J. Mol. Biol. (1991) 218, 449-464

Reaction Mechanism of Alkaline Phosphatase Based on Crystal Structures Two-metal

Ion Catalysis

Eunice E. Kim and Harold W. Wyckoff Department of Molecular Biophysics and Biochemistry Yale University, New Haven, CT 06511, U.S.A. (Received 31 August

1990; accepted 26 November 1990)

Alkaline phosphatase (AI?) is a widely distributed non-specific phosphomonoesterase that functions through formation of -a covalent phosphoseryl intermediate (E-P). The enzyme also catalyzes phosphoryl transfer ,reaction to various alcohols. Escherichia coli AP is a homodimer with 449 residues per monomer. It is a metalloenzyme with two Zn2+ and one Mg2f at each active site. The crystal ‘structure of native E. coli AP complexed with inorganic phosphate (Pi), which is a strong competitive inhibitor as well as a substrate for the reverse reaction, has been refined at 20 A resolution. Some parts of the molecule have been retraced, starting from the previous 2.8 A study. The active site has been modified substantially and is described in this paper. The changes in the active site region suggest the need to reinterpret earlier spectral data, and suggestions are made. Also presented are the structures of the Cd-substituted enzyme complexed with inorganic phosphate at 2.5 A resolution, and the phosphate-free native enzyme at 2% A resolution. At pH 7.5, where the X-ray data were collected, the Cd-substituted enzyme is predominantly the covalent phosphoenzyme (E-P) while the native Zn/Mg enzyme exists in predominantly noncovalent (E .P) form. Implication of these results for the catalytic mechanism of the enzyme is discussed. APs from other sources are believed to function in a similar manner.

1. Introduction Alkaline phosphatase (EC 3.1.3.1, abbreviated as APT) is a non-specific phosphomonoesterase that functions through a phosphoseryl intermediate to produce free inorganic phosphate or to transfer the phosphoryl group to other alcohols (for a review, see Reid & Wilson, 1971; Pernley, 1971; McComb et al., 1979; Coleman & Gettins, 1983aJ). It is a dimeric metalloenzyme with two Zn2 + and one Mg2 + in each active site region. It is found in both prokaryotes and eukaryotes. There are at least four isozymes in man. Although its specific functions in mammals are not known at present, the tissue non-specific variety is strongly implicated in modulation of the mineralization of bone, since hypophosphatasia that results t Abbreviations used: AP, alkaline phosphatase; n.m.r., nuclear magnetic resonance; e.s.r., electron resonance; r.m.s., root-mean-square; p.p.m., parts million. 0022%2836/91/060449-16

$03.00/O

spin per

449

from mutations in the relevant AP gene is primarily a bone disorder (Weiss et al., 1988; Whyte, 1989). Eukaryotic APs are glycosylated and are attached extracellularly to cytoplasmic membrane. In the case of human placental enzyme, it is shown to be anchored by a phosphatidylinositol-glycan moiety (Micanovic et al., 1988). In Escherichia coli the enzyme is involved in the acquisition of phosphate from esters when free inorganic phosphate is depleted. The E. coli enzyme has been studied most extensively by a variety of physicochemical techniques including n.m.r. and e.s.r. studies on both native and substituted APs, and is therefore useful as a prototype. The E. coli dimer (J!f,=94 x 103) has 449 (or 450 if a variable Arg residue at the amino terminus is included) amino acid residues in each of the two chemically identical subunits (Bradshaw et aE., 1981; Chang et al., 1986) and is located in the periplasmic space with neither carbohydrate nor a fatty acid tail. 0 1991 Academic Press Limited

E. E. Kim and PI. W. Wyckofl

450

The overall reaction scheme considering both hydrolysis and transferase activities has been proposed as shown in Scheme I, below. Scheme I Rate-limiting step R,OH E + RrOP =F= E*RrOP fc,

H~OJ

EP

a

E + Pi

_

E+R*OP

e

E-Pfl ERzOP

Rate-limiting step The E-P refers to a phosphoseryl enzyme formed with Serl02 while E. ROP or E * P refers to the noncovalent complexes with substrate or product. The hydrolysis of E-P involves both formation and dissociation of the non-covalent enzyme-phosphate complex (E . P). At acidic pH, the hydrolysis of the phosphoenzyme (E-P) is rate limiting while the dissociation of E . P is the rate-determining step a,t alkaline pH (Chlebowski & Coleman, 1972, 1974; Hull et al., 1976; Chlebowski et al., 1977; Gettins & Coleman, 1983). If a phosphate acceptor such as ethanolamine or Tris is present in high concentration, then the enzyme transfers the phosphate from the substrate to the alcohol (R,OH) (Dayan & Wilson, 1964; Wilson et al., 1964; Gettins et al., 1985). The phosphorylation time constant derived from stopped flow observation of alcohol release is less than 1 ms (Bloch & Schlesinger, 1973). The turnover number for the hydrolysis reaction with p-nitrophenylphosphate(pNPP) as a substrate at pH 8.0, calculated from a specific activity of 1020 prnol of pNPP h- ’ mg- ’ (Chlebowski & Coleman 1974) a.nd a monomer molecular weight of 47 x 103, is 13.3 s-l. If one assumes that all sites are active then the overall retention time is 75 ms at 25 “C. At pH 8.0 the ratio of E-P to E. P is less than 0.01 for the native enzyme and therefore the retention of Pi is about 75 ms, which is consistent with the deduction of 50 ms from n.m.r. saturation transfer (Gettins et al., 1985). I80 exchange time for bound Pi is 6 s with less than one exchange per binding event. Pi binding constants range from 1 to 3 PM (Bloch & Bickar, 1978; Applebury et al., 1970). Using the overall rate as an approximation for the o#-rate of Pi and a binding constant of 0.5 x lo6 M-I the Pi on-rate is calculated to be 6.7 x lo6 M-~ s-l. In recent example one (Chaidaroglou et al., 1988) k,,, and KM were reported to be 13.6 s-l and 7.4 PM, respectively, thus giving k,,,/K, a value of 1.8 x lo6 RI-I s-l per monomer as a lower limit for the substrate on-rate. If the ester substrate o&rate is faster than the phosphorylation rate the on-rate must be increased accordingly. The X-ray structure of E. eoli AP was solved and refined previously at 2.8 A (1 A=O.l nm) resolution (Sowadski et al., 1985). However, there have been long-standing uncertainties about the metal-ligand

constellations, due to poorly defined electron density in the active site region (Wyckoff, 1987). ,41so the top portion of the molecule (residues 380 to 410) was essentially undefined in the electron density map. We have extended the resolution to 2.0 A with completely new dat,a on the rlative enzyme complexed with phosphate. Phosphate is a strong competitive inhibitor as well as a product of the enzyme. The structure has been revised and refined. The metal-ligand constellations have now been clarified. In this paper we describe some of t’he most striking results from the refinement of the structure at 2-O A resolutian. Also we report data on phosphate-free enzyme at 2.8 A resolution and phosphorylated Cd(H)-substituted enzyme at 2.5 A The Cd(H)-substituted enzyme at resolution. pH 7.5, where the X-ray data, were collected. is predominantly E-P while the native enzyme with E.P exists as predominantly ph0sphat.e (Chlebowski et al., 1976). Therefore, these structures bracket reaction intermediates and suggest a, possible mechanism for the enzymatic reaction. The results of the current refinements of structure based on X-ray data indicate a need for reinterpret,ation of some of the earlier spectral data and some suggestions are made. 2. Materials

and Methods

(a) Crystallization

and data

collection

AP from E. coli was purified and crystallized as described (Sowa,dski et al., 1985). Three sets of data were collected: (I) Zn/Mg/Pi refers to the native enzyme

Table 1 Summary Zn/Mg/P, Soaking conditions PH Space group it c

Resolution (A) Total no. of observations No. of unique reflections No. of possible reflections 4x7. Z/o(Z) f&n r%)

of data

Cd/P,

Zn/Mg

65 y. (NHJ2S0,, 61 M-!&is. HCI 10 mm-Zn’+ 50 mx-Cd*” 10 mM-Zn2’ 10 mM-?YlgZ+ 10 mnl-Mg” 2 mM-Pi 2 mx-Pi 7.5

TL5

125

167.41 195-27

1222 168.03 195.64

167.21 195.07

7616

76.71

7656

2-o

2.5

2.8

251,352

?78,818

78,398

42,535

30,762

85,810

44.176

10.6; 3.4 4.7

8.4 6.8

31,539 3TJ 6.8

421,175

All the data were collected on San Diego Multiwire Systems area detectors at Yale. Zn/Mg/P, data set was collected using 2 crystals and 17,164 reflections were used for merging. R,,,= C/Z - l/Z/Z/, where Z is intensity. pH was a,djust.ed by adding HCI.

Reaction Mechanism of Alkaline complexed with phosphate; (2) Zn/Mg refers to the native enzyme, i.e. without phosphate; (3) Cd/P, refers to the Cd(II)-substituted enzyme complexed with phosphate. All the diffraction data reported here were collected using the San Diego Multiwire System area detectors at Yale University. The crystals all belong to the space group 1222 and cell dimensions are given in Table 1 along with other information on data collection. All the data were reduced, scaled and averaged using the software provided with the system. While Cd/P, and Zn/Mg sets were collected on 1 crystal, Zn/Mg/P, data were collected on 2 crystals and the 2 sets were combined using PROTEIN (Steigeman, 1982) by scaling the 2 using a total of 17,164 overlapping unique reflections (b) Structure

rejinement

Refinement was carried out using both fast Fourier (FFT)-implemented stereochemically transform restrained least-squares refinement method (PROFFT: Finzel, 1987) and the simulated annealing refinement method (X-PLOR: Briinger et al., 1987, 1990). The latter involves molecular dynamics simulation at elevated temperature using diffraction data in combination with energy functions to overcome local minima. The calculations were carried out on the VAX 8800 and the COPU’VEX C-l and C-2 at Yale Center for Structural Biology, and the CRAY X-MP at Pittsburgh Supercomputing Center. Progress was slow and frequent model-building interpretation using the program FRODO (Jones, 1978, 1982) on Evans & Sutherland PS300 graphics systems was necessary to correct misplaced atoms. Both 2F,- F, and difference maps as well as omit maps were used throughout. Omit maps were generated with phases determined after several cycles of refinement where relevant regions of the molecules, less than 10% of the total; are omitted from the structure during phasing to avoid possible feedback. Zn/Mg/Pi set was refined first. At the start of this refinement 1 momoner (subunit B) was generated from the other using non-crystallographic 2-fold symmetry, and then refined independently. No restraints were imposed on the metal co-ordination. The starting models for the Cd/Pi and Zn/Mg sets were the refined protein model from the Zn/Mg/P, refinement. Both Cd/P, and Zn/Mg sets were subjected to the annealing method to overcome feedback and explore possible structural differences from the starting model. In the case of Cd/P, Cd(I1) replaces all 3 metals. On the basis of the initial difference Fourier between Cd/P, and Zn/Mg/P, and the omit map (where atoms with a 8.0 i% sphere from Ser102 are omitted in the calculation) Ser102 was modeled as a phosphoserine. The largest peaks on the initial difference map were around the metals. Three metals were labeled Cdl, Cd2 and Cd3, replacing Znl, Zn2 and Mg, respectively. In each case individual isotropic temperature factors were refined. Details of the refinement and the results will be described separately. The co-ordinates have been deposited in the Protein Data Bank (Berstein et al., 1977).

3. Results (a) Rejinement summary The refined model of the phosphate-bound enzyme (Zn/Mg/Pi) includes 6608 protein atoms (non-hydrogen), six metal ions, 296 water molecules and two phosphate ions. The final cycle including

Phosphatase

451

73,328 unique reflections between 5.0 and 2.0 1%led to as crystallographic R-factor (residual error) of 0.184, from an initial value of 0.454, with quite good stereochemistry. The root-mean-square (r.m.s.) deviations from ideality are 0015 J% in bond distances and 2.96” in bond angles. The r.m.s. [shifts from the starting model are 2.46 and 2.95 a for the backbone atoms and 3.53 and 3.96 a for the sidechain atoms for subunits A and B, respectivel;y. During the refinement residues 380-410 were retraced and residues 190-200 were shifted by one residue in both monomers in addition to numerous other minor modifications. Also some changes were made in the active site region of the enzyme. A total of 296 water molecules was located as well as a tightly bound phosphate in each active site. Most of the electron density map is now clearly interpretable except for four residues at the amino end (of the polypeptide chain and another four or five residues on the top of the molecule (404-408). Some sidechains, lysine in particular, are not well defined in the map but they are mostly on the surface Iof the molecule. The map in the active site is now clear and the metal-ligand constellations are well defined. Unless otherwise specified the following description applies to both monomers. (b) Overall

structure

Although the structure has been modified during the refinement, most of the structure of the protein is consistent with the model described earlier (Wyckoff et al., 1983; Sowadski et al., 1985; Wyckoff, 1987). Figure l(a) shows the C” trace of the dimer and Figure l(b) shows a ribbon drawing of a monomer. The enzyme is basically a 2-fold symmetric dimer (100 a x 50 A x 50 A) with two active sites located about 30 a from each other. Each subunit shows typical c~/fl topology with a central ten-stranded b-sheet in the middle flanked by 15 helices of various lengths. In contrast to earlier descriptions, residues 325-334 form a well-behaved a-helix while residues 425-430 are not helical. In fact residues 325-334 become helical upon metal binding and provi’de two of the ligands (Asp327 and His331) to Znl.. After retracting the chain, the top portion of the msolecule contains a three-stranded sheet and a short, helix. Thus the overall structure consists of a central tenstranded b-sheet and a minor three-stranded j-sheet on the top as well as 16 helices, which when combined constitute about 55% of the protein. The dimeric structure is narrow at the bottom and broader on the top both in and perpendicular to the plane of Figure l(a). The monomer is considerably thicker on the side of the central P-sheet on. which the active site is located at the carboxyl endl of the P-sheet. The strands in the central P-sheet are parallel except for one antiparallel insertion, and the sheet is largely classical in that it is connected in a right-handed sense as one proceeds from one strand to the next while the twist of the sheet is lefthanded. The central portion of the sheet is consider-

(bl Figure 1. (a) C” trace of the complete dimer of E. coli alkaline phosphatase (drawn using program SZAZAM developed by A. Perlo) viewed with its non-crystallographic 2-fold axis vertical and its maximum dimension horizontal. The tube radius is 0.5 A and the 3 metals are shown as spheres. The 2 active sites are about 30 a from each other. Strand G of the central /?-sheet is indicated. (b) Ribbon drawing of a monomer of alkaline phosphatase drawn using program RIBBON (Priestle, 1988). The 3 metals are shown as stippled spheres. It consists of a lo-stranded central P-sheet flanked by 15 helices and another 3-stranded P-sheet, and a helix on the top.

Reaction Mechanism of Alkaline

Phosphatase

453

ably flatter than other sheets while the strands at the inner end are curled and provide part of the interface surface.

(c) Active site of phosphate-bound enzyme

(a)

(b)

Figure 2. (a) Zn/Mg/Pi: 2F,-F, map in the active site region of alkaline phosphatase. (b) Cd/P, omit map; atoms within 8.0 I% of Ser102 are omitted from the structure factor and phase calculations. The final refined model is superimposed on the density map. (c) Zn/Mg omit map showing Ser102; atoms within 8.0 .!I of Ser102 are omitted in the calculation of structure factor and phases. The final

refined model is superimposed on the density map.

The active site region can be considered as AsplOl-Ser102-Ala103 and the metal triplet (two Zn2+ and one Mg2+) and their ligands, as well as Arg166 and other amino acids in the immediate vicinity. This site is located at the carboxyl e.nd of the central P-sheet and all the ligands to the three metal ions are provided from one monomer. The electron density map in this region is now well defined except for the disordered hydroxyl group of Ser102, which is involved in phosphorylation and dephosphorylation. The three metals, two Zn2+ (referred as Znl and Zn2) and one Mg2+ are close in space: d(Zn1 -Zn2) = 3.94 A, d(Zn2 -Mg) = 4.88 A, and d(Zn1 - Mg) = 7.09 A in subunit A and 4.18, 466 and 7.08 A in subunit B, respectively (d is distance between metal ions). The angle between Znl-ZnB-Mg is 105” in both subunits. In addition to the tightly bound phosphate (numbered 453), several water molecules were located within the active site, and they form an extensive hydrogen-bonding network. Figure 2(a) shows a 2F0--F, map at the active site region of the molecule in Zn/Mg/P, structure. The active site of the enzyme is shown in Figure 3: (a) shows the metal binding sites in stereo while (b) shows all of the atoms within 100 A of the phosphorus. Hydrogen bonds are indicated by broken lines. The water molecules are numbered 454 to 465 although unequal numbers of water molecules are found in the two subunits. The distances and angles involved in the metal co-ordination spheres are given in Table 2. Table 3 contains the hydrogen-bonding distances. All the residues have very stable conformations as indicated by their low temperature factors (B-values in Table 2). (i) Znl co-ordination Znl is penta-co-ordinated by the imidazole nitrogen atoms of His331 and His412, both carboxyl oxygens of Asp327 and one of the phosphate oxygens with an average metal--1igand distance of 2.07 A as depicted in Figure 4(ab). The polyhedron around the metal can be best described as a pseudo tetrahedral with both carboxyl oxygens of Asp327 occupying one of the apices. His372, which was originally thought to co-ordinate Znl, is not a direct ligand (3.8 A away from Znl), but is hydrogen bonded to one of the carboxyl oxylgens of Asp327 (see Table 3). (ii) Zn2 co-ordination Zn2 is co-ordinated tetrahedrally by the imidazole nitrogen of His370, one of the carboxyl oxy,gens of Asp51 and one of Asp369, and one of the phosphate oxygens as seen in Figure 4(b). The average metal-

E. E. Kim

and H. W. Wyclcoff

(b) Figure 3. (a) Stereo drawing showing the active site of the E. coli alkaline phosphatase. Atoms are shaded by at,om type. Some residues and water molecules are omitted for clarity. (b) The active site region including all the atoms within IO b of the phosphorus atom. Water molecules are labeled as W. Hydrogen bonds are shown as broken lines.

ligand distance is 2.00 il. As mentioned, the OG of Serl02 is disordered. Figure 3 shows a more prominent position for OG.

protocols of simulated annealing refinement were carried out to allow possible alternative solutions, but again the current model was stable.

(iii) Mg co-ordination

(iv) Phosphate

Mg co-ordination can be described as a slightly distorted octahedron with the second carboxyl oxygen of Asp51, one of the carboxyl oxygens of Glu322, hydroxyl of Thr155, and three water molecules (numbered 454 to 456) completing the octahedron (see Fig. 4(c)). The average Mg-0 distance is 2.12 8. Asp153 is not a direct ligand to this metal as previously indicated, but is an indirect ligand in the sense that it forms hydrogen bonds to two (Wat454 and Wat455) of the three water molecules that are co-ordinating Mg. Since some of these assignments differ from previous proposals (on the basis of X-ray crystallography and other spectroscopic st,udies), especially in the case of His372, considerable efforts were made to verify this. First, various omit maps, omitting atoms within 10.0 a of these residues as described in Materials and Methods, were calculated and examined at various stages, but the present interpretation Second, was reinforced. several

Phosphate is co-ordinated to both Znl and Zn2, and the other two oxygens are tightly held by two amino functions of the guanidinium group of Arg166, which is in turn furt’her hydrogen bonded to Asp101 and a water molecule (Wat4.59) that is held by Asp153 and possibly Tyrl69 (see Table 3). The phosphate is further hydrogen bonded to the amide of Ser102 and a water molecule (Wat454) that is co-ordinated to Mg and another water molecule (Wat457) that is bridging to Lys328. Znl , 01, P, 02 and Zn2 are nearly coplanar and the distances and angles in this ring are shown in Figure 5(a). Further geometries are described in the Discussion. The residues 101 (or 102 depending on the definition of the helix) to 112 form a helical stem t’hat is buried between the central b-sheet, and other helices and loops, and this provides a firm anchor for Serl02. The carboxyl group of Asp101 is positioned so that there is no direct contact with either SerlO2

ion

Reaction Mechanism of Alkaline

Geometry around

Phosphatase

455

Table 2 metals in phosphate-bound AP (subunit A/subunit

B)

Znl coordination M...Lt IL4 His331 His412 Asp327 PO,453

NE2 NE2 ODl OD2 01

11.9/143 92/148 7.9117.8 7.9/l&9 21.2/1%7

1.96/2.07 2.04/2.04 2.00/226 2.30/2.53 1.97/2,12

His412 9Mq93.7

L-M-L Asp327 1046/1036 101.1/97.0

angles(“) Asp327 90.q93.0 1582/151.0 57.2/1348

PO,453 110.5/107.5 108.1/112.7 129-3/92-o 87.0/92.0

Zn2 coordination M...L (4 His370 Asp51 $sp369 PO,453

NE2 ODl ODl 02

9.815.8 &2/131 88/1@4 19.4/19.4

2.05j2.01 2.13/2-03 1.79/1.80 197/2.23

Asp51

L-M-L angles(“) Asp369

119.2jll6.3

940/93.6 106+3/102.5

Thr155

Glu322

95,1/92.0

1034/1041 82.7/902

PO,453 102.6/1152 1058/10@8 129.3/128.9

Mg coordination M...L (4 Asp51 Thr155 Glu322 Wat454 Wat455 Wat456

OD2 OGl OE2 0 0 0

%2/1%7 6.2197 7.4/l 1.6 10.4/9+3 8.0/157 69/l 1.0

207/1.96 2.15/2.05 226/1.93 2.03/1.92 1~90/2~00 2,32/2-03

L-M-L angles(“) wat454 91.8/1693 16&g/92,3 104.2/857

wat455

Wat456

176.4j89.6 81.9/171.6 7%2/97,5 90.9/84.8

101.1/92.4 837175.9 152.9/163.3 8@5/8@5 76.7185.5

t M, metal; L, ligand

or the phosphate but it forms hydrogen bonds to the amides of residues 103(3.32 A) and 104(2.98 A) and Argl66. Therefore, it appears that it is functional in positioning the guanidinium group of Arg166 and in helix formation, and in providing part of the electrostatic environment.

(d) Phosphoseryl and phosphate-free enzymes Refinement of Cd/P, (2,5 A) and Zn/Mg (2% A) including only the non-hydrogen protein atoms and metals converged to R-factors of 22.3 and 20.8%, with comparable stereochemistry. respectively, Although some water molecules, mainly in the interior of the protein and the intersubunit region, are prominent in the difference maps, no water molecules are included at this stage. The electron density maps at the active site region of the two structures are shown in Figure 2(b) and (c), respectively. While Cd/P, is shown against a 2F,- F, map, the Zn/Mg model is shown against an omit map in which all the atoms within 8 A from SerlO2 c” were omitted from the structure factor calculation. Cycles of refinement were performed in order to unbias phase information. As a verification of the

metal-ligand constellation, various omit maps around those residues were calculated a,s for Zn/Mg/Pi. They were also subjected to various protocols of simulated annealing. In the case of Cd/P, the refinement shows that the ester oxygen (denoted as OG’) of phosphoserine is 2.3 A from Cd2, and the two oxygens are at hydrogen bonding distance from the guanidinium group of Arg166 similar to what is seen in Zn/lMg/Pi. the remaining oxygen is positioned Therefore, between the two metals, Cdl and Cd2, as shown in Figure 2(b). The average metal-ligand dista,nce is 2.4 A. Details of the Cd-phosphoseryl interaction are shown in Figure 5(b). In the case of Zn/Mg, the hydroxyl group of SerlO2 is clearly co-ordinated to Zn2. The average metal-ligand distance is comparable to that in Zn/Mg/P,. It should be noted that in both Cd/P, and .Zn/Mg there is a positive peak approximately 2.2 A away from the center of Cdl and Znl, respectively, in the difference Fourier, indicating further co-ordination of this site, possibly by a water or other solvent molecule such as ammonia or chloride. There are other positive peaks in the difference map, but further considerations of detailed structure are deferred until higher-resolution data are ava:ilable.

E. E. Kim and H. W. Wyhff

456

Table 3 Hydrogen-bonding x

distances in the active site of AP Distance (d) Subunit A Subunit B

Y

PO&453 01 PO,453 03

PO,453 04 PO,453 04

His331 His370 His412 His412

ND1 ND1 ND1 N

Wat457 0 Wat458 0 A@66 NH1 Wat458 0 SerlOP N Arg166 NH2 Wat454 0 Wat457 0

3.4 3.2 2.64 2.7 3.04 2.61 2.9 32

3.5 3.3 2.62 2.8 2.60 2.87 27 2.8

Gln410 Ala371 ThrlOO Wat463

241 2.66 2.89 3.1

2.43 2.82 2.97 3.0

0 0 0 0

Asp51 ODl Asp51 OD2 Asp327 OD2 Asp369 ODl Asp369 OD2

Gly52 N %x102 OG Wat457 0 SerlO2 OG His372 NE2 Gly52 N His370 N

3.16 2.96 2.8 2.35 3.06 2.99 3.06

3.24 317 2.7 265 332 303 2.95

Glu322 OE2

Wat455 0

2.7

2.8

Thrl55

Wat465 0

2.7

2.7

His372 SD1

Aia373 X Asp330 ODl

2.80 338

2.80 3.27

Asp101 ODl

Arg166 Gly118 Ala103 Ala104 Gly118

NH1 N N N N

318 283 3.35 2.95 322

2.85 3.15 3.23 3.05 3.20

Asp153 OD2

Ala154 wat459 Lys328 Wat454 Wat455

N 0 NZ 0 0

3.14 29 2-75 31 2.9

3.26 3.1 3.06 3.0 2.4

Arg166 0 Argl66 NH2 Arg166 NE

Wat460 0 Wat459 0 Wat459 0

2.6 32 3.0

2.6 3.0 3.0

Lys328 NZ

Wat457 0

27

3.0

ThrlOO 0

Wat462 0

3.0

Ser102 0

Wat465 0

2.9

3.0

Gln152 Ala154 Tyr169 Ala324 Gln410 Glu411

0 N OH 0 0 OEl

Wat460 Wat460 Wat459 Wat456 Wat464 Wa,t464

0 0 0 0 0 0

25 3.2 33 3.1 33 3.3

2.8 32 3.5 3.0

Wat456 Wat459 Wat461 Wat461 Wat463

0 0 0 0 0

Wat457 Wat460 Wat462 Wat463 Wat464

0 0 0 0 0

2.7 31 2.7 3.1 30

2.7 2.4

OG

Asp101 OD2

Asp153 ODl

(e) Comparison

of ZnlMg/P,,

Cd/P, and Zn/Mg

When the structures are superimposed by the least-squares procedure, using only the backbone (N: C, 0; C”) atoms, the r.m.s. difference between

(b)

Figure 4. Co-ordination phosphate-bound

enzyme:

spheres of the 3 metals (a) Znl,

in

(b) Zn2, and (c) Mg.

the models with respect to that of the phosphat,ebound enzyme are 037 A for Cd/P, and 0.17 ,& for Zn/?Jlg. Overall, the molecule remains unchanged except for residues at t,he amino terminus and 404-408 region where the electron density maps are not clear. Other changes involve the residues on the

Reaction

Mechanism

of Alkaline

Phosphatase

457

4. Discussion

(a) Metal-Eigand

(a)

CD1 -I. 2.40

(b)

Figure

5. Schematic drawing of interactions in: (a) metal-phosphate and guanidinium-phosphate interactions in Zn/Mg/P, complex; (b) metal-phosphoryl and guanidinium-phosphoryl interactions in phosphoryl serine intermediate.

surface region of the molecule, mostly involving side-chain atoms. The residues in the active site region remain about the same except for Arg166 whose conformation is slightly altered. However, in the case of Cd/P, there are slight changes in the model in the active site region: (1) the centers of the Cd ions are displaced slightly from the corresponding Zn and Mg positions (by 0.34; 022 and 0.21 A) of Zn/Mg/P,; (2) most of the ligands to the metal ions have moved outwards from the metal ions. Both are understandable and expected, since Cd2+ is appreciably bigger than either Zn2+ or Mg2+ (ionic radii being 097, O-74 and 966 A, respectively). The position of phosphorus is also displaced (by 0.61 A) relative to the protein atoms and the metals, since a covalent bond is formed in the case of Cd/Pi and the tetrahedron around the phosphorus is inverted. Changes in the Zn/Mg are much smaller in magnitude.

interaction

The metal-ligand constellations are now cla.rified and there are substantial changes from earlier findings. Unlike what was reported earlier, His372 does not co-ordinate Znl in any of the complexes reported here, but it is hydrogen bonded to the carboxyl oxygen of Asp327. Also, Asp153 is not a direct ligand to Mg but is hydrogen bonded to two of the three water molecules co-ordinating this metal. It is interesting to note that none of the ligands involves backbone atoms. The average metal-ligand distances observed here are similar to the values reported in the crystal structures of small molecules with the same co-ordination numbers. The imidazole rings of the histidine residues that are co-ordinated to zinc show high variability in the orientation of the ring with respect to the N-Zn bond. Therefore drr-px interactions between the metal and imidazole ring must not be strong. The Zn-N-C angle varies from 114 to 140”. The imidazole rings of His370 and His412 are almost parallel, with a distance of approximately 3.6 A between the two planes, while those of His331 and His372 are almost perpendicular to each other. All three major types of interactions between carboxyl groups and metals (Einspahr & Bugg, 1984; Kim et al., 1985; Carrel1 et al., 1988) are found here: (1) the carboxyl group acting as a unidentate ligand, e.g. Asp369 (co-ordinating Zn2) and G:lu322 (co-ordinating Mg); (2) the carboxyl group acting as a bidentate ligand to one metal, e.g. both oxygen of Asp327 co-ordinating Znl; (3) the carboxyl group bridging two metals, e.g. Asp51 bridging Zn2 and Mg. The detailed geometries are within those reported in crystal structures of small molecules as well as in other metallo-proteins’ refined at high resolution, such as thermolysin (Holmes & Mattews, 1982), concanavalin A (Hardman et aE., 1982) and phospholipase C (Hough et al., 1989), to name a few. In fact the active site of phospholipase C has a metal constellation very similar to that of AP in that: (1) there are three metals in relatively similar orientations and distances; (2) there is a bridging carboxylate group between two metals; and (3) two of the ligands to one metal are provided from the same helix, i.e. His142 and Glu146 from helix E. In the case of AP these are His331 and Asp327 coordinating to Zn2. Note that the sequence order is reversed. In the case of AP this helix becomes disordered when the metal is removed and/or one can imply that the metal binding induces thse helix formation. However, it is interesting to note .that in the case of phospholipase C there is at lea,st one histidine residue in each metal co-ordination sphere and two nitrogen atoms as ligands. There is no further similarity between the two structures, since the phospholipase C is largely helical with no prominent beta sheet.

E. E. Kim and H. W. Wyckqff -

458

(b) interactions involving phosphoseryl

the phosphate moeity

and

The phosphate moeity is very closely associated with all t,hree metals: it bridges Znl and Zn2, and is hydrogen bonded to one (Wat454) of the water molecules co-ordinating Mg. The other two oxygen atoms are tightly held by the amino functions of the guanidinium group of Arg166. Figure 5(a) shows det,ailed schematic drawing of the interactions. This mode of phosphate ion bridging two metals is seen in small molecules (Hayden et aZ.; 1982). However, it is striking to note that the Zn-O-P angles are quite different: about 120” for Znl but almost linear for Zn2 in both subunits. This rat’her unusual linear nat,ure of Zn2-O-P interaction is a result of its immediat,e environment (pre-arranged), which is governed not only by electrostatic interactions but also by van der Waals’ contact,s, which may be quite different from interactions found in small molecules. On the other hand, the interaction between the phosphate and the guanidinium group of Arg166 is symmetrical and nearly parallel. Although this type of guanidyl-phosphate interaction is less frequent (Salunke & Vijayan, 1981), they are observed in propyl guanidinium diethylphosphate (Furberg & Solbakk, 1972) and cation 1 of bis(methylguanidinium) monohydrogen phosphate (Cotton et al., 1974). In the case of E-P the ester oxygen (OG’) of the phosphoserine co-ordinates Cd2 while the remaining oxygen is situated between the two metals (Cdl and Cd2), and therefore in contact with both as shown in Figure 5(b). This results in formation of a fourmembered ring, Cd2-OG’-P-P. A four-membered ring is assumed to be unusual and therefore unstable, although it is seen in bis(ethylenediamine)phosphatocobalt(II1) (Anderson et al., 1977) and [Cd(5’-CMP)(H,O)], (Shiba & Barr, 1978). While there is no noticeable deformation in the Cd(U) compound, in the case of the Co(II1) complex the four-membered ring is rather deformed as shown by the O-P-O angle of 99”. However, the two cases are quite different from what is observed here in that t’here is no second metal ion nearby. The interaction of the phosphoseryl moeity with the guanidinium group remains about the same as that of non-covalently bound phosphate while the interaction with the metals differs significantly.

(c) Reaction

mechanism

As shown in Scheme I, AP exhibits both t’ransferase activity and hydrolase activity. The reaction proceeds through a phosphoserine intermediate, and the transfer is shown to proceed with retention of configuration (Jones et al., 1978). When SerlOQ is replaced by Cys it also proceeds in the same manner in the case of a CyslO2 mutant (Butler-Ransohoff et al.; 1988a). It is generally assumed that each step involves in-line exchange with inversion and this is consistent with many other biological phosphate transfers (Knowles, 1980; Gerlt et nl., 1983).

Figure 6. Schematic drawing of a suggested reaction mechanism of alkaline phosphatase. Left and right a.re from Zn/Mg/P, and Cd/P, data, respectively, while the central pen&co-ordinate intermediate is derived on the basis of these 2 structures. The 2 non-co-ordinating oxygen atoms in each case are hydrogen bonded to Argl66, approaching from the right-hand side of this Figure, as shown in Fig. 5. For other hydrogen bonds see Fig. 3(b). Subskate cleavage and transfer to an acceptor activity are shown at the top. Hydrolysis of the phosphoseryl intermediate and the reverse reaction with inorganic phosphate are shown diagrammatically at the bottom. Metal ions are labeled 1 and 2. Oxygen atoms are depicted by open circles, carbon atoms by filled circles. and phosphorus at’oms by stippled circles.

However, the degree to which the transfer of the phosphate from the substrate to the serine (or the reverse) is associative or dissociative is not known. Diester transfer is normally thought to be associative, proceeding through a pentacovalent intermediate (Gerlt et al., 1983), whereas metal-catalized monoester transfer may be more dissociative, as reported by Hersch1a.g & Jencks (1987). The normal tests are difficult for wild-type AP because of the residence time of phosphate and t’he fact that retention of configuration could still occur with dissociation as long as the PO, group could not flip due to steric restraints. The results of X-ray data described here are fully consistent’ with in-line mechanism involving formation of penta-co-ordinate intermediate. Formation and resolution of this constellation is the central theme of the following discussion. Figure 6 depicts the proposed mechanism showing the observed phosphate-bound E. P on the left and phosphoryi the right, E-P on with the assumed penta-co-ordinate structure at the center. The position and the orientation of the penta-co-ordinate st,ate is derived from data on both Zn/Mg/P, a,nd Cd/P,. The equatoria,l plane of the penta-coordinated phosphate bisects Znl-Zn2 vector and the axial vector 0-P-OG’ is parallel to the Znl-Zn2 vector, giving a symmetry to the in-line bond transfer. The interaction with Argl66 and other

Reaction Mechanism of Alkaline hydrogen bonds positions 02 of phosphate in the plane of the metals and the axial oxygens rather than having two equatorial oxygens straddling this plane in the metal contact region. This arrangement permits the required motion during the catalysis. Resolving the penta-co-ordinate structure into the phosphoryl serine on one hand or the product phosphate or free ester on the other is nearly symmetrical. If one thinks of the central complex as being a metastable intermediate, then the breaking of either axial bond involves a transition state. Which transition state energy is higher and therefore rate limiting overall is unknown. Formation of the phosphoserine involves transfer of the leaving group to Znl, elongation of the P-01 bond, shortening of the P-OG’ bond, weakening the Zn2-OG interaction, motion of phosphorus downward and 02 upward. The leaving group can then be protonated from water, releasing the “product” whether it is ROH or HOH. Formation of free Pi or ROP involves transfer of the OG’-P bond to Zn2, elongating the OG-P contact, shortening the 01-P bond, weakening the 01-Znl interaction, moving phosphorus up and 02 down. Following departure of the product the serine OG’ can be protonated from water and dissociated from Zn2. The release of ROP or Pi from the penta-coordinated constellation can follow the same pathway but there is a significant difference in t’he possibility of a side reaction with Pi. The complexes E * ROPOs and E. HOPOs can be considered similar but the latter can ionize further, releasing a proton and forming a much stronger bond to Zn. In a sense the O-H bond can be transferred to the Zn-0 interaction. Thus there can be an E-OPO, species as well as an E. HOPO, species. Since the HOH-Zn interaction is strong enough to reduce the pK, from 15.7 to near neutral, it is reasonable to assume that the HOP03-Zn interaction reduces the pK from 12.5 to 7 or less. Of these two E/P, species perhaps only the protonated species can form the penta-coordinate intermediate. Certainly, the leaving group in phosphorylating Ser102 from Pi is OH- rather and 02-. When the Pi complex is protonated the proton can be on any of several of the phosphate oxygens. To be in the OH leaving position according to the scheme proposed here the hydrogen would be on the oxygen associated with Znl. If it were on an oxygen associated with Arg166 then the complex would be non-productive with respect to spontaneous phosphorylation. Thus there would be at least three non-productive complexes and the productive complex might be less favored than any of these. The positive charge on Arg166 would repel the proton locally and might enhance or retard protonation of 01 and thus affect the dephosphorylation rate. Hence in this scheme the two metals activate the oxygen of the water and/or alcohol (by Znl) and SerlO2 (by Zn2) as well as stabilizing the appropriate leaving groups and this agrees well with the fact that the two metals are required for rapid reaction (Plocke et al., 1962; Appleburry et al., 1970;

Phosphatase

459

Chlebowski & Coleman, 1976). The distance between the two metals of approximately 4.0 A stabilizes the intermediate and also permits the required motions during the reaction. Two-metal-assisted catalysis, in fact, could be rather general and applicable to other systems. There are at least two other enzymes, namely phospholipase C (Hough et al., 1989) and the exonuclease site of large fragment of -DNA polymerase I (Freemont et al., 1988; Beese & Steitz, 1991), that show basically identical metal constellations and carry out similar hydrolytic reactions. In both of these cases a phosphodiester is the substrate. The situations at Znl and Zn2 are not equivalent for two conspicuous reasons and therefore the symmetry referred to above is far from rigorous. Znl has two histidine ligands and Zn2 has only one. Zn2 is much more deeply buried than Znl as indicated by accessibility calculations using ACCESS, written by Handschumacher and Richards, based on the algorithm by Lee & Richards (1971). Neither position has an ancillary amino acid nucleophile to aid in protonation and deprotonation steps. However, there is a water molecule (Wat465) that is near the Mg co-ordination sphere, and this could be involved in deprotonation of the hydroxyl of Serl02. The electron density maps of Cd/P and Zn/Mg show a positive peak at a similar position, indicating the possible importance of this water. Alternatively, it is possible that water co-ordinating Mg or H-bonding to Lys328 might play a role in proton transfer. The deprotonation of attacking water and/ or alcohol as well as for SerlO2 has been suggested as in the case of carbonic anhydrase (Gettins et al., 1985; Coleman, 1986) and also metal-co-ordinated hydroxide ion is known to promote ester hydrolysis (Hendry & Sargeson, 1989). Fife & Pujari (1988) also reported pronounced effects of divalent meta,l ions on the hydrolysis of phosphate dianion species. The guanidinium group of Arg166 provides ;stabilization of the charged intermediates and transition states along the reaction pathway. It plays an important role in each step of the reaction but is not essential for catalysis. This agrees well with recent studies on various mutants of Arg166 (Chaidarolou et al., 1988; Butler-Ransohoff et al., 1988b) where the mutants are active, but with varied lkinetic parameters. Four hydrogen bonds to Pi in addition to two furnished by Arg166 suggest that redundancy provides tolerance. It should be noted, however, that to date all the wild-type APs from various sources have Arg in this position. In the scheme discussed above the Mg-binding site does not appear to play any direct role in the catalysis steps, although it is well known that addition of Mg enhances catalytic activity by 20% (Bosron et aZ., 1977). It may well be that this site plays an important role structurally by providing the necessary environment for the catalytic steps. The carboxyl group of Asp101 is not directly involved in any of the catalytic steps, but it may serve in the positioning of the guanidinium group of Arg166 and the formation of the 101-112 helix. It is

E. E. Kim

460

and H. W. Wychf

Figure 7. van der Waals’ surface drawing of active site pocket of alkaline phosphatase. Phosphate is shown in the position found in Zn/Mg/P,.

reported that replacement of Zn2 by Mg results in an active enzyme with considerable transferase activity but greatly reduced hydrolytic activity (Coleman & Gettins, 1983aJ). The tetrahedral Zn2 site with one histidine does not seem to be favorable for Mg binding, but evidently at high Mg levels and depleted Zn levels the replacement might be possible. Figure 7 shows the van der Waals’ surface of the active site. Phosphate is drawn at the position found in Zn/Mg/P,. It shows rather a small pocket that is barely large enough to accommodate the phosphate; therefore, the leaving group of the substrate will be left in t’he solvent area. This agrees well with the non-specific nature of the enzyme. The shape of the pocket is puzzling, but perhaps it is t,o allow the motion of displaced water. (d) Comparison

with other spectroscopic

data

Ma,ny spectroscopic studies including e.s.r. and n.m.r. have been carried out with a variety of pure and mixed metal constellations in t’he presence and absence of phosphate and under various conditions 1980; Otvos & Armitage, (see Otvos & Browne2 1980a,b; Coleman & Gettins, 1983a,b; Coleman, 1987, and references therein). These and previous X-ray data have been interpreted as suggesting that there are three histidine residues as ligands to the metal in one metal site, one histidine in a second metal site and no histidine in the third site. Since there are only two histidine ligands to Znl in all three crystal structures reported here rather than the three or four deduced from both previous X-ray data and other spectroscopic studies, some discussion of the evidence is warranted.

The previous X-ray data showed the presence of three histidine residues near Znl so they were ioosely called ligands (Sowadski et aE., 1985) although the electron density did not require this interpretation. The possibility of Asp327 and/or His372 as ligands at different times was suggested later (Wyckoff, 1987). The earliest spectroscopic indication that there are three histidine residues came from an e.s.r. study by Taylor & Coleman (I 972) using the Cu(II)-subst’ituted enzyme. Cu,Mg,AP. Seven hyperfine lines indica,ted the presence of three magnetically equivalent nitrogen atoms as ligands and this led to a suggestion that there were three histidyl ligands to Cu. Accepting this does not neccessarily imply that Zn is similarly co-ordinated, although it appears to be a reasonable posit. On the other hand, the interpretation of the three rr3Cd n.m.r. peaks at 153, 70 and 2 p.p.m., from the Cd(II)-substituted AP by Gett’ins & Coleman (1983a) was based on the e.s.r. and earlier X-ray interpretation, since no firm assignment of the number of nitrogens is possible on the basis of the chemical shifts. n.m.r. data using a,P-[y-13C]dideuteriohistidine AP showed clearly resolved pea,ks for the ten histidine residues in a subunit (Otvos & Browne, 1980). In the holo enzyme only four of these (between 137 and 132 p.p.m.) titrate between pH 5.5 and 9.5. Among the untitrating peaks two are substant’ialiy downfield (at’ 140 and 141 p.p.m.) while four are upfield (between 130 and 126 p.p.m,). All these collapse towards the mid range in the apo enzyme. Three of the upfield peaks show clear splitting due to i13Cd-13C spin coupling when rr3Cd was present in the Znl and Zn2 sites, while they stay as singlets when “‘Cd was substituted (Otvos & Armitage, 1980a). The downfield peaks showed no splitting. These data were interpreted to indicate that there are five histidine residues co-ordinating the metals: t,hree upfield peaks were attributed to co-ordination through NE while the two downfield peaks were att,ributed to co-ordination through ND. NE and ND correspond t,o N, and N,, respectively, in their notation; ND being the nitrogen next to the CG(C,). It is clear from the X-ray crystallographic result’s that the interpretation of the three uptield peaks is valid, but the interpretation for the two downfield peaks differs from our X-ray crystallographic data. It is possible that t’he solution structure differs from the crystal structure, but on the basis of the following considerations we feel that it is probable that some of the spectroscopic data should be reinterpreted. The X-ray crystallographic data show that NE of His372 donates a hydrogen bond t’o one of the carboxyl oxygens of Asp327 (see Table 3) and ND accepts a hydrogen bond from the backbone amide of Ala373, therefore it’ would be singly protonated most of the t)ime. If one assumes that the downfield resonance of CG is governed primarily by the prot,onat’ion state of the adjacent nitrogen, ND (Reynolds et al., 1973), then one would expect His372 to be shifted downfield relative to a free histidine, since it,

Reaction

Mechanism

would be strongly deprotonated due to its hydrogen bond status with an obligate donor, the backbone amide. As for the remaining downfield peak, His86 might be a candidate: NE of His86 is hydrogen bonded to the carbonyl group of Ala415 and ND is at hydrogen bond distance from a water molecule, which is further hydrogen bonded to a backbone amide of ThrlOO and a carbonyl group of Tyr98. Therefore it is reasonable to think that this water acts as a hydrogen bond donor to ND of His86. Both histidine residues are inaccessible in the holo enzyme: His372 is behind the co-ordination sphere of Znl while His86 is located near the subunit interface region, behind the backbone of Asp369His370. It is most likely that they become accessible upon removal of metals as a result of mobility in the Asp369-His370 region, thereby resulting in changes in the spectrum. The hydrogen bond status is the same in both subunits. On the other hand, His162 is a likely candidate for the remaining upfield non-titrating histidine, since it is located in a dip on the surface, making salt-bridge to a carboxylate (Glu134 OE2) and hydrogen bonded to a carbonyl group of Glyl21. Since this is further away from the active site region, it is possible that it would remain the same upon removal of metals. It is interesting to note that His162 is conserved amongst species and a mutant of the adjacent residue Ala161 (E. coli number) to Thr is associated with the inherited disease, hypophosphatasia, in one family (Weiss et al., 1988). The other four histidine residues, His125, 129, 276 and 425, are located ou or near the surface of the protein, and therefore can be titrated. The positions of the inorganic phosphate in Zn(II)AP (E. P) complex and the phosphoryl group in the Cd(II)AP (E-P) covalent intermediate differ from the deductions from phosphate and Cd n.m.r. solution studies (Gettins & Coleman, 1983a,b). J-coupling between 31P and l13Cd was observed in E. P but not in E-P. Furthermore, coupling was seen between Pi and Cdl but not Cd2 (sites A and B in their notation). This coupling was interpreted to indicate inner shell contact when present and no contact when absent. However, the X-ray data here show near-neighbor contact to both metals, 1 and 2, in both the non-covalent and covalent complexes. It is possible that the crystal and solution situations are different or the Zn and Cd complexes are different, but an alternative explanation lies in the nature of the contacts. In the Zn/Mg/P, the P-0-Znl angle is 116” while P-0-Zn2 angle is 175” (see Fig. 5(a)). If the first is more covalent and the second is more ionic and the Cd/P, complex is similar, perhaps this can explain observable J-coupling to Cdl but not Cd2. In the Cd/P, the P-O-Cdl angle is about 155” and therefore may not exhibit coupling. The P-OG-Cd2 angle is 96” but this is the covalently bridging 0 in the P-O-C bonds, not covalently linked to Cd2, and may not support detectable J-coupling. The lack of observable coupling could also be due to heterogeneity and exchange dynamics or to opposing coupling as a function of the P-O-M

qf Allcaline

461

Phosphatase

angle. Thus deductions from the observed couplings are consistent with the X-ray crystallographic (data, but those from the absence of coupling are suspect.

(e) Other APs

There have been several isozymes from both mammalian and other bacterial sources that have been cloned and sequenced in the past few years. When the deduced sequences are aligned witlh the sequence of the E. coli enzyme (25 to 30% identity) after allowing appropriate deletions and insertions, it appears that they all retain the core of the threedimensional structure of the E. coli enzyme with modifications mostly on the surface away from the active site of the enzyme (Kim & Wyckoff, 1989, and references therein for the mammalian enz:ymes; Kaneko et al., 1987; Hulett et al., 1991; Rothschild et al., 1991 for bacterial enzymes). Approximately half of the residues in the central o-sheet are conserved while the other half are replaced by amino acids similar in size or nature; and some of the changes involve pairwise contacting residues. All. the residues serving as a direct ligand to all three rnetals are conserved except for Thr155, which is replaced by Ser in some cases. Also Asp101 and Argl66 are conserved. However, Asp153 and Lys328, whi’ch are directly above the Mg-co-ordination sphere and involved in the hydrogen bonding network, atre replaced by either histidine or tryptophan in other APs. In the case of the mammalian APs His372 is conserved, but this is not universal. By implication, the structure of the active site region of the enzyme appears to be highly conserved, and the reaction mechanism proposed above for the E. coli enzyme should apply to the other APs as well.

We thank Drs J. E. Coleman for providing the protein, A. T. Briinger and S. K. Katti for use of their programs (X-PLOR and van der Waals’ surface diagram) and helpful discussions during refinement, the Center for Structural Biology staff at Yale for technical support. Also we thank Drs Hulett and Claiborne for providing us with sequences of their APs before publication. This work was supported by Public Health Service grant G:M22778 from the Pu’ational Institute of General Medical Science, and supercomputing grant DMB870007P from the National Science Foundation to A. T. Briinger.

References Anderson, B., Milburn, R. M., Harrowfield, J. M. B., Robertson, G. B. & Sargeson, A. M. (1977). Cobalt(III)-promoted hydrolysis of a phosphate ester. J. Amer. Chem. Sot. 99, 2652-2661. Appleburry, M. L., Johnson, B. P. & Coleman, J. E. (1970). Phosphate binding to alkaline phosphatase. J. Biol.

Chem. 245, 4968-4976.

Beese, L. S. & Steitz, T. A. (1991). Structural basis for the DNA exonuclease activity of E. eoli 3’-5’

462

E. E. Kim and H. W. Wyckofl

polymerase I: A two metal ion mechanism. EMBO J., in t.he press. Berstein, F. C., Koetzle, T. F., Williams, G. 9. B., Meyer, E. F.. Jr, Brice, M. D., Rodgers, J. R.. Kennard, O., Shimanouchi, T. & Tasumi, M. (1977). The protein data bank: A computer-based archival file for structures. J. Mol. Biol. 112, macromolecular 535-542. Bloch, W. & Bickar, D. (1978). Phosphate binding to Escherichia coli alkaline phosphatase. J. Biol. Chem. 253, 6211-6217. Bloch, W. & Schlesinger, M. J. (1973). The phosphate content of Escherichia coli alkaline phosphatase and its effect on stopped flow kinetic studies. J. Biol. Chem. 248, 5794-5805. Kosron, W. F., Anderson, R. A., Falk, M. C.; Kennedy, F. S. & Vallee, B. L. (1977). Effect of magnesium on alkaline phosphatase. the properties of zinc Biochemistq, 16, 610-614. Rradshaw, R. A., Cancedda, F.: Ericsson: L. H., Neumann, P. A., Piccoli, S. P., Schlesinger, M. J.; Shriefer; K. & Walsh, K. A. (1981). Amino acid sequence of Es’scherichia co6 alkaline phosphatase. Proc. Nat. Acad. Sci., U.S.A. 78, X73-3477. Rriinger, A. T., Kuriyan, J. & Kaplus, M. (1987). Crystallographic R-factor refinement by molecular dynamics. Science, 235, 458460. Briinger, A. T., Krukowski, A. & Erickson, J. W. (1989). crystrallographic protocols for Slow-cooling refinement by simulated annealing. Acta Crystallogr. sect. A, 46, 585-593. Butler-Ransohoff: J. E., Kendall? D. A., Freeman: S., ,J. R. & Kaiser, E. T. (198%). Knowles, Stereochemistry of phospho group transfer catalyzed by a mutant alkaline phosphatase. Biochemistry, 27, 4i77-4780. Butler-Ransohoff, J. E., Kendall, I). A. & Kaiser, E. T. (19886). Use of site-directed mutagenesis to elucidate the role of arginine-166 in the catalytic mechanism of alkaline phosphatase. Proc. Xat. Acad. Sci., U.S.A. 85, 42764278. Carrel], C. J., Carrell, H. L.. Erlebacher, ,J. & Glusker? J. P. (1988). Structural aspects of metal ioncarboxylate interact,ion. J. ilmer. Chem. Sot. 110. 8651-8656. Chaidaroglou, A., Brezinski, D.