J. Mol. Biol. (1990) 211. 249-254

Three-dimensional Structure of the Complex of Guanylate Kinase from Yeast with its Substrate GMP Thilo Stehle and Georg E. Schulz Institut

fiir

Organ&he Chemie und Biochemie der Universitd Albertstrasse 21, 7800 Freiburg i.Br., F.R.G.

(Received 28 June 1989, and in revised form 9 September 1989) The enzyme guanylate kinase was isolated from baker’s yeast and crystallized as a complex with its substrate GMP. The crystal structure was solved by multiple isomorphous solvent-flattening, restrained least-squares refinement, and simulated replacement, annealing. The current R-factor is 289% at a resolution of 2.0 A. The model is given as a backbone tracing, the GMP binding site is shown in atomic detail. In its major domain (residues 1 to 32 and 82 to 186), the chain fold is closely similar to the adenylate kinases, while the minor domain (residues 33 to 81) differs grossly from the 3-helix fold of the adenylate kinases. Structural homology and mechanistical similarity allow us to assign the AMP site of the adenylate kinases on the basis of the GMP site.

1. Introduction Guanylate kinase (GKaset, ATP : GMP-phosphotransferase; EC 2.7.4.8) is a small monomeric enzyme that catalyzes the reaction MgATP+ (d)GMP=MgADP + (d)GDP. GKase plays an essential role in the biosynthesis of GTP and dGTP. The sequence is known; it consists of 186 amino acid residues with M,=20,548 (Berger et al., 1989). The sequence was aligned to the adenylate kinases; clear homology occurs in regions A and F (Schulz et al., 1986). Region A contains the nucleotide-binding Gly-pattern (Dreusicke & Schulz, 1986), which is found also in the structurally known GDP/GTPbinding c-H-ras protein (DeVos et al., 1988). While adenylate kinase structures have been solved either without substrate (Dreusicke et al., 1988) or with the symmetric double substrate-mimicking inhibitor Ap,A (Egner et aE., 1987; Mueller & Schulz, 1988), crystalline GKase has one of the substrates, GMP, bound, Assuming homology, the GMP position in GKase contributes to the binding site assignment in adenylate kinases, which is being discussed controversially.

TAbbreviations used: GKase, guanylate kinase from baker’s yeast; PEG, polyethylene glycol; m.i.r., multiple isomorphous replacement; s.i.r., single isomorphous replacement; AKlase, cytosolic adenylate kinase from vertebrates; AKr,,ase, adenylate kinase from baker’s yeast; AK,,,ase, adenylate kinase from Escherichia coli: Ap,A,P’,P5-bis(5’-adenosyl)pentaphosphate; r.m.s., root-mean-square. W2-2836/90/01024946

$03.00/O

2. Enzyme Preparation and Crystallization The purification of GKase from baker’s yeast followed established lines (Berger, 1988). Crystals were grown with the hanging drop method at 20°C: the crystallization conditions differed slightly from those used by Berger et al. (1989). The lo-p1 drops contained 7 mg GKase/ml, 40 mw-potassium phosphate (pH 5.5), 1 mM-GMP, lqb (w/v) PEG-1500 (analytical grade, Serva) and 1.0 M-ammonium sulfate as precipitant. As reservoir, we used 500 ~1 of 40 mM-potassium phosphate (pH 55) with 2.0 M-ammonium sulfate. The addition of PEG-1 500 caused an appreciable improvement with respect to crystal size and reproducibility. Crystals grew within 1 to 3 days to a maximum size of 1100 pm x 600 pm x 300 pm. The addition of PEG-1000 instead of PEG-1500 increased the crystal size further, but’ decreased reproducibility. Since no enzyme-free storage buffer could be found, crystals were soaked in the drops and mounted using the drop solvent.

3. Data Collection and Analysis Data were collected on a four-circle diffracto(modified model P2,, Nicolet, 1J.S.A.) at 6°C using nickel-filtered CuKcr radiation as described (Thieme et al., 1981). The omega scan ranges were typically 04” to 66”. Data were collected in ten shells of about 1500 reflections each, which were then scaled using a separate native data set of 300 strong reflections. In order to produce heavy-atom derivatives for m.i.r. phase determination, GKase crystals were soaked with heavy-atom compounds at 20°C. Int’ensity changes were monitored by precession photographs of the hOl-plane (camera Y925, rotating anode generator GX6, Enraf-Nonius, 249 meter

0 1990 Academic Press Limited

1

1.0

2.0

03 104

PIP

K,PtCI,

K,HgKW, VO,Ac,(l

59 5.9

.59

3.6

3.6

2.6

Resol (A)

8.1

243

56

49

Rid r%,)

132

231

32.9

8.0

11.4

17.7

Table 1

0.338 0.683 0399 0.649 0.404 0.692 B254 0.473

0268

0.336

0.246

.T

0330 0.223 0.328 0096 @731 WO82

0.109

@734 0.722

07.71 w712

!/ 0060 0065 0.059 0021 0027 0255 0.035 @024 0,066 0103

0.066

2

parameters

Fractional heavy-atom co-ordinat,es

of heavy-atom

58 76 62

loo

25 27 6 3

19

28

58

Ore. r%,)

56 243

19

57 71

80

40 65 5.5 34 61

Temp. factor (AZ)

2.3 I.1

2 I

1%

14

2.X

CL-59

0.9

0.7

1%

59-3.6

ranges (A)

I.4

3fL26

Phasing power F/IX in

The refinement is based on 1)ickerson el a,l. (1968). The mean figure of merit is @91 in the resolution range CCIto 5.9 A. and 0.55 in the range z to 2.6 3. t MMA. methylrnercuryacetatr; PIP, di-/~-iodobis(ethylenediamine)-di-plati~~um(II)nitrate. $ A drop volume of 5 ~1 is assumed before addition of the heavy-atom compound to the drops. Q The R-fact,or between data sets 1 and 2 is defined as tl=2CIF, - F,j/X(14’, +F,). where E’, and Fz aw the structure factor amplitudes. For K,,,. the data wts arc symmetry-relatt~l reflections on zones h/cl and hki. R,,, compares derivative with native data. /I Qwtaliization and handling in citratr instead of phosphate buffer.

2 2

2

2

1.0

KAuC’I,

15

Time (days)

1.5

Cow.: (mM)

Soaking conditions

MMA

Heavyatom compound?

Re$nement

Structure of Guanylate Kinase-GMP

251

Figure 1. Stereo view of the final solvent-flattened electron density map at 2.6 A resolution with a model of Glu66-Phe67-Ile68-G1u69-Trp7O-Ala71. The cut level is at 15% of maximum density. The current model after simulated annealing using XPLOR (Bruenger et al., 1987) is depicted.

were refined Heavy-atom parameters NL). and phases were determined with the method of Dickerson et al. (1968). The resulting structure factors at 2.6 A resolution (1 A=@1 nm) were then processed by solvent-flattening (Wang, 1985) using the program of Leslie (1987). The known sequence was fitted to the resulting density map with the program FRODO (Jones, 1978) run on a PS-330 graphics display (Evans & Sutherland, U.S.A.). The initial model of GKase was first refined with the TNT program package (Tronrud et al., 1987) in the resolution range 7.0 to 2.6 A. The result was processed by simulated annealing (program XPLOR; Bruenger et al., 1987) in the range 7.0 to 2.0 A. The final model was checked with (Fobs- Fcalc) exp iacalc and V,I,, - Fcad exp ia,,lc maps. All calculations were done on a Vax11/730 (Digital Equipment, U.S.A.).

4. Crystal Properties and Quality of X-ray Data X-ray photographs showed that the crystals diffract to a resolution better than 1.8 8. The space group was determined as P&2,2, with cell dimensions a=b=5@8( +O*l) A and c= 1553( f0.2) A, where the uncertainties reflect the variations between different native crystals. The densities of crystals solvent and were determined as 1.271kO.02) g/ml and 1.16 g/ml, respectively, from which one calculates 1.2( +@2) molecules per asymmetric unit. With one molecule per asymmetric unit, the crystal contains 50% solvent. Native data were collected to 2.0 d resolution from six different crystals. The radiation damage was 4 o/o) on average; in the worst cases it reached 20%. Up to 2.7 A resolution, the_ data were measured twice and merged. The internal R-factor (Table 1) calculated between 540 symmetry-related reflections in the resolution range 00 to 2.0 ir, was 5.0 %. In the range 2.1 A to 2.0 8, the internal R-factor as derived from 44 reflections was still as low as 13.7 %. These values demonstrate the generally good quality of the native data. The derivative data sets were collected

from one crystal each; up to 15% radiation damage was conceded. The Rint values of Table 1 indicate reasonable data accuracy.

5. Multiple Isomorphous Replacement Phases and Solvent-flattening at 2.6 A For m.i.r. analysis, we soaked 36 different heavyatom compounds under various conditions. Six soaks contributed to phase determination as specified in Table 1. As the initial step, the difference Patterson of the K,PtCl,-derivative was interpreted in terms of the main site. All other heavy-atom positions were found by difference Fouriers. Table 1 shows the refined heavy-atom parameters, which were used to calculate m.i.r. phases and a corresponding electron density map at 2.6 A resolution. Beyond 46 i% resolution, however, we obtained essentially only s.i.r. phases because the phasing power of two derivatives dropped below 1.0. It was not possible to follow the whole chain unambiguously in this map. The m.i.r. and s.i.r. phases were then modified by solvent-flattening at 2.6 d resolution, which improved the phase quality substantially. The unit cell was sampled at 60 x 60 x 180 grid points. Firstly, we derived the protein mask from the m.i.r. map using an averaging sphere of 10 d and a level adjustment to 40% solvent. Applying this mask, we ran through four solvent-flattening iterations. The resulting map was used for calculating a second mask in the same manner as the first. Three further iterations were done and a final mask was derived using again the 10 il sphere and 40% solvent, which is appreciably lower than the observed fraction of 50%. With seven more cycles, solvent-flattening was then driven to convergence. The average phase angle changes of the last two cycles were below 05”. A comparison between solvent-flattened and m.i.r. structure factors showed an average phase angle difference of 35”; the mean figure of merit increased from @55 (m.i.r.) to 085. The solvent-flattened map allowed us to trace the polypeptide chain

252

T. Stehle and G. E. Schulz

Table 2 XPLOR stage

rejinement

protocol

(76’

to 2.Od)

Description

Determination of weights WB and WPt Minimization, 10 conjugate gradient steps using soft repulsive potential followed by 40 steps with CHARMM non-bonded potential. bVA= 81300 kcal/mol, WP= 8500 kcal/(mol rad’),$ B= 13 A’, AF=@05 & Molecular dynamics 0.5 ps, T= 2000 K, timestep = 1 fs, AF=@3 A Molecular dynamics 0.25 ps, T= 300 K, timestep = 1 fs. AF=@3 i! Molecular dynamics 0.05 ps, T= 300 K. timestep = I fs, AF=@2 A Minimization, 50 steps with AF=O.Ol All 10 cycles of individual B-factor refinement with standard deviations between B-factors of bonded atoms restrained to 1.5 AZ and B-factors of atoms connected by an angle restrained to 2.0 A2 Minimization, 10 steps with AF=@Ol A The XPLOR program package was received from A. T. Bruenger (Hruenger et al., 1987). The whole process required 19 days central processor unit time on our Vax 1 l/730. t WA. weight for the effective energy term accounting for the diffraction data, E,(XRAY). WP, weight for the effective energy term accounting for the phase information, E,(XRAY). These weights relate the X-ray effective energy terms E,(XRAY) and E,(XRAY) to the empirical potential energy. $ 1 Cal=418 J. 0 AF is a limit,. If any atom movement exceeds AF, the 1st derivatives of the effective energy EA(XRAY)+E,(XRAY) are recalculated. B is the crystallographic temperature factor. 11At this and the following stages, WP was set to zero.

unambiguously over its whole length. After building an initial model into this map, the heavy-atom sites could be assigned to particular residues. The three mercury as well as the two gold sites were located at Cys95. the only cysteine residue in GKase. Platinum was bound to Met61 and MetlOl, the only methionine residues of GKase. The quality of the solventflattened map is demonstrated in Figure 1.

6. Refinement at 2.0 A The initial model of GKase was subjected to a restrained refinement (Tronrud et aE., 1987). After 46 cycles with 6362 independent reflections in the resolution range 7 d to 2.6 A (= 1OOo/o) the procedure was stopped at an R-factor of 34.4% and reasonable model geometry (r.m.s. deviations of 0.10 A for bond lengths and 10” for bond angles). The data were then extended to 2.0 A resolution, and simulated annealing (XPLOR; Bruenger et al., 1987) was started with 13,646 independent reflections in the range 7.0 to 2.0 B (= 100%). After minimization, the molecular dynamics option of XPLOR was run for @5 ps at 2000 K and 0.3 ps at 300 K followed by a second minimization, ten steps of temperature factor refinement and a third minimization. The final R-factor was 28.9o/o at good

model geometry (r.m.s. deviations of 0+12 A in ;t)ontl lengths and 5” in bond angles). A more detailed description of the procedure is given in Table 2. At this stage, a (2F,,,,, - F& exp icl,,,c map showed that most of t,he backbone and the side-chains werf placed correctly and the substrate GMP was wr:Il defined. Only in the loop regions around residues 48, 92 and 137, and at, the three C-terminal residues art’ the densities still weak so that further small barkbone movements have t,o be expected. Moreover. some side-chain conformations are not yet final. ‘I’h(s resulting backbone tracing of GKase is givrrr in Figure 2(a).

7. The Structure Since the catalyzed reaction and the reaction mechanism of GKase resemble those of the adenylate kinases (Miech & Parks, 1965), we had expected that these enzymes are homologous. A sequence comparison yielded 17O/b identical amino acid residues with clear homology only in regions A and F (Berger et al., 1989), which points to a rather distant relationship. Now, the chain fold of GKase shows that the major domain (residues 1 to 32 and 82 to 186 of GKase), which contains A and F, corresponds to the adenylate kinases. The minor domains (residues 33 to 81 of GKase) differ grossly. Here, the chain fold of GKase is a four-stranded b-sheet plus one cl-helix (Fig. 2(a)), whereas the adenylate kinases have three a-helices. A superposition of the C* backbones of GKase and AKlase is shown in Figure 2(b). The surprisingly large structural difference between GKase and AK 1ase accompanying the rather small structural change from adenine to guanine could be related to the substantial solubility difference between these purines. As has been pointed out by Bash et al. (1987), it is much more difficult to extract GMP from water and bind it than to do the same for AMP. Details of the GMP-binding pocket are shown in Figure 3. The guanine ring is buried deeply in a cleft between the two domains of the enzyme and is shielded by three surrounding aromatic residues. The ribose points outwards into solution. At the present stage of refinement, it is clear that at least, Glu69, Ser34, Ser80 and Asp100 interact, with the guanine ring and determine the base specificity of the enzyme. with GKase : GMP A superposition of AKystase :Ap,A based on the chain fold similarity showed that GMP as bound to GKase (Fig. 2(a)) overlaps the AMP moiety at site B of Ap,A as bound to AKystase (Egner et al., 1987); the residual distances are about 3 d, 5 a and 5 A for the phosphate groups, riboses and bases, respectively. The same applies for a superposition of GKase : GMP with AK,,,ase:Ap,A (Mueller & Schulz, 1988). Since the adenylate kinases use ATP and AMP as substrates, the known binding modes of the symmetrical Ap,A molecule do not allow us to distinguish between the mono- and the triphosphate sites. Assuming homology between GKase and the

Structure of Guanylate Kinase-GMP

253

Figure 2. (a) Stereo drawing of the backbone of the current model of the GKase : GMP complex. The substrate GMP is bound in a large cleft, which divides the molecule into 2 domains. (b) Superimposition of the backbones of GKase (thick lines) and AKlase (thin lines). The overlay is based on the central P-pleated sheets, the AKlase model is taken from Dreusicke et al. (1988). The major domain at the rear left-hand side (residues 1 to 32 and 82 to 186) is similar to the equivalent domain in adenylate kinase, whereas the minor domain at the front right-hand side (residues 33 to 81) differs grossly from adenylate kinase. The given view corresponds to the usual adenylate kinase presentation (Dreusicke et aE., 1988). In comparison with (a), the GKase model has been rotated about 40” around an approximately vertical axis. The filled dot marks the center of the purine ring of GMP.

Figure 3. Stereo view of the GMP-binding site of GKase. All residues within 5 A distance from any atom of the bound GMP are shown. They are in 7 chain segments consisting of residues: 10 (Ser), 33 to 38 (Val-Ser-Ser-Thr-Thr-Arg), 41 to 44 (Arg-Ala-Gly-Glu), 50 (Tyr), 69 to 73 (Glu-Trp-Ala-Gln-Phe), 78 to 80 (Tyr-Gly-Ser) and 98 to 103 (Asp-Ile-Asp-MetGln-Gly). The guanine ring of GMP (thick line) forms direct contacts with Ser34, Glu69, Ser80 and AsplOO, and is surounded by 3 aromatic residues (Tyr50, Phe73 and Tyr78). The ribose hydroxyl groups contact the protein only through water (not shown). The phosphate is hydrogen-bonded to the hydroxyl groups of Tyr50 and Tyr78 and the guanidinium groups of Arg38 and Arg41. As compared to Fig. 2(a), the model has been rotated by about 80” around the vertical axis and then by about 20” around the horizontal axis.

254

T. Stehle and G. E. Schulz

adenylate kinases, however, site B of Ap,A can now be assigned as the AMP site. This conclusion does not necessarily imply that site A of Ap,A in AK,,,ase or AK,,,ase is the ATP site. The rather weak binding at this site in AK,,,,ase : Ap,A and in AK,,,ase : Ap,A as well as the lack of sequence homology around this site in the adenylate kinases in general had raised doubts about such a correspondence (Egner et al., 1987). Several proposals for the nucleotide binding sites of adenylate kinases have been derived from nuclear magnetic resonance data (Smith & Mildvan, 1982; Fry et al., 1985, 1986, 1987, 1988; Shyy et al., 1987; Roesch et al., 1989), but none of them assigns the AMP site as we do here. Pai et al. (1977) had detected the AMP site by substrate analog binding to crystalline AKlase, but had suggested that it binds ATP. While the chain folds of GKase and the GDP/GTP-binding c-H-ras protein (DeVos et al., 1988) are different, only the pa-units containing the Gly-pattern (Dreusicke & Schulz, 1986) resemble each other. In relation to these pa-units, the GDP/GTP site of c-H-ras is far away from the GMP site of GKase. It is nearer to site A of AK,,,,ase : Ap,A or AK,,,ase : Ap,A. We thank K. Diederichs for discussions.

References Bash, P. A., Singh, U. C., Langridge, R. C Kollman, P. A. (1987). Science, 236, 564-568. Berger, A. (1988). Dissertation, Universittit Freiburg i.Br. Berger, A., Schiltz, E. & Schulz, G. E. (1989). Eur. J. Biochem. 184, 433443. Bruenger, A. T., Kuriyan, J. & Karplus, M. (1987). Science, 235, 458460. Edited

DeVos, A. M., Tong, I,., Milburn, M. V., AMatias. P. M., Jancarik. J.. Noguchi, S., Nishimura, S.. Miura. K.. Ohtsuka, E. &, Kim, S.-H. (1988). Science, 239. 888-893. Dickerson, R. E., Weinzierl, J. E. & Palmer, R. A. (1968). Acta Crystallogr. sect. B, 24, 997-1003. Dreusicke. D. & Schulz. G. E. (1986). FEBS Letters. 208. 301-304. Dreusicke, D.. Karplus, P. A. & Schulz, (:. E. (1988). J. Mol. Biol. 199, 35!&371. Egner, U., Tomasselli, A. G. &, Schulz. (:. E. (1987). J. Mol. Biol. 195, 649-658. Fry, 1). C., Kuby, S. A. & Mildvan, A. S. (1985). Biochemistry, 24, 4680--2694. Fry. D. C.. Kuby, S. A. & Mildvan. A. S. (1986). I’roc:. LVat. Acad. Sci., U.S. A. 83. 907-911. Fry, I). r.. Kuby, S. A. & Mildvan, A. S. (1987). Biochewkstry, 26, 164551655. Fry. 1). (‘.. Kyler. I). M.. Susi, H., Brown, 1%M.. Kuby. S. A. & Mildvan, A. S. (1988). Biochemistry, 27, 3588-3598. Jones. T. A. (1978). J. Appl. Crystallogr. 11. 268-272. Leslie, A. G. W. (1987). Acta Crystallogr. sect. A, 43, 134-136. Miech, R. P. & Parks. R. E. (1965). J. Biol. (‘hem. 240, 351-357. Mueller, C. M. & Schulz. G. E. (1988). J. Mol. Kiol. 202. 909-912. Pai, E. F., Sachsenheimer, W., Schirmer, R. H. bz Schulz, 0. E. (1977). J. Mol. Biol. 114, 3745. Roesch, P., Klaus. W., Auer, M. & Goody. R. S. ( 1989). Biochemistry, 28, 43184325. Schulz, G. E.. Schiltz, E.. Tomasselli, A. (i.. Frank. I