WO1997042311A1

WO1997042311A1 - Crystal structure of human cytomegalovirus protease

Info

Publication number: WO1997042311A1
Application number: PCT/US1997/007128
Authority: WO
Inventors: Lisette LEGACÉ; Chungeng Qian; Liang Tong
Original assignee: Boehringer Ingelheim Pharmaceuticals, Inc.; Boehringer Ingelheim (Canada) Ltd.
Priority date: 1996-05-07
Filing date: 1997-05-07
Publication date: 1997-11-13

Abstract

The present invention relates to a crystal structure of human cytomegalovirus protease (HCMV PR). This crystal structure is useful for the design and optimization of inhibitors against herpesvirus proteases.

Description

CRYSTAL STRUCTURE OF HUMAN CYTOMEGALOVIRUS PROTEASE

TECHNICAL FIELD OF THE INVENTION

BACKGROUND OF INVENTION

Human cytomegalovirus (HCMV), a herpesvirus, infects up to 70% of the general population and can cause morbidity and mortality in immuno-suppressed individuals (organ transplant recipients, chemo-therapy and AIDS patients) and congenitally infected newborns.¹ HCMV protease (HCMV PR) is essential for the production of mature infectious virions as it carries out a proteolytic processing near the C-terminus (maturation or M-site) of the viral assembly protein precursor (for a review, see 2). HCMV PR is a serine protease,² though with no significant homology to other clans of serine proteases.²-³

SUMMARY OF THE INVENTION

The present invention provides a crystal structure of HCMV PR at 2.0A resolution and establishes the existence of a new polypeptide backbone fold. Ser-132 and His-63 are found in close proximity in the active site, confirming earlier biochemical and mutagenesis studies.² The structure also suggests that the third member of the triad is most likely His-157. A dimer of the protease with an extensive interface is found in the crystal structure. BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1. (A). Schematic drawing²² of the structure of HCMV PR. The β-strands are shown as arrowed ribbons. The active site residues (Ser-132, His-63 and His-157) are shown. The secondary structure elements are labeled starting from the N-terminus, the β-strands numerically (βl through β8) and the α-helices alphabetically (αA through αG). The open end of the β-barrel is toward the upper left corner of the plot. (B). Topological drawing of the backbone fold of HCMV PR. Strand β3 is broken at the cross-over position, and the break points are represented as dark squares in the plot. The secondary structure elements are labeled. The loops are numbered from the N- terminus (LI through L16). The location of Ser-132, His-63 and His-157 is also shown.

Figure 2. Alignment of representative sequences of herpesvirus proteases. A dash (-) represents an amino acid identity with the HCMV PR sequence and a dot (.) represents a deletion. The secondary structure elements (S.S.) are shown and labeled at the bottom of the alignment, where a cross (x) represents a residue that is not included in the current atomic model due to disorder. Residues in red highlight are the catalytic triad, in green are those that form the hydrophobic core in the β-barrel, in yellow those that form the hydrophobic core between the β-barrel and helices αD, αE and αG, and in cyan those that form the dimer interface. HSV-1 — Herpes simplex virus type 1. ILTV — infectious laryngotracheitis virus protease. EBV — Epstein-Barr virus protease.

Figure 3. (A). Superposition of residues Ser-132, His-63 and His-157 in HCMV PR with the catalytic triads of chymotrypsin, subtilisin, and papain. The superposition was done manually, overlapping the C_α atom of Ser-132 and the side chain imidazole ring of His-63 with the equivalent atoms of the other triads. (B). The molecular surface of HCMV PR near the active site.²³

Figure 4. (A). The dimer of HCMV protease. The two-fold axis ofthe dimer is shown as a triangle. The zinc ions are shown as spheres, marking the positions of the active sites. The dimer interface involves essentially residues in helix αF packing against helices αB, αC and αF in the other monomer. The view of this figure is roughly 90° away around the vertical axis from that of Fig. IB. (B). Superposition of the C_α trace of the dimer in 0.15M Na2SU4 with that in 0.4M LiCl. The superposition was done with residues from one monomer only.

Fipure 5.(A)-(T). The atomic coordinates of HCMV PR.

DETAILED DESCRIPTION OF THE INVENTION HCMV PR contains 256 amino acid residues and shares significant sequence homology with the proteases of other herpesviruses.² Besides processing the assembly protein precursor, the protease also catalyzes the cleavage at the R-site (residue 256) to release itself from its full-length gene product. Herpesvirus proteases prefer an Ala residue at ?\ and a Ser (or Ala) residue at Pi'.² HCMV PR also has two internal cleavage sites (after residues 143 and 209). The HCMV PR used in arriving at the present invention contains a mutation (A143Q) to disable one of these sites.⁴ Limited processing at the other site was observed in solution but the crystals contained only the intact protease.

The backbone fold of HCMV protease is different from those of the other clans³ of serine proteases with known structures (chymotrypsin, subtilisin and serine carboxypeptidase)⁵ _' ⁶ and represents a new fold for serine proteases. This also establishes the herpesvirus proteases as a new clan of serine proteases.³ Moreover, an examination of the SCOP database⁷ failed to show any other structures with a similar fold, suggesting that the HCMV PR structure may be the first observation of this protein fold. The structure consists of a seven-stranded mostly anti-parallel β-barrel (βl through β7) which is surrounded by seven helices (αA through αG) (Fig. 1). Alternatively, the β-barrel can be described in terms of two β-sheets, with strands β3, β4, βl and β7 in one sheet and strands β3, β2, β6, β5 in the other. Strand β3 switches between the two sheets on one side ofthe barrel. A short strand of three residues (β8) is hydrogen-bonded to strand β3 but it is not part of the β-barrel. The two β-sheets are arranged such that one opening of the β-barrel is rather narrow whereas the other is somewhat more open. All the amino-acid side chains that point towards the inside of the β-barrel are hydrophobic and highly conserved (Fig. 2). Of the helices, two (αB and αC) are located near the narrow end and four (αD, αE, αF, αG) surround the middle of the barrel. Helix αA represents the few residues of the protein that are located near the open end of the barrel. Another hydrophobic core is formed by the packing of the β-barrel (strands β3, β4, βl and β7) against helices αD, αE and αG. Residues involved in this core are also mostly conserved in sequence (Fig. 2). The only cross-over connection between the β-strands is right-handed and is at the narrow end of the barrel. Five conserved regions that were identified based on sequence comparisons of herpesvirus proteases² correspond mostly to the seven strands of the β-barrel. Therefore, the pattern of sequence conservation of herpesvirus proteases reflects the structural requirement of maintaining the fold.

About 40 residues of HCMV PR are not observed in either molecule of the dimer in the current structure model. These are in four stretches — 1-8 at the N-terminus (loop LI), 46-52 (loop L3), 136-152 (loop L9) and 202-210 (loop LI 3). They correspond to regions of pronounced sequence variability among herpesvirus proteases (Fig. 2). These residues are expected to be on the surface of the protein and are likely to be flexible in structure. The two intemal cleavage sites of HCMV PR are both located in the disordered stretches. It can be expected that their flexibility and surface location may make them accessible to attack by the protease.

As commonly done with the other protease families, we propose that all herpesvirus protease sequences be referred to based on the HCMV PR sequence numbering. Site- directed mutagenesis and affinity labeling studies have identified Ser-132 as the catalytic nucleophile of the herpesvirus protease.² In the structure, Ser-132 is located in strand β5 and is on the outside of the β-barrel. Residue His-63 (in loop L4) is located near Ser-132, in agreement with mutagenesis studies suggesting it to be the second member ofthe triad.² The other conserved histidine residue, His-157 (in strand β6), which was identified to be important (though not essential) for catalytic activity by mutagenesis studies,² is located near His-63. None of the Asp/Glu residues that have been proposed as the third member of the triad based on mutagenesis studies² are located near the His-63 residue. The structure therefore suggests that His-157 is likely to be the third member of the catalytic triad in herpesvirus proteases. The relative positions of the Ser-132, His-63 and His-157 side chains in HCMV PR are similar to those found in other serine/cysteine proteases,⁶-⁸ especially the chymotrypsin family (Fig. 3). Superposition of Ser-132 and His-63 with the equivalent residues in chymotrypsin leads to a general overlap ofthe positions of His- 157 in HCMV PR and Asp- 102 in chymotrypsin as well, providing further support that His-157 may be the third member of the triad in herpesvirus proteases. This is unique as the other known triads generally contain an Asp, Glu or Asn residue as the third member. The catalytic activity of the H157D and H157E mutants, though about three-fold better than that of the H157A mutant, is still about 5-10 times worse than that of the wild-type enzyme (data not shown). Further studies will be needed to determine how His-157 performs its function as the third member of the catalytic triad.

The side chain hydroxyl group of Ser-132 is somewhat disordered in the current structure. Its position as observed here may not correspond to the optimal one for catalysis. The distance between the hydroxyl group and the NE2 atom of His-63 is 3.3A, whereas that between the ND1 atom of His-63 and the NE2 atom of His-157 is 2.8A. The side chain of His-157 is mostly buried (accessible surface area lθA²), whereas that of His-63 is partly exposed to solvent (70A² surface area).

Despite the general similarity in the disposition of the catalytic residues, the difference in the backbone folds of the various enzymes means that the substrate binding regions are located differently relative to the catalytic nucleophile. The oxyanion hole in HCMV PR is probably formed by the main chain amido groups of residues 165 and 166. A water molecule is associated with the two groups in the current structure. The Ser-132 side chain is situated in a depression on the protein surface (Fig. 3). Residues in strands β5 and β6, helices αA, αF and αG, and loops L2, L4 and LI 5 are located near the active site and may contribute to the formation ofthe substrate binding pockets. Residues in helix αB, from the other monomer of the protease dimer, are also located nearby. Residues Cys-161, Arg- 165 and Arg- 166 in this region are highly conserved among herpesvirus proteases. The side chain of Cys- 161 is located next to that of Ser-132 and may be involved in the formation of the Sj (or Si') binding pocket. Residues 136-152, disordered based on the current crystal structure, could be near the active site and might contribute to the binding of substrates.

Residues at the C-terminus of the protease are located in a groove far from the active site and are covered with residues from loop L7, suggesting that there are conformational changes for the C-terminal residues after the cleavage at the R-site. The C-terminal carboxylate group forms a salt bridge with Lys-242, which is conserved to be a Lys or Arg among all herpevirus proteases (Fig. 2).² Lys-254 forms a salt bridge with Glu- 122, which was proposed to be the third member of the triad.² Tyr-253, strictly conserved among herpesvirus proteases, is situated in a hydrophobic pocket with the side chain hydroxyl group forming a hydrogen-bond with the carbonyl oxygen atom of residue 1 16 (loop L7). The structural importance of the Tyr-253 side chain may explain why this residue is conserved at the R-site of all herpesvirus proteases but is never observed at the M-site.²

Despite being a serine protease, the activity of herpesvirus proteases is not easily inhibited by common serine protease inhibitors.² However, Zn⁴^ ions, which normally inhibit cysteine proteases, can inhibit herpesvirus proteases at micromolar concentrations.² X-ray diffraction data to 3.2 A resolution were collected on a crystal of HCMV PR that was soaked overnight in 5mM zinc sulphate. The difference electron density map showed a binding site for Zn^"1-1" in the protease, located near the side chains of His-63 and His-157. This suggests that zinc is an active site inhibitor of herpesvirus proteases, owing to the presence of two histidine residues in the catalytic triad.

The crystals of HCMV PR described herein contain dimers of the protease. Recent studies have shown that HCMV PR exists in a monomer-dimer equilibrium in solution and that the active enzyme may be a dimer.⁹'¹⁰ The dimers as observed here obey proper two-fold symmetry (with the exception of residues 25-46). The rms distance between 179 equivalent C_α atoms of the two monomers is 0.28 A. The dimer interface is formed mostly by helix αF packing against its symmetry-mate and helices αB and αC in the other monomer (Fig. 4). Helix αB has a kink near this interface, probably to prevent steric clashes with helix αF in the other monomer. The buried surface area per monomer is about 1600A², suggesting that the interface is rather extensive. The residues at this interface show reduced variations among herpesvirus proteases (Fig. 2). It may be expected that all herpesvirus proteases can form similar dimers.

The specific activity of herpesvirus proteases is rather low as compared to that of other proteases.² In the presence of 0.5M Na2SO₄, 50% glycerol or many other agents, the activity can be increased by 10 to 100-fold A⁹"¹² Even with this increase, the activity is still significantly lower than that ofthe other proteases. This might be a reflection of a weaker triad in herpesvirus proteases. To identify potential conformational changes in HCMV PR upon the introduction of Na2SO₄, crystals (originally in 0.4M LiCI) were exposed to 0.15M Na2SO4 and a change in the dimer organization was observed (Fig. 4). Changes in the conformations of individual residues may be possible as well, but the crystals in LiCl do not diffract well enough to permit a detailed structural comparison. The catalytic activity of HCMV PR increased about 3-fold by the change from 0.4M LiCl to 0.15M Na2SO4 (data not shown). However, with the current artificial mother liquor, the solubility of Na2SO4 is close to 0.15M and a new set of conditions will be needed to look for any further changes in HCMV PR at higher concentrations ofthe salt.

The increase in the catalytic activity of herpesvirus proteases is due to a decrease in the K_m and/or an increase in the &_Ca_t-^4,9"¹² It is possible that Na2SO4 and other agents stabilize the conformation of the loops that form the substrate binding pockets and/or modify the shape of this binding region to facilitate catalysis. The observed change in the dimer organization in the presence of 0.15M Na2SO₄ may be related to this enhancement of catalytic activity. In addition, the disordered residues, especially 136- 152, might become more ordered in the presence of higher concentrations of Na2SO₄. These residues may also become more ordered upon the binding of inhibitors in the active site. Structure determination of inhibitor complexes of HCMV PR is currently in progress. The structure information presented here, together with that from inhibitor complexes, will greatly help to rationalize the structure-activity relationships of inhibitors as well as to design and optimize new inhibitors against herpesvirus proteases.

The atomic coordinates set forth in Figure 5 (A)-(T) define the three dimensional configuration of HCMV protease. These atomic coordinates are useful for obtaining SAR information relating to HCMV protease inhibitors and candidate inhibitors. For example, the coordinates may be used for: (1) facilitating the structure determination of inhibitor complexes of HCMV protease and (2) identifying, designing and optimizing the structures of HCMV protease inhibitors. The inhibitors of HCMV protease thus identified, designed or optimized would be therapeutically effective anti-herpetic agents. In addition, the SAR information obtained using the atomic coordinates would assist in the rational drug design of HCMV inhibitors.

For example, the crystallographic information from Figure 5(A)-(T) can be directly used in computer assisted methods for rational drug design. Several successful de novo and quasi-de novo approaches have been developed, including DOCK (described in I.D. Kuntz et al., J. Mol. Biol.. 161, p. 269 (1982)) and LUDI (described in H.J. Bohm, J. Comp.-Aided Mol. Design. 6, p. 61 (1992)). In general, DOCK selects from a database of chemical structures, predicting those which are complementary in shape and electrostatics to a particular binding site. LUDI uses a database of chemical fragments and assembles molecules that complement the targeted binding site. These and other computer programs will be well know to those of ordinary skill in the art. Once the relevant atomic data has been analyzed by such programs, candidate inhibitors can be identified, prepared and tested for their ability to inhibit the HCMV protease using known techniques.

It is also possible to use the HCMV protease atomic coordinates to predict how well a candidate inhibitor will bind to the protease. Force field models such as CHARMM (Molecular Simulations Incoφorated, Burlington, MA) or AMBER (P.A. Kollman, UCSF) may be used for such predictions. If a particular functionality is poorly represented, published structural information for molecules containing such functionalities can be analyzed and, if necessary, ab initio calculations can be carried out to determine the preferred conformations and energy differences between conformations. As a result, more accurate parameters describing these functional groups can then be derived for the CHARMM and/or AMBER force fields for use in subsequent calculations.

In order that this invention be more fully understood, the following examples are set forth. These examples are for the puφose of illustration only and are not to be construed as limiting the scope ofthe invention in any way.

EXAMPLES

Several obstacles were encountered and overcome during the structure determination of human cytomegalovirus protease. Dehydration of crystals, by exposing them to higher concentrations ofthe precipitant, reduced the mosaicity ofthe crystals and may have also resolved their microscopic twinning. The initial phase information was obtained with the seleno-methionyl multiple-wavelength anomalous diffraction technique. However, site-specific mutagenesis was required to introduce extra Met residues into the protease. The phase information had to be improved by non- crystallographic symmetry averaging, initially among three "crystal forms". A change in the composition of the artificial mother liquor led to a significant improvement, from 3.0A to 2.0A resolution, in the diffraction quality of the crystals. The experiences reported here may prove useful to structure determination of other proteins.

DETERMINATION OF CRYSTAL STRUCTURE

The crystal structure was determined by the seleno-methionyl multiple-wavelength anomalous diffraction (MAD) technique.¹³ The introduction of two additional Met residues by mutagenesis (T181M, L229M) was crucial for this process. Non- crystallographic symmetry averaging (among crystal forms) was essential to improve the original MAD phases. Crystallization: HCMV PR was expressed in E. coli and purified by anion and cation exchange chromatography. Initial crystallization condition was found by the sparse-matrix sampling method,¹⁴ with a commercial kit (Hampton Research, California). Purified protease was concentrated to about 25 mg/ml in 0.1 M NaAc (pH 5.0), 40mM NaCl and ImM DTT. The reservoir solution contained 16% PEG 4000, 0.1M MES (pH 6.0), 0.4M LiCl, 10% glycerol and 5% t- butanol. The crystals were transferred in a few steps to an artificial mother liquor containing 30% PEG. Data collection: Diffraction data were collected at cryo- temperature on an R-Axis imaging plate system mounted on a Rigaku rotating anode generator, on Fuji imaging plates at the X4A beamline and on a MAR detector at the X25 beamline at the Brookhaven National Laboratory. Diffraction images were processed with DENZO.¹⁵ The crystals belonged to space group PA_\2_\2 but exhibited significant differences in cell parameters (called 'crystal forms' here). There is a dimer ofthe protease in the asymmetric unit, with significant variations in its orientation and position in the different crystal forms. MAD phasing: Initial phase information was obtained with the MADSYS package¹³ based on the MAD data (at 3 A resolution) for the L229M mutant collected at the X4A beamline (crystal form A,

c=215.4A). The Se positions were determined from F& Patterson maps with the program PATSOL.¹⁶ The phase information was used to solve the native crystal forms B (σ=68.5A, c=211.5A) and C (12.1k, 215.3 A) by molecular replacement with the REPLACE suite.¹⁷ Subsequent averaging among the three crystal forms, with a locally-written program (L. T., unpublished), produced an electron density map with recognizable secondary structure elements. The Se and Hg sites, identified from a difference electron density map for a K2Hg(SCN)₄ soak, helped in the development of a preliminary trace and an initial model was built with the program FRODO.¹⁸ It was discovered that replacement ofthe LiCl component in the artificial mother liquor with 0.15M Na2SO4 produced a dramatic improvement (from 3 A to 2A resolution) in the diffraction quality ofthe crystals. A MAD data set to 2.0 A resolution was collected on such a crystal of the T181M/L229M mutant at the X25 beamline. The phasing statistics at 2.5A resolution based on the data, with either the MADSYS package or X- Plor,¹⁹ were rather poor. However, after two-fold averaging at 2.5 A resolution, an excellent electron density map was obtained showing amino acid side chains and the main chain carbonyl groups. The trace developed earlier was found to be essentially correct. Only about 190 residues could be located in the averaged electron density map. Residues 35-45 (helix αA) were found after partial model phase combination with the SIGMAA program.²⁰ Structure Refinement: The structure refinement was carried out with the program X-Plor,¹⁹ for reflections between 6.0 and 2.0A resolution with F>2σ. NCS restraints were used at the beginning stages of the refinement. Residues 26-34 in one monomer and some other missing residues were located in the 2F₀-F_C electron density maps. The current structure model contains 419 residues of the dimer, one sulphate ion and 249 water molecules. The R factor is 22.1% for 30822 reflections between 6.0 and 2.0A resolution (90% complete). The free R factor,²¹ for 7.5% of the reflections, is 28.8%. The rms deviation in bond lengths is 0.010A and that in bond angles is 1.8°. The atomic coordinates have been deposited at the Protein Data Bank.

During the structure determination by the seleno-methionyl MAD technique, several major hurdles had to be tackled and overcome. These included treatment of crystals to reduce mosaicity and to remove microscopic twinning, site-specific mutagenesis to introduce extra methionine residues, non-crystallographic symmetry (NCS) averaging among three crystals to improve the electron density map, and change of artificial mother liquor that improved the diffraction quality of the crystals from 3.0A to 2.0A resolution.

Treatment of Crystals

The protease samples used in the current study all contain the A143Q mutation (to eliminate one of the internal cleavage sites of HCMV protease)⁴. Light scattering studies showed that the protease was mono-disperse in solution and existed as dimers. This observation was made before the appearances of published reports showing the presence of the dimer and suggesting that the dimer is the active form of the enzyme^9,10. Initial crystallization conditions were found with the sparse matrix sampling technique¹⁴ using a commercial kit (Hampton Research). A total of 6 out of 96 conditions gave small crystals. One of these conditions was optimized to produce larger crystals. The reservoir solution contained 0.1M MES (pH 6.0), 16% PEG4000, 0.4M LiCl, 10% glycerol, and 5% /-butanol. The protein was at about 25mg/ml concentration in a solution containing 20mM NaAc (pH 5.0), 80mM NaCl and ImM DTT. The crystals, grown at room temperature, were long tetragonal prisms, the largest of which measured 0.15x0.15x2.0 mm³.

It became clear when these crystals were exposed to X-rays that they were highly mosaic and microscopically twinned. Some of the reflections had an elongated shape or even appeared as two closely-spaced spots in the diffraction image. As expected from the crystal moφhology, the diffraction pattern suggested that the crystal might be tetragonal. However, the a and the b axes of the unit cell were given different lengths by auto-indexing due to the twinning problem. Dehydrating these crystals might reduce the mosaicity and help solve the twinning problem. The PEG concentration in the reservoir was raised in a few steps from 16% to 30%, over 3 to 5 days. The crystals were then transferred to an artificial mother liquor containing 0.1M MES (pH 6.0), 30% PEG4000, 0.4M LiCl, 10% glycerol, and 5% r-butanol. Crystals that were treated in this manner were found to have lower mosaicity. In addition, the diffraction pattern from the treated crystals no longer showed any signs of twinning.

Initial X-ray diffraction data were collected on an R-Axis imaging plate system mounted on a Rigaku RU-200 rotating anode X-ray generator. The crystals diffracted weakly and anisotropically, with the best diffraction generally extending to about 3.θA resolution along c* and about 3.3A along a* (and equivalently b*). The treatment with higher PEG concentrations did not remove the anisotropy in the X-ray diffraction of these crystals. Depending on the treatment protocol, the crystals displayed large variations in cell dimensions, with unit cell volume changes of up to 10% among crystals. The length of the unit cell a axis varied between 68 and 73A, and c axis between 209 and 215 A (Table 1). Reflection data from four different types of crystals were used in this structure determination. These different crystals are loosely called "crystal forms" here (A, B, C, and D; see Table 1). The space group is either P4_\2_\2 or P432)2, with a dimer ofthe protease in the asymmetric unit. The V_m value for a dimer in the asymmetric unit is about 2.4A³/Dalton. For crystal forms A and C, pseudo-extinction in the diffraction pattern for h+k=2n+\ reflections (pseudo C-centering) suggested that the two-fold axis of the NCS dimer might be aligned with the crystal a (or equivalently b) direction. This was also seen in the native Patterson map, which contained a large peak at (0.5, 0.5, 0) (not shown).

Mutagenesis to Introduce Extra Methionine Residues

The search for heavy-atom derivatives for these crystals was hampered by the lack of isomoφhism among crystals. The seleno-methionyl MAD technique represented an alternative for this structure solution¹³. HCMV protease contains three Met residues. However, residue Met-1 is absent in about 50% ofthe protein molecules, as shown by electrospray mass spectrometry, and residue Met-3 is likely to be flexible. This leaves only one Met residue, Met-75, for the 256 residues of the protease. This was not expected to produce enough signals for the MAD experiment. Hence it was necessary to introduce extra methionine residues into the protease.

The aligned amino acid sequences of heφesvirus proteases were examined to find residues in HCMV protease that could be mutated to Met without significantly affecting its activity. Two residues, Leu-222 and Leu-229, were selected for mutagenesis. The basic consideration in this selection was that a hydrophobic residue in HCMV protease could be mutated to Met if it corresponded to a Met residue in other heφesvirus proteases. A hydrophobic residue was chosen so that the resulting Met side chain could be buried and thus ordered. The L229M mutant had about 90% of the catalytic activity of the native protease. The L222M/L229M double mutant, however, was found to have only 10% of the activity. Therefore, the L229M mutant was selected for structural work. To obtain larger signals from the MAD experiments, a second position for mutation was later found at residue 181. The T181M/L229M double mutant had about 60% ofthe catalytic activity.

Seleno-methionyl MAD Phasing with the L229M Mutant

The L229M mutant was used for seleno-methionyl MAD phasing first. The seleno- methionyl protease was produced in a Met-auxotropic E. coli cell line and purified using the same protocol as the native protease. Quantitative incoφoration of Se-Met residues into the protease was confirmed by electrospray mass spectrometry. Crystals of the seleno-methionyl protease were obtained under similar conditions as those for the native protease. Sodium bisulfite (5 mM) was added to the reservoir as an anti¬ oxidant to protect the seleno-methionyl residues.

The X-ray diffraction data were collected at beamline X4A at the Brookhaven National Laboratory. The crystal was flash-frozen, enabling diffraction data at four wavelengths (λl :0.9919A, λ2:0.9793A, λ3:0.9792A, λ4:0.9724A) to be collected (crystal form A, Table 1). The crystal was aligned before the start of data collection. A total of 34 images (2.2° per image, with 0.7° overlap between images) were collected around the c* axis, starting from the a*c* zone. Then, 16 images (1.7° per image, with 0.7° overlap between images) were collected around the b* axis to cover the missing cone. The crystal diffracted to about 3.0A resolution along c* and about 3.3A along a*. The crystal-to-detector distance was 300mm. The diffraction pattern was recorded on Fuji image plates which were scanned with 0.1mm raster step. The diffraction images were processed with DENZO and scaled with SCALEPACK¹⁵. The partial observations were discarded and the Friedel observations were treated separately during scaling.

The individual observations from the four wavelengths were loaded into the MADSYS package¹³ (Table 2). A Patterson map was calculated based on the FA values and inteφretted with the PATSOL program¹⁶. Two strong sites were found in the Patterson map. They were related by a relative translation of (0.5, 0.5, 0), in agreement with the diffraction pattern which showed pseudo C-centering. The Se sites were also solved by direct methods, with the program SHELX²⁴. This revealed two additional minor sites, also related by a relative translation of (0.5, 0.5, 0). One of these minor sites was present in the FA Patterson inteφretation as well. MAD phasing was carried out using the four sites, in both space group P4\2\2 and P4^2_\2. The resulting electron density map for P4\2\2 had clearer solvent boundaries and was pursued further.

The electron density map was of very poor quality, with no recognizable secondary structure elements. Initial attempts at improving the map by solvent-flattening²⁵ were not successful. In the phase combination step, the phases after solvent-flattening were restrained to the original MAD phases and the overall phase shift by solvent-flattening was about 60°. However, as the MAD phases were of very poor quality, it might be better not to restrain the phases in this case. Therefore, the solvent-flattening was carried out without any phase restraints. The overall phase shift was about 70°. The electron density map improved recognizably. A few α-helices could be seen in the map.

NCS Averaging Among Three Crystals

Although some secondary structure elements could be recognized in the electron density map after solvent-flattening, the map could not be traced as the connectivity between these elements was poor. Moreover, not all the secondary structure elements were visible in the map. NCS averaging was then attempted to improve the phases. As the seleno-methionyl crystal exhibited pseudo C-centering, the NCS two-fold was assumed to lie along the b axis of the unit cell. A translation function¹⁶ based on the overlap of NCS-related electron density values clearly showed the position of this two-fold axis in the unit cell. The parameters for the orientation and position of this two-fold axis were then determined more accurately, again by electron density overlap^26,27. Two-fold NCS averaging within this crystal, using a locally-written program, produced only minor improvements in the electron density map.

As mentioned above, various data sets on HCMV protease crystals had been collected on the R-Axis, representing different unit cell parameters and volumes (Table 1). Performing NCS averaging among these different "crystal forms" could produce greater improvement in the electron density. For this puφose, two additional crystal forms (B and C in Table 1) were chosen. The electron density for the dimer in the seleno-methionyl crystal (crystal form A in Table 1) was used as the model to solve these two crystal forms by molecular replacement, with the Replace program package¹⁶. After solvent-flattening, the two-fold axes ofthe dimers in the three crystal forms were correlated with each other by electron density overlap. In agreement with the large differences in unit cell parameters, the orientation and position of the dimers showed significant differences among the crystal forms as well (Table 1). This was originally recognized by the large scaling R factors among the three data sets, which varied between 51 and 58%. The NCS averaging across the three crystal forms was carried out using a locally- written program. The correlation coefficients between observed and calculated structure factors for the three crystal forms increased from 0.75 to 0.85. The average phase change for each crystal form was about 60°. The resulting electron density map could be tentatively traced manually. The two Se positions in each monomer were used for the tracing. The ordered site was found to correspond to the L229M mutation position, whereas the mostly disordered site was found to be Met-75. Several data sets (in crystal form B) collected earlier for crystals soaked in mercury compounds (K₂Hg(SCN)₄, K2Hgl4 and others) were analyzed by difference electron density maps. The best data set showed the presence of 6 weak Hg sites, obeying the NCS two-fold axis. The three Hg sites in each monomer were also used for the tracing. They were found to be associated with Cys-84, Cys-87 and Cys-161, respectively. Finally, the biochemical knowledge that Ser-132 and His-63 are probably close in space was also used in the tracing. An atomic model was built based on this trace with the FRODO program¹⁸.

At this time, the T181M/L229M double mutant was produced and the diffraction data on the seleno-methionyl protein were first collected on the R-Axis. A difference electron density map between this mutant and the native protein showed three sites for each monomer, two of which corresponded to Met-75 and L229M, respectively. The third site was within 3 A of the Thr-181 position in this initial atomic model, giving further support that the trace developed was probably correct.

Improved X-ray Diffraction by a Change of Artificial Mother Liquor

The activity of heφesvirus proteases is enhanced by the presence of high concentrations of salt, such as 0.5- IM sodium sulfate^9"12. The artificial mother liquor for the crystals contained only 0.4M LiCl, and hence was in the low salt condition. In order to determine whether there are any conf ormational changes in HCMV protease in the high salt condition, the LiCl component in the artificial mother liquor was replaced with sodium sulfate. However, only 0.15M Na2SO₄ could be used as higher concentrations were not soluble. Suφrisingly, crystals of HCMV protease that were soaked in this new artificial mother liquor for two days exhibited a significant improvement in the diffraction quality. Diffraction beyond 2.5A was recognized on the R-Axis, and the anisotropic pattern in the diffraction was also absent in crystals treated with sodium sulfate.

Seleno-methionyl MAD data were collected on the T181M/L229M double mutant on the X25 beamline at the Brookhaven National Laboratory (crystal form D in Table 1). The crystal diffracted to 2.0A resolution and the diffraction images were recorded on a Mar imaging plate system. The crystal was not aligned prior to data collection and a total of 25 images (with another 25 images at the inverse-beam position) were collected at each of the four wavelengths. The oscillation range per image was 2.0° and the crystal-to-detector distance was 270mm. The diffraction images were processed with DENZO and scaled with SCALEPACK. The partial observations were used. The data processing was done in two ways. First, data to 2.5A resolution were processed for all four wavelengths, using an overload value of 45000 in DENZO. These data were used in the MAD phasing. Second, data to 2.0A resolution were processed for diffraction images collected at λl, using the default overload value of 105000 in DENZO. This data set was used in subsequent phase improvement and structure refinement.

The phasing statistics to 2.5A resolution from the MADSYS package¹³ were rather poor (Table 3). It was difficult to locate the Se positions from the FA Patterson map. The electron density for the dimer after NCS averaging among the three crystal forms was used as a model to solve the structure of this new crystal by molecular replacement. The resulting phases were used in an anomalous difference electron density map for the data set at λ3, which revealed the 6 Se positions. MAD phasing was also carried out with X-PLOR 4.0¹⁹, with λ4 as the reference data set. The resulting electron density map, from either MADSYS or X-PLOR, was not inteφretable. The average difference between phases from MADSYS and X-PLOR was 60°.

To improve the phases, two-fold averaging was carried out within this crystal form, starting from either the MADSYS or the X-PLOR phases. This led to a phase change of about 65° for each phase set. The average difference between the MADSYS and the X-PLOR phases was reduced to 40° by this averaging. The two electron density maps after the averaging were similar and of very good quality. The average main chain electron density value was raised from about lσ to about 3σ by this averaging. The side chains of amino acid residues and the main chain carbonyl oxygens for many residues could be clearly seen in the electron density map. The initial atomic model was found to be essentially correct and could be easily placed into the new electron density map.

However, there was electron density for only about 190 amino acids out of the 256 total in HCMV protease. The atomic model was used in a partial model phase combination, with the SIGMAA program²⁰, and a new helix was seen in the electron density. This helix, for residues 34-45, in the two monomers showed significant deviation from the two-fold symmetry of the dimer. This may explain its weak electron density in the averaged map. The current atomic model, after structure refinement at 2.0A resolution, is still missing about 40 residues of each monomer of the protease. These residues are in regions of very low sequence homology among the heφesvirus proteases and may be disordered in structure²⁸.

Discussion

The packing of HCMV protease molecules in these crystals is rather loose. The apparent V_m value of 2.4 A³/Dalton seems normal. However, if the disordered residues are excluded from the calculation, the V_m value becomes 2.9 A³/Dalton. The higher mosaicity ofthe original crystals was probably due to this loose packing of the protease molecules. By increasing the PEG concentration in the mother liquor and dehydrating the crystals, a tighter packing arrangement was achieved. This process seemed to have also successfully removed a microscopic twinning problem with these crystals. The introduction of 0.15M sodium sulfate led to a dramatic improvement in the diffraction quality of these crystals. Two possible reasons could be identified based on this crystal structure. First, a sulfate ion was found at the interface between two molecules (not from the same dimer) in the crystal. This could lead to a tighter packing of the protease molecules. Second, sodium sulfate caused a small change in the organization of the HCMV protease dimer in these crystals²⁸. This might also have an impact on the crystal packing, as evidenced by the fact that crystal form D has the shortest c axis length. Other possible effects of sodium sulfate on the structure of HCMV protease (for example, conformational rigidification of the monomer, stabilization of the dimer) may also play a role in this observed improvement in the diffraction. Further experiments showed that a sodium sulfate concentration of 25mM was enough to produce the diffraction enhancement in these crystals.

In crystal forms A and C, the dimer two-fold axis is close to being aligned with the crystal b (or equivalently a) direction. This led to a pseudo C-centered cell. If the dimer two-fold axis were perfectly aligned with b and centered at (0.5, 0.5, 0.125), the unit cell would be perfectly C-centered and a smaller unit cell, with 50A axes along a and b, could be chosen. In other words, the dimer along the a+b direction of the smaller cell deviated from the crystallographic symmetry and became a non- crystallographic dimer in the current unit cell. The center of this dimer were found to be located near (0.5, 0.5, 0.125) or (0.5, 0, 0.125) (Table 1). These two positions are generally not equivalent in the larger unit cell. However, in the smaller unit cell, these two positions are related by an alternative origin translation of (0.5, 0.5, 0) and would therefore be equivalent. The deviation of the dimer two-fold axis would lead to the differentiation between the two positions. For crystals that do not show pseudo C- centering, it would be expected that the deviation should be large enough such that it would be easy to distinguish between the two possible centers. This was found to be the case. If the dimer electron density was used as a model to solve crystal forms B and D by molecular replacement, only one of the centers appeared as the solution from the translation function. In contrast, for crystal forms A and C, both centers appeared as possible solutions, with similar peak heights.

Seleno-methionyl MAD phasing represents a useful alternative to protein structure determination¹³, especially for cases, like this one, where the search for heavy atom derivatives proves difficult. The success ofthe current structure determination did rely on the introduction of extra methionine residues into the protein. The method that was used here for selecting potential sites for mutagenesis into methionines should be appl icabl e to other proteins . NCS averaging is a powerful tool for phase improvement for cases with high local symmetry²⁹. With only a two-fold NCS, the power of averaging is generally weaker. Introduction of additional crystal forms into the averaging process could help the phase improvement. In this case, crystals belonging to the same space group but 5 having significantly different unit cell parameters were utilized as additional crystal forms. The large scaling R factors among the three data sets suggested that they probably represented sufficiently different sampling of the molecular transform to be treated as separate crystal forms. A recognizable improvement in the electron density map was achieved through this averaging, enabling a tentative trace ofthe polypeptide

10 backbone to be established. Greater improvement in the electron density was probably hampered by conformational differences among the crystals and, perhaps more importantly in this case, by the poor quality of the X-ray diffraction data used in this averaging. In comparison, two-fold averaging within crystal form D, using a much better diffraction data set, successfully produced an excellent electron density map at

15 2.5A resolution.

CITED DOCUMENTS

1. Fields, B. N., Knipe, D. M. et al. Eds.Virology, vol. 2, chapters 64-73, (1990). Raven Press, New York. 20 2. Gibson, W., Welch. A. R. & Hall, M. R. T. Perspectives in Drug Discovery and Design, 2, 413-426, (1995).

3. Rawlings, N. D. & Barrett, A. J. Methods in Enzym. 244, 19-61, (1994).

4. Pinko, C, Margosiak, S. A., Vandeφool, D., Gutowski, J. C, Condon, B. & Kan, C.-C. J. Biol Chem. 270, 23634-23640, (1995).

25 5. Ollis, D. L., et al. Protein Eng. 5, 197-211, (1992).

6. Perona, J. J. & Craik, C. S. Protein Science, 4, 337-360, (1995).

7. Murzin, A. G., Brenner, S. E., Hubbard, T. & Chothia, C. J. Mol. Biol. 247, 536- 540, (1995).

8. Yamamoto, A., et al. Biochem. 31, 11305-11309, (1992).

" 30 9. Darke, P. L., Cole, J. L., Waxman, L., Hall, D. L., Sardana, M. K. & Kuo, L. C. J. Biol. Chem. Ill, 7445-7449, (1996). 10. Margosiak, S. A., Vandeφool, D. L., Sisson, W., Pinko, C. & Kan, C.-C. Biochem. 35, 5300-5307, (1996).

1 1. Burck, P. J., et al. J. Virol. 68, 2937-2946, (1994).

12. Hall, D. L. & Darke, P. L. J. Biol. Chem. 270, 22697-22700, (1995). 13. Hendrickson, W. A. Science, 254, 51-58, (1991).

14. Jancarik, J. & Kim, S.-H. J. Appl. Cryst. 24, 409-411, (1991).

15. Otwinowski, Z. In Data Collection and Processing (Sawyer, L., Isaacs, N. & Bailey, S., eds.), pp 56-62, (1993). SERC Daresbury Laboratory, England.

16. Tong, L. & Rossmann, M. G. J. Appl. Cryst. 26, 15-21, (1993). 17. Tong, L. J. Appl. Cryst. 26, 748-751 , (1993).

18. Jones, T. A. J. Appl. Cryst. 11, 268-272, (1978).

19. Brunger, A. T. The X-PLOR manual, version 3.0. Yale University, New Haven, CT, (1992).

20. Read, R. J. Acta Cryst. A42, 140-149, (1986). 21. Brunger, A. T. (1992). Nature, 355, 472-475.

22. Carson, M. J. Mol. Graphics, 5, 103-106, (1987).

23. NichoIls, A., Shaφ, K. A. & Honig, B. Proteins, 11, 281-296, (1991).24. Sheldrick, G. M. (1990). Acta Cryst. A46, 467-473. 24. Sheldrick, G. M. (1990). Acta Cryst. A46, 467-473. 25. Wang, B.-C. (1985). Method Enzymol. 115, 90-112.

26. Tong, L., Choi, H.-K., Minor, W. & Rossmann, M. G. (1992). Acta Cryst. A48, 430-442.

27. Tong, L. & Rossmann, M. G. (1997). Method Enzymol. (eds. Carter, Jr. C. W. & Sweet, R. W.), 276, 594-611. Academic Press, Orlando, Florida. 28. Tong, L., Qian, C, Massariol, M.-J., Bonneau, P. R., Cordingley, M. G. & Lagace, L. (1996). Nature, 383, 272-275. 29. Rossmann, M. G. (1990). Acta Cryst. A46, 73-82, (1990). V

Summary of Crystallographic Information

Crystal Protein Unit Cell Reso¬ No. of No. of "merge Dimer Orientation Dimer Center Form Sample (a, c) (A) lution (A) Obs. Refl. (%) (Φ. ψ)^a (*, y, z)

A A143Q 70.1, 215.4 3.0 313575^b 10770 7.7 90.5, 88.5 (0.512, 0.504, 0.125) L229M Se-Met

ro B 68.5, 211.5 3.0 44026 9432 9.3 ro A143Q 101.6, 85.9 (0.453, 0.034, 0.116)

C A143Q 72.7, 215.3 3.0 157723^c 11714 9.9 84.1, 90.5 (0.508, 0.008. 0.126)

D A143Q 71.0, 209.3 2.0 165253 34045 7.1 99.5, 82.6 (0.447, 0.042, 0.116) L181M L229M Se-Mct

a. φ is me ϊasured from i the unit cell a axis, and ψ is measured from the c axis. b. Reflec tion data fro m the four wav elengths we re merged. c Reflection data from two crystals were merged.

B TALE ¹ Summary of MAD Phasing Statistics for the L229M Mutant

Observed Ratios λl λ2 λ3 λ4 f (_e) f' (e) (15-4A)

λl 0.039 0.045 0.041 0.036 -3.6 1.1

(0.029)

λ2 0 0..005511 0 0..003355 0 0..004444 - -1111..00 3.6

(0.033) ro λ3 0 0..006655 0 0..003388 - -99..11 5.7 (0.031) m to λ4 0 0..005533 - -44..00 4.0 (0.031)

For reflections between 15 and 3A resolution, Λ(FA) = 40%, Δ(Δφ) = 58°.

Observed ratios are Bijvoet difference ratios (diagonal elements) at each wavelength (values in parentheses are for centric reflections) and dispersive difference ratios (off -diagonal elements) between pairs of wavelengths. P(FA) is the residual between the calculated structure factors based on the Se positions and those obtained from the MAD analysis. Δ(Δφ) is the average difference between independent determinations of Δφ from the MAD analysis.

Summary of MAD Phasing Statistics for the T181M/L229M Mutant A. From MADSYS

Observed Rados λl λ2 λ3 λ4 f (_e) f (e) (15-3.5A)

λl 0.069 0.054 0.050 0.052 -3.6 1.1

(0.055)

λ2 00..008800 00..004411 00..005566 -11.2 4.2

(0.058) ro λ3 00..009900 00..005500 -8.8 4.8 g 2 (0.058) m

λ4 00..008811 -4.0 4.1 (0.060)

For reflections between 15 and 2.5A resolution, Λ(FA) = 45%, Δ(Δφ) = 61'

B. From X-PLOR

Resolution (A) 5.0 4.0 3.5 3.2 2.9 2.8 2.6 2.5 Overall

<Figure-of-merit> 0.80 0.73 0.69 0.65 0.60 0.57 0.54 0.53 0.64

While we have described a number of embodiments of this invention, it is apparent that our basic constructions may be altered to provide other embodiments which utilize the products and processes of this invention. Therefore, it will be appreciated that the scope of this invention is to be defined by the appended claims, rather than by the specific embodiments which have been presented by way of example.

Claims

WE CLAIM:

1. A HCMV protease defined by the atomic coordinates in Figure 5(A)-(T), wherein the protease is further characterized by its ability to cleave the HCMV precursor assembly protein to produce a functional assembly protein.

2. The HCMV protease according to claim 1 , wherein the protease is the full- length UL80 gene product containing A143Q, T181M and L229M point mutations.

3. A method for identifying an HCMV protease inhibitor, comprising the step of evaluating the ability of a test compound to inhibit the HCMV protease according to claim 1 or 2.

4. The method according to claim 3, wherein the evaluation is carried out using computer-assisted rational drug design.

5. The method according to claim 4, wherein the computer assisted rational drug design is carried out using DOCK or LUDI.

6. A method for determining the structure of an HCMV protease inhibitor complex, comprising the step of comparing the atomic coordinates ofthe HCMV protease inhibitor complex to the HCMV protease atomic coordinates in Figure 6(A)- (T).

7. The method according to claim 6, wherein the atomic coordinates ofthe HCMV protease inhibitor complex are refined using X-PLOR to yield a three dimensional structure.