Computational design and biochemical characterization of maize nonspecific lipid transfer protein variants for biosensor applications

Authors

  • Eun Jung Choi,

    1. Division of Biology, California Institute of Technology, Pasadena, California 91125, USA
    Search for more papers by this author
    • These authors contributed equally to this work.

  • Jessica Mao,

    1. Division of Biology, California Institute of Technology, Pasadena, California 91125, USA
    Search for more papers by this author
    • These authors contributed equally to this work.

  • Stephen L. Mayo

    Corresponding author
    1. Howard Hughes Medical Institute, Division of Biology and Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, California 91125, USA
    • Howard Hughes Medical Institute, Division of Biology and Division of Chemistry and Chemical Engineering, California Institute of Technology, 1200 E. California Blvd., Pasadena, CA 91125, USA; fax: (626) 568-0934.
    Search for more papers by this author

Abstract

Lipid transfer proteins (LTPs) are a family of proteins that bind and transfer lipids. Utilizing the maize LTP, we have successfully engineered fluorescent reagentless biosensors for the natural ligand of LTPs; this was achieved by using computational protein design to remove a disulfide bridge and attaching a thio-reactive fluorophore. Conformational change induced by ligand titration is thought to affect the fluorescence of the fluorophore, allowing detection of ligand binding. Fluorescence measurements show that our LTP variants have affinity to palmitate that is consistent with wild-type LTP. These molecules have the potential to be utilized as scaffolds to design hydrophobic ligand biosensors or to serve as drug carriers.

Protein-based biosensors recognize specific ligands and relay the binding event to a signal that can be easily detected and measured. The ligand specificity of such sensors and their capacity for reagentless measurement allow them to be used for a wide range of assay methods. Recently, bacterial periplasmic binding proteins (PBPs) with different ligand specificities were tagged with a fluorophore to produce biosensors for their natural ligands, which include sugars, amino acids, dipeptides, and ions (de Lorimier et al. 2002). Using computational protein design techniques, these proteins were engineered into biosensors with specificity for novel ligands such as Zn(II), trinitrotoluene (TNT), L-lactate, serotonin, pinacolyl methyl phosphonic acid (PMPA), and dihydroxyacetone phosphate (Dwyer and Hellinga 2004).

The natural ligands of PBPs are hydrophilic and relatively small. The binding site residues are also largely hydrophilic, thus making it difficult to design variants for binding large hydrophobic ligands. Our aim is to develop a protein platform capable of specifically binding large and/or hydrophobic ligands. For this purpose, we selected the nonspecific lipid transfer protein (LTP) from maize (mLTP). Plant LTPs are a family of proteins known for their ability to bind and transfer lipids. Their biological function is still unknown, but they may be involved in the formation of the cuticle layer, in somatic embryogenesis, and in plant responses to pathogenic stress (Capocchi et al. 2005). The two subfamilies of LTPs (LTP1 and LTP2) share eight conserved cysteines that form four disulfide bridges and have nonpolar binding pockets (Douliez et al. 2000a). Proteins belonging to the LTP1 subfamily are larger (∼90 residues) and bind various phospholipids, fatty acids, and glycolipids, while the smaller (∼70 residues) but more structurally flexible members of the LTP2 subfamily can bind to bulkier sterol molecules as well (Samuel et al. 2002; Cheng et al. 2004).

In this paper, we used mLTP to design a biosensor for its natural lipid ligands. mLTP is a 93-residue basic protein comprised of four α-helices, numbered 1–4, and four disulfide bonds (Fig. 1). Its small size and the availability of a high-resolution crystal structure with a bound ligand make mLTP a good candidate for computational protein design (Shin et al. 1995). A member of the LTP1 subfamily, mLTP has a large tunnel-like hydrophobic cavity with one wide and one narrow opening on either end. The wide opening is thought to be the main entrance and exit site for ligands (Shin et al. 1995; Gomar et al. 1996; Han et al. 2001). Because of this large hydrophobic cavity, apo-mLTP does not have a tightly packed hydrophobic core, and the disulfide bonds are thought to be important for maintaining the tertiary structure of the protein in the absence of ligand. Comparison of the crystal structures for apo-mLTP and holo-mLTP shows little difference, with only the residues in the C terminus exhibiting rmsd over 1 Å (Fig. 1; Shin et al. 1995). Our strategy was to replace mLTP's disulfide bonds one at a time, using the ORBIT protein design software, thus allowing a degree of conformational flexibility that could be exploited for fluorescence-based detection of ligand binding. We hypothesized that adding ligand to a flexible mLTP variant would rigidify the protein and lock it into a single conformation. This conformation change, from flexible to rigid, was expected to cause a fluorescence change of acrylodan, a thiol-reactive environmentally sensitive fluorescent probe, which was conjugated to one of the cysteines of the former disulfide bond. Two out of four acrylodan-conjugated variants showed a decrease in fluorescence with ligand titration, suggesting that the acrylodan becomes more exposed when ligand binds. Kd values derived from the titration data are close to those of the wild-type protein, measured using tyrosine fluorescence, indicating that the designed variants retain native-like binding properties.

Figure Figure 1..

Ribbon diagrams of mLTP. The palmitate-bound mLTP (cyan) is superimposed on the apoprotein (green). The four disulfide bonds are shown in orange. Palmitate is shown in spheres with carbon in magenta and oxygen in red.

Results and Discussion

mLTP designs

mLTP contains four disulfide bridges: C4–C52, C14–C29, C30–C75, and C50–C89. We used the ORBIT protein design software to eliminate each of the disulfide bridges in separate calculations. Of the various designs, five mutants were selected for analysis: C4H/C52A/N55E, C4Q/C52A/N55S, C14A/C29S, C30A/C75A, and C50A/C89E (Fig. 2). The disulfide bridge C4–C52 anchors two helices to each other, with C52 more buried than C4. In the resulting designs for C4–C52 elimination, C4H/C52A/N55E and C4Q/C52A/N55S, residues 4 and 55 form a cross helix hydrogen bond, H4–E55 and Q4–S55, with donor/acceptor heavy atom distances of 2.8 Å (Fig. 2A,B). In the C14A/C29S mutant, a hydrogen bond is designed between S29 and S26 (Fig. 2C). The C30–C75 disulfide bond is surrounded by nonpolar residues, and both cysteines are mutated to alanine in the C30A/C75A mutant (Fig. 2D). The C50–C89 disulfide bridge anchors the C-terminal loop to helix 3. The mutation C89E in the C50A/C89E mutant forms hydrogen bonds with R47, S90, and K54 (Fig. 2E).

Figure Figure 2..

Five variants from designs that individually removed each of the four disulfide bridges. (A) C4H/C52A/N55E, (B) C4Q/C52A/N55S, (C) C14A/C29S, (D) C30A/C75A, and (E) C50A/C89E. Both the wild-type disulfide bonds and the design results are shown in stick with CPK-inspired colors. Putative hydrogen bonds are shown with yellow dashed lines.

Experimental validation

The circular dichroism (CD) wavelength scans of wild-type mLTP and the designed mutants (Fig. 3) show that three of the five mutants (C4H/C52A/N55E, C4Q/C52A/N55S, and C50A/C89E) have CD spectra similar to that of wild-type mLTP, with minima at 208 nm and 222 nm, characteristic of helical proteins. Variants C14A/C29S and C30A/C75A, on the other hand, are not folded properly, with wavelength spectra resembling those of rice LTP with scrambled disulfide bonds (Lin et al. 2004). In the mLTP crystal structure, the C14–C29 and C30–C75 disulfide bonds both link the end of helix 1 (residues 4–18) and helix 4 (residues 65–75), respectively, to helix 2 (residues 27–39) near the narrow opening of the ligand binding cavity. This opening has been suggested to have limited flexibility by the low crystallographic B-factors of the residues around it (Shin et al. 1995). These observations and the fact that the C14A/C29S and C30A/C75A mutants are unfolded imply the importance of these two disulfide bridges in holding the molecule together. In contrast, the two disulfides that were successfully designed out without an apparent major change in the protein's fold, C4–C52 and C50–C89, connect the N-terminal and C-terminal regions of the protein to helix 3 (residues 43–58) near the wide opening of the ligand binding cavity. This opening is thought to be very flexible and is speculated to be the ligand entry and exit site; the apparent greater flexibility in this region of the protein may explain why elimination of the C4–C52 and C50–C89 disulfide bonds is tolerated. In addition, the designed polar interactions in the three variants associated with the C4–C52 and C50–C89 disulfide bonds could help stabilize the structure (Fig. 2A,B,E).

Figure Figure 3..

Circular dichroism wavelength scans of mLTP and designed variants. The slight differences in the ellipticities of the folded proteins are likely due to errors in protein concentration determination.

We determined the thermal stabilities of the mutants in the absence and presence of palmitate and compared them to wild-type mLTP (Fig. 4). The removal of the C4–C52 disulfide bridge significantly destabilizes the proteins relative to the wild type, lowering the apparent Tm by 28°C (Table 1). The removal of the C50–C89 disulfide bridge leads to a Tm only 10°C lower than that of wild type in the absence of palmitate. As evidenced by the increase in the apparent melting temperatures when palmitate is present, the mutants are still able to bind palmitate. There is a much larger gain in stability upon binding palmitate for the C4H/C52A/N55E and C4Q/C52A/N55S mutants (∼20°C) than is observed for the wild-type protein (8°C), suggesting that these mutants might undergo a large conformational change upon ligand binding. The differences in apparent Tms between these palmitate-bound mutants and wild type were 16°C and 18°C, respectively, compared to the 28°C difference observed between the unbound mutants and unbound wild type. Thus, the deletion of disulfide bonds in these mutants decreased the stability of the apoprotein far more than the complexed protein. This observation supports the likelihood that these mutants undergo large conformational changes when shifting from the apo form to the holo form and strengthens the expectation that they will be good candidates for biosensor design.

Table Table 1.. Apparent Tms of mLTP and designed variants
original image
Figure Figure 4..

Thermal denaturations of wild-type mLTP and designed variants in the absence and presence of palmitate. (A) C4H/C52A/N55E; (B) C4Q/C52A/N55S; (C) C50A/C89E.

Unbound C50A/C89E, on the other hand, is 18°C more stable than the C4–C52 mutants and only 10°C less stable than the wild-type protein. The better stability of this mutant could be due to the multiple putative salt bridges and hydrogen bonds that were introduced by the C89E mutation (Fig. 2E). Addition of palmitate raised the Tm by 6°C, similar to the 8°C observed for wild-type mLTP. Since the crystal and solution structures of wild-type mLTP show little change in conformation upon ligand binding (Shin et al. 1995; Gomar et al. 1996), we expect similar behavior for C50A/C89E.

Of the five mutants studied, we consider the C4–C52-derived variants the most promising as scaffolds for the development of a reagentless biosensor. Fluorescent probes can be extremely sensitive to their environment; by conjugating such a probe to the site of conformational change, these mutants could act as reporters for ligand binding.

Protein–acrylodan conjugates

We selected two variants, C4H/C52A/N55E and C50A/C89E, for reintroduction of one of the original cysteine residues in each of them. This did not disturb any of the interactions that were designed for the opposite residue because none of the designed interactions were between the two residues constituting the disulfide bond. This gave us four new mutants: C52A, C4H/N55E, C50A, and C89E. We conjugated acrylodan (Ac), an environment sensitive thiol-reactive fluorophore (Prendergast et al. 1983), to the resulting free cysteine in each protein. Trypsin digest and tandem mass spectrometry of C52A/C4–Ac confirmed that the acrylodan was conjugated to the correct cysteine. Analytical ultracentrifugation of C52A/C4–Ac showed that it is a monomer (data not shown), and gel filtration profiles for all the acrylodan-conjugated variants showed a single dominant peak with retention times similar to C52A/C4–Ac, suggesting that all the other variants are also monomers.

We obtained circular dichroism wavelength scans of the acrylodan-conjugated variants to ensure they were properly folded (Fig. 5). All four conjugates appeared folded, with characteristic helical protein minima near 208 nm and 222 nm; however, the C52A/C4–Ac spectrum was closest to the wild-type mLTP spectrum. Thermal denaturation experiments (Table 1) showed that, when acrylodan is conjugated to C4 or C52, the apoprotein stability is increased, while the stability of the ligand-bound protein is decreased, compared to the C4H/C52A/N55E mutant. For the variants with acrylodan conjugated to C50 or C89, we see a different trend in which the apparent Tms of both the apo- and ligand-bound forms of the protein decrease or stay the same compared to the C50A/C89E mutant. This points to the possibility that, when conjugated to C4 or C52, acrylodan may find its way into the ligand-binding pocket of mLTP, stabilizing the apoprotein. When ligand is added, the acrylodan is displaced by the ligand; thus, the acrylodan does not induce any stabilizing effect on the ligand-bound form of the protein. The fluorescence changes observed with ligand titration also supports this mechanism (see below). The same mechanism has been suggested for the fluorescence change that occurs upon ligand binding to acrylodan-conjugated intestinal fatty acid binding protein (Richieri et al. 1992).

Figure Figure 5..

Circular dichroism wavelength scans of the four protein–acrylodan conjugates and wild-type mLTP.

Ligand binding assays

We titrated the protein–acrylodan conjugates with palmitate to test the ability of the engineered mLTP variants to act as biosensors. As predicted, of the four protein–acrylodan conjugates, C52A/C4–Ac and C4H/N55E/C52–Ac showed the most marked difference in signal when palmitate was added (Fig. 6A). For each titration, the average of the fluorescence intensity time-based measurement at the emission maximum was used to fit a noncooperative binding equation (Dubreil et al. 1997). These fits gave Kds for palmitate of 60 nM for C52A/C4–Ac and 2 μM for C4H/N55E/C52–Ac (Fig. 6B), consistent with the 3 μM Kd that we obtained for wild-type mLTP using tyrosine fluorescence (Douliez et al. 2000b; data not shown).

Figure Figure 6..

Fluorescence titration data. (A) Titration of C52A/C4–Ac with sodium palmitate monitored by fluorescence emission at an excitation wavelength of 363 nm. Addition of sodium palmitate to C52A/C4–Ac decreases fluorescence. Fluorescence monitored at 466 nm and 489 nm, respectively, was used to determine the Kd of C52A/C4–Ac (60 nM) (B) and C4H/N55E/C52–Ac (2 μM) (C).

We have successfully engineered mLTP into a fluorescent reagentless biosensor for its naturally occurring nonpolar ligands. This was achieved using computational protein design to remove one of the disulfide bridges and then attaching a thio-reactive fluorophore to the free cysteine. We believe the observed change in acrylodan signal is a measure of the local conformational change the protein variants undergo upon ligand binding, which causes the displacement of the fluorophore from the hydrophobic binding pocket. Our hypothesis is that the removal of the C4–C52 disulfide bridge provides the N-terminal helix more flexibility and allows acrylodan to insert into the binding pocket. Upon ligand binding, however, acrylodan is displaced and shifts from an ordered nonpolar environment to a disordered polar environment. In addition to the melting temperature data reported above, the observed decrease in fluorescence emission and the red shift of the emission maximum as palmitate is added (Fig. 6A), are also consistent with this hypothesis.

Acrylodan-conjugated mLTP variants can be utilized in various applications. They can be used as scaffolds to design biosensors with specificities for different hydrophobic ligands, as was done with PBP for hydrophilic ligands (Dwyer and Hellinga 2004). Acrylodan-conjugated mLTP variants could also be used in a reliable, sensitive, high-throughput screening method for binding drug compounds. Furthermore, they could be designed to bind a specific ligand and used as drug carriers. Recently, there has been an interest in using proteins as drug carriers due to their high affinity and selectivity for their targets (De Wolf and Brett 2000). The proteins would not only protect potentially unstable molecules from oxidation and degradation, but they could also aid in solubilization and ensure a controlled release of the agents. In a study to determine the suitability of LTPs as drug carriers, various molecules having cosmetic or pharmacological interest were tested for binding to wheat LTP (wLTP). wLTP was found to bind to a variety of molecules with affinities low enough to allow slow release (Pato et al. 2001). mLTP variants designed to bind a specific drug could also allow controlled delivery as has been seen with wLTP.

Materials and Methods

Computational protein design

The crystal structure of mLTP with palmitate bound (PDB ID 1MZM) was energy-minimized for 50 steps using the DREIDING force field in order to regularize the structure (Mayo et al. 1990). Residues were classified as core, boundary, or surface based on solvent accessibility (Dahiyat and Mayo 1997a). Each of the four disulfide bridges was individually “reduced” by deletion of the S–S bond and addition of hydrogens. The corresponding structures were used in designs aimed at stabilizing the respective disulfide bridge deleted variants. The ORBIT protein design program uses an energy function based on the DREIDING force field (Mayo et al. 1990) and includes a Lennard-Jones 12–6 potential with van der Waals radii scaled by 0.9 (Dahiyat and Mayo 1997b), hydrogen bonding and electrostatic terms (Dahiyat et al. 1997; Gordon et al. 1999), and the Lazaridis-Karplus solvation model (Lazaridis and Karplus 1999) with the polar burial energies scaled by 0.6 (O. Alvizo, pers. comm.). In addition, the calculations included a scaled rotamer probability term calculated as 0.3log(Prot) where Prot are rotamer probabilities taken from the May 2002 version of the Dunbrack and Karplus backbone-dependent rotamer library (Dunbrack and Cohen 1997). An algorithm based on the dead-end elimination theorem (DEE) was used to obtain the global minimum energy amino acid sequence and conformation (GMEC) (Pierce et al. 2000; Gordon et al. 2003). For each design, non-proline, non-glycine residues within 4 Å of the two reduced cysteines were included as the first shell of residues and their amino acid identities and conformations were optimized. Residues within 4 Å of the first shell were considered the second shell and their conformations were allowed to change, but their amino acid identities were held fixed. The conformations and identities of the remaining residues were fixed. Based on the results of the initial design calculations, further restricted designs were carried out in which only designed positions making stabilizing interactions were included.

Expression and purification of designed proteins

An Escherichia coli codon-optimized gene encoding the wild-type mLTP amino acid sequence was synthesized and ligated into pET15b (Stratagene) by Blue Heron Biotechnology (www.blueheronbio.com) resulting in an N-terminal His-tagged protein with a thrombin cleavage site (MGSSHHHHHHSSGLVPRGSH…). Inverse PCR mutagenesis was used to construct all variants. The proteins were expressed in BL21(DE3) Gold cells (Stratagene) at 37°C after induction with IPTG (isopropyl-β-D-thiogalactopyranoside). Cells were resuspended in lysis buffer (50 mM sodium phosphate, 300 mM sodium chloride, 10 mM imidazole, pH 8.0) and lysed by passing through an Emulsiflex (Avestin) at 15,000 psi. Soluble protein was isolated by centrifugation at 20,000g for 30 min, loaded onto a Ni-NTA column, and eluted with elution buffer (lysis buffer with 400 mM imidazole). The proteins were further purified by gel filtration chromatography with phosphate buffer (50 mM sodium phosphate, 150 mM sodium chloride, pH 7.5). The molecular weights of the purified proteins were verified by SDS-PAGE and MALDI-TOF. Trypsin digest of the protein followed by mass spectrometer analysis showed that the N-terminal Met was cleaved and that the disulfide bonds are all native. Gel filtration profiles for all mutants looked similar to that of wild-type mLTP, which was verified to be a monomer by analytical ultracentrifugation (data not shown). Protein concentration was determined using the BCA assay (Pierce) with BSA as the standard.

Circular dichroism (CD)

CD data were obtained on an Aviv 62A DS spectropolarimeter equipped with a thermoelectric cell holder. Wavelength scans and thermal denaturation data were obtained from samples containing 50 μM protein. For wavelength scans, data were collected every 1 nm from 190 to 250 nm with an averaging time of 5 sec. For temperature denaturation, data were collected every 2°C from 1°C to 99°C using an equilibration time of 120 sec and an averaging time of 30 sec. Since the thermal denaturations were irreversible, the apparent melting temperatures (Tms) were obtained from the inflection point of the denaturation curves. For thermal denaturations of protein in the presence of palmitate (Sigma Aldrich), 150 μM palmitate, from stock solution in ethanol, was used.

Acrylodan labeling of mLTP variants

Purified proteins were concentrated to 10–20 μM. 6-acryloyl-2-(dimethylamino) naphthalene (acrylodan) (Molecular Probes) was dissolved in acetonitrile and added to the purified proteins in 10-fold excess concentration. The sample was incubated at 4°C overnight. All solutions containing acrylodan were protected from light. Precipitated acrylodan and protein were removed by filtering through 0.2 μm nylon membrane Acrodisc syringe filter (Gelman Laboratory), and the soluble fraction was concentrated. Unreacted acrylodan and protein impurities were removed by gel filtration chromatography with phosphate buffer. The eluents were simultaneously monitored at 280 nm for protein and 391 nm for acrylodan. Fractions were collected when absorbance was observed at both 280 nm and 391 nm. The conjugation reaction looked to be complete, as both absorbances overlapped for the only major protein peak observed. Purified proteins were verified by SDS-PAGE to be of sufficient purity, and MALDI-TOF analysis showed that they correspond to the native disulfide bonded form of the proteins with acrylodan conjugated. Trypsin digest analysis of acrylodan-conjugated proteins showed that the acrylodan was attached at the expected cysteine.

Fluorescence emission spectra and ligand binding assay

Ligand binding was monitored by measuring the fluorescence emission of protein–acrylodan conjugates with the addition of palmitate. Fluorescence measurements were performed on a Photon Technology International Fluorometer equipped with a stirrer at room temperature. The excitation wavelength was set to the excitation maximum for each variant and time-based measurements were taken at the wavelength of the emission maximum for 30–60 sec. Emission wavelength scans were taken at 2-nm intervals and 0.5-sec integration time. A stock solution of palmitate dissolved in ethanol was titrated into 2 mL of 500–1000 nM protein–acrylodan conjugate. The total amount of palmitate in ethanol was never allowed to exceed 1% of the sample volume. The dissociation constants (Kd) were determined by fitting the decrease in fluorescence to an equation for a non-cooperative binding model (Dubreil et al. 1997).

Acknowledgements

We thank Marie Ary for editing the manuscript. This work was supported by the Howard Hughes Medical Institute, the Ralph M. Parsons Foundation, the Army Research Office (Institute of Collaborative Biotechnologies), and an IBM Shared University Research Grant.

Ancillary