Unknown Receptor

4.02.3.1 Pharmacophores

The success of QSAR led to efforts to extend the domain to noncongeneric series, where the structural similarity between molecules active in the same bioassay was not obvious. The work of Beckett and Casey10 on opioids to define parts of active molecules (pharmacophoric groups) essential for efficacy was seminal. Kier further developed the concept of pharmacophore and applied it to rationalize the SAR of several systems.11 Gund and Wipke implemented the first in silico screening methodology with a program to screen a molecular database for pharmacophoric patterns in 1974.12,13

An early example of pharmacophore development involved superimposing apomorphine, chlorpromazine, and butaclamol such that the amines are aligned while maintaining the coplanarity of an aromatic ring.14 This led to a plausible hypothesis of receptor-bound conformations at the dopamine receptor. Least-squares fitting of atomic centers did not allow such an overlap, but the use of the centroid of the aromatic ring with normals to the plane for least-squares fitting accomplished the desired overlap.

There still continues to be method development to generate overlaps of hydrogen-bond donors and acceptors, aromatic rings, etc. to formulate a pharmacophore hypothesis from a set of active compounds for a given receptor/ enzyme. One method developed early at Washington University involved minimization of distances between groups in different molecules assigned by the investigator with no intermolecular interactions. In effect, adding springs caused the groups to overlap as the energy of the entire set of molecules was minimized, excluding any interatomic interactions with the exception of those imposed by the springs (Figure 2).

The results were dependent on the starting conformations of the set of molecules being minimized, and multiple starting conformations were used to generate multiple pharmacophoric hypotheses. Alternative algorithms utilize multiple distance constraints on the molecular ensemble. They also embedded the matrix of constraints into three dimensions utilizing distance geometry15 or systematically determined the set of sterically allowed conformers of each molecule and compared their pharmacophoric patterns in three dimensions.14 A distinct advantage generally exists in using internal coordinates in comparison of molecules, as internal distances are invariant to global rotations and translations.

4.02.3.2 Active-Analog Approach

The early work by medicinal chemists to try and rationalize their SAR with three-dimensional models, as well as the success of Hansch and others in correlating SAR with physical properties, led to exploration of molecular modeling as a

Figure 2 Schematic diagram of minimization approach to overlap of pharmacophoric groups (A with A' with A", B with B' with B", C with C' with C'') by introduction of constraints (springs) with intermolecular interactions ignored and only intramolecular interactions considered.

means of combining the two approaches. Clearly, overall physical properties such as hydrophilicity, steric volume, charge, and molar refractivity would be more meaningful in the context of a specific subsite within the receptor, rather than when considered as an overall molecular average. One expected models with greater resolution and with the ability to discriminate between stereoisomers, for example, as a result of the inclusion of geometrically sensitive parameters. By 1979, the group at Washington University had developed a systematic computational approach to the generation of pharmacophore hypotheses, the Active-Analog Approach, which was disclosed at the American Chemical Society (ACS) National Meeting that year.14 Many, more sophisticated variations16'17 of this approach have subsequently been developed to generate three-dimensional hypotheses regarding molecular recognition.

The basic premise for the Active-Analog Approach was that each compound tested against a biological assay was a three-dimensional question for the receptor. But each molecule was, in general, flexible and could present a plethora of possible three-dimensional arrays of interactive chemical groups. By computationally analyzing the sets of possible three-dimensional pharmacophoric patterns associated with each active molecule, one could find those three-dimensional pharmacophoric patterns common to a set of actives. In the simplest case, each inactive molecule would be geometrically precluded from presenting the given pharmacophoric pattern common to active molecules by steric or cyclic constraints. In practice, inactives capable of presenting the hypothetical pharmacophoric pattern were found often, so another rationale for their inactivity was necessary. Aligning each active molecule to the candidate pharmacophoric pattern allowed determination of the volume requirements of the set of actives. An inactive compound could present the correct pharmacophoric pattern if it competed for extra volume that was occupied by the receptor. When an inactive was aligned with the pharmacophore as scaffold, subtraction of the active volume space could identify such novel requirements in inactives.

Earlier, a Gaussian representation of molecular volume18 that readily allowed mathematical manipulation of atomic volumes was developed. A data set with a set of rigid bicyclic amino acids that inhibited S-adenosylmethionine, the enzyme that synthesizes the active methyl donor, provides the best example of this rationalization.19 In this case, the amino acid portion provided a common frame of reference that revealed the compounds that lost ability to inhibit the enzyme shared a small volume not required by active compounds (presumably required by an atom of the enzyme). No other plausible suggestion for the data set has ever been suggested because the physical properties of this series of actives and inactives were effectively identical, and the amino acid portion was clearly essential for enzyme recognition.

Two other examples of receptor mapping from analysis of SAR data were published on the glucose sensor20 and on the y-amino-butyric acid (GABA) receptor.21

The thesis work of Font on the tripeptide thyrotropin-releasing hormone (TRH), pyroglutamyl-histidyl-prolineamide, is the example of the earliest determination of the receptor-bound conformation of a biologically active peptide using the Active-Analog Approach (Figure 3). Only six torsional angles needed to be specified to determine the backbone conformation and relative position of the imidazole ring of the bioactive conformation. Two alternative conformers were consistent with the conformational constraints required by the set of constrained analogs analyzed. Font designed several polycyclic analogs to test his hypothesis that were intractable for the synthetic procedures available at the time. These compounds served as a catalyst for the design of some novel electrochemical approaches by Moeller of Washington University.22-25 Once the compounds could be prepared, their activity fully supported the receptor-bound conformation derived a decade before.23'25

4.02.3.3 Active-Site Modeling

Crystal structures of protein-ligand complexes (the set of complexes of thermolysin with a variety of inhibitors determined in the Matthews laboratory, for example26) clearly show a major limitation of the pharmacophore assumption. In complexes, ligands do not optimize overlap of similar chemical functionality, but find a way to maintain correct hydrogen-bonding geometry, for example, while accommodating other molecular interactions (Figure 4).

In 1985, ACE was an object of intense interest in the pharmaceutical industry, as captopril and enalapril, the first two approved drugs inhibiting ACE, were being extensively used to treat hypertension. Thus, each pharmaceutical company was contending to design novel chemical structures that inhibited ACE and minimized side effects to gain a piece of the market. Inhibitors of ACE served as a test bed for the Active-Site Mapping where one tries to deduce the receptor-bound conformation of a series of active analogs based on the assumption of a common binding site. Analysis of the minimum energy conformations of eight ACE inhibitors had revealed a common low-energy conformation of the Ala-Pro segment.297 By including additional geometrical parameters, the carboxyl group of enalopril could include the zinc atom with optimal geometry from crystal structures of zinc-carboxyl complexes. Similarly, the sulfhydryl group of captopril could be expanded to include the zinc site as well with additional parameters to allow for appropriate geometrical variation. It is much more reasonable to assume that the groups involved in chemical catalysis and substrate recognition in the enzyme have a relatively stable geometrical relationship, in contrast to chemical groups in a set of diverse ligands. Mayer etal.27 analyzed 28 ACE inhibitors of diverse chemical structure available by 1987, as well as two inactive compounds with appropriate chemical functionality. Based on these data, a unique conformation for the core portion of each molecule interacting with a hypothetical ACE active site was deduced; the two inactive compounds were geometrically incapable of appropriate interaction.

After nearly two decades of attempts, the crystal structure of the complex of lisinopril with ACE was finally determined.28 The common backbone conformation of ACE inhibitors and the location of the zinc atom, hydrogen-bond donor, and cationic site of the enzyme determined by Mayer et al.27 essentially overlaps that seen in the crystal structure of the complex (Figure 5), arguing that, at least for this case, the assumptions regarding the relative geometric stability of groups important in catalysis or recognition are valid.29

Figure 4 Schematics of (a) pharmacophore modeling with assumed ligand groups A = A', B = B', and C = C'; and of (b) active-site modeling with receptor groups X = X', Y = Y', and Z = Z'.

Figure 4 Schematics of (a) pharmacophore modeling with assumed ligand groups A = A', B = B', and C = C'; and of (b) active-site modeling with receptor groups X = X', Y = Y', and Z = Z'.

Figure 5 Overlap29 of crystal structure of complex of the inhibitor lisinopril with ACE and the predicted enzyme-bound conformation of ACE inhibitors by Mayer etal.27 Note overlap between positions of pharmacophoric groups interacting with zinc (orange), C-terminal carboxyl, and carbonyl oxygen of amide, the groups targeted by active-site modeling. The phenyl group common to enalapril analogs such as lisinopril (white ring) was not constrained (green ring) by analogs available at the time of the analysis in 1987. Reproduced with permission from Kuster, D. J.; Marshall, G. R. J. Comput.-Aided Mol. Des. 2005, 19, 609-615.

Figure 5 Overlap29 of crystal structure of complex of the inhibitor lisinopril with ACE and the predicted enzyme-bound conformation of ACE inhibitors by Mayer etal.27 Note overlap between positions of pharmacophoric groups interacting with zinc (orange), C-terminal carboxyl, and carbonyl oxygen of amide, the groups targeted by active-site modeling. The phenyl group common to enalapril analogs such as lisinopril (white ring) was not constrained (green ring) by analogs available at the time of the analysis in 1987. Reproduced with permission from Kuster, D. J.; Marshall, G. R. J. Comput.-Aided Mol. Des. 2005, 19, 609-615.

4.02.3.4 Statistical Modeling

Because QSAR provides information that relates biological activity to molecular properties, a logical extension to QSAR for drug design is to include three-dimensional data in the correlation as well. The combination of QSAR with three-dimensional structural information is known as 3D-QSAR. The success of comparative molecular field analysis (CoMFA), a type of 3D-QSAR by Tripos,30 in generating predictive models was entirely due to a new statistical approach, partial least squares of latent variables (PLS),31 applied to chemistry by Wold of the University of Umea, Sweden. CoMFA probes various interactions at many different points around the molecule. Different types of probes are used to create a set of values in multiple dimensions, and principal component analysis is preformed on the values to provide a correlation of three-dimensional structure and biological activity. The concept that one could extract useful correlations from situations where more variables than observations were present was revolutionary at the time. Traditional linear regression analysis protects the user from chance correlations when too many variables are used. PLS regression recognizes and corrects for cross-correlation between variables and avoids chance correlations in models by systematically determining the sensitivity of the predictability of a model to omission of training data.32

A seminal paper by Cramer33 examined the principal components derived from examining the physical property data of a large set of chemicals from the Handbook of Chemistry and Physics. In effect, only two principal components were responsible for a significant amount of the variance of the data in the model derived. The derivation of chemical principles in terms of those two properties provides a great deal of simplification. An analogous example is the simplification that arises from using internal coordinates, i.e., distances between atoms rather than coordinates, in structural comparisons, enabling the specific global orientation of each molecule to be eliminated from consideration.

Was this article helpful?

0 0
Reducing Blood Pressure Naturally

Reducing Blood Pressure Naturally

Do You Suffer From High Blood Pressure? Do You Feel Like This Silent Killer Might Be Stalking You? Have you been diagnosed or pre-hypertension and hypertension? Then JOIN THE CROWD Nearly 1 in 3 adults in the United States suffer from High Blood Pressure and only 1 in 3 adults are actually aware that they have it.

Get My Free Ebook


Post a comment