PROPERTIES of A GOOD DRUG
Once when doing usability testing on a piece of drug design software, we asked an open-ended question to see what most interested chemists. The question asked was simply “Which of these compounds do you find interesting?”
We had hoped that this type of question would be open enough for the chemists to start asking for new features, such as a chemically relevant way to sort the molecules. Instead, the question told us more about the person we asked it of than about the software. If the subject was trained as a synthetic organic chemist, they chose a molecule that they could easily synthesize. If the subject was
trained in computational drug design, they chose the most drug-like molecules, those that were heterocyclic with multiple functional groups, and few functional
groups known to be highly toxic.
Obviously, both being able to synthesize a molecule and choosing to pursue synthesis of compounds that could potentially be useful drugs are important concerns. This book is not about organic synthesis, some software packages that have been created to help map out possible synthesis routes. This page is devoted to discussing why some compounds are considered to be “drug-like” and others are not. The first section talks about the ways in which compounds are tested for usefulness as a drug. Discussing why compounds pass or fail these tests will begin to give some nonspecific
insight into what features of molecular structure are important in drug design. The second section is devoted to discussing properties of moleculesthat determine whether they have the potential to be good drugs. Finally, we present some exceptions to the typical rules of drug-likeness.
COMPOUND TESTING
“Efficacy” is the qualitative property of a compound having the desired effect on a biological system. In the case of drug efficacy, this means having a measurable ability to treat the cause or symptoms of a disease. “Activity” is the quantitative measure of how much of that compound is required to have a
measured effect on the biological system. Drugs work through binding to a target in the body, or a pathogenic organism such as a virus or bacterium. The target is usually a protein, but in some cases can be DNA, RNA, or another biomolecule. The vast majority of drugs work by inhibiting the action of the target. Unless explicitly stated otherwise, it is assumed in this page that drug activity is obtained through inhibition of the target.
It is important to understand the terminology related to drug testing. As compounds show various levels of activity in the different stages of testing, the compounds will be referred to as “hits,” “leads,” “drug candidates,” “drugs,” and several other terms. Within each pharmaceutical company,
these words have very precisely defined meanings. However, the technical definitions of these terms differ from one company to another. Sometimes, even within the same company, the terminology used within the research and development laboratories will be different from that used by the marketing department, or there may be differences in the quantitative criteria from one project to the next. Thus, researchers must be careful about how their company defines these terms, and to understand the difference in terminology when talking with researchers from other companies. Typically, the term “hit” refers to compounds identified in some initial rounds of screening. Compounds identified as hits typically go through additional rounds of screening. Once screening results have been verified, some readily obtained derivatives will be synthesized or purchased and tested. Once a compound, or often a series of compounds, meets a certain set of criteria, the series will be designated as a lead series. The lead series is then used as the basis for a more comprehensive synthesis of many derivatives and more in-depth analysis, both computational and experimental. The following are some typical criteria that may be necessary to move a compound
series onto the lead development stage:
† concentration-dependent activity
† active in both biochemical and cell-based assay
† below some IC50 threshold (perhaps low micromolar, or down to the
nanomolar range)
† some understanding of the series’ structure–activity relationships
† known binding kinetics
† selectivity assessment
† well-established structure and purity
† stability assessment
† synthetically tractable
† series is patentable
† some path for optimization (creating derivatives) is apparent
† solubility measured
† log D measured
† metabolic liabilities predicted
† pharmacokinetics predicted or measured
† toxicity issues predicted
† potential for significant side effects considered (e.g., whether the drug will block hERG channels, thus resulting in drug-induced cardiac arrhythmia)
Since the precise definitions of terms such as “hit” and “lead” are not universally accepted, compounds at any stage of testing will be referred to in this page using terms that are generically interchangeable within the context of the book, such as “compound,” “molecule,” or “drug.”
In order to test if a compound is effective and to measure its activity, four different types of experimental tests may be used:
† biochemical assays
† cell-based assays
† animal tests
† human tests
For each of these types of tests, the results will give different information about how the compound interacts with the biological system. The following sections give a short discussion of each type of test. Note that crystallography is notmentioned here, since it is usedmore for target development
than for testing of drug efficacy and is thus addressed in the appropriate section of this page.Most of the testing procedures and their details are beyond the scope of this text. The following discussion is intended only to give an introductory description and to establish some terminology. Later in the page, there will be discussions of how these tests fit in with the computational drug design process.
Biochemical Assays
Biochemical assays are usually the cheapest way of experimentally testing compounds. They are also the simplest environment in which the compound can be tested. Thus, biochemical assays are usually the first line of experimental testing of compounds. The term “in vitro assay” is sometimes applied to
biochemical assays or cell-based assays, or both. Biochemical assays entail putting the drug target, usually a protein, in solution with a compound, and measuring whether the compound inhibits
the protein’s activity. Ideally, this can be done using very small quantities of material in the wells of a plate. If possible, it is preferable to measure results colorimetrically. The quantitative measurement of activity is referred to as “scoring” the plates.
In high-throughput screening, there might be a single test for a given compound, in a single well of a plate. This can only give a yes/no result at some activity threshold. Single-well tests are prone to having a percentage of falsepositive and false-negative results, due to random fluctuations in reaction conditions and other factors. These tests are usually used to perform a first-pass screening of large libraries of compounds. Typically, libraries that represent a very wide range of molecular structures are screened in this way. This is done in order to generate an initial list of compounds that may be active. Diverse screenings sometimes identify a class of compounds that inhibit a given target and have not been explored in the past for activity against that target.
Promising compounds are usually retested at a series of different concentrations.
This allows a value of the inhibition constant KI to be computed
from the results. Sometimes, assay results are reported as an IC50 value,
which is the concentration at which the target activity is decreased by 50%.
A passed biochemical assay should indicate that the compound has bound
to the target’s active site, thus having activity (inhibiting the operation of the
target) through a competitive inhibition mechanism. However, it is possible to
have false-positive results. For example, if the results are being measured colorimetrically,
then colored compounds can give false-positive results. There
are sometimes ways to avoid this problem by careful selection of the measuring
wavelengths or by subtracting out the signal from a blank solution.
Compounds that show negative results in a biochemical assay are typically
those that do not bind to the target. However, there can be false-negative
results. For example, these failures can be due to the compound being tested
not being soluble enough to stay in solution.
In general, biochemical assay results are the experimental tests that show the
highest correlation with computational drug design techniques. Since techniques
such as docking, 3D-QSAR, and pharmacophore searches model the
compound in the active site of the target, there is a direct correspondence
between the computational results and biochemical assay results. Often, this
correspondence is used in both directions. Biochemical assay results are
used to verify that appropriate computational techniques have been chosen
for the target being studied. Then computational results are used to flag
when an experimental result should be examined more closely as a possible
false-negative or false-positive result. Once validated (preferably), the computational
techniques can be used to search for potentially active compounds in
massive libraries of chemical structures at a far lower cost than experimentally
testing each compound.
Cell-Based Assays
Cell-based assays (sometimes called in vitro assays) are performed by placing
a compound in solution with a culture of living cells. This adds a level of complexity
to the test conditions. Tests for a compound inhibiting an enzyme
inside the cell may fail because of a lack of efficacy or because the compound
is not sufficiently lipophilic to pass through the cell membrane. Several types
of cell culture assays are often performed. Compounds showing efficacy may
also be tested for cytotoxicity and mutagenicity using cell-based assays.
Cell-based assays are typically more expensive than biochemical assays, but
still fill a valuable role. The greater cost is mostly due to their being more laborintensive
and more difficult to automate. However, there are some assays, such
as ion channel assays, that cannot be done biochemically and therefore must
use living cells. The cell-based assay results give additional information,
such as bioavailability, to decide which compounds should go to an initial
synthesis scale-up and into animal testing.
If a cell-based assay indicates that a compound is active, it is an indication
that activity should be seen if the drug gets into the blood plasma. This is not
always true, since it does not address oral bioavailability, blood–brain barrier
permeability, reaching the correct organ, etc. It does at least give an indication
of correct binding to cell surface receptors, or that the drug is reaching a target
inside a cell.
A failed cell-based assay can mean that the drug is not active against the
target, but there are also other possible reasons for failures. A failure can
mean that the compound is cytotoxic or mutagenic, which can be a disqualifying
criterion in itself. There could also be a bioavailability issue, such as not
reaching a target internal to a cell. Or the compound may be insoluble under
the test conditions.
Another type of cell-based assay is a Caco-2 assay. Caco-2 cells are colon
wall carcinoma cells. There is a Caco-2 cell line that can grow a layer of cells
that aligns to have the efflux pumps (P-glycoprotein and less common ones)
arranged to provide active transport of chemical species from one side to
another. Two tests can be performed to see how much of a given species
will be transferred in each direction. By comparing these two results, it is possible
to get an indication of how much the species will be affected by both active
and passive transport.
The advantage of the Caco-2 assay is that it is one of the few experimental
bioavailability measurements available. Researchers are really interested in
passive absorption in the small intestine, where most substances enter the
body, and in diffusion through the cell membrane. However, since assays
for those are either not available or much more difficult to run, Caco-2
assays are performed with the understanding that these other types of bioavailability
are generally proportional to the Caco-2 assay results. An alternative to
Caco-2 is a MDR1-MDCK cell line assay, which is based on dog kidney cells.
Another alternative is a parallel artificial membrane permeability assay
(PAMPA).
Animal Testing
Animal testing (also called in vivo testing) is much more expensive than
biochemical and cell culture testing. Thus, only a handful of compounds
will be sent on to animal testing. Animal testing typically provides an
indication of whether the compound shows efficacy in a mammal, is toxic,
and is orally bioavailable. Animal testing can also give an initial indication
of whether a potential drug displays severe side effects. Animal testing
typically takes about 6.5 years if testing for toxicity with chronic dosages is
performed. Animal efficacy testing and acute dosage toxicity tests are much
less time-consuming.
Ideally, animal testing should be done with mice, for cost reasons. However,
a necessary prerequisite for efficacy testing is having an animal susceptible to
the disease being treated. Thus, it is sometimes necessary to do animal testing
with rats, rabbits, dogs, primates, armadillos, or other species. For example, the
only animal model for leprosy is the armadillo.
An increasingly important tool is the creation of knockout mice. These are
mice that have one gene turned off through genetic manipulation. This is not
always possible, as some gene deactivations cause prenatal or postnatal mortality
of the mice. Studying the effect of a gene knockout can give information
about the role that that gene plays in the disease under study. It also gives indications
as to how downregulating the function of that gene through the action
of a drug will affect the animal.
Another option is the creation of transgenic mice. These are mice that have
an extra gene, not found in mice, inserted into their genome. If there is not an
animal available that is susceptible to a given disease, it may be possible to
create transgenic mice that are susceptible. For example, transgenic mice
susceptible to polio, which normal mice do not contract, have been created.
There have also been transgenic sheep and chickens created to manufacture
human proteins. At present, the creation of knockout mice and especially
transgenic animals is still a very difficult and expensive task.
In recent years, there have been a number of attempts to find alternatives to
animal testing. In some cases, there are viable alternatives, such as testing on
animal tissues. Where alternatives exist, they are often used both for cost savings
reasons and for humane reasons. However, there are still many diseases
for which there is no alternative to animal testing that is acceptable to the
United States Food and Drug Administration (FDA).
Human Clinical Trials
For every drug, there must be a first time that it is tried on a human being. This
necessarily entails some risk to human life. Human clinical trials have been
designed as a multiple-step process, referred to as phases, in order to minimize
the risk of harming patients while doing the testing necessary to bring the drug
to market as quickly as possible. The following discussion is based on the
clinical trial regime used in the United States, which is regulated by the
FDA. The testing requirements vary greatly from one country to another.
Many countries have an abbreviated testing process for drugs that have already
been approved in another country, particularly if the drug has been approved in
the United States, which has some of the most stringent testing requirements
in the world.
Phase I Clinical Trials. The primary goal of phase I clinical trials
is safety testing. The test is designed to verify that the drug does not result in
any severe side effects that would be harmful to a healthy adult. Thus, the subjects
in a phase I trial do not have the disease that the drug was designed to
treat. Often phase I trial subjects are college students, who are paid a small
sum of money to participate in trials being carried out at university-affiliated
facilities. Phase I clinical trials typically take about a year and a half. About
70% of the drugs submitted to phase I testing are sent on to phase II testing.
Phase II Clinical Trials. Phase II trials are designed to obtain an
initial indication of whether the drug successfully treats the disease in
humans. They also give some initial data about necessary dosages and side
effects. Typically, a few hundred patients at various stages of disease progression
are tested in phase II trials. Phase II typically takes about two years.
About a third of the drugs tested in phase II trials are sent on to phase III testing.
Phase III Clinical Trials. Phase III trials entail giving the drug to a
much larger group of patients. In this phase, dosage regimes are determined
more precisely. There will also be a sufficiently diverse sample of patients
to determine if the necessary dosage is affected by other drugs that the patients
are taking or by patient ethnic background. Patients are also monitored for side
effects, both severe and mild. Typically, a few thousand patients are treated in a
phase III trial. About a quarter of the compounds submitted to phase III trials
are approved for use in the United States.
After phase III testing is completed, the results are submitted to the FDA for
approval as a prescription medication. Data on patient histories continues to be
collected long after FDA approval. This data is later used to justify approval for
sales over the counter (without a prescription). Yet, later in the life of a drug,
the patents expire, making it fair game for generic manufacturers.
There is a considerable amount of variation in clinical testing from one drug
to another. If a drug is administered to patients with only months left to live, it
may be possible to very quickly show that the drug extends life with an acceptable
level of side effects. Drugs that are taken repeatedly over a patient’s life,
such as cold or allergy medications, require very extensive testing, and may be
approved only if side effects are extremely mild.
Most compounds do not fail clinical trials owing to failing to treat the disease.
Most fail owing to ADMET issues (“absorption, distribution, metabolization,
excretion, and toxicity”). Some are eliminated from the body so quickly
that the patient must take doses of the medicine every few hours. Some have
unacceptable side effects. Some have poor oral bioavailability, so massive
quantities of pills must be taken in order to get enough into the bloodstream.
Designing compounds that address all of these issues as well as treating an
illness is one of the reasons that drug design is such a challenging task. The
first step is to look at how various aspects of molecular structure correlate with
drug activity, bioavailability, and toxicity.
MOLECULAR STRUCTURE
The issues discussed above can be translated into constraints on molecular
structure. The following is a discussion of why certain molecular structures
might be desirable or undesirable as drugs. Although many generalizations
can be made, some of these depend upon the particular drug target being
studied, and thus there are exceptions to all of these loose rules. The last
section of this chapter will examine some exceptions to the rules.
Activity
The first concern of drug designers is to create compounds that will have very
high activity. Activity is usually quantified by the order of magnitude of the
biochemical assay inhibition constant KI. Nanomolar activity is usually the
objective of drug design efforts. In rare cases, picomolar activity can be
achieved. It is very difficult to start from a compound with millimolar activity
and modify the molecular structure to obtain a compound with nanomolar
activity. However, that is usually the situation that drug designers face. It is
considerably easier to start with a compound having subnanomolar activity
and modify it to have better bioavailability or less toxicity without sacrificing
too much of the activity.
The majority of drug targets are proteins, such as enzymes or cell surface
receptors. In the majority of cases, the drug acts through competitive inhibition,
by binding to the active site and thus preventing the native substrate
from entering the site. Thus, the first major requirement is that the drug
should be the correct shape to fit in the active site. This is referred to as the
lock-and-key theory of drug action.
The second requirement is that the drug should bind to the active site. This is
accomplished through the positioning of functional groups in the drugmolecule.
For example, if the active site contains a hydrogen bond donor, then the drug
should have a hydrogen bond acceptor positioned to give a hydrogen bond binding
the drug to the active site. Other important interactions arep-system stacking,
positioning of charged groups to form ionic bonds, van der Waals interactions,
and steric hindrance. In a minority of cases, discussed later in this book, the
inhibitor will form a covalent bond with the target’s active site.
Many drugs resemble the target’s native substrate; such drugs are referred to
as substrate analogs. For example, if the native substrate is a protein sequence
including a proline, then the proline ring could be mimicked by a tetrahydrofuran
ring. Understandably, a compound that is very similar to the native
substrate in shape and binding properties will also tend to be accepted into
the active site. Typically, a good drug will be sufficiently different from
the native substrate that it will not react with the target, or be too readily
metabolized by the body.
One very successful tactic is to create drugs that are transition state analogs.
In the case of an enzyme that catalyzes a reaction, the reaction can be described
as an initial complex of reactants, a transition structure, and a complex of products.
A transition state analog resembles the transition structure for the reaction,
but lacks the necessary moiety (i.e., a leaving group) to allow the reaction
to go to completion. The enzyme will accept the drug, form a complex that
resembles the transition structure, and then get stuck at that point unable to
complete the reaction. For example, the first step in the reaction by which
proteases cleave amide bonds is a nucleophilic attack on the carbonyl carbon
in the –N–CO– group. If the drug resembles a modified version of the
native substrate in which that amide is replaced by –C–CO–, –N–PO2–, or
–O–PO2–, it will still undergo nucleophilic attack, but the protease will be
unable to cleave the bond. Some authors treat the terms “substrate analog”
and “transition state analog” as synonymous, and others make the mechanistic
distinction described here.
Experimentally, the positioning of the drug in the active site can be seen
only with great effort by generating crystal structures of the protein with the
drug soaked into the crystal. The experimental analysis of the binding is
most often obtained by measuring an inhibition constant KI with a biochemical
assay. Both fit and binding energy are much easier to analyze computationally.
The primary tool for this analysis is the “docking” technique, in which an
automated algorithm positions the molecule in many different orientations
in the active site to find the lowest-energy orientation. This direct correlation
between computational results and drug activity makes docking the mainstay
of structure-based drug design. Pharmacophore models and 3D-QSAR also
model binding and fit to the active site, although somewhat more indirectly.
Binding to an active site is important, but it is not the only criterion to be
considered. Another important issue is that of specificity. Specificity is the
ability of a compound to bind to the target protein, but not to structurally similar
proteins. Every time a drug binds to the wrong protein, it is a possibilityfor the drug to display unwanted side effects when administered to patients.
Protein structural similarity data is given by a systematic Structural Classification
of Proteins (SCOP), which is cataloged online at http://scop.mrc-lmb.
cam.ac.uk/scop. Other useful protein comparison systems are CATH (Class,
Architecture, Topology, and Homology) at http://www.cathdb.info, and
FSSP (Families of Structurally Similar Proteins) at ftp://ftp.ebi.ac.uk/pub/
databases/fssp/fssp.html and http://srs.ebi.ac.uk/srsbin/cgi-bin/wgetz?-
pageþLibInfoþ-libþFSSP. Typically, docking studies are used to test how
strongly compounds bind in a list of proteins that have active sites very similar
to that of the target protein.Aduplicate set of biochemical assays is often performed
for just a couple of the proteins having the most structurally similar active sites.
Some protein active sites are described as having specificity pockets. These
are empty areas of the active site, which may not be particularly important for
drug binding. However, designing a drug to have a nonpolar functional group
oriented in the specificity pocket can improve the drug’s specificity. The
drug’s ability to bind in the active site of similar proteins will be diminished
by the steric hindrance of this additional functional group.
It is possible to design a drug with too much specificity. This is of particular
importance when designing compounds to inhibit viral proteins. If the compound
fits too tightly in an active site, a minor mutation of the virus can
change the active site slightly, thus preventing the drug from binding.
This is one mechanism by which viruses develop resistance to drugs. When
designing antiviral drugs, it is important to test and design against the protein
structure from several serotypes of the virus.
It is important to remember that the shape of the active site may not
stay fixed. There are some proteins in which the geometry of the active site
changes depending upon what is in the site. This is referred to as having an
induced fit. It is estimated that as many as 50% of proteins operate in this way.
In some cases, individual residues in the active site will have a functional
group that undergoes a conformational change to swing into or out of the
active site. In other cases, an entire loop may change conformation to fold
down on top of the active site. Sometimes, two halves of the protein move in a
clamshell-like motion. A few computational drug design tools are designed to
automatically include active site flexibility in the simulation. Many do not,
thus making it the researcher’s responsibility to run multiple calculations with
likely active site conformations.
Bioavailability and Toxicity
Drug designers must carry out a delicate balancing act between efficacy
and bioavailability. Ideally, drugs should be orally bioavailable so that they
can be administered in tablet form. If a compound is too polar, it will not
enter the bloodstream through passive intestinal adsorption, the primary mechanism
for oral bioavailability. If the compound is too lipophilic, it will be eliminated
from the bloodstream by the liver too quickly. Thus, the rule of thumb is:
A drug should be just lipophilic enough to reach the target, and no more
lipophilic than necessary.
There are, of course, time-released tablets, prodrug formulations, and other
delivery mechanisms for alleviating bioavailability issues. However, drug
designers will be urged to optimize bioavailability as much as is practical without
sacrificing too much efficacy. If the drug has acceptable bioavailability as it
is, then the manufacturing process will generally be less complex and costly.
The liver may eliminate a compound from the body as it is. Or it may break
down the compound, usually through the action of various cytochrome P450
(CYP) enzymes. Some drug development firms are starting to incorporate a
host of CYP interactions with the computational drug design process, to
identify problems with excessively rapid elimination early in the development
process. For more discussion of bioavailability, see Chapter 19.
Because CYP enzymes are responsible for the metabolism of many drugs,
drug interactions can occur when one of the drugs affects the activity of these
enzymes. This is a consideration when a patient is taking multiple drugs.
Taking drugs that are CYP inhibitors generally causes the concentrations of
other drugs in the bloodstream to increase, thus decreasing the needed
dosage. Taking drugs that are CYP inducers tends to cause the patient to
require larger doses of other drugs.
Because the liver functions to eliminate a wide range of compounds from
the bloodstream, many drugs interact with the liver adversely, leading to
liver toxicity (hepatotoxicity). The very molecular properties that help a compound
stay in the bloodstream without being broken down by the liver can
cause the compounds to be suicide inhibitors that harm the enzymes in the
liver. For compounds that are given for a short period of time, such as antibiotics
or anesthetics, a moderate amount of hepatotoxicity is acceptable.
For drugs taken frequently over the patient’s life, such as painkillers or cold
medications, hepatotoxicity becomes a major concern in drug development,
as it will be a significant factor in FDA approval of the drug.
Other types of toxicity are less of a concern in drug design—although that
does not mean that they can be completely ignored. For example, very few
drugs exhibit any neurotoxicity. This is because most drugs are moderately
high molecular weight compounds that mimic small amino acid sequences.
Such compounds are too large to fit in the sodium channel, which is the
target for most neurotoxins. Most potential drugs are, however, tested for
carcinogenicity and mutagenicity.
All substances are poisonous if taking in a sufficiently high dosage. They
only differ in the dosage necessary to see a toxic effect. Toxicity is usually
quantified by the lethal dose value LD50, which is the amount necessary to
kill 50% of test subjects. The units of LD50 are usually mg/kg, meaning the
number of milligrams of compound per kilogram of the subject’s weight.
For drug design purposes, the toxicities most frequently tested in animals
are acute (single dosage) and chronic (multiple dosage), given both orally
and by injection into the bloodstream. During the drug design process, determination
of cytotoxicity is sometimes performed as a convenient cell-based
assay.
Toxicity studies are carried out to determine the safety margin between the
dosage necessary for efficacy and the toxic dosage.
Cells are particularly sensitive to mutagens during mitosis. Thus, many
drugs that can be administered to healthy adults cannot be given to pregnant
women or young children. Mutagenicity is therefore usually included in
drug design and assay testing. The first step in mutagenicity testing is the
Ames test, a cell-based assay for mutagenicity. The computational tests for
mutagenicity are QSAR models designed to predict the results of the Ames
test.
Drug Side Effects
Nearly all drugs exhibit some type of undesirable side effect. There are many
types of potential side effects, such as addiction, drowsiness, allergic reactions,
and dry mouth. In some cases, side effects can be fatal. In many
cases, drugs are approved when side effects occur in only a very small percentage
of the population, or are relatively mild. However, somewhat severe side
effects can be tolerated if the drug is extending life expectancy for patients
with an otherwise fatal condition.
Unfortunately, side effects are usually not identified until human clinical
trials. This means that a very large sum of money has been spent already
when these problems are identified. The cost becomes even higher if the
drug fails clinical trials. The pharmaceutical industry would gladly embrace
a technology that could identify side effects of a drug before it entered
human clinical trials, but at present no reliable solution to this problem exists.
One option for early identification of side effects that is utilized is the observation
of animals in efficacy studies. Unfortunately, this generally only allows
identification of side effects that cause behavioral changes in the animals. For
example, drugs causing moderate drowsiness can be identified by documenting
how much time the animals spend asleep. Many side effects cannot be
identified or may not be present in animals. Likewise, there may be side effects
in animals that are not exhibited in humans.
In theory, it should be possible to predict side effects by computationally
identifying other targets in the body with which the drug could interact.
These targets could be identified through the use of bioinformatics, proteomics,
or high speed docking techniques. Once a side-effect target is identified,
the metabolic pathways in which the target participates could indicate what
type of side effects would be observed. Unfortunately, this schema works
poorly, if at all, in practice, owing to the current limited knowledge of the
proteome, metabolic pathways, and expression profiles.
What is routinely done is to run docking studies and assays against the
proteins that show the highest similarity to the target. The assumption is that
if the drug has sufficient specificity to bind to the intended target, and not
to the protein most similar to that target, then there should be a reasonable
expectation that side effects should be minimized. This is a good first approximation,
but not rigorously true, as can be seen from the example above of binding
to cytochrome P450 enzymes, which promiscuously bind a large selection
of substrates.
Multiple Drug Interactions
Some drugs will have interactions with other drugs, thus preventing both from
being taken simultaneously, or forcing physicians to adjust the dosage accordingly.
Cross reactions are seldom considered in the research and development
phase unless the company plans to market a drug with fewer cross reactions
than their competitor’s product. There is no simple test for side effects of
administering multiple drugs to a patient. Most of this data is gained from
human clinical trials. Some initial indications can be suggested by looking
at drug interaction with cytochrome P450 enzymes. Beyond that, there are
only a few vague thumb rules to guide clinical testing, such as “do not mix
depressants” and “everything interacts with blood thinners.”
Some drug interactions are particularly difficult to anticipate. This is illustrated
rather dramatically by the Fen-Phen story. Fen-Phen was a diet drug that
consisted of a mixture of two existing obesity drugs fenfluramine (trade name
Pondimin) and phentermine. The different drugs displayed opposing side
effects. Fenfluramine is an antihistamine that makes patients drowsy and
phentermine is a mild stimulant that increases the body’s metabolism while
suppressing appetite. The mixture was marketed together in an attempt to
give an obesity drug with minimal side effects. The combination of the
drugs also gave more dramatic weight loss than either drug taken separately.
Unfortunately, the mixture of drugs in Fen-Phen gave problems that had not
been identified from the use of the drugs individually. The most severe of
these side effects were the occurrence of serious cardiac valvular disease, and
primary pulmonary hypertension. In September 1997, at the FDA’s request,
Fen-Phen and fenfluramine were removed from the market, as was dexfenfluramine
(trade name Redux), which is one of the enantiomers of which
fenfluramine is the racemic mixture. After that, a number of law suits regarding
long-termaffects of Fen-Phen-induced heart diseasewere filed againstAmerican
Home Products, the parent company that marketed Fen-Phen. American Home
Products subsequently agreed to a class action settlement valued at $4.75 billion.
METRICS FOR DRUG-LIKENESS
Up to this point, this chapter has discussed the concerns that must be addressed
in designing a drug in general terms. This section will focus on molecular
structure. A number of metrics have been developed for identifying whether
compounds are drug-like. These criteria translate directly or indirectly into
desired molecular size, functional groups, etc. Note that drug-likeness metrics
do not indicate that a compound will be a good drug for any disease. However,
compounds that do not meet drug-likeness criteria often fail to be good drugs
due to poor bioavailability, excessive toxicity or other concerns.
The terms “drug-likeness” and “lead-likeness” have both appeared in the
recent literature. These terms are used in conflicting and overlapping ways
from one article to another. Some authors distinguish “drug-likeness” as
describing structural features, such as certain types of functional groups,
while they use the term “lead-likeness” when describing physical properties,
such as polar surface area. Other authors use the term “lead-likeness” to mean
the use of more restrictive criteria than those used for “drug-likeness,” although
they may be looking at the same features of molecular structure. Some authors
treat the two as being synonymous. This text will use the term “drug-likeness”
for all of the above types of molecular analysis.
One of the criteria most frequently used for drug-likeness is the Lipinski
rule of fives. This rule was not created to be a drug-likeness criterion,
but rather as a criterion for identifying molecules that will be orally bioavailable.
However, pharmaceutical companies push for orally bioavailable
drugs wherever possible, because patients, doctors, and insurance companies
prefer drugs in tablet form. Thus, the two criteria are synonymous
for many drug design projects. The Lipinski rule of fives states that for a
drug to be orally bioavailable, it should meet at least three of the following
criteria:
† The compound should have 5 or fewer hydrogen-bond donors.
† The compound should have a molecular weight of 500 or less.
† The substance should have a calculated log P (ClogP) less than or equal
to 5 (or MlogP 4.15).
† The compound should have 10 or fewer hydrogen-bond acceptors.
Lipinski et al., also note that compound classes that are substrates for biological
transporters are exceptions to the rule. Thus, these criteria only predict
passive intestinal absorption, not active transport. Other oral bioavailability
metrics can also be utilized as drug-likeness metrics. For more information
on oral bioavailability prediction, see Chapter 19.
Many drug likeness metrics have included a log P criterion. Programs to
compute log P quickly and reasonably accurately have been available for
many years. A number of researchers have noted that it would be more appropriate
to use log D, which takes into account the ionization states of the molecule
at a specified pH. Some use log D5.5 5 or log P, 6.25 as a substitute
term in the Lipinski rules. There are now programs for quickly and reliably
computing log D, such as that from ACD/Labs.
Some researchers will use similar rules for identifying reasons why a given
compound might not make a viable drug. For example, Schneider and
Baringhaus noted the following:
† A drug-like molecule should not have more than 5 rotatable bonds
(sometimes referred to as the fifth Lipinski rule).
† A drug-like molecule should not possess a polar surface area exceeding
120A°
2, since it would then not be sufficiently lipophilic to enter the bloodstream
through passive intestinal absorption in the small intestine.
† A drug-like molecule should not have an aqueous solubility (log S)
less than 24.
It should be noted that a perusal of the structures of approved drugs will reveal
a noticeable range of molecular weights, topologies, and functional groups. In
general, drugs that treat central nervous system disorders tend to be at the lower
end of the molecular weight range, with proportionally fewer heteroatoms and
few, if any, very polar functional groups. This may be in part due to the
additional constraint of having to pass through the blood–brain barrier.
Other orally administered low molecular weight drugs, such as some of the
broad spectrum antibiotics, tend to have a greater number of heteroatoms
and thus more polarized charge distributions. At the higher end of the molecular
weight range are steroids, hormones, and macrocyclic antibiotics.
A set of rules similar to the Lipinski rules was developed by Oprea, who
analyzed a somewhat larger database of molecules. He found that 70% of
drugs contain the following.
† 0–2 hydrogen-bond donors
† 2–9 hydrogen-bond acceptors
† 2–8 rotatable bonds
† 1–4 rings
Of course, it would be unwise to use these criteria so stringently that the other
30% of profitable drugs on the market were excluded from consideration.
However, it is reasonable to assume that molecules become less drug-like
(for most applications) as they display larger deviations from these metrics.
Oprea, Bologa, and Olah performed an analysis of a large number of
approved drugs from MDL’s Drug Data Report and the Physicians Desk
Reference to develop the following lead-likeness metric (note the similarity
to the Lipinski rules, except with tighter boundaries):
† molecular weight 460
† 4 ClogP 4.2
† log Sw 25
† number of rotatable bonds 10
† number of hydrogen bond donors 5
† number of hydrogen bond acceptors 9
Oprea et al., note that most drug-likeness and lead-likeness metrics are
developed by comparing databases of drugs with databases of commercially
available chemicals. Since approved drugs are typically more complex than
other chemicals on the market, these metrics become in some part a measure of
structural complexity. They also point out that some other properties to look
for are log D7.4 (log P for ionization states at pH 7.4) between 0 and 3, little
or no binding to cytochrome P450 enzymes, and plasma protein binding
below 99.5%.
Another approach to giving researchers an understanding of drug-likeness
is to look at structural motifs found in existing drugs. This is not necessarily
very useful as a library screening tool, but it can be very relevant to determining
what types of functional groups should or should not be included in
combinatorial library design and other synthesis efforts.
Several papers by Bemis and Murcko present a topological shape analysis
and structural analysis of the approved drugs in the Comprehensive Medicinal
Chemistry (CMC) database of approved drugs, excluding some nondrug listings
such as imaging agents and dental resins. The topological analysis looked
at connectivity only, ignoring element and bond order. Of the drugs analyzed,
94% contained rings, and the remaining 6% were acyclic molecules. There
were 1179 different ring topologies, 783 of which appeared in only a single
drug. There were 32 ring/linker topologies that appeared in 20 compounds
or more. Two of these contained a single ring, with the largest topological category
being drugs with a single six-membered ring. Of the 32 most common
topological categories, 18 contain fused ring systems, only 1 of which has a
bridgehead topology. Of the 32 categories, 18 contain two ring systems
linked together.
The subsequent analysis of drug frameworks by Bemis and Murcko takes
into account the identity of heteroatoms and bond orders. Forty-one such
frameworks are represented at least 10 times in the database. Of these, 25
have noncarbon atoms in the rings. These studies certainly illustrate the
importance of rings and heterocyclic compounds in drug design. They also
suggest some specific ring systems that drug designers would be wise to understand
and consider.
