

Figure 6.

Solvent-accessible surface maps of Δ27-GG (A) and wild-type I-FABP (B) as viewed from directly above the α-helical domain. The binding cavity is sequestered from the solvent by the helical domain in the wild-type protein (magenta), whereas in Δ27-GG, it is completely solvent accessible. The figures were made using a Lee and Richards (1971) contact surface as implemented in MOLMOL (Koradi et al. 1996) with a solvent molecule of radius 1.4 Å. The structure elements are color coded as follows: β-sheet, cyan; α-helices, magenta; all other residues, gray.

Despite the remarkable similarity between the structures of Δ27-GG and Δ17-SG, a predominant difference is seen in the length of the loop between β-strands A and B. This loop consists of 10 residues in Δ17-SG, whereas in Δ27-GG, it is a reverse turn consisting of only three residues. This turn has Type I β-turn geometry, which is the most frequent type of β-turn.

Saturation transfer data analysis

The ratios of ¹⁵N HSQC peak intensities with a 2.0-sec water presaturation pulse (I_sat) and without presaturation of the water resonance (I_nosat) were used to map the backbone order and disorder along the sequence. The general profile of the saturation transfer data is remarkably similar to that of the wild-type protein, taking the deletion of the residues in the helical domain into consideration. In Δ27-GG, residues 54–57 (C-D turn) and 73–77 (E-F turns), which are part of the dynamic portal region in the wild-type protein, exhibit higher extent of amide proton saturation transfer, whereas the rest of the protein structure exhibits low-saturation transfer rates. This is consistent with that observed for the wild-type protein (Hodsdon and Cistola 1997a,b). A notable difference is that residues G44, G86, and N111, which could not be assigned for the wild-type protein under these conditions, are clearly observed in Δ27-GG, possibly implying slight localized differences in the hydrogen bonding for these residues.

Discussion

We constructed a second-generation helix-less variant of I-FABP by deleting 27 residues spanning the distal half of the first β-strand along with those that define the entire helical domain. This deletion included the 10 residues that constitute the ill-defined loop in Δ17-SG. The optimization of loop length led to a more stable and compact all-β-sheet protein.

Comparison of the free energies of unfolding obtained using chemical denaturation experiments and linear extrapolation methods revealed that the stability of Δ27-GG is intermediate between the wild-type and the first generation helix-less I-FABP (Δ17-SG). The Δ27-GG variant is less stable than the wild-type protein, possibly due to the loss of the interdomain hydrogen bonds and van der Waals contacts between the α-helix and β-sheet domains. However, as predicted, the second-generation helix-less variant Δ27-GG with a short reverse turn between the first two strands is more stable than the first generation helix-less protein, Δ17-SG, which had a longer unstructured loop.

Structural studies using NMR revealed remarkable topological similarity between Δ27-GG, the wild-type protein, and Δ17-GG. Like Δ17-SG (Steele et al. 1998), Δ27-GG exhibited an overall β-sheet structure comprised of two five-stranded β-sheets that are oriented nearly orthogonal to each other. The main difference between Δ27-GG and the wild-type is the absence of the α-helical domain. As with Δ17-SG, solvent-accessible surface maps showed that the binding cavity in Δ27-GG is exposed to the solvent (Fig. 6A ▶). In the wild-type protein, the binding cavity is sequestered from the solvent as a result of the interdomain interactions between the α-helical and β-sheet domains.

The closed end of the β-barrel structure of Δ27-GG is effectively covered by the aromatic side chains of the residues Phe 2, Trp 6, Phe 47, Trp 82, and Phe 37. These residues along with Tyr 70, Leu 72, Leu 102, and Tyr 117 constitute the primarily hydrophobic interior of the β-barrel. Most of these residues were shown to interact with the hydrocarbon chain of the fatty acid in the structure of the wild-type I-FABP complexed with palmitate (Sachettini et al. 1989; Hodsdon et al. 1996). The widened gap between β-strands D and E, which is common in the structure of intracellular fatty-acid-binding protein, is also conserved in Δ27-GG. As was observed for the wild-type protein, the nonpolar contacts between the side chains of a number of residues including Ile 58, Val 60, Phe 68, and Tyr 70 fill in this apparent gap.

The chemical shift-derived secondary structure and the tertiary structure of Δ27-GG displayed no extended loops between any neighboring β-strands. The average length of the β-turns is 3.1 residues compared with 3.7 for the first generation helix-less variant. This makes Δ27-GG more compact and streamlined than Δ17-SG.

Ligand-binding studies under stoichiometric conditions showed that Δ27-GG has peculiar fatty-acid-binding properties. It seems to have the ability to bind more than three fatty acids, in stark contrast to the wild-type and Δ17-SG. The oleate-binding isotherm of Δ27-GG has more than three phases. This change in binding characteristics was unexpected, especially in view of the fact that the difference between Δ27-GG and Δ17-SG is a deletion of 10 residues that have no major structural significance. Fatty-acid titrations up to 150 : 1 fatty acid-to-protein mole ratio revealed a weak, but reproducible second binding step in the wild-type protein (Fig. 3A ▶). This is the first evidence that wild-type I-FABP can bind a second fatty acid, albeit very weakly. In the same oleate concentration range, Δ17-SG also has a biphasic binding curve, indicative of the presence of a second binding step.

Two-dimensional NMR experiments were collected on natural abundance proteins complexed with a ¹³C-enriched palmitate in order to confirm our observations of increased ligand-binding stoichiometry. The spectra for Δ27-GG exhibited partially overlapped peaks for the methyl group of the fatty acid. Although this implied increased stoichiometry, it hindered the quantitative estimate of the number of binding sites. However, similar experiments were carried out on other single-site and helix-less variants of I-FABP that revealed well-resolved peaks showing higher-than-expected ligand-binding stoichiomerty. The results for these additional variants are reported in a separate manuscript to be submitted elsewhere.

The structure of Δ27-GG revealed no obvious structural anomaly in the ligand-binding cavity that could provide an explanation for its atypical ligand-binding properties. The only apparent difference between the structures of the two helix-less variants is the length of the loop or turn between the first two strands. Careful scrutiny of the structural significance of the 10 residues spanning the ill-defined loop in Δ17-SG brought to our attention an ion-pair interaction between the side chains of D34 and R126. This is one of several long-range cooperative interactions in the structure of the wild-type I-FABP that is stabilized by the presence of the ligand (see Fig. 7 of Hodsdon and Cistola 1997a). These interactions shift the conformational equilibrium from the open to the closed states of I-FABP. Because D34 is one of the deleted residues in Δ27-GG, the D34/R126 long-range coupling is nonexistent. The absence of this interaction may perturb the equilibrium between the different conformational states, possibly leading to alterations in the number of binding sites in Δ27-GG. Studies of the role of D34/R126 in defining binding stoichiometry will be reported elsewhere.

Mark Distefano and coworkers have shown that V60C mutants of the wild-type I-FABP could be used as artificial enzymes for enantioselective reductive amination reactions of α-keto acids to α-amino acids (Kuang and Distefano 1998; Haring and Distefano 2001; Tann et al. 2001). It has been suggested that the covalent conjugation of the pyridoxamine moiety to Cys 60 positions the cofactor strategically close to the side-chain cationic group of R126, which could result in facilitating the binding of a carboxylate-bearing substrate through electrostatic interaction. In the presence of bound ligand, the wild-type protein favors the closed conformation in which R126 is involved in intramolecular long-range interaction with the carboxylate group of D34. However, in Δ27-GG, this interaction is decoupled, owing to the fact that D34 is part of the 27 residues that were deleted. This frees up the cationic guanidino group of R126, possibly for electrostatic interaction with substrates that contain carboxylate groups. This attribute, along with its favorable stability and open access to the binding cavity, may give Δ27-GG an advantage over the wild-type protein and Δ17-SG as a catalytic scaffold.

Materials and methods

Materials

The mutagenic primers were synthesized by the Nucleic Acid Chemistry Laboratory of Washington University School of Medicine. The high-fidelity PCR kit and the rapid ligation and transformation kits were purchased from Clontech and Boehringer Mannheim, respectively. DNA sequencing was performed using the DNA Sequencing Kit from Perkin Elmer. DNA purification kits were obtained from Promega. Oleic acid was purchased from Sigma Chemicals. The denaturants Ultrapure Guanidine HCL and Ultrapure Urea were obtained from ICN Biochemicals. Uniformly ¹³C-enriched glucose and ¹⁵N-enriched ammonium chloride were purchased from Cambridge Isotope Labs and Isotec.

Protein biosynthesis and purification

The Δ27-GG variant was engineered by deleting 27 contiguous residues (from D9 through N35, inclusive) spanning the helix-turn-helix motif of I-FABP and by inserting a G-G linker to provide a link between the two ends of the deletion site. As described previously (Kim et al. 1996), recombinant DNA techniques were used to construct deletion using 36-nt mutagenic primers that lack amino acid codons corresponding to the deletion site. The sequence of the mutagenic oligomer used was 5′-GGC-ACTTG GAAAGTAGGAGGATTGAAACTGACGATC-3′. PCR protocols were used to amplify the mutagenic DNA inserts. The purified inserts were ligated to a pMON5840-IFABP plasmid and transformed into a DH5α Escherichia coli using high-fidelity ligation and transformation kits. The mutant plasmid was purified and transformed into an MG1655 E. coli expression host cell.

The nonisotope-enriched mutant protein was overexpressed in the E. coli bacteria harboring the pMON plasmids from a recA promoter using nalidixic acid as described by Kim et al. (1996). [¹³C, ¹⁵N]-enriched protein was overexpressed in the host E. coli bacteria using the two-stage process described in Hodsdon et al. (1995).

The same procedures were used to purify the nonisotope-enriched and [¹³C,¹⁵N]-enriched proteins. The harvested cells were partially lysed using the freeze/thaw protocol described previously (Johnson and Hecht 1994). The protein was purified to homogeneity from the soluble fraction of the cell lysate by repeated pressure dialysis in phosphate buffer (20 mM potassium phosphate at pH 7.2), followed by gel filtration chromatography in Sephadex G-50 and Q-Sepharose columns at 4°C. Finally, the protein was delipidated in a lipophylic Sephadex column at 37°C as described previously (Glatz and Veerkamp 1983). The mutant protein appeared as a single band on SDS–polyacrylamide gels and migrated around 11 kD, which is the expected molecular mass. No aggregation was observed during the purification.

The final yield of the purified and delipidated nonisotope-enriched protein from a 4 L high-density fermentation with rich medium was ~2 g. The final yield of a uniformly ¹³C/¹⁵N-enriched protein from a 1.25 L fermentation with minimal medium was 90 mg. The proteins were stored at 4°C.

Conformational stability of proteins

Chemical denaturation experiments were carried out using urea and guanidine hydrochloride (GdnHCl) as denaturants. The equilibrium unfolding and refolding process was monitored by changes in the steady-state tryptophan fluorescence intensities (λ_excitation = 285 nm and λ_emission = 328.5 nm) using PTI Alphascan fluorometer (Photon Technology International) with a PC interface for data collection. A 2.0-μM protein solution (in 20 mM potassium phosphate buffer at pH 7.4) was titrated with a standard denaturant (urea or GdnHCl) in the same buffer conditions, and the equilibrium fluorescence readings were recorded at ambient temperature. Equilibrium was achieved within 3 min and the titrations were performed in triplicate.

To prove the reversibility of the folding process, a 2.0-μM protein was preincubated in a 8.0-M urea (or 2.2 M GdnHCl) for 1 h, and the changes in fluorescence intensity were monitored as the mixture was diluted with potassium phosphate buffer.

After the data were corrected for dilution, the parameters indicative of the energetic stability of the proteins were obtained using linear extrapolation methods by fitting the corrected data to the following equation from Santoro and Bolen (1988):

where Y_obs is the observed fluorescence intensity, [D] is the concentration of denaturant, R is the gas constant, m_n and m_u are slopes of the pre- and post-transition baselines, respectively, and Y_n and Y_u are the intercepts at zero denaturant concentration of the pre- and post-transition baselines, respectively. The m value is a measure of the dependence of the free energy of unfolding, ΔG^o_u, on the concentration of denaturant, [D]. The ΔG^o_u (H₂O) value is the extrapolated free energy of unfolding at zero denaturant concentration. The data fitting process was performed using nonlinear least squares regression as implemented in the program SCIENTIST.

For comparison, the same chemical denaturation experiments were repeated for the wild-type I-FABP and Δ17-SG under identical conditions.

Ligand-binding stoichiometry

A 2.5-μM Δ27-GG apo-protein solution (20 mM sodium pyrophosphate, 135 mM KCl, 10 mM NaCl [pH 9.0], at 20°C) was titrated with potassium oleate in the same buffer, and steady-state fluorescence changes were monitored at equilibrium (λ_excitation = 285 nm, λ_emission = 325 nm). Equilibrium was usually achieved within 2 min. The potassium oleate concentration range for the titration was 0 to 150 μM. No self-association of the oleate was observed for this concentration range on the basis of light scattering.

The titration was repeated for wild-type I-FABP and Δ17-SG under the same experimental conditions. The equilibrium data points were then corrected for dilution and fit to a fluorescence-binding model using the SCIENTIST nonlinear least squares fitting program to assess the minimum number of binding steps or sites.

Circular dichroism

All CD spectra were collected in a 20 mM potassium phosphate buffer at ambient temperature using a Jasco J600 spectropolarimeter. The protein concentrations used for the experiments were 0.11 mg/mL for Δ27-GG, 0.10 mg/mL for wild-type I-FABP, and 0.10 mg/mL for Δ17-SG.

NMR spectroscopy and protein structure calculations

Backbone and side-chain chemical shift assignments

All NMR spectra were collected at 25°C in potassium phosphate buffer (1.8 mM protein, 20 mM potassium phosphate, 135 mM KCl, 10 mM NaCl, 0.5% NaN₃ [pH 7.2]) using a Varian Unity 500 NMR spectrometer, with the exception of the two-dimensional NMR experiments for the assignment of aromatic side chains. The latter experiments were collected using a Varian Unity INOVA600 NMR spectrometer. The sequence-specific backbone resonance assignments were performed using a combination of three-dimensional HNCACB (Kay et al. 1994; Muhandiram and Kay 1994), CBCA(CO)NNH (Grzesiek and Bax 1992a), HNCO (Grzesiek and Bax 1992a), and CBCA(CO)CAHA (Kay 1993) and a two-dimensional ¹⁵N HSQC experiment. The aliphatic side-chain resonances were assigned using a series of three-dimensional TOCSY experiments, which include HCC(TOCSY)NNH (Grzesiek et al. 1993), CC(TOCSY)NNH (Grzesiek et al. 1993), an aliphatic HC(C)H-TOCSY (Kay et al. 1993). Side-chain aromatic ring ¹³C and ¹H resonance assignments were carried out using an aromatic three-dimensional HC(C)H-TOCSY (Kay et al. 1993) along with a series of two-dimensional experiments that include CG(CB)HB, CG(CD)HD, CG(CDCE)HE, and CG(C^aro)H^aro-TOCSY (Prompers et al. 1998). Two cycles of 7.1-kHz FLOPSY-8 spin-lock sequence were used for the CG(C^aro)H^aro-TOCSY with a mixing period of 6.7 msec.

Collection of structural data

Distance restraints were acquired from three different isotope-edited NOESY experiments. The first one was a three-dimensional ¹⁵N-¹⁵N-edited NOESY that correlates the NOE interactions between the amide protons. A mixing time of 150 msec and a relaxation delay of 1.0 sec were used for this experiment. The other three-dimensional NOESY experiments for structural data collection were three-dimensional ¹⁵N-edited and ¹³C-edited NOESY-HSQC. The mixing time for both experiments was 200 msec and the relaxation delay was 1.0 sec. All NOESY spectra were acquired on Varian Unity 500 spectrometer.

Saturation transfer experiments

Gradient- and sensitivity-enhanced ¹⁵N-HSQC experiments were used to collect amide proton saturation transfer data in order to map regions of the backbone order and disorder. Pairs of ¹⁵N-HSQC experiments with and without water presaturation were collected in triplicate using the Varian Unity 500 instrument. The peak intensities of the ¹H/¹⁵N correlations were measured using built-in peak-volume measurement routines in FELIX 2001 (Accelrys, Inc.) and processed in Microsoft Excel.

NMR data processing

Data processing and analysis of all the NMR spectra was carried out using the software VNMR version 6.1 (Varian Associates) and Felix 2001 (Accelrys, Inc) in Silicon Graphics Indy/R5000 and Sun workstations.

NOE assignment and structure calculation using ARIA

The NOE data obtained from the NOESY spectra were assigned in a semiautomatic manner using ARIA software (Ambiguous Restraints for Iterative Assignment, version 1.1.2; Nilges et al. 1997^; Brünger et al. 1998^; Linge 2000^; Linge et al. 2001). This algorithm integrates automated NOE assignment processes with structure calculations. It assigns ambiguous NOEs during the structure calculations using a combination of unambiguous distance restraints and an iterative assignment strategy. Automated pick-peaking processes were carried out using built-in routines in Felix 2001 (Accelrys, Inc.). Initially, 1151 unambiguous NOE restraints were assigned manually and the remaining unassigned NOE cross-peaks were output in the ARIA REGINE format and automatically converted into unassigned distance restraints in the first phase of ARIA. In addition to the manually assigned distance restraints, 176 ψ and φ dihedral angle restraints (derived from the CSI analysis, tolerance of 30°) were used to launch the iterative structure calculation process to reduce the number of possible folds the protein can adopt. The distance restraints were then assigned semiautomatically using the assignment possibility output of the first iteration of the structure calculation. The chemical-shift tolerances were set to 0.04 ppm for the proton dimensions and 0.35 ppm for the indirectly detected heteronucleai dimension for all of the spectra. In each iteration, the 10 lowest energy structures were used as template structure for the next iteration, and the seven best structures were used for restraint violation analysis. This process was repeated three times until no additional assignment was observed as the iterations progressed.

Distance restraints obtained from NOE intensities are usually underestimated due to spin diffusion. This problem is more pronounced when longer mixing times are used in collecting NOESY data. It is a common practice to use wider error bounds to account for this problem. However, this approach may decrease the precision of the calculated structures. ARIA addresses the problem by simulating the spin diffusion network. The NOE intensities are calculated from a given structure using numerical integration of differential equations that govern the relaxation process. Selected structures of an iteration serve as a template for the calculation of the NOE intensities, and these calculated NOE intensities are used to calibrate the distance restraints for the next iteration. Such an iterative calibration process diminishes the inaccuracies of the target distances incurred by spin diffusion.

Electronic supplemental material

The coordinates of an ensemble of 10 structures of Δ27-GG have been deposited in Protein Data Bank (accession code: 1SA8). Two tables summarizing the acquisition and processing parameters for the NMR experiments on apo-Δ27-GG are provided (d27gg.doc) as supplementary material. A plot of the distribution of distance restraints as well as the saturation transfer profile of Δ27-GG are also included in this file.