Molecular Diversity, 5: 75–89, 2000. KLUWER/ESCOM © 2002 Kluwer Academic Publishers. Printed in the Netherlands.

75

Combinatorial chemistry and high-throughput screening in drug discovery: Different strategies and formats Pierfausto Seneci1,∗ & Stanislav Miertus2 1

Nucleotide Analog Pharma AG, Landsbergerstrasse 50, D-80339 München, Germany; 2 International Centre for Science and High Technology of United Nations Industrial Development Organization, Area Science Park, Building L2 Padriciano 99, I-34012 Trieste, Italy (∗ Author for correspondence, E-mail: [email protected])

Received 8 January 2001; Accepted 25 January 2001

Key words: combinatorial chemistry, compound collections, screening, virtual screening

Summary Different strategies for the discovery of novel leads interacting with therapeutically relevant targets are thoroughly presented and discussed, using also three recent examples. Emphasis is given to approaches which do not require extensive resources and budgets, but rather prove how cleverness and creativity can provide active compounds in drug discovery.

Introduction

The extensive use of combinatorial chemistry (CC) and high-throughput screening (HTS) for pharma applications has been growing exponentially in the last years, and some excellent recent reviews covered these topics [1–5]. This short review intends to provide the reader with a critical evaluation of different successful approaches to the discovery of novel and patentable hits [6] which make use of either one or both CC and HTS, with particular focus on the complementarities of these approaches and on the resources (money and scientists) required to correctly apply these approaches to any drug discovery project. Rather than creating another theoretical treaty about hypothetical projects and situations (which seldom occur in the real world . . .), three specific examples by scientists from Lilly and NPS Allelix [7], from The Mayo Clinic [8] and from The University of California [9] were selected to illustrate three of the above approaches also with experimental details. Their extensive illustration will hopefully show their importance as excellent methods for the identification of hits for specific targets and projects and for their further optimization to leads (or lead classes) with potential as actual drug candidates.

The first test case is an example of ‘classical’ HTS – hit optimization, as it is run in major pharma companies; although extensive resources are clearly used, an unambiguous and flexible philosophy of work deserves to be described and commented here. The second case takes advantage of the so-called ‘virtual screening’, i.e. a computational screening approach aimed at the ‘in silico’ discovery of hit structures to be subsequently optimized with little or no chemical and biological efforts. The third case is an excellent example of the use of existing and the design of novel combinatorial-derived chemical diversity for medium–small screening campaigns related to tropical diseases. This is of particular interest for many developing countries; the example stresses the creativity and the cleverness of the academic groups involved, who have repeatedly recycled a common biased library in multiple drug discovery projects, precluding the use of expensive technologies and/or instrumentations.

Random screening of large proprietary collections: When do we become rational human beings? Among the many examples appearing in the literat-

76 ure, we have selected a very recent communication from Eli Lilly and NPS Allelix [7], which deals with the discovery and the optimization of potentiators of 2-amino-3-(5-methyl-3-hydroxy-isoxazol-4yl)propanoic acid (AMPA) receptors, a subtype class of excitatory amino acid (EAA) receptors. This potentiation provokes an enhancement of ion influx through AMPA receptors, and a certain number of positive effects in animals have been observed: examples are nootropic effects [10] and improved performances in learning and memory tasks [11,12]. Some compounds acting as AMPA potentiators were already known at the time of this work (compounds 1–5, Figure 1) [13– 16], but the available data regarding their activity in human studies [17,18] highlighted the need for novel potentiators with greater in vitro potency and better bioavailability. The interest for this target, mainly as a potential source of treatments for cognitive disorders, prompted Lilly towards the search for novel potentiators, possibly belonging to novel, patentable chemical classes; in order to do that, two main things were needed: • first, a large chemical collection of drug-like molecules to be screened for leads, and • second, a reliable and sensitive HTS assay to be used for the screening of the above mentioned collection. If both, or even one of the above two points is missing, a project aimed at the discovery of novel hits by random screening of unbiased, drug-like molecules cannot be commenced. As for Eli Lilly, their chemical collection for screening purposes is surely large and focused on drug-like molecules; this is obtained by applying structural (molecular weight between 250 and 600 or similar, a limited number of rotatable bonds, hydrogen bonds, etc.) and physico-chemical (partition coefficient between 2 and 4, etc.) filters to discard compounds which do not satisfy all, or most of the filter requirements. Any structure showing up as active from an HTS campaign (typically 100 000 to 500 000 compounds tested) will thus become a new hit to be optimized. As previously mentioned, of equal importance is the setup of a biological assay which measures the desired effect (AMPA potentiation) reliably and with an easy detection method, which is robust and thus suitable for automation to become an HTS. NPS Allelix developed proprietary technology and expertise

for such an assay [19], based on cloned human GluR4 receptors expressed in stable HEK-293 cells, and the partnership between the two companies allowed both to benefit from each other’s assets and to be able to start an HTS campaign for the search of AMPA potentiators. The campaign produced a novel hit (6, Figure 2, top), characterized by the presence of a biphenyl fragment, by a fluorine substituent on one of the rings and by the presence of a methyl sulfonamido group β with respect to the other phenyl group. The potency of compound 6 in the assay used for HTS was slightly worse than the best known AMPA potentiator (4, cyclothiazide, Figure 2, top), but still extremely encouraging for the start of a chemical optimization program. The chemical optimization was performed modifying three main areas: (1.) Modification of the sulfonamido substituent (R1 , Figure 2, bottom); (2.) Modification of the α-methyl substituent (R2 , Figure 2, bottom); (3.) Modification of the o-fluorophenyl, both with other substituents on the ring and by replacement of the ring with heterocycles (Ar, Figure 2, bottom). The goal was clearly both to identify more potent compounds and to gather a preliminary SAR to guide further, more focused chemistry efforts. At first, compounds 9a–h bearing various sulfonamido groups were prepared following the simple route depicted in Figure 3. The replacement of methyl with ethyl (9b), isopropyl (9d) or with dimethylamine (isostere of isopropyl, 9h) produced compounds with similar activity to cyclothiazide, thus increasing 3 times the potency of the original lead 6; the isopropyl moiety seemed an optimal choice, as both n-propyl and n-butyl gave significantly less potent compounds (as did bulkier R1 groups). The modification of the α-methyl group was then performed, but a simpler model starting material was considered (10, Figure 4) to avoid complications during the synthetic route depicted in Figure 4. As a result of previous work, the isopropylsulfonamide group was selected as R1 substituent. Despite the simplification in the biaryl structure, significant activity was seen both with α-methyl and ethyl derivatives, hinting at a possible radical change

77

Figure 1. Structures of compounds 1–5.

of the second phenyl group into an alkyl derivative such as t-butyl. The introduction of bulkier groups in the α position was probably not accepted by the target for sterical reasons. Finally, changes in the o-fluorophenyl moiety were evaluated through the synthesis of p-biaryls 12a–h, as was the entire replacement with thiophene as in 13. Both routes to these compounds, containing R1 = iPr and R2 = Me according to the above, are reported in Figure 5. Significant improvements were obtained by the pattern variation on the phenyl ring; groups with different electronic properties (NH2 , CHO, CN, Me) gave excellent results, while others (CF3 , Cl, H, COOH) turned out to be significantly less active. The strong potency of 13 must also be highlighted, as it hints at the replacement of the phenyl ring with heterocycles as an appealing option. In conclusion, the synthesis of around 20 carefully selected compounds has given substantial SAR information and for the best derivative (12h) an increased potency of 150 times versus the original hit 6 was obtained. We can obviously assume that a much larger exploration of the chemical class was performed both before and after the preparation of this communication, and we are surely bound to see in future a more extensive report regarding the SAR and additional properties of these compounds.

Virtual screening of available chemical databases: Cheap, fast and even successful! It often happens that, although the discovery of novel hits would be highly beneficial or even necessary for a given project, either the large chemical collection required or the HTS robust assay (or both) are not available in the research group; this is especially the case for academic groups, among which some have recently devised creative solutions that allowed them to successfully identify valuable hits with significantly less efforts and resources. As an example, The Mayo Clinic [8] uses the socalled ‘virtual screening’ approach, which is almost solely based on computational techniques. It consists of docking iteratively each representative of a large compound database into the active site of an interesting target, thus selecting only a small compound subset for which the theoretical binding affinity is higher than a defined value; this prevents the need for a large compound collection (chemical databases containing hundreds of thousands, or even millions of compounds are available to the public [20–23]) and does not require an HTS assay, as the small compound subset resulting from virtual screening of databases, after being purchased, can be tested with low throughput assays due to its limited size. The target of interest for this work was a zinc metalloprotein, farnesyltransferase (FT), which is well characterized and for which the crystal structure is known [24]. The relevance of FT as a therapeutic target for cancer has been thoroughly validated, as

78

Figure 2. Structures of leads active on PlmII (6 and 4) and of the synthetic targets of the lead optimization effort.

several selective and non-toxic inhibitors are currently undergoing clinical trials [25,26]. The discovery of novel, patentable and potent FT inhibitors would thus be of interest to exploit their potential as safe anticancer drugs. The researchers at Mayo used as a virtual collection the 1998 version of ACD (Available Chemicals Directory, available on the Web [27]), containing 219 930 commercially available chemicals. As this cannot be defined as a drug-like biased collection, they first introduced some filters to discard unwanted entities before starting the real screening, and also to bias the collection towards FT; the whole strategy starting from the ACD collection is reported in Figure 6. The so-called ‘chemical property filter’ retained only those compounds with 300
software program (EUDOC) written by the authors; this ‘loose’ screening was supposed to discard the vast majority of compounds and to retain only those providing a <–35 kcal/mol binding interaction with FT active site. The 1186 survivors were then submitted to a more stringent virtual screening, from which only 313 structures with a <–45 kcal/mol binding were identified (Figure 6). The use of a first crude screen allowed a reduction in the time needed to discard clear ‘no-match’ compounds (>65 000), while the detailed evaluation of the smaller set to refine it with a high resolution docking was operatively feasible. The next computational filter applied related to the solvation potential of compounds, i.e. compounds too hydrophobic or too hydrophilic to access the target were discarded, and solvation was considered also to calculate the binding interactions for the survivors (128, <33 kcal/mol, Figure 6). The 128 compounds were then visually inspected by chemists, who discarded compounds containing reactive/undesired functionalities (thiols, azides, aldehydes, etc.) but also compounds whose commercial availability was not confirmed. This is a highly subjective step, but nevertheless an extremely important one, as to progress a compound containing unstable/reactive groups almost invariably leads to later discovery of either aspecificity or even significant toxicity. Finally, the 27-compound set (almost 10 000-fold reduction of the initial database size) was purchased, and its purity was tested (Figure 6); not very surprisingly, a relevant percentage of these compounds did

79

Figure 3. Synthesis and structures of sulfonamido analogues 9a–h.

not show the requested purity and was discarded to avoid the generation of non-interpretable data from screening with the biological target. Twenty-one compounds were screened on a commercially available FT in vitro assay kit, going up to 500 µM concentrations for the potential inhibitors; the screening (see also Figure 7) produced 4 compounds with good potency (<100 µM inhibition, 19% success rate) and 18 compounds showing at least some inhibition at 500 µM (86% success rate). The structures of compounds 14–17 are shown in Figure 7. The most potent, compound 14, was also found to be active in an assay based on human lung cancer cells, thus proving its ability to penetrate cell membranes. These data and the in vitro potency, higher than the most potent FT inhibitor known (kurasoin A [28], IC50 = 59 mM), make this compound (and its class) very interesting for a lead optimization program. As a final comment, we should say that the reliability and the value of virtual screening on a chemical database is surely inferior to those for an HTS performed on the same collection of physically existing samples; surely several actives are missed during the process, and the theoretical ranking of potencies is often not correct. It is clear, though, that the extreme simplicity, the low cost and the likely occurrence of hit structures to be further optimized make this approach extremely appealing, provided that the necessary computational competences and equipment are available.

Biased and focused combinatorial libraries: Cheap, fast and even successful! Large chemical collections in major pharma companies are normally composed of either proprietary compounds historically prepared during the company’s existence, or of acquired single compounds or subsets coming from vendors. These collections are carefully monitored, and new compounds are added to increase their diversity content but also to bias them towards a more or less defined ‘drug-like’ space by using computational selection methods. Needless to say that the cost required to create such a collection ex novo (typically several hundred thousands of compounds) is prohibitive, and can be afforded only with major investments precluded to small academic or industrial groups. Another, cheaper way to generate significant chemical diversity is represented by combinatorial chemistry. The use of well-assessed chemical routes to multifunctional core molecules, either in solution or on solid phase, and their decoration with various commercially available monomers produces even large libraries of compounds with limited efforts; if the core molecule possesses three decoration points, and 10 commercially available, diverse monomers are used as diversity sets in each position, a total of 1000 diverse molecules are prepared and can be used for screening (Figure 8, top). The same is true for routes where the core entity of the library is assembled by coupling the three diversity sets, rather than being

80

Figure 4. Structures and synthesis of α-methyl modified analogues 11a–f.

Figure 5. Structure and synthesis of o-F phenyl modified analogues 12a–h, 13.

81

Figure 6. Virtual screening of the ACD collection for farnesyltransferase activity.

pre-existent as before (Figure 8, bottom). Both these strategies require only a robust chemical route and 30 commercial compounds to generate diversity; the same route, when the number of monomers is expanded, may lead even to very large libraries and thus to large screening sets for biological assays. The main disadvantage of large primary libraries as sources for novel hits when compared to compound collections is that, while the latter ones are constructed by assembling individual compounds which are also chosen to increase chemical diversity in the collection, the former ones consist of a common structural scaffold which is decorated with monomers, and thus often they span the so-called ‘chemical space’ significantly less than a collection of individuals with the same size.

The obvious conclusion is that, although large collections in major pharma companies also contain several compounds from combinatorial libraries, the vast majority of individuals come from targeted acquisition of individual compounds, and such a collection is ideal for primary screening. The picture changes when we consider biased screening, i.e. the screening of compound sets biased towards certain target classes. These biased sets must contain some common features, and libraries are often an excellent tool to produce these medium–small sets (typically from a few thousands to several tens of thousands individuals). Examples of biased libraries reported in literature include kinases [29,30], proteases [31,32], and G-protein coupled receptors [33]; they all proved to be successful as sources of novel hits, which could also be optimized by making subsequent smaller arrays or individual compounds focused on the hit structure. The example reported here [9], from one of the leading academic groups in the field of combinatorial chemistry, deals with two aspartyl proteases, plasmepsin I and II (Plm I and II), which are appealing and under-exploited targets for antimalarial drugs. The same academic group reported earlier [34] the synthesis and screening of a 1000-member discrete library and of the 39-member focused array derived from it (L1 and L2, Figure 9, top) designed to interact with the active site of cathepsin D (Cat D), another aspartyl protease of human origin; as Cat D shows a high degree of homology with Plm I and II, the decision to test L1 and L2 on at least one of these targets was straightforward. As the crystal structure of Plm II is available [35] and this protease, but not Plm I at the time of this work, could be easily obtained even in large quantities for the setup and run of HTS campaigns, the screening campaign was planned for Plm II (hoping that the extreme degree of homology of the two Plms could even produce potent dual inhibitors!). The screening of L1 and L2 was performed using a standard fluorescence-based protease assay and at a fixed 1 mM concentration; only 13 of the library components showed more than 50% inhibition of Plm II, and the two most potent ones (18 and 19, Figure 9, bottom) were selected as initial submicromolar hits to be further optimized. It is important to underline how the activities on Plm II and Cat D are going somewhat in parallel, and actually how all compounds were much more potent on Cat D than on Plm II; as both 18 and 19 were originally from the focused library L2, they are particularly active (low nM) on cathepsin D. Thus,

82

Figure 7. Real screening on farnesyltransferase: results and structure of active leads 14–17.

Figure 8. The combinatorial strength: diversity through decoration (top) or assembly (bottom) of scaffolds.

83

Figure 9. Structure of Cat D-focused libraries L1, L2 and of PlmII hits 18, 19.

the main objectives of the following optimization have been to increase the in vitro potency, to reach a reasonable selectivity for Plm II and to identify ‘drug-like’ lead compounds to be progressed further. The first round of focused optimization libraries was intended to expand separately the SAR information for substituents R1 –R3 , thus three SP arrays of compounds (L3–L5, Figure 10) bearing similar R1 – R3 substituents to those present in 18 and 19 were prepared following the synthetic route previously assessed by the same research group [36] (Figure 11), which supports on PS the functionalized scaffold 20 [37,38] and then by means of four steps transforms it in the desired peptidomimetics; final acidic release of the THP linker produces the free hydroxyethylamine isosteres 22. L3 was prepared first, using 74 monomers (isocyanates, isothiocyanates or carboxylic acids), and screened; among the individuals, compound 23 (Figure 10) was the best and produced a 2–3 times improvement in comparison with 18 and 19. This side

chain was then selected as a substituent for libraries L4 and L5. L4 was prepared then from 54 amines and screened; none of the new R2 groups showed improved activity (Figure 10), thus the native R2 substituent was kept for the synthesis of library L5. Finally, L5 was prepared using 44 monomers (isocyanates, isothiocyanates or carboxylic acids) and screened; the monomer set used for the synthesis of L3 was reduced in size because several of the original monomers would not stand the reaction conditions leading to the final compounds (i.e., the SnCl2 mediated reduction of azido group 21, Figure 11). Among the library individuals, compound 24 (Figure 10) represented the first breakthrough, as its potency was increased 40 times with respect to 18 and 19, and in vitro activities on Cat D and Plm II for the first time appeared comparable. As a second optimization cycle, the simultaneous variation of R1 and R3 was examined in the 80-member library L6 (Figure 12); probably these positions were preferred to R2 as the library L4 did not

84

Figure 10. Structure of PlmII-focused libraries L3-L5 and of leads 23, 24.

85

Figure 11. Synthetic strategy to libraries L1–L7.

show any improvement, thus the R2 substituent was deemed essential as it was and not pursued further. Among the criteria chosen to select the monomers, preference was given to low MW monomers to try to reduce the MW of the compounds (e.g., compound 24 has a MW of 752 that could be too much to allow its penetration through biological membranes). Several compounds showed interesting inhibitory activities, and among them compound 25 (Figure 12) showed a slightly better Plm II inhibition, although the selectivity towards Cat D was worse than that of 24; its MW, though, was over 100 Da less than 24. The last optimization round was made by a single, all-mutated-residues library L7 (Figure 13), containing 320 compounds and prepared using a slightly revised SP strategy [32] that allowed to introduce limited diversity also at the R8 position. This library included 10 monomers at R1 and 8 at R3 , but also 2 at R2 , using 4-methoxy phenethyl as a smaller analogue of the best R2 substituent, and 2 at R8 , using the hydroxyethyl isostere of leucine as a smaller replacement

of the usual phenylalanine isostere (Figure 13); again, these two changes aimed to reduce the overall MW of novel Plm II inhibitors. The library produced highly active compounds, such as 26 and 27, but especially of importance is a first significant observed selectivity versus Cat D, as in 28 and 29 (Figure 13). These last compounds were then characterized more in detail, confirming all the good preliminary results but adding also the lack of inactivation in vivo by serum binding and the even slightly stronger inhibition of Plm I for some of them (compounds 26–28). It is impressive how a library tailored on a specific enzyme (L1, Cat D) could also be used to detect initial hits on another member of the family of aspartyl proteases, and how a small optimization program (five libraries, around 600 compounds) allowed to increase in vitro inhibition up to 100 times (compare 19 and 27), and to increase the selectivity factor versus Cat D up to 200 times (compare 19 and 28). In our opinion, this is a clear example of how biased libraries can be an

86

Figure 12. Structure of PlmII-focused library Lg and of the optimized lead 25.

excellent and extremely versatile source of novel hits, and thus of drug discovery projects, for large classes of therapeutically relevant targets, provided that the chemistry assessment of the selected synthetic route confirms its robustness and allows the use of highly diverse monomers in the hit discovery/optimization phases. Conclusions and outlook The use of large (≈1 000 000 compounds or more), diverse compound collections and of robust and automated HTS assays is surely the most complete and powerful approach for the discovery of novel hits; this is especially true for non-characterized targets deriving from genomics-based target identification, or for known families of receptors for which screening campaigns have not been extremely successful (i.e., inhibitors of protein–protein interactions). The creation and the use of large compound collections, though, requires among other necessities: • the access to various sources of chemical diversity, such as individual compounds, small arrays, natural products, etc.;

• large expenses to maintain, or even increase the level of representativity of the collection via the computational selection of individual compounds or small arrays and their synthesis or purchase; • expensive facilities and instrumentation for the storage and handling of the compound collection; • expensive facilities and instrumentation for the generation of HTS sets and for their automated screening; • the access to databases and software to electronically control and direct all of the above during an HTS campaign; • the periodical quality control of the collection samples to maintain its high purity and reliability. The frequent run of HTS campaigns (say, several tens per year) is enough to substantiate the expense necessary to create and support the above infrastructure, both in terms of money and resources; this is the reason why all major pharma companies have these assets. It is nevertheless important to increase the level of rationality in selecting compound sets for an HTS campaign, to increase the success rate of each campaign, to reduce the size of a screening set while keeping its diversity content, and to

87

Figure 13. Structure of PlmII-focused library L7 and of optimized (26, 27) and selective (28, 29) leads.

reduce the consumption rate of the collection itself; techniques and technologies such as computational chemistry, miniaturization, informatics and automation will play a major role in switching from random screening campaigns to knowledge-driven screening. Virtual screening campaigns or biased combinatorial libraries represent useful tools to increase the information available to plan actual HTS campaigns. Even more importantly, they allow to obtain the needed outcome (one or more valuable hits) with much less effort and they open the field of hit discov-

ery to smaller, but creative players as was shown in the previous examples. The major mistake that could be made, is to be really convinced that there is only one good way to run screening campaigns, and to access the right chemical diversity to be the source of novel hits. Many things are still at least confused in this area, starting from real and effective metrics to prove unequivocally the effectiveness of different approaches. It is clear, though, to all skilled scientists in the field that this ‘Holy Grail’ approach does not exist and that each

88 target, or assay, or set of compounds will influence strongly the progress of the hit discovery project. Flexibility and creativity, thus, are and increasingly will be the keywords for the success of players in this area where novel targets are being generated continuously, and where gold mines are surely still awaiting to be discovered and exploited.

15.

16.

References 17. 1. Dolle, R.E., Comprehensive survey of combinatorial library synthesis: 1999, J. Comb. Chem., 2 (2000) 383–433. 2. Weber, L., High-diversity combinatorial libraries, Curr. Opin. Chem. Biol., 4 (2000) 295–302. 3. Valler, M. and Green, D., Diversity screening versus focussed screening in drug discovery, Drug Discov. Today, 5 (2000) 286–293. 4. Herzberg, R.P. and Pope, A.J., High-throughput screening: New technology for the 21st century, Curr. Opin. Chem. Biol., 4 (2000) 445–451. 5. Bosse, R., Illy, C., Elands, J. and Chelsky, D., Highthroughput screening in the new millennium, Drug Discov. Today HTS Suppl., 1 (2000) 42–47. 6. The term hit in this review defines a structurally defined chemical entity with good activity in the in vitro assay used during the screening campaign and with general adherence to the accepted criteria to define a drug-like molecule (Lipinsky’s ‘rule of five’). 7. Ornstein, P.L., Zimmerman, D.M., Arnold, M.B., Bleisch, T.J., Cantrell, B., Simon, R., Zarrinmayeh, H., Baker, S.R., Gates, M., Tizzano, J.P., Bleakman, D., Mandelzys, A., Jarvie, K.R., Ho, K., Deverill, M. and Kamboj, R.K., Biarylpropylsulfonamides as novel, potent potentiators of 2-amino3-(5-methyl-3-hydroxyisoxazol-4-yl)-propanoic acid (AMPA) receptors, J. Med. Chem., 43 (2000) 4354–4358. 8. Perola, E., Xu, K., Kollmeyer, T.M., Kaufmann, S.H., Prendergast, F.G. and Pang, Y.-P., Successful virtual screening of a chemical database for farnesyltransferase inhibitor leads, J. Med. Chem., 43 (2000) 401–408. 9. Haque, T.S., Skillman, G.A., Lee, C.E., Habashita, H., Gluzman, I.Y., Ewing, T.J.A., Goldberg, D.E., Kuntz, I.D. and Ellman, J.A., Potent, low-molecular-weight non-peptide inhibitors of malarial aspartyl protease plasmepsin II, J. Med. Chem., 42 (1999) 1428–1440. 10. Lee, C.R. and Benfield, P., Aniracetam: An overview of its pharmacodynamic and pharmacokinetic properties and a review of its therapeutic potential in senile cognitive disorders, Drugs Aging, 4 (1994) 257–273. 11. Staubli, U., Rogers, G. and Lynch, G., Facilitation of glutamate receptors enhances memory, Proc. Natl. Acad. Sci. USA, 91 (1994) 777–781. 12. Thompson, D.M., Guidotti, A., Di Bella, M. and Costa, E., 7Chloro-3-methyl-3,4-dihydro-2H-1,2,4-benzothiadiazine S,Sdioxide (IDRA 21), a congener of aniracetam, potently abates pharmacologically induced cognitive impairments in patas monkeys, Proc. Natl. Acad. Sci. USA, 92 (1995) 7667–7671. 13. Gouliaev, A.H. and Senning, A., Piracetam and other structurally related nootropics, Brain Res. Rev., 19 (1994) 180– 222. 14. Yamada, K.A. and Tang, C.-M., Benzothiadiazides inhibit rapid glutamate receptor desensitization and enhance

18.

19.

20. 21. 22. 23. 24.

25.

26. 27. 28.

29.

30.

glutamatergic synaptic currents, J. Neurosci., 13 (1993) 3904– 3915. Bertolino, M., Baraldi, M., Parenti, M., Braghiroli, D., Di Bella, M., Vicini, S. and Costa, E., Modulation of AMPA/kainite receptors by analogues of diazoxide and cyclothiazide in thin slices of rat hippocampus, Recept. Channels, 1 (1993) 267–278. Granger, R., Staubli, U., Davis, M., Perez, Y., Nilsson, L., Rogers, G.A. and Lynch, G.A., A drug that facilitates glutamatergic transmission reduces exploratory activity and improves performance in a learning-dependent task, Synapse, 15 (1993) 326–329. Lynch, G., Kessler, M., Rogers, G., Ambros-Ingerson, J., Granger, R. and Schehr, R.S., Psychological effects of a drug that facilitates brain AMPA receptors, Int. Clin. Psychopharmacol., 11 (1996) 13–19. Lynch, G., Granger, R., Ambros-Ingerson, J., Davis, C.M., Kessler, M. and Schehr, R.S., Evidence that a positive modulator of AMPA-type glutamate receptors improves delayed recall in aged humans, Exp. Neurol., 145 (1997) 89–92. Fletcher, E.J., Nutt, S.L., Hoo, K.H., Elliott, C.E., Korczak, B., McWhinnie, E.A. and Kamboj, R.K., Cloning, expression and pharmacological characterization of a human glutamate receptor: hGluR4, Recept. Channels, 3 (1995) 21–31. Cambridge Structural Database, 202 146 compounds. http://www.ccdc.cam.ac.uk. MACCS-II Drug Data Report, 113 842 compounds. http://www.mdli.com/dats/pharmdb.html. National Cancer Institute 3D-Database, 126 554 compounds. http://dtp.nci.nih.gov. Available Chemicals Directory – Screening Compounds, 1 280 658 compounds. http://www.mdli.com. Park, H.W., Boduluri, S.R., Moomaw, J.F., Casey, P.J. and Beese, L.S., Crystal structure of protein farnesyltransferase at 2.25 Angstroems resolution, Science, 275 (1997) 1800–1804. Njoroge, F.G., Taveras, A.G., Kelly, J., Remiszewski, S., Mallams, A.K., Wolin, R., Afonso, A., Cooper, A.B., Rane, D.F., Liu, Y.T., Wong, J., Vibulbhan, B., Pinto, P., Deskus, J., Alvarez, C.S., del Rosario, J., Connolly, M., Wang, J., Desai, J., Rossman, R.R., Bishop, W.R., Patton, R., Wang, L., Kirschmeier, P., Bryant, M.S., Nomeir, A.A., Lin, C.-C., Liu, M., McPhail, A.T., Doll, R.J., Girijavallabhan, V.M. and Ganguly, A., (+)-4-[2-[4-(8-Chloro-3,10-dibromo6,11-dihydro-5H-benzo[5,6]-cyclohepta[1,2-b]-pyridin-11yl)-1-piperidinyl]-2-oxo-ethyl]-1-piperidinecarboxamide (SCH-66336): A very potent farnesyl protein transferase inhibitor as a novel antitumor agent, J. Med. Chem., 41 (1998) 4890–4902. Rawls, R.L., Puzzling promise of protein prenylation, Chem. Eng. News, (1998) 67–69. Available Chemicals Directory, 278 093 compounds. http://www.mdli.com. Sunazuka, T., Hirose, T., Tian, Z. M., Uchida, R., Shiomi, K., Harigaya, Y. and Omura, S., Syntheses and absolute structures of novel protein farnesyltransferase inhibitors, kurasoins A and B, J. Antibiot., 50 (1997) 453–455. Norman, T.C., Gray, N.S., Koh, J.T. and Schultz, P.G., A structure-based library approach to kinase inhibitors, J. Am. Chem. Soc., 118 (1996) 7430–7431. Gray, N.S., Wodicka, L., Thunnissen, A.-M.W.H., Norman, T.C., Kwon, S., Espinoza, F.H., Morgan, D.O., Barnes, G., LeClerc, S., Meijer, L., Kim, S.-H., Lockhart, D.J. and Schultz, P.G., Exploiting chemical libraries, structure, and genomics in the search for kinase inhibitors, Science, 281 (1998) 533–538.

89 31.

32.

33.

34.

Lee, C.E., Kick, E.K. and Ellman, J.A., General solid-phase synthesis approach to prepare mechanism-based aspartyl protease inhibitor libraries. Identification of potent cathepsin D inhibitors, J. Am. Chem. Soc., 120 (1998) 9735–9747. Gehlhaar, D.K., Bouzida, D. and Rejto, P.A., Reduced dimensionality in ligand-protein structure prediction: Covalent inhibitors of serine proteases and design of site-directed combinatorial libraries, ACS Symp. Ser., 719 (1999) 292–311. Mason, J.S., Morize, I., Menard, P.R., Cheney, D.L., Hulme, C. and Labaudiniere, R.F., New 4-point pharmacophore method for molecular similarity and diversity applications: Overview of the method and applications, including a novel approach to the design of combinatorial libraries containing privileged structures, J. Med. Chem., 42 (1999) 3251–3264. Kick, E.K., Roe, D.C., Skillman, A.G., Liu, G., Ewing, T.J.A., Sun, Y., Kuntz, I.D. and Ellman, J.A., Structure-based design and combinatorial chemistry yield low nanomolar inhibitors of cathepsin D, Chem. Biol., 4 (1997) 297–307.

35. Silva, A.M., Lee, A.Y., Gulnik, S.V., Majer, P., Collins, J., Bhat, T.N., Collins, P.J., Cachau, R.E., Luker, K.E., Gluzman, I.Y., Francis, S.E., Oksman, A., Goldberg, D.E. and Erickson, J.W., Structure and inhibition of plasmepsin II, a hemoglobindegrading enzyme from Plasmodium falciparum, Proc. Natl. Acad. Sci. USA, 93 (1996) 10034–10039. 36. Kick, E.K and Ellman, J.A., Expedient method for the solidphase synthesis of aspartic acid protease inhibitors directed toward the generation of libraries, J. Med. Chem., 38 (1995) 1427–1430. 37. Ghosh, A.K., McKee, Sp.P., Lee, H.Y. and Thompson, W.J.A., A facile and enantiospecific synthesis of 2(S)- and 2(R)-[1 (S)azido-2-phenethyl]oxirane, J. Chem. Soc., Chem. Commun., (1992) 273–274. 38. Caron, M., Carlier, P.R. and Sharpless, K.B., Regioselective azide opening of 2,3-epoxy alcohols by [Ti(O-i-Pr)2 (N3 )2 ]: Synthesis of α-amino acids, J. Org. Chem., 53 (1988) 5185– 5187.