Effects of local protein environment on the binding of diatomic molecules to heme in myoglobins. DFT and dispersion-corrected DFT studies
- 239 Downloads
The heme-AB binding energies (AB = CO, O2) in a wild-type myoglobin (Mb) and two mutants (H64L, V68N) of Mb have been investigated in detail with both DFT and dispersion-corrected DFT methods, where H64L and V68N represent two different, opposite situations. Several dispersion correction approaches were tested in the calculations. The effects of the local protein environment were accounted for by including the five nearest surrounding residues in the calculated systems. The specific role of histidine-64 in the distal pocket was examined in more detail in this study than in other studies in the literature. Although the present calculated results do not change the previous conclusion that the hydrogen bonding by the distal histidine-64 residue plays a major role in the O2/CO discrimination by Mb, more details about the interaction between the protein environment and the bound ligand have been revealed in this study by comparing the binding energies of AB to a porphyrin and the various myoglobins. The changes in the experimental binding energies from one system to another are well reproduced by the calculations. Without constraints on the residues in geometry optimization, the dispersion correction is necessary, since it improves the calculated structures and energetic results significantly.
KeywordsCarbon monoxide DFT calculations Dispersion correction Myoglobins Oxygen
As respiratory proteins, Mb and Hb must favor the binding of O2 compared to CO and NO to avoid suffocation; the latter two molecules are ubiquitous in biology. The binding affinity of CO to an iron porphyrin is 105-fold larger than that of O2. However, this ratio is reduced by several orders of magnitude when the heme is embedded in the protein matrix . Clearly, the intermolecular interactions by the surrounding protein polypeptide assist in the discrimination against endogenous CO inhibition.
The underlying mechanism of ligand discrimination in Mb has attracted much interest from both experimentalists and theorists. It was once a central point of view in the field that the bent geometry of CO observed in the previous X-ray  and neutron  crystal structures of MbCO, is the major cause of the reduced affinity for CO. This is a steric hindrance mechanism suggested by Collman et al.  more than 35 years ago, who argued that the strategically placed distal histidine residue (His64) in the protein pocket, positioned over Fe, could accommodate the bent O2, but would inhibit binding of the normally upright CO. However, later X-ray crystal structures of MbCO [5, 6] show very little bending angles (relative to upright position), and sophisticated infrared experiments [7, 8] have ruled out the possibility of severe FeCO bending, as have the different computations [9, 10]. Many site-directed mutagenesis studies  also do not give a result consistent with this hypothesis. It is now generally accepted that the earlier structures [3, 4] and idea (a strongly bent FeCO unit as the reason for the discrimination)  were incorrect.
An alternative explanation of this difference is centered on a stabilizing hydrogen bond (H-bond) interaction between the distal histidine and the bound ligand (see Fig. 1b). It is suggested that the distal histidine side chain, having an acidic NH group, can H-bond more favorably with bound O2 than with CO. This idea has been supported by neutron diffraction , infrared (IR) spectroscopy , NMR , and theoretical [14, 15] studies, which demonstrated an important influence of polarity on the binding of O2, but not CO, to the heme group.
There have been many theoretical studies of the interaction of AB (AB = CO, O2, NO) with a model heme. Iron porphine (FeP) with an axial ligand L such as imidazole (Im), pyridine (Py), or NH3 has been used to mimic the heme. Its AB complex FeP(L)(AB) is therefore used as model for heme(AB). Very early theoretical studies (e.g., refs. [16, 17, 18, 19]) were performed with semiempirical, Hartree-Fock, or Xα methods at a fixed geometry. Those studies were limited to a qualitative analysis based on orbital energies, populations, and atomic charge distributions, etc. Later, density functional theory (DFT) methods have been applied to FeP(L)(AB) (e.g., refs. [14, 15, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33]). Several recent DFT studies [14, 15, 25, 26, 27, 28, 31, 33] have included the protein environment in the calculations. Spiro and co-workers  considered a large heme(AB) model system that includes the 13 closest, surrounding residues. They compared Mb to H64I, but just reported the relative energies between Mb-AB and H64I-AB (AB = CO, O2) (see Experimental data and previous calculated results section). Strickland et al.  calculated the Fe-CO binding energy in MbCO with a quantum mechanics/molecular mechanics (QM/MM) method; Mb was shown to weaken the Fe-CO bond by 5.8 kcal mol−1, which is significantly larger than experiment (∼1.15 kcal mol−1, see Experimental data and previous calculated results section). Shaik and co-workers  performed a QM/MM study of MbO2, focusing mainly on the effects of the protein on the nature of the Fe-O2 bonding. In the previous QM/MM calculations [14, 15, 25, 28, 31], the QM region included at most one residue (His64) in addition to FeP(Im)(AB), thereby neglecting the polarization effects from other surrounding residues and the strain energy stored in the protein itself . More recently, Cole et al.  used a rather approximate linear-scaling DFT + U method to investigate the ligand discrimination in Mb; however, the accuracy of the calculated results depends greatly on the choice of the adjustable parameter U.
What is the precise interaction energy between the bound AB and the distal protein environment? This energy is unknown precisely from experiment or previous quantum chemical calculations. What is known from experiment is either the free energy of MbAB deduced from equilibrium constant data  or the enthalpy from calorimetry or temperature variations of the equilibrium constant . The interaction energy between bound AB and the distal histidine-64 (His64) has been evaluated with DFT methods [14, 15, 26], but the effects of the other heme pocket components on the bound AB are unknown. Thus, we have question (2):
In addition to His64 (see Fig. 1b), how large is the energetic contribution from the other neighboring residues to the binding of AB to the heme? The previous DFT studies of heme(AB) that considered the protein environment have mainly included a single His64 in the calculations. The energetic contribution from the other neighboring residues has not been investigated in detail.
- (3)Can the experimental kinetic data (ligand dissociation rate constant kAB) for various mutants be exactly correlated with the calculated ligand binding energies [Ebind(AB)] for a residue substitution in the globins? There is a relationship between the change (Δ) in Ebind(AB) and the change in kAB: ΔEAB = kTΔln kAB  (see Table 1). Although experimental studies [34, 35, 36, 37, 38] yield very useful information about the selectivity of Mb, it is not always possible to provide an explanation of the results obtained for the various mutants.
Por in benzenec
Por (in vacuum)
In this work, both DFT and dispersion-corrected DFT calculations have been carried out to investigate the heme-AB binding energies (AB = CO, O2) in a wild-type Mb and two mutants (H64L, V68N) of Mb, where H64L and V68N represent two different, opposite situations. The His64 to leucine (Leu) substitution causes a large decrease in O2 affinity, while there is a significant increase in O2 affinity when the valine-68 in Mb is replaced with asparagine (Asn). It would be of interest to compare the two situations.
Although DFT has proven to be efficient in calculations on heme-AB complexes, it is argued [42, 43, 44, 45] that the DFT methods in common use today do not properly describe the long-range dispersion interactions. This is because the present DFT methods are thought to be “local” or at most “semilocal” theories , even if they include gradients or higher-order derivatives; but dispersion is a long-range, nonlocal electron correlation effect. In recent years, there has been considerable interest in overcoming the lack of dispersion forces in standard DFT and a number of dispersion correction approaches have been proposed in the literature . One most practical and successful approach in the field has been adding a (semi-)empirical correction of the form C6R−6 to a density functional scheme to yield a DFT + Edisp model (it is denoted as DFT-D by some other authors ). Meanwhile, many modifications of the DFT + Edisp approach have also been made (see DFT and dispersion-corrected DFT methods section).
Recently, we assessed the performance of DFT + Edisp for several large biological systems . Three versions of Grimme’s dispersion correction methods  were tested; they are labeled as DFT-D1, -D2, and -D3, respectively (see DFT and dispersion-corrected DFT methods section). DFT-D1 was shown to yield structures and energetic results which are adequate. In contrast, the -D2 and -D3 approaches place the residues too close to the heme and their calculated relative binding energies are in poor agreement with experiment in most cases, but the newer -D3 version is an improvement over the old -D2 one. More recently, a revised version of DFT-D3 was reported by Grimme et al. . It uses the rational damping proposed by Becke and Johnson (see DFT and dispersion-corrected DFT methods section). This variant, labeled as -D3(BJ), has been implemented in the updated ADF program. The current ADF also includes the density-dependent dispersion correction (dDsC) method developed by Steinmann and Corminboeuf  (see DFT and dispersion-corrected DFT methods section).
Considering the fact that the residues around the heme(AB) are linked to the protein backbone, constraints are imposed on the residues in the geometry optimization here. In the previous work , the distal residues above the heme plane were allowed to move freely in geometry optimization. A comparison is made between the results obtained with and without the constraints in geometry optimization.
In addition to exploring the effects of the local protein environment in the various myoglobins, the present work examines the specific role of His64 in the distal pocket of Mb in somewhat more detail than other studies in the literature.
The more recently developed -D3(BJ) and -dDsC dispersion correction methods are tested on some of the present systems.
Experimental data and previous calculated results
(histidine-64 → isoleucine) mutated myoglobin.
(histidine-64 → leucine) mutated myoglobin.
(valine-68 → asparagine) mutated myoglobin.
Three sets of experimental binding energies are given for CO/O2 to a porphyrin. The first set (0.78 eV for CO and 0.53 eV for O2) refers to the measured dissociation barriers for chelated protoheme dissolved in benzene . The binding energies of the second set (0.79 eV for CO and 0.55 eV for O2), which are given also for chelated protoheme in benzene, are estimated based on the measured dissociation rate constants kAB and on the formula ΔEAB = kTΔln kAB ; they are shown to be very close to those of the first set. The third set refers to the dissociation barriers for Mb, corrected for the absence of the protein environment ; they should represent the binding energies of AB to a porphyrin in vacuum, and are seen to be notably different from those for chelated protoheme in benzene. Compared to the values for Por in vacuum, the binding energy of O2 to Mb is increased by 0.26 eV, while the binding energy of CO to Mb is reduced by as much as 0.05 eV. Therefore, the discrimination energy between O2 and CO by Mb should be as large as 0.31 eV, considerably higher than 0.18 eV that is usually cited in the literature . The values for Por in benzene may not be best suited for comparison with computational results because they are not measured in the gas phase. It is known that the binding properties of AB to chelated protoheme can be different when the porphyrin is dissolved in different solvents .
Calculated Mb-AB binding energies (Ebind, eV) and MbAB⋅⋅⋅His64 hydrogen-bond energies (EHbond, eV) from the literature
FeP(NH3)(AB) with one His64
FeP(Im)(AB) with one His64 + surrounding point charges (QM/MM method)
Heme(AB) with one His64 (QM/MM method)
Heme(AB) with 13 surrounding residues
Spiro and co-workers  carried out DFT computations on a large heme(AB) model system that includes the 13 closest, surrounding residues. They compared Mb to H64I, but just reported the relative energies between Mb-AB and H64I-AB (AB = CO, O2); no information is provided about the absolute binding energies for Mb-AB and H64I-AB, nor is comparison made between Por-AB and Mb-AB.
After the above brief review of the experimental data and previous calculated results, we report, in the following, both our DFT and dispersion-corrected DFT calculations for the AB ligands on Por, Mb, H64L, and V68N. Recent calculations [54, 55] performed with the DFT + Edisp technique for weakly bound systems have been very promising; the new method is expected to yield improved quantitative results.
Two optimization procedures are used: (1) Considering the fact that the residues around the heme(AB) are anchored to the polypeptide in the protein matrix, constraints are imposed on the residues in the geometry optimization; i.e., the terminal amino nitrogen atoms are fixed according to the crystal structure. (2) The distal residues above the heme plane are allowed to move freely in geometry optimizations. It would be of interest to give a comparison of the results of the two situations.
Comparison of the calculated Mb-O2 binding energies (Ebind, eV) based on the crystal structures of the MbCO and MbO2 proteins (i.e., in MbCO and MbO2)a
In MbCO (code 1BZR)
In MbO2 (code 1A6M)
in V68N(CO) (code 1M6C)
in V68N(O2) (code 1MNO)
The systems used for both H64L(CO) and H64L(O2) are based on the crystal structure of H64L(CO) (code 2MGC) . We chose H64L rather than H64I because the crystal structure of the CO complex of H64L is available, on which the calculations can be based directly. (H64I was used by Spiro and co-workers  in their calculations, where His64 in Mb was manually replaced by isoleucine.) However, no crystal structure is available for native H64L(O2) to the best of our knowledge. We suppose here that the structure of the O2 complex is also similar to that of the CO complex, with the exception of the Fe-O-O angle, which is ca. 120°.
DFT and dispersion-corrected DFT methods
All calculations used the Amsterdam Density Functional (ADF) program package - ADF2012.01 [58, 59]. Three density functionals were used in the calculations. They are BP (which contains Becke’s 1988 gradient correction for exchange  plus Perdew’s 1986 gradient correction for correlation ), revPBE (revised Perdew-Burke-Ernzerhof functional  proposed in 1998 by Zhang and Yang ), and B3LYP (Becke’s 1993 three-parameter hybrid functional  using the Lee-Yang-Parr correlation functional ). The heme-AB binding energies prove to be difficult to predict theoretically. We recently tested a large number of density functionals on the FeP(Im)(AB) systems (Im = imidazole; AB = CO, NO, O2) and found that relatively satisfactory results for the various FeP(Im)-AB binding energies were obtained with the meta-GGA functionals BLAP3 and Bmτ1 . But in the ADF program, these functionals are treated in a non-self-consistent (non-SCF) manner and so they are not suited for the present calculations on the rather large systems here (because a very high level of numerical-integration accuracy needs to be set in the non-SCF calculations of binding energies  and this is computationally too expensive for large systems). Among the various functionals tested, B3LYP and revPBE provided excellent binding energies for the AB = CO and O2 ligands, respectively (B3LYP for CO and revPBE for O2), and so they are adopted here.
Three versions of Grimme’s dispersion correction have been implemented in ADF; they are labeled  as DFT-D1, -D2, and -D3, respectively. The damping functions in the different versions are given in Supplementary material.
More recently, a revised version of DFT-D3 was reported by Grimme et al. . It uses the rational damping proposed by Becke and Johnson  (see Supplementary material). The variant that uses Becke and Johnson’s damping, labeled as -D3(BJ), has been implemented in the updated ADF program (ADF2012.01). The current ADF also includes the density-dependent dispersion correction (dDsC) method developed by Steinmann and Corminboeuf [50, 68] (Supplementary material).
The relatively early -D1 version  was designed mainly for noncovalent interactions between molecules, assuming that within a covalent molecule, the effects of correlation are covered well by the particular exchange-correlation (XC) functional (although it may be argued that in the DFT + Edisp approach, the dispersion part is damped away at shorter distances and should not affect these). This version has been shown to give good results [45, 47, 55]. -D2  is a slightly modified version of -D1, following a strategy in which the DFT description is restricted to shorter electron correlation ranges and to describe the medium to larger ranges by damped C6R−6 terms. Thus, this newer version can be used for normal molecules. However, DFT-D2 has the tendency of overbinding at shorter intermolecular distances, probably because of a double-counting of correlation effects . The -D3 version was developed recently  and contains many modifications and new features. In this work, the -D1, -D3, -D3(BJ), and -dDsC dispersion correction methods have been used.
The Slater-type orbital (STO) basis set used is the standard ADF-TZP, which is a triple-ζ plus one polarization function set. Frozen-core techniques  were used here to reduce the computational cost. To obtain accurate results, the valence set on Fe included sub-valence 3s and 3p shells. For C, N, and O, 2s and 2p were considered as valence shells. The other shells of lower energy, i.e., [Ne] for Fe and [He] for C/N/O, were described as core and kept frozen. By using the large TZP basis sets, the basis set superposition error (BSSE) was found to be small and could be negligible1.
Relativistic corrections for the valence electrons were calculated by the quasi-relativistic (QR) method . For the open-shell states, the unrestricted Kohn-Sham (KS) spin-density functional approach was adopted.
Results and discussion
Calculated FeP(4-EtIm) − AB binding energies (Ebind) at the BP optimized structure and at the structure optimized with the respective individual functional itself
Ebind[FeP(4-EtIm) − AB], eV
AB = CO
AB = O2
The calculated structural parameters of interest are all provided in Supplementary material, together with available experimental (X-ray) crystal structural data. Since the deoxy forms of myoglobin are relevant to the myoglobin-AB binding energies (see Binding energies of AB to porphyrin and the various myoglobins section), the Supplementary material also presents the calculated structural parameters of FeP(4-EtIm) without the AB ligand.
The d6 FeII iron in deoxyHeme can exhibit three spin states, namely S = 0 (low spin, singlet), S = 1 (intermediate spin, triplet), and S = 2 (high spin, quintet). The geometry optimization was performed for each spin state. Three critical coordination parameters of deoxyHeme are RCt(4N)⋅⋅⋅N(eq) (distance between the center of the porphyrin ring and the equatorial, porphinato nitrogen), RCt(4N)⋅⋅⋅Fe (distance between the center of the porphyrin ring and Fe), and RFe-N(ax) (axial Fe-imidazole bond length). RCt(4N)⋅⋅⋅N(eq) is a measure of the porphinato core size and RCt(4N)⋅⋅⋅Fe represents the displacement of Fe out of the 4N-plane toward the axial imidazole ligand. Binding an AB ligand to the five-coordinate complex moves the Fe back into the plane. But RCt(4N)⋅⋅⋅Fe in heme(AB), though small, is not zero experimentally and may be different in different compounds . Hence, we have also presented the calculated RCt(4N)⋅⋅⋅Fe values for each heme(AB). The other structural parameters of interest for a heme(AB) moiety include RFe-AB (the axial Fe-AB bond length), RA-B (the A-B bond length), and ∠FeAB (the Fe-A-B bond angle). Heme(CO) has a closed-shell singlet ground state; the strong field of the CO ligand makes the six-coordinate system low spin. In the case of AB = O2, the ligand field is relatively weak. There is an electron transfer from Fe to O2, yielding an FeIII − O2− unit in the complex. Therefore, the ground state of heme(O2) is either a triplet (S = 1) (dxy)2(dxz)2(dyzα)1(O2-πg*α)1 or an open-shell singlet (S = 0) (dxy)2(dxz)2(dyzα)1(O2-πg*β)1. According to our calculations, the triplet is somewhat lower in energy than the open-shell singlet for this complex. The ground state of FeP(L)(O2) is shown to be different with different L ; a weaker Fe-L bond more likely leads to an open-shell singlet ground state for the oxy complex.
For the external structure of heme in a myoglobin, we give the distances between each of the Fe-A-B atoms and their nearest neighboring C (or N) and H atoms from each residue. These atoms are shown in Fig. 3 and indicated with big balls.
FeP(4-EtIm) and FeP(4-EtIm)(AB)
The calculated structural parameters of deoxyHeme can be rather different for different spin states. From Table S1 (Supplementary material), we see that RCt(4N)⋅⋅⋅N(eq) (core size) and RCt(4N)⋅⋅⋅Fe (Fe out-of-plane displacement) for S = 2 are significantly larger than those in the lower-spin states. The experimental data used to compare with FeP(4-EtIm) are those measured for deoxyMb, which is high spin. We should point out here that the X-ray crystal structural data for the five-coordinate, synthetic FeTPP(2-MeIm) and FeTpivPP(2-MeIm) compounds are available in the literature . 4-EtIm is similar to 1-MeIm, but rather different from 2-MeIm; see ref.  for more details. Concerning FeP(4-EtIm)(AB) (Table S2), we have X-ray structural data on comparable, synthetic FePor(L)(AB) compounds (here Por = TPP or TpivPP) .
For both FeP(4-EtIm) and FeP(4-EtIm)(AB), there is good quantitative agreement between the pure DFT calculations and experimental structures. The dispersion correction results in a shortening of the bonds between Fe and surrounding atoms; the axial Fe-N(ax) distance is decreased by 0.03–0.05 Å. But the other bonds Fe-N(eq) and Fe-AB are shortened only slightly by the dispersion correction. There are no notable differences among the DFT-D3, -D3(BJ), and -dDsC optimized structural parameters.
According to the RCt(4N)⋅⋅⋅Fe values, there is a notable out-of-plane displacement of Fe even in the six-coordinate system for AB = CO, where RCt(4N)⋅⋅⋅Fe amounts to −0.03 to −0.04 Å. The negative value just denotes that Fe is displaced toward the AB ligand. Fe in FeP(4-EtIm)(O2) can be thought to lie in the porphyrin 4N plane because RCt(4N)⋅⋅⋅Fe in this complex is about −0.01 Å. The trend in RCt(4N)⋅⋅⋅Fe from AB = CO to AB = O2 is consistent with the order of the binding strengths of the AB ligands to Fe (see Binding energies of AB to porphyrin and the various myoglobins section).
Except for ∠FeAB, the experimental structural parameters of FePor(L)(AB) are in general quite close to those measured for MbAB. The Fe-A-B bond angle is shown to be smaller in MbAB than in FePor(L)(AB), which can be attributed to the interaction of bound AB with the protein environment in MbAB.
From Por(AB) to MbAB, all the calculated R values undergo little changes. But the situation for ∠FeCO is different. The calculated Fe-C-O angles in MbCO vary from 174° to 179°, depending on the method used. The experimental Fe-C-O angle in the high-resolution MbCO crystal structure (code 1BZR) from ref.  is 171°, comparable to the DFT and DFT-D1 calculations without constraints on the residues in geometry optimization. The corresponding change in the calculated Fe-O-O angles is less pronounced; they vary from 129° to 131° and are about 3–5° smaller than those calculated in the pure porphyrin (∼134°). The crystal structural data of MbO2 (code 1A6M) from ref.  are probably less accurate, which give a significantly smaller Fe-O-O angle (123°) than the calculations. Within the internal structure of heme, the DFT and DFT + Edisp optimized parameters are quite similar. That is, the dispersion effects between the heme and the protein environment do not notably affect the internal structure of the heme.
However, the external structure of heme in the myoglobin can be changed very much by dispersion effects. Owing to the attractive property of the dispersion interaction, the DFT + Edisp calculations give, in most cases, considerably shorter distances between the FeAB atoms and the protein residues than the DFT ones. As expected, the calculations with pure DFT greatly overestimate the intermolecular distances in MbAB, even when the terminal amino nitrogen atoms are frozen in the geometry optimization. Examining the tables in Supplementary material, we see that the distances optimized with DFT-D1 compare in most cases favorably with the X-ray crystal structural data for each of the MbAB, H64L(AB), and V68(AB) systems. With the DFT-D1 method, moreover, the geometry optimizations with and without constraints on the residues give similar structures, indicating that the modeled heme site is quantum mechanically stable. In a few cases for MbO2, H64L(CO), or V68N(AB), the DFT-D1 optimized distances are still significantly larger than the experimental ones, even when dispersion corrections are taken into account. It should be pointed out that the crystal structural data of a myoglobin measured with relatively low resolution may not be accurate enough. For example, the experimental Fe-C-O angle of 156° for the H64L(CO) structure (code 2MGC) cannot be considered to be correct (see next subsection). Various factors may influence the accuracy of the crystal structural data of a myoglobin2. Nevertheless, we see that the experimental distances for MbCO measured with high resolution (code 1BZR) are very well reproduced by the calculations with DFT-D1.
On going from DFT-D1 to DFT-D3, the optimized intermolecular distances are further shortened and they are now too short for Phe43, Ile10, and Leu29 as compared to experiment. The optimization with constraints on the residues (i.e., constrained optimization) improves the results in some cases. DFT-D3(BJ) gives much longer distances than DFT-D3 for Leu29 and so improves the agreement between the calculation and experiment, but this is true only when the constrained optimization is performed. With the constraints on the residues, the DFT-dDsC optimized structural parameters are comparable to those obtained with DFT-D3(BJ). Without constraints, DFT-dDsC appears to yield better results than DFT-D3(BJ) or DFT-D3 for the structure.
The structure of the Mb moiety in MbAB is changed notably when CO is replaced with O2. The distal His64 group in MbO2 comes so close to the O2 molecule that it can form a fairly strong H-bond to the oxygen bound to Fe. This leads to an elongation of the other Fe⋅⋅⋅residue distances in MbO2.
In this mutant, the Fe-C-O bond angles calculated with every method are all about 178°; their deviation from linearity is only ca. 2°. This appears to be smaller than that in MbCO (∼6°). Nevertheless, the internal structure of the heme changes little from MbCO to H64L(CO) according to the calculations. The same is true for AB = O2. But there are some differences in the positions of the residues between the MbCO and H64L(CO) structures; the Fe⋅⋅⋅residue distances are usually shorter in the former than in the latter. However, there are no significant differences in the Fe⋅⋅⋅residue distances between MbO2 and H64L(O2) according to the calculations.
When the optimized structure of H64L(CO) is compared to the X-ray crystal structure of the mutant, large differences can be seen. The measured Fe-C-O angle is very bent (156°) and there is a large out-of-plane displacement of Fe toward the axial His93 ligand. The other measured R values within the heme are also significantly larger than the calculated ones, and they indicate that heme(CO) has some character of deoxyHeme. The X-ray crystal structural data for H64L(CO) cannot be considered to be reliable. No native crystal structure is available for H64L(O2).
When Val68 in Mb is replaced with Asn, the linearity of the FeCO unit is restored. Val68 in MbCO is in close van der Waals contact with the bound ligand and has been suggested to be responsible for the small distortion of Fe-C-O from linearity . Certain calculations support this argument. We note in the above section that the bending of Fe-C-O is smaller in H64L(CO) than in MbCO. This is because the His64 → Leu substitution makes Val68 shift away from the heme center; theVal68 carbon atom in H64L(CO) is at longer distance from the bound ligand than in MbCO. As a result, the steric hindrance from Val68 to the bound CO is weakened from MbCO to H64L(CO).
The V68N mutation involves the substitution of the isopropyl side chain of Val68 by the acetamide side chain of Asn, a replacement which is expected to alter the polarity of the distal heme pocket in myoglobin. The orientation of the Asn68 side chain is well defined with its -NH2 group toward the bound CO; such an orientation is expected to favor oxygen binding by providing an additional H-bond to the highly polar Fe-O2 complex.
A notable feature of the V68N(CO) crystal structure is that the distal His64 side chain has swung out of the pocket away from the CO ligand (see Fig. 3c). However, such an upward movement of His64 does not occur in the structure of native V68N(O2) (Fig. 3d). But one part of the experimental V68N(O2) structure appears to be unusual: Fe is displaced greatly to the proximal side; the Ct(4N)⋅⋅⋅Fe distance is as large as 0.13 Å. According to the calculations, however, the Fe atom is almost in the heme plane for the six-coordinate O2 complex, there being no difference from the position in MbO2 and H64L(O2).
Binding energies of AB to porphyrin and the various myoglobins
Calculated relative energies (Erelative, eV) for selcted states of FeP(4-EtIm) and deoxyMb
S = 1
S = 0
S = 2
S = 1
S = 0
S = 2
S = 1
S = 0
S = 2
Calculated binding energies (Ebind, eV)a of CO and O2 to pure porphyrin (Por) [i.e., FeP(4-EtIm)] and to the models of wild-type Mb and two mutants (H64L, V68N) of myoglobin with DFT and DFT-D1
Calculated relative binding energies (ΔE, eV) between myoglobin-AB and Por-AB, ΔE(myoglobin-AB/Por-AB)
Some other calculated relative binding energies (ΔE, eV) and the evaluated discrimination energies (ΔΔE, eV)
ΔΔE (discrimination E)a
ΔΔE (discrimination E)b
A comparison of the calculated energies (Ebind, ΔEbind) with DFT and the various DFT + Edisp methods (−D1, -D3, -D3(BJ), -dDsC) are reported in Supplementary material. -D3(BJ) and -dDSc have only been tested on Por(AB) and MbAB (AB = CO, O2). We have shown in Structure section that the DFT-D3 approach gives too short distances between the residues and the heme moiety in the myoglobins. This may imply that the dispersion energies in these large biological molecules are probably significantly overestimated by DFT-D3. As a result, its calculated relative energies are in poor agreement with experiment in most cases. Some improvements of the results are obtained with DFT-D3(BJ) when constraints on the residues are imposed in the geometry optimization. With this optimization procedure, DFT-dDSc gives better results than DFT-D3(BJ) for ΔEbind(Mb-O2/Por-O2), but provides poorer results for ΔEbind(Mb-CO/Por-CO). Without constraints on the residues in the geometry optimization, the DFT-dDsC approach is shown to yield very poor results for both ΔEbind(Mb-O2/Por-O2) and ΔEbind(Mb-CO/Por-CO).
In contrast, the DFT-D1 approach yields the model structures which agree very well with the X-ray crystal structural determinations in high resolution, and its geometry optimizations with and without constraints on the residues give similar structures. As a results, the DFT-D1 calculated energies with the different optimization procedures are similar as well and agree well with the experimental ΔEbind data. In the following, the effects of the local protein environment are discussed mainly based on the results obtained with DFT-D1 and with the constrained optimization, unless otherwise stated.
For comparison, Por-AB is also of interest as it is the reference point for discussing the effects of the local protein environment on the heme-AB binding. The calculated binding energies Ebind(Por-AB) with the BP functional are 1.56 and 0.71 eV for AB = CO and O2, respectively3. They are larger than the experimental data (CO: 0.85 eV, O2: 0.44 eV) by 0.71 and 0.27 eV, respectively for the two ligands. The revPBE functional performs much better than BP and gives Ebind(Por-O2) which is in excellent agreement with experiment. Nevertheless, this functional still overestimates Ebind(Por-CO) by 0.46 eV. While the pure GGA functionals seriously overbind for AB = CO, the hybrid functional B3LYP clearly provides a superior result for this ligand (0.81 eV). But it gives too small a binding energy for AB = O2 (0.04 eV).
When the dispersion correction is made, the binding energies obtained with DFT-D3 increase by ca. 0.14, 0.24, and 0.26 eV for the BP, revPBE, and B3LYP functionals, respectively (see Supplementary material). The dispersion contributions to Ebind are similar for AB = CO and O2. Since BP and revPBE already give too large a binding energy for Por-CO and revPBE works very well for Por-O2, the addition of the dispersion term causes the performance of the two functionals to deteriorate. The DFT-D3(BJ) method gives similar results to those of DFT-D3 for AB = CO; but there is a reduction of 0.05–0.1 eV in the calculated Ebind(Por-O2) on going from -D3 to -D3(BJ). Some increase, ranging from 0.02 to 0.16 eV depending on functional, in the DFT-dDsC calculated Ebind(Por-AB) is obtained as compared to DFT-D3.
For the B3LYP functional, the dispersion term greatly improves the result for Ebind(Por-O2). The repulsive behavior of B3LYP for Por-O2 seems to indicate inadequate treatment of dispersion effects in DFT, but the dispersion correction shows deficiency when this functional is applied to Por(CO). It would be wrong to assume that the interaction energy obtained with these functionals is dispersion-free; some form of compensation for the missing component (dispersion) of the interaction energy must exist.
We now put heme(AB) in Mb and examine the influence of the local protein environment on the heme-AB binding. First looking at the results obtained with pure DFT, the calculated Mb-CO binding energies are 1.52, 1.25, and 0.77 eV for the BP, revPBE, and B3LYP functionals, respectively; they are 0.04–0.06 eV smaller than those for Por-CO, the ΔEbind values being close to the experimental one (−0.05 eV). Without constraints on the residues in geometry optimization, the relative binding energies between Mb-CO and Por-CO would be −0.11 to −0.15 eV; the constrained optimization improves the results of pure DFT. When a dispersion correction is made, there are little changes in Ebind(Mb-CO) for BP and B3LYP. But the DFT-D1 calculation with revPBE gives an Mb-CO binding energy which is similar to the Por-CO one. It is shown that the DFT-D1 calculated Ebind(Mb-CO) values with and without constraints on the residues in geometry optimization, are almost the same; the same is true for Mb-O2. This is somewhat different from the case of pure DFT.
Calculated dispersion energies (in eV) between the heme(AB) moiety and the surrounding residues in AB-myoglobins, Edisp[heme(AB) − (residues)], and the dispersion energies between the deoxyHeme and the surrounding residues in deoxy-myoglobins, Edisp[deoxyHeme − (residues)]
Edisp[heme(AB) − (residues)]
Edisp[deoxyHeme − (residues)]
AB = CO
AB = O2
S = 1
S = 0
S = 2
in Mb (code 1BZR)
in H64L (code 2MGC)
in V68N (code 1M6C)
To examine the specific role of His64 in the distal pocket, additional calculations were also performed on a model system that includes only His64 (case 1) and on another model system that excludes it from the residues (case 2). In both cases, the DFT and DFT-D1 calculated Mb-CO binding energies are very similar and close to the Por-CO one. For AB = O2, however, the two cases give rather different results. In case 1, the calculated Mb-O2 binding energy is similar to that for the system that has all five residues; it is therefore significantly larger than the Por-O2 one. In case 2, however, the exclusion of His64 in the system leads to a large (∼0.15 eV) decrease in Ebind and the calculated Mb-O2 binding energies are only slightly larger than the Por-O2 ones. These results further support the previous conclusions  which show that His64 provides the dominant electrostatic stabilization for bound O2 and the contributions of the other residues to the discrimination between O2 and CO are small. The dispersion effects in case 1 are also found to destabilize the bound O2. We note that the calculations on such a model do not give the experimental trend that the binding energy is lowered from Por-CO to Mb-CO; they give an opposite trend instead.
According to the experimental data (Table 1), the heme-CO binding energy is slightly lowered by 0.01 eV when His64 in Mb is replaced with Leu. The DFT and DFT-D1 calculations with BP show that Ebind(H64L-CO) is 0.01–0.02 eV smaller than Ebind(Mb-CO) (see Table 8), in good agreement with the experiment value. The calculations with the other functionals yield slightly large binding energies for H64L-CO than for Mb-CO. Without constraints on the residues in geometry optimization, the ΔEbind(Mb-CO/H64L-CO) values calculated with pure DFT are rather negative, in strong disagreement with experiment. However, the experiment-theory agreement is improved greatly by making a dispersion correction.
In the case of AB = O2, the decrease in Ebind from Mb-AB to H64L-AB is significant, 0.14 eV according to experiment, 0.12–0.18 eV from the DFT calculations, and ∼0.10 eV based on the DFT-D1 calculations. Somewhat large ΔEbind(Mb-O2/H64L-O2) values are obtained without constraints on the residues in geometry optimization. The protein environment in H64L with His64 replaced by Leu has some stabilizing effect on the bound O2. The experimental relative binding energy of H64L-O2 vs. Por-O2, ΔEbind(H64L-O2/Por-O2), is as large as 0.12 eV, which is in good agreement with the DFT-D1 calculations (0.07–0.14 eV). We note that the calculated H64L-O2 binding energies are not notably different from the Mbno His64-O2 ones.
In contrast to H64L, the binding of AB to heme is enhanced by replacing Val68 in Mb with Asn, whose amide group has the capacity to serve as a H-bond donor (see V68N(AB) section and Fig. 3c or d). According to the calculations, the V68N-CO binding energy is 0.04–0.10 eV higher than the Mb-CO one, depending on the functional used. The experimental relative energy ΔEbind(V68N-CO/Mb-CO) is 0.02 eV. The interaction between Asn and bound CO is somewhat overestimated by the calculations. Again, the constrained optimizations greatly improve the results of the pure-DFT calculations.
The binding energy of O2 to heme is increased by 0.08 eV from Mb to V68N according to experiment. The DFT-D1 calculated relative binding energies ΔEbind(V68N-O2/Mb-O2) are 0.08–0.17 eV, depending on the functional used; they are in reasonable agreement with experiment. The results of revPBE are improved significantly when a dispersion correction is made. On the other hand, the experimental relative binding energy of V68N-O2 vs. Por-O2 (0.34 eV) is well reproduced by the DFT-D1 calculations (0.28–0.41 eV).
The effects of local protein environment on the binding of diatomic molecules (CO, O2) to heme in myoglobins have been studied in detail with both DFT and dispersion-corrected DFT methods. Several dispersion correction approaches, -D1, -D3, D3(BJ), and -dDsC, were tested in the calculations. In the -D1 approach, the dispersion correction (Edisp) is calculated only for noncovalent interactions between molecular fragments and Edisp within a covalent interaction is not calculated. It is shown that DFT-D1 performs very well, ensuring structural and energetic features in close agreement with experiment. Surprisingly, satisfactory energetic results are also obtained with the pure DFT method when constraints on the residues are imposed in the geometry optimization. This may be ascribed to error cancellations in the DFT calculated results.
By calculating and comparing the binding energies of O2 and CO to a porphyrin and the various myoglobins, more details about the interaction between the protein environment and the bound ligand can be revealed. Based on the dissociation barriers for Mb, corrected for absence of the protein environment, the binding energy of O2 to Mb is 0.26 eV (1 eV = 23.06 kcal mol−1 = 96.5 kJ mol−1) increased as compared to that of a free porphyrin, while the binding energy of CO to Mb is reduced by 0.05 eV. The calculated results are in good agreement with both experimental data and the observed trend. Although the protein environment gives a favorable H-bond interaction for bound CO, it destabilizes the heme-CO binding. This is because the Mb-AB binding energy is also related to the energy of optimized deoxyMb, and the non-covalent interaction between the heme and the protein environment is enhanced when AB departs. Olson and Phillips  attributed most of this inhibition to the requirement of water displacement from the distal pocket. (The distal residues often bring a water molecule into the protein, which must be displaced before ligands can bind.) An alternative argument is that there exists repulsive interaction between the His64 side chain and the bound CO ligand . The calculations show that this inhibition results from a change in the interaction between heme and residues with and without the AB ligand. In the case of O2, the very strong H-bond interaction for this ligand over-compensates very much the mentioned enhanced, extra interaction. Therefore, we can still see a great stabilization of bound O2 by the protein environment on going from Por to Mb.
The specific role of histidine-64 in the distal pocket has been examined in more detail in this study than in other studies in the literature. According to the calculations, the stabilization of heme-O2 in Mb by a single His64 is ca. 0.20 eV, whereas the sum of the energetic contributions from the other residues to the heme-O2 binding is about 0.05 eV. The small distortion of the Fe-C-O moiety from linearity in the high-resolution crystal structure of MbCO  is due to the steric hindrance from Val68 to the bound CO. This is consistent with previous calculated results of Spiro and co-workers . The changes in the experimental binding energies from Mb-AB to H64L-AB or to V68N are also well reproduced by the calculations.
The performances of the advanced -D3, -D3(BJ), as well as -dDsC dispersion correction methods appear to be questionable for large biological molecules. These methods take into account intramolecular dispersion contribution for a covalent molecule. There are, however, some disputes as to whether this is desirable , since various approximate DFT methods that include GGA or hybrid GGA have been remarkably successful in describing a wide variety of strongly interacting systems. This may imply that the contribution of the dispersion energy, which is missing from the DFT treatment, is compensated for to a certain degree when DFT methods are applied to covalent systems. But the uncorrected DFT methods have not met equal success in describing weak interactions and a dispersion correction has to be made in this case. The DFT-D1 approach can give satisfactory results not only for small noncovalent systems  but also for large ones [47, 55]. The application of the dispersion-corrected DFT methods to covalent systems is a new subject and may need further investigations. Recently, a steeper damping function was adopted by Chai and Head-Gordon  in their ωB97X-D functional so as to avoid a double-counting of correlation effects at short range; meanwhile the functional is reparametrized in the presence of the empirical dispersion correction. This functional might be useful for the study of the present systems, but it has not been implemented in the ADF program yet.
In fact, we do not need to consider BSSE in the present calculations at all because we do not calculate intermolecular binding energies in this work.
The crystal coordinates are not the experimental raw data; they are a product of an involved series of model building, crystallographic refinement, manual examination, and rebuilding, possibly involving mistakes in interpretation.
Previous calculations with all-electron method (ref. ) or with an increased accuracy of the numerical integration (ref. ) could give FeP(4-EtIm)-AB binding energies which are somewhat larger than the present ones. Owing to error cancellations, the present calculated results are actually in closer agreement with experiment than the previous ones.
This work was supported by Award Number SC1-HL096018 from the National Heart, Lung, and Blood Institute. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Heart, Lung, and Blood Institute or the National Institutes of Health (NIH). The ADF calculations were run on a QuantumCube™ QS32-2800C computer from Parallel Quantum Solutions, LLC.
- 1.Stryer L (1981) Biochemistry, 4th edn. Freeman, New YorkGoogle Scholar
- 10.McMahon MT, deDios AC, Godbout N, Salzmann R, Laws DD, Le H-B, Havlin RH, Oldfield E (1998) An experimental and quantum chemical investigation of CO binding to heme proteins and model systems: a unified model based on 13C, 17O, and 57Fe nuclear magnetic resonance and 57Fe moessbauer and infrared spectroscopies. J Am Chem Soc 120:4784–4797CrossRefGoogle Scholar
- 44.Hesselmann A, Jansen G, Schütz M (2005) Density-functional theory-symmetry-adapted intermolecular perturbation theory with density fitting: a new efficient method to study intermolecular interaction energies. J Chem Phys 122:14103–14120Google Scholar
- 51.Springer BA, Egeberg KD, Slighar SG, Rohlfs RJ, Mathews AJ, Olson JC (1989) Discrimination between oxygen and carbon monoxide and inhibition of autooxidation by myoglobin. Site-directed mutagenesis of the distal histidine. J Biol Chem 264:3057–3060Google Scholar
- 59.ADF2012.01. SCM, Theoretical Chemistry, Vrije Universiteit, Amsterdam, Netherlands. See also (http://www.scm.com)
- 71.Jameson GB, Molinaro FS, Ibers JA, Collman JP, Brauman JI, Rose E, Suslick KS (1980) Models for the active site of oxygen-binding hemoproteins. Dioxygen binding properties and the structures of (2-methylimidazole)-meso-tetra(alpha, alpha, alpha, alpha-o- pivalamidophenyl)porphyrinatoiron(II)-ethanol and its dioxygen adduct. J Am Chem Soc 102:3224–3237CrossRefGoogle Scholar