Clinical Science (2000) 99, 255260 (Printed in Great Britain) 255

2 0 0 0 B A Y E R L E C T U R E

Science, art and drug discovery:

a personal perspective

Simon F. CAMPBELL
Pfizer Central Research, Sandwich, Kent CT13 9NJ, U.K.

A B S T R A C T

The research programme that started in 1985 led to the approval of Sildenafil (Viagra2), in 1998,
as the first oral treatment for male erectile dysfunction. The initial project objective was the
design and synthesis of novel inhibitors of phosphodiesterase that would increase tissue levels
of cGMP, and that could be beneficial for the treatment of cardiovascular conditions. Starting
from zaprinast, a weak phosphodiesterase inhibitor, computer modelling guided rational
medicinal chemistry to achieve significant increases in potency and selectivity for the
5-isoenzyme within a novel series of pyrazolopyrimidinones. Optimization of structureactivity
relationships and pharmacokinetic properties led to sildenafil, which proved essentially devoid
of cardiovascular activity in clinical trials. However, the emerging role of nitric oxide and cGMP
in controlling blood flow in the penis suggested that sildenafil would have a beneficial effect on
erectile function. This hypothesis was confirmed by extensive clinical trials in nearly 5000
patients and the Food and Drug Administration approved sildenafil in March 1998 for male
erectile dysfunction. Sildenafil is now available in over 100 countries and more than 150 million
tablets have been dispensed worldwide. The sildenafil research programme reflects a traditional
approach to drug discovery, but pressures to improve productivity have prompted major
investments in genome sciences and new technologies. The impact of these initiatives on the
drug discovery paradigm will be discussed, particularly with respect to shortening time scales
between identifying gene sequences and submitting innovative products for regulatory approval.

INTRODUCTION Consequently, the pharmaceutical industry is under

The pharmaceutical industry invests between 10 and
20 % of annual sales revenues in research and develop-
ment, which is far higher than in other research-based
sectors. However, transforming an idea into a marketed
medicine can take between 10 and 20 years, and costs
have risen to over $500 million, a 10-fold increase since
1976. Moreover, pharmaceutical research is associated
with significant risk, since many drug candidates fail
during safety and clinical evaluation, while only one in
three marketed agents yields a return on investment.
intense pressure to increase productivity, and drug
discovery is experiencing a paradigm shift, whereby the
explosion in genome sciences and new technologies is
being harnessed to produce innovative therapies within
shorter time scales. While it is expected that new targets
and screening hits will be identified at unprecedented
rates, innovative science is still rate limiting in trans-
forming leads into drug candidates. Moreover, since the
structures of many receptors and enzymes are unknown,
successful drug discovery teams must blend scientific
data with experience and intuition to develop robust

This paper was presented as the 2000 Bayer Lecture at the Spring Meeting of the Medical Research Society, held at the Royal
College of Physicians, London, on 22 May 2000.
Key words: clinical trials, combinatorial chemistry, cGMP, drug discovery, erectile dysfunction, Food and Drugs Administration
approval, genomics, high throughput screening, nitric oxide, pharmacogenetics, phosphodiesterases 5, sildenafil (Viagra2), target
validation.
Abbreviations: ANF, atrial natriuretic factor ; ED, erectile dysfunction ; PDE, phosphodiesterase.
Correspondence and present address: Dr S. F. Campbell, Church Cliff House, Kingsdown, Deal, Kent CT14 8AT, U.K.

# 2000 The Biochemical Society and the Medical Research Society

256 S. F. Campbell
hypotheses for drugreceptor interactions that guide
rational design and synthesis. These challenges will be
illustrated with a personal account of the research
programme that led to sildenafil (Viagra2 ; for a previous
account see [1]), followed by some thoughts on how
drug discovery might evolve over the next decade.
THE ROAD TO SILDENAFIL
Atrial natriuretic factor (ANF), cGMP and
phosphodiesterase (PDE)
The research programme that led to the discovery of
sildenafil originated from our interest in ANF, an
endogenous peptide with vasodilator and natriuretic
properties. ANF exerts its physiological roles by stimu-
lating guanylate cyclase to increase tissue levels of cGMP,
although this second messenger is degraded rapidly by a
specific PDE. In 1985, we suggested that compounds that
blocked the destructive action of this PDE would
preserve tissue levels of cGMP and hence would
potentiate the vasodilator and natriuretic effects of ANF.
We expected these PDE inhibitors to show potential for
the treatment of hypertension and other cardiovascular
indications [2].
Rational design of PDE5 inhibitors
In 1985, little was known about the physiological role of
cGMP (Figure 1, structure 1) or the specific PDE (later
characterized as PDE5) to support our theory, and there
were no potent, selective PDE5 inhibitors available for
laboratory studies. However, a search of the literature
uncovered zaprinast (Figure 1, structure 2), a weak (IC
&!
2n0 M), but non-selective, inhibitor of the enzyme that
showed vasodilator activity in vitro, and lowered blood
pressure in the anaesthetized dog. However, the com-
pound possessed additional pharmacological properties,
and had been progressed to the clinic as an anti-allergy
agent.
Despite such a broad profile, zaprinast provided a
starting point for a rational medicinal chemistry pro-
gramme aimed at increasing both potency and selectivity
for PDE5. First, comparison of the electronic distribution
in zaprinast and cGMP, the physiological substrate for
PDE, showed that the dipole moments of the triazolo-
pyrimidinone and guanine ring systems were similar in
both magnitude and direction (9n5 and 7n4 respectively).
This suggested that both heterocycles could be
recognized in a similar manner within the PDE active
site, and that modification of the parent triazolo-
pyrimidinone could lead to enhanced enzyme affinity.
Secondly, computational studies confirmed that cGMP
preferred to adopt a syn conformation and that there was
scope within the zaprinast skeleton to introduce isosteric
replacements for the cyclic phosphate moiety. Finally,
the X-ray structure of zaprinast showed an intramol-
# 2000 The Biochemical Society and the Medical Research Society
O
H N
N
H2N N N H O H
O Pr O N N
N
OH N
N
O O
P
O O 1 2
Me
H O
Et O N N
N
N 3 R1 = Me; R2 = H
R1
R2 4 R1 = Pr; R2 = SO2 N N Me
Figure 1 Structures of cGMP (1), zaprinast (2), a pyrazolo-
pyrimidinone template (3) and sildenafil (4)
Me, methyl ; Et, ethyl ; Pr, propyl. See the text for further details.
ecular hydrogen bond between the aryl oxygen and
pyrimidinone hydrogen atoms that maintained co-
planarity, and this part of the molecule was therefore left
essentially unchanged.
These rationale were pursued by the chemistry team,
who initially probed a series of heterocyclic replacements
for the parent ring system in zaprinast. After studying
several alternatives, a pyrazolopyrimidinone (Figure 1,
structure 3) proved to be superior, with a 10-fold increase
in potency against PDE5 (IC 0n3 M), and reduced
&!
activity against other members of this extended enzyme
family [3]. The next step was to exploit compound (3) as
a template to probe the spatial region occupied by the
cyclic phosphate moiety in the natural substrate. Thus it
was proposed that extension of the 3-methyl group would
be beneficial, as would introduction of a tetrahedral
sulphonamide as an isosteric alternative for the natural
cyclic phosphate (Figure 1, structure 1).
These ideas were brought together in 1989 with the
synthesis of UK 92 480 (sildenafil) (Figure 1, structure 4),
which provided a 100-fold increase in PDE5 inhibitory
activity (IC 3n5 nM) over zaprinast, together with un-
&!
precedented selectivity over other PDE isoenzymes [4].
The piperazinosulphonamide moiety in sildenafil also
plays an important role in improving physicochemical
properties by reducing lipophilicity (logD, 2n7) and in-
creasing solubility. Sildenafil was shown to be a com-
petitive inhibitor of PDE5 (Ki 2 nM), which confirmed
our design principles, and gave a dose-related (10100
nM) increase (24-fold) in cGMP levels in canine
coronary artery sections in vitro. Importantly, levels of
cAMP, which has been implicated in cardiac arrythmias,
remained unchanged [4]. Sildenafil also potentiated the
anti-vasoconstrictor activity of glyceryl trinitrate, and
increased intracellular cGMP levels in response to sodium
nitroprusside in intact cells [5].

Corpus
cavernosum
Cell membrane
GTP
GMP
Figure 2 Mechanistic pathways for NO and sildenafil in erectile function
Adapted from Palmer, E. (1999) Chemistry in Britain, January 1999 issue, 2426, with permission.
Sildenafil in the clinic : an initial setback
Despite this pharmacological profile, we were disap-
pointed when the initial clinical performance of sildenafil
in patients with coronary heart disease fell short of our
expectations. However, rather than lose heart, it was
decided to push the compound to its limit with a 10-day,
multiple-dose study in volunteers, where it was expected
that some clues to the clinical potential of sildenafil
would emerge. In the event, a variety of effects were
observed, including head and muscle ache, indigestion,
and some increase in erectile function.
Nitric oxide (NO), PDE and sildenafil :
a mechanistic rationale for erectile
dysfunction (ED)
This final observation was most intriguing, and we
decided to investigate the pharmacological basis for the
effect of sildenafil on erectile function, and to determine
the potential for use in patients suffering from ED
(impotence). These clinical and laboratory studies with
sildenafil coincided with a burgeoning interest in the
physiological role of NO, an endogenous compound that
was nominated Molecule of the Year by Science in 1992.
More recently, three of the pioneers of NO research were
awarded a Nobel Prize for Physiology or Medicine in
1998. NO has been shown to produce a range of
pharmacological effects through stimulation of guanylate
cyclase but, by 1992, a strong case was emerging for a
specific role in controlling erectile function. For example,
NO mediates relaxation of human corpus cavernosum
smooth muscle [6,7], which can be impaired in patients
suffering from ED.
These observations, and in-house studies, were
brought together within a unified hypothesis (Figure 2)
Science, art and drug discovery 257
NO
Guanylate cyclase Erection
cGMP Smooth
muscle
relaxation
PDE
Sildenafil
whereby, on sexual stimulation, NO is released from
non-adrenergic, non-cholinergic nerve endings in the
penis, where it stimulates guanylate cyclase to increase
cGMP levels in the corpus cavernosum. This second
messenger then initiates the smooth muscle relaxation
that increases blood flow to the penis and leads to an
erection, which is subsequently terminated when cGMP
is degraded by PDE5.
It was immediately realized that, by blocking the
action of PDE5, sildenafil would potentiate the natural
activity of NO, and would improve erectile function in
conditions where NO release or smooth muscle relax-
ation are impaired in the penis. Importantly, sildenafil
would only be effective in response to the neurogenic
activity that accompanies sexual stimulation, since the
compound has little or no direct relaxant effect on
contracted human corpus cavernosum [8].
Sildenafil, a major medical advance for the
treatment of ED
In parallel with our laboratory studies, project clinicians
carefully examined the patient population suffering from
ED, which is a common and serious condition [9] that
may afflict up to 30 million men in the U.S.A. alone. In
addition, it was apparent that there were no defined
protocols for the study of an orally administered agent
for this indication. Accordingly, the innovative step of
developing the International Index of Erectile Function
was taken to provide a detailed questionnaire that helps
patients and partners record the extent of ED, and the
impact of drug treatment [10]. Clearly, utility in patients
would require a response after acute administration of
sildenafil, rather than the multi-dose regimen used in the
10-day volunteer study.
# 2000 The Biochemical Society and the Medical Research Society
258 S. F. Campbell
Gene Function Target
Figure 3 Flow chart from gene sequence to drug product
With a clear understanding of the mechanism of action,
and with clinical methodology in place, Phase 2 trials
with sildenafil in patients suffering from ED started in
1994. Towards the end of 1997, a database on nearly 5000
patients was submitted to the Food and Drugs Adminis-
tration (FDA), and approval for use in ED was granted
on March 27, 1998, after fast-track review. These ex-
tensive clinical studies showed that sildenafil produced a
dose-related effect in patients, with improved erectile
function of 65, 79 and 88 % reported after single
administration of 10, 25 and 50 mg respectively. More-
over, the response to sildenafil was similar whether the
underlying cause of ED was diagnosed as organic (68 %),
psychogenic (84 %) or mixed (77 %). Overall, toleration
is excellent, side effects such as headache and indigestion
tend to be mild [11], and the low discontinuation rate
(2n5 %) is essentially the same as for placebo (2n3 %). At
the 1999 meeting of the American Urology Society, it
was reported that 93 % of patients who participated in a
Viagra2 study remained satisfied with the drug after 2
years. A small number of patients (3 %) experienced
some transient visual disturbance that probably resulted
from weak interaction of sildenafil (IC 3438 nM) with
&!
PDE6, the predominant PDE isoform in the retina [4].
Sildenafil can be used for ED patients with a range of
co-existing conditions, such as diabetes, hypertension,
spinal cord injury and following radical prostatectomy,
and can be co-administered with a variety of common
medications. When used as directed, the incidence of
cardiovascular effects has been similar for patients taking
sildenafil and for a population with the same age and risk
profile, both during the clinical development phase and in
more extensive post-approval surveillance [12]. How-
ever, the use of sildenafil is contra-indicated in patients
who use regular, or intermittent, nitrates in any form
[13].
Sildenafil is a major medical and scientific advance that
provides the first effective treatment for ED that can be
administered orally, and has a highly favourable safety
profile. The product is now available in over 100
countries, some 200 000 prescriptions are written each
week and more than 150 million tablets have been
dispensed worldwide. Millions of patients suffering from
ED have benefitted from effective treatment with
sildenafil, and this distressing condition is now much
better understood, and is discussed more openly, than
previously. However, it is sobering to recall that the
sildenafil discovery programme started in 1985, and that
the innovative and dedicated efforts of some 1500 Pfizer
staff, and many external colleagues, were required over
13 years to bring the product to approval.
# 2000 The Biochemical Society and the Medical Research Society
Screen Lead Candidate Product
DRUG DISCOVERY IN THE 21st CENTURY
The research programme that led to sildenafil provides an
example of rational compound design, sequential med-
icinal chemistry and expert pharmacological evaluation
that has characterized pharmaceutical research for the
past decades, and may continue to serve as a model for the
future. However, time scales for such a classical approach
are lengthy, failure rates are high, and the escalating costs
of drug discovery and development may become
unsustainable. Even so, most major pharmaceutical com-
panies have set ambitious targets of doubling, or trebling,
productivity during this decade, and are making major
investments, internally and externally, in genome sciences
and new technologies. These initiatives are expected to
revolutionalize the drug discovery paradigm (Figure 3) by
shortening time scales between identifying a gene se-
quence of interest and submitting a new product for
regulatory approval.
Current bottlenecks are assigning function to novel
genes and proteins, and choosing the most relevant
biological targets for in-depth pursuit. Moreover, while
high-throughput screening initiatives often identify
structures with binding affinities in the 10 M region,
significant chemistry effort is required to transform these
weak hits into lead series that confirm that the desired
pharmacological profile is attainable. When all of the
target biological properties are expressed in a single
molecule, a candidate is nominated for pre-clinical
development, although the chances of reaching approval
are only around 1 in 12. Clearly, there are many
opportunities for improving productivity throughout
this drug discovery paradigm (Figure 3), but there is
considerable debate over which intervention points will
provide the greatest advantage.
There have been significant investments in positional
cloning, where specific gene defects have been correlated
with some diseases, but few tractable drug targets have
emerged. Searching for genes associated with disease may
be more fruitful, since single-nucleotide polymorphism
analysis has been claimed to locate regions of DNA
associated with migraine, Type II diabetes and psoriasis.
However, alternative estimates of the number of single-
nucleotide polymorphisms required to find genes related
to common diseases have risen to between 600 000 and 1
million, which has significant cost and time implications.
Dramatic improvements in DNA sequencing tech-
niques have brought completion of the human genome in
sight, and there are expectations that this will usher in a
new era of biological understanding and drug discovery
opportunities. Although deposition of the complete

human genome sequence in public databases will be an
outstanding scientific achievement, and emotional land-
mark, these data have little intrinsic value without
functional and biological correlates. A rough analogy
might be to compare collecting the names of everyone
living in Framingham, without documenting their roles
and relationships, with the richness of the database from
the Framingham Heart Study.
Delineating the function of genes and their encoded
proteins is a significant challenge, and there are major
initiatives in bioinformatics to attempt to keep pace with
data mountains that seem to increase exponentially. The
dream of translating a gene sequence into a three-
dimensional protein structure in silico is still some way
off, and IBM has announced a $100 million investment to
build a petaflop computer that will be 500 times faster
than todays machines to tackle the challenge of protein
folding. In parallel, structural genomics initiatives aim to
determine three-dimensional protein structures on an
industrial scale and to assign function by shape com-
parison with known biological mediators, since structure
is better conserved than sequence during evolution.
Various technologies are available to knock-out gene
function in vitro and in vivo and to study resultant
changes in phenotype, but none appears to have broad-
scale application, particularly for genes of unknown
function. For example, these are few more emotive or
high-profile genes than BRCA1 and BRCA2, which have
been implicated in breast cancer, yet, after years of
research, a link with DNA repair is only just emerging.
Transcript profiling using DNA microarrays is being
widely used to measure differential gene expression in
normal and diseased tissues, but inconsistent protocols
make comparison between laboratories difficult, since
there is no consensus on what level of change is
significant. A complementary approach is to focus on
changes in protein levels, since these may be more
relevant than differences in gene expression. However, an
extra level of complexity may be introduced, since post-
translational modification allows a single gene to furnish
different proteins, and there is a poor correlation
between mRNA levels and the proteins that mediate
biological processes.
Many of these in silico, in vitro and in vivo techniques
are often described as target validation but, strictly
speaking, this accolade can only be claimed after dem-
onstration of clinical efficacy. These laboratory
experiments simply represent a series of hurdles any
target must cross in order to prioritize multiple options
and increase confidence in final selection. On a more
general note, it is not obvious how many new protein
targets will emerge from the human genome initiative in
addition to the 500 or so that have been exploited to date.
A recent estimate of 500010 000 may be realistic [14], but
these may add detail to established biological pathways
rather than providing a wealth of novel mechanisms.
Science, art and drug discovery 259
After target selection, most research groups take
advantage of the massive improvements in screening
technologies that enable hundreds of thousands of
compounds to be assayed within a few days in order to
identify lead molecules for synthetic pursuit. However,
while it has been estimated that the number of screening
data points has risen from 200 000 in 1990 to 50 million in
1999, concerns have been expressed that few leads or
development candidates have emerged from this para-
digm [15]. One factor in this perceived lack of success
may be that combinatorial chemistry has been widely
used to generate thousands of similar compounds, but
limited synthetic options have not allowed access to the
rich molecular architectures often required for inter-
action with biological targets. In response, emphasis has
now shifted to building compound files with maximum
structural richness and diversity, since designed libraries
should have higher hit rates.
Once a lead has been identified, automated high-speed
parallel synthesis may be appropriate, since wide vari-
ation of substituents around the pharmacophore template
is often required to optimize potency\selectivity and in
vivo performance. An important paradigm shift has been
to focus on drug like lead series with physicochemical
and pharmacokinetic properties consistent with the
rigorous demands of subsequent development. In silico
predictions of preferred molecular properties are being
developed, and lead series selection can be further
enhanced by high-speed docking of virtual compound
libraries with target proteins on the computer screen.
Even so, transforming a hit from a high-throughput
screen into a drug candidate is still the rate-limiting step,
since innovative medicinal chemistry and astute pharma-
cological evaluation have not yet been automated !
Significant gains will be realized through new methods
for predictive toxicology, although it is unlikely that
regulatory authorities will dispense with animal safety
studies in the short term. However, transcript profiling
is being used to demonstrate changes in gene\protein
expression associated with toxic events, so that multiple
lead series can be prioritized, and unnecessary safety
evaluation eliminated.
Only 1 in 12 development candidates reaches the
market, and attrition is roughly 50 % at both the pre-
clinical and clinical stages. It is sobering to realize that
these traditional failure rates and productivity losses
reflect an era when a significant proportion of develop-
ment candidates were improved versions of drugs with
established clinical efficacy, and that these figures may
deteriorate as the focus turns to unproven mechanisms.
Clearly, continued efforts to improve target selection and
to reduce compound-related failure will be essential. In
addition, it will be important to correlate pharmaco-
genetic and surrogate markers with drug response during
Phase 2 trials, so that the significant cost and time scales
for Phase 3 can be reduced through more appropriate
# 2000 The Biochemical Society and the Medical Research Society
260 S. F. Campbell
patient selection. For example, it has recently been shown
that DNA microarrays can be used to display differences
in gene expression in various cancers, and, for diffuse
large B-cell lymphoma, patients could be stratified with
respect to survival, and presumably treatment regimens.
Moreover, if these techniques can identify those patients
that may be subject to adverse events, this could also
facilitate clinical evaluation and avoid the idiosyncratic
reactions that can lead to drug withdrawal.
POSTSCRIPT
The scientific programme that led to sildenafil provides
many examples of the innovation, multi-disciplinary
teamwork and clear decision making that are essential for
success in any research project. However, these well-
established principles may be challenged as drug dis-
covery undergoes a major paradigm shift to harness
advances in genome sciences and new technologies in
order to reduce time scales and improve productivity. No
doubt these ambitions will be realized, but an over-
reliance on automation, robotics and massive data sets is
unlikely to stimulate the high level of creativity that is
fundamental for the discovery of innovative new
medicines. Therefore it will be important to integrate
new scientific advances into an environment that builds
on traditional skills, fosters multidisciplinary interactions
between teams and individuals, and is primed to exploit
Pasteurs Dictum that chance favours the prepared
mind .
ACKNOWLEDGMENTS
It has been a privilege to act as spokesperson for the many
hundreds of Pfizer staff who contributed to the discovery
and development of sildenafil.
# 2000 The Biochemical Society and the Medical Research Society
REFERENCES
1 Palmer, E. (1999) Making the love drug. Chem. Britain,
January, 2426
2 Campbell, S. F. and Roberts, D. A. (1995) Pfizer Research
Proposal
3 Beavo, J. A. (1995) Cyclic nucleotide phosphodiesterases ;
functional implications of multiple isoforms. Physiol. Rev.
75, 725748
4 Wallis, R. M., Corbin, J. D., Francis, S. H. and Ellis, P.
(1999) Tissue distribution of phosphodiesterase families
and the effects of sildenafil on tissue cyclic nucleotides,
platelet function, and the contractile responses of
trabeculae carneae and aortic rings in vitro. Am. J. Cardiol.
83, 3C12C
5 Moreland, R. B., Goldstein, I. and Traish, A. (1998)
Sildenafil, a novel inhibitor of phosphodiesterase type 5 in
human corpus cavernosum smooth muscle cells. Life Sci.
62, 309318
6 Knispel, H. H., Goessl, C. and Beckmann, R. (1992) Nitric
oxide mediates relaxation in rabbit and human corpus
cavernosum smooth muscle. Urol. Res. 20, 253257
7 Rajfer, J., Aronson, W. J., Bush, P. A., Dorey, F. J. and
Ignarro, L. J. (1992) Nitric oxide as a mediator of
relaxation of the corpus cavernosum response to
nonadrenergic, noncholinergic neurotransmission. N. Engl.
J. Med. 326, 9094
8 Ballard, S. A., Gingell, C. J., Tang, K., Turner, L. A., Price,
M. E. and Naylor, A. M. (1998) Effects of sildenafil on the
relaxation of human corpus cavernosum tissue in vitro and
on the activities of cyclic nucleotide phosphodiesterase
isozymes. J. Urol. 159, 21642171
9 Feldman, H. A., Goldstein, I., Hatzichristou, D. G.,
Krane, R. J. and McKinley, J. B. (1994) Impotence and its
medical and psychosocial correlates : results of the
Massachusetts Male Ageing Study. J. Urol. 151, 5461
10 Rosen, R. C., Riley, A., Wagner, G., Osterloh, I. A.,
Kirkpatrick, J. and Mishra, A. (1997) The international
index of erectile dysfunction (IIEF) : a multidimensional
scale for assessment of erectile dysfunction. Urology 49,
822830
11 Goldstein, I., Lue, T. F., Padma-Nathan, H., Rosen, R. C.,
Steers, W. D. and Wicker, P. A. (1998) Oral sildenafil in
the treatment of erectile dysfunction. N. Engl. J. Med. 338,
13971404
12 Zusman, R. M., Morales, A., Glasser, D. B. and Osterloh,
I. H. (1999) Overall cardiovascular profile of sildenafil
citrate. Am. J. Cardiol. 83, 35C44C
13 Zusman, R. M. (ed.) (1999) Cardiovascular data on
sildenafil citrate. Am. J. Cardiol. 83, 1C44C
14 Drews, J. (2000) Drug discovery : a historical perspective.
Science 287, 19601964
15 Lahana, R. (1999) How many leads from HTS ? Drug
Discovery Today 4, 447448