1

MOLECULAR DYNAMICS SIMULATIONS OF THE LIGAND-BINDING DOMAIN OF AN NMDA-RECEPTOR

Samantha L. Kaye, Mark S.P. Sansom and Philip C. Biggin* Department of Biochemistry, University of Oxford, South Parks Road, Oxford, OX1 3QU, U.K.

Running Title: Molecular Dynamics Simulations of NR1 *Address correspondence to: Philip Biggin, Structural Bioinformatics and Computational Biochemistry,

Department of Biochemistry, University of Oxford, South Parks Road, Oxford, OX1 3QU. U.K. Tel: +44 1865 275255; Fax: +44 1865 275273; E-mail: philip.biggin@bioch.ox.ac.uk

The mechanism of partial agonism at N-methyl-D-aspartate (NMDA) receptors is an unresolved issue, especially with respect to the role of protein dynamics. We have performed multiple molecular dynamics simulations (7x20ns) in order to examine the behaviour of the ligand-binding core of the NR1 subunit with a series of ligands. Our results show that water plays an important role in stabilizing different conformations of the core and how a closed-cleft conformation of the protein might be stabilized in the absence of ligands. In the case of ligand-bound simulations with both full and partial agonists, we observed that ligands within the binding cleft may undergo distinct conformational changes, without grossly influencing the degree of cleft-closure within the ligand-binding domain. In agreement with recently published crystallographic data, we also observe similar changes in backbone torsions corresponding to the hinge-region between the two lobes for the partial agonist, D-cycloserine. This observation rationalizes the classification of D-cycloserine as a partial agonist and should provide a basis with which to predict partial agonism in this class of receptor by analyzing the behaviour of these torsions with other potential ligands. N-methyl-D-aspartate (NMDA) receptors are ligand-gated ion channels that play a major role in learning and memory (1). They have also been implicated in a number of injury or disease states including for example, schizophrenia (2-5). NMDA receptors are a distinct subtype of ionotropic glutamate receptor (6) in that they require both glutamate and glycine for activation and membrane depolarization (7). The receptor is a heterotetrameric cation channel typically comprised of two NR1 subunits which contain the

glycine binding site along with two NR2A-D glutamate-binding sites. The heterogeneity of the complex is further extended if the third subtype (NR3), which also binds glycine, is also taken into account. The architecture of each subunit of NMDA receptors is similar to that of non-NMDA receptors. Each subunit is comprised of an amino-terminal binding domain (ATD) which shows homology to the Leucine-Isoleucine-Valine-Binding-Protein (LIVBP) protein (8), 3 transmembrane helices (M1, M2 and M3), a re-entrant P-loop, and a ligand binding domain which is comprised of two discontinuous polypeptide chains S1 and S2 that form two distinct lobes or sub-domains, D1 and D2 (Fig. 1A). An isolated D1D2 ligand-binding core has been shown to bind ligands with similar affinity as wild-type receptors (9,10). Recently, the crystal structure of the ligand-binding domain of the NR1 subunit has been solved by X-ray diffraction with both full (9) and partial (10) agonists. The overall fold and structure (Fig. 1B) is very similar to those observed for crystal structures of the ligand binding domains for the (non-NMDA) GluR2 (11-13), GluR5 (14,15) and GluR6 receptors (15,16). For recent reviews see (17-19). The major difference is an insertion of 30 residues into loop 1 which also contains 2 disulphide bridges. Interestingly, the mechanism of partial agonism in NMDA receptors has been proposed to differ from the mechanism in non-NMDA receptors (10). Partial agonists for NMDA receptors appear capable of promoting a degree of domain closure comparable with that observed for full agonists. Furthermore, whether they are full or partial agonists can be discriminated by the conformation of the second strand (residue 750 to 755) of the hinge region. Whilst the crystallographic

http://www.jbc.org/cgi/doi/10.1074/jbc.M512728200The latest version is at JBC Papers in Press. Published on March 2, 2006 as Manuscript M512728200

Copyright 2006 by The American Society for Biochemistry and Molecular Biology, Inc.

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

2

structures have been a leap forward in our understanding of these receptors, it is important to remember that proteins at physiological temperatures can exhibit important motions that determine water mobility in binding pockets (20) through to large domain motions (21). Our results show not only the importance of receptor dynamics but also how the mechanism of the partial agonist D-cycloserine can be reconciled in light of recent data.

COMPUTATIONAL PROCEDURES

System Preparation. Coordinates for the crystal structures of the NR1 with glycine (1PB7), D-serine (1PB8), D-cycloserine (1PB9), 1-aminocyclopropane-1-carboxylic acid (ACPC) (1Y20) and 5,7-dichlorokynurenic acid (DCKA) (1PBQ) were downloaded from www.rcsb.org (22). A Closed-Apo state was generated by removing the ligand from the glycine-bound structure and an Open-Apo state by removing the ligand from the DCKA-bound structure. The chemical structures of the ligands are shown in Fig. 1C. Table 1 summarizes the simulations. The N and C termini were acetylated and amidated respectively to mimic the continuation of the peptide chain. Missing side chains were built in using the WHATIF “complete a structure” web service (http://swift.cmbi.kun.nl/WIWWWI/). As the initial structures did not have loop 1 resolved in the crystal structure we used Modeller 7v7 (23) to generate the missing residues. Since beginning this research a crystal structure with an intact loop 1 has been solved (PDB code: 1Y20), comparison of the modeled loop is very similar to this crystal conformation for the closed cleft simulations (RMSD ≤1.15 Å) but not for the open cleft ones (RMSD 4.3 Å). This can be explained by the resolved crystal loop structure being from a closed ligand binding domain. Presumably this loop was not originally observed as it has high mobility compared with the rest of the structure. Indeed, the B-factors for loop 1 from the 1Y20 structure are generally higher than for the rest of the protein. The ligands D-cycloserine, ACPC and DCKA were parameterized using PRODRG (24).

Simulation Parameters. The resulting

structures were solvated in a cubic box with

~25,000 SPC (25) waters and counter-ions added to ensure overall electrostatic neutrality. All simulations were performed with Gromacs 3.1.4 (26), with the ffgmx force field and the NVT ensemble at 300K using the Berendsen thermostat. The pressure was coupled with the Berendsen barostat with a coupling constant, tau of 1.0 ps. Electrostatics were calculated using the Particle Mesh Ewald method (27,28) with a short-range cut-off of 10 Å. The time step for integration was 2 fs. The LINCS algorithm (29) was used to restrain bond lengths. Each system was subjected to a 200 ps dynamics run with the protein restrained (10 kjmol-1Å-2 on all heavy atoms). This was followed by 20 ns of free simulation. Except for RMSD calculations only the last 18ns was used for analysis. Hydrogen bonds were calculated with 30o angle and 3.5 Ǻ distance cut-offs. All calculations were performed on a Linux cluster.

RESULTS AND DISCUSSION Protein Motion. An initial evaluation of structural drift is provided by analysis of the Cα atom root mean square deviations (RMSDs) from the initial structures as a function of time. We omitted loops 1 and 2 (residues 413-450 and 486-497 respectively) from this analysis in order to focus on the overall dynamics of the fold. Each simulation shows an initial jump of ~1.5 Å followed by slower relaxation (Fig. 2A). The closed cleft simulations; Gly, DS, DCS, ACPC and Closed-Apo (only the Gly simulation shown in Fig. 2A for clarity) all seem to converge to a similar RMSD (1.75 Å) indicating that the closed-cleft form of the protein exhibits little conformational drift on these timescales. The RMSD of our simulations reveals that the open-cleft forms of the protein (DCKA and Open-Apo simulations) exhibit a much larger conformational drift from the initial structure (up to 4 Å in the case of the Open-Apo simulation). Visual inspection of the Open-Apo simulation revealed that it appears to undergo a transition from an open to a more closed form of the protein between 5 and 8 ns. This is examined in more detail later.

In addition to conformational drift we can also examine flexibility as a function of residue number. Fig. 2B shows the root mean square fluctuations (RMSF) for the Gly and Open-Apo simulations. There are four peaks in all of the

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

3

simulations that show a large degree of fluctuation compared with the rest of the protein. These peaks correspond to residues in two major loops (Fig 1B); residues 416-417, 426-427, and 443-449 are in loop 1 and residues 492-493 are located in loop 2. The first three peaks indicate a large degree of flexibility in the protein structure of loop 1. Eight residues in this region were disordered in the crystal structure of all but ACPC, presumably due to this high flexibility. In the complete receptor, loop 1 has been postulated to interact with a subunit on an adjacent dimer (9) which would stabilize it.

Interestingly, in the Open-Apo simulation, residues 533-544 and 663-737 show a larger degree of fluctuation compared to the other simulations. These residues are located in the D2 sub-domain (residues 533-537 are in the inter-sub-domain strands). If one considers the Cα RMSF of this simulation after the cleft has closed (8-20 ns, data not shown), the degree of fluctuation is reduced, implying that a closed cleft conformation stabilizes this region of the protein.

Unlike in other iGluRs, in the NMDA receptors the extent of sub-domain separation appears not to be linked to agonist efficacy. Both partial and full agonists confer essentially the same degree of sub-domain closure on the ligand-binding domain (although antagonists continue to force the sub-domains apart). Fig. 2C shows the inter-sub-domain separation (calculated as the distance between the centers of masses of the two sub-domains) as a function of time. The closed-cleft simulations (Gly, DS, DCS, ACPC and Closed-Apo) all maintain a similar degree of domain closure (only DS shown for clarity in Fig. 2C). Thus, at least on these timescales, partial agonists do not appear to lead to partial domain opening. The DCKA simulation shows a slight initial closure (about 1Å over ~0.75 ns) to a separation of ~25.5 Å which is maintained for the duration of the simulation. At 5ns the Open-Apo simulation exhibits an initial rapid closing followed by a slower closure to approximately the same degree of domain separation as the DS simulation by ~17 ns.

Essential dynamics analysis of the Open-Apo simulation show that the mechanism of closing is comprised of two major motions. The first is a clamshell closing motion as shown in Fig. 3A. The second is analogous to a twisting of one

side of D1 and D2 toward each other and is shown in Fig. 3B. These motions can occur concurrently. This has been observed before for this class of protein (30,31), thus we are reasonably confident that this motion is genuine.

Although sub-domain closure has been observed before in simulations of these and other proteins (31,32) and evidence suggests that the ligand-binding domain exists in an open/closed equilibrium in the apo state (33) we are cautious not to over-interpret this motion. The simulation of the ligand-binding domain in the absence of the rest of the subunit may have led to a higher likelihood of a closure event due to the lack of an opposing force that would be provided by the transmembrane ion channel domain.

Inter-sub-domain Contacts. In the crystal structures of NR1 there are a limited number of direct contacts formed across the binding left. Gln405 interacts with Trp731 directly and via a water with Asp732. Asn499, Gly485 interact with Gln686 and Lys493 with Glu712. In addition to these, during the simulations, we commonly observe interactions between 13 residues in D1 and 10 residues in D2 (Table 2). Perhaps the most interesting of these are between Thr518 and both Ser688 and Asp732, all of which form part of the ligand binding pocket.

In the absence of a ligand in the binding cleft it might be expected that the sub-domains would open ready to accept a ligand. However, the Closed-Apo conformation that we created by removing the glycine ligand from PDB structure 1PB7 does not undergo any significant increase in degree of sub-domain closure throughout the 20 ns simulation. It is likely that this stability is due to the absence of the transmembrane domain. Since the role of the ligand-binding domain is to open the ion channel, in the absence of the channel there is no opposing force acting to open the ligand-binding domain.

Water in the Binding Pocket. During the

Open-Apo simulation, we observed protein motion that results in a conformation that is more closed than the Closed-Apo structure. However, the difference is small (less than 1 Å) and the extent of the closure is almost identical to that of the D-serine bound structure (Fig. 2C). We are therefore confident that the conformation is genuine.

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

4

In order to characterize the closed conformation in the absence of a ligand more fully we examined the nature of the interactions of water with the binding pocket. As the Closed-Apo simulation is somewhat artificial, we used a section of trajectory (17-20ns) from the Open-Apo simulations that corresponds to a closed form of the protein based on the degree of closure observed (Fig 2C). 175 different hydrogen bonds with 55 unique waters are formed in this time-frame, thus indicating that substantial water exchange is possible in the closed-cleft form of the protein in the absence of ligand.

We looked at key waters in the binding pocket of the ligand-bound simulations in more detail. In particular we have considered the sites of three crystal waters (A, B and C. See Fig. 4A) that were found in the same location in the glycine, D-serine and D-cycloserine bound structures and therefore might play a key role in ligand binding. Sites A and B are positioned between the ligand and D2, C is located proximal to the ligand amino group between D1 and D2.

To define the sites of the crystal waters we considered the charged atoms within 4 Å of each crystal water. Table 3 shows the number of unique waters that enter each of these sites during 20 ns of the Gly, DS, DCS and ACPC simulations.

From these results it is clear that Gly undergoes far more water exchanges than either DS, DCS or ACPC (although it is still a rare event). This is probably due to the larger size of D-serine, D-cycloserine and ACPC.

In the DS simulation both waters A and B quickly leave the binding pocket and are not replaced in 20 ns. For Gly, waters at sites A and B are present but undergo several exchanges during the 20ns. Exchanges for site A in the Gly simulation are shown in Fig. 4B. This is in contrast to the DCS simulation where waters are bound to both sites and undergo fewer exchanges, but which do appear to coincide with ligand movement. Fig. 4C shows the water behaviour corresponding to site A. The behaviour of water in this site may also be linked to alternative orientations of DCS within the binding pocket (see Fig 6B). Overall, this suggests that waters in sites A and B are not important in binding D-serine and that, at least in the case of Gly, the waters may act as a substitute to the H-bonding sidechain of D-serine. There are no waters in site A in the ACPC

crystal but water in site B remains tightly bound with no exchanges throughout 20 ns. C remains tightly bound in all four simulations (data not shown).

The Ligand Binding Pocket. The binding pocket was defined by considering all residues in the D1 and D2 lobes that interact (inter-atomic distance <3.2 Å as also described by (9)) with the ligand in the crystal structures 1PB7, 1PB8, 1PB9, 1Y20 and 1PBQ. This results in a binding pocket defined by Pro516, Thr518, Arg523 which reside in D1 and Ser688 and Asp732 which are part of D2. In addition to these residues we identified 7 new residues which are also capable of interacting with the ligand (Table 4). Three of these (Gln405, Phe484 and Thr486) are located in D1 whilst the others (Ser687, Val689, Trp731 and Ser756) are in D2. None of these contacts were long lived but it is of interest that Ser756, which is in the hinge region, is capable of interacting directly with the ligand as this may suggest a method through which the ligand can directly influence the hinge. The ligand remains in the binding site during all the simulations, although in both the Glycine and D-cycloserine simulations there is movement of the ligand within the binding pocket which affects the hydrogen bonding patterns of the ligand. These ligands were able to adopt different orientations within the binding pocket that did not change the degree of sub-domain separation (Fig 2C). The transitions for both of these ligands were sharp and involved a loss of interaction between the ligand and Pro516.

When in the starting conformation (Fig. 5A), the glycine ligand interacts Asp732, Ser688, Arg523, occasionally with Pro516 and with the amide group of Thr518. In addition, the side chain of Thr518 interacts with Asp732. During the first ~13 ns of simulation we observe two types of ligand movement before settling back into the crystal conformation. Firstly, a rotation of the amino group so that it faces D2, during which the carbonyl group continues to interact with Thr518 (Fig. 5B and D). Secondly, we observe a complete flip where the ligand acts as a rigid body so that both amino and carbonyl groups are facing D2 (Fig. 5C). When this occurs contacts between the ligand and Thr518 are broken and often a water molecule moves into the gap between Thr518 and

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

5

the ligand preventing Thr518 interacting with Asp732 in D2. In the DCS simulation the ligand acts as a rigid body. In the starting conformation D-cycloserine interacts with the amino groups of Thr518 and Arg523 via it’s carbonyl oxygen and with Pro516, Asp732 and the side chain of Thr518 via it’s amino group (Fig. 6A). As in the glycine-bound structure, the Thr518 side chain also interacts with Asp732. When the ligand moves at ~4 ns it continues to interact with all residues in the binding pocket except with Pro516. The amino group of the ligand is able to interact with Ser688 but the positioning of the ligand means that Thr518 and Asp732 no longer interact directly but via the ligand instead (Fig. 6B). This flipped conformation persists to the end of the 20 ns simulation. Hydrogen Bonds in the Hinge. The hinge region of the ligand-binding domain was examined in closer detail as differences between the conformation adopted in the presence of full and partial or ant-agonist were reported. The hinge region is comprised of two strands (residues 537-539 and 750-755) and inter-strand hydrogen bonds in this region differ according to the conformation adopted by the hinge (9).

One key hydrogen bond forms between the carbonyl group of Leu538 and the amino group of Phe753. This is present in all NR1 crystal structures to date and is independent of the ligand bound. The second inter-strand hydrogen bond is variable. In the presence of the full agonists glycine and D-serine, Leu538 forms a second hydrogen bond, between it’s amino group and the carbonyl group of Phe754 (similar to that shown in Fig 7A). However, in the presence of partial agonists ACPC, ACBC and cycloleucine, and the antagonist DCKA, the amino group of Leu538 forms a hydrogen bond with the carbonyl group of Phe753 (similar to that shown in Fig. 7B). Interestingly, the crystal structure with the partial agonist D-cycloserine bound (1PB9) has a hinge hydrogen bond pattern like those seen in the presence of full agonists (Fig. 7A).

In all ligand-bound cases (except DCS) the hinge conformation was unchanged from the starting structure. As might be anticipated in the DCS simulation, the hinge stably adopts a more partial agonist-like conformation with the carbonyl

group of Phe753 facing across the inter-strand region (Fig. 7B). This conformation occurs within ~5ns and persists until the end of the 20 ns simulation. This rotation appears to be closely associated with the observed movement of D-cycloserine in the binding pocket. As can be seen in the plot of the backbone dihedral angle of Phe753 (Fig. 7C), rotations around the backbone occur frequently in the first 5 ns of the simulation. Stabilization occurs within 1 ns of the ligand changing binding mode. These observations provide us with an indication of how partial agonists may confer their mode of action: Partial agonists are able to move within the binding cleft and adopt different binding modes and our results suggest that some of these binding modes affect the conformation of the hinge whilst others do not. It is thus possible to see how a spectrum of partial agonists might act depending on how flexible they are and how readily they can influence the conformation of the hinge. Full agonists on the other hand might exhibit considerable movement (as we observe here for glycine) within the binding pocket, but at the same time exert no effect on the hinge region.

CONCLUSIONS

We have demonstrated here that the dynamics of the NR1 subunit of the NMDA receptor are an important consideration with respect to their mode of action and in particular how partial and full agonists might be distinguished based upon their influence of the local conformations of the binding pocket. In addition, large scale conformational changes were also observed that provide a plausible pathway for a transition between the open-cleft and closed-cleft forms of the domain. The Open-Apo transition and the general motion of the trajectories as decomposed by the eigenvector analysis reveals the principal motions of the ligand-binding core to be a lobe twisting motion and a hinge-motion in agreement with previous observations (30,31). The speed of the closing transition observed during the Open-Apo simulation is fast (nanosecond timescale) and is currently beyond the resolution of experimental techniques for single molecules. This may be important in the context of protein ensembles (21); For example, recent NMR experiments have

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

6

indicated such considerations might be especially relevant for enzyme catalysis (34). The closed-cleft simulations showed little drift in terms of RMSD (Fig. 2A) and thus provided a good opportunity to analyze ligand-protein and binding pocket-water interactions. The movement of the ligands within the site indicates that there is potential for the protein-ligand system to adopt multiple conformations that induce the same degree of domain closure. For the NR1 subunit, unlike the AMPA receptor subunit GluR2, recent crystallographic evidence has suggested that partial agonism is not related to the extent of cleft closure but rather to the efficacy of the ligand to induce a change in the conformation of the hinge region. In the crystal structure of NR1 complexed with the partial agonist D-cycloserine, the conformation of the hinge region was found to be “full agonist like”. Our simulations here show that D-cycloserine may in fact be capable of inducing the change in backbone hydrogen bonding of the inter-sub-domain strands that was found for other partial agonists crystallized (10). We also show that partial agonists are able to maintain full domain closure in the NR1 subunit and do not lead to partial domain opening. It is also interesting to note that although the glycine ligand was able to move substantially (changing hydrogen bonds with the binding pocket), it did not induce the change in the inter-domain strand backbone and thus remains consistent with the profile of a full agonist. Thus we suggest that molecular dynamics simulations of the compounds within the receptor will be a useful tool in predicting agonist efficacy, by examining the influence of ligands on the local conformational dynamics of the hinge region.

REFERENCES 1. Nakazawa, K., Quirk, M. C.,

Chitwood, R. A., Watanabe, M., Yeckel, M. F., Sun, L. D., Kato, A., Carr, C. A., Johnston, D., Wilson, M. A., and Tonegawa, S. (2002) Science 297, 211-218

2. Tsai, G., and Coyle, J. T. (2002) Ann. rev. Pharmacol. Toxicol. 42, 165-179

3. Hashimoto, K., Fukushima, T., Shimizu, E., Komatsu, N., Watanabe, H., Shinoda, N., Nakazato, M., Kumakiri, C., Okada, S.-I., Hasegawa, H., Imai, K., and Iyo, M. (2003) Arch Gen Psychiatry 60, 573-576

4. Laruelle, M., Kegeles, L. S., and Abi-Dargham, A. (2003) Ann. N.Y. Acad. Sci. 1003, 138-158

5. Mueller, H. T., Haroutunian, V., Davis, K. L., and Meador-Woodruff, J. H. (2004) Mol. Brain. Res. 121, 60-69

6. Dingledine, R., Borges, K., Bowie, D., and Traynelis, S. F. (1999) Pharmacol. Rev. 51, 7-61

7. Cull-Candy, S., Brickley, S., and Farrant, M. (2001) Curr. Opin. Neurobiol. 11, 327-355

8. Sack, J. S., Saper, M. A., and Quiocho, F. A. (1989) J. Mol. Biol. 206, 171-191

9. Furukawa, H., and Gouaux, E. (2003) EMBO J. 22, 2873-2885

10. Inanobe, A., Furukawa, H., and Gouaux, E. (2005) Neuron 47, 71-84

11. Armstrong, N., Sun, Y., Chen, G.-Q., and Gouaux, E. (1998) Nature 395, 913 - 917

12. Armstrong, N., and Gouaux, E. (2000) Neuron 28, 165-181

13. Hogner, A., Kastrup, J. S., Jin, R., Liljefors, T., Mayer, M. L., Egebjerg, J., Larsen, I. K., and Gouaux, E. (2002) JMB 322, 93-109

14. Naur, P., Vestergaard, B., Skov, L. K., Egebjerg, J., Gajhede, M., and Kastrup, J. S. (2005) FEBS Letts. 579, 1154-1160

15. Mayer, M. L. (2005) Neuon 45, 539-552

16. Nanao, M. H., Green, T., Stern-Bach, Y., Heinemann, S. F., and Choe, S. (2005) Proc Natl Acad Sci USA 102, 1708-1713

17. Madden, D. R. (2002) Nat. Rev. Neurosci. 3, 91-101

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

7

18. Gouaux, E. (2003) J. Physiol. 554.2, 249-253

19. Mayer, M. L. (2005) Curr. Opin. Neurobiol. 15, 282-288

20. Arinaminpathy, Y., Sansom, M. S. P., and Biggin, P. C. (2006) Mol. Pharm. 69, 11-18

21. Gerstein, M., Lesk, A. M., and Chothia, C. (1994) Biochemistry 33, 6739-6749

22. Berman, H. M., Westbrook, J., Feng, Z., Gilliland, G., Bhat, T. N., Weissig, H., Shindyalov, I. N., and Bourne, P. E. (2000) Nucleic Acids Res. 28, 235-242

23. Fiser, A., and Sali, A. (2003) Meths. Enzym. 374, 461-491

24. Schuettelkopf, A. W., and van Aalten, D. M. F. (2004) Acta Cryst. D 60, 1355-1363

25. Hermans, J., Berendsen, H. J. C., van Gunsteren, W. F., and Postma, J. P. M. (1984) Biopolymers 23, 1513-1518

26. Lindahl, E., Hess, B., and D., v. d. S. (2001) J. Mol. Model 7, 306-317

27. Darden, T., York, D., and Pedersen, L. (1993) J. Chem. Phys. 98, 10089-10092

28. Essman, U., Perera, L., Berkowitz, M. L., Darden, T., Lee, H., and Pedersen, L. G. (1995) J. Chem. Phys. 103, 8577-8593

29. Hess, B., Bekker J, Berendsen HJC and Fraaije JGEM. (1997) J Comput Chem 18, 1463-1472

30. Pang, A., Arinaminpathy, Y., Sansom, M. S. P., and Biggin, P. C. (2003) FEBS Lett. 550, 168-174

31. Pang, A., Arinaminpathy, Y., Sansom, M. S. P., and Biggin, P. C. (2005) Proteins: Struct. Funct. Bioinf. 61, 809-822

32. Roccatano, D., Mark, A. E., and Hayward, S. (2001) 2001 310, 1039-1053

33. Madden, D. R., Armstrong, N., Svergun, D., Perez, J., and Vachette, P. (2005) JBC 280, 23737-23642

34. Eisenmesser, E. Z., Millet, O., Labeikovsky, W., Korzhnev, D. M., Wolf-Watz, M., Bosco, D. A., Skalicky, J. J., Kay, L. E., and Kern, D. (2005) Nature 438, 117-121

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

8

FOOTNOTES SLK acknowledges the MRC for a studentship. MSPS and PCB thank the Wellcome Trust for support. We also thank the Oxford Supercomputing Centre.

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

9

Tables. Table 1. Summary of simulations. Simulation* PDB

code

Ligand bound Agonist

Classification

Crystallographic

Resolution (Å)

Gly 1PB7 Glycine Full 1.35

DS 1PB8 D-serine Full 1.45

DCS 1PB9 D-cycloserine Partial 1.60

ACPC 1Y20 1-aminocyclopropane-1-carboxylic acid Partial 1.40

DCKA 1PBQ 5,7-dichlorokynurenic acid Antagonist 1.90

Closed-Apo 1PB7 None 1.35

Open-Apo 1PBQ None 1.90

*All simulations were 20ns. Table 2. Common inter-domain contacts occurring in at least 3 of the 7 simulations. D1 residue D2 residue

Glu406 Gln742/Lys743

Thr442 Ser713

Ser443 Ser713

Lys483 Lys685

Phe484 Gln686

Thr486 Gln686

Gln487 Arg695

Glu488 Gln686/Arg694

Thr518 Ser688/Asp732

Asn520 Ser688/Arg695

Asn521 Arg695

Glu522 Arg695

Arg523 Gln686/Ser687/Ser688

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

10

Table 3: Number of waters that enter specific sites in the ligand binding pocket in 20 ns.

Table 4: Novel ligand binding residues. Residue and Lobe Gly DS DCS ACPC DCKA

Gln405 (D1) Yes Yes Yes No No

Phe484 (D1) No No Yes No No

Thr486 (D1) No No No No Yes

Ser687 (D2) Yes No No No No

Val689 (D2) Yes No No No No

Trp731 (D2) Yes No Yes No No

Ser756 (D2*) No No Yes No No

* = hinge residue. Figures. Figure 1. (A) shows a schematic of the overall fold of an NMDA subunit. (B) shows a cartoon of the

ligand-binding domain based upon the starting structure of our Gly simulation with loop 1 modelled. (C)

shows chemical structures of the ligands.

Figure 2. (A) Root mean square deviation (RMSD) of the Cα atoms excluding loops 1 and 2 (residues

413 to 450 and 486 to 497 respectively). Of the closed-cleft simulations (Gly, DS, DCS, ACPC and

Closed-Apo), only Gly is shown for clarity (black). DCKA is shown as a dark grey line and Open-Apo as

a light grey line. (B) shows the RMSF the Gly (black line) and the Open-Apo (light grey line) (C)

shows the inter-sub-domain separation as defined as the distance between the centers of mass of D1 and

D2 Cα atoms (not including loop residues). Gly (black line), DCKA (dark grey line) and the Open-Apo

(light grey line).

Subsite Gly DS DCS ACPC

A 7 2 2 0

B 8 2 2 1

C 3 1 4 1

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

11

Figure 3. Reduced representations of the Cα trace showing (A) eigenvector 1 (a clamshell closing

motion) and (B) eigenvector 2 (a twisting of one side of D1 and D2 toward each other). The arrows on

the protein indicate the vector of movement for the average of 5 Cα in that region. The grey arrows

indicate the general direction of the movement.

Figure 4. (A) Ligand binding pocket of NR1 with Glycine showing the position of crystal waters in sub-

sites A, B and C. For the analysis of the trajectories, sub-sites were defined as protein atoms that were

within 4 Å of the crystal waters. We then measured the distance of water oxygens to these center of

masses. (B) shows results from the A sub-site of the Gly simulation. The crystal water diffuses away

from the binding pocket during the equilibration phase and is replaced by W388 (at t = 940 ps) which is a

crystallographic water located on the surface of the protein ~ 28 Å away from sub-site A in the crystal

structure. W2121 and W4204 are waters added during the solvation process that diffuse into the binding

pocket during the course of the simulation. (C) shows results for the A sub-site for DCS simulation.

W387 is the crystallographic water located in sub-site A whilst W319 is the crystallographic water

originally observed in sub-site B.

Figure 5. (A) Conformation and orientation of glycine within the binding pocket at the start of the

simulation (t = 0 ns). (B) A snapshot (t = 4 ns) whereby glycine changes its N-Cα-C-O torsion to change

its orientation within the binding pocket. (C) A snapshot of glycine in a flipped conformation (t = 11 ns)

whereby the amino group no-longer interacts with P516, but continues to interact with D732. (D)

Behaviour of the N-Cα-C-O torsion angle of the glycine ligand.

Figure 6. (A) Snapshot of D-cycloserine (indicated as DCS) in the binding pocket at the start (t = 0 ns)

and (B) at t = 4 ns. The amino group of DSC is seen to swing away from P516.

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

12

Figure 7. (A) Hydrogen pattern at the start of the simulation between the L538 and F753/F754. In this

conformation the L538 amino group is hydrogen bonded to the F754 carbonyl group. (B) The change in

hydrogen-bonding after the D-cycloserine had flipped (at t = 4 ns. See Figure 5B). L538 amino group is

now hydrogen-bonded to F754 carbonyl group. (C) The behaviour of the torsions of F753 shows a

distinct transition for the phi torsion angle (dark grey) and the psi torsion angle (light grey). There is also

a distinct transition for the phi torsion angle of F754 (black).

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

A Figure 1

B

COO

NH3+

OO

NH3+

OH

ONO

NH3+

NH

O

O

O Cl

Cl

glycine D-serine

D-cycloserine 5,7-dichlorokynurenic acid

IN

OUT

M3M1 M2

ATD

D1

D2

P

CTD

D1

D2

loop 2

loop 1

ACPC

O O

H3N+

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

A Figure 2

0 10 20time (ns)

0

1

2

3

4

5R

MS

D (Å

)

8 184 6 1614122

B

394 444 494663

713 763 8130

1

2

3

4

5

6

7

RMSF

(Å)

residue numberC

22

24

26

28

30

time (ns)0 10 208 184 6 1614122

inter

-lobe

sepa

ratio

n (Å

)

loop 1 loop 2{ {

544

D2 (535-758)D1 D1

GT linker

Gly

DCKA

Open-Apo

GlyOpen-Apo

Open-Apo

Gly

DCKA

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

A Figure 3

B

D1

D2

D1

D2

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

AFigure 4

B

C

AB

CGly

P516T518

R523

S688

D732

0 5 10 15 20time (ns)

0

4.0

2.0

3.0

1.0

Dis

tanc

e (Å

)

W 388W 2121W 4204

0 5 10 15 20time (ns)

0

4.0

2.0

3.0

1.0

Dis

tanc

e (Å

)

W 319W 387

DCS flip

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

A Figure 5

B

C

time (ns)

R523

P516T518

D732S688

R523

P516T518

D732S688

GLY

GLY

t = 0 ns

t =11 ns

0 5 10 15 20-200

-100

0

100

200

tors

ion

angl

e (°

)

D

R523

P516T518

D732

S688

GLY

t =4 ns

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

A Figure 6

B

R523 P516

T518

D732S688

DCS

t = 0 ns

R523

P516T518

D732S688

DCS

t = 4 ns

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

AFigure 7

B

t = 0 ns

L538F754

F753

t = 5 ns

L538

F753

F754

C

0 5 10 15 20time (ns)

-200

-100

0

100

200

300

Ang

le (°

)

F753 ψ

F753 φ

F754 φ

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from

Samantha L. Kaye, Mark S.P. Sansom and Philip C. BigginMolecular dynamics simulations of the ligand-binding domain of an NMDA-receptor

published online March 2, 2006J. Biol. Chem. 

  10.1074/jbc.M512728200Access the most updated version of this article at doi:

 Alerts:

  When a correction for this article is posted• 

When this article is cited• 

to choose from all of JBC's e-mail alertsClick here

by guest on October 13, 2020

http://ww

w.jbc.org/

Dow

nloaded from