Original article
Drug-like properties and the causes of poor solubility and poor permeability

https://doi.org/10.1016/S1056-8719(00)00107-6Get rights and content

Abstract

There are currently about 10 000 drug-like compounds. These are sparsely, rather than uniformly, distributed through chemistry space. True diversity does not exist in experimental combinatorial chemistry screening libraries. Absorption, distribution, metabolism, and excretion (ADME) and chemical reactivity-related toxicity is low, while biological receptor activity is higher dimensional in chemistry space, and this is partly explainable by evolutionary pressures on ADME to deal with endobiotics and exobiotics. ADME is hard to predict for large data sets because current ADME experimental screens are multi-mechanisms, and predictions get worse as more data accumulates. Currently, screening for biological receptor activity precedes or is concurrent with screening for properties related to “drugability.” In the future, “drugability” screening may precede biological receptor activity screening. The level of permeability or solubility needed for oral absorption is related to potency. The relative importance of poor solubility and poor permeability towards the problem of poor oral absorption depends on the research approach used for lead generation. A “rational drug design” approach as exemplified by Merck advanced clinical candidates leads to time-dependent higher molecular weight, higher H-bonding properties, unchanged lipophilicity, and, hence, poorer permeability. A high throughput screening (HTS)-based approach as exemplified by unpublished data on Pfizer (Groton, CT) early candidates leads to higher molecular weight, unchanged H-bonding properties, higher lipophilicity, and, hence, poorer aqueous solubility.

Introduction

The literature emphasis on combinatorial chemistry and the screening of up to million(s) of compounds tends to obscure the fact that the world of drug-like compounds is quite limited, and that most of the information content related to desirable drug-like properties is contained in relatively small numbers of compounds. Filters for selecting drug-like compounds (Brennan, 2000), or schemes for differentiating between nondrug and drug-like compounds are based on analysis of libraries of a few thousand to 50 000 compounds (Sadowski & Kubinyi, 1998, Shah et al., 1998). The ability to construct large experimental screening libraries and the very large estimates of chemistry space accessible to low molecular weight compounds has been coupled with the concept of “maximal chemical diversity” (Martin et al., 1995). This coupling results in the argument that screening efficiency and the probability of a useful screening active will be increased if screening libraries span as large a volume of chemistry space as possible. For example, a recent chemical diversity analysis suggests that “in order to ensure nanomolar ligands to any given target, a library of at least 24 million molecules will be required” (Wintner & Moallemi, 2000). In contrast to this estimate, current experience suggests that clinically useful drugs exist as small tight clusters in chemistry space. For example, the estimated number of drug targets is only about 500 (Drews, 2000). Maximal chemistry diversity is an inefficient library design strategy, unless there are vast numbers of useful undiscovered targets. Moreover, while maximal chemistry diversity is possible in-silico, it remains to be seen whether it is possible in an experimental setting. The theme of large chemistry space and small target space applies to the screening arenas of absorption, distribution, metabolism, and excretion (ADME). The author contends that ADME chemistry space is much simpler than pharmacological target chemistry space. The result is that simple filters and rules work for ADME Clark, 1999, Lewis, 2000, Lipinski et al., 1997, but not for pharmacological targets. Acceptable ADME space may be considered as a smaller subset of pharmacology target space, which is in turn likely a very small subset of chemistry space. Paradoxically, for large data sets, pharmacological target SAR prediction is easier than for ADME prediction. Much of the reason lies in the multi-mechanism nature of current ADME screens, as opposed to the typical single mechanism of pharmacological target screens. The theme of information content related to desirable drug-like properties in relatively small numbers of compounds applies to the question of the relative importance of poor solubility or poor permeability in the problem of poor oral absorption. The time-dependent analysis of simple properties from Merck and Pfizer clinical candidates illustrates that the research approach to lead generation strongly influences solubility and permeability. As targets become more complex, a Merck-like “rational drug discovery” approach tends to poorer permeability, while a Pfizer (Groton, USA)-like high throughput screening (HTS) approach tends to poorer solubility.

Section snippets

Drugs and chemistry space

The world of drug-like compounds is limited in that there are currently only about 10 000 drug-like compounds. Drug-like is defined as those compounds that have sufficiently acceptable ADME properties and sufficiently acceptable toxicity properties to survive through the completion of human Phase I clinical trials. Compounds that survive through Phase I and into Phase II clinical efficacy studies are conveniently identified by the presence of a United States Adopted Name (USAN), International

Maximal chemical diversity. Fact or fiction?

From HTS experience, we know that for the more tractable targets, i.e., g-protein coupled receptors, ion channels, and kinases, we can reliably detect hits from screening “diverse” chemical libraries in the size range of 105–106 (Spencer, 1999). How do we reconcile the experimental finding that HTS works in finding hits with the dismal prediction for screening success from the aforementioned thought experiment? The answer is that our “diverse” screening libraries are anything but diverse. True

ADME and pharmacological target dimensionality in chemistry space

ADME and chemical reactivity-related toxicity is very different from biological receptor activity in its occupancy of chemistry space. Compounds with biological receptor activity exist in small, tight clusters. The actual number of existing in-vivo drug targets is very small, and has been estimated at 417 in total (Drews, 1996). However, the chemistry space they occupy is very large. Hence, the chemical descriptor space can be very large. For example, the dimensionality of chemistry drug space

Screen for the target or ADME first?

Which is better to do first? Select for biological receptor activity or select for properties related to “drugability,” i.e., ADME/toxicity. Currently, the industry practice is to screen for the receptor activity first, and then to follow with the “drugability” properties as a fast second. However, the order of this process could well change. Consider the problem faced when dealing with a new biological therapeutic target. A priori, in general, nothing will be known about the structural

Solubility, permeability, and potency

What level of permeability or solubility is needed to minimize poor absorption? Fig. 1 shows a bar graph that we distribute to our medicinal chemists that answers this question. It depicts the minimum acceptable solubility in μg/ml that is required for an orally active drug. The vertical columns are grouped in sets of three and show the minimum thermodynamic aqueous solubility (at pH 6.5 or 7.0) that is required for low, medium, and high permeability values (Ka) at a particular clinical dose.

Relative importance of poor solubility and permeability

What is the relative importance of poor solubility and poor permeability towards the problem of poor oral absorption? In a global sense, the relative importance may depend at least in part on an organization's research approach. Specifically, it could depend on how leads are generated. Two extremes can be identified. In one extreme, leads are generated from an empirical HTS, and, in the other extreme, leads are generated in some type of “rational drug design” process. This could encompass a

Conclusions

This article has focused on the themes of chemistry space and on the information content found in small data sets. Against these themes are portrayed the concepts of maximal chemical diversity in combinatorial chemistry and the differences in dimensionality between ADME and pharmacological screening. The author argues that drugs are not uniformly distributed in chemical space, and that true “maximal chemical diversity” is unobtainable in an experimental setting. The proposition is made that

References (18)

There are more references available in the full text version of this article.

Cited by (3029)

View all citing articles on Scopus
View full text