Using Molecular Dynamics to Probe the Structural Basis for Enhanced Stability in Thermal Stable Cytochromes P450

Abstract

High temperature molecular dynamics (MD) has been used to assess if MD can be employed as a useful tool for probing the structural basis for enhanced stability in thermal stable cytochromes P450. CYP119, the most thermal stable P450 known, unfolds more slowly during 500K MD simulations compared to P450s that melt at lower temperatures, P450cam and P450cin. A comparison of the 500K MD trajectories shows that the Cys ligand loop, a critically important structural feature just under the heme, in both P450cin and P450cam completely unfolds while this region is quite stable in CYP119. In CYP119 this region is stabilized by tight nonpolar interactions involving Tyr26 and Leu308. The corresponding residues in P450cam are Gly and Thr, respectively. The in silico generated Y26A/L308A CYP119 double mutant is substantially less stable than wild type CYP119 and the Cys ligand loop unfolds similar to P450cam. The MD thus has identified a potential “hot spot” important for stability. As an experimental test of the MD results, the Y26A/L308A double mutant was prepared and thermal melting curves show that the double mutant exhibits a melting temperature (Tm) 16°C lower than wild type CYP119. Control mutations that were predicted by MD not to destabilize the protein were also generated and the experimental melting temperature was not significantly different from the wild type enzyme. Therefore, high temperature MD is a useful tool in predicting the structural underpinnings of thermal stability in P450s.

Cytochromes P450 catalyze the hydroxylation of a wide range of substances and represent one of Nature’s largest enzyme families. Given the widespread occurrence of P450s throughout the biosphere, it was not too surprising when P450s were first discovered in thermophilic organisms (1). CYP119 from the acidothermophilic archaeon, Sulfolobus acidocaldarius, is the most thermal stable P450 known with a melting temperature ≈90°C compared to ≈55°C for the well known cytochrome P450cam (2, 3). Given that P450s have the potential for catalyzing industrially and medically important hydroxylation reactions, an important goal is to understand the structural basis for such stability and then engineer the stability of CYP119 into the P450 of interest thus generating a robust and useful catalyst. The crystal structure of CYP119 (2, 4) revealed an unusual and extensive aromatic network not found in other P450s, which suggested that this may be one feature contributing to the stability of CYP119. Indeed, converting two of the Trp residues, 4 and 281, involved in the aromatic stacking to Ala lowered the melting temperature by 15°C (5) while a single Phe to Ser mutation lowered Tm by about 10°C (6). Even so, many other comparative studies indicate several factors like salt bridges, shorter loops, and more compact packing contribute to enhanced stability (7).

In recent years molecular dynamics (MD) has emerged as an important tool in studying protein unfolding and stability (8-14). The main limitation is that the time scale of protein unfolding at room temperature is far too long for MD simulations of fully solvated protein models, but high temperature MD in the range of 400-600K speeds up the unfolding process to the ns time range and thus is accessible by MD (10, 15, 16). A detailed analysis of high vs. low temperature MD indicates that the unfolding path is not significantly different and that high temperature MD in the range of 498K over a shorter trajectory provides a realistic picture of stability and unfolding (8). Moreover, MD has been used successfully as a guide to determine where to introduce mutations for the purpose of enhancing stability (11, 17). Given these advances in the use of MD to study stability and unfolding we were encouraged to use similar approaches to try to understand the structural basis for stability in P450s. Here we report MD simulations for several P450s including CYP119 coupled with experimental mutagenesis studies that reveal a heretofore, unexpected “hot spot” in P450s that is critically important in conferring enhanced stability.

Experimental Procedures

MD Simulations

Although CYP119 is able to catalyze fatty acid hydroxylation in an artificially reconstituted monooxygenase system (18), the functional role of CYP119 remains unclear and thus there is no simple choice as to which P450 should serve as the mesophilic counterpart. Therefore, we chose P450cam and its close homolog P450cin. The structure of CYP119 has been solved with (2) and without (4) the ligand 4-phenylimidazole bound in the active site. There is a substantial difference in the two structures with the ligand-free form having a wide open active site, while the ligand bound form is in a closed conformation. P450cam exhibits only the closed conformation with and without a ligand bound (19). Therefore, we chose to compare the closed forms of both structures and started with the 4-phenylimidazole complexes of CYP119 (1F4T) and P450cam (1PHD) (20). Included in the P450s subjected to MD simulation were the 4-phenylimidazole complex of the thermal stable CYP231A2 (2FRC) from thermoacidophilic archaeon Picrophilus torridus (21), CYP175A1 (1N97) from Thermus thermophilus HB27 (22), and P450cin (1T2B) from Citrobacter braakii (23). For CYP175A1 an unknown ligand was found to be bound to the heme iron in the original structure determination so this was replaced with 4-phenylimidazole. Both CYP231A2 and CYP175A1 can be considered orphan P450s since the function of each remains unknown. Similarly, the substrate in P450cin was removed and 4-phenylimidazole modeled into the active site using the 4-phenylimidazole-P450cam complex as a guide. In this way all P450s had the same ligand bound at the active site. P450cin was included as an additional mesophilic “control” since P450cin and P450cam are close homologues and were expected to behave similarly in high temperature MD simulations.

The Amber 9.0 suite was used for all calculations (24). The ff99 forcefield provided with the Amber 9.0 package was used for the protein, while the heme-Cys ligand parameters were as described earlier (25). Parameters for the 4-phenylimidazole were derived with Antechamber and the Gaff forcefield (26) using the BCC charging scheme (27, 28). Ligands like phenylimidazole form a covalent bond with the iron and this bond was constrained with a force constant of 200 kcal/Å to the 2.21Å distance observed in the P450cam-4-phenylimidazole crystal structure. The angle defined by the Cys sulfur-heme Fe and 4-phenylimidazole N atom was 92° with a force constant of 10kcal/Å. The P450 structures were stripped of crystallographic waters and then solvated with box of TIP3 waters using 10Å cushion. Counterions were added to maintain a net neutral charge. The total number of atoms including explicit solvent for P450cam (CYP101), P450cin (CYP176A), CYP119, CYP231A2, and CYP175A1 was 36,549, 39795, 41386, 31,570, and 40,069, respectively.

The structures were prepared for production MD runs by first energy minimization for 1000 cycles with all heavy atoms except water molecules fixed in position. This was followed by a short 10 ps MD run to allow just the water molecules to relax and 1000 cycles of energy minimization with all atoms allowed to move. Production runs for the 300 K 30ns simulations were carried out with a 2 fs time step and coordinates saved every 10 ps. Temperature and pressure were held constant through weak coupling with a 1 ps pressure relaxation time and Langevin dynamics using a collision frequency of 1 ps⁻¹. Periodic boundary conditions were used with a Particle Mesh Ewald implementation of the Ewald sum for the description of long-range electrostatic interactions (29). A spherical cutoff of 8.5 Å was used for nonbonded interactions. Bonds involving hydrogen atoms were constrained using SHAKE (30). For the 500 K runs, the starting point was the end of the 30 ns 300 K trajectory. Trajectories were analyzed with the Ptraj program provided with the Amber 9.0 suite.

Simulations at 500K initiated from the final structure and velocities extracted from the 30ns runs. It was necessary to run the 500K simulations for only 8ns since both P450cam and P450cin were substantially unfolded by 8ns. Two methods were employed for simulations with mutants. In method 1 mutations were generated in silico by replacing specific residues with Ala. The replacements were implemented by modifying the topology file and final frame of the 30ns trajectory. Therefore, we start with an equilibrated structure and solvent box thus avoiding a long 300K simulation. The mutant structure was equilibrated for 1ns at 300K followed by an 8ns 500K run. In method 2 we started from the beginning and introduced the mutation into the crystal structure with a new solvent box followed bya full 30ns 300K simulation followed by8ns 500K runs. There was no significant difference in the rmsd of backbone atoms using method 1 or 2. As a result, the “short cut” method 1 was employed.

Site-Directed Mutagenesis

The CYP119 expression plasmid was kindly provided by Prof. Paul Ortiz de Montellano (Univ. Calif., San Francisco). Mutations were introduced into CYP119 gene, using Stratagene’s QuikChange Site-Directed Mutagenesis kit (La Jolla, CA), and In-Fusion PCR cloning kit (Clontech). The primers required were obtained from Operon Technologies (Alameda, CA). The Y26A, and L308A mutations were built into the WT Cyp119 gene individually. The primers used were the following (positions of the mutations are underlined), Y26A_for: TGG CAG GTG TTT TCC GCT AGG TAC ACA AAG G; Y26A_Rev: C CTT TGT GTA CCT AGC GGA AAA CAC CTG CCA; L308A_for: CCG AAC CCA CAC GCA AGC TTT GGG TCT GG; L308A_rev: CC AGA CCC AAA GCT TGC GTG TGG GTT CGG. Once the mutant sequence of the Y26A was confirmed by DNA sequencing, it was used as a scaffold for the next step, where the L308A mutation was introduced using the same method, thereby generating the Y26A/L308A double mutant. The double mutation W147A/Y174A was introduced into the WT Cyp 119 using the following oligos W147A:GATAAGGAGAAGTTCAAAGAGGCGTCAGACTTAGTCGCATTCAG;Y174A:G GTAAGAAGTACCTTGAGTTAATAGGTGCTGTGAAGGATCATCTAAATTCAG and In-Fusion 2.0 dry-down PCR cloning kit. The sequences of all mutants were confirmed by DNA sequencing.

Expression & Purification

All of the proteins were expressed in JM109DE3 cells. The mutant plasmids were transformed into JM109 cells, and a single colony was selected for expression. The cells were grown in 2xYT (tryptone, 16 g, yeast extract, 10 g, NaCl, 5 g, per liter) media supplemented with 100 ug/mL ampicillin. Bacteria were grown to an optical density of 0.6 at 600 nm before protein expression was induced with 250uM isopropyl β-D-thiogalactopyranoside (IPTG). Upon induction, the temperature was reduced to 26°C, the shaking intensity was reduced to 80 rpm, and the cells were harvested 19hr after induction.

Cells were then lysed by sonication but heat precipitation of the crude lysate was avoided and, instead the lysate was centrifuged at 17,000g for 30min and the resulting clear supernatant was loaded onto a 20 mL Ni-NTA column. The column was washed with at least 10 column volumes of column buffer containing 25 mM imidazole. Pure protein was eluted with 50mM sodium phosphate, pH 7.4 buffer containing 100 mM imidazole. The red fractions were pooled and concentrated. Purity was confirmed by SDS-PAGE.

Tm Determination

The thermal melting temperatures were measured by following the absorbance of the heme Soret peak at a linear temperature rate of 0.5°C/min on a Cary 3E UV-vis spectrophotometer equipped with a temperature controller. All measurements were carried out in 50mM Na phosphate buffer pH 7.4. All the reagents and enzymes necessary for site-directed mutagenesis were purchased from New England Biolabs Inc. (Beverly, MA), Invitrogen Corp. (Carlsbad, CA), and Qiagen Inc. (Valencia, CA). Chromatography columns were purchased from Qiagen Inc. All other chemicals were molecular biology grade.

Results and Discussion

Room Temperature MD

Each of the P450s, with the exception of P450cin, used in this study were subjected to 30ns room temperature MD simulations which was sufficient for equilibration as judged by stabilization of the rmsd (root mean square deviation) of backbone atoms (Fig. S1) from the starting crystal structures. P450cin was run for only 10ns but this was sufficient for equilibration.

Thermal stable proteins often are thought to be more rigid than their mesophilic counterparts, which might be reflected in more rigidity during MD simulations. Atomic fluctuations thus were analyzed by computing B factors for backbone atoms using snapshots from 12-30ns. P450cin was not included in this analysis since the room temperature MD was run for only 10ns. Table I provides the B factor for backbone atoms as a function of secondary structure while Fig. 1 highlights key helices in all P450s. Overall there is little difference in B factors between the various P450s. A comparison between the MD and experimentally determined crystallographic B factors might be considered useful. However, in some of the crystal structures surface helices, such as the F and G helices in P450cam, are involved in crystal contacts that tend to decrease B factors.

Table 1.

Average B factor for backbone atoms (Ų)

Secondary Structure	P450cam (CYP101)	CYP119	CYP231A2	CYP175A1
overall	17.6	18.8	17.4	17.20
C helix	17.9	18.2	10.8	8.9
D helix	28.7	17.5	11.9	12.3
E helix	11.9	8.4	5,7	7.9
F helix	19.7	28.7	24.8	27.2
G helix	13.5	27.3	21.8	34.5
H helix	15.7	25.1	18.1	23.7
I helix	6.9	8.7	9.4	9.3
J helix	13.3	15.1	31.0	9.5
K helix	8.9	6.6	5.9	4.5
L helix	15.5	11.6	9.1	5.2
Beta strands	7.2	13.1	13.6	13.3
Overall for secondary structure	14.5	16.4	15.6	14.2

Figure 1.

CYP119 with some key helices labeled. The F and G helices and the F/G loop are cyan colored. Also shown are two interacting aromatic residues, Trp147 and Tyr174.

In addition to the overall B factors being very similar, the various elements of secondary structure also are similar. As expected helical regions buried in the core of the protein or contacting the heme such as the I and L helices have low B factors. However, the F and G helices in all P450s exhibit higher B factors. The F and G helices and the connecting F/G loop are generally thought to undergo a large open/close motion in P450s which enables substrates to enter and products to leave. Note, however, that in the thermophilic P450s the F and G helices are more flexible than in P450cam. Another method for visualization of motions is to carry out a principal component analysis (31). Using Ptraj in Amber 9.0 the covariance matrix for Cα atoms computed over the last 20ns of the simulation was diagonalized, thus providing a series of eigenvectors. Projection of these vectors onto the structure enables visualization of the primary low frequency motions. Visualization was achieved using Interactive Essential Dynamics (32) and the graphics program VMD (33). Movements along the first eigenvector in the F and G helical region is shown in Fig. 2. It is clear that the F and G helices experience a greater range of motion in CYP119 than in P450cam. In agreement with these results a much longer 200ns MD simulation of CYP119 shows substantial motion of the F/G helical region (34).

Figure 2.

The F and G helical regions in P450cam and CYP119. The red and blue structures represent the extreme ends of the motion defined by the primary eigenvector obtained from the principal component analysis of the 300K 30ns MD trajectories.

That CYP119 is actually more flexible in the F/G helical region than P450cam while the rest of the structure is about the same as P450cam may at first seem counter-intuitive since rigidity normally is associated with increased stability and is counter to the prevalent view that at room temperature, thermophilic proteins are more rigid and become flexible and functional only at elevated temperatures (35-38). However, an increasing number of both experimental and computational studies lead to a much different view. A recent neutron scattering experiment comparing E. coli dihydrofolate reductase with its thermophilic counterpart found that the thermophilic enzyme is intrinsically more flexible (39). Similar conclusions were made from deuterium exchange experiments with α-amylases (40). MD simulations of diverse proteins (9, 12, 14, 41) also came to the conclusion that the thermal stable protein are no more rigid that their mesophilic cousins. Moreover, as noted by Lazaradis et al (42), there is no thermodynamic advantage to rigidity. In fact, just the opposite may be true since greater flexibility in the folded protein means there is less of a conformational entropic penalty in going from the unfolded to the folded state. It thus appears that the CYP119 fits the pattern of not being more rigid than its mesophilic counterpart and in one region of the structure is actually more flexible.

High Temperature MD

We next subjected the P450s to 500K MD runs for 8ns. We selected 500K since previous studies have shown that 400-600K are reasonable for following protein unfolding on the ns time scale (15, 16). The starting point was the end of the 30ns 300K simulation. The rmsd from the starting crystal structures over the course of the unfolding trajectories are shown in Fig. 3A while Fig. 3B shows the radius of gyration for P450cam and CYP119 as a function of time. As might be expected, P450cam and P450cin exhibit the largest rmsd over time, which indicates P450cam and P450cin are unfolding. There also is a greater variation and increase in the radius of gyration for P450cam compared to CYP119 indicative of a looser less compact structure. This is more evident in Fig. 4 which shows the 8ns 500K structures compared to the starting crystal structures. By 8ns both the F and G helices are substantially unfolded in both P450s. However, in P450cam the beta structure in the upper left region of the molecule (Fig. 1) and the A helix are unfolded while in CYP119 the homologous region remains folded. Analysis of the 500K trajectories also enables a quantitative estimate on the rate of secondary structure loss. In P450cam about 40% of the helices are lost by 4ns and 50% by 8ns. In CYP119 about 20% of the helices are unfolded by 4ns and 35% by 8ns. We also checked the 500K MD runs as a function of starting point. Others (8) have started from the energy minimized crystal structure without first equilibrating the structure at 300K. As shown in Fig. S2 there is essentially no difference between starting at the end of the 30ns 300K simulation or the energy minimized crystal structure.

Figure 3.

A) RMSD of backbone atoms (Cα, N, and C) for several P450s over the 8ns trajectories at 500K. B) Radius of gyration for P450cam and CYP119 over the 8ns 500K trajectories.

Figure 4.

Models of the starting crystal structures and the last saved structure at 8ns from the 500K simulation.

A closer view of some functionally important regions illustrates additional major differences between P450cam and CYP119. Fig. 5 shows the Cys ligand loop found in all P450s. This is a β–bulge segment that forms a tight turn under the heme. Note that for CYP119 this region is very stable over the entire 8ns at 500K but completely unfolds in P450cam. We repeated the 500K MD simulations by either taking velocities from the 20ns 300K MD structure rather than the 30ns velocities or reinitializing with a new set of velocities. Either way the Cys ligand loop unfolds in P450cam but remains stable in CYP119. One way of quantitating this local unfolding is to follow the key H-bond distance that helps to hold the Cys ligand loop together. All P450s have a H-bond between the carbonyl O atom of the highly conserved Phe (cyan color in Fig 5) and the peptide NH of the Cys ligand. This H-bond distance is 2.9Å in both P450cam and Cyp119 at the beginning of the 500K MD runs. In P450cam this distance increases to 18.1Å, 9.0Å, and 11.0Å at the end of the three 500K MD runs while in CYP119 the increase is to 3.1Å in all 3 500K MD runs. The overall picture that emerges from these studies is that the helical core is fairly stable in P450s whereas the N-terminal region, the nearby β structure, and the Cys ligand loop are “hot spots” in P450cam and P450cin and unfold first during the 500K simulation. We thus focused attention on these hot spot regions.

Figure 5.

Beginning and ending structures from the 500K simulations for P450cam, CYP119, and CYP119 mutants. Note that the Cys ligand loop completely unfolds in P450cam and the Y26A/L308A CYP119 mutant but not in wild type CYP119 or the W147A/Y174A mutant.

In Silico Mutagenesis

A closer examination of the Cys ligand loop provides some testable reasons for the differences in stability. As shown in Fig. 6 P306, L308, F292, V291, and Y26 in CYP119 form a tight hydrophobic cluster at the N-terminal end of the Cys ligand loop. The corresponding residues in P450cam are T348, S346, N332, E331, and G68. This indicates the N-terminal region of the protein (Gly68 in P450cam and Tyr26 in CYP119) provides greater stability to the Cys ligand loop in CYP119. We already know from the MD work so far that a hot spot for early unfolding is the N-terminal part of the molecule. Therefore, as the N-terminal region unfolds, the Cys ligand loop is destabilized. In CYP119 the additional nonpolar interactions in this region provides additional stability.

Figure 6.

Detailed models of the Cys ligand loop in P450cam and CYP119.

To test this hypothesis we generated in silico mutants of CYP119 and converted Tyr26 and Leu308, which form part of a tight cluster in CYP119 but not in P450cam, to Ala. We also generated the two single mutants. These mutants then were subjected to 500K MD runs. The rms deviation plots are shown in Fig. 7. For comparison, P450cam and wild type CYP119 are included. The double mutant unfolds more rapidly than wild type CYP119 independent of starting structure and velocities (Fig. S3). Not only is the double mutant substantially less stable than wild type but the Cys ligand loop unfolds similar to P450cam (Fig. 5). It thus appears that the clustering of nonpolar residues ties the C-terminal end of the protein (Leu308) to the N-terminal end (Tyr26) and that the loss of this interaction results in substantial destabilization. The single mutants also unfold faster than the WT but in both these mutants the Cys ligand loop remains intact at the end of the 500K MD run.

Figure 7.

RMSD plots for the 8ns trajectories at 500K for CYP119, P450cam, and the in silico generated Y26A/L308A CYP119 mutant.

Experimental Test

To test if Tyr26 and Leu308 indeed are important for stabilization, three mutants were generated: Y26A, L308A, and Y26A/L308A. These were purified and thermal melting curves determined (Fig. 8A). The double mutant exhibits a Tm=76°C compared to 92°C for wild type CYP119, a significant 16°C decrease in stability. The individual single mutants are more stable than the double mutant but still less stable than wild type CYP119. We also carried out a control study by mutating two interacting nonpolar residues that were predicted by MD simulations not to alter stability. By visually examining the CYP119 structure, we selected Trp147 and Tyr174 as two interacting aromatic residues that are close together in sequence and should have little effect on stability. Moreover, these two residues form interactions between the F and G helices (Fig. 1) which are the two most flexible helices in the structure and thus, mutations at these sites are likely to have little effect on stability. The 500K MD run of the in silico generated W147A/Y174A double mutant predicts that these two mutations do not alter stability (Fig. 8B). In addition, the Cys ligand loop does not unfold in this mutant (Fig. 5). We also experimentally generated this mutant and the thermal melting curve is shown in Fig. 8A. The W147A/Y174A double mutant exhibits a Tm just a few degrees lower than wild type. There thus is excellent agreement between the MD predicted stability of mutants and experimental thermal melting curves.

Figure 8.

A) Thermal melting curves for CYP119 and various mutants. On the Y axis the minimum and maximum optical densities at 415nm were normalized to be 0 and 100, respectively. B) RMSD plots for the 8ns trajectories at 500K for CYP119, P450cam, and the in silico generated W147A/Y174A CYP119 mutant.

Conclusion

The main goal of this study was to assess the use of MD simulations as a tool to help predict which regions of CYP119 might be especially important in conferring stability. What was not evident in the static crystal structures, but suggested from MD, is that the Cys ligand loop region (Fig. 3) and the tighter clustering on nonpolar groups in CYP119 is important for enhanced thermal stability, as compared to P450cam. It should be noted that this region is not part of the hydrophobic core of the protein and, in fact, Tyr26 and Leu308 are partially solvent exposed. Also noteworthy is that the W4A/W281A double mutant melts 15°C lower than wild type (5), and, like Tyr26 and Leu308, these two Trp residues help to tie N- and C-terminal regions together in a partially exposed nonpolar cluster. The 500K simulations show that the N-terminal region and the nearby β structure are particularly unstable at elevated temperature so perhaps one strategy CYP119 uses to enhance stability is to tether N- and C-terminal segments together via enhanced nonpolar interactions relative to P450cam. The experimental test of this prediction clearly shows that Tyr26 and Leu308 play important stabilization roles since converting these two groups to Ala lowered Tm by about 16°C.

The obvious next question is to ask if MD can be used to predict which mutants will increase stability. This is a much more challenging problem. For example, if we want to mimic the Tyr26-Leu308 interaction found in CYP119 in P450cam, we would need to replace a Gly and Thr with larger nonpolar residues. This will cause other local and likely unfavorable packing problems. We did attempt to mimic the Tyr26-Leu308 by introducing a S-S bond but this mutant did not exhibit enhanced stability using the high temperature MD test. Disulfide bonds, however, have highly restricted stereochemical requirements and are thus not likely to be generally useful for mimicking nonbonded intramolecular interactions. Salt bridges provide another possibility since there are far fewer stereochemical restraints and given that such ionic interactions are on the surface, there is little problem with altering nonpolar packing. Indeed, in a study similar to ours, a salt bridge found to be important in the thermophilic protein was engineered into the mesophilic counterpart resulting in enhanced stability of the mesophilic protein (17). It thus may be possible to mimic nonbonded intramolecular interactions that tie together residues distant in sequence with engineered salt bridges. In summary, this study illustrates that MD is a useful tool for predicting regions in P450s that are particularly important in conferring enhanced stability and perhaps can help streamline mutagenesis experiments.

Abbreviations

MD

molecular dynamics

rmsd

root mean square deviation