Journal Articles

Ligand Scaffold Replacement
using MOE Pharmacophore Tools

Alain Deschênes and Elizabeth Sourial
December 2007
Chemical Computing Group Inc.
 

Introduction

Scaffold hopping is an approach used to discover new chemical classes by replacing a portion (the scaffold) of a known compound, while preserving the remaining chemical groups, under the assumption that they are important for biological activity. For example, in peptidomimetic efforts, certain peptide sidechains are the groups that are to be preserved during a search for a new linker (the scaffold) that preserves the presentation of the sidechain groups with the receptor. In many cases, the scaffold is a ring system whose substituents are the groups that must be preserved:

In principle, 2D methods such as substructure or similarity search can be used, however, scaffold hopping generally is most successful when attempted with 3D methods.

CAVEAT [Lauri 1994] is one of the early programs developed for 3D scaffold replacement. The CAVEAT methodology is based on specifying two or more bonds linking the scaffold and R-groups in a 3D bioactive conformer of a lead compound (the query bonds). These bonds define 3D vectors in which the origin of the vector is an atom on the existing scaffold and the terminus of the vector is an atom of the R-group to be preserved. A 3D database of candidate scaffold conformations is searched for molecules that, if suitably substituted, would coincide with the substitution bonds of the defined vectors. The 3D scaffold database contains conformations of molecules as well as annotations that encode the location of potential substituents. A sample linker is shown in Figure 1.

Figure 1. A 3D linker conformation with potential exit vectors.

By overlaying the potential R-group bonds from the scaffold database molecules with the defined query bonds, the bond lengths, angles, and orientation of the original R-groups are preserved. If the bonds overlay sufficiently well, then a potential new scaffold will have been found. The advantages of the CAVEAT methodology are that a) the queries are easily specified (select two scaffold to R-group bonds called the link bonds) and b) the search can be performed rapidly (overlay vector segments in 3D).

Figure 2. Sample scaffold replacement query. The original ligand has two R-groups defined. A pair of (Link,Link2) features are used to preserve each bond. Searching a database of candidate linkers gives new potential replacement scaffolds.

The original CAVEAT databases consisted of hydrocarbon ring systems or small alkane substituents of small ring systems, although in principle, any database can be searched. Vendor catalogs can be preprocessed by removing “decorations” (e.g., with RECAP [Lewell 1998] rules) and retaining the unique fragments which are then subjected to conformational search. Alternatively, structural databases such as the CSD can be similarly processed to produce 3D fragments more directly.

A more recent method is called Recore [Maass 2007]. Like CAVEAT, Recore requires the definition of at least two exit vectors, but this approach allows additional pharmacophore-type features or constraints that a candidate scaffold must satisfy — for example, the existence of a hydrogen bond acceptor at a specific location in space. Recore rapidly searches a specially indexed 3D database although the principles are the same as those of CAVEAT which conducts a linear search. The addition of pharmacophore constraints or filters on the output, however, increases utility of the method. The combination of CAVEAT methodology with pharmacophore-type methodologies provides a key advantage over the more structurally sensitive 2D methods. ?Specifically, the important interactions provided by the scaffold can be preserved while retaining the required geometry of the attached R-groups. The resulting re-scaffolded lead is then more likely to be a viable new chemical direction for lead optimization.

In a structure-based context (where the starting bioactive conformation is often obtained) further constraints can be imposed. The receptor volume can be used as a shape guide to eliminate candidate scaffolds that would clash with the receptor. In this article, we present the MOE scaffold replacement method which combines the CAVEAT methodology with full pharmacophore discovery capabilities as well as the ability to include structure-based information, definition of chemistry rules through SMARTS patterns as well as efficient database searching. We believe that these enhancements are a significant advancement in 3D scaffold replacement techniques.

Methodology

The fundamental approach is to combine CAVEAT style functionality with the MOE pharmacophore tools. Specifically, we will define the substituent vectors with special pharmacophore annotations called Link annotations. Link-type annotations denote points of substitution on a candidate scaffold molecule as well as the locations of potential R-group substituents. For example, in Figure 1, the yellow spheres are the Link annotations. Some are placed on the heavy atoms of the molecule while others are projected away from the molecule. The heavy atom annotations are the possible substitution points and the projected annotations are placed at bonded distances and angles consistent with the hybridization of the heavy atom. The idea is that one Link-type annotation is placed at the origin of a CAVEAT vector and one at the vector terminus. During a pharmacophore search, a constraint is imposed that ensures that both heavy and projected annotations match simultaneously or not at all. In this way, a pair of Link-type annotations simulates the CAVEAT “exit vector”.

There are three types of link annotations:

  1. Link: annotates a scaffold heavy atom substitution point with at least one (implicit) hydrogen.
  2. Link2: annotates projected locations of potential sp2 R-groups.
  3. Link3: annotates projected locations of potential sp3 R-groups.

A typical query will consist of (Link,Link2) or (Link,Link3) pairs located at positions intended to represent scaffold / R-group bonds. The Link2 or Link3 feature is placed on an existing R-group heavy atom (~1.5 Å away from the corresponding scaffold atom) and the Link feature is intended to match a heavy atom of a new scaffold to be found in a database of scaffold or linker molecules (see Figure 1). A matching scaffold is then likely to present the existing R-group in the proper direction. Link2 query features are used when the R-group atom is sp2. Link3 query features are used when the R-group atom is sp3. For example, when attempting to match a piperazine scaffold with an aromatic R-group, a Link2 query feature should be used so that the piperazine nitrogen conjugated planar geometry is used (and not the nitrogen tetrahedral geometry).

Scaffold molecules are annotated by an automatic procedure. Only C, N, O, S, and P atoms with 1, 2, or 3 heavy neighbors and at least one (possibly implicit) hydrogen are candidates for Link features; other atoms are not considered as candidates. The following rules assume that the foregoing conditions are satisfied. Not all these atoms will be given Link annotations; substituents on freely rotatable single bonds will be avoided (e.g., sp2-sp3 rotors).

Q = any 1°, Xi = any Q = 1° {C,N+}, Y = 2° {O,S,N-} N = 2° or 3°, Q1 = 4-coord, Q2 = any
Y ≠ 1°, Q = {OH, SH, NH2, PH2} Q = {C,N,P,S}, Q = 3° sp3, Ri ≠ H

For primary sp3 centers with one substituent (e.g., 1,1,1-trichloroethane), there are three potential exit vectors in a tetrahedral geometry. Each point is annotated with Link2&Link3 since the local geometry is not affected by a substituent's hybridization. A similar rule is used for tertiary sp3 centers where there is only one exit vector to preserve the tetrahedral geometry.

Q = 2° aromatic Q = 2° sp2 Q = 1° sp2

sp2 centers have exit vectors to retain the trigonal planar geometry. For example, substitution on an aromatic nitrogen (e.g., pyridine) will happen at 120° from the ring atoms regardless whether the substituent connected atom is sp2 or sp3 hybridized. Other secondary and primary sp2 centers will be annotated with one and two exit vector, respectively, retaining the trigonal planar geometry.

N = 2° amide not in 3,4-ring N = 1° amide not in 3,4-ring N = 2° in NCN+ not in 3,4-ring
N = 1° in NCN+ not in 3,4-ring

Secondary sp3 centers have exit vectors to retain tetrahedral geometry. However, an extra exit vector is added if a bond is formed with an sp2 center, resulting in a trigonal geometry.

For primary and secondary amides, the geometry on the nitrogen must be trigonal planar because of the delocalization. For a primary amide, two exit vectors are defined, and for a secondary amide one exit vector is defined at 120° to retain the trigonal planar geometry. Similarly, a primary NCN+ will have two possible exit vectors while a secondary NCN+ group will have one retaining the trigonal planar geometry. These rules are independent of the hybridization of the connected substituent atom. Also, note that cis peptide formation is avoided.

O = 1° OH on Caro, Xi = 2°

Carboxylic acids have a single exit vector defined at 120° from the C-O-H center. Any hydroxyl group substituted on an aromatic ring (e.g., phenol) will have two exit vectors forming a 120° trigonal planar center ensuring coplanarity with the ring atoms.

 
   

sp centers only have one possible exit vector defined to be colinear with the triple bond.

 
Q = 2° {C,N} sp3, Xi ≠ H  

A secondary amine can be used to show the difference between the geometry of Link2 and Link3. As shown in the figure below, each nitrogen can adopt either a tetrahedral geometry (Link3) or extend the π plane of the connected substituent. In the case of piperazine, an exit vector for substitution of cyclohexane would require a Link3 feature, while an exit vector for substitution of benzene would require a Link2 feature. The Link2 feature would enforce a flat nitrogen extending the π plane of the benzene ring.

Figure 3: Differentiation between Link2 and Link3 annotations. Left) benzene substitution for Link2 annotations Right) cyclohexane substitution for Link3 in a tetrahedral geometry.

Note that in most cases, Link2&Link3 annotations are used for the projected annotations. For 2° nitrogen and carbon atoms, there is a distinction between Link2 and Link3. Link2 projections are in the potential π system plane and Link3 projections are in tetrahedral formation.

The general strategy for scaffold replacement is therefore:

  1. Obtain the structure of the active site and a template ligand.

  2. Identify substituent locations from expert knowledge of the system or with the help active site analysis tools such as ligand interaction diagrams, electrostatic maps, or contact statistics.

  3. Assign special pharmacophore features to preserve the bonds that link the scaffold with each substituent.

  4. Add excluded volumes to avoid clashes with the receptor and the substituents.

  5. Include any additional pharmacophore feature(s) from key interaction(s) present in the scaffold of the native ligand.

  6. Search a database of candidate linkers and select replacement scaffolds.

  7. Connect the ligand and the R-group(s) to form the new molecule.

Results and Discussion

The most reliable pharmacophore information can be obtained from high resolution crystal structures or high-quality docking results. Here, we will examine Factor Xa in complex with the M55532 ligand (PDB:1IOE). Factor Xa is a vitamin-K-dependent serine protease that is responsible for the generation of thrombin from prothrombin in the coagulation cascade. As a result, Factor Xa is an important target for the development of anticoagulant drugs [Davie 1991] [Kastenholz 2000]. Before creating special pharmacophore queries for scaffold replacement, one must first identify the R-groups that are to be preserved.


Figure 4. Ligand Interaction Diagram for Factor Xa (IIOE). The active site residues are represented as follows: polar residues in pink, hydrophobic residues in green, acidic residues with a red contour ring, basic residues with a blue contour ring. Green and blue arrows indicate hydrogen bonding to sidechain and backbone atoms respectively. A naphthyl icon represents a π-π stacking interaction, while a benzene with a + represents a cation-π interaction. Blue “clouds” on ligand atoms indicate the solvent exposed surface area of ligand atoms (darker and larger clouds means more solvent exposure). Light-blue “halos” around residues indicate the degree of interaction with ligand atoms (larger, darker halos means more interaction). The dotted contour reflects steric room for methyl substitution. The contour line is broken if it is closest to an atom which is fully exposed.

The ligand interaction diagram [Clark 2007] in Figure 4 shows important interactions between the M55532 ligand and Factor Xa. The 4-amino pyridine group (in the hydrophobic D pocket) forms backbone hydrogen bond interactions with Thr98 and Glu97 as well as cation-π interactions with Phe174 and Thr99. The 4-amino pyridine group also forms a π-π stacking interaction with Phe174, which accommodates hydrophobic and positively charged functional groups [Stubbs 1995].

At the entrance of the S1 pocket, the Gln192 and Gly216 residues have a strong hydrophobic interaction with the ligand (as indicated by the large, dark halo around the residue in Figure 4). A hydrogen bond is also formed between the lactam carbonyl oxygen of M55532 and backbone nitrogen of Gly218.

Figure 5. An Electrostatic Map of Factor Xa calculated from the receptor structure without the ligand. Negative preferences in red drawn at ~2 kcal/mol and neutral in green plotted at 3kcal/mol.

Figure 5 shows electrostatic isocontours in the active site of Factor Xa (1IOE) along with a re-entrant surface and pocket labels. The electrostatic isocontours are drawn by Electrostatic Maps which is an implementation of a non-linear Poisson-Boltzmann equation solver to the prediction of electrostatically preferred locations of hydrophobic, negative and positive regions in a receptor active site. The Factor Xa structure was prepared for electrostatic analysis by assigning standard ionization states, adding protons to satisfy valence requirements and calculating partial charges using the MMFF94 forcefield [Halgren 1996]. In Figure 5, the green contour shows the hydrophobic regions and the red contour shows the negative regions. The positive regions of the electrostatic maps was minimal and is not displayed for clarity.

Figure 6. The electrostatic isocontours of the Factor Xa receptor produced by Electrostatic Maps with the M55532 crystal ligand.

Figure 6 shows the electrostatic isocontours, produced by the Electrostatic Maps application, superimposed onto the crystal ligand (M55532). The receptor structure and surface are hidden to illustrate the correspondence between the prediction and the ligand features. The strong hydrophobic region overlays with the ligand 4-amino pyridine group in the D pocket. Notwithstanding the carboxylate near the S1 pocket (Asp189), there is a large hydrophobic region that overlays well with ligand napthyl fragment. There are no other strongly hydrophobic regions in the active site. In addition, the map shows a preference for a negative feature (red) which overlays with the carbonyl oxygen atom on the ligand.

The electrostatic analysis shows that the strongly interacting groups are the 4-amino pyridine group in the D pocket and the chloro-naphthyl group in the S1 pocket. The remainder of M55532 (in the P pocket) can be considered a “linker” and a good candidate for scaffold replacement.

Figure 7. The M55532 ligand with scaffold drawn with red bonds and R-groups drawn in black.

The bonds that link the (red) scaffold to the 4-amino pyridine and chloro-napthyl groups are marked with arrows. A new scaffold should have a conformation that would, upon substitution, preserve the orientation of the non-scaffold groups. In particular, the bonds marked with arrows should overlay closely with the corresponding bonds of a new substituted scaffold. A Link feature is placed on each scaffold atom that is bonded to a group to preserve, and a Link2 feature is placed on each substituent atom bonded to the existing scaffold. This creates a 4-point pharmacophore query.

The four point query (with 0.3 Å radii for the Link and Link2 features) was used to search a database containing potential linkers. The database contained the conformations of over 21,372 linkers and scaffold fragments, which were prepared from more than 40 commercial catalogs by removing R-groups and other “decorations” based on chemical patterns. The search returned 8,489 conformations from 1,553 distinct molecules and are shown overlayed with M55532 in Figure 8.

Figure 8. 1,553 candidate linkers overlayed with M55532 in the active site of Factor Xa, many of which penetrate the van der Waals surface of the receptor.

In the linker hits, both bond lengths and bond angles are satisfied for substitution of both the 4-amino pyridine group and the napthalene group. However, a large number of potential linkers have van der Waals clashes with the atoms of the receptor or the atoms of the R-groups. This highlights a key limitation of pure CAVEAT style pharmacophore queries in which geometric constraints alone are defined and steric clashes are not considered — a large number of false positives would be produced. To avoid such clashes, the pharmacophore query should be augmented with excluded volume constraints on the R-groups as well as around the binding site (within 2.2 Å) for modeling receptor shape.

In addition, more traditional features such as hydrogen bond donors and acceptors can be added to the query. In the case of Factor Xa, the M55532 ligand had a hydrogen bond to Gly216 and the electrostatic maps show a strong preference for a negative feature at that location.

The augmented query consists of:

  1.  four link features (Link&Link2 to preserve two bonds); and
  2.  a 1.0 Å acceptor pharmacophore feature (Acc); and
  3.  excluded volumes on the R-groups; and
  4.  a union of excluded volume constraints for receptor shape.
This query results in fewer hits as shown in Figure 9.

Figure 9. 16 candidate linkers overlayed with M55532 in the active site of Factor Xa. Excluded volumes and one acceptor feature were used in a 5-point pharmacophore query.

Table 1 shows some of the returned potential scaffolds. Not all of the candidate scaffolds connect a nitrogen to the pyridine ring. This is a problem since the scaffold nitrogen is required to form the 4-amino pyridine group to retain the key basic feature for binding in the D pocket.

Table 1: Structures of potential linker scaffolds between the napthalene and pyridil R-groups in M55532

Fortunately, the MOE pharmacophore tools allow for SMARTS patterns in the query. In this case, a boolean expression is added to ensure that the basic nitrogen is part of the scaffold. The Link query feature of the scaffold is replaced with Link & "N". The resulting query produces two distinct candidate linkers shown below.


Figure 10. The pharmacophore query with two candidate replacement scaffold. The query consists of a) exit vectors defined by (Link,Link2) pairs drawn in yellow (0.3 Å). The Link feature on the scaffold atom connecting to the pyridyl ring is further specified with Link & "N" to preserve the nitrogen atom, b) an acceptor feature (Acc) drawn in cyan, c) two excluded volumes drawn in red to avoid steric clashes with the R-groups (2.6 Å and 3.6 Å), and d) an excluded volume around the receptor (2.2 Å). In each case, the native ligand carbons are colored in gray and a candidate replace scaffold carbons are colored green.

Both candidate linkers preserve the substituent bond vectors, do not clash with the receptor and contain a nitrogen atom satisfying the requirement for a hydrogen bond acceptor with Gly216. In addition, the 4-amino pyridine group is maintained in both scaffolds required for the creation of a salt bridge with Glu97.

Summary

We have shown how MOE's pharmacophore tools are used to perform scaffold replacement experiments. In similar fashion to CAVEAT, an exit vector is defined using pairs of special “Link” pharmacophore features. A pair of Link features on a scaffold atom and a connected atom from the R-group are defined for each bond that needs to be preserved. Choosing R-groups is done by identifying key interactions between the native ligand and the receptor. Active site analysis tools such as Ligand Interactions, Contact Statistics, and Electrostatic Maps can also act as a guide in choosing appropriate R-groups.

MOE's pharmacophore tools show several advantages to traditional computational scaffold replacement techniques. Unlike the pure CAVEAT methodology alone, MOE's Link features can be used in conjunction with other query features, as well as volumes to create more sophisticated queries that preserve important scaffold interactions or avoid van der Waals clashes with the receptor. Ad hoc SMARTS expressions can be incorporated to enforce specific chemical group requirements. In addition, no special preparation is required for the linker database — any 3D conformation database can be searched.

The combination of active site analysis tools and pharmacophore-based scaffold replacement methods means that scaffold replacement can be routinely performed in structure-based design projects.

References

[Clark 2007] Clark, A.M; Labute, P. 2D Depiction of Protein-Ligand Complexes J. Chem. Inf. Model. 47 (2007) 1933-1944.
[Davie 1991] Davie, E.W; Fujikawa, K.; Kisiel, W.; The Coagulation cascade: initiation, maintenance, and regulation. Biochemistry 30 (1991) 10363-1-370.
[Halgren 1996] Halgren, T.A.; The Merck Force Field; J. Comp. Chem. 17 (1996) 490-641.
[Kastenholz 2000] Kastenholz, M.A.; Pastor, M.; Cruciani, G.; Haaksma, E.E.J.; Fox, T.; GRID/CPCA: A New Computational Tool to Design Selective Ligands. J. Med. Chem. 43 (2000) 3033-3044.
[Lauri 1994] Lauri, G., Bartlett, P. A.; CAVEAT: A Program to Facilitate the Design of Organic Molecules; J. Comp. Aided Mol. Des. 8 (1994) 51-66.
[Lewell 1998] Lewell, X.Q., Judd, D.B., Watson, S.P., Hann, M.M.; RECAP — Retrosynthetic Combinatorial Analysis Procedure: A Powerful New Technique for Identifying Privileged Molecular Fragments with Useful Applications in Combinatorial Chemistry; J. Chem. Inf. Comput. Sci. 38 (1998) 511-522.
[Maass 2007] Maass, P.; Schulz-Gasch, T.; Stahl, M.; Rarey, M.; Recore: A Fast and Versatile Method for Scaffold Hopping Based on Small Molecule Crystal Structure Conformations; J. Chem. Inf. Model. 27 (2007) 390-399.
[Takano 2007] Takano Y.; Koizumi M.; Takarada R.; Takimoto Kamimura M.; Czerminski R.; Koike T.; Computer-aided design of a factor Xa inhibitor by using MCSS functionality maps and a CAVEAT linker search; Journal of Molecular Graphics and Modelling 22 (2003) 105-114.
[Stubbs 1995] Stubbs, M.T.; Huber, R.; Bode, W.; Crystal Structure of Factor Xa Specific Inhibitors in Complex with Trypsin: Structural Grounds for Inhibition of Factor Xa and Selectivity Against Thrombin. FEBS Lett. 375 (1995) 103-107.