Mechanistic Insights Into Oxidosqualene Cyclizations Through Homology Modeling

Mechanistic Insights into Oxidosqualene Cyclizations through Homology Modeling

TANJA SCHULZ–GASCH, MARTIN STAHL Molecular Design, Pharmaceutical Division, F. Hoffmann–La Roche AG, Pharmaceuticals Division, Discovery Technologies, Bldg. 092/2.10D, CH-4070 Basel, Switzerland

Received 27 February 2001; Accepted 28 May 2002

Abstract: 2,3-Oxidosqualene cyclases (OSC) are key enzymes in sterol biosynthesis. They catalyze the stereoselective cyclization and skeletal rearrangement of (3S)-2,3-oxidosqualene to lanosterol in mammals and fungi and to cycloartenol in algae and higher plants. Sequence information and proposed mechanism of 2,3-oxidosqualene cyclases are closely related to those of squalene-hopene cyclases (SHC), which represent functional analogs of OSCs in bacteria. SHCs catalyze the cationic cyclization cascade converting the linear triterpene squalene to fused ring compounds called hopanoids. High stereoselectivity and precision of the skeletal rearrangements has aroused the interest of researchers for nearly half a century, and valuable data on studying mechanistic details in the complex enzyme-catalyzed cyclization cascade has been collected. Today, interest in cyclases is still unbroken, because OSCs became targets for the development of antifungal and hypocholesterolemic drugs. However, due to the large size and membrane-bound nature of OSCs, three-dimensional structural information is still not available, thus preventing a complete understanding of the atomic details of the catalytic mechanism. In this work, we discuss results gained from homology modeling of human OSC based on structural information of SHC from Alicyclobacillus acidocaldarius and propose a structural model of human OSC. The model is in accordance with previously performed experimental studies with mechanism-based suicide inhibitors and mutagenesis experiments with altered activity and product speciﬁcity. Structural insight should strongly stimulate structure-based design of antifungal or cholesterol-lowering drugs.

Key words: cation-␲ interaction; homology modeling; lanosterol synthase; 2,3-oxidosqualene; squalene-hopene cyclase

Introduction supposed to be required for initiation. Site-directed mutagenesis experiments identified a highly conserved aspartic acid as an Oxidosqualene cyclases (recommended name lanosterol synthase, essential residue for catalysis,9,10 which is thought to play the role E.C. 5.4.99.7) of mammals and fungi catalyze the highly stereo- of the proton donor for epoxide activation. During the subsequent selective cyclization of linear (3S)-2,3-oxidosqualene 1 to an in- complex cyclization process, starting from a prefolded chair– termediate protosterol cation 2 and subsequent skeletal rearrange- boat–chair conformation of 2,3-oxidosqualene 1, four CAC dou- ment to lanosterol 3 (Fig. 1).1–8 ble bonds are successively converted to COC single bonds, lead- Complexity, efficiency, and high stereoselectivity of cyclase ing to the tetracyclic protosterol cation 2 (Fig. 1). Quantum catalyzed reactions make them prime examples for studying mul- chemical investigations indicate that the replacement of a CC ␲ ␴ tiple enzyme functions and interesting tools in synthetic chemistry. -bond (65 kcal/mol bond energy) by a CC -bond (85 kcal/mol Ϫ Triterpene cyclization products represent precursors of all ho- bond energy) is highly exothermic by 20 kcal/mol. Calculations panoids and steroids, for example, cholesterol, glucocorticoids, also showed that nearly no activation energy for an intramolecular estrogens, androgens, and progesterones. Finally, cyclases are of process with prefolded conformation (Fig. 1) is required and that the addition of a tertiary cation to a tertiary sp2 carbon atom can high economical interest as targets in development of antifungal, be expected to occur as a barrierless collapse.11 Exothermic reac- hypocholesterolemic, and phytotoxic drugs. tions of conformationally flexible substrates such as 2,3-oxido- The enzyme-catalyzed cyclization of 2,3-oxidosqualene 1 is initiated through electrophilic (protic) epoxide activation. Because 2,3-oxidosqualene is stable in neutral media and in glacial acetic Correspondence to: T. Schulz–Gasch; e-mail: tanja.schulz-gasch@ acid at room temperature for about 1 day,9 a strong acid was roche.com

Figure 1. (Left) OSCs catalyze the conversion of 2,3-oxidosqualene 1 to lanosterol 3 via the protosterol cation 2. Initial substrate chair–boat–chair conformation of 2,3-oxidosqualene 1, putative conformations after A-, B-, and C-ring closure (2,3-oxidosqualene cation numbering), and skeletal rearrangement of protosterol cation 2 (lanosterol cation numbering) through 1,2-shifts of hydride and methyl groups are shown. (Right) SHCs catalyze the conversion of squalene 4 to hopene 5 and diplopterol 6. Initial all-chair conformation of squalene 4 and putative conformations after A-, B-, C-, D-, and E-ring closure (squalene cation numbering) are shown. Mechanistic Insights into Oxidosqualene Cyclizations through Homology Modeling 743 squalene 1, including reactive cationic intermediates, are difficult sequence against the PDB27 retrieved only structural information on to control in solution and often lead to a broad variety of products. A. acidocaldarius SHC and verifies the template selection. In contrast, the cyclase-catalyzed reactions show enormously high Pairwise sequence alignment of human OSC and A. acidocal- structural and stereochemical control. darius SHC shows 41% similarity and rather low identity of 26% High product specificity in cyclases is believed to be achieved (Fig. 2). Thus, protein modeling can be used to generate a crude through several factors: (1) by enforcing substrates to occupy overall skeleton of human OSC and interpretation of entire model prefolded conformations, (2) by progression of reaction via rigidly properties is speculative. Wendt et al.16 identified 34 residues held, partially cyclized carbocationic intermediates, and (3) by (His451 as second sphere residue excluded) participating in active stabilization of intermediate carbocations by cation-␲ interac- site formation of A. acidocaldarius SHC. Exclusively comparing tions12–15 of frequently occurring aromatic residues in the active these active site residues leads to increased similarity of 50% and site, thus preventing early truncation of the cyclization cascade by an identity of 38%. Similarity is even higher (65% sequence deprotonation or nucleophilic addition of solvent molecules. similarity and 50% sequence identity) if only those residues are The enzyme’s role in subsequent skeletal rearrangement of the aligned that are close to a modeled hopene position28 in the active protosterol cation 2 through 1,2-hydride and -methyl group migration site of SHC, while residues forming the bottom (residues close to is considered to be secondary. The enzyme is thought to protect Trp196) and entrance tunnel are neglected (Fig. 3A). Obviously, intermediate cations from being quenched or the reaction being trun- sequence similarity decreases from the catalytic Asp at the top of cated by early deprotonation and promote migration of hydride and the active site to the bottom of the hydrophobic tunnel. Thus, a methyl substituents toward the lanosterol cation. This assumption is superposition of the SHC structure with the OSC model backbone supported by observations on nonenzymatic, Lewis -acid-catalyzed shows high consistency around the catalytic Asp (Fig. 3A). Based cyclization experiments of 2,3-oxidosqualene 1, which also lead to on high identity of active site residues, properties like size, shape, 2,3 products arising from methyl group migration. and physico-chemical properties of the human OSC active site can The terminal step of the enzyme catalysis process is the dep- be predicted with some confidence. This, in turn, allows verifica- rotonation of the lanosterol cation 2, which is again believed to 16 tion of the mechanistic hypothesis on a molecular level. Compar- take place under rigorous steric control of the enzyme. ison with experimental results on suicide inhibitors and mutagen- Although significant improvements have been made in the field of esis studies should further allow a refinement of the model membrane protein expression and crystallization,17,18 X-ray structural structure. information on OSC is still lacking. In cases where structure elucida- Homology modeling of human OSC is based on the crystal tion fails, model building of target proteins by exploiting homologies structure of A. acidocaldarius SHC. To examine the suitability of to proteins of known structure (templates) becomes important. A. acidocaldarius SHC as a template structure measuring sequence We present homology modeling studies on human OSC based similarity and identity, secondary structure elements of human on a crystal structure of Alicyclobacillus acidocaldarius SHC. OSC were predicted (independent from A. acidocaldarius SHC) Proposed reaction mechanism (Fig. 1) and sequences (Fig. 2) of using the Web application SSpro229,30 and compared to secondary SHCs are closely related to OSC and therefore imply structural structure elements of SHC. In accordance to A. acidocaldarius similarity of the two cyclase families. The generated three-dimen- SHC, the main predicted secondary structure element in human sional structural information on human OSC is compared to the ␣ n Ͼ widely accepted cyclization mechanism via a prefolded chair– OSC are -helices. Helices of human OSC ( 4) match similar boat–chair conformation of 2,3-oxidosqualene 1 and differences to regions like helices of A. acidocaldarius SHC in the sequence ␣ ␣ the SHC catalyzed cyclization of squalene 4 are identified on a alignment of the two cyclases (Fig. 2, helices 1to 22 as 31 ␤ molecular level. Consistency of the model structure with available numbered by Wendt et al. ). Furthermore, five short -strands are Յ experimental data on mechanism-based suicide inhibitors10,19–24 assigned in the SHC structure (n 5). For all of them close ␤ Յ and site-directed mutagenesis studies25,26 is clearly demonstrated. -strands (n 5) in the human OSC sequence are predicted by The three-dimensional structural information on human OSC pro- PSIPRED (Fig. 2). Good agreement in distribution of secondary vides new insight into the cyclization and deprotonation of 2,3- structure elements gives evidence for similar folding of the two oxidosqualene 1. cyclases and further proves that A. acidocaldarius SHC is an The information gained from the model should help researchers appropriate template for homology modeling of human OSC. to upgrade their knowledge of the nature of OSC-mediated con- To verify the stereochemical quality of the homology model version at the atomic level, for example, size and shape of active structure of human OSC, geometrical parameters were checked site, specific hydrophobic and polar interactions responsible for and a 3D profile of the protein was generated. stereochemical control, and should also set the stage for structure- MOE’s protein report function indicates no insufficiencies in based design of potent OSC inhibitors. angles and bond lengths for the minimized model structure of human OSC. Ramachandran et al.32 showed, that due to confor- mational restrictions, a plot of the peptides ␾ and ␺ angles in Results and Discussion protein structures could be used as a quality measure. Generally, three sets of ␾, ␺ angle ranges are defined (CORE, ALLOWED, GENEROUS) according to statistical occurrence in the PDB.33 Sequence Alignment, Secondary Structure Prediction, and Figure 4 shows a Ramachandran plot of the human OSC model Homology Modeling structure. Proline and glycine residues were excluded because of A search with the molecular modeling package MOE (settings: 20% their atypical dihedral angle distributions. As it can be seen from identity, Z-score 5, see Methods for details) with the human OSC Figure 4, there are eight residues (1.2% of overall human OSC 744 Schulz–Gasch and Stahl • Vol. 24, No. 6 • Journal of Computational Chemistry

Figure 2. Sequence alignment of Homo sapiens OSC (Hs), Arabidopsis thaliana cycloartenol synthase (At), Saccharomyces cerevisiae OSC (Sc), and Alicyclobacillus acidocaldarius SHC (Aa). Predicted secondary structure elements of human OSC and secondary structure elements of A. acidocaldarius SHC are included. Red rectangles indicate ␣-helices, yellow rectangles ␤-sheets. Numbering of ␣-helix and ␤-sheet of A. acidocaldarius SHC corresponds to Wendt et al.31 sequence) not belonging to either the CORE, ALLOWED, or darius SHC. Thus, disallowed ␾, ␺ conformations result from GENEROUS region. Compared to dihedral angle distributions of inaccuracies in the construction algorithms used in homology 121,879 residues from 463 structures extracted from the PDB modeling rather than from bad template structure selection. (Table 1), where 1.3% of all residues adopt disallowed conforma- 3D proﬁles (Fig. 5) of OSC model and SHC crystal structure tions,33 this is an acceptable overall number of residues to be show positive average numbers over all residues, indicating side outliers in the model structure. Furthermore, all of the observed chains to be in a preferable environment. Average 3D–1D scores eight outliers are solvent-exposed residues and do not participate per residue for the OSC model and SHC crystal structure are 37.9 in formation of the active site. All outliers are situated at inserts or and 54.0, respectively. The higher score for SHC, however, indi- gaps in the sequence alignment of human OSC and A. acidocal- cates a better overall packing of residues for the crystal structure. Figure 3. (A) Superposition of ribbons from A. acidocaldarius SHC crystal structure (bold gray) and modeled human OSC (thin red) structure. Atoms of 2,3-oxidosqualene after A-ring closure with calculated Conolly surface are included to show the location of the hydrophobic tunnel. Atoms of the catalytic Asp455 at the top of the cavity and of Trp196 at the bottom of the cavity (human OSC numbering) are also included. All sequence inserts and gaps with n Ն 2 are located at solvent-exposed positions. High consistency in superposition of backbones can be observed around the catalytic Asp455. (B) Atoms of the catalytic Asp455 and 2,3-oxidosqualene (C6 cation) in prefolded chair–boat–chair conformation. The methyl group at C10 is indicated by a larger ball. The empty sphere indicates the C10 methyl group position for hypothetical all-chair conformation of the substrate. Superposition of aligned ribbon fragments I and II show reversed steric properties of the two cyclases enforcing 2,3-oxidosqualene in OSC catalysis in a chair–boat–chair conformation. (C) Superposition of SHC (gray) and OSC (red) ribbon fragments stabilizing cations after D-ring formation. In A. acidocaldarius SHC Phe601 and Phe605 stabilize the tertiary cation at C19 after Markovnikov D-ring closure of squalene 4 and promote ring expansion to form six-membered D-ring including a secondary cation at C18. Aromatic side chains are in an appropriate position to stabilize a cation at C18. In human OSC, an appropriately positioned ␲-density of aromatic side chains is lacking and rearrangement of the shown protosterol cation C20 (2 in Fig. 1) to a six-membered D-ring is not observed. 746 Schulz–Gasch and Stahl • Vol. 24, No. 6 • Journal of Computational Chemistry

Cys456 and Thr502 is supported by sequence alignment of OSCs from various species9 (S. cerevisiae, C. albicans, S. pombe, H. sapiens, and R. norvegicus), because all show conserved Cys456 and Thr502 residues (S. pombe shows conservative substitutions). Acidity of the catalytic Asp might also be increased by hydrogen bonding to ordered water molecules, like it was observed for A. acidocaldarius SHC.28 The crystal structure shows a small cavity close to the catalytic Asp, which is separated from solvent- exposed surface by hydrogen bonding between Arg623 and Glu454. The same motif can be observed in the human OSC model structure. Furthermore, anchimeric assistance to COO cleavage by proximate and optimally oriented ␲-nucleophilic double bonds and stabilization of intermediate cations by a high ␲-electron density or aromatic residues in the active site, like it is observed for a manually modeled prefolded chair–boat–chair conformation of 2,3-oxidosqualene 1 in the human OSC model, can contribute to increased acidity of the catalytic Asp. Thus, the model on human OSC explains how acidity of catalytic Asp455 is increased and activation barrier of epoxide protonation is reduced to a level, allowing an enzyme aspartic acid to initiate the cyclization cascade Figure 4. Ramachandran plot of ␾, ␺ angle distribution in the homol- by protonation. ␾ ␺ ogy model structure of human OSC. Gray dots indicate , angles Experimental34,36–41 and theoretical11,34 investigations dem- within the CORE, GENEROUS, and ALLOWED region, black trian- onstrated a rapid catalytic cyclization to be promoted not only by gles represent OUTLIERS. an optimally placed, highly acidic Asp residue but also by an optimally prefolded conformation of the substrate. Based on experimental and theoretical studies it is widely accepted that initi- Mechanism ation by protonation and A-ring formation represents a concerted process. Molecular level differences between SHCs and OSCs mediate Prefolded conformations of the first ring (chair) and first cat- differences in the reaction mechanisms (Fig. 1). These differences, ionic intermediates (tertiary cation) of closed A-rings are identical which eventually lead to the formation of the different products for substrates of both cyclases (Fig. 1). Positive charges are located lanosterol 3, hopene 5, and diplopterol 6, are discussed in the above the molecular plain of squalene and 2,3-oxidosqualene 1. In following by comparing the interactions of reaction intermediates accordance with this high consistency in mechanisms, 8 out of 10 and transition structures with the enzyme. Major reaction steps like residues participating in formation of the active site close to the protonation, individual cyclization steps, and deprotonation are A-ring are conserved. Only two of the residues are substituted: discussed separately, although there exists evidence for concerted SHC Asp374 is changed to human OSC Val453 and SHC Asp377 processes,9,11,34 for example, protonation and A-ring closure. to human OSC Cys456. As has been mentioned previously, For olefin protonation in SHC catalysis, a strong Brønsted acid Cys456 is supposed to be one of the catalytic Asp activating is required. Based on crystallographic data28 it was proposed that residues in human OSC. Conserved aromatic residues Phe444, acidity of catalytic Asp376 in its protonated state is increased by Tyr503, Trp581, and Tyr704 (OSC numbering) stabilize the first hydrogen bonding to protonated His451 (hydrogen bond distance tertiary cation at C6. Because there are no gaps or inserts in the 2.8 Å) and additionally through an ordered water molecule. The sequence alignment close to those residues (Fig. 2), side chains in acidity of the pair Asp376:His451 seems to be sufficient to pro- the model structure of human OSC show similar positions and tonate squalene. In contrast, for OSC an adequate activating resi- orientations like those in the SHC crystal structure. due at His451 position is missing and His451 (SHC) is aligned The first mechanistic differences in squalene and 2,3-oxido- with Ala536 (P48449 numbering for human OSC throughout the squalene 1 cyclization occur in B-ring formations. Squalene 4 must text). Because epoxides are more readily protonated than olefins, Wendt et al.16 suggested an isolated aspartic acid in OSC may ␾ ␺ suffice for initiation. However, experiments showed that 2,3-oxi- Table 1. Statistical Occurrence of , Angles in the PDB ␾ ␺ dosqualene is completely stable in neutral media and in glacial Compared to the , Angles Distribution in the Homology Modeling acetic acid at room temperature for about 1 day.9 Structure of Human OSC. Different effects might contribute to increase acidity of the catalytic Asp455 or reduce the activation barrier for epoxide av. PDB OSC Model Region (%) (%) protonation, for example, proximate hydrogen bond partners35 close to the catalytic Asp. The catalytic Asp455 in the human OSC Core 81.9 73.2 model is in hydrogen bond distance to Cys456, Thr502, and Allowed 14.8 22.6 Cys533. Single-point mutation of Cys533 (Cys540Ala of S. cer- Generous 2.0 3.0 evisiae OSC), however, proved this residue not to be essential for Outside 1.3 1.2 catalytic activity.10 Nevertheless, a catalytic role of residues Mechanistic Insights into Oxidosqualene Cyclizations through Homology Modeling 747

Figure 5. XSAE generated 3D proﬁles of the homology modeling structure of human OSC (A) and the crystal structure of A. acidocaldarius SHC (PDB entry 2SQC). adopt a chair-like prefolded conformation of the B-ring chain and the methyl group at C10 in the prefolded all-chair conformation of a tertiary cation after B-ring closure at C10, which is homologous squalene 4 (Fig. 3B). The four residues Gly598–Phe601 form the to the previous cation at C6. The methyl groups at C6 and C10 are opposing part of the active site below the molecular plain of located above the molecular plain of squalene 4 (Fig. 1). In squalene 4 in SHC. contrast, 2,3-oxidosqualene 1 reacts from an energetically unfa- Obviously, a positively charged Lys is present above the mo- vored boat conformation. Its tertiary cation at C10 and the corre- lecular plain of 2,3-oxidosqualene 1 in the OSC active site. Thus, sponding methyl group are below the molecular plain of 2,3- cation location above the molecular plain of 2,3-oxidosqualene oxidosqualene 1 (Fig. 1). would lead to unfavorable electrostatic interactions with the pos- The homology model of human OSC allows an analysis of itively charged Lys331 side chain. C10 cation location below the steric and electronic effects that lead to the observed mechanistic molecular plain, however, can be well stabilized by ␲-stacking differences. Two fragments in the alignment of SHC and OSC with side chain of residue Trp581 in human OSC. It is also obvious have to be considered in detail: that there is a reversion in steric properties between the two cyclases. Human OSC has an inserted amino acid above and a 260 600 deleted residue below the molecular plain of 2,3-oxidosqualene, ¦¦ A.a. SHC: GG-I · · · GTGF thus preventing the methyl group at C10 to be located above the H.s. OSC: TKSI · · · GV-F molecular plain and avoiding the substrate to take the energetically ¦¦ favored chair conformation. Conversely, some additional space for 330 694 the methyl group at C10 is offered below the molecular plain through a single gap in the OSC sequence. Residues Gly259, Gly260, and Ile261 of A. acidocaldarius The importance of forcing the B-ring in a prefolded boat SHC participate in the formation of a loop in the active site above conformation and avoiding proximity to Lys331 can be seen from the molecular plain of squalene and are close to the C10 cation and mutagenesis studies performed with SHC.42 A single point mutation Leu607Lys introduces a Lys in the hydrophobic tunnel of A. acidocaldarius SHC. The position of the Lys607 side chain is close to prefolded the B- and C-ring of squalene. This introduction of a positively charged residue in A. acidocaldarius SHC and proximity of the positive charge to intermediate cations truncates the cyclization cascade after B-ring formation and the cation is deprotonated to lead to the bicyclic product ␥-polypodatetraene 7 (Fig. 6). To summarize this step, the model structure of human OSC gives detailed insight into electronic and steric pressure of the Figure 6. A. acidocaldarius SHC single mutant Leu607Lys leads to enzyme to enforce speciﬁc, energetically unfavored (boat vs. formation of ␥-polypodatetraene 7 and Phe605Lys to 17-isodammara- chair) substrate conformations to avoid truncation of cyclization 20(21),24-diene 8. by early deprotonation. 748 Schulz–Gasch and Stahl • Vol. 24, No. 6 • Journal of Computational Chemistry

For C-ring formation, fragments of both OSC and SHC sub- terol cation 2 terminates the cyclization cascade in OSC cycliza- strates must adopt a prefolded chair conformation (Fig. 1). The tion. secondary cations at C14 are generated in an anti-Markovnikov This observation is strongly supported by mutagenesis studies sense. Cyclization catalysis of substrate analogues provided evi- of A. acidocaldarius SHC.42 The single point mutation Phe605Lys dence that C-ring formation occurs through five-membered ring in A. acidocaldarius SHC truncates the cyclization cascade after closure (Markovnikov) followed by subsequent ring expan- formation of a five-membered D-ring and the tertiary cation is sion.43–45 Further support for the rearrangement is provided deprotonated to give 17-isodammara-20(21), 24-diene 8 (Fig. 6). through computational studies.11 Thus, the absence of aromatic side chain of Phe605 in SHC, as As in the formation of the B-ring, cation location is the major observed in the model OSC (Fig. 3C), prevents direct anti-Mark- difference between OSC and SHC mechanisms during this step. In ovnikov six-membered D-ring formation or ring-expansion from SHC, the cation at C15 is positioned above the molecular plain of five-membered D-ring to six-membered D-ring. Both pathways squalene, whereas in OSC the cation at C15 is below the molecular would include a secondary cation at C18 and stabilizing effects of plain of the substrate. Because Lys331 in the model structure of ␲-stacking through Phe605 are lacking in this SHC mutant. human OSC is situated directly above the two subsequent cationic Skeletal rearrangement through 1,2-shifts of hydride and centers at C10 and C15 (Fig. 3B), there is again an electrostatic methyl substituents (Fig. 1) convert the protosterol cation 2 to force to place the cation below the molecular plain of the substrate lanosterol C8 or C9 cation. Intervention of OSC during rearrange- and to avoid proximity to Lys331. Steric pressure through the ment is supposed to be minor since even the nonenzymatic Lewis enzyme, however, will only play a secondary role for the prefolded acid-catalyzed cyclization of 2,3-oxidosqualene affords at least C-ring conformation in both enzymes, because it is the energeti- one product arising from methyl group migration.2 Homology cally favorable chair conformation that is required for C-ring modeling, however, indicates that OSC does indeed play a role in formation. the skeletal rearrangement. Figure 7 shows high ␲-electron density SHC cyclization proceeds after C-ring closure with formation around intermediate cations at C6 and C10, which are formed after of a six-membered D-ring. Chemical studies strongly suggest that A- and B-ring closure. Conversely, ␲-electron density around D-ring cyclization in SHC proceeds through a five-membered cations at C15 after C- and protosterol cation 2 after D-ring D-ring closure to afford a Markovnikov cation followed by ring formation is strongly reduced. Therefore, the positive charge mi- expansion to yield a six-membered D-ring.46 The six-membered grates during the rearrangement towards high ␲-electron density D-ring includes a secondary cation at C18. For OSC the cycliza- from protosterol cation 2 to lanosterol cation at C8 and C9 within tion cascade ends after formation of the five-membered D-ring to the active site of the modeled OSC (Fig. 7). form the tertiary protosterol cation 2 (Fig. 1). Rearrangement to a Thus, the enzyme’s role in skeletal rearrangement is to shift six-membered ring including a secondary cation is not observed equilibrium between the protosterol cation 2 toward carbocations for OSC. at the C8 and C9 position of lanosterol cation. For the structural interpretation of this reaction step one has to After skeletal rearrangement to C8 and C9 lanosterol cation the consider a fragment of the sequence alignment and resulting ef- cyclization is terminated by proton removal. This deprotonation fects on the model structure of human OSC (Fig. 3C): has to be under rigorous enzymatic control to avoid product scattering on the final reaction step. OSCs catalyze the highly 600 selective deprotonation to form lanosterol 3, which is the thermo- ¦ dynamically most stable deprotonation product (Saytzeff product) A.a. SHC: GFPGDF with a tetrasubstituted double bond. Other possibilities to depro- human OSC: ϪFNKSC tonate the C8 and C9 lanosterol cations could lead to formation of ¦ kinetic products like 9␤–⌬7-lanosterol 9 and parkeol 10 (Fig. 8), 700 which include trisubstituted double bonds, and also to formation of cycloartenol 11 containing a cyclopropyl ring. Site-directed mu- The two residues Phe601 and Phe605 in A. acidocaldarius SHC tagenesis studies47 showed that Glu45, which is located at the participate in formation of the active site below the molecular plain bottom of the hydrophobic tunnel in A. acidocaldarius SHC par- of squalene and are close to the secondary cation at C18 in the ticipates in the terminal deprotonation step. In human OSC, six-membered D-ring. This cation at C18 is accommodated exactly Gly107 occupies the SHC Glu45 position and no other basic between the two aromatic Phe side chains, which are in optimal residue is located in the bottom of human OSC to terminate position for efficient ␲-cation stabilization, which enable interme- catalysis by deprotonation of protosterol cation 2. The model diate formation of a secondary cation. structure of human OSC shows His232 to be the closest basic The above sequence alignment shows Phe605 to be substituted residue and to be well positioned to serve as a catalytic base for by Cys700 in human OSC, while Phe601 is conserved (Phe696). deprotonation of lanosterol cation (Fig. 7). His232 is highly con- But, due to a gap in human OSC in front of Phe696, the orientation served within the OSC family.10 of backbone and side chain atoms of the conserved Phe residues in To summarize the deprotonation step, SHC has a basic residue SHC and OSC are different (Fig. 3C). Phe696 is no longer in an (Glu45) at the bottom of the hydrophobic tunnel in an appropriate appropriate position to stabilize a cation at C18. Thus, the absence distance to deprotonate the C23 cation after E-ring closure without of efficient ␲-cation stabilization by aromatic side chains, caused affording skeletal rearrangement. In contrast, human OSC shows by substitution of one Phe and a different side chain position for no comparably positioned basic residue to allow direct C20 cation the second Phe, prevent OSC to form a secondary cation at C18 deprotonation of the protosterol cation 2, and therefore, has to (anti-Markovnikov). Therefore, formation of the tertiary protos- support cation location at C8 and C9 of the lanosterol cation rather Mechanistic Insights into Oxidosqualene Cyclizations through Homology Modeling 749

Figure 7. Orientation of the lanosterol cation (orange) and side chain of catalytic Asp455 (ball-and-stick representation, orange) in the human OSC model structure are shown. Aromatic side chains (stick representation) participating in the active site formation of modeled human OSC structure are included. Migration of positive charge from protosterol cation 2 to lanosterol C9 cation is indicated by a dashed arrow along the substrate skeleton. ␲-Electron density of aromatic side chains decreases from catalytic Asp455 to Trp196 at the bottom of the hydrophobic tunnel. The migration of positive charge through 1,2-shifts of hydride and methyl substituents occurs toward higher ␲-electron density of aromatic side chains in the active site, thus shifting the equilibrium toward lanosterol C8 and C9 cation. His232 is essential for deprotonation and proximate to the lanosterol C8 and C9 cation.

than at C20 in the protosterol cation 2. C8 and C9 lanosterol cation mutant and to 0% cycloartenol and 75% lanosterol by the double positions are proximate to His232, a residue located in the middle mutant Tyr410Thr Ile481Val.26 9␤–⌬7–Lanosterol 9 and parkeol of the hydrophobic tunnel. 10 are formed in different amounts as side products by all three mutated enzymes. Lanosterol, cycloartenol, 9␤–⌬7–Lanosterol Correlation of the OSC Model Structure to Experimental and parkeol all represent deprotonation products of lanosterol C8 Studies to Verify the Quality of the Model Structure and C9 cations, thus indicating similar overall mechanisms except the ﬁnal deprotonation steps. Homology modeling studies provide OSCs catalyze the stereoselective cyclization and skeletal rear- interpretation of these diverse deprotonation paths. For human rangement of (3S)-2,3-oxidosqualene to lanosterol 3 in mammals OSC, Val453 and Thr380 occupy the corresponding mutated po- and fungi. Their counterpart in algae and higher plants is cyclo- sitions in the cycloartenol synthase and are responsible for the artenol synthase (EC 5.4.99.8), which catalyzes conversion to cycloartenol 11 (Fig. 8). Cyclization mechanism and skeletal re- observed deprotonation to the thermodynamic product (tetra-sub- arrangement is identical for the two enzymes. Differences occur stituted double-bond) lanosterol 3. Obviously, increasing the size 3 3 only in the ﬁnal deprotonation step, which lead to the formation of of one or both of the two residues Val Ile and Thr Tyr leads different products lanosterol 3 and cycloartenol 11 (Fig. 8). As it to formation of the kinetic products, cycloartenol 11 (includes ␤ ⌬ can be expected from high consistency in mechanisms, global cyclopropyl fragment), 9 – 7–lanosterol 9 and parkeol 10 (three- sequence alignment of A. thaliana cycloartenol synthase and hu- fold substituted double bonds), because corresponding positions man OSC shows high similarity 55.2% and an identity of 45.8% for deprotonation are more easily accessible. (Fig. 2). According to the homology models, only three active-site res- Product distribution of A. thaliana cycloartenol synthase can be idues in human OSC are different from those in A. thaliana altered from 99% cycloartenol 11 by the wild-type enzyme, to 56% cycloartenol synthase (Arg361Lys, Tyr410Thr, and Ile481Val). cycloartenol and 24% lanosterol by the Ile481Val single mutant, High similarity and identity of active sites is to be expected 0% cycloartenol and 65% lanosterol by the Tyr410Thr single because of high mechanistic consistency. As mentioned above, the 750 Schulz–Gasch and Stahl • Vol. 24, No. 6 • Journal of Computational Chemistry

Figure 9. The single-point mutations Val454Ala and Val454Gly in Arabidopsis thaliana cycloartenol synthase cause signiﬁcant production of the monocyclic triterpenes achilleol 12 and camelliol 13.

methyl substituents above the molecular plain of 2,3-oxidosqualene. Thus, the A-ring of 2,3-oxidosqualene is held rigidly and in optimal position for subsequent boat conformation of the B-ring. Optimal formation of the boat conformation is essential because Figure 8. OSC catalyze the stereoselective cyclization and skeletal otherwise a proximate and optimally oriented ␲-nucleophilic dou- rearrangement of (3S)-2,3-oxidosqualene to lanosterol 3 in mammals ble bond for COC bond formation is lacking, and reaction rates and fungi. Their counterpart in algae and higher plants are cycloartenol could be slowed down. As a result, early truncation of cyclization synthases (EC 5.4.99.8), which catalyze the conversion to cycloartenol by deprotonation of the C6 cation through close residues like 11. Mutagenesis of two active site residues of cycloartenol synthase His232 could occur. Because Leu and Val show similar steric and changes product specificity and leads also to formation of lanosterol 3, electronic properties as Val, mutation of wild-type Val to Leu or ␤ ⌬ 9 - 7-lanosterol 9, and parkeol 10. Ile respectively show invariant production of lanosterol. Replace- ment of wild-type Val by a Phe residue, however, leads to inacti- vation of the cyclase. This is understandable from the spatial prefolded chair conformation of the B-ring in human OSC is arrangement shown in Figure 11. The phenyl side chain would be enforced through sterical and electronical (Lys331) effects. As a situated in front of catalytic Asp and occupy part of the space proof of concept of this former hypothesis, A. thaliana cycloarte- where the A-ring is located. Thus, initiation of cyclization cascade nol synthase, which also requires the B-ring in an energetically would be inhibited by preventing the substrate to get close to unfavorable chair conformation uses the same motifs—an inserted catalytic Asp and being protonated. amino acid above the molecular plain and a deletion below the Studies on mechanism-based suicide inhibitors also provide molecular plain of the substrate compared to the SHC structure. results capable to verify the human OSC model structure. Mech- Thus, the methyl group at C10 is sterically enforced to be situated below the molecular plain. Furthermore, like observed in the human OSC model, a positive charge is introduced in the mainly hydrophobic tunnel of the cycloartenol synthase model (Lys331 in human OSC and Arg361 in A. thaliana cycloartenol synthase), thus causing the carbocation at C10 to be situated below the molecular plain of 2,3-oxidosqualene 1 and preferring a boat-like conformation of the substrate. Further mutagenesis studies have been performed with OSC from S. cerevisiae. The single mutants Val454Ala and Val454Gly of S. cerevisiae OSC cause significant production of the monocyclic triterpenes achilleol 12 and camelliol 13 (Fig. 9).25 The fact that Val454 is conserved in OSCs of various species9 (Saccharo- myces cerevisiae, Candida albicans, Schizosaccharomyces pombe, Figure 10. The specific function of the Val453 side chain in OSCs (H. Homo sapiens, and Rattus norvegicus), already indicates a highly sapiens numbering) is the rigorous control of substrate orientation and specific role of this active site residue. Thus, discussing experi- prefolded conformation in cyclases. The dot-dash line represents the mental results concerning Val454 gained for S. cerevisiae OSC on crude borders of the hydrophobic tunnel. The black structure of 2,3- basis of structural information of human OSC seems reasonable. oxidosqualene represents a perfectly oriented conformation. The gray Val454 of S. cerevisiae OSC corresponds to Val453 in human structures of 2,3-oxidosqualene however, represent conformations OSC. Homology modeling gives insight into position and orien- which are not optimally oriented in the hydrophobic tunnel. His232 is tation of Val453 and allows detailed interpretation of specific the residues responsible for the final deprotonation step. Reduction of function of this residue. As illustrated in Figures 10 and 11, Val453 to smaller residues like Ala and Gly leads to early deprotona- Val453 is responsible for rigorous fixation and correct substrate tion products 12 and 13, and to a disturbed orientation of A-rings, orientation in the hydrophobic tunnel. The side chain atoms of which prevent substrates to occupy an optimal prefolded conformation Val453 show van der Waals interactions with C4 and the two and position. Mechanistic Insights into Oxidosqualene Cyclizations through Homology Modeling 751

suggested cyclization and rearrangement mechanism of 2,3-oxidosqualene 1 and to previously performed experimental mutagenesis and mechanism-based suicide inhibitor studies, including studies performed on A. acidocaldarius SHC and on A. thaliana cycloartenol synthase. Furthermore, homology modeling of human OSC allowed detailed insight on molecular level properties of the binding site and provided new aspects to contradictory opinions. Protic epoxide activation through catalytic Asp455 was shown to be supported by close hydrogen bond partners and activation through ordered water molecules seems to be feasible. The activation barrier is further lowered through stabilization of intermediate cations by a high ␲-electron density or aromatic residues in the active site. Residues causing steric and electrostatic pressure of the enzyme to force the B-ring in a prefolded boat conformation were identified. Molecular level differences resulting in truncation of cyclization after five-membered D-ring formation for OSC were detected and correlate well with mutagenesis experiments performed on A. acidocaldarius SHC. The role of the enzyme during skeletal rearrangement, which in the past was assumed to be Figure 11. Orientation of Val453 in human OSC and A-ring in chair secondary, is obviously to push the cation position to higher conformation of 2,3-oxidosqualene 1. The calculated Conolly surface ␲-electron density of aromatic side chains in the active site of is included to demonstrate steric fixation of the A-ring through the side human OSC. Thus, the final lanosterol cation 2 gets in an appro- chain of Val453. priate position to be deprotonated by His232 and lead to selective formation of lanosterol 3. Mutagenesis studies of those active site residues, which were identified to be responsible for specific tasks anism-based suicide inhibitors are capable of covalently labeling in differentiating between squalene and 2,3-oxidosqualene cycliza- the enzyme. After digestion and sequencing, labeled residues can tion, should strongly be stimulated by now available structural be identified as active site residues, and their position in the active information on human OSC. Knowledge of size, shape, and site relative to catalytic Asp at the top of the hydrophobic tunnel physico-chemical properties of the binding site of human OSC can be estimated. gained by homology modeling studies offers a new way to apply For checking the correct placement of residues in the model efficient structure-based drug-design methods on human OSC, structure of human OSC, experimental results from suicide inhib- aiming to develop more potent antifungal, hypocholesterolemic, itors (Fig. 12) were compared to the model OSC structure.10,46 and phytotoxic drugs with high economical interest. Tritiated compounds 14, 15, and 16 produce reactive mono- and Furthermore, the present article provides detailed experiences bicyclic cations and label Tyr510 of S. cerevisiae OSC. Tyr510 gained in homology model building, quality estimation, and re- corresponds to Tyr503 in human OSC, and is in the model struc- finement of model structure and also examination and inclusion of ture located close to the catalytic Asp (Fig. 8). For successive previous experimental work to validate model structures. Model occurring cations in the cyclization cascade the distance to the catalytic Asp, the initiation point of the catalysis, increases. There- fore, early occurring mono- and bicyclic cations of suicide inhibitors should label residues close to catalytic Asp. This is indeed observed for 14, 15, and 16 labeling human OSC corresponding active-site residue Tyr510. Cyclization of 20-oxaoxidosqualene 17 and diene 18 (Fig. 12) are expected to produce reactive cations in the vicinity of the D-ring and therefore label residue further away from the catalytic Asp than, for example, from the previously described labeling of Tyr510. Compounds 17 and 18 label His234 in S. cerevisiae OSC (His232 human OSC). In accordance, His232 in the model OSC structure is further away from catalytic Asp than Tyr510. The assumption that later occurring carbocations during catalyzed cyclization label active-site residues more distant from the catalysis initiation point catalytic Asp is fulfilled by the modeled OSC structure.

Conclusion and Outlook

We provided structural information on human OSC gained from Figure 12. Suicide inhibitors that label OSCs at different residues in homology modeling studies, which is in good accordance with the the active site. 752 Schulz–Gasch and Stahl • Vol. 24, No. 6 • Journal of Computational Chemistry building will play an important role in the era of exploding al.33 have shown that these quantities are significantly correlated to amounts of sequence information to support drug discovery re- the resolution at which a protein structure was solved. search, where detailed structural information on binding sites is The homology models of human OSC and cycloartenol syn- required. thase were verified for correct ␾ vs. ␺ angles of amino acid residues.33 A 3D profile was generated using the Roche in-house tool XSAE. A residue-wise list of 1D3D profile scores is generated. A large positive number indicates that a residue is in its Materials and Methods preferred environment in terms of solvent exposure and occlusion by polar and hydrophobic atoms. In good structures the average Hard- and Software over a window of 21 residues should not become negative.53 However, some residues of model structures showed slightly neg- Molecular Modeling studies were performed on a Silicon Graphics ative average numbers, mainly due to buried side chains of argi- Octane R12000 Workstations. MOE 2001.01 (Chemical Comput- nine and lysin at the protein surface. Manual reorientation of side ing Group Inc., Montreal, Quebec, Canada) was used for sequence chains to solvent-exposed surface led to positive numbers through- alignment and homology modeling. The quality of the structure out the model. was checked by a 3D profile generated with XSAE version 1.5 (Clemens Broger, F. Hoffman–La Roche AG, Switzerland). Man- ual modeling of substrates was carried out with the modeling Minimization package Moloc (Gerber Molecular Design, Basel, Switzerland). Resulting model structures were refined by energy minimization Final structures were minimized with Amber 6 (Scripps, UCSF). using the AMBER6.0 all-atom force field. Minimizations were carried out for 20 iterations of simplex minimization followed by 54 Experimental Data Selection 1500 steps of conjugate gradient minimization.

Sequence information on human OSC,48 S. cerevisiae OSC,49 Manual Modeling of Substrates cycloartenol synthase from A. thaliana50 and A. acidocaldarius SHC51 were taken from the Swiss-Prot database (H. sapiens OSC: C8 and C9 lanosterol cations, protosterol cation, and lanosterol P48449, S. cerevisiae OSC: P38604, cycloartenol synthase from A. were manually modeled in the active site of homology models of thaliana: P38605, A. acidocaldarius SHC: P33247). The template human OSC and cycloartenol synthase from A. thaliana using the structure for homology modeling, the crystal structure of SHC modeling package Moloc. from A. acidocaldarius (PDB entry 2SQC, resolution 2.00 Å),28 was obtained from the Protein Data Bank (PDB).27 Mutations D376C and C435S were reverted to wild-type residues. All water Acknowledgment and inhibitor molecules, as well as the B-chain were removed from the structure. T. Schulz-Gasch and M. Stahl thank D. Bur for introduction on OSC modeling and all colleagues at Roche Basel for stimulating Sequence Alignment discussions and a supportive research atmosphere.

Sequence alignment was performed using the MOE sequence alignment module applying the BLOSUM 40 substitution ma- References trix.52 The alignment was checked for correct alignment of motifs and catalytic residues. Erroneously aligned residues were manually 1. Corey, E. J.; Russey, W. E.; Ortiz de Montellano, P. R. J Am Chem constrained within the MOE interface.31 Soc 1966, 88, 4750. 2. van Tamelen, E. E.; Willett, J. D.; Clayton, R. B.; Lord, K. E. J Am Chem Soc 1966, 88, 4752. Homology Modeling 3. van Tamelen, E. E.; Willet, J.; Schwartz, M.; Nadeau, R. J Am Chem Soc 1966, 88, 5937. Homology modeling was performed using the homology modeling 4. van Tamelen, E. E. J Am Chem Soc 1982, 104, 6479. module of MOE. P48449 and P38605 were used as model se- 5. Yamamoto, S.; Lin, K.; Bloch, K. Proc Natl Acad Sci USA 1982, 63, quences. As a primary template, the refined PDB entry 2SQC was 110. selected. MOE was set up to generate ten energy minimized 6. New, W. D. Recent Adv Phytochem 1990, 24, 283. intermediate models. Different homology models are the result of 7. Cattel, L.; Ceruti, M. Physiology and Biochemistry of Sterols; Patter- permutational selection of different loop candidates and side-chain son, G. W.; New, W. E.; Eds.; American Oil Chemists’ Society: rotamers. As a final model the best intermediate model according Champaign, IL, 1992; chap. 3. to MOE’s packing evaluation function was chosen and subsequent 8. Abe, I.; Rohmer, M.; Prestwich, G. D. Chem Rev 1993, 93, 2189. 9. Corey, E. J.; Cheng, H.; Baker, C. H.; Matsuda, S. P. T.; Li, D.; Song, “fine” energy minimization (MOE default settings, RMS gradient X. J Am Chem Soc 1997, 119, 1277. 0.005) of this model was performed. 10. Corey, E. J.; Cheng, H.; Baker, C. H.; Matsuda, S. P. T.; Li, D.; Song, A protein report concerning stereochemical quality of protein X. J Am Chem Soc 1997, 119, 1289. structure coordinates can be obtained from the MOE interface. The 11. Jenson, C.; Jorgensen, W. L. J Am Chem Soc 1997, 119, 10846. report includes observed and statistically expected values of the 12. Corey, E. J.; Matsuda, S. P. T.; Baker, C. H.; Ting, A. Y.; Cheng, H. mean and standard deviation of various measured angles. Morris et Biochem Biophys Res Commun 1996, 219, 327. Mechanistic Insights into Oxidosqualene Cyclizations through Homology Modeling 753

13. Zoltewicz, J. A.; Maier, N. M.; Fabian, W. M. F. J Org Chem 1998, 63, 33. Morris, A. L.; MacArthur, M.; Hutchinson, E. G.; Thornton, J. M. 4985. Proteins 1992, 12, 345. 14. Miklis, P. C.; Ditchﬁeld, R.; Spencer, T. A. J Am Chem Soc 1998, 120, 34. Gao, D.; Pan, Y.; Byun, K.; Gao, J. J Am Chem Soc 1998, 120, 4045. 10482. 35. Gandour, R. D. Bioorganic Chem 1981, 10, 169. 15. Gallivan, J. P.; Dougherty, D. A. Proc Natl Acad Sci USA 1999, 96, 36. Corey, E. J.; Lin, K.; Jautelat, M. J. J Am Chem Soc 1968, 90, 2724. 9459. 37. Crosby, L. O.; van Tamelen, E. E.; Clayton, R. B. J Chem Soc Chem 16. Wendt, K. U.; Schulz, G. E.; Corey, E. J.; Liu, D. R. Angew Chem Int Comm 1969, 532. Ed 2000, 39, 2812. 38. van Tamelen, E. E.; James, D. R. J Am Chem Soc 1977, 99, 950. 17. Ostermeier, C.; Iwata, S.; Ludwig, B.; Michel, H. Nat Struct Biol 39. Corey, E. J.; Staas, D. D. J Am Chem Soc 1998, 120, 3526. 1995, 2, 842. 40. Medina, J. C.; Kyler, K. S. J Am Chem Soc 1988, 110, 4818. 18. Pautsch, A.; Vogt, J.; Model, K.; Siebold, C.; Schulz, G. E. Proteins 41. Xiao, X. Y.; Sen, S. E.; Prestwich, G. D. Tetrahedron Lett 1990, 31, 1999, 34, 167. 2097. 19. Abe, I.; Prestwich, G. D. J Biol Chem 1994, 269, 802. 42. Schmitz, S.; Fu¨ll, C.; Glaser, T.; Albert, K.; Poralla, K. Tetrahedron 20. Abe, I.; Liu, W.; Oehlschlager, A. C.; Prestwich, G. D. J Am Chem Lett 2001, 42, 883. Soc 1996, 118, 9180. 43. Corey, E. J.; Virgil, S. C.; Sarshar, S. J Am Chem Soc 1991, 113, 21. Abe, I.; Zheng, Y. F.; Prestwich, G. D. Biochemistry 1998, 37, 5779. 8171. 22. Stach, D.; Zheng, Y. F.; Perez, A. L.; Oehlschlager, A. C.; Abe, I.; 44. Hoshino, T.; Sakai, Y. Chem Commun 1998, 1591. Prestwich, G. D.; Hartmann, P. G. J Med Chem 1997, 40, 201. 45. van Tamelen, E. E.; Sharpless, K. B.; Hanzlik, R.; Clayton, R. B.; 23. Cattel, L.; Ceruti, M. Crit Rev Biochem Mol Biol 1998, 33, 353. Burlingame, A. L.; Wszolek, P. C. J Am Chem Soc 1967, 89, 7150. 24. Zheng, Y. F.; Oehlschlager, A. C.; Georgopapadakou, N. H.; Hartman, 46. Abe, I.; Dang, T. Y.; Zheng, Y. F.; Madden, B. A.; Feil, C.; Poralla, P. G.; Scheliga, P. J Am Chem Soc 1995, 117, 670. K.; Prestwich, G. D. J Am Chem Soc 1997, 119, 11333. 25. Joubert, B. M.; Hua, L.; Matsuda, S. P. T. Org Lett 2000, 2, 339. 47. Dang, T.; Prestwich, G. D. Chem Biol 1999, 7, 643. 26. Herrera, J. B. R.; Wilson, W. K.; Matsuda, S. P. T. J Am Chem Soc 48. Baker, C. H.; Matsuda, S. P. T.; Liu, D. R.; Corey, E. J. Biochem 2000, 122, 6765. Biophys Res Commun 1995, 213, 154. 27. Bernstein, F. C.; Koetzle, T. F.; Williams, G. J. B.; Meyer, E. F. Jr.; 49. Corey, E. J.; Matsuda, S. P. T.; Bartel, B. Proc Natl Acad Sci USA Brice, M. D.; Rodgers, J. R.; Kennard, O.; Shimanouchi, T.; Tasumi, 1994, 91, 2211. M. J Mol Biol 1977, 112, 525. 50. Corey, E. J.; Matsuda, S. P. T.; Bartel, B. Proc Natl Acad Sci USA 28. Wendt, K. U.; Lenhart, A.; Schulz, G. E. J Mol Biol 1999, 286, 175. 1993, 90, 11628. 29. Altschul, S. F.; Madden, T. L.; Schaffer, A. A.; Zhang, J.; Zhang, Z.; 51. Ochs, D.; Kaletta, C.; Entian, K.-D.; Beck–Sickinger, A.; Poralla, K. J Miller, W.; Lipman, D. J. Nucleic Acids Res 1997, 25, 3389. Bacteriol 1992, 174, 298. 30. Baldi, P.; Brunak, S.; Frasconi, P.; Pollastri, G.; Soda, G. Bioinfor- 52. Henikoff, S.; Henikoff, J. G. Proteins 1993, 17, 49. matics 1999, 15, 937. 53. Lu¨thy, R.; Bowie, J. U.; Eisenberg, D. Nature 1993, 356, 83. 31. Wendt, K. U.; Poralla, K.; Schulz, G. E. Science 1997, 227, 1811. 54. Case, D. A.; Pearlman, D. A.; Caldwell, J. W.; Cheatham, T. E., III; 32. Ramachandran, G. N.; Ramakrishnan, C.; Sasisekharan, V. J Mol Biol DeBolt, S.; Ferguson, D.; Seibel, G.; Kollman, P. Comp Phys Com- 1963, 7, 95. mun 1995, 91, 1.