QUARTERLY

The oligomeric metalloenzymes protein phosphatases dephosphorylate OH groups of Ser/Thr or Tyr residues of proteins whose actions depend on the phosphorus signal. The catalytic units of Ser/Thr protein phosphatases 1, 2A and 2B (PP1c, PP2Ac and PP2Bc, respectively), which exhibit about 45% sequence similarity, have their active centers practically identical. This feature strongly suggests that the unknown structure of PP2Ac could be successfully homology-modeled from the known structures of PP1c and/or PP2Bc. Initially, a theoretical model of PP1c was built, including a phosphate and a metal dication in its catalytic site. The latter was modeled, together with a structural hydroxyl anion, as a triangular pseudo-molecule (Zno or Mno), composed of two metal cations (double Zn2+ or Mn2+, respectively) and the OH- group. To the free PP1c two inhibitor sequences R29RRRPpTPAMLFR40 of DARPP-32 and R30RRRPpTPATLVLT42 of Inhibitor-1, and two putative substrate sequences LRRApSVA and QRRQRKpRRTI were subsequently docked. In the next step, a free PP2Ac model was built via homology re-modeling of the PP1c template and the same four sequences were docked to it. Thus, together, 20 starting model complexes were built, allowing for combination of the Zno and Mno pseudo-molecules, free enzymes and the peptide ligands docked in the catalytic sites of PP1c and PP2Ac. All models were subsequently subjected to 250-300 ps molecular dynamics using the AMBER 5.0 program. The equilibrated trajectories of the final 50 ps were taken for further analyses. The theoretical models of PP1c complexes, irrespective of the dication type, exhibited increased mobilities in the following residue ranges: 195-200, 273-278, 287-209 for the inhibitor sequences and 21-25, 194-200, 222-227, 261, 299-302 for the substrate sequences. Paradoxically, the analogous PP2Ac models appeared much more stable in similar simulations, since only their "prosegment" residues 6-10 and 14-18 exhibited an increased mobility in the inhibitor complexes while no areas of increased mobility were found in the substrate complexes. Another general observation was that the complexes with Mn dications were more stable than those with Zn dications for both PP1c and PP2Ac units.

261, 299-302 for the substrate sequences.Paradoxically, the analogous PP2Ac models appeared much more stable in similar simulations, since only their "prosegment" residues 6-10 and 14-18 exhibited an increased mobility in the inhibitor complexes while no areas of increased mobility were found in the substrate complexes.Another general observation was that the complexes with Mn dications were more stable than those with Zn dications for both PP1c and PP2Ac units.
The reversible phosphorylation/dephosphorylation of proteins by protein kinases/phosphatases is the most important switch in the regulation of a majority of cellular activities, from growth and metabolism to memory.Unlike the kinases, all showing a uniform structure, there are several families of protein phosphatases (PP) with little or no sequence and structural mutual homology (Jia, 1997).Thus, the eukaryotic protein tyrosine phosphatases differ from the serine/threonine phosphatases, which further subdivide into subclasses (Barford, 1996).Initially, the serine/threonine phosphatases were classified into four groups PP1, PP2A, PP2B and PP2C on the basis of differences in their biochemical properties (Ingebritsen & Cohen, 1983).To date it is known that the functional enzymes work in hetero-associates, composed of the catalytic subunit and one or more variable regulatory subunits, which are tissue dependent and a target substrate (Barford, 1996;¯o³nierowicz & Hemmings, 1996;Cohen, 1997;Klee et al., 1998;Berndt, 1999;Millward et al., 1999;Aggen et al., 2000).It is also known that the catalytic units of PP1, PP2A and PP2B (PP1c, PP2Ac and PP2Bc, respectively) are homologous in sequence, structure and mechanism of action, and differ from that of PP2C (Barford, 1996;Jia, 1997;Herzig & Neuman, 2000).The latter differences were evidenced by experiments on solution of PP1c structure (Egloff et al., 1995;1997;Goldberg et al., 1995), PP2B (Griffith et al., 1995;Kissinger et al., 1995) and PP2C (Das et al., 1996).
The PP1 and PP2B catalytic subunits consist of a central b sandwich made of a sheet of six b strands, covered by three a helices and three other b strands on the lower side, and a sheet of five b strands, covered on its top by an all-helical domain.The two metal ions, separated by about 3.5 Å, are coordinated by a phosphate and several strictly conserved His, Asp, Asn and Arg residues provided by the two faces of the b sandwich.To-gether, they define the deeply nested catalytic center, flanked from its upper side by a unique babab scaffold, comprising three b strands from the upper slice of the b sandwich and two a helices from the all-helical domain (Fig. 1).A fairly long substrate trench, kinked by about 120°in its midpoint at the PP1c active center and thus divided into a pre-phosphorylation N-terminal part and a post-phosphorylation C-terminal section, is incoming from the south-east and continuing to the west (Fig. 1).In PP1c the N-terminal groove is highly acidic and the C-terminal groove is hydrophobic yet this is not the case in the homologous PP2Ac and PP2Bc (Goldberg et al., 1995).The reported PP1c and PP2Bc structures contain iron, manganese and/or zinc ions, depending on crystallization conditions (Egloff et al., 1995;1997;Goldberg et al., 1995;Griffith et al., 1995;Kissinger et al., 1995).There is a debate as to what cations support the native PP1, although the most likely candidate is the combination of Zn 2+ and Fe 2+ (Chu et al., 1996;Cohen, 1997).
The purpose of this work is an endeavor to study the delicate problem of reliable molecular simulations of the catalytic subunit of serine/threonine phosphatases in particular, and metalloenzymes in general.So far, there are only a few PP1c simulations reported, with the metal dications treated quite crudely, e.g. by Lindvall et al. (1997).A simple and workable parameterization and implementation of the metal cations is a particular objective of this project.We propose and test a concept of a triangular pseudo-molecule, made of two metal cations and a structural OH -so parameterized regarding the geometry, force constants and charges as to adequately mimick the properties of the experimental PP1c complex (Egloff et al., 1995;Barford, 1996).For this test case we have chosen two uniform metal di-cations, viz.2Zn 2+ and 2Mn 2+ , easy enough to parameterize, adequate for structure compari-sons, and reported to be active in vitro, when incorporated into PP1 (Chu et al., 1996).The reason for not including a more realistic combination of both Zn 2+ and Fe 2+ (Chu et al., 1996;Cohen, 1997) in a single di-cation is that such a "nonsymmetrical" double-metal center is a quite hard case for exact geometry-optimization and subsequent parameterization.Subsequently we studied the effect of these two uniform metal di-cations on binding inhibitor sequences, the phosphorylated dodecapeptide R 29 RRRPpTPAMLFR 40 (since here on: acronym DAR), a fragment of DARPP-32 (Dopamine-and cyclic AMP-regulated neuronal phosphoprotein; Williams et al., 1986) and the phosphorylated tridecapeptide R 30 RRRP-pTPATLVLT 42 (acronym I-1) a fragment of inhibitor 1 (Inh-1) (Huang & Glinsmann, 1976).DARPP-32 (202 amino-acid residues) and Inh-1 (171 residues), although quite differently distributed in mammalian organisms, are mutually homologous, potent and selective protein inhibitors against PP1c and much weaker ones against PP2A and PP2B.A phosphorylated threonine (T34 in DARPP-32 and T-35 in Inh-1) is a prerequisite of their inhibitory activity.Their minor modifications (e.g.pT35D mutation in Inh-1) confer on them a comparative inhibitory power against both PP1 and PP2A (Herzig & Neuman, 2000;Endo et al., 1996).In addition, we also studied the effect of binding of two putative substrate sequences LRRApSVA (acronym 7-pep) and QRRQ-RKpSRRTI (acronym 11-pep), of which the latter is a PP1c substrate while the former is an excellent PP1Ac substrate (¯o³nierowicz, personal communication).It has recently been shown that the sequence around pT 34 in DARPP-32 (Kwon et al., 1997;Huang et al., 1999) and pT 35 in I-1 (Endo et al., 1996) interacts with PP1c at a site other than the interaction site for the (R/K)(V/I)XF consensus sequence, conserved in all PP1 regulatory units, close to N-termini in DARPP-32 and I-1, whose docking site in PP1c has recently been determined (Egloff et al., 1997) to be in the lower-right hand side of PP1c in Fig. 1  All ab initio calculations were done using the GAMESS MO package (Schmidt et al., 1993) and the triple zeta basis set plus d polarisation functions on all heavy atoms (Lee et al., 1996).The partial charges for new residues were derived using the RESP (restrained electrostatic potential) procedure (Bayly et al., 1993), as recommended by the AMBER 5.0 manual (Case et al., 1997).For the non-standard residues phosphothreonin, phosphoserine and PO 4 3-the force-field (Lennard-Jones, force constants) parameters were taken from the AMBER 5.0 (Case et al., 1997) parameter library while the missing charges were calculated as above, except that the 6-31G ** basis set was used for the computation.The resulting new parameters are given in Fig. 2 in the paper by WoŸniak et al. (2000).

Design and parameterization of Zno and Mno residues, the pseudo-molecules composed of 2Zn
Starting models and simulations for PP1c×Zno×Tpo or PP1c×Mno×Tpo.The chain A of the PP1c structure with a covalently attached marine toxin microcystin at C 273 (Goldberg et al., 1995), deposited in the Brookhaven Protein Data Bank (PDB; Bernstein et al., 1977) in file 1fjm, was chosen as a template for the starting structures of the PP1c with two metal cations and the phosphate in the active center.The removal of microcystin and the copying of the missing PO 4 3from PP2Bc (chain A in file 1tco; Bernstein et al., 1977) onto the active center of PP1c via a superimposition of the PP1c and PP2Bc catalytic centers, have been described in detail by WoŸniak et al. (2000).Having the PO 4 3-(Tpo) configuration copied to the PP1c catalytic center, the two metals plus the structural OH -were defined as a Zno residue (see above) and the free PP1c was initially relaxed to release all intramolecular bad contacts and/or those involving Zno and structural water(s).This relaxation consisted of the first three stages in Table 1, executed in vacuo.The most involved stage 2 (the constrained simulated annealing protocol, CSA) was executed using protocols, described in detail elsewhere (WoŸniak et al., 2000) with constraints individually customized to each of the analyzed systems.Subsequently, the system was immersed in the box of TIP3P water (Jorgensen et al., 1983) of a size enabling a cutoff for electrostatic interactions at 14 Å, and all consecutive stages (cf.  1, which was then repeated and followed by all consecutive stages. Starting models and simulations for PP2Ac×Zno×Tpo or PP2Ac×Mno×Tpo.Based on the PP1c, PP2Ac and PP2Bc sequence alignment (Goldberg et al., 1995) and using SYBYL 6.2 (1996) program package, the protein sequence typical of PP1c comprised within the PP1c×(Zno/Mno)×Tpo was computer mutated into that of PP2Ac to give PP2Ac×(Zno/Mno)× Tpo, retaining the molecular architecture of the former.To obtain a full agreement between the linear and the three-dimensional PP1c-PP2Ac alignment, the first six residues from PP1c were removed and two more deletions (Y 114 -P 115 and K 221 ) and one insertion of G (between Y 144 and N 145 ) on the PP1c sequence/structure were introduced and the associated strains initially relieved using force field and tools implemented in SYBYL.Six structural water molecules, present in the PP1c experimental template, were removed from the PP2Ac star model.The initial PP2Ac model was superimposed onto the PP1c starting model for control of the agreement of the conserved residues.Further procedures, starting from the next minimalization on, involved AMBER and followed those described for PP1c×Zno×Tpo and PP1c×Mno×Tpo in the former section.

Vol. 48
Protein phosphatases 1 and 2A 39  plexes.The docking and subsequent simulation procedures are described in detail for the PP1c×Zno×DAR complex taken as a prototype.The dodecapeptide sequence from the PP1c inhibitor/regulatory unit DARPP-32, with the phosphothreonine-34 in the center, R 29 RRRPpTPAM-LFR 40 (DAR), was manually docked with the assumption of the maximal complementarity between the substrate cleft and the bioligand (Goldberg et al., 1995).Thus, the RRRR sequence was fitted into the N-terminal acidic section of the cleft, the phosphate from the phosphorylated threonine was constrained exactly to the initial position occupied by the free phosphate (see above), and the consecutive residues were docked into the C-terminal hydrophobic section of the substrate cleft.The precise initial fit was achieved using a combined constrained minimization and constrained simulated annealing (CSA), in accordance with the protocol included in Table 1.The Mno complex was obtained by a computer mutation of Zno into Mno after stage 3 in Table 1.Full relaxation past stage 3 of Table 1 was achieved similarly as for the free PP1c.Other PP1c and PP2Ac complexes.The starting complexes with other ligands were obtained via proper computer mutations of either PP1c× (Zno/Mno)×DAR or PP2Ac×(Zno/Mno) ×DAR complexes, followed by CSA with gradually released constraints, carefully selected and typical for each individual case, as described above for the specific example of PP1c×Zno×DAR.Subsequently, the productive simulations were carried out accordingly.
Computations and display.Model building was done using SYBYL (1996) program package.All molecular simulations were done using AMBER 5.0 program (Case et al., 1997) on either the POWER CHALLENGE SGI computer (at the Academic Computer Center in Gdañsk TASK) or T3E computer (at the Interdisciplinary Center for Mathematical and Computational Modeling, ICM, at the University of Warsaw). Figure 1 was prepared using the MOLMOL program (Koradi et al., 1996) Figs. 2 and 4-7 using the PlotMTV program (Toh, 1995) and Fig. 3 using the Rasmol program (Sayle, 1996).

Molecular models of the PP1c and PP2Ac catalytic units
Both our PP1c and PP2Ac models should basically be both compatible with and diverse from their respective twins modeled formerly (Gauss et al., 1997;Aggen et al., 1999).The reason for compatibility is the same strategy of modeling PP2Ac from the PP1c solid state template (Goldberg et al., 1995), taking advantage of their 50% sequence homology.A reason for diversity is the level of relaxation of the models in this work, being far more extensive (including about 210-230 ps unconstrained MD runs at 300K in periodic box filled with water) than in the work quoted (5 ps dynamics/minimization at 300K in vacuo, with the enzyme body more than 7-8 Å away from the ligand completely constrained, and the metal ions excised from the catalytic center).Although a direct comparison of the two models is impossible without the respective sets of coordinates, one can believe that the former models of PP1c and PP2Ac must very closely resemble the solid-state structure of PP1c (Goldberg et al., 1995) because of their negligible MD relaxation.Hence, our comparisons of the resulting enzyme-ligand complexes with the starting ones (see below) should actually also reflect a level of similarity/diversity with those from the former work (Gauss et al., 1997;Aggen et al., 1999).

Unoccupied catalytic units
The four maps for the "free" PP1c×(Zno/ Mno)×PO 4 3-and PP2Ac×(Zno/Mno)×PO 4 3-catalytic units are given in Fig. 2. A summary of the enzymes' amino-acid residues, whose C a carbons have moved in MD by more than 5.5 Å away from their starting positions, is given in Table 2 while the sequence-averaged C a -based root-mean-square (RMS) deviations of the final versus the starting snapshots in the MD simulations are given in Table 3.It is seen that the unoccupied PP1c with the Zno dication (RMS = 3.63 Å) appears to be the most flexible of all the systems under study (see below, Fig. 2 vs. 4-7 and Table 3) and, in particu-

Table 2. A summary of the motional characteristics of the models under study a
alizes that the PP1c model of the free enzyme is based on the experimental structure of the covalent PP1c-inhibitor complex (Goldberg et al., 1995), while the PP2Ac unit is clearly homologymodeled from the PP1c template.Furthermore, while the experimental starting model of PP1c comprised six structural water molecules, the homology-modeled PP2Ac was devoid of this water.However, inspection of the four fully relaxed PP2Ac variants demonstrated that they accommodated 3-4 structural water molecules in the sites analogous to those in PP1c.In these four variants of unoccupied enzyme units, the relaxed active centers are well conserved.This can be inferred from a careful inspection of the data in Table 2, keeping in mind that the PP1c active center is defined by the following 12 residues: D64, H66, D92, D95, R96, N124, H125, H173, D208, R221, H248 and Y272, and the probable PP2Ac active center is, by homology, defined by their respective counterparts: D57, H59, D85, D88, R89, N115, H116, H165, D200, R212, H239 and Y263 (Goldberg et al., 1995).It is seen that none of these residues are listed in Table 4 for either free PP1c or PP2Ac.This is confirmed by the low RMS values in Table 4 between the starting and the relaxed geometries of the active centers of the two enzymes.
The deviations between the starting and the relaxed configurations range from 1.05 Å for PP2Ac×Zno×PO    The values relating to the differences between the respective starting and MD-relaxed configurations are on the diagonal.

Ligand-substituted catalytic units
The MD trajectories for the enzymes docking four peptide ligands DAR, I-1, 7-pep and 11-pep, each in the combinations of the PP1c and PP2Ac enzyme pairs with either Zno or Mno dications, are given as the consecutive sets of four maps in Figs.4-7, respectively.A summary of the enzymes amino-acid residues, whose C a carbons have moved in MD by more than 5.5 Å away from their starting positions, is continued in Table 2.The sequence-averaged C a -based RMS deviations of the final snapshots from the starting ones in the MD simulations are given in Table 3.The last two columns in Table 2 indicate the motions common for the Zno and Mno-comprising PP1c and PP2Ac, while the last two rows indicate the motions common for the two inhibitor-enzyme and the two substrate-enzyme complexes.As can be seen from Table 3, greater flexibility in the PP1c than in the PP2Ac complexes is regularly observed, which is easy to infer from the comparison(s) among the relevant RMS values in Table 3 and among the relative numbers of amino-acid residues experiencing excessive motions in specific PP1c-ligand complexes versus PP2Ac-ligand complexes, as demonstrated in Table 2 and Figs.4-7.No significant differences in the motional characteristics between the Zno-and Mno-comprising catalytic units were observed, irrespective whether it was the PP1c or PP2Ac catalytic unit.Contrary to the overall flexibilities, the relaxed active centers are usually well conserved, which is confirmed by the relatively low RMS values (Table 5), being a measure of similarities among the 16 relaxed and starting active center configurations simultaneously for both PP1c and PP2Ac complexes.The deviations between the starting and the relaxed configurations range from 0.84 Å for the PP1c×Mno×DAR complex to 2.62 Å for the PP1c×Mno×I-1 complex (see the diagonal in Table 5) and those among the relaxed configurations range from 0.66 Å between the PP2Ac×Zno×7-pep and PP2Ac×Mno×I-1 complex to 2.06 Å between the PP1c×Zno×7-pep and PP2Ac×Mno×11-pep complex (see the upper triangle in Table 5).The differences between the free and the ligand-occu-pied active centers, although not shown, are of a comparable order.From the inspection of Figs. 2 and 4-7 one can easily see the time points at which the (Zno/Mno)×(phosphate/ligand) constraints (at 60-90 ps) were released (see Methods).The re-laxed structures of the free PP1c units, either with Zn 2+ or Mn 2+ dication center in the active sites differ to a considerable extent from the starting (x-ray) structure of PP1c (see Tables 2, 3 and Figs.2A and 2B).Apart from the PP1c N-and C-termini, the most significant differences and ac-Vol.48 Protein phosphatases 1 and 2A 45   toxin, here the empty hydrophobic groove gets partly filled with the dropping helix D due to a hinge-like feature of the aD-aE loop.Helix D, while lowered towards the hydrophobic trough, may cause helix G with the aG-aH loop, and the b8-b9 loop on the opposite sides of itself to move and deform most extensively (cf.Fig. 1).Interestingly, the pattern of interactions between Zno/ Mno and the conserved residues pointed out above is essentially unperturbed (not shown).Contrary to the free PP1c subunits, the complexed PP1c structures, both with the Zno or Mno dication center in the active sites, very little differ from the starting (x-ray) structure of PP1c (see Figs. 4AB-7AB).For instance, in the PP1c×Zno × DAR complex, apart from the N-terminal prosegment and the C-terminus, the only noticeable motion (up to 3.5 Å) is experienced by Asp197-Glu198 in the C-terminal section of the aG-aH loop (see Fig. 4A vs. Fig. 1 and Table 2).Similarly, the pattern of interactions between Zno/Mno and the conserved residues, and the orientation of the di-nuclear ion center relative to "canonical" orientation observed in the crystal structures (Egloff et al., 1995;Griffith et al., 1995;Kissinger et al., 1995) remain essentially unperturbed (not shown).The relaxed DAR and I-1 inhibitor ligands comply with the demand for, on the one hand, good interactions between their four Arg residues and the acidic groove and, on the other, for packing their mainly non-polar C-termini in the hydrophobic groove (Goldberg et al., 1995).However, the detailed interactions between any specific peptide ligand and the PP1c×Zno vs. the same ligand and PP1c×Mno model differ considerably (not shown).
As concern the PP2Ac×(Zno/Mno) models of unknown experimental structure, they were modeled via homology on the framework of PP1c, given the 49% sequence identity shared between the two catalytic units (Dawson & Holmes, 1999).Despite this high homology, PP2Ac essentially loses most of the acidic character in the groove termed "acidic" in PP1c (Goldberg et al., 1995), as a result of the following four amino acid replacements lining the floor of this groove: D220P in PP1c numbering (or D211P in the PP2Ac numbering) in the b8-b9 loop, E252M (E243M) in the b10-b11 loop, E256N (E247N) in the b11 sheet and A259H (A250H) in the b11-b12 loop (cf.Fig. 1).One should expect this dramatic change of properties to significantly affect the binding of the four ligands, comprising 2 (7-pep) to 4 (11-pep, I-1 and DAR) basic Arg and/or Lys residues at their N-termini.In particular, they should appear to bind more loosely the enzyme with these termini.In this respect, it is surprising that both the free PP2Ac×(Zno/Mno)×PO  The values relating to the differences between the respective starting and MD-relaxed configurations are on the diagonal.tory maps in Figs. 2, 4-7, that the PP2Ac structures, irrespective of the metal dication and ligand, undergo but small changes in the simulations.This is also clearly seen in Table 2 in the scarce occurrence of PP2Ac residues experiencing deviations from the starting structures larger than 5.5 Å, limited only to some residues from extra-globular loops.This feature can hardly be reconciled with any differences in biological selectivity of the ligands towards PP1c vs. PP2Ac catalytic units.
Currently, it is believed that a combination of Zn 2+ and Fe 2+ (Cohen, 1997;Chu et al., 1996) is a probable candidate for the PP1c di-cation center in vivo.While we have currently been working on such a dication×OH -complex, in view of this work one can doubt that even a properly parameterized pseudomolecule "ZnFeO" could account for subtleties pertinent to the proper choice of the cation(s) and its detailed role in the PP1c mechanism, let alone PP2Ac whose structure is only a "qualified guess" in this work.In view of the rapidly growing evidence (over 600 hundred papers per year) on the regulatory mechanisms of the protein phosphatases it becomes clear that, apart of the metal di-cation type, an essential role is attributed to the abundance of variable regulatory protein units, depending on specific tissues and targeting the catalytic units to specific tasks.For instance, a complex mode of binding of the DARPP-32 and Inh-1 inhibitors to PP1c has been proven recently.Both DARPP-32 and Inh-1 involve multiple sequence fragments for high affinity and selectivity (Endo et al., 1996;Kwon et al., 1997; see also introduction) binding to different sites on PP1c (Huang et al., 1999).Apart from the sequence involving pThr and its surroundings, there is still a short KIQF sequence close to the N-terminus (8-12 in Inh-1 and 7-11 in DARPP-32) which is a requisite for their binding and selectivity.Furthermore, this motif is more or less conserved in a quite wide spectrum of diverse PP1c regulatory units (Aggen et al., 2000).A volatile binding of the phosphorylated dodecapeptide in the catalytic groove may thus argue for a complex mechanism for binding Inh-1 and DARPP-32 to PP1c, involving in a cooperative way at least two distal sequence fragments.
In this work we introduced the test case Zno and Mno pseudo-molecules (-residues), composed of an either Zn 2+ or Mn 2+ dication center, respectively, nesting a structural OH -nucleophile between the metal nuclei, so that the total charge of either Zno or Mno was +3.For their incorporation into the AMBER force field, Zno and Mno were structurally optimized and parameterized using advanced ab initio methodology so as to meet geometrical, mechanical and electrostatic requirements (see Fig. 2 in Methods), typical of the crystal structures of PP1c.Our results demonstrate that both Zno and Mno behave well in typical MD simulations and thus prove the concept of pseudomolecules to be useful.Simultaneously, the differences between Zno and Mno, as reflected in the parameterization, seem to be not precise enough for being specifically accented in the respective simulations.

Figure 1 .
Figure 1.The standard view of PP1c, a stereo diagram.The figure is based on the structure, deposited as PDB file 1fjm (Goldberg et al., 1995) modified as described in Methods for the starting model of PP1c×Zno×Tpo.The metal cations and the structural OH -are represented as magenta balls while the phosphate (orange) and the conserved residues (see Methods) in the active center (black) are shown as sticks.The secondary structure is defined as in Goldberg et al. (1995) and labelled accordingly.The figure was prepared using the Molmol program (Koradi et al., 1996).
40E.WoŸniak-Celmer and others 2001 lar, it is more flexible than the PP1c with the Mno dication (RMS = 2.65 Å).Surprisingly, the unoccupied PP2Ac models with both the Zno and Mno dications appear much less flexible than their PP1c counterparts (RMS = 1.43 Å and 1.51 Å, respectively).This is even more surprising if one re-Vol.48 Protein phosphatases 1 and 2A 41

Figure 2 .
Figure 2. The contour plots typical of the collective MD trajectories in four free enzyme models: A. PP1c×Zno×Tpo, B. PP1c×Mno×Tpo, C. PP2Ac×Zno×Tpo, D. PP2Ac×Mno×Tpo, represented by the C a sequential positions along the vertical axes, over the time range of about 300 ps (horizontal axes).The contours are drawn every 1.5 Å, in agreement with the color scale defined on the right.They denote departure, evolving in time, of any C a relative to its starting position.For instance, the build-up of the green-to-orange traces for A: PP1c×Zno×Tpo and B: PP1c×Mno×Tpo between residues 128-136 corresponds to a flapping-down movement of helix D due to the Da-Ea hinge towards the trough of the hydrophobic groove, see text. b

Figure 3 .
Figure 3.The overlapping active centers with key residues labeled and shown as sticks and the protein chains shown as thin backbones.A: The worst fit corresponding to RMS = 2.03 Å for PP1c×Zno×Tpo (blue) vs. PP2Ac×Zno×Tpo (green) and labeled in accordance with the PP1c sequence.B: The best fit corresponding to RMS = 0.91 Å for PP2Ac×Zno×Tpo (blue) vs. PP2Ac×Mno×Tpo (green) and labeled in accordance with the PP2Ac sequence.

Table 5 .
The matrix of the C a -based RMS deviations among the starting and the MD-relaxed active center geometries of the four ligand-bound PP1c and PP2Ac units with either Zno or Mno dications a .a RMS in Å.The maximum and the minimum values typical of each of three sub-blocks (framed): (PP1 or PP2A)×Zno, (PP1 or PP2A)×Mno and (PP1 or PP2A)×(Zno or Mno) are given in bold and italics, respectively.This also applies to the diagonal, subdivided into two sets depending on which dication (Zno or Mno) is incorporated in the enzyme.b

Table 1
) were executed until 38 E. WoŸniak-Celmer and others 2001 full release of the constraints and the final productive molecular dynamics (MD) run.In order to save time, all MD simulations were carried out with frozen carbon-hydrogen covalent bonds (procedure SHAKE; Brooks III et al., 1988).PP1c×Mno×Tpo was obtained by a computer mutation of Zno into Mno, after stage 3 in Table

Table 4 .
The most poorely and most closely overlapping free-enzyme active centers are shown in Fig.3.

Table 3 . The sequence-averaged C a -based RMS deviations between the final and the starting snapshots in the MD simulations of the systems studied a,b
aThe upper values: RMS in Å.bThe lower values: MD total time(s) in ps.

Table 4 . The matrix of the C a -based RMS deviations among the starting and the MD-relaxed active center ge- ometries of the four ligand-free, PO 4 3--bound PP1c and PP2Ac units with either Zno or Mno dications a
a RMS in Å.