Research article

Microsomal prostaglandin E synthase-1 exhibits one-third-of-the-sites reactivity

Shan He, Yiran Wu, Daqi Yu, Luhua Lai


mPGES-1 (microsomal prostaglandin E synthase-1) is a newly recognized target for the treatment of inflammatory diseases. As the terminal enzyme of the prostaglandin production pathway, mPGES-1 inhibition may have a low risk of side effects. Inhibitors of mPGES-1 have attracted considerable attention as next-generation anti-inflammatory drugs. However, as mPGES-1 is a membrane protein, its enzymatic mechanism remains to be disclosed fully. We used MD (molecular dynamics) simulations, mutation analysis, hybrid experiments and co-IP (co-immunoprecipitation) to investigate the conformation transitions of mPGES-1 during catalysis. mPGES-1 forms a homotrimer with three substrate-binding sites (pockets). In the MD simulation, only one substrate molecule could bind to one of the pockets and form the active complex, suggesting that the mPGES-1 trimer has only one pocket active at any given time. This one-third-of-the-sites reactivity enzyme mechanism was verified further by hybridization experiments and MD simulations. The results of the present study revealed for the first time a novel one-third-of-the-sites reactivity enzyme mechanism for mPGES-1, and the unique substrate-binding pocket in our model constituted an active conformation that was suitable for further enzymatic mechanism study and structural-based drug design against mPGES-1.

  • drug design
  • enzyme mechanism
  • inflammation
  • membrane-associated proteins in eicosanoid and glutathione metabolism (MAPEG)
  • microsomal prostaglandin E synthase-1 (mPGES-1)
  • one-third-of-the-sites reactivity


mPGES-1 (microsomal prostaglandin E synthase-1) is a newly recognized target for the treatment of inflammatory diseases [1]. NSAIDs (non-steroidal anti-inflammatory drugs) are taken by more than 30000000 people each day around the world. However, almost all of the currently available NSAIDs have certain types of side effects (e.g. stomach damage). To gain more insights into network-based anti-inflammatory drug design, the dynamic properties of the AA (arachidonic acid) metabolic network have been studied by us previously [2,3]. In this AA metabolic network, several enzymes, including leukotriene A4 hydrolase, have been investigated for their enzymatic mechanisms and inhibitor/activator design [4,5]. mPGES-1 is the terminal enzyme of the production pathway of prostaglandins (one type of inflammatory mediator) and is not coupled to any downstream enzymes in the enzymatic cascade. Constitutive levels of mPGES-1 are normally low, and it is highly up-regulated by pro-inflammatory stimuli. Therefore, mPGES-1 is a novel attractive target with a low risk of side effects [6].

However, the enzymatic reaction mechanism of mPGES-1 has not been studied much and insufficient information is available to guide rational inhibitor design. Site-directed mutagenesis was used to explore the contribution of residues to the enzymatic reaction [7,8]. Hammarberg et al. [7] reported that the mutation of Arg126 in mPGES-1 to an alanine residue changed the isomerase activity to reductase activity. This observation verified Arg126 as a key catalytic residue. Pawelzik et al. [9] investigated the effects of different inhibitors on rat/human mPGES-1 and identified key residues as gatekeepers for the active site of the enzyme. Their work is helpful for predicting the substrate-binding site, yet a comprehensive understanding of the enzymatic structure and reaction mechanism remains a great challenge.

mPGES-1 is a member of the MAPEG (membrane-associated proteins in eicosanoid and glutathione metabolism) protein family, which also includes MGST (microsomal glutathione transferase)-1, MGST-2, MGST-3, LTC4S (leukotriene C4 synthase) and FLAP (5-lipoxygenase-activating protein) [10]. MGST-1 is the first member of the protein family for which crystal structures have been determined [11,12]. On the basis of these crystal structures, two comparative models of mPGES-1 were built and reported by two groups: one by Hamza et al. [13], and the other by Xing et al. [14]. Even before the high-resolution crystal structure of MGST-1 was published, Zhan and colleagues built the first model of the substrate-binding domain of mPGES-1 (one pocket) using an ab initio structure prediction approach [15]. After a 3.2 Å (1 Å=0.1 nm) resolution crystal structure of MGST-1 became available, they built a new structural model of the mPGES-1 trimer using comparative modelling with MD (molecular dynamics)-simulation refinement and experimental validation [13]. On the basis of this model, several important residues for ligand binding were identified and their computational predictions about these residues were supported by the corresponding experimental tests [16]. However, because both comparative models were mainly based on the structure of MGST-1, the cofactor GSH in their models took an extended conformation like that in the MGST-1 structure as pointed out by Pawelzik et al. [9]. Recent advances in membrane protein studies and structural biology produced an explosion of crystal structures for MAPEG family proteins including FLAP [17], LTC4S [18] and mPGES-1 [19]. In the crystal structures of LTC4S and mPGES-1, GSH took a U-shaped conformation. As the residues co-ordinating GSH were highly conserved in the MAPEG family, the U-shaped GSH-binding mode should be conserved among MAPEG enzymes. Compared with the U-shaped conformation, the extended conformation of GSH in MGST-1 might correspond to a structurally distorted form [20]. Therefore the problem of MGST-1 crystal structure might influence the comparative modelling of mPGES-1 which used the structure of MGST-1 as the template structure, especially for the GSH-binding site.

Jegerschöld et al. [19] determined the structure of mPGES-1 by electron crystallography to a resolution of 3.5 Å. The mPGES-1 molecule in this structure took a closed inactive conformation that was not accessible by the substrate PGH2 (prostaglandin H2). As the crystal structure of LTC4S was solved with an open conformation to a resolution of 2.15 Å, by comparing these two structures they speculated that a conformation change might occur during binding and turnover of the substrate. Enlightened by this idea, we attempted to build an open conformation structural model of mPGES-1 and used MD simulations to investigate the conformation transition between the open and closed state. During the MD simulation, only one substrate molecule was observed to bind to the pocket and form an active complex, suggesting that the mPGES-1 trimer might only have one pocket active at any given time. This one-third-of-the-sites reactivity enzyme mechanism was further supported by hybridization experiments and prolonged MD simulations.



The plasmid DNA harbouring the full-length cDNA of the PIG12 gene was obtained from Addgene [Addgene plasmid 16506: pBK-CMV Pig12 (CMV is cytomegalovirus)]. PGH2, anti-(mPGES-1) monoclonal antibody, mPGES-1 Western blot-ready control and the PGE2 EIA (enzyme immunoassay) kit were purchased from Cayman Chemical. POPC (1-palmitoyl-2-oleoyl phosphatidylcholine) was obtained from Avanti Polar Lipids. Immunoprecipitation Kit Dynabeads® Protein G and DynaMag™-2 magnets were purchased from Invitrogen. An anti-His6-tag antibody and a rabbit polyclonal secondary antibody against mouse IgG were purchased from Abcam. Other reagents were obtained from Sigma–Aldrich unless indicated otherwise.

Cloning and plasmid construction

The coding region of mPGES-1 from the plasmid DNA was PCR-amplified using the primer pair 5′-CGTGTACATATGCCTGCCCACAGCCTG-3′ and 5′-CTAGAATTCACAGGTGGCGGGCCGCTTC-3′. The NdeI/EcoRI fragment was ligated into a pET30a (+) vector. Three CGG codons of mPGES-1 encoding the Arg40, Arg73 and Arg122 were changed to CGT codons using a Muta-direct kit (SBS) to improve mPGES-1 expression [21]. For further co-IP (co-immunoprecipitation) analysis, we also constructed His–mPGES by ligating the NdeI/EcoRI fragment into a pET28a (+) vector. After ligation, an extended His6-tag (MGSSHHHHHHSSGLVPRGSH) was attached to the N-terminus of mPGES-1. Therefore the protein bands of the fusion protein His–mPGES (172 amino acids, theoretical molecular mass of 18 kDa) and mPGES-1 (152 amino acids, theoretical molecular mass of 16 kDa) could be separated by SDS/PAGE. The protein-coding region of the resulting expression vector was sequenced for verification.

Protein expression and microsome preparation

Human recombinant mPGES-1 was overexpressed in the Rosetta (DE3) strain of Escherichia coli. Recombinant cells were cultivated at 37°C in LB (Luria–Bertani) medium containing 30 μg/ml kanamycin and 34 μg/ml chloramphenicol until an attenuance value of D600=1.0 was reached. Protein expression was induced by the addition of IPTG (isopropyl β-D-thiogalactopyranoside) to a final concentration of 1 mM, and the cells were grown for a further 12 h at 25°C. Cells were harvested by centrifugation at 5000 g for 20 min at 4°C.

The cell pellet from a 1 litre culture was resuspended in lysis buffer [1 mM EDTA, 1 mM PMSF and 10% (v/v) glycerol (pH 7.4)] and lysed by ultrasonication (350 W, 99× 3 s bursts with 5 s rest between each burst) until homogeneous. Insoluble material was separated by centrifugation (12000 g for 30 min at 4°C), and the supernatant was then ultracentrifuged at 52000 rev./min, for 1 h with a Hitachi rotor (S58A-0026). The membrane pellet was washed in activity assay buffer [2.5 mM GSH, 0.1 M potassium phosphate buffer (pH 7.4) and 1% (v/v) glycerol] once and then resuspended in 2 ml of solubilizing buffer [2.5% (w/v) POPC in activity assay buffer]. The total membrane protein concentrations were determined using a bicinchoninic acid protein assay kit (Biomed Laboratories) with BSA as a standard. According to the concentration, the microsome samples were diluted to a final concentration of 10 mg/ml. WT (wild-type) and mutant mPGES-1 microsome samples and negative control (microsome prepared with Rosetta cells without IPTG induction) were all prepared in this manner.

Enzymatic activity assay

mPGES-1 activity was measured by quantifying the conversion of PGH2 into PGE2 according to a method reported previously [22]. The microsome was diluted in activity assay buffer to the desired concentration. Because the substrate PGH2 was labile, it was always kept on dry ice until just prior use. Prior to incubation, both the substrate and samples were transferred on to ice for 2 min for temperature equilibrium. PGH2 was added to each well of a 96-well plate, and the reaction was started by adding 100 μl of the samples. After reacting at 4°C for 1 min, the reaction was terminated by adding 150 μl of stop solution (50 mM FeCl2 and 100 mM citric acid) to lower the pH to 3. PGE2 in the reaction mixture was quantified using the PGE2 EIA kit (Cayman Chemical).

Site-directed mutagenesis and activity comparison between the WT and mutants

Mutagenesis was performed using a Muta-direct kit (SBS). The plasmid pET30a (+) containing WT mPGES-1 was mutated to obtain the following mutants: R38A, R38S, H72A, N74A, E77A, R110A, Y117A, Y117F and R126A. The plasmid pET28a (+) containing His–mPGES was mutated to obtain His–R126A. All DNA sequences of the mutants were verified by DNA sequencing. The protein expression and activity assay of mutants were performed as described for WT. As the expression levels of mutants and WT might be different, the mPGES-1 content in the microsome samples might be different, which would affect the comparison of their activities. To eliminate the influence of enzyme concentration, we divided the enzymatic activity by the mPGES-1 content in the sample (determined by Western blot analysis, see Supplementary Figure S3 and Supplementary Table S3 for quantification details at to obtain the unit activity. Then, for comparison, the unit activities of mutants were represented as the relative activity compared with that of the WT.

Hybridization experiment

Microsome samples with a constant molar ratio (10:1) of R126A/WT were mixed for various periods to investigate the mixing process. The incubation temperature was 25°C, with gentle shaking at 500 rev./min. To reach the 10:1 molar ratio, we analysed the mPGES-1 expression level in the R126A and WT samples by Western blotting (see Supplementary Figure S3C lanes 4–5 and Figure 3C lanes 1 and 3). As their expression levels were very similar, 50 μl of WT and 500 μl of R126A samples were mixed directly. Then, maintaining a constant incubation time (28 h), different molar ratios of R126A/WT were mixed in the same manner while keeping the amount of WT constant [23]. A control experiment using WT and a negative control was performed in the same way. The activity of the final enzyme mixture was determined using the PGE2 EIA kit.


Prior to co-IP, microsome samples with equal amounts of WT and His–R126A were mixed at 25°C with rotation for 28 h. Co-IP was performed using an immunoprecipitation kit (Invitrogen) according to the manufacturer's protocol. Briefly, Dynabeads Protein G (1.5 mg) were incubated with 3 μg of antibody [using an anti-His6-tag antibody for anti-His IP (immunoprecipitation) and an anti-mPGES-1 antibody for anti-mPGES IP] for 20 min with rotation at 25°C. The antibody-bound Dynabeads were then gently washed and microsome sample mixture (typically 500 μl) was added. After 1 h of incubation, the supernatant was removed (collected for further Western blot analysis) and the Dynabeads were washed three times. The Dynabeads were then resuspended in 20 μl of elution buffer and 20 μl of 2×SDS loading buffer, followed by heating at 70°C for 10 min.

The immunoprecipitated proteins in the eluent were resolved by 15% Tricine-SDS/PAGE and transferred on to PVDF membranes (Millipore) for Western blot analysis. Primary antibody (1:500 anti-mPGES-1 antibody for anti-mPGES-1 Western blot and 1:1000 anti-His6-tag antibody for anti-His Western blot) and rabbit polyclonal secondary antibody against mouse IgG were used together with the BCIP (5-bromo-4-chloroindol-3-yl phosphate)/NBT (Nitro Blue Tetrazolium) Color Development Substrate (Promega) to visualize the proteins. A control experiment using a WT and R126A mixture and a parallel experiment using R126A and His–WT mixture were performed in the same manner.

Construction of the mPGES-1 open conformation structural model with PGH2 and structure refinement using MD simulation

As mPGES-1 was solved in a closed conformation (PDB code 3DWW) and LTC4S was solved with an open conformation (PDB code 2UUH), we compared and superimposed these two structures using the program PyMOL (; Figure 2). The main difference lay in the bending direction of the cytoplasmic half of helix 1 and helix 2 (active region) over a hinge fixed by the Lys26–Asp75 salt bridge, as Jegerschöld et al. [19] suggested. Therefore we manually bent this region over the hinge with reference to the corresponding helix in LTC4S to ‘open’ the substrate-binding pockets. As mPGES-1 is a trimer that has three possible substrate-binding sites, all three sites were ‘opened’ using the same procedure initially. As the structure was determined without the presence of substrate, we docked the substrate PGH2 into the binding site. Prior to the molecular docking of PGH2, the tentative open conformation structure was optimized by energy minimization and MD simulation with the protein embedded into the POPC phospholipid bilayer surrounded by water molecules using the GROMACS software package version 4.0.5 with the GROMOS96 43a1 Forcefield [25]. The final system contained three protein chains, three GSH moieties, 98 POPC moieties, 3877 water molecules and 21 Cl ions (a total of 21191 atoms in a 63 Å×63 Å×76 Å simulation box). All simulations were run under constant pressure (1 bar) and temperature (310 K) using the Parrinello–Rahman method [26] for pressure control and the Nose–Hoover method [27] for temperature control. A 12-Å cut-off was used for the non-bonded interactions, and long-distance electrostatic interactions were treated using the particle mesh Ewald method [28]. The bonds with hydrogen atoms were constrained by the LINCS algorithm [29], and the time step was set to 1.0 fs. Harmonic restraint with a stiffness of 100 kJ·mol−1·nm−2 was applied on the α-carbon atoms of proteins to avoid unwanted pocket closing.

PGH2 was then docked to the three pockets separately by AutoDock 4 using a genetic algorithm with the following parameters: ga_num_evals=10000000, ga_pop_size=300 and ga_run=100. Arg116, Tyr120, Gln124 and Gln130 were treated as flexible residues to aid the docking. The final conformation of PGH2 was selected with consideration of binding energy, clustering, and proper protein–substrate interactions. This mPGES-1–PGH2 complex structure was further optimized by MD simulation with progressively restraint reduction until full relaxation was achieved. MD simulations of 100 ns (notated as refining-MD simulation) were required to obtain a stable open conformation complex structural model of mPGES-1 (notated as open conformation, see Supplementary Table S1 and Supplementary Figure S1 at for MD simulation details).

Construction of pseudo-all open structure model of mPGES-1 and prolonged MD simulations

To elucidate the conformation transition mechanism and confirm the unsymmetrical structure of the trimer, we also built a pseudo-all open structure and studied its conformational changes during MD simulation. The all open structure was built using the second monomer of mPGES-1 with GSH (the open conformation) to generate a symmetric trimer. After repacking the enzyme to the membrane environment and energy minimization, a 50 ns fully relaxed MD simulation was performed to investigate the conformation transition elicited by this change (notated as simulation A). The energy minimization steps and MD simulation parameters were the same as the modelling section (see above). The reverse transition from the all closed structure (PDB code 3DWW) was also studied by 50 ns fully relaxed MD simulations in the membrane environment (notated as simulation B). Simulation A and B were performed in three independent runs, and at least two trajectories exhibited similar results. The final structure after MD simulation (after-MD structure) was taken directly as the last frame.

Inhibitor docking

We collected mPGES-1 inhibitors that had available IC50 values based on the cell-free microsome assay in the literature (a total of 117 molecules belonging to 19 different types, see Supplementary Table S2 at The inhibitors were docked to the substrate-binding pocket using the program AutoDock 4 (genetic algorithm with parameters of: ga_num_evals=50000000, ga_pop_size=300 and ga_run=50). Docked structures were clustered on the basis of the RMSD (root mean square deviation) value, and only the structure with the lowest energy in the largest cluster was selected as the best docking result. The correlation between experimentally determined IC50 values and calculated Kd values was analysed.


Construction of the open conformation structural model of mPGES-1 with PGH2 and MD simulation refinement

mPGES-1 forms a homotrimer with three substrate-binding pockets and each of the pockets is buried at the interface of two adjacent monomers. mPGES-1 catalyses the conversion of PGH2 into PGE2 and requires GSH as an essential cofactor for activity. The mPGES-1 structure (PDB code 3DWW) was determined in complex with GSH, but without substrate [19]. As there were no reports about the binding ratio of PGH2 to the enzyme trimer, we tentatively docked PGH2 in all three pockets. Surprisingly, these three PGH2 molecules moved differently during the MD simulation (Figure 1). We numbered the pockets clockwise as pocket-1, pocket-2 and pocket-3 for clarity. To represent the extent to which the substrate entered the pocket, we used the distance between the oxygen atom of PGH2 and α-carbon atom of Arg73 because Arg73 was located at the bottom of the pocket and positioned on the same horizontal plane as PGH2. In pocket-1 and pocket-3, the distance between PGH2 and pocket grew from 14 Å to more than 17 Å, whereas in pocket-2 the distance was reduced from 14 Å to less than 12 Å. Therefore, during the refining-MD simulation, only the substrate near pocket-2 entered the pocket to form an active complex, whereas the other substrates moved away from the pocket. This implied that the mPGES-1 trimer might have only one pocket active at any given time. The cofactor GSH remained tightly bound only in pocket-2. Recent studies of a homologous protein MGST-1 indicated that its three pockets had different affinities for GSH with one high-affinity (Kd=20 μM) and two low-affinity (Kd=2.5 mM) sites [30]. Our MD simulation results agree well with these experimental data.

Figure 1 Open conformation structural model of mPGES-1

(A) Final model of mPGES-1 (the three monomers are coloured green, magenta and cyan) with cofactor GSH (yellow) and substrate PGH2 (blue). The distances between the oxygen atoms of PGH2 and α-carbon atoms of Arg73are labelled to represent the extent to which the substrate entered the pocket. (B) The PGH2–Arg73 distance variation during the MD simulation. (CE) Closer view of pocket-1, -2 and -3. The PGH2–GSH and PGH2–Arg126 distances were labelled to illustrate the substrate-binding mode. (F) Hypothetical enzyme reaction mechanism involves an attack of the GSH thiolate and protonation of Arg126 [19].

In pocket-2, the binding mode of PGH2 was suitable for enzymatic reaction. The endoperoxide bridge of PGH2 formed a hydrogen bond with the guanidine group of Arg126 (3.4 Å) and was close to the thiol group of GSH (4.4 Å). This was consistent with the reaction mechanism proposed by Jegerschöld et al. [19], which involved the participation of GSH and Arg126. Tyr117 and Tyr130 were also close to the endoperoxide bridge of PGH2 (within 5 Å).

Validation of important residues identified in the model by mutagenesis

In pocket-2, seven residues (Arg38, His72, Asn74, Glu77, Arg110, Arg126 and Tyr117) formed hydrogen bonds with GSH or PGH2, which may play important roles in enzyme catalysis. Although there were mutation data published for six of them (except for Arg38) by different research groups [7,8,19], for comparison we mutated all seven residues and compared their activities (see Supplementary Figure S2 at for a comparison between the results of the present study and the published data). The enzymatic activity of all mutants decreased dramatically to less than 30%, which verified the importance of these residues (Figure 2). We also examined the less important residues (defined as residues that after their mutation mPGES- activity remained above 50%) identified by published mutation data [7,8,15,19]. In pocket-2 of our model, the unimportant residues were not located near the substrate-binding pocket, in good agreement with the experimental results (see Supplementary Figure S2 for details).

Figure 2 Important residues validated by mutagenesis

(A) Important residues in pocket-2 identified by our model that formed hydrogen bonds with GSH. (B) Relative enzymatic activities of mPGES-1 WT and mutants. The WT activity was set as 100%. The enzymatic activity was measured by quantifying the conversion of PGH2 into PGE2 as described in the Experimental section. Results are means±S.E.M. (n=3).

In summary, the conformation of pocket-2 can explain the reaction mechanism well, and the catalytic residue Arg126, and the important and unimportant residues for enzyme reaction can be predicted correctly. Thus the structure of pocket-2 provides a good representation of the substrate-binding conformation.

Hybridization experiment and co-IP

To further verify the one-third-of-the-sites reactivity mechanism, we performed hybridization experiments by mixing the inactive R126A mutant (M) with the active WT enzyme (W) to produce hybrid trimers. Considering Arg126 was located far from the subunit interface and therefore was unlikely to influence the hybrid trimer formation, we selected the R126A mutant as the inactive mutant. There were four possible species (WWW, WWM, WMM and MMM) in the enzyme mixture. For the mPGES-1 trimer, there were three possible enzymatic mechanisms. (i) Mechanism 1: all-sites reactivity (only WWW is active). In this situation, if we add the mutant to the WT sample, the activity of the enzyme mixture will decrease as more hybrids will be formed. (ii) Mechanism 2: one-third-of-the-sites reactivity (WWW, WWM and WMM are all active). One-third-of-the-sites reactivity means that the mPGES-1 trimer needs at least one active monomer to be active, and therefore WWW and both hybrids (WWM and WMM) should be active. In this situation, adding increasing amount of the mutant to the WT would result in the increasing formation of active hybrids, and the total activity would increase. The maximum activity of the enzyme mixture would be three times that of the WT as one WWW could recombine to form three WMM hybrids; (iii) mechanism 3, two-thirds-of-the-sites reactivity (both WWW and WWM are active). In this situation, the activity of the enzyme mixture would initially increase and then decrease when a higher concentration of the mutant is mixed with the WT. This is because during the process of mixing, the WT and mutant first form the active hybrid WWM, giving an increase in the total activity. When more inactive mutant is added the WWM hybrid will be transformed into the inactive hybrid WMM and the activity of the mixture would decrease. The turning point should occur when the mutant and WT are in an equal molar ratio. At this point, the total concentration of active WWW and WWM will be the same as those of the inactive WMM and MMM. Increasing the concentration of the mutant after this point would result in the formation of additional inactive species, thereby reducing the activity of the enzyme.

On the basis of the differences between these three possible enzymatic mechanisms, we designed two hybridization experiments (Figure 3). First, we mixed samples containing a constant molar ratio (10:1) of R126A/WT for various time periods to investigate the mixing process. Figure 3(A) shows that the experimentally measured enzymatic activity increased with incubation time and reached equilibrium at approximately 28 h. This result excluded the possibility of mechanism 1 and supported mechanism 2. We then kept the incubation time at 28 h and mixed different molar ratios of R126A/WT (keeping the concentration of WT constant). The experimental enzymatic activity increased as the R126A/WT molar ratio increased (even after R126A/WT=1) and increased by a factor of approximately three when R126A/WT=10. This result excluded the possibility of mechanism 3 and supported mechanism 2. We also performed a control experiment by adding the negative control to the WT, upon which no increase in activity was observed. Therefore the activity increases observed in the two experiments should be caused by the addition of R126A, and the experimental results supported mechanism 2 (one-third-of-the-sites reactivity). For the second experiment, we also calculated the theoretical activity curve for the three possible enzymatic mechanisms, and the experimental value was in good agreement with the one-third-of-the-sites reactivity mechanism (see Supplementary Figure S4 at for a detailed analysis).

Figure 3 Hybridization experiments and co-IP

(A) Microsome samples with a constant molar ratio (10:1) of R126A/WT were mixed for various periods to investigate the mixing process. A control experiment was also performed using a WT and negative control (NC; microsome prepared with Rosetta cells without IPTG induction) mixture. The enzymatic activity was measured by quantifying the conversion of PGH2 into PGE2 as described in the Experimental section. The activity of the enzyme mixture was represented by the relative activity (the starting point of the WT+R126A activity curve when the incubation time was 0 h was set as 100%). Results are means±S.E.M. (n=3). (B) Microsome samples with different molar ratios of R126A/WT were mixed for 28 h before the activity assay. A control experiment was also performed using a WT and negative control mixture. The activity of the enzyme mixture was represented by the relative activity (the starting point of the WT+R126A activity curve when the R126A/WT molar ratio equalled 0 was set as 100%). Results are means±S.E.M. (n=3). (C) Samples containing equal amounts of WT (theoretical molecular mass of 16 kDa) and His–R126A (theoretical molecular mass of 18 kDa) (or parallel experiments using R126A and His–WT samples) were mixed for 28 h and then immunoprecipitated with anti-mPGES-1 antibody (mPGES IP*) or anti-His6-tag antibody (His IP). Both the immunoprecipitated proteins (IP) and the supernatant (S) were analysed. The samples were resolved on two identical gels, which were analysed by Western blotting (WB) using the anti-mPGES-1 antibody (top panel, anti-mPGES-1 Western blot) or anti-His6-tag antibody (bottom panel, anti-His6 tag Western blot). The additional band detected above 18 kDa in the anti-His6 tag Western blot was a non-specific band. Lanes 1 and 2 are WT and His–R126A microsome samples respectively. Lanes 5–7 are co-IP analysis of WT and His–R126A mixture. Lanes 3 and 4 and 8–10 are a parallel experiment using R126A and His–WT mixture. A representative result of three independent experiments is shown. A control experiment using WT and R126A mixture was also performed (see Supplementary Figure S3C at for details). Molecular mass in kDa is given on the left-hand side.

We have assumed that the mutant and WT can form hybrid trimers when mixed in solution. To verify the hybrid trimer formation, we further conducted co-IP analysis (Figure 3C). To distinguish the mutant and WT monomer, we attached an extended His6 tag (MGSSHHHHHHSSGLVPRGSH) to the N-terminus of R126A. The protein bands of the fusion protein His–R126A (172 amino acids, theoretical molecular mass of 18 kDa) and WT (152 amino acids, theoretical molecular mass of 16 kDa) can be separated by SDS/PAGE. We also tested the enzymatic activities of WT, His–WT, R126A and His–R126A. The attachment of the His6 tag had no effect on the enzymatic activity (WT and His–WT had comparable activity and both R126A and His–R126A had no activity; see Supplementary Figure S3B for details). IP of the WT and His–R126A mixture with the anti-mPGES-1 antibody pulled down all detectable mPGES-1 moieties (both WT and His–R126A) from the supernatant, which can be used to represent the total amount of WT and His–R126A in the mixture. In comparison, IP with the His6-tag antibody pulled down all His–R126A and a large percentage of WT from the supernatant. Because IP of the WT and R126A mixture with the anti-His6-tag antibody could not pull down WT or R126A from the supernatant (see Supplementary Figure S3C for control experiment details), the Western blot analysis using the anti-His6-tag antibody could not detect WT or R126A bands (Figure 3C). There was no cross-reaction between the WT and anti-His6-tag antibody. Therefore pull down of the WT by the anti-His6-tag antibody should be mediated by its interaction with His–R126A, indicating hybrid trimer formation between the mutant and WT. We also performed a parallel experiment using R126A and His–WT mixture for co-IP analysis. The anti-His6-tag antibody pulled down both His–WT and R126A from the supernatant. This result further supported the formation of hybrid trimers of mutant and WT.

In summary, the co-IP analysis confirmed that the mutant and WT can form hybrid trimers in solution, and the two hybridization experiments showed that the activity of the mixture increased with increasing hybrid formation. Therefore the results provided evidence for the one-third-of-the-sites reactivity mechanism.

MD simulation of conformation transition process

We performed simulation A (from pseudo-all open structure) and simulation B (from all closed structure) to further elucidate the conformation transition mechanism and verify the unsymmetrical structure of the trimer (Figure 4). To represent the extent to which the substrate-binding pocket opened, we used the distance between the α-carbon atoms of Leu39 and Arg126, as they were located at the termini of helices 1 and 4 in an approximately horizontal direction. The three pockets in the after-MD structure were named pocket-1′, pocket-2′ and pocket-3′ for simulation A and pocket-1″, pocket-2″ and pocket-3″ for simulation B.

Figure 4 Simulation A (from pseudo-all open structure) and simulation B (from all closed structure)

(A) Superimposition of mPGES-1 (magenta and green) and LTC4S (white). The main difference lay in the bending direction of the cytoplasmic half of helix 1 and helix 2 (active region light magenta) over a hinge fixed by the Lys26–Asp75 salt bridge (arrow). The RMSD fluctuation of Cα atoms of pocket-2′ during simulation A is shown on the right. The maximum (with a light magenta bar below, the bar is coloured following the left-hand panel) corresponds with the active region. (B and C) Simulation A–simulation B. Closing path from open (white) to closed (magenta and green) and opening path from closed (white) to open (magenta and green) suggested by the simulation and the Leu39–Arg126 distance variation during the simulation. The distance between the α-carbon atoms of Leu39 and Arg126 was labelled to illustrate the extent of opening of the substrate-binding pocket. The cofactor GSH is coloured yellow, and the after-MD structure is shown in pink. (D and E) Simulation A–simulation B. Molecular surface images of the mPGES-1 trimer before (top panel) and after (bottom panel) the MD simulations. The three pockets are numbered from left to right.

During simulation A, two pockets closed quickly and transformed to the closed conformation, with only pocket-3′ remaining open. This result suggested that there was a constraint force between the monomers and simultaneous opening of the three pockets was not permitted. Figures 4(D) and 4(E) illustrate the molecular surface representations of the conformations. In the open conformation, the pocket was open, and the cofactor GSH could be observed from outside of the binding pocket. After the MD simulation, the distances between the two monomer pairs decreased and the monomers entirely surrounded GSH making it inaccessible to the substrate. The MD simulation also demonstrated how the conformation transition occurred. As Jegerschöld et al. [19] solved the crystal structure of mPGES-1 in a closed conformation, they speculated that it could transform to the open conformation by bending the cytoplasmic halves of helix 1 and helix 2 (active region) over a hinge fixed by the Lys26–Asp75 salt bridge. The maximal RMSD fluctuation of Cα during the simulation (using pocket-2′ as an example, with the other two pockets showing similar results) corresponded with this region, supporting the proposed transformation mode.

Simulation B indicated that only one pocket (pocket-2″) opened during the MD simulation, whereas the other two pockets remained closed. As our model was constructed by manually bending helix 1 to open the pocket, this simulation validated that our tentative open conformation was transformable from the experimentally determined closed conformation. Furthermore, it suggested that the mPGES-1 trimer tended to open only one pocket at any given time, supporting the one-third-of-the-sites reactivity mechanism.

Superimposition of the after-MD structures and open/closed conformation revealed that, whereas in simulation A the pockets fully transformed to the closed conformation (the Leu39–Arg126 distance was reduced from 13 Å to 8 Å), in simulation B the pocket was only partially open (the Leu39–Arg126 distance grew from 8 Å to 12 Å). To reach the fully open conformation, we attempted to extend simulation B to 100 ns (simulation B+) or manually placed a molecule of PGH2 near one pocket (simulation C) (see Supplementary Figure S3 for details). However, the conformation transition still could not be completed and PGH2 could not be docked in the partially open pocket. Recently, Nury et al. [31] performed 1 μs MD simulations of an ion channel in a fully hydrated membrane environment, and the conformation transition was incomplete at the end of simulation. They speculated that the conformation transition mediated by helix bending was a slow process, especially in the lipid bilayer environment. Therefore although the ideal method of building an open conformation structural model of mPGES1 may be extending simulation B until a fully open conformation was obtained, its implementation was limited to computational demands. To strike a balance between speed and accuracy, we used a manually generated initial configuration combined with a refining-MD simulation strategy to build the open conformation structural model of mPGES-1, as described in the Experimental section. As our model was well supported by the experimental results, this strategy may also be useful for other slow-process studies.

Validation of the pocket-2 structure by the molecular docking of inhibitors

In our model, pocket-2 provided a good representation of the active substrate-binding conformation and therefore should be the target conformation of mPGES-1 inhibitors. We performed a rigorous test of this model by docking all of the collected inhibitors (117 molecules belonging to 19 different types, see Supplementary Table S2) in this pocket and analysed the correlation between the experimental pIC50 and predicted pKd values (Figure 5). The correlation coefficient R reached 0.74, which was good for docking studies based on a large set of inhibitors. This result further supported pocket-2 as an active substrate-binding pocket.

Figure 5 Molecular docking of inhibitors

Published mPGES-1 inhibitors (117 molecules belonging to 19 different types) were docked to the active substrate-binding pocket-2 using the program AutoDock 4. Correlation analysis was performed between experimental pIC50 and calculated pKd values. The correlation coefficient R was 0.74.


In biological systems, proteins often self-associate to form dimers or higher-order oligomers [32]. Protein oligomerization is a key factor in regulating the biological functions of proteins, such as enzymes, ion channels, receptors and transcription factors. In previous studies, the regulatory roles of protein dimerization were extensively explored [3335]; however, much fewer reports concerning trimeric proteins are known. The most difficult problem is the relative paucity of experimental data for trimeric proteins. A search of the Brenda enzyme database (a comprehensive enzyme information system containing 5117 different enzymes, indicates that among 611 human enzymes with PDB entries, 248 enzymes are dimers, whereas only 27 are trimers.

The results of the present study have proved for the first time that the mPGES-1 trimer has one-third-of-the-sites reactivity. Recent studies of MGST-1 indicate that its three pockets have different affinities for GSH and exhibit similar one-third-of-the-sites reactivity [30]. Therefore this enzymatic mechanism might be conserved in the MAPEG family. For the one-third-of-the-sites reactivity of mPGES-1, there are two possible ways of activity regulation: mPGES-1 undergoes breathing motion and randomly opens one pocket to be active (random way) or a sequential catalytic process similar to that of ATP synthase occurs. Although mPGES-1 shares a similar 3-fold structure and one-third-of-the-sites reactivity with ATP synthase, the structure of ATP synthase is much more complex, which consists of at least 22 subunits and is powered by a proton gradient, whether this molecular motor can be accomplished by a simple homotrimer structure remains an open question [36]. Furthermore, the sequential catalytic process implies that inactivating one site will lead to complete inactivation [37]. Our hybridization experiment revealed that hybrid species WMM and WWM had comparable activity with that of the WT enzyme, which indicated that the random way was more plausible.

However, unlike the sequential catalytic process for which it is easy to explain why the mPGES-1 trimer has only one-third-of-the-sites reactivity, if mPGES-1 randomly opens one pocket to be active, why does it require oligomerization to form homotrimers? In principle, one-third-of-the-sites reactivity lowers the capacity of enzyme and as a consequence might not be favoured by evolution. On the basis of the results of the present study and those of previous studies, we propose that in mPGES-1 the active pocket needs inactive pockets for both functional and structural purposes. mPGES-1 relies on an open-to-closed conformation shift to perform its enzymatic reaction; a single pocket may fluctuate in three directions and this flexibility hinders efficient transition motion. A simple solution to this is associating three pockets together to form a homotrimer (Figure 6). Griffin et al. [38] reported that for dihydrodipicolinate synthase, the homotetrameric structure was more rigid than the dimeric form, thus reducing excessive relative motions and improving the enzymatic activity. In mPGES-1, the two inactive pockets may also sterically stabilize the active site and only permit the motion responsible for the conformation transition. This steric stabilization effect may be fulfilled by the close packing of helices 1 and 2 and cofactor GSH, forming an annular interaction network (Figure 6B). Furthermore, mPGES-1 may form homotrimers for structural purposes. In addition to the substrate-binding pocket, the three subunits of mPGES-1 form a central cavity (Figure 6C). Residues located inside the cavity are mainly polar residues such as arginine and glutamic acid. The folding of α-helical membrane proteins has been conceptualized by a two-stage model [39]. The first stage is membrane insertion and secondary structure formation, and the second stage is helix–helix association. Polar residues located in the interface of the subunits can form interhelical hydrogen bonds in the second stage that contribute to protein stability substantially. We believe that these two effects explain the enzymatic mechanism of the mPGES-1 trimer. However, other explanations may also be possible, and further experimental evidence is needed to unravel the dynamics of the entire reaction process.

Figure 6 Possible mechanism for mPGES-1 one-third-of-the-sites activity

(A) Different motion mode for the single pocket (top panel) and associated three pockets (bottom panel). (B) Annular interaction network formed by the close packing of helices 1 and 2 (shown in cylinder) and cofactor GSH (molecular surface representation). (C) Central cavity (black dot) in the mPGES-1 trimer calculated using the program Cavity [40].

In conclusion, we have identified a novel one-third-of-the-sites reactivity mechanism for the mPGES-1 trimer. An open conformation structural model of mPGES-1 was also built, which behaved well in known inhibitor docking and structure–activity relationship studies. This active conformation model of mPGES-1 can be further used in novel inhibitor discovery and other related studies.


Shan He performed the experiments and computational work, analysed data and wrote the paper; Yiran Wu and Daqi Yu guided and optimized the computational work and edited the paper prior to submission; Luhua Lai, the scientific supervisor, initiated the study and conceived the research, edited and approved the final paper.


This work was supported, in part, by the Ministry of Science and Technology of China and the National Natural Science Foundation of China [grant numbers 10721403 and 90913021].


We thank the Shanghai Supercomputer Center of China for providing high-performance computing resources, and Changsheng Zhang and Yaxia Yuan for helpful discussions.

Abbreviations: AA, arachidonic acid; co-IP, co-immunoprecipitation; EIA, enzyme immunoassay; FLAP, 5-lipoxygenase-activating protein; IP, immunoprecipitation; IPTG, isopropyl β-D-thiogalactopyranoside; LTC4S, leukotriene C4 synthase; MAPEG, membrane-associated proteins in eicosanoid and glutathione metabolism; MD, molecular dynamics; MGST, microsomal glutathione transferase; mPGES-1, microsomal prostaglandin E synthase-1; NSAID, non-steroidal anti-inflammatory drug; PGH2, prostaglandin H2; PNP, purine nucleoside phosphorylase; POPC, 1-palmitoyl-2-oleoyl phosphatidylcholine; RMSD, root mean square deviation; WT, wild type


View Abstract