Article pubs.acs.org/biochemistry
Structural Determinants of Improved Fluorescence in a Family of Bacteriophytochrome-Based Infrared Fluorescent Proteins: Insights from Continuum Electrostatic Calculations and Molecular Dynamics Simulations Mikolaj Feliks,†,‡,§ Céline Lafaye,†,‡,§ Xiaokun Shu,∥,⊥ Antoine Royant,*,†,‡,§,# and Martin Field*,†,‡,§ †
Université Grenoble Alpes, Institut de Biologie Structurale (IBS), F-38044 Grenoble, France CNRS, IBS, F-38044 Grenoble, France § CEA, IBS, F-38044 Grenoble, France ∥ Department of Pharmaceutical Chemistry, University of California, San Francisco, California 94158, United States ⊥ Cardiovascular Research Institute, University of California, San Francisco, California 94158, United States # European Synchrotron Radiation Facility, F-38043 Grenoble, France ‡
S Supporting Information *
ABSTRACT: Using X-ray crystallography, continuum electrostatic calculations, and molecular dynamics simulations, we have studied the structure, protonation behavior, and dynamics of the biliverdin chromophore and its molecular environment in a series of genetically engineered infrared fluorescent proteins (IFPs) based on the chromophore-binding domain of the Deinococcus radiodurans bacteriophytochrome. Our study suggests that the experimentally observed enhancement of fluorescent properties results from the improved rigidity and planarity of the biliverdin chromophore, in particular of the first two pyrrole rings neighboring the covalent linkage to the protein. We propose that the increases in the levels of both motion and bending of the chromophore out of planarity favor the decrease in fluorescence. The chromophore-binding pocket in some of the studied proteins, in particular the weakly fluorescent parent protein, is shown to be readily accessible to water molecules from the solvent. These waters entering the chromophore region form hydrogen bond networks that affect the otherwise planar conformation of the first three rings of the chromophore. On the basis of our simulations, the enhancement of fluorescence in IFPs can be achieved either by reducing the mobility of water molecules in the vicinity of the chromophore or by limiting the interactions of the nearby protein residues with the chromophore. Finally, simulations performed at both low and neutral pH values highlight differences in the dynamics of the chromophore and shed light on the mechanism of fluorescence loss at low pH. infrared fluorescent protein”). A directed evolution approach subsequently led to the design of a protein with significantly improved infrared fluorescence, IFP1.4.3 Another variant, IFP2.0, was later obtained with a higher binding affinity for the chromophore biliverdin.4 The evolution tree of IFP1.4, shown in Figure 1, contains three intermediate proteins, IFP1.1, IFP1.2, and IFP1.3. Proteins from DrCBD to IFP1.2 occur as dimers, whereas proteins from IFP1.3 to IFP2.0 are monomeric. To date, only a few attempts, both experimental and computational, have been made to understand at the molecular level the origins of fluorescence in different proteins derived
P
hytochromes are photoreceptors used by green plants, algae, bacteria, and fungi to absorb light and convert it into physiological signals.1,2 They can absorb light in the red and farred regions of the electromagnetic spectrum. Efforts have been made to genetically engineer alternatives to the natural phytochromes that possess improved fluorescent properties.3,4 These properties include an increase in the quantum yield, an increase in the extinction coefficient, and a shift of the fluorescence maximum closer to the infrared region of the spectrum. Engineering of infrared fluorescent proteins has now become a new, emerging field.5 The chromophore-binding domain of the bacteriophytochrome from Deinococcus radiodurans (DrCBD) has attracted much attention.6 Significantly, a single-point mutant of this domain has been shown to be fluorescent in the near-infrared range7 and has been called IFP1.0 (for “first version of an © 2016 American Chemical Society
Received: March 31, 2016 Revised: July 18, 2016 Published: July 29, 2016 4263
DOI: 10.1021/acs.biochem.6b00295 Biochemistry 2016, 55, 4263−4274
Article
Biochemistry
In this paper, we have combined continuum electrostatic calculations and molecular dynamics simulations to study the protonation behavior and dynamics of the weakly fluorescent bacteriophytochrome and its six fluorescent mutants. We have used the available crystal structures of the bacteriophytochrome-based proteins as well as our newly determined, 1.11 Å resolution structure of IFP1.4, which provides an increased level of detail compared to that of the previously determined structure. The objective of our study has been to identify structural factors responsible for the gradually improving fluorescent properties within the given set of mutants by focusing on the dynamics of the monomeric versions of all the proteins. The knowledge of these structural factors would provide a rational basis for the future design of fluorescent proteins based on the biliverdin chromophore (or another bilin) with improved fluorescent properties.5
■
METHODS Protein Expression and Purification. The sequences of IFP1.1, IFP1.2, IFP1.4, and IFP2.0 with a C-terminal six-His tag were inserted into the modified pBad expression vector containing the heme-oxygenase-1 gene.3 Recombinant proteins were expressed and purified as described previously.4 The purity of the protein solutions was confirmed by sodium dodecyl sulfate−polyacrylamide gel electrophoresis. Final concentrations were determined by UV−visible absorption spectroscopy using calculated molar absorption coefficients at 280 nm of 32555 M−1 cm−1 for IFP1.1 and IFP1.2, 35915 M−1 cm−1 for IFP1.4, and 34045 M−1 cm−1 for IFP2.0. Crystallization of IFP1.4. Mutagenesis of residue 307 has been shown to improve crystal quality.9 We managed to crystallize the E307Y mutant of IFP1.4 using the hanging-drop method at 293 K at a concentration of 14 mg/mL in a solution consisting of 26% PEG 400 and 0.1 M sodium acetate (pH 5.0), i.e., under conditions similar to those used for IFP2.0.4 Crystals grew in 6 days, compared to 1 day for IFP2.0. X-ray Diffraction Data Collection and Structure Refinement. Prior to diffraction experiments, crystals were flash-cooled in liquid nitrogen because PEG 400 at this concentration is a cryoprotectant. X-ray diffraction experiments were performed at the European Synchrotron Radiation Facility (Grenoble, France). Data were collected at 100 K at a wavelength of 0.976 Å on beamline ID29.19 Diffraction data sets were processed using XDS,20 and intensities were scaled and reduced with AIMLESS.21 The crystal belongs to the C2 space group and diffracted to 1.11 Å resolution. The solvent content is 46%. Structural refinement was conducted with Refmac5 using anisotropic B factors for all atoms.22 As already described for IFP2.0, the thioether bond between Cys24 and the chromophore biliverdin is particularly sensitive to X-rays. Structural and experimental data have been deposited in the Protein Data Bank (PDB) as entry 5AJG. Crystallographic data statistics can be found in Table 2. pKa Measurements. The pKa values of the various IFPs were assessed using the measurement of their fluorescence in solutions at increasing pH values using phosphate-citrate buffer (from pH 2.6 to 7.2). Each buffer solution was obtained by mixing various amounts of 0.1 M citric acid and 0.2 M dibasic sodium phosphate.23 All samples (4 μL of concentrated IFP at 1 mg/mL with 33 μL of buffer at a given pH) were loaded onto a 96-well plate and placed in a fluorescence plate reader (Synergy H4, BioTek). Fluorescence was excited at 640 nm, and the fluorescence emission signal was integrated between
Figure 1. Evolution tree of D. radiodurans bacteriophytochrome-based infrared fluorescent proteins. Fluorescence quantum yields are indicated in parentheses as previously determined.3,4,10 Reversed mutations in IFP2.0 are underlined.
from the bacterial phytochrome. The structure of DrCBD was first determined in 2005 by Forest and co-workers at 2.5 Å resolution and represented the first structure of any fragment of a phytochrome.8 The resolution of this DrCBD structure was later improved to 1.8 Å because of a surface mutation.9 As a follow-up, three different structures of IFP1.0 were determined at resolutions as high as 1.24 Å, revealing that the crucial His207 residue could adopt two orthogonal orientations of its imidazole group.10 Finally, the same group determined the structure of IFP1.4 at 1.65 Å resolution, revealing that the addition of mutations decreased structural heterogeneity in the chromophore-binding cavity and induced a relative movement of two secondary elements of the protein.11 These observations agreed with the structure of IFP2.0 at 1.14 Å resolution that we determined shortly before this latter work.4 Other IFPs that have been investigated include the bright Wi-Phy, designed by the Forest group, which differs by only two mutations from its parent DrCBD and for which the structure has been determined at 1.75 Å resolution.10 A naturally monomeric fragment of a phytochrome from a Bradyrhizobium species was also forced to evolve into the definitively monomeric IFPs, namely, mIFP12 and its markedly blue-shifted variant iBlueberry.13 A distinct series of IFPs with emission maxima ranging from 670 to 720 nm14 have been developed by Verkhusha and co-workers from the bacteriophytochromes RpBphP2 and RpBphP6 from Rhodopseudomonas palestris.6 So far, little direct structural information about these proteins has been obtained.15,16 Likewise, a molecular dynamics study of a modeled structure of iRFP, derived from RpBphP2, suggested that the improved fluorescent properties of the biliverdin chromophore were due to an increased tilt of its terminal ring, a reduction in the number of water molecules with which it interacts, and a significant decrease in flexibility.17 Finally, another bacteriophytochrome from R. palestris, RpBphP1, was forced to evolve into the bright blue-shifted infrared FP BphP1-P1, for which a structure could be obtained.18 It is noteworthy that the mechanism of the blue shift in iBlueberry and BphP1-FP is identical and is due to the binding of the chromophore to a cysteine residue of a different domain than for other IFPs. 4264
DOI: 10.1021/acs.biochem.6b00295 Biochemistry 2016, 55, 4263−4274
Article
Biochemistry
pDynamo,25,31 interfaced to the external solver of the Poisson− Boltzmann equation, MEAD,32 and to the Monte Carlo sampling program, GMCT.33 Protonation state energies and titration curves were calculated using pDynamo’s own routines.31 The following parameters were set for the Poisson−Boltzmann continuum electrostatic model. Charges and radii for the protein and chromophore atoms were taken from the CHARMM27 force field28 and literature,29,30 respectively. To construct the volume of the protein, an ionexclusion layer of 2.0 Å and a solvent probe of 1.4 Å were used. The dielectric constants of the protein (εp) and the solvent (εs) were set to 4 and 80, respectively. The solvent was assigned an ionic strength (I) of 100 mM. The calculations were performed at 300 K. Electrostatic potentials were calculated using four grids, each consisting of 1213 nodes, with focusing steps at resolutions from 2.0 to 0.25 Å. The protonation probabilities of titratable residues were estimated for pH values ranging from 0 to 14. For every pH step, the MC calculation consisted of 100 equilibration scans and 3000 production scans. The chromophore was treated as a nontitratable site, and its protonation state was fixed either to the fully protonated form or to the form deprotonated at the B pyrrole ring, depending on the particular calculation. The carboxylic groups at rings B and C of the chromophore were treated separately as titratable sites. The parameters of glutamate (model pKa and atomic charges) were used for the treatment of these groups. Molecular Dynamics Simulations. Before the molecular dynamics simulations, the proteins were assigned protonation states according to the previous electrostatic calculations. For each protein, we performed simulations with the fully protonated chromophore at either low or neutral pH (4 or 7, respectively). Additionally, the simulations were repeated with the chromophore deprotonated at the B pyrrole ring at both pH values (see the Supporting Information for details). The simulations were performed using NAMD.34 We used custom Tcl scripts and the VMD35 package to prepare the models for the simulations as well as to analyze the resulting trajectories. In the first step, each protein structure was solvated in a rectangular cell of explicit water molecules, using the TIP3P water model. A 9 Å wide buffer of solvent molecules was used between the protein and the boundary of the periodic cell. Counterions, chloride and sodium, were added appropriately to balance the non-zero charge of each of the proteins. The final models consisted of 29 × 103 to 31 × 103 atoms, depending on the protein. For each protein, the simulation was preceded by 1000 steps of geometry optimization of the entire system using periodic boundary conditions. The initial temperature was set to 20 K. Both the heating stage and the consecutive production simulation were performed in the NPT ensemble, under a pressure of 1 bar. During the heating stage, the temperature of the system was increased in steps of 10 K/ns from 20 to 300 K. At the end of each heating window, 10000 equilibration steps were performed. In the last window, the number of such steps was doubled. The production run was performed at a final temperature of 300 K for 10 ns, using a time step of 2 fs. Bonds involving hydrogen atoms were constrained using the SHAKE method36 to decrease the computational cost of the simulation. During the simulations, we monitored the most important geometrical parameters of the system. These included, for example, the six dihedral angles of the chromophore and the root-mean-square deviation (rmsd) of its atoms. To ensure that the system had reached equilibrium, we traced the root-meansquare deviation of the chromophore atoms (see Figure S5 of
660 and 800 nm. Each curve was measured in triplicate and normalized against the value at pH 6.0, corresponding to maximal emission for each variant. Preparation of the Protein Models. The protein models used in our study were prepared on the basis of the highestresolution crystal structure, when available [DrCBD, IFP1.0, IFP1.4, and IFP2.0 (see Table 1 for the known structures of Table 1. Protein Structures, Crystallographic and ComputerGenerated, Used To Construct the Continuum Electrostatic and Molecular Dynamics Models protein
PDB entry
resolution (Å)
source 9
year
comments − − generated from IFP1.0 generated from IFP1.0 − − This Work lower resolution lower resolution
DrCBD IFP1.0 IFP1.1
2O9C 3S7O −
1.45 1.24 −
Wagner Auldridge10 this work
2007 2012 −
IFP1.2
−
−
this work
−
IFP1.4 5AJG 1.11 this work 2015 IFP2.0 4CQH 1.14 Yu4 2014 Other Available Structures of Phytochromes Not Used in DrCBD 1ZTU 2.50 Wagner8 2005 IFP1.4
4O8G
1.65
Bhattacharya11
2014
IFPs)]. IFP1.1 and IFP1.2, for which no structures are available, were modeled using the crystal structure of IFP1.0. The MODELLER24 package was used to perform the mutations as well as to repair unresolved atoms or side chains. All structures missing a loop between residues 127 and 152 were completed using the loop present in the structure of DrCBD. DrCBD, IFP1.0, IFP1.1, and IFP1.2 are dimers in solution, and their geometries were generated by applying the appropriate symmetry operations in pDynamo.25 For electrostatic calculations, we used both the dimeric and monomeric geometries. For molecular dynamics simulations, only the monomeric geometries were used. The initial preparation of the protein models was conducted in CHARMM.26,27 Molecular mechanics parameters for the protein were taken from the CHARMM27 force field.28 The parameters for the biliverdin chromophore in its fully protonated form were taken from the literature.29,30 The parameters for the chromophore deprotonated at the B pyrrole ring were refined on the basis of the parameters of the fully protonated chromophore. The refinement procedure is described in section S3 of the Supporting Information. During the initial preparation of the models, all titratable protein residues were set to their standard protonation states at pH 7; i.e., aspartates and glutamates were deprotonated, histidines doubly protonated, and all other residues protonated. The side chains of the mutated and repaired residues, the added loop fragment, and all hydrogen atoms were subsequently geometryoptimized in CHARMM. During the optimizations, the coordinates of the other parts of the proteins were kept fixed at their crystallographic positions. Determining the Protonation States of Titratable Residues. In the next step, we performed Poisson−Boltzmann electrostatic calculations combined with a Monte Carlo (MC) titration to evaluate protonation states of titratable residues in all models. The electrostatic calculations were performed using 4265
DOI: 10.1021/acs.biochem.6b00295 Biochemistry 2016, 55, 4263−4274
Article
Biochemistry the Supporting Information for the rmsd evolution in different proteins). For the important dihedral angles of the chromophore, we also plotted the distributions and tabulated the fitted normal distribution parameters based on the complete simulations, as well as on their first and second halves (see Figures S9−S12 and Tables S1 and S2).
■
RESULTS AND DISCUSSION Structure of IFP1.4. We determined the structure of IFP1.4 at 1.11 Å resolution and compared it to that of the parent DrCBD, to that of the closest progenitor for which a structure has been determined, IFP1.0, and to that of its descendant IFP2.0 (Table 2). For DrCBD and IFP1.0, we chose the Table 2. Data Collection and Refinement Statisticsa for IFP1.4 space group cell dimensions a, b, c (Å) α, β, γ (deg) resolution (Å) Rmerge I/σI completeness (%) multiplicity resolution (Å) no. of reflections Rwork, Rfree no. of atoms protein biliverdin water/ion B factor (Å2) protein biliverdin water/ion root-mean-square deviation bond lengths (Å) bond angles (deg) a
Data Collection C2 95.4, 53.0, 66.2 90, 90.9, 90 38.4−1.11 (1.13−1.11) 3.4 (62.3) 17.6 (2.0) 98.5 (92.1) 4.3 (3.7) Refinement 27.3−1.11 (1.14−1.11) 121711 0.146 (0.227), 0.168 (0.245)
Figure 2. Comparison between the chromophore environment of the newly determined structure of IFP1.4 and those of (a) IFP1.0 (PDB entry 3S7O) and (b) IFP2.0 (PDB entry 4CQH). Color code: IFP1.0, magenta; IFP1.4, cyan; IFP2.0, yellow. This figure was prepared with PyMOL.46
which was introduced between IFP1.0 and IFP1.1, has been shown to affect only the absorption and fluorescence emission maxima, but not the fluorescence quantum yield,11 thereby highlighting the importance of the A288V mutation.4 Finally, comparison of the DrCBD and IFP1.0 structures shows that the orientation of the side chain of residue 207 has effects on the positioning of Tyr263 and the loop bearing residues 204−206 (Figure S2). pKas of IFPs. We determined the pKas of IFP1.1, IFP1.2, IFP1.4, and IFP2.0. IFP1.1 and IFP1.2 have the highest pKa values (4.8), followed by IFP1.4 (4.6) and IFP2.0 (4.2) (Figure S3). The low pKa of IFP2.0 corresponds to the mutation of the ionizable residue His207 into a threonine, while the decreased pKa of IFP1.4 indicates a better shielding of the chromophore from bulk solvent. Protonation States of Biliverdin. For each of the studied proteins, we calculated the probabilities of protonation states of titratable residues. In addition to titratable protein groups, the bacterial phytochrome binds the biliverdin chromophore (Figure 3), which could also be a titratable group. In principle, the chromophore can bind or release one proton at each of the four pyrrole rings. However, there is a consensus in the literature about the protonation state of the chromophore. Raman spectroscopy and other experiments suggest that the bacteriophytochrome contains predominantly the fully protonated form of biliverdin.17,37 Theoretical studies using timedependent density functional theory indicate that the chromophore likely exists in its fully protonated form,38,39 as it was found that the calculated absorption spectrum of the deprotonated chromophore was shifted by 15−60 nm in comparison to the experimentally observed one. On the other hand, light- and X-ray-induced deprotonation of the biliverdin
2525 43 263 17.1 10.0 25.8 0.014 1.94
Values in parentheses are for the highest-resolution shell.
structures with the highest resolution, 2O9C9 and 3S7O,10 respectively. We previously determined the 1.14 Å structure of IFP2.0 (PDB entry 4CQH).4 Residual density on the chromophore showed that the two enantiomers already observed in IFP1.010 and the 1.65 Å structure of IFP1.410 are also present in our structure; only the enantiomer with the highest conformation occupancy was modeled. Comparison of IFP1.4 with IFP1.0 primarily shows that His207 has two orthogonal orientations of the imidazole ring, as already observed in a lower-resolution structure of IFP1.0.10 This conformational difference appears only to induce the movement of loop 197−206 and that of Tyr263 toward the chromophore (Figure 2a). Comparison of IFP1.4 with IFP2.0 (with an rmsd value of all Cα atoms of only 0.27 Å) shows that the chromophore environment has hardly changed (Figure 2b). The largest differences occur around residue 207 (His207 in IFP1.4 and Thr207 in IFP2.0), whereas there are only minor effects in the vicinity of the reversion mutation V186M and the quasi-isosteric mutation F198Y. Indeed, the V186M mutation, 4266
DOI: 10.1021/acs.biochem.6b00295 Biochemistry 2016, 55, 4263−4274
Article
Biochemistry
variant and the protonation state of the chromophore. It is has already been shown for IFP1.43 that fluorescence ceases below 4 but reaches its maximum at around pH 6, and this has been confirmed for IFP1.1, IFP1.2, and IFP2.0 (Figure S3). The range of pH values between these two values is associated with a sharp increase in fluorescence. Therefore, the protonation state of His207 must control the fluorescence of the chromophore. Although the protonatable side chain of His207 is oriented toward the solvent, its different protonation states may result in different hydrogen bond networks that interact with the chromophore. As we will show in the molecular dynamics section, changing the protonation states of both histidines, His260 and His207, leads to different interaction patterns between water molecules and the region of the chromophore-binding pocket. Because some of the proteins, namely, DrCBD and variants IFP1.0−IFP1.2, occur as dimers, we also performed comparative electrostatic calculations using the geometries of the complete dimers. From these calculations, we conclude that the addition of the second monomer does not noticeably change the protonation behavior of the residues in the first monomer. Exceptions are limited to a few residues at the interface between the two monomers (Figures S7 and S8). For example, in the monomer, Glu306 has a high probability of being protonated at low pH values, but in the dimer, it always remains deprotonated because of interactions with the nearby Arg141 from the second monomer. Another example is His138, which in the dimer becomes ε-protonated, because it interacts closely with Arg100 from the other monomer. Molecular Dynamics Simulations. Because of the very slight differences in the protonation behavior of titratable residues between the monomers and dimers, we used the geometries of the monomers for our molecular dynamics simulations. We assumed that changes in protein dynamics due to the interactions from the additional monomer would not be sufficiently significant to justify the use of the complete dimers, which would be computationally more costly. Moreover, the dimerization interface and the region around the chromophorebinding pocket, on which we focus in this study, are located on opposite sides of the monomer. For each of the studied proteins, we performed simulations at two values of pH, namely pH 4 and 7. Two separate situations were considered because the fluorescence of IFPs was shown to cease at low pH and reach its maximum near neutral pH, most likely because of the ionization of different groups in the protein. The pH value of a simulation was regulated by adjusting the protonation states of all titratable protein residues in accordance with the previous electrostatic calculations. The protein-bound biliverdin was fixed to be in its fully protonated form regardless of the pH. However, we have also studied the less likely case in which the chromophore is deprotonated at its B pyrrole ring (see sections S2−S4 of the Supporting Information). Statistical Distribution of the Important Dihedral Angles. During the simulations, we traced the values of the six dihedral angles of the chromophore located between the pyrrole rings (see Figure 3 for the definitions and numbering of the dihedrals). Finally, we calculated and compared statistical distributions of the dihedrals in different protein variants. Figure 4 shows the plotted distributions, and Table 3 lists the fitted normal distribution parameters μ and σ, describing the mean value of the dihedral and its spread, respectively, calculated from the last 5 ns of the 10 ns molecular dynamics
Figure 3. Close-up of the fully protonated biliverdin chromophore as seen in the binding pocket of DrCBD (crystal structure with added and geometry-optimized hydrogens). Arrows indicate rotations around central bonds of the crucial dihedrals that define the planarity and rigidity of the chromophore. Green circles indicate individual pyrrole rings of the chromophore. The pyrrole water is shown in the middle. The dashed line indicates the point of covalent binding of the chromophore to Cys24.
chromophore have been experimentally confirmed,40,41 reinforcing the notion that one of the pyrrole protons may be quite labile. In this paper, we focus primarily on the situation in which the chromophore is protonated at all pyrrole rings in the whole spectrum of pH. For the sake of completeness, however, we have also examined the less probable scenario in which the chromophore is deprotonated at one of the pyrrole rings at neutral pH. We chose the B-deprotonated form based on the comparison of gas-phase energies and geometries of different protonation forms of the chromophore. The rationale behind our approach was that the fluorescence of IFP1.4,3 as well as the other proteins (Figure S3), was shown to be highly pHdependent. The fluorescence is at a maximum in the pH range of 5.5−6.5 but decreases at pH
