Non combinatorial chemistry
1 like1,084 views
1) Non-combinatorial chemistry allows for the design, synthesis, purification and screening of individual compounds selected from a virtual library, rather than all combinations of reagents like in traditional combinatorial chemistry. 2) The authors developed a system called LiCRA that harnesses efficient non-combinatorial chemistry design and high-throughput automated purification in a closed loop for hit-to-lead optimization. 3) Quality is a priority in their non-combinatorial approach, from compound design through product analysis and registration.
![DDT Vol. 6, No. 15 August 2001 research focus reviews
involved in the purification process.
There are eight semi-preparative HPLC
systems [Gilson 306 gradient systems,
with Gilson 215 liquid handlers config-
ured as both autosamplers and fraction
collectors, operating under Unipoint
control software (Gilson, Middleton,
WI, USA)]. The HPLCs all use simple
generic gradients to effect separation
and purification (acetonitrile–0.1% tri-
fluoroacetic-acid gradients running
from 5–95% organic over 8 min at 4 ml
min–1 with a total run time of 15 min).
Each 96-well compound plate has a
barcode generated from LiCRA linked
to database records describing the 96
expected compounds and assigning an
empirical formula to each well. Each of
Figure 3. A view of part of the autopurification system: the linear track ORCA® robot
the 96 compounds is fractionated by
plate-handler (Beckman Coulter, Fullerton, CA, USA) is housed in a recirculating safety
gradient HPLC into bar-coded 48-well cabinet. The MS and evaporative light scattering unit are situated below the bench on
plates, collecting up to five fractions the right-hand side. Some of the eight semi-prep HPLCs can be seen in the
per compound using UV-triggered frac- background.
tion collection. The 48-well plates are
manually loaded onto a carousel and
an ORCA® robot then automatically (a) (b)
transfers the plate to a Gilson 231 auto-
sampler for analysis (Fig. 3). Each
fraction well is analyzed by flow
injection MS (Micromass® Platform
LC, Micromass® UK, Manchester, UK)
that operates in parallel with evap-
orative light scattering (Sedex Sedere
55, Thomson Instrument Company,
Chantilly, VA, USA). The MS identifies
the fraction or fractions with the (c) (d)
expected empirical formula for each
of the 96 compounds throughout
the series of 48-well fraction plates.
The evaporative light-scattering data
provides an estimate of the amount
of compound present in the fraction
by reference to a calibration file8. The
correctly identified fractions that
have sufficient sample mass are then
Figure 4. (a) A plate view from the LiCRA system, illustrating the connectivity between
automatically selected and compressed
the 96-well plate of products and all of the associated structural, analytical and synthetic
into a bar-coded plate of tared 5 ml information. The colour coding illustrates success (green) or failure (red) of the
Micronics tubes to be evaporated to chemistry at the well level. (b) A typical dual wavelength UV chromatogram of the crude
dryness (Genevac, Ipswich, UK). After product from a library synthesis. (c) Positive- and negative-ion electrospray MS of the
desired fraction (fraction 61, vial 13) from the purification of this crude product. (d) A
autopurification, all the MS and chro- typical 1[H] NMR showing the quality of compounds obtained using this autopurification
matographic data on any compound system.
can be accessed from within LiCRA.
783](https://image.slidesharecdn.com/noncombinatorialchemistry-110918111138-phpapp02/85/Non-combinatorial-chemistry-5-320.jpg)
Non combinatorial chemistry
- 1. DDT Vol. 6, No. 15 August 2001 research focus reviews The application of non-combinatorial chemistry to lead discovery Jeremy Everett, Mark Gardner, Frank Pullen, Graham F. Smith, Mike Snarey and Nick Terrett Non-combinatorial chemistry is a powerful technology for the synthesis combinatorial strategies1–4. There was no of large numbers of compounds, with complete control over the properties doubt that the new technology and the ready availability of automated synthesis offered of those compounds. We have developed a Library Creation, Registration opportunities for increased productivity in and Automation system (LiCRA), which harnesses an efficient non- both file enrichment and lead optimization. combinatorial chemistry design and synthesis engine, together with However, it was a certainty that chemistry high-throughput automated purification. This LiCRA system also operates would not permit easy success when there was such a casual disregard of the basic tenets in a closed loop mode for hit-to-lead optimization, and contains an of good science – careful experimental plan- integrated IT system that controls and facilitates all aspects of the ning and meticulous observation of results. operation from design to registration. Quality has been our watchword, Indeed, we commented in 1995 that in the fu- from the quality of compound design through to the quality of the ture of drug discovery through combinatorial chemistry, ‘quality, rather than quantity, will products. become the new goal’5. Thus, library synthesis reverted towards the w The launch of combinatorial chemistry old values of quality, both in compound *Jeremy Everett Mark Gardner onto an unsuspecting pharmaceutical indus- design and product analysis. Although this Frank Pullen try in the early 1990s resulted in several fran- inevitably resulted in fewer compounds being Graham F. Smith tic efforts as companies tried to maintain a prepared overall, we have now reached a stage Mike Snarey Nick Terrett competitive edge through the generation and where those that are made are the products of Medicinal Technologies screening of compounds in unprecedented more considered design, meticulous protocol Pfizer Global R&D numbers and at an unprecedented rate. The development and thorough structural analy- Sandwich focus at that time was solid-phase synthesis sis. Medicinal chemists who exploit high-speed Kent, UK CT13 9NJ and the generation of compound mixtures, chemical techniques are increasingly making *tel: +44 (0)1304 616161 resulting in library arrays of many millions of single compounds, depending primarily on fax: +44 (0)1304 656259 compounds. These combinatorial libraries re- the proven legacy of solution-phase chem- e-mail: jeremy_everett@ sandwich.pfizer.com quired considerable analytical detective work istry. The products are often purified, or sub- to ascertain whether the desired compounds jected to rigorous analysis, before being sent had been made. Frequently, the library ap- for registration and screening. proaches were frustrated by restrictive solid- We have now reached a phase in the devel- phase chemistry of limited success or general- opment and application of high-speed chem- ity. Furthermore, the observation of biological istry that permits library compounds that are activity in assays of mixtures frequently re- active in biological screens to be identified sulted in a wild-goose chase that employed without deconvolution, fractionation or tag sophisticated or tedious deconvolution or decoding, and, more significantly, there are decoding, pursuing hits that often vanished: much fewer false-positives. Because the prep- the false-positive problem. Having suffered aration of targeted single compounds removes several resource-intensive yet fruitless pur- the requirement for the synthesis of combi- suits, in the mid- to late-1990s, many compa- natorial arrays, we refer to our approach as nies were forced to radically rethink their being ‘non-combinatorial’. 1359-6446/01/$ – see front matter ©2001 Elsevier Science Ltd. All rights reserved. PII: S1359-6446(01)01876-1 779
- 2. reviews research focus DDT Vol. 6, No. 15 August 2001 This article describes the non-combinatorial approach developed and used by our Library Design and Production Group, and demonstrates that a combination of rational design, careful synthesis and compound purification be- H2N OH H2N OH H2N OH fore testing can generate high quality biological data that O O O result in novel leads and new drug discovery programmes. Non-combinatorial chemistry O Non-combinatorial chemistry is a phrase coined to express H2N OH H2N OH H2N OH the difference in strategy between our technique and com- O O binatorial chemistry as it is more commonly applied. Drug Discovery Today Combinatorial chemistry involves the use of all combi- Figure 1. Six amino acids. Your task is to choose the two most nations of a set of reagents to produce an ordered array of diverse amino acids that should be selected as monomers for products. For example, a combinatorial amide library syn- file enrichment. This is representative of the problem faced in thesis starting with 80 amines and 60 acids would make combinatorial file enrichment, in which a small number of 4800 amide products (80 × 60). In non-combinatorial monomers have to be selected in an attempt to represent the properties of a much larger set. chemistry, the same reagent sets can be used, but the prod- ucts would be individually chosen for synthesis, such that the 80 amines and 60 acids might be used to select any used to select the representative building blocks. However, number of products from 80 up to the combinatorial limit this is an approach with significant drawbacks; a typical of 4800, and still use each reagent. This approach is also situation that might be faced in selecting 20 amino acids called ‘cherry-picking’ by some groups. The key differential from a set of 100 is presented in Fig. 1. is to allow the design, synthesis, purification and screening As a subset of the whole exercise of choosing 20 from of any group of individual compounds from a virtual 100, two amino acids might have to be chosen from the library, not just arrays of compounds. The drive for this set of six presented in Fig. 1. If you have been involved strategy is the desire to exploit virtual libraries that are as in such an exercise, you will recognize the similarity of large as possible in a powerful and flexible manner, while this situation with examples from your own experience. retaining total control over the properties of each product. Although software packages might be able to make a The approach is important in both file enrichment and selection, the truth is that there is no correct way to lead optimization. makethis choice for a file enrichment library. Whichever pair you choose, they clearly cannot adequately represent Application of non-combinatorial chemistry to file all six. Once you have made your selection of a pair of enrichment amino acids, remember that in this library of 8000 com- Non-combinatorial chemistry is important in file enrich- pounds, each of those amino acids will be present in 400 ment in that it allows the synthesis of diverse sets of com- products (in combination with the 20 selected amines and pounds, using as many monomers as possible, not just a 20 selected carboxylic acids). The other four amino acids small subset. The reason why this is important is best will be present in no products at all. illustrated by an example. Consider a three component An alternative approach is to construct the 8000 library, represented as A–B–C. Let us assume that A are compounds in a non-combinatorial manner, using all the amines, B are amino acids and C are acids, so the products available building blocks. In this example, if each of the will be diamides. It can be easy to acquire 500 amines, 100 components A, B and C is used equally, each amine and amino acids and 500 carboxylic acids from commercial carboxylic acid would be present in 16 (8000/500) prod- sources, giving a virtual library size of 25 million ucts, and each amino acid in 80 (8,000/100) products. compounds. Clearly, it is unlikely that 25 million single Arbitrary choices of monomers no longer need to be made compounds will be synthesized using parallel synthesis. – all can be used in the file enrichment library – thus pro- Therefore, a choice has to be made as to how to exemplify viding much greater diversity of products. The preparation this virtual library (with perhaps 8000 examples) in a for synthesis of such a library does take longer than for a compound collection for screening purposes. A typical combinatorial array (it takes longer to weigh and solubilize approach might be to select 20 amines, 20 amino acids and 1100 components than 60). In addition, automation is 20 carboxylic acids to give a combinatorial array of 8000 now essential for synthesis because neat arrays that could examples. Frequently, diversity-based methods might be be handled manually are no longer being used. 780
- 3. DDT Vol. 6, No. 15 August 2001 research focus reviews A further significant benefit of the 20,000 non-combinatorial approach is that you can choose not to synthesize spe- Count of products 16,000 cific products, and thus precisely tailor the physicochemical properties of the 12,000 non-combinatorial set. For example, it is straightforward to calculate the prop- 8000 erties of the products and to ensure 4000 that all of those to be synthesized 0 should comply with certain distribu- 150 190 230 270 310 350 390 430 470 510 550 590 630 670 110 tions of molecular weight or clogP values, Lipinski’s rule of five6 or other Molecular weight desired properties. An example of the 20,000 effect of this flexibility can be seen in Count of products Fig. 2. 16,000 The standard combinatorial array would give something approaching a 12,000 normal distribution of a property such 8000 as molecular weight. However, with non-combinatorial synthesis, you can 4000 synthesize only those products with 0 150 190 230 270 310 350 390 430 470 510 550 590 630 670 110 molecular weights below a given value or in a given value-range. Our in-house Molecular weight library design software, S1D (selection Drug Discovery Today in one dimension, i.e. by additive or approximately additive fragmental Figure 2. An illustration contrasting property distribution in a combinatorial selection with that in a non-combinatorial selection. Well-defined property cut-offs can be readily properties, such as molecular weight or achieved via non-combinatorial approaches. clogP) allows for the application of these design principles and has been used to select diverse libraries of specific design for file be searched by computational techniques, such as pharma- enrichment for three years. S1D makes pseudo-random cophore searching or crystal structure ‘docking’, and the selections of products from a virtual library in a way that hit-lists synthesized as they are or manipulated further. attempts to use each monomer an equal number of times, but with boundary conditions such that the calculated Non-combinatorial synthesis using LiCRA properties of each product (molecular weight, clogP, and Our first significant problem was that at the time we so on) fall within predetermined limits. S1D links to the started, nobody had considered that non-combinatorial Library Creation, Registration and Automation system synthesis would ever be a user requirement of ‘combina- (LiCRA). Both LiCRA and S1D were written by the torial chemists’, and still many do not. We were drawn to chemoinformatics and Discovery Research informatics our solution by the knowledge that our HTS groups used groups in Pfizer Global R&D, Sandwich, UK. ‘cherry-picking’ software on Tecan liquid handling robots (Tecan UK, Reading, UK) to choose active compounds from Application of non-combinatorial chemistry to lead HTS screening plates. However, no software was available optimization at that time for our library synthesis task, because of our Although non-combinatorial synthesis offers significant unique in-house requirements, and we decided, therefore, benefits in file enrichment, the strategy is particularly pow- to design and build LiCRA ourselves. erful once there are some design principles to apply, such When our group started work in 1997, we used each as when an active compound or a crystal structure is in piece of automation in isolation to try to understand what hand. Non-combinatorial synthesis gives the medicinal inputs and outputs were important for us to monitor and chemist complete freedom to choose any selection of change. In general, most commercial automation runs on single compounds from a given virtual library to expand its own software with minimal opportunities for broader upon this structural information. Virtual libraries can also integration. A flat file can often be read to load sample data 781
- 4. reviews research focus DDT Vol. 6, No. 15 August 2001 and a further file is often written to the workstation to log All chemical incubations are performed off-line in simple automation activity. Our first attempt at synthesis software shakers or ovens, and all products are sent through our simply read flat files from a shared network drive and pro- in-house autopurification system (see later). Samples are duced an input file for the next step of the process. This reformatted into 96-well Micronics plates (Micronics, proved successful and was flexible in the event of machine Redmond, WA, USA) for registration that allows a final errors. User intervention and file manipulation, however, quantitation by weighing to be performed using a Bohdan was causing problems with tracking, and serious human Weighing Station (Mettler-Toledo Myriad, Royston, UK). errors were increasing. The user is then presented with quality-control data on a The latest version of LiCRA is based around an Oracle plate-by-plate basis to allow review of spectra and individual database (Oracle UK, Reading, UK) of the individual com- sample data before registration. pound batches that we have chosen to make. Their non- LiCRA has built-in links to several chemical structure combinatorial selection means that, compared with array databases to allow easy analysis of monomer solubility, re- synthesis, there is much less common data from one batch activity and other properties. Product structures are also in a library to the next. Our main reasons for moving to a stored within the LiCRA database to allow more chemically database-centred system were to increase storage and secu- intuitive quality control. Because all synthesis information rity and to facilitate data analysis and trend recognition. is stored within the database, trends in solubility and reac- Compliance with data storage and retrieval requirements tivity of individual monomers can be tracked to enhance for patent purposes made us look into this more robust the quality of future libraries. When the chemist is satis- approach. fied with the library products, a structure definition file, The LiCRA user deals with a Visual Basic interface, which which contains all relevant information, is produced and treats all the compounds in a library together (e.g. reaction sent to our compound control centre for registration to the temperature, molar equivalents) but tracks most other data corporate database. at the individual compound level (e.g. yield, purity and protecting groups). A simple input interface that requires Automated purification of compounds from non- the product structure in MDL (Molecular Design Ltd, combinatorial synthesis Camberley, UK) Structure definition (SD) file format, The non-combinatorial chemistry approach to file enrich- and an ASCII (American Standard Code for Information ment and lead optimization, as described earlier, generates Interchange) file containing compound numbers with con- large numbers of single compounds, all of which need to stituent monomers, is all that is required to load a library. be purified to minimize the presence of impurities and the Typical libraries range from 200 to 2000 compounds. In generation of false-positives in biological screens. Analysis, planning a synthetic scheme, volumes, equivalents and so purification and quantitation of large numbers of com- on, can be entered for a whole library but can also be pounds in 96-well plate format has been viewed by many altered on a monomer-by-monomer basis. An automatic as a highly labour intensive process. Systems are available reservation can be made against the monomer inventory for high-throughput purification (e.g. the Biotage Parallex system and lists of quantities printed if required. Monomers Flex™ system – Dyax, Charlottesville, VA, USA). However, that appear in the library design but are not present at all, there are currently no commercial purification systems on or are in insufficient quantity in the monomer store, are the market that can effectively process, quantify and track flagged for intervention. The LiCRA inventory control large numbers of compounds in an automated way. The system interfaces directly with accurate balances during key to any automated purification system is a fully inte- weighing and calculates the exact amount of each monomer grated sample tracking process, coupled with proven weighed. The weighed monomer samples are bar-coded purification technology and linked to a user-friendly soft- and loaded into racks for synthesis. LiCRA produces a re- ware interface. A useful review of automated purification port that warns when monomers are below a predetermined methodologies and emerging supercritical fluid chro- re-order level, so that the inventory can be maintained. matography-based methods is available7. The LiCRA synthesis system uses several standard and We have developed an autopurification system based on robust pieces of commercial hardware. All solution dis- gradient HPLC separation, off-line MS analysis of fractions pensing activities are performed by Tecan Genesis liquid and quantification by light scattering that has the capacity handling robots, which have a proprietary interface for to purify up to 100,000 compounds per year at the 10–50 mg non-combinatorial sample distributions. Specific racks and scale. The software interface (OmniTrak, developed by boxes have been custom-made to enable inert atmosphere, Aitken Scientific, Thame, UK) is fully integrated with corrosive reagent and solid-phase chemistry capabilities. LiCRA, and controls all the HPLC, MS and robotic hardware 782
- 5. DDT Vol. 6, No. 15 August 2001 research focus reviews involved in the purification process. There are eight semi-preparative HPLC systems [Gilson 306 gradient systems, with Gilson 215 liquid handlers config- ured as both autosamplers and fraction collectors, operating under Unipoint control software (Gilson, Middleton, WI, USA)]. The HPLCs all use simple generic gradients to effect separation and purification (acetonitrile–0.1% tri- fluoroacetic-acid gradients running from 5–95% organic over 8 min at 4 ml min–1 with a total run time of 15 min). Each 96-well compound plate has a barcode generated from LiCRA linked to database records describing the 96 expected compounds and assigning an empirical formula to each well. Each of Figure 3. A view of part of the autopurification system: the linear track ORCA® robot the 96 compounds is fractionated by plate-handler (Beckman Coulter, Fullerton, CA, USA) is housed in a recirculating safety gradient HPLC into bar-coded 48-well cabinet. The MS and evaporative light scattering unit are situated below the bench on plates, collecting up to five fractions the right-hand side. Some of the eight semi-prep HPLCs can be seen in the per compound using UV-triggered frac- background. tion collection. The 48-well plates are manually loaded onto a carousel and an ORCA® robot then automatically (a) (b) transfers the plate to a Gilson 231 auto- sampler for analysis (Fig. 3). Each fraction well is analyzed by flow injection MS (Micromass® Platform LC, Micromass® UK, Manchester, UK) that operates in parallel with evap- orative light scattering (Sedex Sedere 55, Thomson Instrument Company, Chantilly, VA, USA). The MS identifies the fraction or fractions with the (c) (d) expected empirical formula for each of the 96 compounds throughout the series of 48-well fraction plates. The evaporative light-scattering data provides an estimate of the amount of compound present in the fraction by reference to a calibration file8. The correctly identified fractions that have sufficient sample mass are then Figure 4. (a) A plate view from the LiCRA system, illustrating the connectivity between automatically selected and compressed the 96-well plate of products and all of the associated structural, analytical and synthetic into a bar-coded plate of tared 5 ml information. The colour coding illustrates success (green) or failure (red) of the Micronics tubes to be evaporated to chemistry at the well level. (b) A typical dual wavelength UV chromatogram of the crude dryness (Genevac, Ipswich, UK). After product from a library synthesis. (c) Positive- and negative-ion electrospray MS of the desired fraction (fraction 61, vial 13) from the purification of this crude product. (d) A autopurification, all the MS and chro- typical 1[H] NMR showing the quality of compounds obtained using this autopurification matographic data on any compound system. can be accessed from within LiCRA. 783
- 6. reviews research focus DDT Vol. 6, No. 15 August 2001 Using this technology, library chemists can purify their R2 R1 compounds without involving themselves in the detailed Cl N R5 1) R1R2NH R5 understanding of HPLC or MS, as all the identification, N N analysis and quantification is carried out in an automated 2) R 3R4NH R3 R6 N Cl R6 N N way. Compounds purified in this high-throughput and R4 automated fashion are of comparable quality with those R2 R1 produced by traditional synthesis and purification technol- Cl N ogies; they are typically of 90–95% purity (Fig. 4). All the R5 1) R1R2NH R5 N N compounds are tracked from their synthesis to registration N N 6 6 using a combination of OmniTrak and LiCRA, and all the R 2) R3R4NH R Cl N analytical data for a particular compound are linked to that R4 R3 tracking process. R2 R5 N Cl 1) R1R2NH R5 N N R1 Application of S1D and LiCRA non-combinatorial R3 chemistry to hit-to-lead optimization R6 N Cl 2) R3R4NH R6 N N We have applied the S1D and LiCRA systems to several tar- R4 get types. The example that follows illustrates the power of R1, R2, R3, R4 = H, lower alkyl, cycloalkyl, substituted alkyl, substituted cycloalkyl, heteroaryl, benzyl the technology to optimize some weak hits coming from a kinase screen to generate multiple, potent leads for chem- R5, R6 = H, lower alkyl, cycloalkyl, substituted alkyl, istry follow-up. Five thousand compounds were selected substituted cycloalkyl, heteroaryl, benzyl, fused aryl, fused heteroaryl, fused cycloalkyl using S1D for non-combinatorial synthesis from a virtual Drug Discovery Today library of 300,000 compounds that was designed specifi- Figure 5. The design for some of the diamino-substituted cally for a kinase screen. Figure 5 illustrates the type of aromatic heterocycles used in the kinase virtual library. This chemistry used in this library subset. Figure 6a shows that library employed simple, solution-phase, chlorine displacement the only activity observed on screening these compounds chemistry using large available monomer sets in 96-well polytetrafluoroethylene plates. came from three members of a subset of 700 compounds made from just two templates: all other templates (4300 compounds) were inactive. The three active compounds avoid the selection of high molecular weight and lipophilic were repeatedly tested and had IC50 values of 5–10 µM. The compounds that would occupy the top right-hand side of hits were followed up by non-combinatorial synthesis of a the charts, because of the way in which the monomer lists closed-loop subset of 693 compounds from the same vir- were ordered. The fifth and sixth loops delivered three dif- tual library, focussing in on the regions of chemical space ferent series of selective and potent compounds with activ- occupied by the three original hits and related regions ity against the kinase in the region of 10–50 nM. All of the (Fig. 6b). This subset gave a respectable 2.3% hit rate at 2500 rule-of-five6 compliant compounds made in the six 10 µM. Furthermore, repeating the screening with solid closed-loops were synthesized as singletons, each one of samples for some of the more active compounds gave the which was analyzed and autopurified as described above; same Ki values, reinforcing our confidence in the identity all with less than one ‘person-year’ of effort. The leads we and quality of the parallel synthesis-derived liquid samples discovered have provided the basis for a chemistry-staffed used in the screens. lead-optimization project, which is now under way. The early SAR information generated from this first closed-loop prompted the synthesis of a second loop of Conclusions 345 compounds, which gave an impressive 66% hit rate We have demonstrated that non-combinatorial chemistry against the enzyme at 10 µM (Fig. 6c). These compounds is an effective method for screening-file enrichment and had improved solubility, and exhibited weaker activity and for lead optimization. We believe that whatever strategy is independent SARs against key kinase-selectivity targets. used for library chemistry, the quality of design and the Another 1400 compounds were made in four succeeding quality of the products are of paramount importance; we closed-loops, in the course of which the virtual library was have achieved this with our S1D and LiCRA systems. We expanded from 300,000 to 4,600,000 compounds by the intend to build upon this system in the future and add new addition of further custom and commercial monomers. A design tools and new synthetic capabilities to expand the summary of the outcome of all six closed-loops is shown in scope and quality of the virtual chemistry space accessible Fig. 6d. It is clear from Fig. 6a–d that S1D has been used to to our chemists. 784
- 7. DDT Vol. 6, No. 15 August 2001 research focus reviews (a) (b) (c) (d) Figure 6. (a) Non-combinatorial selection of 700 compounds from a virtual library of 300,000 in two templates: yellow and dark-blue diamonds. The three active compounds are circled in red. Each axis of the chart consists of a list of the amine monomers used in each position around the aromatic core. The x-axis corresponds to the amines used to displace the most reactive chlorine, the y-axis to the amines used in the displacement of the second, less reactive chlorine. The amines were ordered by similarity (according to a judgement made by medicinal chemists; see Fig. 7 for design outline). (b) A corresponding view of the 693 compounds made in the first closed loop in response to the three hits found in Fig. 8a. The original 700 compounds screened are represented as green dashes; the 693 new compounds in the first loop are shown as blue diamonds (inactive) and red circles (active); the three original hits upon which the SAR was based are still shown circled in red. In this display, some of the compounds synthesized sit on top of one another in the chart, as multiple templates sometimes shared the same monomer pairings; this third dimension of the data is missing from this compressed 2D view. (c) Corresponding figure for the second closed-loop; all previous compounds are now shown as green dashes, and the second loop compounds as blue diamonds (inactive) or red circles (active); again, this is a compressed 2D view of a 3D dataset. (d) A compressed 2D representation of all six closed-loops (CLs). The compounds from the six loops are represented with the following symbols: CL1, yellow triangles; CL2, magenta squares; CL3, blue diamonds; CL4, brown circles; CL5, open red diamonds; CL6, blue dashes; two groups of file compounds are shown as crosses and as crossed crosses. Acknowledgements References 1 Labaudiniere, R.F. (1998) RPR’s approach to high-speed parallel We are grateful for the expert assistance provided in hard- synthesis for lead generation. Drug Discov. Today 3, 511–515 ware development, automation, information systems, 2 Hird, N.W. (1999) Automated synthesis: new tools for the organic chemistry and screening by the following Pfizer colleagues: chemist. Drug Discov. Today 4, 265–274 3 Dolle, R.E. (2000) Comprehensive survey of combinatorial library Mark Kemp, John Taylor, Pam Greengrass, Duncan Armour synthesis: 1999. J. Comb. Chem. 2, 383–433 and their teams; Ruth Tucker, Mark Lord, Ian Parsons, 4 Coates, W. J. et al. (2000) Successful implementation of automation in medicinal chemistry. Drug Discov. Today 5, 521–527 Jason Robson, Derek Hearn and Wilma Keighley and her 5 Terrett, N.K. et al. (1995) Combinatorial synthesis – the design of team; James Merson and his team; and George Perkins, compound libraries and their application to drug discovery. Tetrahedron 51, 8135–8173 Chris Selway, Mike Stace, Mark Taylor and Adrian Wright. 6 Lipinski, C.A. et al. (1997) Experimental and computational approaches We are also grateful to Andy Kemp of Aitken Scientific, Ian to estimate solubility and permeability in drug discovery and development settings. Adv. Drug Deliv. Rev. 23, 3–25 Hibbard and Grant Cameron of Anachem and Jeremy Batt 7 Ripka, W.C. et al. (2001) High-throughput purification of compound of Micromass for their productive collaboration on the libraries. Drug Discov. Today 6, 471–477 8 Fang, L. et al. (2000) Evaluation of evaporative light-scattering detector autopurification system, and to David Webb of PA Consulting for combinatorial library quantitation by reversed phase HPLC. for project management of LiCRA and S1D. J. Comb. Chem. 2, 254–257 785