15
Abstract The Human Genome Project has reached an important stage with the recent de- claration that the first draft is now complete. In the new post-genomic world, the problem facing scientists is how to maximize the use of the newly acquired data for improving health care. It is estimated that the number of therapeutic targets available for drug discovery will increase from the current number of 600–1,000 to perhaps as many as 5,000–10,000. In addition to the challenges that this num- ber provides to the pharmaceutical industry, there is the issue of sequence varia- tion through single nucleotide polymorphisms (SNPs), some of which may im- pact upon the way that the body handles drug treatment. This may be due to a di- rect effect on the binding site of the protein target through non-conservative al- terations of the amino acid sequence, or else through indirect effects on drug me- tabolizing enzymes. In order to meet these challenges, the marriage of structural proteomics and computer-aided small molecule design will provide opportunities for creating new molecules in silico; these may be designed to bind to selected pharmacogenetic variants of a protein in order to overcome the non-responsive- ness of certain patient groups to a particular medicine. The basic aspects of these technologies, and their applicability to selected targets showing structural varia- tion, form the basis of this chapter. 7.1 Introduction Much has been written concerning the expense and problems encountered by the modern pharmaceutical industry in developing novel drugs. As a further diffi- culty, the patient population (market) is becoming fragmented due to genetic vari- ation in the response to medicines resulting from alterations in the drug target or in the metabolism of the compound once ingested. The relatively new discipline of pharmacogenetics is concerned with the inheritance of these variations, mea- sured using single nucleotide polymorphism (SNP) analysis at the level of genom- ic DNA. Pharmacogenomics, on the other hand, is relevant when the genetic vari- 143 7 Pharmacogenomics and Drug Design Philip Dean, Paul Gane and Edward Zanders Pharmacogenomics: The Search for Individualized Therapies. Edited by J. Licinio and M.-L. Wong Copyright © 2002 Wiley-VCH Verlag GmbH & Co. KGaA ISBNs: 3-527-30380-4 (Paper); 3-527-60075-2 (Electronic)

Pharmacogenomics || Pharmacogenomics and Drug Design

  • Upload
    ma-li

  • View
    217

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Pharmacogenomics || Pharmacogenomics and Drug Design

Abstract

The Human Genome Project has reached an important stage with the recent de-claration that the first draft is now complete. In the new post-genomic world, theproblem facing scientists is how to maximize the use of the newly acquired datafor improving health care. It is estimated that the number of therapeutic targetsavailable for drug discovery will increase from the current number of 600–1,000to perhaps as many as 5,000–10,000. In addition to the challenges that this num-ber provides to the pharmaceutical industry, there is the issue of sequence varia-tion through single nucleotide polymorphisms (SNPs), some of which may im-pact upon the way that the body handles drug treatment. This may be due to a di-rect effect on the binding site of the protein target through non-conservative al-terations of the amino acid sequence, or else through indirect effects on drug me-tabolizing enzymes. In order to meet these challenges, the marriage of structuralproteomics and computer-aided small molecule design will provide opportunitiesfor creating new molecules in silico; these may be designed to bind to selectedpharmacogenetic variants of a protein in order to overcome the non-responsive-ness of certain patient groups to a particular medicine. The basic aspects of thesetechnologies, and their applicability to selected targets showing structural varia-tion, form the basis of this chapter.

7.1Introduction

Much has been written concerning the expense and problems encountered by themodern pharmaceutical industry in developing novel drugs. As a further diffi-culty, the patient population (market) is becoming fragmented due to genetic vari-ation in the response to medicines resulting from alterations in the drug target orin the metabolism of the compound once ingested. The relatively new disciplineof pharmacogenetics is concerned with the inheritance of these variations, mea-sured using single nucleotide polymorphism (SNP) analysis at the level of genom-ic DNA. Pharmacogenomics, on the other hand, is relevant when the genetic vari-

143

7

Pharmacogenomics and Drug DesignPhilip Dean, Paul Gane and Edward Zanders

Pharmacogenomics: The Search for Individualized Therapies.Edited by J. Licinio and M.-L. Wong

Copyright © 2002 Wiley-VCH Verlag GmbH & Co. KGaAISBNs: 3-527-30380-4 (Paper); 3-527-60075-2 (Electronic)

Page 2: Pharmacogenomics || Pharmacogenomics and Drug Design

ation leads to changes in protein expression or conformation of key ligand bind-ing sites. This may result in loss of drug efficacy due to lack of binding; when theloss of binding occurs in drug metabolizing enzymes, this may result in levels ofcompound in the blood that exceed safety thresholds.

While these aspects of human genetics are problematical for both patients andpharmaceutical companies alike (for obviously different reasons), there are waysforward; these are emerging from new drug discovery technologies, including insilico approaches to drug design. The relevant technologies will be described inthis chapter, but first we shall discuss the background to lead discovery and phar-macogenomics by concentrating on the structural aspects of the protein targetsfor small molecule drugs.

7.2The Need for Protein Structure Information

Traditional drug discovery has relied upon the chemical modification of biologicallyactive natural products or high-throughput screening of compound libraries to ob-tain “hits” that may be converted to “leads” and ultimately drugs. This process isgenerally inefficient, since the nature of compound collections used for screeningis often a reflection of the historical activity of the company in question. This meansthat most compounds will be limited in coverage of the variety of targets encoun-tered in drug research, namely enzymes, receptors and ion channels [1]. Even theadvent of combinatorial chemistry in the mid 1990s has failed to deliver a noticeableincrease in good drug leads (these issues are discussed in [2]).

The problems highlighted above have been compounded by the increase in tar-gets afforded by the genome sequencing projects, culminating in the recent publi-cations of the human sequence [3, 4]. The realization that it will be impossible tofind suitable small molecule candidates for every potential drug target using high-throughput screening alone, has driven the search for alternatives based on an un-derstanding of the three-dimensional structure of the protein. The structural infor-mation on the ligand-binding site may then be used for the in silico design ofcompounds that make strong interactions with appropriate residues within thesesites; alternatively, existing small molecule structures may be docked into the sitesand optimized using medicinal chemistry techniques.

The availability of three-dimensional protein structures is clearly one of therate-limiting steps in this process. Publicly available structures, derived by X-raydiffraction or NMR techniques, are deposited in the Protein Data Bank (PDB),and currently number over 15,000 entries. There is considerable redundancy inthis, however, with the number of single entries for human proteins being ap-proximately 500. Due to the technical difficulties associated with certain classes ofprotein there is a strong bias towards enzymes in the database. This excludes,therefore, the G protein-coupled receptors (GPCR) and ion channel classes thatmake up a large proportion of drug targets. A number of public and private struc-tural proteomics initiatives have been established in order to increase the number

7 Pharmacogenomics and Drug Design144

Page 3: Pharmacogenomics || Pharmacogenomics and Drug Design

and variety of structures that may be used for studies of protein folding or drugdiscovery [5]. Where structures are not available, it is possible in many instancesto create a homology model using the structure of a related protein as a guide [6].The success of drug design corresponds to the % identity between the sequenceand the known structure. Despite these advances (including the structural deter-mination of the GPCR rhodopsin [7]), it has to be accepted that some proteinswill not yield to current structural techniques. This does not mean, however, thatin silico drug design techniques cannot be used with these proteins, since ligand-based design can be employed (see Sect.7.9).

7.3Protein Structure and Variation in Drug Targets – the Scale of the Problem

Variation in the response of individual patients to medicines is an important issuethat is currently being addressed by the pharmaceutical industry [8]. It would beuseful to have some idea of the scale of the problem through determining thenumber of SNPs in the human genome and their relative effects on both the levelof expression and functional activity of the protein target. From the perspective ofrational drug design, it is necessary to consider the alteration of the tertiary struc-ture of the translated protein in which function is retained, but not the binding ofan existing drug. This is directly analogous to the situation with the microbial tar-gets (e.g., HIV reverse transcriptase) in which sequence variations affect the struc-ture of the target, resulting in the problem of drug resistance [9].

An analysis of 1.42 million non-redundant human SNPs has been recently pub-lished by an international collaborative group [10] (plus additional commentary[11]). The majority of these are in repetitive regions, whereas 60,000 lie withincoding and untranslated regions of exons. Bearing in mind that many more SNPsare being identified in the public and private domains, the total number that arepotentially able to disrupt protein structure, while small in comparison with thetotal genome complement, is still likely to be significant in terms of pharmaceuti-cal research opportunities. This study gave an overview of the total number ofSNPs available within the public domain as of the first quarter of 2001.

An analysis of polymorphisms in coding regions was published by Lander’sgroup in 1999 [12]. They identified 560 SNPs (392 in coding regions) in 106genes of relevance to cardiovascular disease, neuropsychiatry and endocrinology.Only a minority of polymorphic changes gave rise to non-conservative amino acidsubstitutions, most likely due to evolutionary selection against deleterious muta-tions in the human genome.

Nevertheless, there is considerable activity in identifying SNPs within protein tar-gets for current marketed drugs. The whole purpose of the study of genetic varia-tions in drug responses is to identify patients who may benefit from a particularmedicine and avoid wasteful (or dangerous) prescription to others [8]. Examples ofdrug targets and detoxification systems that have been studied using SNP or othermutational analyses are listed in Table7.1 (see [13] for a more comprehensive list).

7.3 Protein Structure and Variation in Drug Targets – the Scale of the Problem 145

Page 4: Pharmacogenomics || Pharmacogenomics and Drug Design

These comparatively early studies are beginning to highlight a number ofpoints, including the fact that some SNPs are specific for particular ethnic popula-tions. In addition, some drug targets appear not to have amino acid sequence vari-ation as a result of SNP polymorphism, and therefore will not require a numberof different ligands to accommodate this. However, since variations in tertiarystructure resulting from mutations in the coding region offer opportunities forstructure-based design, some relevant examples will be discussed in detail below.

7.4Mutations in Drug Targets Leading to Changes in the Ligand Binding Pocket

7.4.1�2-Adrenergic Receptor

This well characterized drug target for anti-asthma medications represents one ofthe earliest examples of a natural mutation leading to an alteration in ligand bind-ing [20]. Mutation of Thr164 to Ile in the fourth transmembrane-spanning domain

7 Pharmacogenomics and Drug Design146

Tab. 7.1 Examples of SNPs that influence drug metabolism or disease state

Drug target Disease Comments Reference

NMDAR1 receptor Schizophrenia SNPs in coding region,but no functional signifi-cance

14

Quinone oxidoreductase,Sulfotransferase

Drug metabolism 6 out of 22 coding regionSNPs gave amino acidsubstitutions

15

75 candidate genes forblood pressurehomeostasis

Hypertension 874 candidate SNPs with387 within coding se-quence; 54% of these pre-dicted to change proteinsequence

16

ThiopurineS-methyltransferase

Drug metabolism SNPs in promoter region,introns and 3�UTR, butnot coding region

17

41 candidate genes Ischemic heart disease SNPs restricted to ethnicgroup (Japanese) used instudy

18

CYP3A Drug metabolism SNPs cause alternativesplicing and protein trun-cation

19

�2-adrenergic receptor Asthma SNPs cause alteration ofligand binding

20

BCR-ABLtyrosine kinase

Cancer Point mutation causes re-sistance to STI-571 com-pound

21

Page 5: Pharmacogenomics || Pharmacogenomics and Drug Design

of this G protein-coupled receptor occurs at low frequency in the populationtested, but gives rise to altered ligand binding through interaction with an adja-cent Ser165 as illustrated in Figure 7.1.

This example is one where the accurate three-dimensional structure of the pro-tein is unknown; under these circumstances, it is necessary to create a computermodel. The development of inhibitors that are designed to overcome the effects ofthis mutation could not be based on the accurate structure of a ligand-bindingsite; here, ligand-based design would be appropriate (see Sect.7.9).

7.4.2STI-571 and BCR-ABL

STI-571 is a 2-phenylamino pyrimidine chemotherapeutic agent that targets thetyrosine kinase domain of the BCR-ABL oncogene present in chronic myelogen-ous leukemia (CML) [22]. After many years of research into compounds that areselective for specific oncogenes, this compound is showing real clinical benefitagainst CML [23]. Unfortunately, due to the selective pressure on tumor cells toevolve resistance, a specific mutation in BCR-ABL inhibits the action of STI-571[21]. Since crystal structures of ligand bound into the wild-type protein are avail-

7.4 Mutations in Drug Targets Leading to Changes in the Ligand Binding Pocket 147

Fig. 7.1 Mutation of Thr164 toIle (cyan) in the fourth transmem-brane region of the �2-adrenergicreceptor leads to steric interfer-ence with Ser165 (green) withinthe active site of the receptor.(A) general domain structure andposition of the residues, (B) wildtype, (C) mutant showing interac-tion of Ile164 with Ser165 (see colorplates, p. XXXII).

Page 6: Pharmacogenomics || Pharmacogenomics and Drug Design

able (PDB: 1IEP), it is possible to visualize the effects of a C-T nucleotide muta-tion resulting in a substitution of Thr315 by Ile (Figure 7.2). The change of aminoacid results in disruption of a key hydrogen bond between oxygen on the threo-nine and a secondary amine nitrogen on STI-571.

In this example, structure-based design of inhibitors to the mutant enzyme maybe undertaken using available X-ray structures and homology models.

7.5Resistance to Human Immunodeficiency Virus (HIV)

The human retrovirus HIV can be controlled using chemotherapy directed at thereverse transcriptase and aspartyl protease encoded by the viral genome; as withother microbial pathogens, however, resistance to drug therapy becomes a majorproblem. Figure 7.3 shows a crystal structure (PDB: 1HXW) of the HIV protease,where mutated amino acids (shown in cyan) lead to disrupted binding of the clini-cally effective inhibitor ritonavir [24].

7.6In silico Design of Small Molecules

Changes induced by SNPs (or other mutations) in the protein targets of drugsmay result in the medicine being ineffective. In this case, the relevant diseasemay remain untreated, or else an expensive drug discovery effort will have to beundertaken to identify new molecules that are able to affect the mutant protein.Given that most current small molecule discovery strategies rely on expensive

7 Pharmacogenomics and Drug Design148

Fig. 7.2 Mutation of Thr315 (A) to Ile (B) in BCR-ABL interferes with the binding of STI-571(see color plates, p. XXXIII).

Page 7: Pharmacogenomics || Pharmacogenomics and Drug Design

high-throughput screening and lead optimization programs, this is an unattractiveprospect. Advances in computer-aided design, however, may make this processmore efficient and less costly through creation of virtual small molecules that canby optimized for interaction with the mutant protein before any chemical synthe-sis is undertaken. The remainder of this chapter will be concerned with an over-view of this exciting and rapidly evolving area of research.

7.7Automated Drug Design Methods

The pharmaceutical industry faces a large increase in the number of therapeutictargets available for exploration. New methods must be developed for drug discov-ery that can be automated, are cheap and that optimally explore the chemicalspace available for drug design. The number of different possible chemical struc-tures has been estimated to be of the order 10200, with perhaps 1060 structureswith drug-like properties [25]; these numbers can be compared with the estimatedmass of the visible universe at 1056 g. Thus it would not be possible to make a sin-gle gram of each potential drug molecule. In contrast, the number of differentchemical structures synthesized is small by comparison, and has not yet reached108. Clearly, in silico methods will have to be developed to cope with searchesthrough the vastness of chemical space.

Design paradigms fall into two types:� structure-based methods that use three-dimensional structural information

about the site,� ligand-based methods that can be used for drug discovery in the absence of

structural information.

7.7 Automated Drug Design Methods 149

Fig. 7.3 Binding of the HIV pro-tease inhibitor ritonavir. Amino acidshighlighted in cyan are mutated in re-sistant strains of the virus and tendto occur at the extremities of the in-hibitor (see color plates, p. XXXIII).

Page 8: Pharmacogenomics || Pharmacogenomics and Drug Design

The applicability of structure-based methods is critically dependent on the avail-ability of relevant structural data. The Structural Genomics Initiative [5] is an in-ternational project that seeks to provide high-quality crystallographic or NMR dataabout new proteins. While promising for future drug discovery efforts, the rate ofintroduction of new structures into the public domain is still too slow. To speedthis up, alternative methods for the generation of three-dimensional data usinghomology modeling techniques have been introduced. These are less precise, butshow increasingly better performances as elucidated by the Computer Aided-Structure Prediction (CASP) challenges [26]. The logical next step would be toautomate protein structure prediction in order to provide extensive coverage of thehuman proteome. One of the main drivers for this effort is the need for three-di-mensional structural models of whole gene families for use in drug design.

Ligand-based design methods are built upon identified molecular similarities inknown active ligands. The similarity methods have to cope with finding unspecifiedpartial molecular similarities within the ligand data sets for freely flexible ligands.There are often many solutions to molecular similarity and choices have to be madeas to which one to use. Figure 7.4 summarizes the two approaches to drug design.

7.8Structure-Based Drug Design

The automation of structure-based design can be divided into key areas for devel-opment:� identification of binding sites,� automated analysis of binding sites,� prioritization of strategies for design within sites,� de novo design of scaffolds (active templates),� elaboration of scaffolds into combinatorial libraries.

7 Pharmacogenomics and Drug Design150

Fig. 7.4 Scheme for drug design utilizinggenomic data. The left track illustrates theprocess of drug design based on structural

determination of binding sites. The right trackoutlines design from known ligands.

Page 9: Pharmacogenomics || Pharmacogenomics and Drug Design

The identification of binding sites can be initiated from protein sequence informa-tion. Many sites show conservation of structural motifs; for example, many enzy-matic catalytic sites have highly conserved local sequences within a gene family.Many of these motifs can be identified at the annotation stage of ascribing a func-tion to a novel gene. Furthermore, the identification of co-factor binding sites canalso provide additional clues to function. At the moment, the prediction of the mo-lecular structure of the substrate has not been solved precisely, although dockingalgorithms could provide clues for potential substrate fitting into binding sites.

Ligands bind to their sites through specific molecular interactions, including hy-drogen bonds, electrostatic interactions, steric interactions, interactions through hy-dration and hydrophobic interactions. This pattern lies at the surfaces of both ligandand binding site, so that complementary interactions are optimized for efficient bind-ing. If the target proteins have 3D coordinate data, computer algorithms can be usedto automatically analyze the binding features and categorize them on a grid map ofthe protein surface. The features mapped onto grid points are known as site points.Structurally and functionally related proteins may then be compared by superposingthe binding sites and aligning the pattern of chemical features. This facility is impor-tant in pharmacogenomics where we may wish either to design compounds that bindto a class of sites within related proteins, or to design compounds specific for a par-ticular member of the protein family. In the latter case, the aim is to acquire selec-tivity. Figure 7.5 illustrates the family of caspase enzymes; caspases 1, 4 and 5 form aseparate group – the ICE sub-family, the rest are in the CED-3 subfamily.

The CED-3 sub-family shows very similar three-dimensional structures. Fig-ure 7.6 shows the superpositions of the C� atoms. The structural backbones ofthe proteins within the family are closely superposed.

The concept of dividing sites into a collection of site points for design providesa method for directing design to specific regions. In some respects this processhas parallels to drug design from pharmacophores where a small number ofpoints are used to search compound collections by virtual screening. 4-point phar-macophores provide three-dimensionality to the search. However, the number ofsite points encountered in a site, often about 30, creates a combinatorial problemfor selection [27]. Suppose that there are n site points; if r site point interactionsare needed for a molecule to have measurable affinity, the number of combina-tions of r site points, C(n,r) is given by

7.8 Structure-Based Drug Design 151

Fig. 7.5 Sequence similarities ofthe caspase family.

Page 10: Pharmacogenomics || Pharmacogenomics and Drug Design

C�n� r� � n���r��n � r���

C(n, r) is a maximum when r = n/2. For the case where n = 30 and r = 5, thereare 140000 subsets of 5 site points. Each subset would be a strategy for design.However, some strategies would lead to preferred designs, e.g., a strategy that in-cluded interactions with the catalytic residues would be expected to consistentlyperform well in designing inhibitors. Moreover, if there were crystallographic evi-dence of substrates or known inhibitors binding to certain residues, these wouldform a basic set for de novo design. Other methods for prioritizing design strate-gies could be built on hydrogen bonding regions, where there are overlaps in hy-drogen bonding probability. In the pharmacogenomic problem of design within afamily of sites, the choice of design strategy is paramount for obtaining selectivity.The selection of site points should be such that the set chosen for the target pro-tein should be as different as possible from nearly corresponding sets within thefamily of proteins.

The affinity of the designed drug molecule for its site is governed by the free en-ergy of interaction �G. �G can be determined experimentally from the equation

�G � �RT ln Kb

Where Kb is the experimental binding constant, R is the gas constant and T is thetemperature. The free energy has an enthalpic and entropic component.

�G � �H � T�S

Where �H is the change in enthalpy on binding and �S is the change in entropy.

7 Pharmacogenomics and Drug Design152

Fig. 7.6 Superposed C� atoms from aselection of proteins in the CED-3 familyof caspases.

Page 11: Pharmacogenomics || Pharmacogenomics and Drug Design

De novo methods for drug design attempt to build novel molecules within speci-fied regions of the site. Given the large number of molecules that could be cre-ated, it is important to assess the relative differences in predicted binding affinityfor the designed ligands and the site. This is achieved using a scoring functionbased on the equations described above.

A composite scheme for de novo design is shown in Figure 7.7.Molecules are built from small molecular fragments, the latter being derived

from databases of drug molecules such as the World Drug Index, or from proprie-tary compound collections. Fragments are labeled for attachment and substitutionpositions. Molecules are then built into the site by a stochastic assembly processfrom randomly selected fragments. This process is controlled by chemical rulesthat govern the addition, removal or exchange of a fragment on the evolving skele-ton. At each modification, the assembled structure is fitted in the site by posi-tional and conformational transformations, with the scoring function being usedto monitor the prediction of the free energy of the interaction between ligand andsite. An optimal assembly procedure is performed using combinatorial optimiza-tion routines, such as simulated annealing or genetic algorithms.

Simulated annealing is an ergodic optimization method, meaning that givensufficient time, the algorithm converges on the optimal solution irrespective ofthe initial starting position. If time constraints are placed on the algorithm runtime, it behaves non-ergodically and different solutions can be obtained that areacceptable, but suboptimal. In structure assembly this is important, since a varietyof molecular structures can be produced, thus giving a choice of different chemis-tries. These solutions can be analyzed statistically to identify different classes ofstructures that fit a chosen starting strategy.

Constraints can be placed on the structure assembly to limit the size of the mol-ecule being generated so that it can be used as a template for further decoration bycombinatorial chemistry. Virtual combinatorial chemical libraries can be createdround the designed template to explore regions close to where the template has beendesigned to bind. The in-site enumeration of combinatorial libraries designed for aparticular site is very important for chemical genomics, since it offers a way of de-signing molecules to distinguish between mutational differences in a particular site.

7.8 Structure-Based Drug Design 153

Fig. 7.7 Scheme for de novo design.

Page 12: Pharmacogenomics || Pharmacogenomics and Drug Design

7.9Ligand-Based Drug Design

In the absence of useful three-dimensional information about the site, the de-signer can only build novel molecular structures based on molecular similaritywith existing active molecules. Ligand-based design requires the resolution of anumber of technical issues. These are itemized below.1. A set of molecules is needed to extract spatially distributed chemical features

for inclusion in drug design.2. Structurally dissimilar, active molecules provide more information about the

site than a series of active close structural homologs.3. The molecules in the initial ligand set may have numerous torsion angles

making identification of the active conformation difficult.4. The choice of molecular similarity procedure needs to be made from a wide

variety of available methods, preferably including one that includes a searchfor partial molecular similarity. This allows a more useful set of similarities tobe identified.

5. Molecules within the ligand set could have different binding modes whichneed to be identified before selecting a subset for a design strategy.

6. Molecules within the ligand set, with similar binding modes, need to besuperposed such that the site points inferred from the corresponding ligandpoints show common spatial positions.

7. The constraining supersurface of the set of molecules needs to be constructedto provide limits for automated design procedures.

8. Since there is no information about the site, a scoring function based on freeenergy methods cannot be used.

A molecular similarity method that is applicable to this method of design has recentlybeen published as the algorithm SLATE [28]. This algorithm takes a set of moleculesand compares them in a pairwise fashion for molecular similarity. The molecules areallowed to flex. In the case of potential hydrogen bonding interactions between theligand points and possible complementary site points, site points are projected fromthe ligand surfaces. Hydrophobic regions are handled as local regional centroids. Themolecular description is reduced to a set of points with associated properties. Thesesets of points can be matched using distance geometry and the search optimized bysimulated annealing through conformational space. The procedure generates a simi-larity matrix; partial similarity is handled by null correspondences [27].

An application of this procedure to the design of novel compounds for the hista-mine H3 receptor has been published [29]. Similarity data suggested that four hydro-gen bonding site points, together with two lipophilic regions could be identified with acommon spatial distribution. Furthermore, the constraining supersurface showedstrong similarity with each of the selected ligand conformations. From these in silicodesigns, histamine H3 antagonists were synthesized and showed high affinity for thesite. Thus computational methods could be used to design potent ligands without anyprior knowledge of the sequence or structure of the receptor.

7 Pharmacogenomics and Drug Design154

Page 13: Pharmacogenomics || Pharmacogenomics and Drug Design

7.10Future Directions

New developments in pharmacogenomics will impact on drug design at threemain levels� the interaction of the drug with its receptor binding site,� the absorption and distribution of the drug,� the elimination of the drug from the body.

If we are ever to achieve personalized drug therapies, the above issues will have tobe addressed.

The bulk of this chapter has discussed SNPs in terms of mutated drug bindingsites and the resulting changes that occur in response to medicines. In each ofthese cases, drug design strategies may need to be modified to develop drug mole-cules that would be selective and effective in the mutated site.

Absorption, in general, is treated as a physicochemical transport process basedon computations of logP (the octanol/water partition coefficient) and solubilitygoverned by factors such as polar surface area on the molecule. It is conceivablethat SNPs in drug transporter genes will affect the pharmacokinetic properties ofcompounds and, therefore, these may have to be taken into consideration in thedesign process.

Elimination mechanisms may include active excretion of drug molecules fromcells, as with multiple drug resistance that arises in cancer chemotherapy. Pa-tients with genotypes for particular MDR pathways could be identified and alter-native treatments provided. Pharmacogenetic differences in the response to drugscan often be related to population differences in drug metabolizing enzymes.Crystal structures, or homology models of these enzymes can be used to screencompounds designed de novo before synthesis has begun. This can be done at twolevels:� The virtual compounds can be input into metabolism prediction programs,

such as Metabolexpert [30] or Meteor [31], to identify principal pathways of ex-pected drug metabolism.

� The virtual compounds can be screened against structural models of the meta-bolizing enzymes, including the known SNP variants. These procedures are be-coming widely adopted for the cytochrome P450 isozymes involved in oxidativedrug metabolism.

Genomic pre-screening of patients for SNP mutations in drug metabolism willimprove the utility of clinical trials by focusing attention on the response of spe-cific subpopulations to drug treatment. The consequence of this approach is thatit should be possible to industrialize the creation of individual therapies, althoughat a significant increase in cost.

7.10 Future Directions 155

Page 14: Pharmacogenomics || Pharmacogenomics and Drug Design

7.11Conclusions

Pharmacogenomics will have a major influence on drug discovery, through drugdesign, as well clinical practice. The subtle changes in biomolecular structurecaused by small SNP-induced amino acid alterations will profoundly affect thesearch for personalized medicines. In the future, greater emphasis will be placedon very precise differences in drug structures that select between mutated pro-teins. As well as being scientifically challenging, this approach will clearly havesignificant economic effects on the pharmaceutical industry.

7 Pharmacogenomics and Drug Design156

7.12References

1 Drews JJ. Drug discovery: A historicalperspective. Science 2000; 287:1960–1964.

2 Bailey D, Brown D. High-throughputchemistry and structure-based design:survival of the smartest. Drug DiscoveryToday 2001; 6:57–59.

3 International Human Genome Sequenc-ing Consortium. Initial sequencing andanalysis of the human genome. Nature2001; 409:860–921.

4 Venter JC, Adams MD, Myers EW, Li

PW, Mural RJ, Sutton GG et al. The se-quence of the human genome. Science2001; 291:1304–1351.

5 Burley SK. An overview of structuralgenomics. Nature Struct Biol 2000; 7(Suppl):932–934.

6 Maggio ET, Ramnarayan K. Recent de-velopments in computational proteomics.Trends Biotechnol 2001; 19:266–272.

7 Palczewski K, Kumasaka T, Hori T,

Behnke CA, Motoshima H, Fox BA, Le

Trong I, Teller DC, Okada T, Sten-

kamp RE, Yamamoto M, Miyano M.

Crystal structure of rhodopsin: a G pro-tein-coupled receptor. Science 2000;289:739–745.

8 Roses AD. Pharmacogenetics and thepractice of medicine. Nature 2000;405:857–865.

9 Richman, DE. HIV chemotherapy. Na-ture 2001; 410:995–1001.

10 The International SNP Map WorkingGroup. A map of human genome se-quence variation containing 1.42 million

single nucleotide polymorphisms. Nature2001; 409:928–933.

11 Marth G, Yeh R, Minton M, Donald-

son R, Li Q, Duan S, Davenport R,

Miller RD, Kwok PY. Single-nucleotidepolymorphisms in the public domain:how useful are they? Nature Genet 2001;4:371–372.

12 Cargill M, Altshuler D, Ireland J,

Sklar P, Ardlie K, Patil N, Shaw N,

Lane CR, Lim EP, Kalyanaraman N, Ne-

mesh J, Ziaugra L, Friedland L, Rolfe

A, Warrington J, Lipshutz R, Daley

GQ, Lander ES. Characterization of sin-gle-nucleotide polymorphisms in codingregions of human genes. Nature Genet1999; 3:231–238.

13 Evans WE, Relling MV. Pharmacoge-nomics: translating functional genomicsinto rational therapeutics. Science 1999;286:487–491.

14 Rice SR, Niu N, Berman DB, Heston

LL, Sobell JL. Identification of single nu-cleotide polymorphisms (SNPs) andother sequence changes and estimationof nucleotide diversity in coding andflanking regions of the NMDAR1 recep-tor gene in schizophrenic patients. MolPsychiatry 2001; 6:274–284.

15 Iida A, Sekine A, Saito S, Kitamura Y,

Kitamoto T, Osawa S, Mishima C, Na-

kamura Y. Catalog of 320 single nucleo-tide polymorphisms (SNPs) in 20 qui-none oxidoreductase and sulfotransferasegenes. J Hum Genet 2001; 46:225–240.

Page 15: Pharmacogenomics || Pharmacogenomics and Drug Design

7.12 References 157

16 Halushka MK, Fan JB, Bentley K, Hsie

L, Shen N, Weder A, Cooper R, Lip-

shutz R, Chakravarti A. Patterns ofsingle-nucleotide polymorphisms in can-didate genes for blood-pressure homeo-stasis. Nature Genet 1999; 22:239–247.

17 Seki T, Tanaka T, Nakamura Y. Genom-ic structure and multiple single-nucleo-tide polymorphisms (SNPs) of the thio-purine S-methyltransferase (TPMT) gene.J Hum Genet 2000; 45:299–302.

18 Ohnishi Y, Tanaka T, Yamada R, Sue-

matsu K, Minami M, Fujii K, Hoki N,

Kodama K, Nagata S, Hayashi T, Ki-

noshita N, Sato H, Sato H, Kuzuya T,

Takeda H, Hori M, Nakamura Y. Identi-fication of 187 single nucleotide poly-morphisms (SNPs) among 41 candidategenes for ischemic heart disease in theJapanese population. Hum Genet 2000;106:288–292.

19 Kuehl P, Zhang J, Lin Y, Lamba J, As-

sem M, Schuetz J, Watkins PB, Daly

A, Wrighton SA, Hall SD, Maurel P,

Relling M, Brimer C, Yasuda K, Ven-

kataramanan R, Strom S, Thummel K,

Boguski MS, Schuetz E. Sequence di-versity in CYP3A promoters and charac-terization of the genetic basis of poly-morphic CYP3A5 expression. NatureGenet 2001; 27:383–391.

20 Green SA, Cole G, Jacinto M, Innes

M, Ligett SB. A polymorphism of thehuman �2 adrenergic receptor within thefourth transmembrane domain alters li-gand binding and functional propertiesof the receptor. J Biol Chem 1993;268:23116–23121.

21 Gorre ME, Mohammed M, Ellwood K,

Hsu N, Paquette R, Rao PN, Sawyers

CL. Clinical resistance to STI-571 cancertherapy caused by BCR-ABL gene muta-tion or amplification. Science 2001;293:876–880.

22 Druker BJ, Tamura S, Buchdunger E,

Ohno S, Segal GM, Fanning S, Zim-

mermann J, Lydon NB. Effects of a se-lective inhibitor of the Abl tyrosine ki-

nase on the growth of Bcr-Abl positivecells. Nature Med 1996; 2:561–566.

23 Druker BJ, Sawyers CL, Kantarjian H,

Resta DJ, Reese SF, Ford JM, Capde-

ville R, Talpaz M. Activity of a specificinhibitor of the BCR-ABL tyrosine kinasein the blast crisis of chronic myeloid leu-kemia and acute lymphoblastic leukemiawith the Philadelphia chromosome N.Engl J Med 2001; 344:1084–1086.

24 Boden D, Markowitz M. Resistance tohuman immunodeficiency virus type 1protease inhibitors. Antimicrob AgentsChemother 1998; 42:2775–2783.

25 Klebe G. Virtual screening: An alterna-tive or complement to high throughputscreening. Perspectives in Drug Discov-ery and Design 2000; 20:9 (preface).

26 http://PredictionCenter.llnl.gov27 Dean PM. Defining molecular similarity

and complementarity for drug design. In:Molecular Similarity in Drug Design(Dean PM, ed.), Blackie Academic & Pro-fessional 1995; 1–23.

28 Mills JEJ, De Esch IJP, Perkins TDJ,

Dean PM. SLATE: a method for thesuperposition of flexible ligands. J Com-put Aided Mol Design 2001; 15:81–96.

29 De Esch IJP, Mills JEJ, Perkins TDJ,

Romeo G, Hoffmann M, Wieland K,

Leurs R, Menge WMPB, Nederkoorn

PHJ, Dean PM, Timmerman H. Devel-opment of a pharmacophore model forhistamine H3 receptor antagonists, usingthe newly developed molecular modelingprogram SLATE. J Med Chem 2001;44:1666–1674.

30 Darvas F. Predicting metabolic pathwaysby logic programming. J Mol Graph1988; 6:80–86.

31 Greene N, Judson PN, Langowski JJ,

Marchant CA. Knowledge-based expertsystems for toxicity and metabolism pre-diction: DEREK, StAR and METEOR.SAR and QSAR in Environmental Re-search. 1999; 10:299–314.