Properties of Combinatorial Libraries for Drug Development

Generally, the drug development process aims to synthesize chemical entities which are, apart from their biological effects on specified targets, orally bioavail-able. A major criterion for good absorption of these compounds is that certain physiological properties will allow them to be reasonably absorbed into the gastrointestinal system. In a landmark paper, Lipinski and coworkers [1] from Pfizer were looking for factors which influence absorption and permeability of drugs and which could be considered even at the stage of the drug discovery process. They studied a database of 2245 drug compounds which had passed clinical phase I and had entered clinical phase II. Lipinski and coworkers assumed that these compounds had good absorption properties since insoluble and poorly permeable substances had been eliminated from further research at an earlier stage of the drug development process. The properties which were taken into account had to be easily calculable and could be expected to have a pronounced effect on permeabil-

Handbook of Combinatorial Chemistry. Drugs, Catalysts, Materials. Vol. 2. Edited by K. C. Nicolaou, R. Hanko, and W. Hartwig Copyright © 2002 WILEY-VCH Verlag GmbH, Weinheim ISBN: 3-527-30509-2

ity. These authors identified: (1) molecular weight, (2) lipophilicity, (3) number of hydrogen bond donor groups (i.e. number of NH and OH bonds), and (4) number of hydrogen bond acceptors (i.e. number of N + O) as the key properties which have an essential effect on the permeation through lipid bilayers and therefore on the intestinal absorption process. Properties 1, 3, and 4 are obviously easily calculable from the structural formula. As a measure of lipophilicity, Lipinski and coworkers used the partition coefficient of the substance between n-octanol and water, P, which can be approximated by increment systems. The authors used two such systems. The first was the Pomona College Medicinal Chemistry program which calculated log P values (C log P) from structural fragments and gave very accurate results, but failed in many instances because of fragments in the molecule which were not assigned in the program. The other system they used had been described by Moriguchi et al. [2]. This system gives less but still reasonably accurate results and allows the calculation of M log P directly from the structural formula [2].

Lipinski et al. determined the distribution of these four properties among the drug database and found cut-off levels for each parameter such that @ 90% of the drugs were within these levels. From these, they stated that poor absorption and permeation are more likely when:

1 the molecular weight (MW) is >500;

3 the number of N + O is >10 (H-bond acceptors);

5 compound classes which are substrates for biological carriers are exceptions to the rule.

These became known as ''Lipinski's rule of 5''. Lipinski and coworkers found that no more than 10% of drugs obeyed any combination of two of these rules. For example, only 1% of the drugs had both MW and log P outside the cut-off levels. Orally bioavailable substances which violate the first four rules mainly belong to a few therapeutic categories, i.e. antibiotics, antifungals, vitamins, and cardiac glycosides. These substances are assumed to be substrates for naturally occurring transporters.

Using these results, Lipinski and coworkers examined the lead generation process at Pfizer, which is representative of that used by many large pharmaceutical companies. The advent of high-throughput screening (HTS) allowed testing of huge numbers of chemical compounds. In contrast to earlier techniques, it was not necessary to obtain thermodynamically stable aqueous solutions of the compounds for biological testing: the compounds were delivered as dimethyl sulfoxide (DMSO) stock solutions, which allowed for testing of very lipophilic compounds. At the same time, combinatorial chemistry enabled the synthesis of vast numbers of test compounds for in vitro HTS screening. Lipinski and coworkers compared the properties of test compounds which were synthesized 1986 and 1994 (pre- and post-HTS era) (Table 25.1). They found that test compounds were becoming much

25.2 Properties of Combinatorial Libraries for Drug Development | 727 Tab. 25.1. Distribution of M log P and molecular weight in the pre- and post-HTS era [1].


M log P





90 75 50

514 415 352

726 535 412

more lipophilic and heavier. Both these factors led to the discovery of a number of highly active compounds in vitro but caused severe problems in developing orally bioavailable drugs owing to insufficient solubility and permeability in vivo. Obviously, the strategy in combinatorial chemistry of taking a multifunctional central building block and decorating it with organic residues runs the risk of generating very high MW compounds which are furthermore very lipophilic since all polar groups (acids, amines, alcohols, and phenols) would be masked by combinatorial variation (acid amides, ureas, ethers, esters). Although this can be useful during the late stages of the optimization process, it is not a good procedure for generating lead compound libraries.

In accordance with these results, Teague et al. [3] studied the drug optimization process at Astra Zeneca. They found that only very few lead compounds have such desirable properties and that it would be very unlikely for one of them to pass on to clinical development directly. Generally, potency and pharmacokinetic profile have to be further optimized - this goal is accomplished by the addition of suitable groups and side-chains and causes an increase in MWand lipophilicity of the drug compared with the initial lead. Teague et al. divided the properties of lead structures identified by HTS into three groups with respect to affinity, MW, and log P:

1 Intermediate affinity (> 0.1 mM), low MW (< 350), and low C log P (< 3). These are "classical" lead structures which can be optimized by introducing lipophilic groups to increase potency and improve pharmacokinetic properties.

2 High affinity (« 0.1 mM), high MW (» 350), and low C log P (< 3). These leads are often derived from natural products and can be optimized by derivatization to improve pharmacokinetics while potency is retained.

3 Intermediate affinity (> 0.1 mM), high MW (» 350), and high C log P (> 3). These leads are often generated in HTS assays of combinatorial libraries. An optimization of these drug-like leads normally proves difficult, since the affinity results from many nonoptimized interactions between lead and target and a further optimization produces very lipophilic, poorly soluble compounds.

Teague et al. concluded that combinatorial libraries for lead identification should be fundamentally different from those for lead optimization. Lead structure libraries have to produce compounds with low MW (100-350) and low C log P (1.03.0). Focused libraries based on such leads can quickly improve drug properties.




(0 s


c a>










010101010101010101010101010 YYCNeNcoco^t^i-ioiotptpivop oooooooooooooo


Molecular Weight

Fig. 25.1. Molecular weight distribution of drugs that have good and poor oral bioavailability [4].

Although there is general agreement in the literature on the influence of lip-ophilicity and H-bond donor and H-bond acceptor properties on the bioavailability of drugs, the influence of MW is controversial. A set of 286 marketed drugs was examined for the relationship between MW and oral bioavailability and a boundary of 40% blood levels after oral application was chosen, above which no bioavailability problems should be expected [4]. Of the total set of 286 drugs, 168 lay above the 40% level and 118 lay below. As expected, about 50% of drugs that perform well had a MW between 250 and 350 and only very few had a MW > 500. Surprisingly, however, the drugs with low bioavailability had a quite similar distribution, with an average MW of between 350 and 400, and only a few compounds had MW > 500 (Fig. 25.1). When those drugs with obvious metabolic instabilities (such as b-lactams), with very high lipophilicities (which cause poor water solubility), and with quaternary amines were subtracted from the 116 inferior performers, 97 substances remained whose correlation between MW and oral availability did not significantly differ from the distribution of the well-bioavailable compounds. This indicates that MW is not a primary cause of low bioavailability but that high MW is very often correlated with factors such as high lipophilicity and poor solubility and with a high number of heteroatoms.


Differentiation of Drug-like and Nondrug-like Compounds

Since an accessible virtual library is much larger than one that is actually synthesized, the first criterion when choosing compounds for synthesis should be that they have reasonable drug-like properties. For example, toxic and reactive substructures which can react with proteins can cause false-positive read-outs in screening assays and should, therefore, be eliminated (Fig. 25.2) [5]. Several authors have tried to differentiate drugs from nondrugs by looking at the

25.3 Differentiation of Drug-like and Nondrug-like Compounds I 729

il o

Sulfonyl halldes

R-^H Aldehydes

Acyl halldes

R X Alky I halldes

Was this article helpful?

0 0

Post a comment