CHAPTER 3

Combinatorial Chemistiy
DOUGLAS R. HENRY

The term f)arath/,'m s/si/i is an overused one. hut in the mid- use. In 1963. Merrifield introduced the efficient synthesis
1980s a true paradigm shift occurred in the way new drugs of peptides on a solid support or resin (Fig. 3-2).' This made
are synthesized and screened for activity. Prior to then, most the rapid, automated synthesis of peptides possible, and
drug compounds were synthesized in milligram quantities earned Merrilield a Nobel Prize in 1984. A key feature of
in a serial one-at-a-time fashion. After synthesis, the com- his approach is the attachment of a growing peptide chain
pound was sent to a biologist, who tested it in several in loan inert polymer bead. tisually about 100 4um in diameter.
vitro assays and returned the results to the chemist. Based composed of polystyrene cross-linked with divinyl bcnzcne.
on the assay results, the chemist would apply sonic struc- Such beads were originally designed for size exclusion chro-
ture—activity relationship (SARI or use chemical intuition matography. The beads can be immersed in solvents.
to decide what changes to make in future versions of the washed, heated. etc.. and when the synthesis is complete.
molecule to improve activity. Using this iterative process, a the beads can be filtered l'rorn solution, and the reaction
chemist would be able to synthesize only a handful of struc- products can be cleaved front the polymer. yielding pure
tures per week. Since the yield of marketable drugs from products. A Hungarian chemist. Arpad Furka, realized that
compounds synthesized and tested is only about I in 10.0(X). Merrifield's approach could be extended to allow the sytithe-
the road to success has been a long and expensive one, taking civ of all possible combinations of a given set of amino acids
6 to 12 years and costing S5(X) to $800 million per drug. in a limited number of steps. He accomplished this by split-
In the mid- 1980s. this approach to drug synthesis changed ting and remixing portions of the peptide-bound resin at each
dramatically with the introduction ol combinatorial chemis- step in the synthesis (Fig. 3-3). His description of the use
try. The drug discovery process becanic a highly parallel of combinatorial chemistry to synthesize polypeptides ap-
one, in which hundreds or even thousands of structures could peared in the Hungarian patent literature in 1982. Appar-
be synthesized at one time. Interestingly, biologists had for ently, it is the lirst literature reference to a combinatorial
some lime been using high-throughput screening HTS) to chemistry experiment.2
perform their in vitro assays, running assays in 96-well mi- As seen in Figure 3-3. the advantage of split-and-mix syn-
crotiter plates and even using laboratory robotics for pi- thesis is that all 27 tripeptides can be synthesized in just
petting and analysis. The bottleneck had become the synthe- three steps, instead of 27 steps. The disadvantage of this
sis of the compounds to test. Chemists realized that syntheses approach is that in the end, one obtains three mixtures of
could also be conducted by using a parallel approach. The beads with tripcptides attached, rather than the pure com-
term conrb,,satoru,l chemistry was coined to refer to the par— pounds themselves. If activity is detected in one of the mix-
aDd generation of all possible co,ubinaiions of substituents tures, it becomes necessary to go back and resynthesize some
uc components in a synthetic experiment. Whereas the yield or all of the structures in that mixture, to see which tripeptide
fmm a serial synthesis is a single compound. the yield from a is responsible for the activity. As we shall see, various meth-
combinatorial synthesis is a chemical liltrar. Figure 3-I ods for tagging and deconvoluting combinatorial libraries
shows two common types of chemical libraries—a generic have been devised that reduce or eliminate the need for re-
library, based on a single parent or scaffold structure and mul- synthesis.
tiplesubstituentsorresidues. and a mixture library, containing The first combinatorial chemistry experiments were ap-
a variety of structure types. The total number of structures in plied to the study of epitopes—the short sequences of amino
alibrary iseitherthe product of the various nunibersofsubstit- acids responsible for antibody recognition and binding to
nents (for a generic library) or the total number of structures proteins. Early researchers used solid-phase resin beads in
in a mixture. The goal of conihinatonal chemistry is to be able vials. microtiter plates. colutnns, and porous plastic mesh
to synthesize, purify, chemically analyze. and biologically "tea bags" They also used brush-like arrays of plastic pins.
test all the structures in the library, using a.s few synthetic cx- at the ends of which compounds could be synthesized. Other
perimenisas possible. This chapterdescrihes how combinato- media that have been used include paper and polymer sheets
rial chemistry and HTS are being used in drug design and dis- and glass chips—basically anything that can immobilize a
cosery to find new lead structures in a sluwter time. structure for the purpose of exposing it to reagents and sol-
vents (Fig. 3-4).
Peptides. of course, make poor oral drug molecules be-
HOW IT BEGAN: PEPTIDES AND OTHER cause they hydrolyze in the acidity of the stomach. As com-
LINEAR STRUCTURES binatorial methods were applied to the synthesis of drugs, a
need developed for methods of generating small (molecular
Combinatorial chemistry was first applied to the synthesis weight. <500) nonpeptide molecules as potential drug leads.
of peptides, since a convenient method for the automated Among the first alternatives that were studied were Chiron's
synthesis of these compounds already in svidesprcad "peptoids' '—molecules in which the variation occurs in
43
44 and Tt-',ibaok of Organic Medicinal and Pharn,areuziial C!wn,,.szrv

A1 = H, 4-OH

A2 = 5-Cl, 4-COOH+5C1

A3 = CH3. CH(CH3)2. CH2Ph(4-OH). CH2Ph,

A4 = H, CH3, CH2CH3. CH2CH=CH2, CH2Ph

Figure 3—1 • Generic (a) and mixture

(b) combinatorial libraries. The size of the
generic library is the product of the various
numbers of substituenis (here, 2 x 2 x 4
x 5. or 80). The size of the mixture library
is simply the total number of structures.

R2

R2
BOCNH
N
NH2 NI-IBOC
0
DCC 0
Figure 3—2 • Merrifield synthe-
sis on a polymer bead support.
The growing peptide chain is at-
tached to a polymer support, usu-
ally in the form of small beads.
The next amino acid (bearing R2)
is attached, and its protecting
group (BOC) is removed with CF3COOH Repeat
N
acid/base treatment. BOC. butyl- NH2
oxycarbonyl. DCC, dicyclohexyl- NEt3
carbodiimide.

Split & Mix

OAA Q-AB 0-AC

0-BA 0-BB 0-BC

& Mix
0-AAA Q-AAC
0-BAA 0-BAB 0-BAC
0-CAA 0-CAB 0-CAC
0-ABA 0-ABB 0-ABC
0-BBA 0-BBB 0-BBC
Q-CBA Q-CBB Q-CBC Figure 3—3 • Split-and-mix synthesis of tripeptides In the first step, all the
Q-ACc beads in a given container have a single monopeptide. These are all mixed
Q.ACA Q-ACB
together, then split into three aliquots and re-treated, attaching a second
Q-BCA 0-BCB 0-BCC peptide. After just one more step, all 27 possible combinations exist, spread
0-CCA Q-CCB Q-CCC among the three containers.
 Chapter 3 • Comb jmug,riul 45

Seal

Polypropylene

Resin

Label
b

•IeIIOSIS .•••••••
'OIl,.,.
.I,.,.,., ,,,ll•..
...,l,l• It
"I'.,
•IIIIIOSIS
d
FIgure 3—4 • Various approaches to immobilizing and separating chemical compounds during combinato-
rial synthesis have been devised, a. Tea bag synthesis. b. Pins and "lollypops. c. Dots on cellulose. d.
Spatial arrays on microchips

molecular <500) the attachment to the aunide nitro- in this way limited the total number of compounds to just
gee (Fig. Although these structures could potentially 204. Nevertheless, several potent ligands were found, includ-
place side chain functional groups in positions similar to ing a nanomolar a-adrencrgic inhibitor (Fig. 3-6a) and a
those on the corresponding peptides. they differ significantly similarly active ia-opiate receptor ligand (Fig. 3-6b).6 Be-
in that they lack peptide hydrogen bonds and chiral centeN. cause of the ease of synthesis, other classes of linear chain
They also show more rotational flexibility than the cone-
pcptides. since the peptoid amide bonds show less
0
double-bond character than those in peptides. Miller demon-
strated the stability of peptoids to a number of enzymes. 0
including chymotrypsin. papain. pepsin, and carboxypepti-
dase Zuckerman et demonstrated in 1994 that biolog-
ically active peptoids could be obtained by using combinato-
dat chemistry. He used 24 monomers to generate tripeptoids.
each of which had one hydroxylic. one aromatic, and one
diverse side chain. Limiting the composition of ihe peptoids

Peptide

b
Figure 3—6 • Biologically active peptoids. Compound a is an
Peptoid
a-adrenergic inhibitor, while compound b is a receptor
Figure 3—5 • Comparison of peptide and peptoid. ligand.
 _____

46 Wilson and Giscolds Textbook of Organic Medicinal and Pharmaceutical Chemisir

NH2

R3
R2 Attach R2 Fmoc P2 I

to solid amino Cycllze

support acid
— __O — __o
0

Ri RI

Figure 3—7 • Synthesis of t ,4-benzodiazepines on a solid support. Fmoc 9-fluorenylmethoxycarbonyl, a

common organic protecting group.

molecules have been investigated. These include oligonucle- alcohol-deterrent, uterine-relaxant, antineoplastic, anticon-
otides (DNA and RNA). oligoureas. and carbohydrates. vulsant. antiulcerative. analgesic. antiarthritic. and sedative
structures, among many others. For this reason, the demon-
stration of the solid-phase synthesis of these structures vir-
tually opened the door to the use of combinatorial chemistry
DRUG-likE MOLECULES in drug discovery.
The real advance in combinatorial chemistry for drug discov- In the decade since the first drug molecules were gener-
ery purposes was the introduction of synthetic methodology ated by using combinatorial chemistry, solid-phase synthe-
to yield true drug-like structures. Bunin and Ellman7 in 1992 ses have been discovered for most common classes of drug
demonstrated the synthesis of I .4-benzodiazepine com- structure. Some examples of these are shown in Figure 3-8.
pounds. using three components: a 2-aminobenzophenone, Of necessity, the reactions that can be performed in combina-
a protected amino acid, and an alkyl halide (Fig. 3-7). Al- torial chemistry are simpler than many reactions that a chem-
though we normally think of benzodiazepines as muscle re- ist using standard synthetic procedures can perform. Ex-
laxants and tranquilizers, a search for drug structures con- tremes of temperature and pressure, the use of highly caustic
taining the benzodiazepine scaffold returns antiviral, reagents, inert atmospheres, and multistep reactions are gen-

P4

HO

P2

Figure 3—8 • Examples of drug-like

structures that can be generated by solid-
phase synthesis. (Structures drawn from
Balkenhohi, F., et al.: Combinatorial syn-
thesis of small organic molecules. Angew.
P3 Chem. Int. Ed. EngI. 35:2288—2337,
1996.)
 Chapter 3 • Co,,,bi,nuorial C'Iie,nis:rv 47

avoided in combinatorial chemistry. Also, the reac- Other classes of therapeutically important compounds that
should not yicld alternative products. The yield of reac- have been synthesized successfully by using solid-phase
tions in combinatorial chemistry should be high (80% or combinatorial chemistry include carbohydrdtcs and naturdl
hut this can often be achieved by using an excess of products.9 Polysaccharides are important For various carbo-
Ivagent and then washing the beads afterward to remove hydrate—protein interactions. Carbohydrate antibiotics, in-
the excess. An important recent advance in combinatorial cluding vancomycin and aminoglycosides. have been the tar-
chemistry is the use of microwave heating in place of stan- gets of combinatorial chemistry, as well as complex
dard heating methods.5 oligosaccharides like the one shown in Figure 3-9.'° A van-

a HO b

B2O
B2
NHFmec
o 0

— R2

e R3

FIgure 3—9 • Carbohydrate and natural product synthetic targets. a. Bauhinia purpurea lectin ligand
analogues. b. Vitamin D3 analogues. c. Erythromycin analogues. d. Neocarzinostatin anticancer agents. a.
Galanthamine cholinesterase inhibitors. In most cases, the molecules are assembled by connecting large
subfragments in a small number of synthetic steps rather than by attempting total synthesis. Fmoc, 9-
fluorenytmethoxycarbonyl, a common organic protecting group.
48 Wilsoti and Gj.cvo/r/'x Textbook of Medicinal and Pharnzareuiieal Che,nixgr,

ely of natural products arc being studied, mainly in the areas swell in organic solvents, allowing the free diffusion of sol-
at' infectious disease and cancer but also as scaffolds for vent and reagent into the interior of the bead and greatly
many other therapeutic categories. Figure 3-9 shows exam- expanding the available area for the attachment of product.
ples of natural products that have been studied. including The polymers are inert, except the functional groups to
vitamin D analogues. crythromycin-like antibiotics, antican- which the molecules arc attached. In general, the compounds
cer neocarzinostatlns. and galanthamine. a cholinestcrasc in- to he synthesized arc not attached directly to the polymer
hibitor.° molecules. They are usually attached using a "linker"
moiety that (a) enables attachment in a way that can be
easily reversed without destroying the molecule that is being
SUPPORTS AND LINKERS synthesized and (b) allows some room for rotational freedom
of the molecules attached to the polymer. Sometimes, the
Most solid-state combinatorial chemistry is conducted by molecules attached to the polymer are used directly as sub-
using polymer beads IC) to 750 in diameter. These heads strates in in vitro assays without removing them from the

HF

Product

Memiteld resin (peplide products)

Produci

P AM resin (peptide or carboxylic acid products)

cl-c
Product

1% TFA

Trityl resin (carboxylic acid products)

-N Product

Nucleophile

NM BA resin (peptide products)

Figure 3—10 • Common linker functional

H2 I-I 1%TFA groups and the reagents that cleave the product.
Note that different tinkers provide different spac-
ing and steric freedom between the product anc
the support. HF, hydrofluoric acid; TFA, trifluoro
ADCC resin (amine products) acetic acid.
 Chapter 3 • ('mnhinc,torial CIu'n,isirv 49

polymer. In such cases. if the molecules are too tightly described above, solution-phase combinatorial chemistry
packed on the polymer. the enzyme molecules cannot gain often leads to a mixture of products. Imagine reacting a set
access to the substrates. A similar situation exists br access of 10 amincs with 10 acid chlorides, all in one flask, and
by sonic of the reagents and catalysts used to synthesize the with the reactants and conditions chosen so that no reaction
molecules. lit general. about I mmnol of linker is attached of amines with amines or chlorides with chlorides occurs.
per gram of solid support. The types of solid supports that only reactions between amines and chlorides. The result
are used include would be a mixture of 1(X) amides, one for each possible
combination of amine and acid chloride. The resulting mix-
• Polystyrene resins. Polystyrene cross-linked with divinyl ture could then be tested for activity, under the assumption
bcnzcnc ZibOLII I cross-linking>. These are common resins that the inactive amides did not interfere with binding of
used in site exclusion chromatography. active molecules (not always a valid assumption). If activity
• TentaGel resins. Polystyrene in which some of the phenyl is found, smaller subsets of amines and chlorides can he
groups have polyethylene glycol mPEG) groups attached in
tested to eventually find the structure(s) responsible fur ac-
position. Thc free OH groups of the PEG allow the
attachment of compounds to be synthesited. tivity.
• Polyacrylamide resins. Like "simper glue.' these resins swell Researchers have gone one step further by reacting multi-
betmer in polar solvents and, since they contain amide bonds. plc kinds of reactants together to produce some rather aniaz-
more closely resemble biological materials. ing mixtures. Figure 3-Il shows an example of a four-com-
• Glass and ceramic heads. Not a type of organic resin hut ponent Ugi reaction that yields, alter appropriate further
sometimes used when high-temperature or high-pressure reac- transformation of the intermediate product, a mixture of car-
tions are needed. boxylic acids, esters and thioesters. pyrroles. I .4-benzodi-
azepine-2.5-diones. and even a monosaccharide.'4 Despite
To support the attachment of a synthetic target, the poly- the diversity of the chemistry. the yields of products in such
mer is usually modified by equipping it with a linker or mixture-based experiments are often linind to he about 90%
anchor group. Such groups must be stable under the reaction or better. Although this is an extreme example of a multi-
conditions, but they must be susceptible to a "cleavage" component reaction, it illustrates the utility of solution-phase
reaction that allows removal of the product. Sonic common chemistry forgenerating great diversity in chemical libraries.
linkers arc shown in Figure 3-la. along with the reagents An approach that is intermediate between solid-phase
that cleave the prraiuct from the resin. chemistry and solution-phase chemistry is to use soluble
Some specialized linkers have been developed to meet
polymers as a support for the product. PEG is a common
particular reaction or product conditions. So-called Iraceless
vehicle in many pharmaceutical preparations. Depending on
linkers can be cleaved from the resin with no residual func-
the degree of polymerization. PEG can he liquid or solid at
tionality left. This allows the attachment of aryl and alkyl
room temperature and show varying degrees of soluhility in
products that do not have OH or NH functionality. These
aqueous and organic solvents. Each molecule of PEG has
linkers usually include a silyl group that is sen-
an OH group at either end:
sitive to acids amid can be cleaved to give unsubstituted
phenyl or alkyl products. A class of linkers known as
"safety-catch'' linkers are inert to the synthesis conditions
hut have to be chemically transformed to allow final libera- By converting one OH group to a methyl ether
lion of the product fromn the resin. Typically, two reactions (MeO—PEG—OH), it is possible to attach a carboxylic acid
are required to break the linker (hence the name). A rather functionality to the free OH and use solution-phase comuhina-
elegant approach to linker chemistry is to use linkers that tonal chemistry to synthesize, for example. iV-aryl-sulfon-
are sensitive to ultraviolet WV) light, The Affymux group amide structures.° The resulting mixture of PEG-bound sul-
has used these in the synthesis of carboxylic acid and carbox- fonamides can be separated by use of chromnatography.
amide products,L Finally, some groups have used linkers Another type of soluble support is dendnimers. These are
that can only he cleaved by cnzymues)' large, highly branched molecttles with terminal amino
groups that can be used like the OH groups of PEG fur the
attachment of products. Finally, a class of molecules known
as fluorous phases are a form of "liquid Teflon," consisting
SOLUTION-PHASE COMBINATORIAL mainly of long chains of (—CF2-) groups attached to a sili-
con atom. When these phases are used as a soluble support
CHEMISTRY
for synthesis, the resulting product can be readily separated
Most ordinary synthetic chemistry takes place in solution. from any organic solvents and reaction by-products by ex-
When a reaction must be modified to accommodate a solid tracting the reaction mixture with fluorocarbon solvents."
support, it takes time and resources to develop and optimize A unique application using complementary DNA as a "sup-
the reaction conditions. Indeed, a combinatorial chemist may port" has been reported by Harvard To "en-
spend months designing a solid-phase reaction and gathering courage" pairs of molecules in solution to react under mild
the necessary materials but then conduct the entire synthesis conditions, they attach short strands of complementary DNA
in a matter of hours or days! Many reactions cannot ever be or RNA to the structures to "zip" the structures together
run on solid supports because of poor yields or failed reac- and promote reaction. The DNA is then removed, yielding
tions. For these reasons, there has been much interest in product that would not otherwise be synthesized. Using this
using solution-phase chemistry for the preparation of combi- method makes it possible to prevent reaction of certain pairs
natorial libraries. Unlike one-bead one-compound synthesis of structures ax well.
50 Wilson wul Textbook of. Mediiinul and

R1—COOH + R2—NH2 t NC -

0 R3 H

0
R2
R2 0

0
/
R3
H
/ 0

Figure 3—11 u Four-component Ugi reaction reported by Keating and Armstrong."1 A combinatorial mix-
ture of the intermediate cyclohexenyl amide can be split into several portions, and each can be further
reacted to give a variety of products, all of which will be combinatorial. (Reaction redrawn from description
from Keating, T. A., and Armstrong. R. W: Molecular diversity via a convertible isocyanide in the Ugi four-
component condensation J. Am. Chem. Soc. 117:7842—7843, 1995.)

he several) in the third group. Since it is in the third group.

POOLING STRATEGIES know aC in position 3 iv needed for activity. We synthesii.e
a smaller library of the structures, in three groups: AAC +
Although some solid-phase combinatorial chemistry is con- RAC + ('AC. ARC + BI3C ± CRC. and ACC + BCC +
ducted by use of the one-bead one-compound strategy. CCC. We do this by skipping the last set of pooling shown
chemists have devised numerous other approaches to pool- in Figure 3-3. Now when we screen these we find
ing reactants and inlemsediales to getlerate libraries. The activity in the middle group of beads. This tells us that a It in
goal is generally to achieve a balance between the simplicity position 2 is required activity. The final step is to synthesise
of mixing everything together in one step but then having ABC. BI3C. and ('BC. keeping iheiti separate, and screen each,
to 'deconvolute" the resulting mixture and working with to find ABC a.s active structure.
• SubtractIve deconvolution. This is similar to iterative decon-
more, but smaller, mixtures. II has been likened to someone
solution but uses negative logic, namely, leave out a functional
giving you a rake and a magnet and telling you to go find group, and if activity is absent, the functional group that is
and describe a needle in a field of hay. You can make one missing must be needed for activity. This is particularly useful
big haystack you know contains the needle, then have to for quantitative structure—activity relationship (QSAR)-type
deal with ever-smaller "subhaystacks." or you can use more studies in which. say. a —Cl group is placed at several posi.
clever approaches, such as dividing the field into regions. tions on a phenyl ring. The entire library is screened as a
using overlapping regions. etc. The major approaches that mixture to get the activity level. If activity is detected.
have been used include the following: a set of sublibraries is prepared, with each missing one build-
ing block (subtr.ictiott ofa Functional group). Sublibraries that
• One-bead one-compound strategy. With this strategy, a spe- are missing functional groups front the active compound(s)
cific quantity of beads is allocated for cacti possible structure will be less active thait the parent library. The least active
in the library: those beads Contain only molecules of the given subhibraries the most important functional groups. A
library member. The beads may be tagged in various ways reduced library containing only these luncticinal groups is then
Isec the next section) to help identify the synthetic compound. prepared, and the most active compounds are identified by
The advantage of one-bead one-compound strategy is the viol- either one.compcnind synthesis or iterative dcconvolution.
plicity of analysis and screening. The disadvantage is keeping • Bogus-coin detection. This begins with generating and
the beads separate and having to deal with a large number of screening the ctitire library as a single mixture. If activity is
syntheses in parallel. As advances were made in robotics and detected. the building blocks are divided into three group,. (a,
automation. the problems were reduced, and today, probably fI. atid yt, and additional sublibrarics arc prepared. In these
most combinatorial experiments involve a one-head one-corn- subsets. the nuttther of building blocks From the a group is
pound strategy. decreased, the tiumber front (./ group is increased, and the
• Iterative deconvolution. This is the strategy ftrst described number front the y group is unchanged. The resulting effect
20 years ago when combinatorial chemistry was started. Reex- on acttvcty tup. down. ccnehanged suggests which group of
amine Figure 3-3 and imagine starting at the bottom nI the building blocks was conlributing most to activity. This ap.
Iigam with three groups of beads. Each group has beads bear- proach is applied iteratively to zoom in on the groups that are
ins a variety olcompounds. but a given structure only appears most active.
in one of the groups. Suppose lhc active structure is ABC (we • Orthogonal pxling. The tenhl orthrim,.o,tal means perpendicti-
pretend here there is only one—in reality there will probably lar or uncorrelatcd. In this type of pooling, we the
 Chapter 3 • Combinatorial Clwmis;n 51

functional groups to be considered into sets of libraries. A. B. ity for the stationary phase in the column, and they exit or
C. etc.. which can contain mixtures of the same compounds. elate from the column at difl'erent times. They are detected
However, the functional groups are distributed such that any by some optical method (UV absorption. fluorescence, re-
subset in A and B shares only one functional group. For exam-
fractive index. etc.) that gives rise to peaks on a graphical
ple, if we have a very small library of structures—au. ab. and
readout. Sometimes, the output from the column is passed
ac—we might put na and ab into group A, an and ac into
group B, and ab and ac into group C. If ab is the active sttuc- into a spectrophototneter or mass spectrometer to generate
tare, screening A. B. and C would show activity in A and C, a spectrum for each fraction of the output. These spectra can
but not in B. telling us that ub (the only structure in common) be interpreted to determine the structure of the compound
is the active one. that caused a given peak. It is also possible to use much
• PositIonal scanning. This is a noniterative screening strategy larger chromatographic columns and run preparative HPLC
in which a subset library is created with a single building to separate up to several milligrams of material for further
block fixed at one position and all building blocks in the other analysis or biological assay.
positions. In principle, by selecting the lunctional group from Chromatographic separations and analyses can be fully
the most active subset at each position. the most active com-
automated. Thus, a chemist can place all the reaction vessels.
pound overall is discovered. This ignores interactions between
building blocks, which complicate the results.
microtiter plates, etc. from a combinatorial experiment into
racks and use a robotic system to draw samples. inject them
Certain problems with mixtures must be considered when into the HPLC, and collect the data output into computer
pooling. Complex mixtures with only one or a few active file.s or databases—all without further intervention from the
stniclures can have solubility problems, especially if the chemist (except to wash the dishes!). For this reason, speed
compounds are poorly soluble. The inactive compounds con- and solvent handling are special concerns with combinatorial
tribute to the total ionic concentration but not to the activity. experiments. One approach that has been adopted to speed
Sometimes, compounds that have a common scaffold will up analyses and reduce the amount of solvent that must he
have many active species, arising from the scaffold and not consumed is .supercrilical fluid (SFC),
the substiluents. Thus, many poorly active structures may Here. the solvent is not a common organic solvent such as
show additivity of activity, leading us to think the mixture acetone or ethanol. Instead, it is a pressurized gas like CO2
contains a single active structure (false-positive results). Fi- that evaporates from the output. leaving pure compound be-
nally. partial binding of inactive structures can sometimes hind. Another advantage of SFC is speed: since the solvent
prevent an active structure from showing full activity (false- molecules are small, diffusion is rapid, and separations take
negative results). place in about half the time of ordinary HPLC separations
or less. Finally, the amount of "solvent" that is consumed is
significantly lower with SFC. A disadvantage is that certain
compounds may not separate as well under SFC as under
DETECTION. PURIFICATION. AND ANALYSIS i-IPLC.'8
JR spectroscopy is often applied in combinatorial chemis-
Detection, analysis, and purification of combinatorial librar- try. Since IR light can be reflected from materials, one can
ies places high demands on existing analytical techniques analyze resin beads directly, without cleaving the products
because (a) the quantities to be analyzed are very small, from them. Since the loading of product on any given bead is
sometimes picomoks of material, (b) the analysis should very small, usually computer-enhanced methods like Fourier
be nondestructive, to allow recovery of the compound if transform JR (FTIR) are needed to enhance the very small
possible, and (c) the methods must be suitable for rapid, spectral signal from one or a few beads. Interestingly, the
parallel analysis—analysis cannot be the rate-limiting step shape of the beads has been found to affect the IR spectra
in the procedure. No single analytical technique can lit all results, and flattened rather than spherical beads give
lIre requirements. so usually some "hyphenated" analytical stronger IR signals)9 NMR spectroscopy gives more struc-
techniques are used, for example. high-performance liquid tural information than IR or UV spectroscopy. but it has
chromatography with a mass spectrometer detection system traditionally not been nearly as sensitive. Compounds arc
IHPLC-MS). We describe this and other techniques in this normally cleaved from solid support before analysis by
section, NMR, since NMR on solid resin or on resin swollen by
Chromatography is usually the first step in the analysis solvent gives broadened peaks and low resolution. A type
of a combinatorial mixture. If we start with solid-phase of NMR called magic angle spinning NMR, in which the
chemistry, we chemically cleave the compounds from the sample is inserted into the magnetic field at an angle of
support and filter off the beads, giving a solution containing about 550, reduces the peak broadening and has been used
the compounds we synthesized. If the solution contains just to analyze swollen polymer beads directly. Recent improve-
a single compound. we might use a spectrophotometer, to ments and the use of "nanoprobcs" have allowed NMR
measure infrared (IR) and ultraviolet (UV) absorbance or analysis of I00-mjz beads bearing less than 80() pniol of
fluorescence directly, or even nuclear magnetic resonance compound. Other NMR techniques that have been used to
INMR) spectroscopy. to determine the structure of the com- analyze combinatorial mixtures include various "two-di-
pound in solution. If the solution contains a mixture of com- mensional" (2D) NMR techniques that use multiple bag-
pounds, one must separate them before determining their netic fields. HPLC-NMR, capillary ekctrophoresis coupled
structures. HPLC is a standard approach. A sample of the to NMR (CE-NMR), and even NMR to detect the binding
mixture is injected into the flow of solvent entering a chro- of drugs to receptors to identify active agents. This latter
matographic column. The components in the mixture travel technique has been termed SAR mm'iI!m NMR.24t
down the column at different rates, depending on their affin- Mass spectrometry (MS) is the technique most widely
52 tViIxo,i alul Gi.cvold s jeribr,ok of !ile'diri,,aI and PI,ar,nace,aical ('he,njsiry

used for combinatorial library analysis. The measurements 0 H

can he made on resin beads directly, a wide range of com-
Peplide library
pounds can be analyzed, and MS analyses can be highly
automated. Included among a number of MS techniques in
C)— Linker — N"-'
H
use arc
a Primer-DNA tag.Primer
Electrospray ionization. A solution containing the corn-
pounds to be analy,.ed is passed into a macs spcctronteter
through an electrically charged capillaty. The droplets that Linker2 — N — Peptide library
H
emerge from lime capillary hear strong electric charges theft-
selves, and they literally "exphxlc" into smaller and smaller Linker, — Primer-DNA lag.Primer
droplets and eventually into singly charged ions that are de-
tected by the mass spcctromemer.
linker2 — N — Peptide library
• Matrlx.as,sisted laser desorptionllonlzation tlmc.of-flight H
(MALI)l-TOF). Quite a mouthful, ii simply means the sample (
is embedded in some solid matris (e.g.. 2.5.dihydroxybenzoic Linker2— N — Peplicte library
acid) and then bonibanied with a laser. Sample ntokcules axe H
vaporized and ionized in a 'gentle' fashion that allows Figure 3—12 • Two ways of attaching DNA tags to solid sup-
whole-molecule ions of the sample to be analyzed. The analy- ports. a. A DNA tag is attached with each peptide molecule via
sis is dune with use of a time-of-flight analyzer, in which ions a bifunctional serine residue. b. The DNA and peptide groups
of different mass travel different distances in a given amount have separate linkers
of lime.
• Other less-used MS techniques. These include secondary.
ion MS (SIMS) in which the sample is hit by a metal ion
rather than tIme electron beam itself, and Fourier translorm MS. to beads on which peptides or pcptoids were being built.2'
Since there are 20 possible amino acids and only four nuelco-
A very important use of MS in combinatorial chemistry tide bases, enough bases must be attached at each amino
is in quality control of combinatorial libraries. As much as acid addition step to identify properly the amino acid being
possible. we would like to have pure compounds generated attached. Although three bases are used in the DNA genetic
in high yield, with no side reactions or by-products. We code, it is customary to use up to six bases for library tagging.
also need to verify that every component actually exists in For decoding. the DNA tag is amplilied by use of the poly-
a library (i.e.. that no reactions failed). Only MS provides ,nerase ihau, reunion (PCR). the same reaction that is used
the sensitivity and versatility to perform this checking with in forensic DNA analysis. For this reason, the chemical tag
both solid-phase and solution-phase libraries must also bear PCR primer sequences.
TWO types of anchors have been used to connect the DNA
tags to the solid support (Fig. 3-12). (none type, the growing
DNA chain is attached to the a carbon of a serine group
ENCODING COMBINATORIAL LIBRARIES that is anchored to the solid support by a linker molecule.
The growing peptide chain is attached to the serine atnino
Once we have found a mixture or sublibrary that shows bio- group, possibly through a spacer. In the second type of an-
logical activity, how do we determine exactly which struc- chor, the DNA chain has its own anchor to the solid support.
ture or structures are responsible for the activity? We can in this case, fewer DNA tags are attached than the number
purify and analyze as described in the previous section. but of polypeptide molecules.22
if no direct analysis is available, we need to encode or tag If DNA tags cannot be used, one can label beads by using
the support or the molecules themselves. using physical or a "binary" approach. Suppose we are bttilding a tripeplide
molecular "barcodes." An obvious approach that can be with four possible amino acids at each position. We can use
used only with small libraries is to physically label each vial binary digits to encode which amino acid is at a given posi-
of one-bead one-compound resin. This may be practical for tion us follows (each binary "number" is read from the
a few lens of compounds, hut what if we have a library right) (Table 3-I): Thus, using 1K different tags (3 x 6) we
of 32.(X)0 compounds, or even I .000,000 compounds. in a can encode br any of the (4' = 64) members of the library.
mixture? Clearly. there is a need for a more automated means For example, if the product is Ala-Gly-Lys. the encoding
of identifying the structures that are in the library.
The most common approach to encoding solid-phase li-
braries is to attach a chemical tag to the resin beads as the
target molecule gets synthesized. 'rypically. at each step in TABLE 3-1 Binary Encoding of a Tripeptide.
the reaction, a tag is attached that is unique for the given UsIng 18 PossIble Tags
step. For example, if we are creating a tripeptide and we
have 10 possible amino acids at each position, we need to Amino Add Position I PosItion 2 Position 3
attach either a single tag that says tripeptide on this
bead has amino acid Ala at position I, Phc al position 2, Ala 000000 (N) 1*) OIl IX) (10 (Ni

and Gly at position 3." or we need to attach three different Plie 000001 1)00100 010000
tags, one for each position. Cay 000010 11010(5) 100015)
One of the earliest types of chemical encoding was the l.ys (1000 II 00 11(51 110000
attachment of oligonucleotides (usually single-strand DNA)
 Chapter 3 • C€nnbinumria! Oiemisirv 53

would be 00000000100011 0000, and 3 of the 18 higher-density screening platforms. The standard layout for
possible tags would be attached to the support, along with HTS has been a 96-well microtiter plate (12 x 8). Denser
the tripeptide. It is common to use polyhalogenated aromatic formats, up to 1,536 wells per plate, are increasingly being
compounds as lags, such as used. This requires advances in liquid handling, precision
of detection, and laboratory automation.
One of the first activities in developing a HTS assay is
selecting the target. About 500 targets are currently being
used by drug companies. Of these, cell membrane receptors.
mostly G-protein—coupled receptors, make up the largest
group (about 45% of the total). Enzymes make up the next
largest group (28%), followed by hormones (lIck), un-
knowns (7%). ion channels (5%), nuclear receptors (2%).
and finally DNA (2%).25 It is expected that the annotation
where X represents some combination of halogen atoms. The of the human genome will add additional targets. although
halogens make the tags show up clearly in MS analysis of the rate of this addition is not known. New targets must he
the mixture, and by varying the chain length a, the tag can part of some regulatory pathway in the cell and should be
be made flexible enough not to interfere with attachment of sensitive to some disease state, not be expressed all the time
the product. Other chemical tags that have been used include and everywhere in the cell.
isotopically labeled peptides and dyes. The next concern in I-IFS is the library to be screened.
When it is not possible to use chemical tags, one must Throughout our discussion, we have perhaps offered the
physically label the solid or liquid support itself. One alterna- impression that a given library for a particular project was
tive is to use radiofrequency encoding, in which tiny mi- the only set of compounds that were ever screened for activ-
crochips are added to the resin or to the solution phase. As ity. In fact, much HTS involves screening compounds that
various reactions are conducted to generate the products, at
are part of the corporate storehouse of compounds synthe-
each step a radiofrequency signal is stored in the microchip.
sized in the past, or they may be a library purchased from
This signal can be recalled to identify the sequence of reac-
a vendor. Such libraries usually Consist of microtiter plates
tions that generated the product (a similar principle is used
containing frozen or dried samples of compound—perhaps
when your dog or cat gets a small identification (ID) pellet
only micrograms per well. The size of such libraries may
implanted under the skin of the neck). Laser optical encoding
range from a few thousand compounds to nearly a million.
is yet another approach, in which the solid support consists
The cost of completely screening such a library against just
of a ceramic chip covered with a polypropylene—polystyrene
a single assay may amount to over $300,000, so such large-
polymer solid phase. The barcode pattern is actually burned
scale screens are conducted rather infrequently, compared
into the ceramic at each step in the reaction and is decoded
visually with use of a microscope. Finally, one can embed with routine day-to-day screens. It has been estimated that
semiconductor particles into the solid phase that fluoresce one must screen at least 120,000 "quality" compounds (i.e..
at different wavelengths. These are called "quantum dots" diverse drug-like structures) to discover a single-lead
by their manufacturer.23
for a therapeutically sound target.25'
As discussed above in the section on pooling strategies.
one can reduce the screening effort by pooling groups of
structures and running assays on mixtures of compounds.
This also conserves reagents and biological material, has
HIGH-THROUGHPUT SCREENING (HTS)
smaller storage requirements, and requires fewer personnel.
Without the ability to screen libraries rapidly for activity, There are potential problems with pooling. A number of
there would be no combinatorial chemistry. Fortunately, the factors limit the number of different compounds we can test
biologists are just as adept at developing rapid high-through- in a given well, including ionization, reactivity, and solubil-
pat assays as the chemists are at generating structures. HTS ity. Compounds can enter a screening program in a nonran-
is an extremely broad topic. encompassing enzymes. organ- dom order, such that a given assay plate may have com-
elks, cells, various tissues, whole organs, and even whole- pounds that are highly similar structurally. This may give
animal testing, via cassette dosing. This section briefly dis- rise to false-positive hits. False-negative hits are less likely
cusses only a part of the role of HTS in drug discovery, with to arise from pooling. Another concern is the use of repli-
emphasis on a few recent cates—compounds from the same series—in a given assay.
Successful HTS programs integrate several activities, in- If only one representative of a given series is present. the
cluding target identification (genomics and molecular biol- chance of missing that series as a possible lead series is
ogy groups), reagent preparation (protein expression and Pu- greater than if multiple members are present. Therefore, it
rification groups), compound management (information is common to include several members of each series in a
management group), assay development (biologist and phar- given assay when possible.
macologist), and high-throughput library screening (biolo- To be effective, a given compound must dissolve com-
gists and chemists). Formerly, these activities were handled pletely in the assay medium. It is common to add a small
separately, and multiple handoffs of samples were involved. amount (1%) of dimethyl sulfoxide (DMSO) to the assay to
It is becoming more common to integrate the activities and assist solvation. The best concentration of compound to use
share expertise. This increases efficiency of the screening is somewhat debatable. High concentrations (10 jaM and
process. Another route to increasing efficiency is a move to above) often lead to more false positives than screening at
54 Wilson and Gi.;wld's of Organic Medicinal and P/,arnzaceutieal C/wn,is:rv

a low concentration (3 suM). The reason for this may be .cc-in:illant—a compound that fluoresces in the near presence
nonspecific binding at the higher concentration. of the radioactive substrate (near being about 20 zrn)—is
Just as there are several ways to dctcct and identify mem- added to the mixture. The lysed and unlysed substrate binds
bers of a combinatorial library, there are many ways so mea- to the beads, and if the radioactive part of the substrate is
sure activity in I-ITS assays. Any such method must be accu- still attached, the bead will fluoresce. If not, the radioactive
rate. reproducible. and have a high signal.to-noise ratio (SI parts of the substrate floating in the solution will be too far
N). Typically, the result of FITS is a qualitative (yes/no) from the beads to cause any fluorescence. The presence of
or semiquantitative one (high-medium-low), rather than a fluorescence implies that the test compound inhibited the
precise value (e.g.. K180 or The methods for detection enzyme. The advantage of SPA over filtration is that no
in FITS fall into the categories of nonradiometric and radio. filtering of the solution is needed, so beads can be added
metric. directly to the assay mixture in wells or test tubes. Also.
Nonradiometric methods include absorbance, fluores- special scintillation fluid is not needed. The beads for SPA
cence, and luminescence spectroscopy. Enzyme assays are can be engineered to attach a variety of substrate types.
a common example. The assay is usually run at or below Other HTS assay advances include the use of microorga-
the value of substrate. svith only about 5% of the nisms such as bacteria and yeast, the cloning and expression
substrate consumed during the assay, and multiple enzyme of mammalian receptors in microorganisms, probing pro-
turnovers occur during the assay. Sometimes enzyme reac- tein—protein interactions, and very importantly. DNA and
tions arc coupled, especially if the target reaction does not protein arrays. These are too involved to discuss here, but
produce a product that can be detected directly in the assay. excellent reviews exist.24 V The increasing use of HTS to
An example is carboxypeptidase. which is coupled to the screen for a molecule's absorption, distribution, metabolism.
reduction of NADP to NADPH. giving rise to absorbance excretion, and toxicity (ADMET) properties has been cov-
at 340 nm. ered as well.28
Radiometric methods include filtration and scintillation
proximity assay (SPA). These assays use radioisotopes. so
safe storage and handling are of concern. In filtration assay. VIRTUAL (IN SILICO) SCREENING
a radioactive substrate bound to a capture group is cleaved
by its enzyme. removing the radioactivity from the capture Virtual, or in silico, screening refers to the use of computers
group. The mixture is filtered through special filter paper to predict whether a compound will show desired properties
that the capture group sticks to. but everything else passes or activity on the basis of its two-dimensional (2D) or three-
through. A scintillation fluid is added, and the radioactivity dimensional (3D) chemical structure or its physicochemical
of the filter is measured. The degree to which the radioactiv- properties. The motivation for using virtual screening arises
ity is retained measures the strength of the inhibition (Fig. from the flood of new structures coming from combinatorial
3-l3a). chemistry. the expense and time required to run conventional
SPA is a newer, simpler method (Fig. 3-I3b). We start HTS. the ethical concerns about using animal tissue instead
with the same radioactive substrate, which may not necessar- of predictive models, and an increasing failure rate for struc-
ily need a capture group. The enzyme and potential drug are tures coming out of combinatorial programs. In general, a
added, causing the cleavage of the substrate to some degree. virtual screening program attempts to answer one or both of
Now, instead of filtering, a special resin bead coated with a these questions:

Enzyme CO—Substrate, Substrate(i ).—__ CC—Substrate,

CO—Substrate, * Substrate(')

Enzyme inhibitor CO—Substrate, — SubsIrate(')

CO—Substrate, - Substrate(')

FILTER
a

Enzyme CG—Substiale, Subslrate(') CO—Substrate,

Substrate(')
CO—Substrate,
CG—Substrale, — Substrate(')
Enzyme I nhlbltor CO—Substrate, — Substrate(')

b SPA Effect

Figure 3—13 • Comparison of filtration and scintiUation proximity assays. a. In filtration assay, the enzyme.
substrate, and inhibitor are mixed; the uninhibited enzyme splits the radioactive portion (*) off the substrate,
and filtering the mixture, followed by measuring the radioactivity of the filter, tells how much inhibition
has occurred. b. In SPA, the same mixture is treated with resin beads containing a scintillant that fluoresces
only in close proximity to the radioactive source. Any radioactivity that was split off by the enzyme does
not need to be filtered in SPA.
 Chapter 3 U Cwnbi,wrorial CIw,nistrv 55

I. Will i particulur compound show sufficient binding iou known drugs. This led Lipinski to enumerate some rules for the
rL'ccpisr? rejection of structures. He proposed rejecting any structures
2 Will a particular compound possess any undesirable ADMET that fail two or more of the following criteria:
properties?

• Molecular weight should be <5(X).

To answer these questions, we must build computer • Number of hydrogen bond donors (NH. OH should be fewer
utodds of the interaction of drugs with receptors (docking, than 5.
molecukir modeling, quantum mechanics) and models for • Number of hydrogen bond acceptors t—N—. —0—.
predicting ADMET properties on the basis of chemical struc- —S-—) should be fewer than 10.
ture (QSAR models). Like much of drug discovery, the vir- • Calculated log P value should be less than 5.
tual screening process is a cyclic otie. We stan with a collec-
Since the number 5 shows up in many of these criteria.
tion of structures and run predictive models on them.
they have become known as the "rule of five." These and
generating some subset of "best" structures. We then test
similar rules have become widely adopted by drug compa-
these slnictUres in real assays or screens to see if the predic-
nies in their discovery programs. They ball under the general
tions were accurate. Finally, we incorporate information
category of "business rules' '—guidelines for the proper
learned From the real assays back into our predictive models
conduct of business and research. One should be cautiouS
to improve them for future useY1
though and use them as guidelines and not strict rules, to
Models tor the prediction of binding arise from the field
avoid missing important drug structures that happen to fail
of molecular modeling. This includes molecular mechanics
the rules but still show high activity and acceptable ADMET
predicting the 3D structure olmolecules from the standpoint
properties.34 The "rule of five" is designed to be a yes/no
of the atondc nuclei) and quantum mechanics (predicting
filter for the rejection of structures. More quantitative
the 3D structure of molecules from the standpoint of their
models designed to predict some value or level of property
electrons). We can generate fairly accurate 3D structures of
have also been developed. Examples include predicting
stand-alone molecules with either of these approaches. We
Caco-2 cell permeability (Caco-2 cells are human intestinal
can also model the interaction of tWo molecules and predict
epithelial cells that can be layered and studied in vitro)33
whether they will show hydrogen bonding or electrostatic
and predicting binding to cytochrume P-450 (a major liver
or lipophilie interactions, If we know (he 3D structure of a
enzyme involved in metabolism),35
receptor, we can predict whether a given compound will
"fit'' into the receptor with sufficiently tight binding to pre-
sent normal substrates from binding. i.e.. an enzyme inhibi-
tor (or drug). Typically, this does not consider any effects CHEMICAL DIVERSITY AND LIBRARY
of transport, metabolism, interaction with solvent. etc. Con-
DESIGN
sequently. the existing models for the binding of drugs to
receptors are rather crude, with errors of 50% or more. The universe of drug-molecule--sized compounds that could
Nevertheless, for the purpose of screening, this is often ade- be synthesized is enormous. For example, a polypcptidc with
quate. II a structure is predicted to show tight binding, it 100 residues could regenerate I 005° or 1040 possible struc-
will probably be synthesiied and testcd in a real HTS assay. tures. If just I g of each were synthesized, it would he a
loobtain a more accurate estimate of its activity.30 An inter- mass comparable to the mass of the known universe (about
esting application of high-throughput docking is the United 1050 kgh) (Peterson. J,: Universe in the Balance. jVeii' Scien-
Devices screensuver.'5t This is a PC program that can be tist 168:26. 2000), Clearly, a combinatorial chemist nced.s
downloaded to run in the background and participate in to select carefully the scaffolds and building blocks that will
worldwide projects to screen large databases of structures make up the library to be synthesized. Since the 1980s.
again.ct such targets as cancer and anthrax. chemists have adopted many computational techniques to
We can combine the prediction of binding to a receptor aid in the design and selection of library structures. These
with the design ofomolecules. lfwe start with a set of building techniques have been borrowed largely from the fields of
blocks and a known receptor, we can align complementary QSAR and molecular modeling. This section describes snore
building blocks with pockets in the receptor, in conforma- of the most common computational approaches to library
tions that masintize hydrogen bonding and other interac- design.
lions, When the fragments are aligned, we can then connect Typically in drug design there are lour possible scenarios
the fragtiients with appropriately sited spacer fragments to for the amount of information we are starting with37
"build" a drug molecule within the confines of the receptor.
lids approach is sometimes called de not'u or structure- I. We do not know the structure of the receptor, and we do not
hosed drug design. As our models for binding to receptors have any known ligands or inhibitors. This is mainly a case for
inuiprove. this approach is becoming more popular. ' high-throughput in vitro screening nt evisting drug databa.scs
In 1997. researchers at Ptizcr looked at the previous 10 and libraries, in the hope thai sonic structure may hind to the
scars' worth of drug design. including all of their combinato- receptor and serve as a sturtittg point for lead development.
Often it is straightforward to (md existing structures with micro.
rial chemistry efforts.33 They found that since the introduc-
molar potency. The problem then becomes one ot deciding how
tion of combinatorial chemistry, chemists had tended to de- to modify the structures to yield ones that arc novel from a
sign larger. more complex, and more lipophilic structures patent standpoint and show suitable hiophannaceutical proper.
thutu in mite past. .Siiice the structures were SUpposed to be ties.
desioned as lead structures rather than optimized drug mole- 2. We do not know the structure of the receptor. hut we do have
euks, it was becoming timore diflicult to optimize them into known ligands or inhibitors. This was once the most commotu
56 Wilson and Gisvold'x Textbook of Medicinal and Plior,nacriaical Clu'n,,srrv

situation in drug design. The usual approach to selecting since- brury. we want to look for small changes in the httsic struc-
lures br Further testing is to lind structures in chemical data- ture, to find the best comnhintttion of suhstituents that will
h;ises that are similar in structure and properties to the known yield the highest activity.
inhibitors. A common approach is to develop a phannacophscre The first task is to define and qttantify chetnical dit'er.sit.
model (icr the receptor, based on supenmposing structures with
This requires a delinition oichemicul .thnilariit. since diver-
known activity. This can then be used as a scan!i query in a
chemical structure database to find structures that have the same
sity is essentially the opposite of similarity. There are several
functional groups in the same relative positions. ways to quantify tuolecular sitnilarity. Two common ap-
3. We do know the structure of the receptor and we base known proaches in combinatorial chemistry are (a) to define simi-
ligands or inhibitors. This is increasingly becoming the state of larity as the closeness of structttres to each other in the space
affairs br known receptors. There are about 100 known recep- of some physicochensical or topological elcscnptoni (e.g..
tor—ligand complexes published in the Protein Data Bank that log P. solubility. or polar molecular surface area) and (hj to
have relevance to existing human illness. Pharmaceutical com- delitie similarity as the nutnber olsimple structural features
panics have many more examples. As soon as one drug company the compounds have in common (e.g.. carbonyl groups.
patents a drug with activity for a known receptor, other compa- phenyl rings. etc.).
nies are qtmick to use this information to try to develop novel
Consider the molecules in Table 3-2. If the first molecule,
structures with greaier potency, longer-lasting activity, fewer
side effects. etc. This is c.spccially (rue if the market for the
diethylstilbestrol. is used as a similarity probe in a database
agent is a large one. The techniques for finding new drugs in this of drug structures. several other structures can be found
situation combine (a) pharmacophore discovery using known (some of which are estrogens, hut many of which have other
ligands with (b) docking and molecular modeling of new struc- activities). The tuost similar structures do have estrogenic
tures to see how well they fit into the receptor. This is knowit activity, and they have values similar to those of some of
as virtual and it is becoming more important all the the other descriptors in the table. Sonie descriptors by them-
time. The quality of virtual screening results depends on the selves would be poor predictors of molecular similarity. For
quality of the protein and drug molecule structures we start with example. t)OC can bind many structures with sitnilar log P
and the accuracy of the mathematical functions that predict the values but with different structural characteristics. For this
degree of binding from such factors as steric. electrostatic, hy-
reason, there has been some controversy about what the
drogen bonding, and lipophilic interactions between the drug
''best'' approach to similarity and diversity is.38 In the de-
and the receptor.
4. In the final scenario, we know the structure of the receptor, but sign of exploratory libraries, the goal is to sample a wide
we do not have known ligands or inhibitors. This is a less. likely variety of chemical structures, so it is reasonable to use di-
case than scenario 3, but the decoding of the hutnan genome versity measures that focus on structural differences, such
may change that. 01 the 3(1.0(X) or more genes that arc being as 2D molecttlar similarity. We select molecules that are
annotuted front the genome. it is estimated that perhaps dissimilar to each tither. In the design of optimization librar-
may have some relevance to human disease. This would repre- ies, the gnal is often twolold: to increase potency and to
sent 3.000 new enzyme targets for drug designers to work with. optimize ADMET properties.Minor changes in the structure
Researchers expect 10 or more new targets to be elucidated each arc usually made at this point, so all the structures show
year. so this promises to be a long-term project. In many cases,
high 21) similarity to each other. But tninor changes in sub-
the enzyme or structural protein that a gene encodes may be
dirncult to crystallize for x-ray analysis or to study in solution stituents can change physicochemical properties and modify
using NMR. In such cases, protein molecular modeling may be the interaction of the drug with the receptor. so the other
used to predict the structure of the enzyme. A common approach descriptors in Table 3-2 are <then more useful in the lead
is lmonu,logy modeling, in which sequences of the unknown pro. optimization stage of a project. An example of a simple
1cm lime assigned secondary and tertiary structttrcs based on change that causes a large decrease in binding affinity is
those oh' known proteins with similar sequences. IBM has em- seen in Figure 3-14. where the replacement of a H atom by
barked on a project known as "Blue Gene" to try to predict a methyl group reduces activity 80-IbId.39
the structures and receptor Sites of all the protein products of
the human genome. over a period of a few years, using super-
computers. Once the enzyme structures are known or predicted.
virtual screening can be used to find drug molecules that might
tit into the receptor.

Assuming that one of (he above scenarios holds for a given

drug discovery project. one commonly proceeds through a
sequence of chemical libraries on the way to lead discovery
and optimization. This process has several phases, and the
libraries that are designed at any given phase show varying
levels of diversity. size, and specificity for the given recep-
tor. The goal of library design is to select a subsel of mole-
cules from some larger collection such that we are taking CH3
samples from various regions of chemical space. If the li-
brary is an initial or exploratory library, we want to sample
as much chemical space as possible, so we seek structures R - H (6.7 nM)
with high diversity, If the library is a morefo<-u.sed library. R Me (470 nM)
we want to sample a smaller region of chemical space in Figure 3—14 • Changing a H atom to a methyl group has a
the neighborhood of the most active structure discovered in large effect on binding to the ru-aminobutyric acid (GABAA}
the exploratory library. If the library is an oplimization Ii- receptor.
 Chapter 3 U ChL',ni.r:r% 57

TABLE 3-2 SimIlarity of Sthactures to Diethyistilbestrol (DES). Using Various Descriptors

Polar No. of No. of No. of 20 Molecular
Surface N-Bond H-Bond Rotatable Similarity
Molecule Log P Area Acceptors Donors Bonds to DES

40.5 2 2 2 100

7.91 29.5 I 2 5 76

6.7K 52.6 0 4 8 56

5.68 40.5 2 2 5 4))

5.34 20.2 I I I) 36

5.58 40.5 2 2 4 33
58 Wilson (i.srold'.s Te.ttl,ook of Organic Medicinal and I'han,uus'iaica! Che,nistrv

Having decided on some measure(s) of diversity. the next designation is just I or (I). We start by generating random
step is to actually select structures from the starting collec- strings of Is and Os to represent the genome. Each string is
tion. Typically, we select either whole compounds for HTS measured for the diversity the structures in the string represent.
or reagents to be used in a particular synthetic step. The This is called the fitswss of the string. Then, we apply the
genetic operations of mutation. crossover, and reconihination.
selection process consists of using similarity/diversity mea-
to generate new populations of Is and Os.Some of these will
sures to pick compounds that sample chemical space in a have higher fitness values (more diversity) than others. Over
manner appropriate for the library being generated. Thus. time, as we repeat the genetic operations, the overall genome
for an exploratory library, we would pick diverse structures will tend It) improve toward more diverse collections of struc-
with as much structural variation as possible. perhaps subject tures.
to some rules about size. lipophilicity. etc. For an optimiza-
tion library, we would pick substituents for a given scaffold Most of these methods, including calculating descriptors
that span traditional QSAR space. which includes steric, and measuring molecular similarity arc part of combinatorial
electronic, and lipophilic properties of the substituents. chemistry software systems. These systems are provided by
The methods of selecting structures come mainly from molecular modeling and by chemical information companies
statistics and QSAR.4° They include the following: such as Accelrys. Daylight. MDL. and Tripos. They usually
function in the context of a chemical structure database. Such
• Random selection.Here we let the computer pick structures
databases can store 2D structures. 3D models, reactions, ge-
at random from the initial collection. We may "bias" the
r.mdom selection by littering the structures as they are picked neric structures, building blocks, and all the physicochemical
and rejecting ones that we know we do not want or that are and inventory data in a single repository. sometimes called
too similar to structures already picked. a duo, ;t'are/rou.se. Chemical data are indexed and accessed
• Visual selection. If we have selected, for example, 10 descrip- by structure or structure ID. Biological data are typically
tors for our chemical space, it is difficult to lind display meth- indexed and accessed by assay or test ID. For this reason.
ods that can display data in high dimensions. There are projec- most drug companies have traditionally stored chemical and
tion methods in statistics that can make a "shadow" of the biologicitl data separately and used different systems for ac-
points in high-dimensional space onto two or three dimen- cess. In the past few years. it has become common to store
sions, at which point a standard two-dimensional (2D) or both chemical structures and biology data in relational data-
three-dimensional (3D) scaflerplot can be used to see how the
bases such as Oracle. This trend toward integration of the
points are distributed and to select compounds. A standard
method for this projection is principal campollenix onalssis two types of data is motivated in large part by the flood
(PCA). In this approach. we generate a couple of new descrip- of information that combinatorial chemistry and HTS are
tors, the principal components. that are linear combinations generating. Most large pharmaceutical companies are syn-
of the original descriptors (pe, = w,x1 + + --) with thesizing and testing times as many structures today
each descriptor (v) weighted 1w) according to how much it as they were 15 to 20 years ago. with a similar increase in
contributes to the overall variation of the data. the amount of information that must be collected, organized.
• Binning. If we partition chemical space into regions, like stored, utid interpreted. The drug companies are taking a cue
squares on a checkerboard, we can pick molecules from each from large retailers and financial vendors and adopting "data
region to build our library. Probably some regions will havc mining" techniques to search for hidden associations, clus-
many structures in them, and we may pick several from such
ters, and predictive relationships in the mountains of data
areas. Some regions will be empty. These are termed hok.s in
our collection, anti we may want to design structures with they arc collecting itt their databases.4'
property values that would help fill these holes.
• Cluster analysis. This method involves using statistical pro-
cedures that try to discover natural groupings of compounds
on the basis of similarity or distance between them. Cluster REPORT CARD ON COMBINATORIAL
methods function either by partitioning space on the basis of CHEMISTRY: HAS IT WORKED?
the density of the compounds (e.g.. K-means clustering) or by
linking the compounds in a tree-like structure (hierarchical A report published in 1998 showed that virtually all of time
clustering). The result is to assign each compound to a group major pharmaceutical firms had adopted combinatorial
or clustcr. Compounds within a cluster are more similar or
chemistry to some extent in their drug discovery process.42
closer to each other, on average, than to compounds in other
clusters.
The degree of adoption ranged from 16 to 100% of new
• Experimental design. If we have selected a subset of struc- drug synthesis, with an average of 66%. Clearly, the phanna-
tures from our collection, we can use any of several statistical ceutical firms have a big stake in combinatorial chemistry,
measures to quantify the diversity of our subset. Further, there but has it really worked out? As of this writing, the phamia-
are statistical selection procedures that, if followed, will make ceutical firms are suffering large increases in research and
it more likely (but not guarantee) that we pick more and more development expenses, but with a decline in the number of
diverse subsets. These are optimization procedures, and they new drugs in the pipeline. Many finns have a large fraction
are widely used in the design of experiments, statistical sur- of their blockbuster drugs (>$l billion in sales per year)
veys. etc. An example of a common procedure is one called going off patent in the next few years. Part of the problem has
0-optimal selec:io,,. which is designed to pick a subset of
been the pursuit of only a few, highly profitable, therapeutic
poiuits that are as widely separated from each other as possible.
• Genetic algorithms. This optimitation procedure is inspired targets. For example, there are at least seven Stalin
by the way genetics and natural selection work. To use a ge- cholesterol drugs on the market: the most prolitable one.
netic algorithm, we pretend that our collection is an artificial Lipitor, currently collects about 57 billion in sales per year
'genome." Each structure in the collection is a base or string for its developer. Another problem has been the marketing
of bases in the gcnome (instead of A. T, 6, and C, the base of drugs that appeared to be safe, even throughout clinical
 Chapter 3 a Combinatorial Chemistry 59

riuls, hut were later found to cause serious and even fatal chemistry to build up their in-house libraries of structures
side effects (e'., Seldane and Baycol). that could be "mined" for activity against newly discovered
There is tithe question that combinatorial chemistry has receptors. A typical pharmaceutical firm has access to infor-
been effective in generating large numbers of lead structures. mation on 10 to 20 million structures from commercial
pharmaceutical companies began using combinatorial sources (various chemical software vendors and the Amen-

TABLE 3-3 Examples of Lead Structures Obtained by Combinatorial Chemistry

Structure Source Target Mechanism

Merek III I intcgruse Block Strut integration

ci SinhkKiine Human AntagoniM: cognithe

leeeliam Scrotonin receptor

CF3 Abbou intcrlcukin-2 Cytokine Inititsitirm

N—N

CF3 Plizer Earnesyl trunufetase a Inhibition

Cl

Parkc Davis KDO-8-P synthetasc Inhibition; antibiictorijl

60 lViI.con and Gici'old's Textbook of Organic Medicinal and Pharmaceutical Chemistry

can Chemical Society Chemical Abstracts Service). In addi- Journals

tion. large companies have their own multimillion-com-
pound databases. Golehiowski et describe how lead Combinatorial Chemist re and High-Tl,rou.c.hput Screen
structures with a wide variety of activity have been obtained ing—Bentham Publishers
with use of combinatorial chemistry. Sonic examples are Drug Discovery Today—Reed Elsevier
shown in Table 3-3, demonstrating the variety of structural Journal of Chemical Information and Computer Sti
ences—American Chemical Society
types that have been generated. An industry perspective pub-
Journal of Combinatorial Chen,isirs—American Cheinica
lished in 2001 reported 46 compounds in human clinical Society
trials that originated from HTS of libraries that were identi- Modern Drug Discovery—American Chemical Society
fled between 1992 and Molecular Diversitv—Kluwer
What can be argued is whether the goal of generating lead Nature Reviews Drug Discovery—Nature Publishing Group
structures is sufficient, in light of an increasing rejection rate Trends in Biotechnology (T1BT&'H)—Elscvicr
of candidate drugs in clinical trials, caused by side effects
and other ADMEI'-related failures. Most researchers would Videos
agree that we need to predict the "drugability" of a lead
better before much testing, if any, is done. As mentioned in Che,nicai Diversity: Applications of Computational Approaches
the section on virtual screening, much work is being devoted Washington, DC. American Chemical Society. 1995.
to the development of better in vitro and computational Chemical Diversity: Synthetic Techniques of Co,nbinatoria,
methods for predicting ADMET properties. Alternatives to Chemi,csry. Washington. DC. Anerican Chemical Society
combinatorial chemistry arc appearing in the literature. An 1995.
examfle is the "non-combinatorial" approach of Everett
et al. These authors argue that the goal of combinatorial Web Sites
chemistry should be the quality, not the quantity, of leads.
Some trends that are appearing in the literature include (a) lutp://lvsvw.comhi.web.com—Corporate-sponsored web portal.
smaller libraries, a few thousand carefully selected structures Accessed Dec. 3, 2002.
rather than 250,000 hastily designed ones; (b) more attention http://www.combkhent.neilhonte/login.a.sp—Recent develop.
menu.. Accessed Dec. 3. 2002.
to ADMET properties in the early phases of drug discovery;
hsrp://www.contbinarorial.com—Weh site for The Combinato-
(c) miniaturization of syntheses and assays, using ,nicroflu- rial Index text. Accessed Dcc. 3, 2002.
idics and nanoiechnologv. both for speed and to conserve hup://www.geocities.eo,n/ResearchTriangle/Lab/4688/
resources; and (d) an integration of genomic and combinato- combinatorial chen,istrv.l,tm—llnofficial Combinatorial
rial chemistry technology for better use human genome in- Chemistry Web site. Updated Mar. 5. 2002.
formation in the design of new drugs.46 Most chemists agree http://wwiv.microarrays.org—tjnivcrsity of Califomia at San
that combinatorial chemistry, after 20 years of evaluation. Francisco site. Accessed Dcc. 3, 2002.
is a vital, but not the only, implement in the drug discovery
toolkit that should be used. Like other tools, it can be applied
intelligently to great benefit, or it can be misused,
COMBINATORIAL CHEMISTRY
TERMINOLOGY
RESOURCES FOR COMBINATORIAL The following terms are some of the most common used in
CHEMISTRY combinatorial chemistry and HIS. More complete glossaries
can be found in Beck-Sicklinger. A., and Weber. P.: Combi.
natorial Strategies in Biology and Chemistry. New York.
Beck-Sicklinger. A.. and Weber. P.: Combinatorial Strategies in John Wiley & Sons. 2002, and in MacLean. D., et al.: Glos-
Biology and Chemistry. New York. John Wilcy & Sons. 2002. sary of terms used in combinatorial chemistry. I. Comb.
(The finest short introduction available) Chem. 2:562—578. 2000.
Bunin. B. A.: The Combinatorial Index. New York, Academic
Press, 1998 (a comprehensive, chemistry-oriented reference). ADMET (also ADME. ADMET-FK): The collection of a mole.
Crarnik. A. W.. and DeWitt. S. H. (eds.): A Practical Guide to cule's properties related to absorption, distribution. meiabo-
Combinatorial Chemistry. Washington. DC. American Chem- lism. excretion, toxicity, and phartniscokinetics, These facwrs
ical Society, 1997. are being increasingly considered in combinatorial library de-
Fennin. H.. Combinatorial Chemistry—A Practical Approach. sign, to yield molecules that will be more suitable as
Oxford. UK. Oxford University Press. 2000. (Laboratory ex- Aptasner: RNA molecule that displays specific binding to a
periments) target, usually a protein. Aptamers arc often used in
(Those. A. K.. and Viswanadhan. V. N.: Combinatorial Library rays in place of antibodies, to bind peptide ligands.
Design and Evaluation. Principles. Software Tools, and Appli- Array synthesIs: The form of parallel synthesis in which the
cations in Drug Discovery. New York, Marcel Dekkcr, 200!. reaction vessels are maintained in a particular spatial arrange.
Gordon, E. M., and Kerwin, J. J. F. (eds.): Combinatorial Chem- ment, such as a grid in a microtiter plate. Such arrays generated
istry and Molecular Diversity in Drug Discovery. New York, on a microscopic basis are termed a spatially addressable
Wiley-Lies, 1998. library.
Terrett. N.: Combinatorial chemistry. In Compton, R. G.. Davies. Backbone: A linear scaffold to which suhstituents are attached.
S. 0.. and Evans. J. (edt.). Oxford Chemistry Masters. Oxford. Common backbones include the a carbon backbones of pep.
UK. Oxford University Press. 1998. (A brief, highly readable tides and peptoids.
introduction) Bead: A spherical particle of solid support. Typically 50 to (K)
 Chapter 3 . ('o,nbi,,awrio! Chcn,i.un 61

par or dianreter. they swell in solvent. allowing access by A directed library lies midway between an initial e.qth;ra:i.-nr
synthetic reagents for reaction, washing. etc. The loading on library and a final libr,irv in its size and overall
head is the amount of synthetic target that can be attached diversity.
To ii sinnlc head, which is in the nanomolar range. Diversity: The "unrelatedness" of a set of, for example. build-
Binary encoding: Encoding technique of a library based on the ing blocks or members of a combinatorial library, Measured
presence or absence of tags on a bead. Thus, the sequence using physicochemical orctructural descriptors, a set with high
01 ItCh would encode thc presence of three of six possible diversity spans a larger fraction of "chemical space." Cluster
tugs. The number of combinations that can he encoded is 2'. analysis is one technique used to quantify diversity.
where it is the number of positions in the string. Dynamic library: A mixture of compounds in a dynamic cqui.
Binning: A computational procedure to allow selecting chciiiicul lihriutn with, for example, a synthetic process. If a receptor
structures across a wide range of diversity. The structures are is introduced into the system, the equilibrium will shift to
into bins on the basis of common physical or chemical produce mote of the compounds that bind tightly with the
structures. receptor.
Building block: One of a set of interchangeable reagents that Encoding: The process of adding a chemical or electronic tag
can be used in the synthesis of a generic library. to a bead for the purpose of "recording" the sequence of
Capacity: Theoretical amount of material that could be atiached reaction steps to which the bead has been exposed. By th'-od-
toa bead. Because of steric hindruncc of the synthetic target. ing the resulting tag, perhaps by treating a DNA tag with
it may he greater than the actual amount. polymcr-.tse chain reaction and analyzing the oligonucietrtide.
Capillary electrophoresis: Method of separating components the exact nature of the synthetic target on the bead can be
of a nrixture by placing the mixture at one end of a capillary determined.
Oiled with gel. A continuous gradient of electronic charge Enumeration: The process of explicitly describing all of the
across the capillary causes the components to separate. much specific structures that a generic structure or library contains.
like a chmmatograpliic separation but based on charge, size. Epitope: The region of a protein strand that is rccogtiized by
and shape of the molecules. an antibody.
Cleavage: The process of releasing a compound from a solid Fingerprint: An army of numbers In1, nj.. -. ) that numerically
support, allowing assay or analysis in solution. Special re- represents a given structure as values of physicochemical or
agents or even enzymes may he used to release the compound structural descriptors. Commonly, the ttutnbers are binary It)
without reacting with or ahienng it. or I). but they may also be counts (whole numbers) or values.
Clinter analysis: Statistical or pattern recognition technique to Flow cytometry: Technique characterizing or separating par.
group a set of structures into ' natural" groupings or clusters tides such as beads or cells, often on the basis of their tluores-
tin the ha.sis of physicochemical or structural properties. It is cence. Used to separate beads that have biologically active
similar to binning in its result, and both methods are corn- molecules attached.
manly used to select a representative sample of structures, Fluorous synthesis: An approach to solution-phase synthesis
either for screening or as building blocks for combinatorial that uses highly fitiorinated compounds as soluble supports
synthesis. for combinatorial chemistry. The addition of water or organic
CombInatorial: Relating to combinations of objects. solvents causes a phase separation of the fluorinated support
Combinatorial chemistry: Using a combinatorial process to for subsequent cleavage of the synthetic target structure.
prepare sets of compounds from building blocks. Generic structure: General structural formula of a library. con-
Combinatorial lIbrary: A set olcompounds prepared by combi- sisting of a .waffold(parent structure) pius rt'.cidue's (K groups).
natorial chemistry. A simple example is
Cross-linking: The property of a polymer used in a solid SUpport Genetic algorithm: Method of library design by selecting sub.
such that long strands of polymer are interconnected at various stituents for a library in a stepwisc fashion, bused on the fitness
points by relatively short sequences—much like rungs on a of the resulting library for some purpose (e.g.. biological activ-
flexible ladder. Cross-linking affects the properties of the ity). At each step, the substituents arc modified by use of the
polymer, including its ability It) swell in different solvents. genetic principles of recombination. crossover, mutation. etc.
Decode: To "read'' a chemical or electronic tag attached to a Selection of the "fittest" conrbinatiot,s of substituenis yields
bead or other solid support, for the purpose of determining a library thai is locally optimal for the given purpose.
the of reaction steps that were applied to the given Green fluorescent protein (GFP): A protein isolated fromjelly-
head. This allows determining the composition of the synthetic Itch that has its own fluorescence. It can be modified at various
target on the head, positions to generate molecules that fluoresce at different
l)econvolute: To make the results of a combinatorial experiment wavelengths. The DNA for this protein can be inserted into
less complex, usually by backtracking and reanalyzing or re- the genomes of cells to give them a fluorescent label.
synthesizing a subset of the structures iii the library. The goal High-throughput screening (HTS): The process for rapidly
of deconvolution is to determine which of a mixture of com- assessing the activity of samples from a combinatorial library
pounds in actually responsible for activity. or other compound coliection. usuaily done by running parallel
Dendrimer: A polymer having a very highly branched structure. assays in plates of 96 or more wells. A screening rate of
Dendrimers can be used in place ol' solid supports for attach- 100.000 assays per day is termed ul:ra/sigh-throughpui
ment of synthetic targets, and then they can be separated by screening.
ming size exclusion chromatography. Hit: A compound that has some required level of activity.
Descriptor: A numerical representation of a molecular property, HPLC: High-performance liquid chromatography. Solvent is
cithera bulk property (like log P) ora two-dimensional (2D) or pumped under high pressure through a chromatographic col-
three-dimensional (3D) structural property. When descriptors umis containing a very finely divided support. The compounds
encode the presence or absence of a property, they are usually in the mixture separate according to their affinity for the sup-
represented by Is and Os. and the collection of descriptors is port and elate from the column at different times, to be de-
called a fingerprint of the molecule, tected by use of sonic optical or even mass spectronietric de-
Directed (focused) library: A library that uses a limited number tector (HPLC-MS),
of building blocks chosen on the basis of information or some In sillco screening: See i'iriual screening.
hypothesis that defines the functionalities needed for activity. Lead compound: First compound in the development of a drug
62 Wilson and Gisi'ald's Textbook of Organic Medicinal and Pharmaceutical Chemistry

that has the desired biological and physicochemical properties. tive model for some property or response. The model consists
It typically has micromolar potency, and by optimizing various of input (the input data), a set of "hidden" nodes.
positions of the molecule, the potency can be increased to and one or more output nodes (the predictions). Each node
nanomolar. at which point it would be considered for drug behaves like a neuron, with a threshold value of input below
candidacy. which it will not "fire" any output. The interconnection of
Library: A collection of structures, either a generic library nodes allows the network to deal with complex nonlinear rela-
(based on some scaffold plus multiple residues) or a mixture tionships. The network is trained by iteratively adjusting the
library (containing diverse scaffolds). The number of specific weights at the nodes on the basis of the difference between
structures in a library is either the product of the numbers of the observed and the predicted output.
residues possible at each variable position (for a generic Ii- OmIssion library: Strategy for identifying active library
brary) or simply the sum of the number of structures (rot a bers by the systematic omission of building blocks from ntis-
mixture library). tures. Observation of reduced activity in a certain pool sug-
Linker: A chemical chain that connects the solid or soluble gests that the building block that was omitted in that poo1
support to the synthetic target in a combinatorial experiment. contributes to activity.
The linker is decomposed when the desired compound is One-bead one-compound strategy: The earliest strategy for
cleaved from the support. solid-phase combinatorial chemistry, in which each bead has
Llplnski "rule of live": A set of criteria for predicting the oral molecules of only a single structure attached rather than a
bioavailability of a compound on the basis of simple molecular mixture of structures.
properties (molecular weight. <500: log P. <5; number of Orthogonal design: (a) Using protecting groups or linkers in a
hydrogen-bond donors. <5; and number of hydrogen-bond combinatorial experiment that do not interfere with each other
acceptors. <10). Typically, the criteria are applied to a library chemically; or (b) a pooling strategy in which a given library
to filter structures from the library before any synthesis takes member appears in more than one pool, mixed with other
place. Any structure exceeding two or more criteria is rejected. members. Pools have only one structure in common, so a hit
LiquId-phase chemistry: The process of using a large, soluble in several pools implies that a given member is responsible
molecule at. the support for a combinatorial chemistry experi- for the activity.
ment. Peptold: Oligomer of repenting N-substituted glycines that can
Loading: Characteristic property 01 a solid support that de. emulate a peptide. without being susceptible to acid degrada-
scribes the amount of a specific chemical species that can be tion in the gastrointestinal tract.
attached synthetically to a unit mass of support. Phage display: Use of bacteriophage viruses as vessels for pre-
Mapping: Analyzing the sequence of a protein with regard to senting short peptide segments of their native surface proteins.
a desired property to identify the residues involved in binding By varying the gene sequences of the phages in a combinato-
or activity. Typically, this involves generating short, overlap- rial manner, libraries of peptides can be generated and tested.
ping sequences of the protein, perhaps on a microarray. and Pharmacophore: The ensemble of steric. electronic, and lipo-
testing for activity. philic factors needed to ensure interaction of a drug molecule
Markush structure: A type of structure representation in which with a given receptor. Pharmacophores are most useful in
very general terms, such us alkyl or alcohol, can be used to searching a three-dimensional (3D) structure database and in
describe the substituents in a generic structure. Used in the filtering structures for virtual screening.
patent literature and adapted for combinatorial chemistry pub- Photolithography: The process by which successive masking
lications. generates light patterns that direct chemical transformations
Member: Either (a) a particular substituent at a given position in in certain areas of a photosensitive surface. Coupling different
a generic structure (an k-group member) or (h) an enumerated building blocks to discrete sites on the surface gives rise to
structure of a generic library, which corresponds to a given spatially addressable microarrays of compounds.
selection of substituents (a member of a library). Pin: An elongated device in which the tip acts as a solid support.
Mesh size: The density of wires in a sieve; also, a term In de- An army of pins, fitting into wells of a microtiter plate. cmi
scribe the size of particles. A 1(10- to 200-mesh particle will be used for parallel synthesis of a combinatorial library.
pass through a 100-mesh filter but be trapped by a 200-mesh Polyethylene glycol (PEG): Polymer widely used in pharmscy
filter and consist of particles 75 to 150 in diameter. as an ointment base. It has been used as a soluble support and
Microarray: Masks can be used in the same way that stencils u.s a linker in combinatorial synthesis. Its structure is
ate used to make printed circuit boards, to either allow or
block UV light from causing chemical reactions in a small
defined area. In this way. a library of hundreds or thousands Polymerase chain reaction(PCR): Technique for amplification
of compounds can be synthesi,.ed in a grid layout, over a very of small amounts of DNA. starting with a few molecules and
small area, perhaps a fraction of a square inch. The resulting yielding sufficient material to analyse the sequence. Also
microarray can be exposed to a given receptor, and examining widely used in forensic DNA analysis.
the chip under UV light can reveal which structures have Pool: (a) A subset of a given combinatorial library: or (h) the
bound to the receptor. process of combining and mixing library components.
Mimetics: Compounds that share the desired properties of other Pool/split (spilt and mix, split and pool): Strategy for assenl•
molecules (e.g.. the affinity for a given receptor) but do not bling a combinatorial library. The solid support is divided into
share the undesirable properties, such a.'. susceptibility to pro- portions, each of which is subjected to a given reaction with
teases. An important cla.ss of mimetics is peptide mimelics. a single building block. Pooling the portions gives a single
Mimotope: A compound that imitates an cpitopc; typically, a hatch with a mixture of components. Repeating the split. react,
nonpeptide sequence that can bind to a particular antibody. and mix sequence results in a library in which each discrete
Mimotopet. were an important class of compounds studied in bead of solid support carries a single library member (one'
early combinatorial chemistry experiments. bead one-compound strategy). The number of members equals
Monomer: A member of a set of building blocks that can be tine product of the number of building blocks incorporated at
repeatedly incorporated into a library leg., amino acids in a each step (i.e.. fully combinatorial).
peptide library). Positional scan: Strategy for identifying individual compoundi
Neural network: Computational procedure to generate a predie- of interest in a library. A collection of sublibranes is prepared,
 Chapter 3 • Co,nbinawrial Chemistry 63

in number to the total number of building blocks in the Selectivity: Measure of a compound's tendency to hind only to
whole library. In each pool, one substituent position is held a certain target receptor. Drug molecules should have high
constant by incorporating a single building block while the selectivity us well as potency.
other positions use all possible building blocks. Activity in a Solid support: insoluble, functionalized polymeric material to
given pool implies that the given substituent at the given posi- which library members or reagents may be attached, often via
turn is essential for activity. a linker, allowing them to be readily separated from solvent.
PrincIpal components analysis: Computational approach to re- excess reagent. etc. Typically, the solid support swells in sol-
duce the dimensionality (i.e.. the number of variables) in a vent, allowing reactions to occur in the interior of the bead
data analysis, by weighting variables according to their contri- or other form. This greatly increases the available surface area
bution to the overall variation. A plot of points in the first for reaction.
two principal components is like a two-dimensional (2D) Soluble support: Typically, a large molecule that is soluble in
"shadow" of the multidimensional data that can be used to some solvents and, upon the addition of other solvents, sepa-
(lad clusters and relationships among the points (i.e.. coin- rates into phases. The molecule can serve as a support for
pounds), attaching members of a combinatorial library. The advantage
Property space: Multidimensional representation of a set of over solid supports is that reactions are more coniplete in
compounds as points in space. Each axis of the space repre- liquid phase. Sometimes termed liquid-phase chemistry. An
sents some descriptor, either whole property or computed from example of a soluble support is PffG, which is soluble in polar
the two-dimensional (2D) or three-dimensional (3D) chemical solvents but separates 1mm organic ones.
stnictuge. Compounds that are similar to each other chemically Spacer: Same as a linker.
will cluster together in property space. A further assumption Spatially addressable: Having the ability to identify part or all
used in library design is that structurally similar compounds of the structure of a library component or pool from its physi-
will share similar biological activity. cal location in a grid or array.
I'rotecting group: Chemical group that reversibly blocks func- Spot synthesis: Solid-phase synthesis at certain points (spots) on
tional groups ott a synthetic target, to prevent them from enter- a two.dimensiotial (21)) surface (e.g.. a cellulose membrane).
ing into undesired side reactions. For example, an OH group Recently, the techniques of ink-jet printing have been applied
might be protected by converting it to an ester, then hydrolyz- to spot synthesis, yielding very dense arrays of compounds.
ing the ester buck to an alcohol when the synthesis is complete. Sublibrary: A subset of a combinatorial library in which. For
Radlofrequency encoding: The process of embedding into solid example, the substituent at one position is held constant while
supports the minute electronic devices that emit rudiotre- other positions are varied. Also called a pool.
quency signals upon stimulation with an electromagnetic Supercritical fluid chromatography (SFC): HPLC using a
source. The signals can be used to track the reaction history ''solvent" such as liquid CO2 under high pressure. The advan-
of the given bead and thus the makeup of the compounds tage is that the carrier evaporates, simplifying the detection
attached. of the compounds as they elute from the chromatographic
Ratio encoding: Strategy in which the quantities of tags on ii column.
bead give itttormation about the compound identity rather than Tag: A nonrcaciis'e chentical functionality attached to a solid
tile nature of the tags. support thaI carries information about the reaction history of
Residue: The portion of a chemical structure that can be identi- the given support and titus can at least partially idetitify the
fied us coining front a particular building block, such as the attached synthetic target. An exuniple is attaching various
alunine residue in a polypeplide. In a generic structure, the DNA bases to the bead at each synthetic step. The resulting
rx'sulucs ore the substiluents that correspond to the R-groups oligonucleotide can be multiplied by using PCR and identified
in the uructure. analytically.
Resin: Insoluble polymeric material to which linkers, synthetic Tea bag: A type ol' reaction vessel consisting of a porous mesh
targets, and tags are attached. Soitietintes. resins arc simply hag that encloses the resin but allows passage of reagents and
used to scavenge side of a reaction. In chromatogra- solvents. Several tea bags cult be immersed in a given reagent
phy. resin heads are used to separate compounds by size or and then manipulated from one reagent to the next to generate
by the charge ott a molecule. combinatorial libraries.
Resynthesis: Preparation of individual members or subsets of Virtual library: A combinatorial library that has no physical
combinatorial library, to l'ollow up on sonic property ut existence; rather, it exists in a computer or on paper. Such
interest. libraries eon he generated automatically and screened against
Reverse transcriptase: An enzyme that can reverse transcribe physiochentical tillers like time "rule of five" or be docked
RN/s into its correspotmding DNA. into receptors by use of niolecular modeling.
Ribozyme: RN/s molecule with enzyme catalytic activity. Virtual screening: The selection of compounds by evaluating
Robotic system: An automated system, usually controlled by a their fitness by use of contptttttionmtl model. Also called in
cimitipuler, to transfer materials by physical movement of a .vilit'o
delivery device or by movement of' the reaction vessels. Ro-
botic systems can be general purpose, and capable of being
reprogrammed to do a variety of different tasks, or may be REFERENCES
specialiied. as part of a turnkey system. I. Merrifield. K. it.: Solid ptta'.c peptide synthesis. I. TIme synthesis of a
Safety-catch linker: A linker that is cleaved by performing two teirapeplide. J. Am. ('hem. 85:2t49'-2154. 1963.
reactions instead of the customary one. This provides greater 1. Furka. A.: Combinatorial cheitmistry: 20 years... Drug Discovery
control over the timing of the release of the synthetic target. Today 7:1 2002.
Scaffold: The core portion of a generic struCture, common to 3. tloughten. K A.: mimethod hmr mIme rapid so)id-phasc synthesis
oh large numbers ot pcpmides: spcrilieimy miIamumigcn—aiutitssiy interaction
all the imenthets of the library. Note that in the ease of Ph-
at level of individual amino Proc. Nat). Acad. Sri. U. S. A.
thc seatThld consists two disconnected
82:5l31—5t35, 985,
Iraginents, separated by the R1 substituent. 4. Simimmumu. R. J.. Kania. R. S.. 7uckcrnmann. K. N,. em it).: t'cpmoids: a
Setnenger resin: A resin introduced into a combinatorial experi- niusiular approach to drug discovery. t'roc. Nail. Acad. Sci. U. S. A.
ment to react with undesirable materials (excess reagent or 89:9367—937). 992.
side products) and remove them from the experiment. 5. Miller. S. M . em it).: Promeotytme sftslies of lummnolugous peptide and N—
64 Wilson and Tn'xgbor,k of Organic Medicinal and Pharrnaci'uik'al Chemistry

substituted glycine pcpioid uligomer.. Bioorg. Med. Chem. Leit. 4: 26. Spencer. K. W.: High-throughput screening of historic collections: ob-
2657—2662. 1994. servations on file sire, biological targets, and file diversity Biotechnol,
Ii. Zuckertuan, R. N.. ci aL: Discovery of nanotiwlar ligands for 7.truns- }Siocng. 61:61—67. 1998.
incnthratie G.pnrteiit-coupled receptors from a diver.e 27, Venton, U. L., and Woodbury, C. P.: Screening combinatorial libraries.
glycine peptoid library. J. Med. Chcni.37:2fi78—2685. 1994. Chcmometr. Intell. Lab. Syst. 48:131 — ISO. 1999.
7. Bunin, B. A and ElIman. 3. A.: A gener.il and expedient method liii 28. Tarbit. M. H.. and Berman, 3.: High.throughput approaches forevaluat-
the solid-phase synthesis of I ,4trenrodiarepine derivatives. J. Ant. ing absorption, distribution, metabolism and excretion properties of
Chem. Six. 14:10997-10998. 1992. lend compounds. Cure. Opin. Chem. Biol. 2:411—416. 1998.
8, Lartred, M.. md Hallbcrg. A.: Micruwave.assisted high.speed chentis- 29. Walters. W. P.. Stahl. M. T.. and Murcko. M. A.: Virtual screening—an
try: a new tcchnique in drug discovery. Drug Discovery Today 6: overview. Drug Discovery Today 3:160—178. 1998.
.UJ6-416. 2001. 30. Shoichet. B. K.. et at.: Lead discovery using molecular docking. Cure.
9. Balkcnhohl. P.. ci at.: Combinatorial synthesis at small organic mole- Opin. Chem. Biol. 6:439—446. 2002.
cules. Angew. Chem. tnt. Eel. Engl.35:2288—2337. 1996. 31. htmp'J/www.grid.orgiprojeus/patnot,htm, United Devices agent soft.
TO. Murvaurelle. I.. A.. and Seeberger. P. H.: Combinatorial carbohydrate ware download. 2002,
chemistry. Curr. 0gm. Chein. Biol. 6:289—296. 2002, 32. tOOhtn. 3.-i.. and Stahl, M.: Structure-based library design: molecular
II. Nielsen, 3.: Combinatorial synthesis of natural products. Cure. Opin. modelling merges with combinatorial chemistry. Cwr. Opmn. Chem.
Chent. Biol. 6:297—305, 2002. Biol. 4:283—286, 2000.
12. holmes, C. P.: Model studies For nev. o.nitrobenzyl photolabile Tinkers: 33. Lipmnshi. C. A.. et al: E,xpeilinental and coniputational approaches to
suhstiitrcni effects on the rates of photochemical cleavage. J. Org. estimate solubility and permeability in drug discovery and development
Chem. 62:2370—2381). 1997. settings. Adv. Drug Delis. Rev. 23:3—25. 1997.
13. Reents, R.. Jeyaraj. U. A., and Waldman. H.: Enrymatically cleavahic 34. Opera. T. I.: Virtual screening in lend discovery: a viewpoint. Mole.
linker groups in polymer-supported synthesis. Drug Discovery Today cules 7:51—62, 2002.
7:71—75, 2(102. 35. Opera, T. I.. and Gotil'ries. J.: Toward minintalistie modeling of oral
14. Keatiog. T. A,. and Armstrong. K. W.: Molecular diversity via a con- drug absorption. J, Mol. Model. Graph. 17:261 —274. 1999.
verttble isocyanide in the Ugi Four-component condensation. i. Am. 36. Zuegge. 3., ci al.: A Fast virtual screening filter for cytochromr P450
Chetti. Soc. 117:7842—7843. 1995. 3A4 inhibition liability of compound libraries. QSAR 21:249—256.
2002.
15. Han. H., et il.: l.iquid.pha.sn combinatorial synthesis. Proc. NatI. Acud.
37. Adang. E. P.. and l'lmrrrnkens. P. H. H.: The contribution oFcombinato-
Sci. U, S. A. 92:6419—6423. 1995.
rial chemistry to lead generation: an interim analysis. Med. ('hem.
16. Siuder. A., ci at.: Iluornus synthesis: a fluorous.phu.sc strategy for
8:985—998, 2001.
itttproving separation efficicncy in organic sytithesis. Science 275:
38. Spellmeycr, 0. C.. and Gruotenhuis, p. o. 3.: Recent developments in
823—826. 1997.
molecular diversity: computational approaches to combinatorial cheni-
17 Caldemne, C. let at.: Directing otherwise incompatible reactions in a
istry. Ann. Rep. Med. 0mm. 34:287—296. 1999.
single solution using I)NA-templatedorganic synthesis Angew. Chem.
39. Jacobsen. E. 3.. et al.: imidarol I ,5-alquinoxnline ureas as
tnt. Ed. EngI. 41:4t04.—4lll%. 2(Xl2.
high affinity GABA—a ligand of dual functionality. 3. Med. Cheat.
18. Hughes. I.. and Hunter. D.: Techniques for analysis and purification in 42:1123. 1999.
higlt-thrcrughput chemistry. Cure. 0gm. Chem, Biol. 5:243—247.2001. 40. Martin, E. 3.. and Cntchlow. R. 0.: Beyond mere diversity: tailoring
19. Van. B.. and Kunaaeuvel. G.: Probing solid-phase reactions by monitor- combinatorial libr,tries for drag discovery. 3. Comb. Chem. 1:32-45.
ing the IR bands of compounds on a single 'flattened' resin head. 1999.
Tetrahedron 52:843-848. 1996. 41. Opera. T. I.. ci al.: Chemical information managetncnt in drug discov-
20. Shuker. S. B., ci at.: Discovering high'aflinity ligands for proteitis: ery: opiinmiring the cotimpulalional and combinatonal chemistry inter,
SAR by NMR. Science 274:1531—1534. 1996. Faces. 3. Mol. Model. Gr.mph. 18:512—524. 2000.
21. Brenner. S.. and Lerner. K. A.: Encoded combinatorial chemistry. Proc. 42. Merritt. A. 1.: Uptake of new technology in lead oplimieation fur drag
Nail. Acad. Sci, U.S. A. 89:5381—5383. 1992. discovery. Drag Discovery Today 3:505—510. 1998.
22. Nielsen. 1., Brenner. S.. and Janda. K. U.: Synthetic methods for the 43. Golebiowski, A., Kloptenstein. S. R., and Portlock, 0. E.: Lead com-
impletnentatititi of encoded combinatorial chemistry. 3. Am. Chem. pounds discovered from libranes. Curr, 0gm. Chent. Biol. 5:273—284.
Sire. 115:9812—9813. 1993. 2001.
23. http://www.qdots.csitn. Quantunt dots. 2002. 44. Owens. 3.. et uI.: News in brieflmiscellaneous—Itrst H'l'S doug cantli.
24. Limnitro, 3. A., ci al.: HTS in the new itiillenniuitt—thr role of pharma- dates reach clinical trials. Drag Discoxcry Today 6:230. 2001.
colirgy and flexibility. J. Pharmacol. Toxicol, Methods 44:273 -289, 45. Everett. 3., ci al.: The application of non-combinatorial chemistry to
2(881. lead discovery. Drug Discovery Today 6:779—785. 2001.
25. Drews. J.: Drug discovery: a historical perspective. Science 287: 46. Lear. 0. R.. Nash. H. M.. and Jindal. S.: Chetnical ligands.genomics,
1960—1964. 21)00. and drag discovery. Drug Discovery Today 5: 145—156. 2000.
CHA PT ER 4
Metabolic Changes of Drugs and
Related Organic Compounds
STEPHEN J. CUTLER AND JOHN H. BLOCK

Metabolism plays a central role in the elimination of drugs

and other foreign compounds (xe,:ohiotk.s) from the body. GENERAL PATHWAYS OF DRUG
A solid understanding of drug metabolic pathways is an es- METABOUSM
sential tool for pharmacists in their role of selecting and
Drug metabolism reactions have been divided into two cate-
monitoring appropriate drug lherjpy for their patients. Most
gories: phase! (fiinctionalization) and phase!! (coI,jugalion)
organic compounds entering the body are relatively lipid
soluble (ilpoplillic). To be absorbed, they must traverse the
reacttons.' Phase I, or functionalization reactions, include
oxidative. reductive, and hydrolytic biotransformations
lipoprotein membranes of the lumen walls of the gastrointes-
(Table 4_l).C The purpose of these reactions is to introduce
tinal (GIl tract. Then, once in the bloodstream, these mole-
a functional polar group(s) (e.g.. OH. COOH, NHa. SH) into
cults can diffuse passively through other mcinbrane.s and be
the xenobiotic molecule to produce a more water soluble
distributed effectively to reach various target organs to exert
their pharmacological actions. Because of reabsorption in
compound. This can be achieved by direct introduction of
the functional group (e.g.. aromatic and aliphatic hydroxyla.
the renal tubules. lipophilic compounds are not excreted to
any substantial extent in the urine. Xenohiotics then meet
tion) or by modifying or "unmasking" existing tunctionali-
ties (e.g.. reduction of ketones and aldehydes to alcohols;
their metabolic fate through various enzyme systems that
oxidation of alcohols to acids: hydrolysis of ester and amides
change the parent compound to render it more water soluble
(hs'dropltilw). Once the metabolite is sufficiently waler solu-
ble, ii may be excreted from the body. The statements above
chow thai a working knowledge of the ADME (absorption,
distribution. metabolism, and excretion) principles is vital TABLE 4—1 General Summary of Phase I and Phase Il
for successful determination of drug regimens. Metabolic Pathways
If lipophilic drugs. or xenohiotics, were not metabolized to
Phase t or Functlonallzatlon Reactions
polar, readily excretable water-soluble products, they would
Oxidatinc reactions
remain indefinitely in the body. eliciting their biological ci- Oxidation of aromatic ntolctios
Thus, the formation of water-soluble metabolites not Oxidation ot ok'tins
only enhances drug elimination, but also leads to compounds Oxidation at henzytic. allytie carbon atoms. mid curbon stems o to
thai are generally pharmacologically inactive and relatively earhonyt and iminca
nontoxie. Consequently, drug metabolism reactions have tra- Oxidation at atipharic and aticyctic carbon atoms
O*idation involving curbon—hcicroatomri systems:
ditionally been regarded as (!ewx,calion (or detrixifkazion) Carbon—nitrogen systems and aromatic anuses: inctudrs
processes.' Unfortunately, it is incorrect to assume that drug N..dcatkyiaiion. dcjminarion. N'oxitk formation.
nietabolisin reactions are always detoxifying. Many drugs N-hydroxytamiun)
are hiotransformed to pharmacologically active rnetabolites, Carbon—oxygen srtcms tO-draikytatimml
These metabolites may have significant activity that contrib- Carbon—sulfur systems IS-deatkyiiition. S-oxidaiion, and
dcsuifwauion)
utes substantially to the pharmacological or toxicological Oxidation of alcohols mind aidehyden
effect(s) ascribed to the parent drug. Occasionally, the parent Other misctliuncomts oxidalivc reactions
compottnd is inactive when administered and must be meta- Reductive Reactions
helically converted to a biologically active drug (metabo- Reduction til aidrhydcs and kctonet
lite).1 These types of compounds are referred to as pro- ReductIon of ,iiuu and azo compounds
Miscellaneous reductive reurciion.s
In addition, it is becoming increasingly clear that not
all meinbolites are nontoxic. Indeed, many adverse effects Hydrolytic Reactions
Hydrolysis of esters and amudc.s
e.g.. tissue necrosis, carcinogenicity. teratogenicity) of Hydration ot epoxidcs and arene oxides by epoxide hydruse
drugs and environmental contaminants can be attributed di-
reedy to the formation of chemically reactive metabolites Phase H or Conjugation Reactions
that are highly detrimental to the dy.4t' This concept is Gtueurnnic acid conjugation
more important when the patient has a disease state that Sutfatc conjugation
inhibits or expedites xenohiutic metabolism. Also, more and Conjugation with glycine. glittamine, and other amino acids
more drug metaboliics are being tbund in our sewage sys- Gtuiaihione or mtrvupturie acid conjugation
Acetylation
rena. These compounds may be nontoxic to humans but
Methytation
hanniul to other animals or the environment.

65